1 /* Copyright (c) 2004-2006, Roger Dingledine, Nick Mathewson.
2 * Copyright (c) 2007-2010, 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"
19 #include "routerlist.h"
22 static void bw_arrays_init(void);
23 static void predicted_ports_init(void);
25 /** Total number of bytes currently allocated in fields used by rephist.c. */
26 uint64_t rephist_total_alloc
=0;
27 /** Number of or_history_t objects currently allocated. */
28 uint32_t rephist_total_num
=0;
30 /** If the total weighted run count of all runs for a router ever falls
31 * below this amount, the router can be treated as having 0 MTBF. */
32 #define STABILITY_EPSILON 0.0001
33 /** Value by which to discount all old intervals for MTBF purposes. This
34 * is compounded every STABILITY_INTERVAL. */
35 #define STABILITY_ALPHA 0.95
36 /** Interval at which to discount all old intervals for MTBF purposes. */
37 #define STABILITY_INTERVAL (12*60*60)
38 /* (This combination of ALPHA, INTERVAL, and EPSILON makes it so that an
39 * interval that just ended counts twice as much as one that ended a week ago,
40 * 20X as much as one that ended a month ago, and routers that have had no
41 * uptime data for about half a year will get forgotten.) */
43 /** History of an OR-\>OR link. */
44 typedef struct link_history_t
{
45 /** When did we start tracking this list? */
47 /** When did we most recently note a change to this link */
49 /** How many times did extending from OR1 to OR2 succeed? */
50 unsigned long n_extend_ok
;
51 /** How many times did extending from OR1 to OR2 fail? */
52 unsigned long n_extend_fail
;
55 /** History of an OR. */
56 typedef struct or_history_t
{
57 /** When did we start tracking this OR? */
59 /** When did we most recently note a change to this OR? */
61 /** How many times did we successfully connect? */
62 unsigned long n_conn_ok
;
63 /** How many times did we try to connect and fail?*/
64 unsigned long n_conn_fail
;
65 /** How many seconds have we been connected to this OR before
68 /** How many seconds have we been unable to connect to this OR before
70 unsigned long downtime
;
71 /** If nonzero, we have been connected since this time. */
73 /** If nonzero, we have been unable to connect since this time. */
76 /* === For MTBF tracking: */
77 /** Weighted sum total of all times that this router has been online.
79 unsigned long weighted_run_length
;
80 /** If the router is now online (according to stability-checking rules),
81 * when did it come online? */
83 /** Sum of weights for runs in weighted_run_length. */
84 double total_run_weights
;
85 /* === For fractional uptime tracking: */
86 time_t start_of_downtime
;
87 unsigned long weighted_uptime
;
88 unsigned long total_weighted_time
;
90 /** Map from hex OR2 identity digest to a link_history_t for the link
91 * from this OR to OR2. */
92 digestmap_t
*link_history_map
;
95 /** When did we last multiply all routers' weighted_run_length and
96 * total_run_weights by STABILITY_ALPHA? */
97 static time_t stability_last_downrated
= 0;
100 static time_t started_tracking_stability
= 0;
102 /** Map from hex OR identity digest to or_history_t. */
103 static digestmap_t
*history_map
= NULL
;
105 /** Return the or_history_t for the OR with identity digest <b>id</b>,
106 * creating it if necessary. */
107 static or_history_t
*
108 get_or_history(const char* id
)
112 if (tor_mem_is_zero(id
, DIGEST_LEN
))
115 hist
= digestmap_get(history_map
, id
);
117 hist
= tor_malloc_zero(sizeof(or_history_t
));
118 rephist_total_alloc
+= sizeof(or_history_t
);
120 hist
->link_history_map
= digestmap_new();
121 hist
->since
= hist
->changed
= time(NULL
);
122 digestmap_set(history_map
, id
, hist
);
127 /** Return the link_history_t for the link from the first named OR to
128 * the second, creating it if necessary. (ORs are identified by
131 static link_history_t
*
132 get_link_history(const char *from_id
, const char *to_id
)
134 or_history_t
*orhist
;
135 link_history_t
*lhist
;
136 orhist
= get_or_history(from_id
);
139 if (tor_mem_is_zero(to_id
, DIGEST_LEN
))
141 lhist
= (link_history_t
*) digestmap_get(orhist
->link_history_map
, to_id
);
143 lhist
= tor_malloc_zero(sizeof(link_history_t
));
144 rephist_total_alloc
+= sizeof(link_history_t
);
145 lhist
->since
= lhist
->changed
= time(NULL
);
146 digestmap_set(orhist
->link_history_map
, to_id
, lhist
);
151 /** Helper: free storage held by a single link history entry. */
153 _free_link_history(void *val
)
155 rephist_total_alloc
-= sizeof(link_history_t
);
159 /** Helper: free storage held by a single OR history entry. */
161 free_or_history(void *_hist
)
163 or_history_t
*hist
= _hist
;
164 digestmap_free(hist
->link_history_map
, _free_link_history
);
165 rephist_total_alloc
-= sizeof(or_history_t
);
170 /** Update an or_history_t object <b>hist</b> so that its uptime/downtime
171 * count is up-to-date as of <b>when</b>.
174 update_or_history(or_history_t
*hist
, time_t when
)
177 if (hist
->up_since
) {
178 tor_assert(!hist
->down_since
);
179 hist
->uptime
+= (when
- hist
->up_since
);
180 hist
->up_since
= when
;
181 } else if (hist
->down_since
) {
182 hist
->downtime
+= (when
- hist
->down_since
);
183 hist
->down_since
= when
;
187 /** Initialize the static data structures for tracking history. */
191 history_map
= digestmap_new();
193 predicted_ports_init();
196 /** Helper: note that we are no longer connected to the router with history
197 * <b>hist</b>. If <b>failed</b>, the connection failed; otherwise, it was
198 * closed correctly. */
200 mark_or_down(or_history_t
*hist
, time_t when
, int failed
)
202 if (hist
->up_since
) {
203 hist
->uptime
+= (when
- hist
->up_since
);
206 if (failed
&& !hist
->down_since
) {
207 hist
->down_since
= when
;
211 /** Helper: note that we are connected to the router with history
214 mark_or_up(or_history_t
*hist
, time_t when
)
216 if (hist
->down_since
) {
217 hist
->downtime
+= (when
- hist
->down_since
);
218 hist
->down_since
= 0;
220 if (!hist
->up_since
) {
221 hist
->up_since
= when
;
225 /** Remember that an attempt to connect to the OR with identity digest
226 * <b>id</b> failed at <b>when</b>.
229 rep_hist_note_connect_failed(const char* id
, time_t when
)
232 hist
= get_or_history(id
);
236 mark_or_down(hist
, when
, 1);
237 hist
->changed
= when
;
240 /** Remember that an attempt to connect to the OR with identity digest
241 * <b>id</b> succeeded at <b>when</b>.
244 rep_hist_note_connect_succeeded(const char* id
, time_t when
)
247 hist
= get_or_history(id
);
251 mark_or_up(hist
, when
);
252 hist
->changed
= when
;
255 /** Remember that we intentionally closed our connection to the OR
256 * with identity digest <b>id</b> at <b>when</b>.
259 rep_hist_note_disconnect(const char* id
, time_t when
)
262 hist
= get_or_history(id
);
265 mark_or_down(hist
, when
, 0);
266 hist
->changed
= when
;
269 /** Remember that our connection to the OR with identity digest
270 * <b>id</b> had an error and stopped working at <b>when</b>.
273 rep_hist_note_connection_died(const char* id
, time_t when
)
277 /* If conn has no identity, it didn't complete its handshake, or something
278 * went wrong. Ignore it.
282 hist
= get_or_history(id
);
285 mark_or_down(hist
, when
, 1);
286 hist
->changed
= when
;
289 /** We have just decided that this router with identity digest <b>id</b> is
290 * reachable, meaning we will give it a "Running" flag for the next while. */
292 rep_hist_note_router_reachable(const char *id
, time_t when
)
294 or_history_t
*hist
= get_or_history(id
);
296 char tbuf
[ISO_TIME_LEN
+1];
300 if (!started_tracking_stability
)
301 started_tracking_stability
= time(NULL
);
302 if (!hist
->start_of_run
) {
303 hist
->start_of_run
= when
;
306 if (hist
->start_of_downtime
) {
309 format_local_iso_time(tbuf
, hist
->start_of_downtime
);
310 log_info(LD_HIST
, "Router %s is now Running; it had been down since %s.",
311 hex_str(id
, DIGEST_LEN
), tbuf
);
313 log_info(LD_HIST
, " (Paradoxically, it was already Running too.)");
315 down_length
= when
- hist
->start_of_downtime
;
316 hist
->total_weighted_time
+= down_length
;
317 hist
->start_of_downtime
= 0;
319 format_local_iso_time(tbuf
, hist
->start_of_run
);
321 log_debug(LD_HIST
, "Router %s is still Running; it has been Running "
322 "since %s", hex_str(id
, DIGEST_LEN
), tbuf
);
324 log_info(LD_HIST
,"Router %s is now Running; it was previously untracked",
325 hex_str(id
, DIGEST_LEN
));
329 /** We have just decided that this router is unreachable, meaning
330 * we are taking away its "Running" flag. */
332 rep_hist_note_router_unreachable(const char *id
, time_t when
)
334 or_history_t
*hist
= get_or_history(id
);
335 char tbuf
[ISO_TIME_LEN
+1];
337 if (!started_tracking_stability
)
338 started_tracking_stability
= time(NULL
);
341 if (hist
->start_of_run
) {
342 /*XXXX We could treat failed connections differently from failed
343 * connect attempts. */
344 long run_length
= when
- hist
->start_of_run
;
345 format_local_iso_time(tbuf
, hist
->start_of_run
);
347 hist
->weighted_run_length
+= run_length
;
348 hist
->total_run_weights
+= 1.0;
349 hist
->start_of_run
= 0;
350 hist
->weighted_uptime
+= run_length
;
351 hist
->total_weighted_time
+= run_length
;
354 log_info(LD_HIST
, "Router %s is now non-Running: it had previously been "
355 "Running since %s. Its total weighted uptime is %lu/%lu.",
356 hex_str(id
, DIGEST_LEN
), tbuf
, hist
->weighted_uptime
,
357 hist
->total_weighted_time
);
359 if (!hist
->start_of_downtime
) {
360 hist
->start_of_downtime
= when
;
363 log_info(LD_HIST
, "Router %s is now non-Running; it was previously "
364 "untracked.", hex_str(id
, DIGEST_LEN
));
367 format_local_iso_time(tbuf
, hist
->start_of_downtime
);
369 log_info(LD_HIST
, "Router %s is still non-Running; it has been "
370 "non-Running since %s.", hex_str(id
, DIGEST_LEN
), tbuf
);
375 /** Helper: Discount all old MTBF data, if it is time to do so. Return
376 * the time at which we should next discount MTBF data. */
378 rep_hist_downrate_old_runs(time_t now
)
380 digestmap_iter_t
*orhist_it
;
387 history_map
= digestmap_new();
388 if (!stability_last_downrated
)
389 stability_last_downrated
= now
;
390 if (stability_last_downrated
+ STABILITY_INTERVAL
> now
)
391 return stability_last_downrated
+ STABILITY_INTERVAL
;
393 /* Okay, we should downrate the data. By how much? */
394 while (stability_last_downrated
+ STABILITY_INTERVAL
< now
) {
395 stability_last_downrated
+= STABILITY_INTERVAL
;
396 alpha
*= STABILITY_ALPHA
;
399 log_info(LD_HIST
, "Discounting all old stability info by a factor of %lf",
402 /* Multiply every w_r_l, t_r_w pair by alpha. */
403 for (orhist_it
= digestmap_iter_init(history_map
);
404 !digestmap_iter_done(orhist_it
);
405 orhist_it
= digestmap_iter_next(history_map
,orhist_it
)) {
406 digestmap_iter_get(orhist_it
, &digest1
, &hist_p
);
409 hist
->weighted_run_length
=
410 (unsigned long)(hist
->weighted_run_length
* alpha
);
411 hist
->total_run_weights
*= alpha
;
413 hist
->weighted_uptime
= (unsigned long)(hist
->weighted_uptime
* alpha
);
414 hist
->total_weighted_time
= (unsigned long)
415 (hist
->total_weighted_time
* alpha
);
418 return stability_last_downrated
+ STABILITY_INTERVAL
;
421 /** Helper: Return the weighted MTBF of the router with history <b>hist</b>. */
423 get_stability(or_history_t
*hist
, time_t when
)
425 unsigned long total
= hist
->weighted_run_length
;
426 double total_weights
= hist
->total_run_weights
;
428 if (hist
->start_of_run
) {
429 /* We're currently in a run. Let total and total_weights hold the values
430 * they would hold if the current run were to end now. */
431 total
+= (when
-hist
->start_of_run
);
432 total_weights
+= 1.0;
434 if (total_weights
< STABILITY_EPSILON
) {
435 /* Round down to zero, and avoid divide-by-zero. */
439 return total
/ total_weights
;
442 /** Return the total amount of time we've been observing, with each run of
443 * time downrated by the appropriate factor. */
445 get_total_weighted_time(or_history_t
*hist
, time_t when
)
447 long total
= hist
->total_weighted_time
;
448 if (hist
->start_of_run
) {
449 total
+= (when
- hist
->start_of_run
);
450 } else if (hist
->start_of_downtime
) {
451 total
+= (when
- hist
->start_of_downtime
);
456 /** Helper: Return the weighted percent-of-time-online of the router with
457 * history <b>hist</b>. */
459 get_weighted_fractional_uptime(or_history_t
*hist
, time_t when
)
461 unsigned long total
= hist
->total_weighted_time
;
462 unsigned long up
= hist
->weighted_uptime
;
464 if (hist
->start_of_run
) {
465 long run_length
= (when
- hist
->start_of_run
);
468 } else if (hist
->start_of_downtime
) {
469 total
+= (when
- hist
->start_of_downtime
);
473 /* Avoid calling anybody's uptime infinity (which should be impossible if
474 * the code is working), or NaN (which can happen for any router we haven't
475 * observed up or down yet). */
479 return ((double) up
) / total
;
482 /** Return an estimated MTBF for the router whose identity digest is
483 * <b>id</b>. Return 0 if the router is unknown. */
485 rep_hist_get_stability(const char *id
, time_t when
)
487 or_history_t
*hist
= get_or_history(id
);
491 return get_stability(hist
, when
);
494 /** Return an estimated percent-of-time-online for the router whose identity
495 * digest is <b>id</b>. Return 0 if the router is unknown. */
497 rep_hist_get_weighted_fractional_uptime(const char *id
, time_t when
)
499 or_history_t
*hist
= get_or_history(id
);
503 return get_weighted_fractional_uptime(hist
, when
);
506 /** Return a number representing how long we've known about the router whose
507 * digest is <b>id</b>. Return 0 if the router is unknown.
509 * Be careful: this measure increases monotonically as we know the router for
510 * longer and longer, but it doesn't increase linearly.
513 rep_hist_get_weighted_time_known(const char *id
, time_t when
)
515 or_history_t
*hist
= get_or_history(id
);
519 return get_total_weighted_time(hist
, when
);
522 /** Return true if we've been measuring MTBFs for long enough to
523 * pronounce on Stability. */
525 rep_hist_have_measured_enough_stability(void)
527 /* XXXX021 This doesn't do so well when we change our opinion
528 * as to whether we're tracking router stability. */
529 return started_tracking_stability
< time(NULL
) - 4*60*60;
532 /** Remember that we successfully extended from the OR with identity
533 * digest <b>from_id</b> to the OR with identity digest
537 rep_hist_note_extend_succeeded(const char *from_id
, const char *to_id
)
539 link_history_t
*hist
;
540 /* log_fn(LOG_WARN, "EXTEND SUCCEEDED: %s->%s",from_name,to_name); */
541 hist
= get_link_history(from_id
, to_id
);
545 hist
->changed
= time(NULL
);
548 /** Remember that we tried to extend from the OR with identity digest
549 * <b>from_id</b> to the OR with identity digest <b>to_name</b>, but
553 rep_hist_note_extend_failed(const char *from_id
, const char *to_id
)
555 link_history_t
*hist
;
556 /* log_fn(LOG_WARN, "EXTEND FAILED: %s->%s",from_name,to_name); */
557 hist
= get_link_history(from_id
, to_id
);
560 ++hist
->n_extend_fail
;
561 hist
->changed
= time(NULL
);
564 /** Log all the reliability data we have remembered, with the chosen
568 rep_hist_dump_stats(time_t now
, int severity
)
570 digestmap_iter_t
*lhist_it
;
571 digestmap_iter_t
*orhist_it
;
572 const char *name1
, *name2
, *digest1
, *digest2
;
573 char hexdigest1
[HEX_DIGEST_LEN
+1];
574 or_history_t
*or_history
;
575 link_history_t
*link_history
;
576 void *or_history_p
, *link_history_p
;
581 unsigned long upt
, downt
;
584 rep_history_clean(now
- get_options()->RephistTrackTime
);
586 log(severity
, LD_HIST
, "--------------- Dumping history information:");
588 for (orhist_it
= digestmap_iter_init(history_map
);
589 !digestmap_iter_done(orhist_it
);
590 orhist_it
= digestmap_iter_next(history_map
,orhist_it
)) {
593 digestmap_iter_get(orhist_it
, &digest1
, &or_history_p
);
594 or_history
= (or_history_t
*) or_history_p
;
596 if ((r
= router_get_by_digest(digest1
)))
600 base16_encode(hexdigest1
, sizeof(hexdigest1
), digest1
, DIGEST_LEN
);
601 update_or_history(or_history
, now
);
602 upt
= or_history
->uptime
;
603 downt
= or_history
->downtime
;
604 s
= get_stability(or_history
, now
);
607 uptime
= ((double)upt
) / (upt
+downt
);
611 log(severity
, LD_HIST
,
612 "OR %s [%s]: %ld/%ld good connections; uptime %ld/%ld sec (%.2f%%); "
613 "wmtbf %lu:%02lu:%02lu",
615 or_history
->n_conn_ok
, or_history
->n_conn_fail
+or_history
->n_conn_ok
,
616 upt
, upt
+downt
, uptime
*100.0,
617 stability
/3600, (stability
/60)%60, stability
%60);
619 if (!digestmap_isempty(or_history
->link_history_map
)) {
620 strlcpy(buffer
, " Extend attempts: ", sizeof(buffer
));
621 len
= strlen(buffer
);
622 for (lhist_it
= digestmap_iter_init(or_history
->link_history_map
);
623 !digestmap_iter_done(lhist_it
);
624 lhist_it
= digestmap_iter_next(or_history
->link_history_map
,
626 digestmap_iter_get(lhist_it
, &digest2
, &link_history_p
);
627 if ((r
= router_get_by_digest(digest2
)))
632 link_history
= (link_history_t
*) link_history_p
;
634 ret
= tor_snprintf(buffer
+len
, 2048-len
, "%s(%ld/%ld); ", name2
,
635 link_history
->n_extend_ok
,
636 link_history
->n_extend_ok
+link_history
->n_extend_fail
);
642 log(severity
, LD_HIST
, "%s", buffer
);
647 /** Remove history info for routers/links that haven't changed since
651 rep_history_clean(time_t before
)
653 int authority
= authdir_mode(get_options());
654 or_history_t
*or_history
;
655 link_history_t
*link_history
;
656 void *or_history_p
, *link_history_p
;
657 digestmap_iter_t
*orhist_it
, *lhist_it
;
660 orhist_it
= digestmap_iter_init(history_map
);
661 while (!digestmap_iter_done(orhist_it
)) {
663 digestmap_iter_get(orhist_it
, &d1
, &or_history_p
);
664 or_history
= or_history_p
;
666 remove
= authority
? (or_history
->total_run_weights
< STABILITY_EPSILON
&&
667 !or_history
->start_of_run
)
668 : (or_history
->changed
< before
);
670 orhist_it
= digestmap_iter_next_rmv(history_map
, orhist_it
);
671 free_or_history(or_history
);
674 for (lhist_it
= digestmap_iter_init(or_history
->link_history_map
);
675 !digestmap_iter_done(lhist_it
); ) {
676 digestmap_iter_get(lhist_it
, &d2
, &link_history_p
);
677 link_history
= link_history_p
;
678 if (link_history
->changed
< before
) {
679 lhist_it
= digestmap_iter_next_rmv(or_history
->link_history_map
,
681 rephist_total_alloc
-= sizeof(link_history_t
);
682 tor_free(link_history
);
685 lhist_it
= digestmap_iter_next(or_history
->link_history_map
,lhist_it
);
687 orhist_it
= digestmap_iter_next(history_map
, orhist_it
);
691 /** Write MTBF data to disk. Return 0 on success, negative on failure.
693 * If <b>missing_means_down</b>, then if we're about to write an entry
694 * that is still considered up but isn't in our routerlist, consider it
697 rep_hist_record_mtbf_data(time_t now
, int missing_means_down
)
699 char time_buf
[ISO_TIME_LEN
+1];
701 digestmap_iter_t
*orhist_it
;
705 open_file_t
*open_file
= NULL
;
709 char *filename
= get_datadir_fname("router-stability");
710 f
= start_writing_to_stdio_file(filename
, OPEN_FLAGS_REPLACE
|O_TEXT
, 0600,
718 * FormatLine *KeywordLine Data
720 * FormatLine = "format 1" NL
721 * KeywordLine = Keyword SP Arguments NL
722 * Data = "data" NL *RouterMTBFLine "." NL
723 * RouterMTBFLine = Fingerprint SP WeightedRunLen SP
724 * TotalRunWeights [SP S=StartRunTime] NL
726 #define PUT(s) STMT_BEGIN if (fputs((s),f)<0) goto err; STMT_END
727 #define PRINTF(args) STMT_BEGIN if (fprintf args <0) goto err; STMT_END
731 format_iso_time(time_buf
, time(NULL
));
732 PRINTF((f
, "stored-at %s\n", time_buf
));
734 if (started_tracking_stability
) {
735 format_iso_time(time_buf
, started_tracking_stability
);
736 PRINTF((f
, "tracked-since %s\n", time_buf
));
738 if (stability_last_downrated
) {
739 format_iso_time(time_buf
, stability_last_downrated
);
740 PRINTF((f
, "last-downrated %s\n", time_buf
));
745 /* XXX Nick: now bridge auths record this for all routers too.
746 * Should we make them record it only for bridge routers? -RD
747 * Not for 0.2.0. -NM */
748 for (orhist_it
= digestmap_iter_init(history_map
);
749 !digestmap_iter_done(orhist_it
);
750 orhist_it
= digestmap_iter_next(history_map
,orhist_it
)) {
751 char dbuf
[HEX_DIGEST_LEN
+1];
752 const char *t
= NULL
;
753 digestmap_iter_get(orhist_it
, &digest
, &or_history_p
);
754 hist
= (or_history_t
*) or_history_p
;
756 base16_encode(dbuf
, sizeof(dbuf
), digest
, DIGEST_LEN
);
758 if (missing_means_down
&& hist
->start_of_run
&&
759 !router_get_by_digest(digest
)) {
760 /* We think this relay is running, but it's not listed in our
761 * routerlist. Somehow it fell out without telling us it went
762 * down. Complain and also correct it. */
764 "Relay '%s' is listed as up in rephist, but it's not in "
765 "our routerlist. Correcting.", dbuf
);
766 rep_hist_note_router_unreachable(digest
, now
);
769 PRINTF((f
, "R %s\n", dbuf
));
770 if (hist
->start_of_run
> 0) {
771 format_iso_time(time_buf
, hist
->start_of_run
);
774 PRINTF((f
, "+MTBF %lu %.5lf%s%s\n",
775 hist
->weighted_run_length
, hist
->total_run_weights
,
776 t
? " S=" : "", t
? t
: ""));
778 if (hist
->start_of_downtime
> 0) {
779 format_iso_time(time_buf
, hist
->start_of_downtime
);
782 PRINTF((f
, "+WFU %lu %lu%s%s\n",
783 hist
->weighted_uptime
, hist
->total_weighted_time
,
784 t
? " S=" : "", t
? t
: ""));
792 return finish_writing_to_file(open_file
);
794 abort_writing_to_file(open_file
);
798 /** Format the current tracked status of the router in <b>hist</b> at time
799 * <b>now</b> for analysis; return it in a newly allocated string. */
801 rep_hist_format_router_status(or_history_t
*hist
, time_t now
)
803 char sor_buf
[ISO_TIME_LEN
+1];
804 char sod_buf
[ISO_TIME_LEN
+1];
807 int up
= 0, down
= 0;
810 if (hist
->start_of_run
) {
811 format_iso_time(sor_buf
, hist
->start_of_run
);
814 if (hist
->start_of_downtime
) {
815 format_iso_time(sod_buf
, hist
->start_of_downtime
);
819 wfu
= get_weighted_fractional_uptime(hist
, now
);
820 mtbf
= get_stability(hist
, now
);
825 " weighted-time %lu\n"
826 " weighted-uptime %lu\n"
828 " weighted-run-length %lu\n"
829 " total-run-weights %lf\n",
830 up
?"uptime-started ":"", up
?sor_buf
:"", up
?" UTC\n":"",
831 down
?"downtime-started ":"", down
?sod_buf
:"", down
?" UTC\n":"",
833 hist
->total_weighted_time
,
834 hist
->weighted_uptime
,
836 hist
->weighted_run_length
,
837 hist
->total_run_weights
842 /** The last stability analysis document that we created, or NULL if we never
843 * have created one. */
844 static char *last_stability_doc
= NULL
;
845 /** The last time we created a stability analysis document, or 0 if we never
846 * have created one. */
847 static time_t built_last_stability_doc_at
= 0;
848 /** Shortest allowable time between building two stability documents. */
849 #define MAX_STABILITY_DOC_BUILD_RATE (3*60)
851 /** Return a pointer to a NUL-terminated document describing our view of the
852 * stability of the routers we've been tracking. Return NULL on failure. */
854 rep_hist_get_router_stability_doc(time_t now
)
858 if (built_last_stability_doc_at
+ MAX_STABILITY_DOC_BUILD_RATE
> now
)
859 return last_stability_doc
;
864 tor_free(last_stability_doc
);
865 chunks
= smartlist_create();
867 if (rep_hist_have_measured_enough_stability()) {
868 smartlist_add(chunks
, tor_strdup("we-have-enough-measurements\n"));
870 smartlist_add(chunks
, tor_strdup("we-do-not-have-enough-measurements\n"));
873 DIGESTMAP_FOREACH(history_map
, id
, or_history_t
*, hist
) {
875 char dbuf
[BASE64_DIGEST_LEN
+1];
876 char header_buf
[512];
878 digest_to_base64(dbuf
, id
);
879 ri
= router_get_by_digest(id
);
881 char *ip
= tor_dup_ip(ri
->addr
);
882 char tbuf
[ISO_TIME_LEN
+1];
883 format_iso_time(tbuf
, ri
->cache_info
.published_on
);
884 tor_snprintf(header_buf
, sizeof(header_buf
),
887 "relevant-flags %s%s%s\n"
888 "declared-uptime %ld\n",
889 dbuf
, ri
->nickname
, ip
,
891 ri
->is_running
? "Running " : "",
892 ri
->is_valid
? "Valid " : "",
893 ri
->is_hibernating
? "Hibernating " : "",
897 tor_snprintf(header_buf
, sizeof(header_buf
),
898 "router %s {no descriptor}\n", dbuf
);
900 smartlist_add(chunks
, tor_strdup(header_buf
));
901 info
= rep_hist_format_router_status(hist
, now
);
903 smartlist_add(chunks
, info
);
905 } DIGESTMAP_FOREACH_END
;
907 result
= smartlist_join_strings(chunks
, "", 0, NULL
);
908 SMARTLIST_FOREACH(chunks
, char *, cp
, tor_free(cp
));
909 smartlist_free(chunks
);
911 last_stability_doc
= result
;
912 built_last_stability_doc_at
= time(NULL
);
916 /** Helper: return the first j >= i such that !strcmpstart(sl[j], prefix) and
917 * such that no line sl[k] with i <= k < j starts with "R ". Return -1 if no
918 * such line exists. */
920 find_next_with(smartlist_t
*sl
, int i
, const char *prefix
)
922 for ( ; i
< smartlist_len(sl
); ++i
) {
923 const char *line
= smartlist_get(sl
, i
);
924 if (!strcmpstart(line
, prefix
))
926 if (!strcmpstart(line
, "R "))
932 /** How many bad times has parse_possibly_bad_iso_time parsed? */
933 static int n_bogus_times
= 0;
934 /** Parse the ISO-formatted time in <b>s</b> into *<b>time_out</b>, but
935 * rounds any pre-1970 date to Jan 1, 1970. */
937 parse_possibly_bad_iso_time(const char *s
, time_t *time_out
)
941 strlcpy(b
, s
, sizeof(b
));
943 year
= (int)tor_parse_long(b
, 10, 0, INT_MAX
, NULL
, NULL
);
949 return parse_iso_time(s
, time_out
);
952 /** We've read a time <b>t</b> from a file stored at <b>stored_at</b>, which
953 * says we started measuring at <b>started_measuring</b>. Return a new number
954 * that's about as much before <b>now</b> as <b>t</b> was before
958 correct_time(time_t t
, time_t now
, time_t stored_at
, time_t started_measuring
)
960 if (t
< started_measuring
- 24*60*60*365)
962 else if (t
< started_measuring
)
963 return started_measuring
;
964 else if (t
> stored_at
)
967 long run_length
= stored_at
- t
;
968 t
= now
- run_length
;
969 if (t
< started_measuring
)
970 t
= started_measuring
;
975 /** Load MTBF data from disk. Returns 0 on success or recoverable error, -1
978 rep_hist_load_mtbf_data(time_t now
)
980 /* XXXX won't handle being called while history is already populated. */
982 const char *line
= NULL
;
984 time_t last_downrated
= 0, stored_at
= 0, tracked_since
= 0;
985 time_t latest_possible_start
= now
;
989 char *filename
= get_datadir_fname("router-stability");
990 char *d
= read_file_to_str(filename
, RFTS_IGNORE_MISSING
, NULL
);
994 lines
= smartlist_create();
995 smartlist_split_string(lines
, d
, "\n", SPLIT_SKIP_SPACE
, 0);
1000 const char *firstline
;
1001 if (smartlist_len(lines
)>4) {
1002 firstline
= smartlist_get(lines
, 0);
1003 if (!strcmpstart(firstline
, "format "))
1004 format
= tor_parse_long(firstline
+strlen("format "),
1005 10, -1, LONG_MAX
, NULL
, NULL
);
1008 if (format
!= 1 && format
!= 2) {
1010 "Unrecognized format in mtbf history file. Skipping.");
1013 for (i
= 1; i
< smartlist_len(lines
); ++i
) {
1014 line
= smartlist_get(lines
, i
);
1015 if (!strcmp(line
, "data"))
1017 if (!strcmpstart(line
, "last-downrated ")) {
1018 if (parse_iso_time(line
+strlen("last-downrated "), &last_downrated
)<0)
1019 log_warn(LD_HIST
,"Couldn't parse downrate time in mtbf "
1022 if (!strcmpstart(line
, "stored-at ")) {
1023 if (parse_iso_time(line
+strlen("stored-at "), &stored_at
)<0)
1024 log_warn(LD_HIST
,"Couldn't parse stored time in mtbf "
1027 if (!strcmpstart(line
, "tracked-since ")) {
1028 if (parse_iso_time(line
+strlen("tracked-since "), &tracked_since
)<0)
1029 log_warn(LD_HIST
,"Couldn't parse started-tracking time in mtbf "
1033 if (last_downrated
> now
)
1034 last_downrated
= now
;
1035 if (tracked_since
> now
)
1036 tracked_since
= now
;
1039 log_warn(LD_HIST
, "No stored time recorded.");
1043 if (line
&& !strcmp(line
, "data"))
1048 for (; i
< smartlist_len(lines
); ++i
) {
1049 char digest
[DIGEST_LEN
];
1050 char hexbuf
[HEX_DIGEST_LEN
+1];
1051 char mtbf_timebuf
[ISO_TIME_LEN
+1];
1052 char wfu_timebuf
[ISO_TIME_LEN
+1];
1053 time_t start_of_run
= 0;
1054 time_t start_of_downtime
= 0;
1055 int have_mtbf
= 0, have_wfu
= 0;
1058 long wt_uptime
= 0, total_wt_time
= 0;
1061 line
= smartlist_get(lines
, i
);
1062 if (!strcmp(line
, "."))
1065 mtbf_timebuf
[0] = '\0';
1066 wfu_timebuf
[0] = '\0';
1069 n
= sscanf(line
, "%40s %ld %lf S=%10s %8s",
1070 hexbuf
, &wrl
, &trw
, mtbf_timebuf
, mtbf_timebuf
+11);
1071 if (n
!= 3 && n
!= 5) {
1072 log_warn(LD_HIST
, "Couldn't scan line %s", escaped(line
));
1078 int mtbf_idx
, wfu_idx
;
1079 if (strcmpstart(line
, "R ") || strlen(line
) < 2+HEX_DIGEST_LEN
)
1081 strlcpy(hexbuf
, line
+2, sizeof(hexbuf
));
1082 mtbf_idx
= find_next_with(lines
, i
+1, "+MTBF ");
1083 wfu_idx
= find_next_with(lines
, i
+1, "+WFU ");
1084 if (mtbf_idx
>= 0) {
1085 const char *mtbfline
= smartlist_get(lines
, mtbf_idx
);
1086 n
= sscanf(mtbfline
, "+MTBF %lu %lf S=%10s %8s",
1087 &wrl
, &trw
, mtbf_timebuf
, mtbf_timebuf
+11);
1088 if (n
== 2 || n
== 4) {
1091 log_warn(LD_HIST
, "Couldn't scan +MTBF line %s",
1096 const char *wfuline
= smartlist_get(lines
, wfu_idx
);
1097 n
= sscanf(wfuline
, "+WFU %lu %lu S=%10s %8s",
1098 &wt_uptime
, &total_wt_time
,
1099 wfu_timebuf
, wfu_timebuf
+11);
1100 if (n
== 2 || n
== 4) {
1103 log_warn(LD_HIST
, "Couldn't scan +WFU line %s", escaped(wfuline
));
1111 if (base16_decode(digest
, DIGEST_LEN
, hexbuf
, HEX_DIGEST_LEN
) < 0) {
1112 log_warn(LD_HIST
, "Couldn't hex string %s", escaped(hexbuf
));
1115 hist
= get_or_history(digest
);
1120 if (mtbf_timebuf
[0]) {
1121 mtbf_timebuf
[10] = ' ';
1122 if (parse_possibly_bad_iso_time(mtbf_timebuf
, &start_of_run
)<0)
1123 log_warn(LD_HIST
, "Couldn't parse time %s",
1124 escaped(mtbf_timebuf
));
1126 hist
->start_of_run
= correct_time(start_of_run
, now
, stored_at
,
1128 if (hist
->start_of_run
< latest_possible_start
+ wrl
)
1129 latest_possible_start
= hist
->start_of_run
- wrl
;
1131 hist
->weighted_run_length
= wrl
;
1132 hist
->total_run_weights
= trw
;
1135 if (wfu_timebuf
[0]) {
1136 wfu_timebuf
[10] = ' ';
1137 if (parse_possibly_bad_iso_time(wfu_timebuf
, &start_of_downtime
)<0)
1138 log_warn(LD_HIST
, "Couldn't parse time %s", escaped(wfu_timebuf
));
1141 hist
->start_of_downtime
= correct_time(start_of_downtime
, now
, stored_at
,
1143 hist
->weighted_uptime
= wt_uptime
;
1144 hist
->total_weighted_time
= total_wt_time
;
1146 if (strcmp(line
, "."))
1147 log_warn(LD_HIST
, "Truncated MTBF file.");
1149 if (tracked_since
< 86400*365) /* Recover from insanely early value. */
1150 tracked_since
= latest_possible_start
;
1152 stability_last_downrated
= last_downrated
;
1153 started_tracking_stability
= tracked_since
;
1159 SMARTLIST_FOREACH(lines
, char *, cp
, tor_free(cp
));
1160 smartlist_free(lines
);
1164 /** For how many seconds do we keep track of individual per-second bandwidth
1166 #define NUM_SECS_ROLLING_MEASURE 10
1167 /** How large are the intervals for which we track and report bandwidth use? */
1168 #define NUM_SECS_BW_SUM_INTERVAL (15*60)
1169 /** How far in the past do we remember and publish bandwidth use? */
1170 #define NUM_SECS_BW_SUM_IS_VALID (24*60*60)
1171 /** How many bandwidth usage intervals do we remember? (derived) */
1172 #define NUM_TOTALS (NUM_SECS_BW_SUM_IS_VALID/NUM_SECS_BW_SUM_INTERVAL)
1174 /** Structure to track bandwidth use, and remember the maxima for a given
1177 typedef struct bw_array_t
{
1178 /** Observation array: Total number of bytes transferred in each of the last
1179 * NUM_SECS_ROLLING_MEASURE seconds. This is used as a circular array. */
1180 uint64_t obs
[NUM_SECS_ROLLING_MEASURE
];
1181 int cur_obs_idx
; /**< Current position in obs. */
1182 time_t cur_obs_time
; /**< Time represented in obs[cur_obs_idx] */
1183 uint64_t total_obs
; /**< Total for all members of obs except
1184 * obs[cur_obs_idx] */
1185 uint64_t max_total
; /**< Largest value that total_obs has taken on in the
1186 * current period. */
1187 uint64_t total_in_period
; /**< Total bytes transferred in the current
1190 /** When does the next period begin? */
1192 /** Where in 'maxima' should the maximum bandwidth usage for the current
1193 * period be stored? */
1195 /** How many values in maxima/totals have been set ever? */
1197 /** Circular array of the maximum
1198 * bandwidth-per-NUM_SECS_ROLLING_MEASURE usage for the last
1199 * NUM_TOTALS periods */
1200 uint64_t maxima
[NUM_TOTALS
];
1201 /** Circular array of the total bandwidth usage for the last NUM_TOTALS
1203 uint64_t totals
[NUM_TOTALS
];
1206 /** Shift the current period of b forward by one. */
1208 commit_max(bw_array_t
*b
)
1210 /* Store total from current period. */
1211 b
->totals
[b
->next_max_idx
] = b
->total_in_period
;
1212 /* Store maximum from current period. */
1213 b
->maxima
[b
->next_max_idx
++] = b
->max_total
;
1214 /* Advance next_period and next_max_idx */
1215 b
->next_period
+= NUM_SECS_BW_SUM_INTERVAL
;
1216 if (b
->next_max_idx
== NUM_TOTALS
)
1217 b
->next_max_idx
= 0;
1218 if (b
->num_maxes_set
< NUM_TOTALS
)
1220 /* Reset max_total. */
1222 /* Reset total_in_period. */
1223 b
->total_in_period
= 0;
1226 /** Shift the current observation time of 'b' forward by one second. */
1228 advance_obs(bw_array_t
*b
)
1233 /* Calculate the total bandwidth for the last NUM_SECS_ROLLING_MEASURE
1234 * seconds; adjust max_total as needed.*/
1235 total
= b
->total_obs
+ b
->obs
[b
->cur_obs_idx
];
1236 if (total
> b
->max_total
)
1237 b
->max_total
= total
;
1239 nextidx
= b
->cur_obs_idx
+1;
1240 if (nextidx
== NUM_SECS_ROLLING_MEASURE
)
1243 b
->total_obs
= total
- b
->obs
[nextidx
];
1245 b
->cur_obs_idx
= nextidx
;
1247 if (++b
->cur_obs_time
>= b
->next_period
)
1251 /** Add <b>n</b> bytes to the number of bytes in <b>b</b> for second
1254 add_obs(bw_array_t
*b
, time_t when
, uint64_t n
)
1256 /* Don't record data in the past. */
1257 if (when
<b
->cur_obs_time
)
1259 /* If we're currently adding observations for an earlier second than
1260 * 'when', advance b->cur_obs_time and b->cur_obs_idx by an
1261 * appropriate number of seconds, and do all the other housekeeping */
1262 while (when
>b
->cur_obs_time
)
1265 b
->obs
[b
->cur_obs_idx
] += n
;
1266 b
->total_in_period
+= n
;
1269 /** Allocate, initialize, and return a new bw_array. */
1275 b
= tor_malloc_zero(sizeof(bw_array_t
));
1276 rephist_total_alloc
+= sizeof(bw_array_t
);
1278 b
->cur_obs_time
= start
;
1279 b
->next_period
= start
+ NUM_SECS_BW_SUM_INTERVAL
;
1283 /** Recent history of bandwidth observations for read operations. */
1284 static bw_array_t
*read_array
= NULL
;
1285 /** Recent history of bandwidth observations for write operations. */
1286 static bw_array_t
*write_array
= NULL
;
1287 /** Recent history of bandwidth observations for read operations for the
1288 directory protocol. */
1289 static bw_array_t
*dir_read_array
= NULL
;
1290 /** Recent history of bandwidth observations for write operations for the
1291 directory protocol. */
1292 static bw_array_t
*dir_write_array
= NULL
;
1294 /** Set up [dir-]read_array and [dir-]write_array. */
1296 bw_arrays_init(void)
1298 read_array
= bw_array_new();
1299 write_array
= bw_array_new();
1300 dir_read_array
= bw_array_new();
1301 dir_write_array
= bw_array_new();
1304 /** We read <b>num_bytes</b> more bytes in second <b>when</b>.
1306 * Add num_bytes to the current running total for <b>when</b>.
1308 * <b>when</b> can go back to time, but it's safe to ignore calls
1309 * earlier than the latest <b>when</b> you've heard of.
1312 rep_hist_note_bytes_written(size_t num_bytes
, time_t when
)
1314 /* Maybe a circular array for recent seconds, and step to a new point
1315 * every time a new second shows up. Or simpler is to just to have
1316 * a normal array and push down each item every second; it's short.
1318 /* When a new second has rolled over, compute the sum of the bytes we've
1319 * seen over when-1 to when-1-NUM_SECS_ROLLING_MEASURE, and stick it
1320 * somewhere. See rep_hist_bandwidth_assess() below.
1322 add_obs(write_array
, when
, num_bytes
);
1325 /** We wrote <b>num_bytes</b> more bytes in second <b>when</b>.
1326 * (like rep_hist_note_bytes_written() above)
1329 rep_hist_note_bytes_read(size_t num_bytes
, time_t when
)
1331 /* if we're smart, we can make this func and the one above share code */
1332 add_obs(read_array
, when
, num_bytes
);
1335 /** We wrote <b>num_bytes</b> more directory bytes in second <b>when</b>.
1336 * (like rep_hist_note_bytes_written() above)
1339 rep_hist_note_dir_bytes_written(size_t num_bytes
, time_t when
)
1341 add_obs(dir_write_array
, when
, num_bytes
);
1344 /** We read <b>num_bytes</b> more directory bytes in second <b>when</b>.
1345 * (like rep_hist_note_bytes_written() above)
1348 rep_hist_note_dir_bytes_read(size_t num_bytes
, time_t when
)
1350 add_obs(dir_read_array
, when
, num_bytes
);
1353 /** Helper: Return the largest value in b->maxima. (This is equal to the
1354 * most bandwidth used in any NUM_SECS_ROLLING_MEASURE period for the last
1355 * NUM_SECS_BW_SUM_IS_VALID seconds.)
1358 find_largest_max(bw_array_t
*b
)
1363 for (i
=0; i
<NUM_TOTALS
; ++i
) {
1364 if (b
->maxima
[i
]>max
)
1370 /** Find the largest sums in the past NUM_SECS_BW_SUM_IS_VALID (roughly)
1371 * seconds. Find one sum for reading and one for writing. They don't have
1372 * to be at the same time.
1374 * Return the smaller of these sums, divided by NUM_SECS_ROLLING_MEASURE.
1377 rep_hist_bandwidth_assess(void)
1380 r
= find_largest_max(read_array
);
1381 w
= find_largest_max(write_array
);
1383 return (int)(U64_TO_DBL(w
)/NUM_SECS_ROLLING_MEASURE
);
1385 return (int)(U64_TO_DBL(r
)/NUM_SECS_ROLLING_MEASURE
);
1388 /** Print the bandwidth history of b (either [dir-]read_array or
1389 * [dir-]write_array) into the buffer pointed to by buf. The format is
1390 * simply comma separated numbers, from oldest to newest.
1392 * It returns the number of bytes written.
1395 rep_hist_fill_bandwidth_history(char *buf
, size_t len
, bw_array_t
*b
)
1399 or_options_t
*options
= get_options();
1402 if (b
->num_maxes_set
<= b
->next_max_idx
) {
1403 /* We haven't been through the circular array yet; time starts at i=0.*/
1406 /* We've been around the array at least once. The next i to be
1407 overwritten is the oldest. */
1408 i
= b
->next_max_idx
;
1411 if (options
->RelayBandwidthRate
) {
1412 /* We don't want to report that we used more bandwidth than the max we're
1413 * willing to relay; otherwise everybody will know how much traffic
1414 * we used ourself. */
1415 cutoff
= options
->RelayBandwidthRate
* NUM_SECS_BW_SUM_INTERVAL
;
1417 cutoff
= UINT64_MAX
;
1420 for (n
=0; n
<b
->num_maxes_set
; ++n
,++i
) {
1422 if (i
>= NUM_TOTALS
)
1424 tor_assert(i
< NUM_TOTALS
);
1425 /* Round the bandwidth used down to the nearest 1k. */
1426 total
= b
->totals
[i
] & ~0x3ff;
1430 if (n
==(b
->num_maxes_set
-1))
1431 tor_snprintf(cp
, len
-(cp
-buf
), U64_FORMAT
, U64_PRINTF_ARG(total
));
1433 tor_snprintf(cp
, len
-(cp
-buf
), U64_FORMAT
",", U64_PRINTF_ARG(total
));
1439 /** Allocate and return lines for representing this server's bandwidth
1440 * history in its descriptor.
1443 rep_hist_get_bandwidth_lines(void)
1446 char t
[ISO_TIME_LEN
+1];
1448 bw_array_t
*b
= NULL
;
1449 const char *desc
= NULL
;
1452 /* opt [dirreq-](read|write)-history yyyy-mm-dd HH:MM:SS (n s) n,n,n... */
1453 len
= (67+21*NUM_TOTALS
)*4;
1454 buf
= tor_malloc_zero(len
);
1460 desc
= "write-history";
1464 desc
= "read-history";
1467 b
= dir_write_array
;
1468 desc
= "dirreq-write-history";
1472 desc
= "dirreq-read-history";
1476 format_iso_time(t
, b
->next_period
-NUM_SECS_BW_SUM_INTERVAL
);
1477 tor_snprintf(cp
, len
-(cp
-buf
), "%s %s (%d s) ",
1478 desc
, t
, NUM_SECS_BW_SUM_INTERVAL
);
1480 cp
+= rep_hist_fill_bandwidth_history(cp
, len
-(cp
-buf
), b
);
1481 strlcat(cp
, "\n", len
-(cp
-buf
));
1487 /** Update <b>state</b> with the newest bandwidth history. */
1489 rep_hist_update_state(or_state_t
*state
)
1493 smartlist_t
**s_values
= NULL
;
1494 time_t *s_begins
= NULL
;
1495 int *s_interval
= NULL
;
1496 bw_array_t
*b
= NULL
;
1498 len
= 20*NUM_TOTALS
+1;
1499 buf
= tor_malloc_zero(len
);
1505 s_begins
= &state
->BWHistoryWriteEnds
;
1506 s_interval
= &state
->BWHistoryWriteInterval
;
1507 s_values
= &state
->BWHistoryWriteValues
;
1511 s_begins
= &state
->BWHistoryReadEnds
;
1512 s_interval
= &state
->BWHistoryReadInterval
;
1513 s_values
= &state
->BWHistoryReadValues
;
1516 b
= dir_write_array
;
1517 s_begins
= &state
->BWHistoryDirWriteEnds
;
1518 s_interval
= &state
->BWHistoryDirWriteInterval
;
1519 s_values
= &state
->BWHistoryDirWriteValues
;
1523 s_begins
= &state
->BWHistoryDirReadEnds
;
1524 s_interval
= &state
->BWHistoryDirReadInterval
;
1525 s_values
= &state
->BWHistoryDirReadValues
;
1529 SMARTLIST_FOREACH(*s_values
, char *, val
, tor_free(val
));
1530 smartlist_free(*s_values
);
1532 if (! server_mode(get_options())) {
1533 /* Clients don't need to store bandwidth history persistently;
1534 * force these values to the defaults. */
1535 /* FFFF we should pull the default out of config.c's state table,
1536 * so we don't have two defaults. */
1537 if (*s_begins
!= 0 || *s_interval
!= 900) {
1538 time_t now
= time(NULL
);
1539 time_t save_at
= get_options()->AvoidDiskWrites
? now
+3600 : now
+600;
1540 or_state_mark_dirty(state
, save_at
);
1544 *s_values
= smartlist_create();
1547 *s_begins
= b
->next_period
;
1548 *s_interval
= NUM_SECS_BW_SUM_INTERVAL
;
1550 cp
+= rep_hist_fill_bandwidth_history(cp
, len
, b
);
1551 tor_snprintf(cp
, len
-(cp
-buf
), cp
== buf
? U64_FORMAT
: ","U64_FORMAT
,
1552 U64_PRINTF_ARG(b
->total_in_period
));
1553 *s_values
= smartlist_create();
1554 if (server_mode(get_options()))
1555 smartlist_split_string(*s_values
, buf
, ",", SPLIT_SKIP_SPACE
, 0);
1558 if (server_mode(get_options())) {
1559 or_state_mark_dirty(get_or_state(), time(NULL
)+(2*3600));
1563 /** Set bandwidth history from our saved state. */
1565 rep_hist_load_state(or_state_t
*state
, char **err
)
1567 time_t s_begins
= 0, start
;
1568 time_t now
= time(NULL
);
1573 smartlist_t
*s_values
= NULL
;
1574 bw_array_t
*b
= NULL
;
1576 /* Assert they already have been malloced */
1577 tor_assert(read_array
&& write_array
);
1583 s_begins
= state
->BWHistoryWriteEnds
;
1584 s_interval
= state
->BWHistoryWriteInterval
;
1585 s_values
= state
->BWHistoryWriteValues
;
1589 s_begins
= state
->BWHistoryReadEnds
;
1590 s_interval
= state
->BWHistoryReadInterval
;
1591 s_values
= state
->BWHistoryReadValues
;
1594 b
= dir_write_array
;
1595 s_begins
= state
->BWHistoryDirWriteEnds
;
1596 s_interval
= state
->BWHistoryDirWriteInterval
;
1597 s_values
= state
->BWHistoryDirWriteValues
;
1601 s_begins
= state
->BWHistoryDirReadEnds
;
1602 s_interval
= state
->BWHistoryDirReadInterval
;
1603 s_values
= state
->BWHistoryDirReadValues
;
1606 if (s_values
&& s_begins
>= now
- NUM_SECS_BW_SUM_INTERVAL
*NUM_TOTALS
) {
1607 start
= s_begins
- s_interval
*(smartlist_len(s_values
));
1610 b
->cur_obs_time
= start
;
1611 b
->next_period
= start
+ NUM_SECS_BW_SUM_INTERVAL
;
1612 SMARTLIST_FOREACH(s_values
, char *, cp
, {
1613 v
= tor_parse_uint64(cp
, 10, 0, UINT64_MAX
, &ok
, NULL
);
1616 log_notice(LD_HIST
, "Could not parse '%s' into a number.'", cp
);
1619 add_obs(b
, start
, v
);
1620 start
+= NUM_SECS_BW_SUM_INTERVAL
;
1625 /* Clean up maxima and observed */
1626 /* Do we really want to zero this for the purpose of max capacity? */
1627 for (i
=0; i
<NUM_SECS_ROLLING_MEASURE
; ++i
) {
1631 for (i
=0; i
<NUM_TOTALS
; ++i
) {
1638 *err
= tor_strdup("Parsing of bandwidth history values failed");
1639 /* and create fresh arrays */
1640 tor_free(read_array
);
1641 tor_free(write_array
);
1642 read_array
= bw_array_new();
1643 write_array
= bw_array_new();
1649 /*********************************************************************/
1651 /** A list of port numbers that have been used recently. */
1652 static smartlist_t
*predicted_ports_list
=NULL
;
1653 /** The corresponding most recently used time for each port. */
1654 static smartlist_t
*predicted_ports_times
=NULL
;
1656 /** We just got an application request for a connection with
1657 * port <b>port</b>. Remember it for the future, so we can keep
1658 * some circuits open that will exit to this port.
1661 add_predicted_port(time_t now
, uint16_t port
)
1663 /* XXXX we could just use uintptr_t here, I think. */
1664 uint16_t *tmp_port
= tor_malloc(sizeof(uint16_t));
1665 time_t *tmp_time
= tor_malloc(sizeof(time_t));
1668 rephist_total_alloc
+= sizeof(uint16_t) + sizeof(time_t);
1669 smartlist_add(predicted_ports_list
, tmp_port
);
1670 smartlist_add(predicted_ports_times
, tmp_time
);
1673 /** Initialize whatever memory and structs are needed for predicting
1674 * which ports will be used. Also seed it with port 80, so we'll build
1675 * circuits on start-up.
1678 predicted_ports_init(void)
1680 predicted_ports_list
= smartlist_create();
1681 predicted_ports_times
= smartlist_create();
1682 add_predicted_port(time(NULL
), 80); /* add one to kickstart us */
1685 /** Free whatever memory is needed for predicting which ports will
1689 predicted_ports_free(void)
1691 rephist_total_alloc
-= smartlist_len(predicted_ports_list
)*sizeof(uint16_t);
1692 SMARTLIST_FOREACH(predicted_ports_list
, char *, cp
, tor_free(cp
));
1693 smartlist_free(predicted_ports_list
);
1694 rephist_total_alloc
-= smartlist_len(predicted_ports_times
)*sizeof(time_t);
1695 SMARTLIST_FOREACH(predicted_ports_times
, char *, cp
, tor_free(cp
));
1696 smartlist_free(predicted_ports_times
);
1699 /** Remember that <b>port</b> has been asked for as of time <b>now</b>.
1700 * This is used for predicting what sorts of streams we'll make in the
1701 * future and making exit circuits to anticipate that.
1704 rep_hist_note_used_port(time_t now
, uint16_t port
)
1710 tor_assert(predicted_ports_list
);
1711 tor_assert(predicted_ports_times
);
1713 if (!port
) /* record nothing */
1716 for (i
= 0; i
< smartlist_len(predicted_ports_list
); ++i
) {
1717 tmp_port
= smartlist_get(predicted_ports_list
, i
);
1718 tmp_time
= smartlist_get(predicted_ports_times
, i
);
1719 if (*tmp_port
== port
) {
1724 /* it's not there yet; we need to add it */
1725 add_predicted_port(now
, port
);
1728 /** For this long after we've seen a request for a given port, assume that
1729 * we'll want to make connections to the same port in the future. */
1730 #define PREDICTED_CIRCS_RELEVANCE_TIME (60*60)
1732 /** Return a pointer to the list of port numbers that
1733 * are likely to be asked for in the near future.
1735 * The caller promises not to mess with it.
1738 rep_hist_get_predicted_ports(time_t now
)
1744 tor_assert(predicted_ports_list
);
1745 tor_assert(predicted_ports_times
);
1747 /* clean out obsolete entries */
1748 for (i
= 0; i
< smartlist_len(predicted_ports_list
); ++i
) {
1749 tmp_time
= smartlist_get(predicted_ports_times
, i
);
1750 if (*tmp_time
+ PREDICTED_CIRCS_RELEVANCE_TIME
< now
) {
1751 tmp_port
= smartlist_get(predicted_ports_list
, i
);
1752 log_debug(LD_CIRC
, "Expiring predicted port %d", *tmp_port
);
1753 smartlist_del(predicted_ports_list
, i
);
1754 smartlist_del(predicted_ports_times
, i
);
1755 rephist_total_alloc
-= sizeof(uint16_t)+sizeof(time_t);
1761 return predicted_ports_list
;
1764 /** The user asked us to do a resolve. Rather than keeping track of
1765 * timings and such of resolves, we fake it for now by treating
1766 * it the same way as a connection to port 80. This way we will continue
1767 * to have circuits lying around if the user only uses Tor for resolves.
1770 rep_hist_note_used_resolve(time_t now
)
1772 rep_hist_note_used_port(now
, 80);
1775 /** The last time at which we needed an internal circ. */
1776 static time_t predicted_internal_time
= 0;
1777 /** The last time we needed an internal circ with good uptime. */
1778 static time_t predicted_internal_uptime_time
= 0;
1779 /** The last time we needed an internal circ with good capacity. */
1780 static time_t predicted_internal_capacity_time
= 0;
1782 /** Remember that we used an internal circ at time <b>now</b>. */
1784 rep_hist_note_used_internal(time_t now
, int need_uptime
, int need_capacity
)
1786 predicted_internal_time
= now
;
1788 predicted_internal_uptime_time
= now
;
1790 predicted_internal_capacity_time
= now
;
1793 /** Return 1 if we've used an internal circ recently; else return 0. */
1795 rep_hist_get_predicted_internal(time_t now
, int *need_uptime
,
1798 if (!predicted_internal_time
) { /* initialize it */
1799 predicted_internal_time
= now
;
1800 predicted_internal_uptime_time
= now
;
1801 predicted_internal_capacity_time
= now
;
1803 if (predicted_internal_time
+ PREDICTED_CIRCS_RELEVANCE_TIME
< now
)
1804 return 0; /* too long ago */
1805 if (predicted_internal_uptime_time
+ PREDICTED_CIRCS_RELEVANCE_TIME
>= now
)
1807 // Always predict that we need capacity.
1812 /** Any ports used lately? These are pre-seeded if we just started
1813 * up or if we're running a hidden service. */
1815 any_predicted_circuits(time_t now
)
1817 return smartlist_len(predicted_ports_list
) ||
1818 predicted_internal_time
+ PREDICTED_CIRCS_RELEVANCE_TIME
>= now
;
1821 /** Return 1 if we have no need for circuits currently, else return 0. */
1823 rep_hist_circbuilding_dormant(time_t now
)
1825 if (any_predicted_circuits(now
))
1828 /* see if we'll still need to build testing circuits */
1829 if (server_mode(get_options()) &&
1830 (!check_whether_orport_reachable() || !circuit_enough_testing_circs()))
1832 if (!check_whether_dirport_reachable())
1838 /** Structure to track how many times we've done each public key operation. */
1840 /** How many directory objects have we signed? */
1841 unsigned long n_signed_dir_objs
;
1842 /** How many routerdescs have we signed? */
1843 unsigned long n_signed_routerdescs
;
1844 /** How many directory objects have we verified? */
1845 unsigned long n_verified_dir_objs
;
1846 /** How many routerdescs have we verified */
1847 unsigned long n_verified_routerdescs
;
1848 /** How many onionskins have we encrypted to build circuits? */
1849 unsigned long n_onionskins_encrypted
;
1850 /** How many onionskins have we decrypted to do circuit build requests? */
1851 unsigned long n_onionskins_decrypted
;
1852 /** How many times have we done the TLS handshake as a client? */
1853 unsigned long n_tls_client_handshakes
;
1854 /** How many times have we done the TLS handshake as a server? */
1855 unsigned long n_tls_server_handshakes
;
1856 /** How many PK operations have we done as a hidden service client? */
1857 unsigned long n_rend_client_ops
;
1858 /** How many PK operations have we done as a hidden service midpoint? */
1859 unsigned long n_rend_mid_ops
;
1860 /** How many PK operations have we done as a hidden service provider? */
1861 unsigned long n_rend_server_ops
;
1862 } pk_op_counts
= {0,0,0,0,0,0,0,0,0,0,0};
1864 /** Increment the count of the number of times we've done <b>operation</b>. */
1866 note_crypto_pk_op(pk_op_t operation
)
1871 pk_op_counts
.n_signed_dir_objs
++;
1874 pk_op_counts
.n_signed_routerdescs
++;
1877 pk_op_counts
.n_verified_dir_objs
++;
1880 pk_op_counts
.n_verified_routerdescs
++;
1883 pk_op_counts
.n_onionskins_encrypted
++;
1886 pk_op_counts
.n_onionskins_decrypted
++;
1888 case TLS_HANDSHAKE_C
:
1889 pk_op_counts
.n_tls_client_handshakes
++;
1891 case TLS_HANDSHAKE_S
:
1892 pk_op_counts
.n_tls_server_handshakes
++;
1895 pk_op_counts
.n_rend_client_ops
++;
1898 pk_op_counts
.n_rend_mid_ops
++;
1901 pk_op_counts
.n_rend_server_ops
++;
1904 log_warn(LD_BUG
, "Unknown pk operation %d", operation
);
1908 /** Log the number of times we've done each public/private-key operation. */
1910 dump_pk_ops(int severity
)
1912 log(severity
, LD_HIST
,
1913 "PK operations: %lu directory objects signed, "
1914 "%lu directory objects verified, "
1915 "%lu routerdescs signed, "
1916 "%lu routerdescs verified, "
1917 "%lu onionskins encrypted, "
1918 "%lu onionskins decrypted, "
1919 "%lu client-side TLS handshakes, "
1920 "%lu server-side TLS handshakes, "
1921 "%lu rendezvous client operations, "
1922 "%lu rendezvous middle operations, "
1923 "%lu rendezvous server operations.",
1924 pk_op_counts
.n_signed_dir_objs
,
1925 pk_op_counts
.n_verified_dir_objs
,
1926 pk_op_counts
.n_signed_routerdescs
,
1927 pk_op_counts
.n_verified_routerdescs
,
1928 pk_op_counts
.n_onionskins_encrypted
,
1929 pk_op_counts
.n_onionskins_decrypted
,
1930 pk_op_counts
.n_tls_client_handshakes
,
1931 pk_op_counts
.n_tls_server_handshakes
,
1932 pk_op_counts
.n_rend_client_ops
,
1933 pk_op_counts
.n_rend_mid_ops
,
1934 pk_op_counts
.n_rend_server_ops
);
1937 /*** Exit port statistics ***/
1939 /* Some constants */
1940 /** To what multiple should byte numbers be rounded up? */
1941 #define EXIT_STATS_ROUND_UP_BYTES 1024
1942 /** To what multiple should stream counts be rounded up? */
1943 #define EXIT_STATS_ROUND_UP_STREAMS 4
1944 /** Number of TCP ports */
1945 #define EXIT_STATS_NUM_PORTS 65536
1946 /** Top n ports that will be included in exit stats. */
1947 #define EXIT_STATS_TOP_N_PORTS 10
1949 /* The following data structures are arrays and no fancy smartlists or maps,
1950 * so that all write operations can be done in constant time. This comes at
1951 * the price of some memory (1.25 MB) and linear complexity when writing
1952 * stats for measuring relays. */
1953 /** Number of bytes read in current period by exit port */
1954 static uint64_t *exit_bytes_read
= NULL
;
1955 /** Number of bytes written in current period by exit port */
1956 static uint64_t *exit_bytes_written
= NULL
;
1957 /** Number of streams opened in current period by exit port */
1958 static uint32_t *exit_streams
= NULL
;
1960 /** Start time of exit stats or 0 if we're not collecting exit stats. */
1961 static time_t start_of_exit_stats_interval
;
1963 /** Initialize exit port stats. */
1965 rep_hist_exit_stats_init(time_t now
)
1967 start_of_exit_stats_interval
= now
;
1968 exit_bytes_read
= tor_malloc_zero(EXIT_STATS_NUM_PORTS
*
1970 exit_bytes_written
= tor_malloc_zero(EXIT_STATS_NUM_PORTS
*
1972 exit_streams
= tor_malloc_zero(EXIT_STATS_NUM_PORTS
*
1976 /** Reset counters for exit port statistics. */
1978 rep_hist_reset_exit_stats(time_t now
)
1980 start_of_exit_stats_interval
= now
;
1981 memset(exit_bytes_read
, 0, EXIT_STATS_NUM_PORTS
* sizeof(uint64_t));
1982 memset(exit_bytes_written
, 0, EXIT_STATS_NUM_PORTS
* sizeof(uint64_t));
1983 memset(exit_streams
, 0, EXIT_STATS_NUM_PORTS
* sizeof(uint32_t));
1986 /** Stop collecting exit port stats in a way that we can re-start doing
1987 * so in rep_hist_exit_stats_init(). */
1989 rep_hist_exit_stats_term(void)
1991 start_of_exit_stats_interval
= 0;
1992 tor_free(exit_bytes_read
);
1993 tor_free(exit_bytes_written
);
1994 tor_free(exit_streams
);
1997 /** Helper: compare two ints. */
1999 _compare_int(const void *x
, const void *y
) {
2000 return (*(int*)x
- *(int*)y
);
2003 /** Return a newly allocated string containing the exit port statistics
2004 * until <b>now</b>, or NULL if we're not collecting exit stats. */
2006 rep_hist_format_exit_stats(time_t now
)
2008 int i
, j
, top_elements
= 0, cur_min_idx
= 0, cur_port
;
2009 uint64_t top_bytes
[EXIT_STATS_TOP_N_PORTS
];
2010 int top_ports
[EXIT_STATS_TOP_N_PORTS
];
2011 uint64_t cur_bytes
= 0, other_read
= 0, other_written
= 0,
2012 total_read
= 0, total_written
= 0;
2013 uint32_t total_streams
= 0, other_streams
= 0;
2015 smartlist_t
*written_strings
, *read_strings
, *streams_strings
;
2016 char *written_string
, *read_string
, *streams_string
;
2017 char t
[ISO_TIME_LEN
+1];
2020 if (!start_of_exit_stats_interval
)
2021 return NULL
; /* Not initialized. */
2023 /* Go through all ports to find the n ports that saw most written and
2025 for (i
= 1; i
< EXIT_STATS_NUM_PORTS
; i
++) {
2026 total_read
+= exit_bytes_read
[i
];
2027 total_written
+= exit_bytes_written
[i
];
2028 total_streams
+= exit_streams
[i
];
2029 cur_bytes
= exit_bytes_read
[i
] + exit_bytes_written
[i
];
2030 if (cur_bytes
== 0) {
2033 if (top_elements
< EXIT_STATS_TOP_N_PORTS
) {
2034 top_bytes
[top_elements
] = cur_bytes
;
2035 top_ports
[top_elements
++] = i
;
2036 } else if (cur_bytes
> top_bytes
[cur_min_idx
]) {
2037 top_bytes
[cur_min_idx
] = cur_bytes
;
2038 top_ports
[cur_min_idx
] = i
;
2043 for (j
= 1; j
< top_elements
; j
++) {
2044 if (top_bytes
[j
] < top_bytes
[cur_min_idx
]) {
2050 /* Add observations of top ports to smartlists. */
2051 written_strings
= smartlist_create();
2052 read_strings
= smartlist_create();
2053 streams_strings
= smartlist_create();
2054 other_read
= total_read
;
2055 other_written
= total_written
;
2056 other_streams
= total_streams
;
2057 qsort(top_ports
, top_elements
, sizeof(int), _compare_int
);
2058 for (j
= 0; j
< top_elements
; j
++) {
2059 cur_port
= top_ports
[j
];
2060 if (exit_bytes_written
[cur_port
] > 0) {
2061 uint64_t num
= round_uint64_to_next_multiple_of(
2062 exit_bytes_written
[cur_port
],
2063 EXIT_STATS_ROUND_UP_BYTES
);
2066 tor_asprintf(&buf
, "%d="U64_FORMAT
, cur_port
, U64_PRINTF_ARG(num
));
2067 smartlist_add(written_strings
, buf
);
2068 other_written
-= exit_bytes_written
[cur_port
];
2070 if (exit_bytes_read
[cur_port
] > 0) {
2071 uint64_t num
= round_uint64_to_next_multiple_of(
2072 exit_bytes_read
[cur_port
],
2073 EXIT_STATS_ROUND_UP_BYTES
);
2076 tor_asprintf(&buf
, "%d="U64_FORMAT
, cur_port
, U64_PRINTF_ARG(num
));
2077 smartlist_add(read_strings
, buf
);
2078 other_read
-= exit_bytes_read
[cur_port
];
2080 if (exit_streams
[cur_port
] > 0) {
2081 uint32_t num
= round_uint32_to_next_multiple_of(
2082 exit_streams
[cur_port
],
2083 EXIT_STATS_ROUND_UP_STREAMS
);
2085 tor_asprintf(&buf
, "%d=%u", cur_port
, num
);
2086 smartlist_add(streams_strings
, buf
);
2087 other_streams
-= exit_streams
[cur_port
];
2091 /* Add observations of other ports in a single element. */
2092 other_written
= round_uint64_to_next_multiple_of(other_written
,
2093 EXIT_STATS_ROUND_UP_BYTES
);
2094 other_written
/= 1024;
2096 tor_asprintf(&buf
, "other="U64_FORMAT
, U64_PRINTF_ARG(other_written
));
2097 smartlist_add(written_strings
, buf
);
2098 other_read
= round_uint64_to_next_multiple_of(other_read
,
2099 EXIT_STATS_ROUND_UP_BYTES
);
2102 tor_asprintf(&buf
, "other="U64_FORMAT
, U64_PRINTF_ARG(other_read
));
2103 smartlist_add(read_strings
, buf
);
2104 other_streams
= round_uint32_to_next_multiple_of(other_streams
,
2105 EXIT_STATS_ROUND_UP_STREAMS
);
2107 tor_asprintf(&buf
, "other=%u", other_streams
);
2108 smartlist_add(streams_strings
, buf
);
2110 /* Join all observations in single strings. */
2111 written_string
= smartlist_join_strings(written_strings
, ",", 0, NULL
);
2112 read_string
= smartlist_join_strings(read_strings
, ",", 0, NULL
);
2113 streams_string
= smartlist_join_strings(streams_strings
, ",", 0, NULL
);
2114 SMARTLIST_FOREACH(written_strings
, char *, cp
, tor_free(cp
));
2115 SMARTLIST_FOREACH(read_strings
, char *, cp
, tor_free(cp
));
2116 SMARTLIST_FOREACH(streams_strings
, char *, cp
, tor_free(cp
));
2117 smartlist_free(written_strings
);
2118 smartlist_free(read_strings
);
2119 smartlist_free(streams_strings
);
2121 /* Put everything together. */
2122 format_iso_time(t
, now
);
2123 tor_asprintf(&result
, "exit-stats-end %s (%d s)\n"
2124 "exit-kibibytes-written %s\n"
2125 "exit-kibibytes-read %s\n"
2126 "exit-streams-opened %s\n",
2127 t
, (unsigned) (now
- start_of_exit_stats_interval
),
2131 tor_free(written_string
);
2132 tor_free(read_string
);
2133 tor_free(streams_string
);
2137 /** If 24 hours have passed since the beginning of the current exit port
2138 * stats period, write exit stats to $DATADIR/stats/exit-stats (possibly
2139 * overwriting an existing file) and reset counters. Return when we would
2140 * next want to write exit stats or 0 if we never want to write. */
2142 rep_hist_exit_stats_write(time_t now
)
2144 char *statsdir
= NULL
, *filename
= NULL
, *str
= NULL
;
2146 if (!start_of_exit_stats_interval
)
2147 return 0; /* Not initialized. */
2148 if (start_of_exit_stats_interval
+ WRITE_STATS_INTERVAL
> now
)
2149 goto done
; /* Not ready to write. */
2151 log_info(LD_HIST
, "Writing exit port statistics to disk.");
2153 /* Generate history string. */
2154 str
= rep_hist_format_exit_stats(now
);
2156 /* Reset counters. */
2157 rep_hist_reset_exit_stats(now
);
2159 /* Try to write to disk. */
2160 statsdir
= get_datadir_fname("stats");
2161 if (check_private_dir(statsdir
, CPD_CREATE
) < 0) {
2162 log_warn(LD_HIST
, "Unable to create stats/ directory!");
2165 filename
= get_datadir_fname2("stats", "exit-stats");
2166 if (write_str_to_file(filename
, str
, 0) < 0)
2167 log_warn(LD_HIST
, "Unable to write exit port statistics to disk!");
2173 return start_of_exit_stats_interval
+ WRITE_STATS_INTERVAL
;
2176 /** Note that we wrote <b>num_written</b> bytes and read <b>num_read</b>
2177 * bytes to/from an exit connection to <b>port</b>. */
2179 rep_hist_note_exit_bytes(uint16_t port
, size_t num_written
,
2182 if (!start_of_exit_stats_interval
)
2183 return; /* Not initialized. */
2184 exit_bytes_written
[port
] += num_written
;
2185 exit_bytes_read
[port
] += num_read
;
2186 log_debug(LD_HIST
, "Written %lu bytes and read %lu bytes to/from an "
2187 "exit connection to port %d.",
2188 (unsigned long)num_written
, (unsigned long)num_read
, port
);
2191 /** Note that we opened an exit stream to <b>port</b>. */
2193 rep_hist_note_exit_stream_opened(uint16_t port
)
2195 if (!start_of_exit_stats_interval
)
2196 return; /* Not initialized. */
2197 exit_streams
[port
]++;
2198 log_debug(LD_HIST
, "Opened exit stream to port %d", port
);
2201 /*** cell statistics ***/
2203 /** Start of the current buffer stats interval or 0 if we're not
2204 * collecting buffer statistics. */
2205 static time_t start_of_buffer_stats_interval
;
2207 /** Initialize buffer stats. */
2209 rep_hist_buffer_stats_init(time_t now
)
2211 start_of_buffer_stats_interval
= now
;
2214 typedef struct circ_buffer_stats_t
{
2215 uint32_t processed_cells
;
2216 double mean_num_cells_in_queue
;
2217 double mean_time_cells_in_queue
;
2218 uint32_t local_circ_id
;
2219 } circ_buffer_stats_t
;
2222 smartlist_t
*circuits_for_buffer_stats
= NULL
;
2224 /** Remember cell statistics for circuit <b>circ</b> at time
2225 * <b>end_of_interval</b> and reset cell counters in case the circuit
2226 * remains open in the next measurement interval. */
2228 rep_hist_buffer_stats_add_circ(circuit_t
*circ
, time_t end_of_interval
)
2230 circ_buffer_stats_t
*stat
;
2231 time_t start_of_interval
;
2232 int interval_length
;
2233 or_circuit_t
*orcirc
;
2234 if (CIRCUIT_IS_ORIGIN(circ
))
2236 orcirc
= TO_OR_CIRCUIT(circ
);
2237 if (!orcirc
->processed_cells
)
2239 if (!circuits_for_buffer_stats
)
2240 circuits_for_buffer_stats
= smartlist_create();
2241 start_of_interval
= circ
->timestamp_created
>
2242 start_of_buffer_stats_interval
?
2243 circ
->timestamp_created
:
2244 start_of_buffer_stats_interval
;
2245 interval_length
= (int) (end_of_interval
- start_of_interval
);
2246 stat
= tor_malloc_zero(sizeof(circ_buffer_stats_t
));
2247 stat
->processed_cells
= orcirc
->processed_cells
;
2248 /* 1000.0 for s -> ms; 2.0 because of app-ward and exit-ward queues */
2249 stat
->mean_num_cells_in_queue
= interval_length
== 0 ? 0.0 :
2250 (double) orcirc
->total_cell_waiting_time
/
2251 (double) interval_length
/ 1000.0 / 2.0;
2252 stat
->mean_time_cells_in_queue
=
2253 (double) orcirc
->total_cell_waiting_time
/
2254 (double) orcirc
->processed_cells
;
2255 smartlist_add(circuits_for_buffer_stats
, stat
);
2256 orcirc
->total_cell_waiting_time
= 0;
2257 orcirc
->processed_cells
= 0;
2260 /** Sorting helper: return -1, 1, or 0 based on comparison of two
2261 * circ_buffer_stats_t */
2263 _buffer_stats_compare_entries(const void **_a
, const void **_b
)
2265 const circ_buffer_stats_t
*a
= *_a
, *b
= *_b
;
2266 if (a
->processed_cells
< b
->processed_cells
)
2268 else if (a
->processed_cells
> b
->processed_cells
)
2274 /** Stop collecting cell stats in a way that we can re-start doing so in
2275 * rep_hist_buffer_stats_init(). */
2277 rep_hist_buffer_stats_term(void)
2279 start_of_buffer_stats_interval
= 0;
2280 if (!circuits_for_buffer_stats
)
2281 circuits_for_buffer_stats
= smartlist_create();
2282 SMARTLIST_FOREACH(circuits_for_buffer_stats
, circ_buffer_stats_t
*,
2283 stat
, tor_free(stat
));
2284 smartlist_clear(circuits_for_buffer_stats
);
2287 /** Write buffer statistics to $DATADIR/stats/buffer-stats and return when
2288 * we would next want to write exit stats. */
2290 rep_hist_buffer_stats_write(time_t now
)
2292 char *statsdir
= NULL
, *filename
= NULL
;
2293 char written
[ISO_TIME_LEN
+1];
2294 open_file_t
*open_file
= NULL
;
2297 int processed_cells
[SHARES
], circs_in_share
[SHARES
],
2298 number_of_circuits
, i
;
2299 double queued_cells
[SHARES
], time_in_queue
[SHARES
];
2300 smartlist_t
*str_build
= smartlist_create();
2301 char *str
= NULL
, *buf
=NULL
;
2304 if (!start_of_buffer_stats_interval
)
2305 return 0; /* Not initialized. */
2306 if (start_of_buffer_stats_interval
+ WRITE_STATS_INTERVAL
> now
)
2307 goto done
; /* Not ready to write */
2309 /* add current circuits to stats */
2310 for (circ
= _circuit_get_global_list(); circ
; circ
= circ
->next
)
2311 rep_hist_buffer_stats_add_circ(circ
, now
);
2312 /* calculate deciles */
2313 memset(processed_cells
, 0, SHARES
* sizeof(int));
2314 memset(circs_in_share
, 0, SHARES
* sizeof(int));
2315 memset(queued_cells
, 0, SHARES
* sizeof(double));
2316 memset(time_in_queue
, 0, SHARES
* sizeof(double));
2317 if (!circuits_for_buffer_stats
)
2318 circuits_for_buffer_stats
= smartlist_create();
2319 smartlist_sort(circuits_for_buffer_stats
,
2320 _buffer_stats_compare_entries
);
2321 number_of_circuits
= smartlist_len(circuits_for_buffer_stats
);
2322 if (number_of_circuits
< 1) {
2323 log_info(LD_HIST
, "Attempt to write cell statistics to disk failed. "
2324 "We haven't seen a single circuit to report about.");
2328 SMARTLIST_FOREACH_BEGIN(circuits_for_buffer_stats
,
2329 circ_buffer_stats_t
*, stat
)
2331 int share
= i
++ * SHARES
/ number_of_circuits
;
2332 processed_cells
[share
] += stat
->processed_cells
;
2333 queued_cells
[share
] += stat
->mean_num_cells_in_queue
;
2334 time_in_queue
[share
] += stat
->mean_time_cells_in_queue
;
2335 circs_in_share
[share
]++;
2337 SMARTLIST_FOREACH_END(stat
);
2338 /* clear buffer stats history */
2339 SMARTLIST_FOREACH(circuits_for_buffer_stats
, circ_buffer_stats_t
*,
2340 stat
, tor_free(stat
));
2341 smartlist_clear(circuits_for_buffer_stats
);
2343 statsdir
= get_datadir_fname("stats");
2344 if (check_private_dir(statsdir
, CPD_CREATE
) < 0)
2346 filename
= get_datadir_fname2("stats", "buffer-stats");
2347 out
= start_writing_to_stdio_file(filename
, OPEN_FLAGS_APPEND
,
2351 format_iso_time(written
, now
);
2352 if (fprintf(out
, "cell-stats-end %s (%d s)\n", written
,
2353 (unsigned) (now
- start_of_buffer_stats_interval
)) < 0)
2355 for (i
= 0; i
< SHARES
; i
++) {
2356 tor_asprintf(&buf
,"%d", !circs_in_share
[i
] ? 0 :
2357 processed_cells
[i
] / circs_in_share
[i
]);
2358 smartlist_add(str_build
, buf
);
2360 str
= smartlist_join_strings(str_build
, ",", 0, NULL
);
2361 if (fprintf(out
, "cell-processed-cells %s\n", str
) < 0)
2364 SMARTLIST_FOREACH(str_build
, char *, c
, tor_free(c
));
2365 smartlist_clear(str_build
);
2366 for (i
= 0; i
< SHARES
; i
++) {
2367 tor_asprintf(&buf
, "%.2f", circs_in_share
[i
] == 0 ? 0.0 :
2368 queued_cells
[i
] / (double) circs_in_share
[i
]);
2369 smartlist_add(str_build
, buf
);
2371 str
= smartlist_join_strings(str_build
, ",", 0, NULL
);
2372 if (fprintf(out
, "cell-queued-cells %s\n", str
) < 0)
2375 SMARTLIST_FOREACH(str_build
, char *, c
, tor_free(c
));
2376 smartlist_clear(str_build
);
2377 for (i
= 0; i
< SHARES
; i
++) {
2378 tor_asprintf(&buf
, "%.0f", circs_in_share
[i
] == 0 ? 0.0 :
2379 time_in_queue
[i
] / (double) circs_in_share
[i
]);
2380 smartlist_add(str_build
, buf
);
2382 str
= smartlist_join_strings(str_build
, ",", 0, NULL
);
2383 if (fprintf(out
, "cell-time-in-queue %s\n", str
) < 0)
2386 SMARTLIST_FOREACH(str_build
, char *, c
, tor_free(c
));
2387 smartlist_free(str_build
);
2389 if (fprintf(out
, "cell-circuits-per-decile %d\n",
2390 (number_of_circuits
+ SHARES
- 1) / SHARES
) < 0)
2392 finish_writing_to_file(open_file
);
2394 start_of_buffer_stats_interval
= now
;
2397 abort_writing_to_file(open_file
);
2401 SMARTLIST_FOREACH(str_build
, char *, c
, tor_free(c
));
2402 smartlist_free(str_build
);
2406 return start_of_buffer_stats_interval
+ WRITE_STATS_INTERVAL
;
2409 /** Free all storage held by the OR/link history caches, by the
2410 * bandwidth history arrays, by the port history, or by statistics . */
2412 rep_hist_free_all(void)
2414 digestmap_free(history_map
, free_or_history
);
2415 tor_free(read_array
);
2416 tor_free(write_array
);
2417 tor_free(last_stability_doc
);
2418 tor_free(exit_bytes_read
);
2419 tor_free(exit_bytes_written
);
2420 tor_free(exit_streams
);
2421 built_last_stability_doc_at
= 0;
2422 predicted_ports_free();