rephist.c 67 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888889890891892893894895896897898899900901902903904905906907908909910911912913914915916917918919920921922923924925926927928929930931932933934935936937938939940941942943944945946947948949950951952953954955956957958959960961962963964965966967968969970971972973974975976977978979980981982983984985986987988989990991992993994995996997998999100010011002100310041005100610071008100910101011101210131014101510161017101810191020102110221023102410251026102710281029103010311032103310341035103610371038103910401041104210431044104510461047104810491050105110521053105410551056105710581059106010611062106310641065106610671068106910701071107210731074107510761077107810791080108110821083108410851086108710881089109010911092109310941095109610971098109911001101110211031104110511061107110811091110111111121113111411151116111711181119112011211122112311241125112611271128112911301131113211331134113511361137113811391140114111421143114411451146114711481149115011511152115311541155115611571158115911601161116211631164116511661167116811691170117111721173117411751176117711781179118011811182118311841185118611871188118911901191119211931194119511961197119811991200120112021203120412051206120712081209121012111212121312141215121612171218121912201221122212231224122512261227122812291230123112321233123412351236123712381239124012411242124312441245124612471248124912501251125212531254125512561257125812591260126112621263126412651266126712681269127012711272127312741275127612771278127912801281128212831284128512861287128812891290129112921293129412951296129712981299130013011302130313041305130613071308130913101311131213131314131513161317131813191320132113221323132413251326132713281329133013311332133313341335133613371338133913401341134213431344134513461347134813491350135113521353135413551356135713581359136013611362136313641365136613671368136913701371137213731374137513761377137813791380138113821383138413851386138713881389139013911392139313941395139613971398139914001401140214031404140514061407140814091410141114121413141414151416141714181419142014211422142314241425142614271428142914301431143214331434143514361437143814391440144114421443144414451446144714481449145014511452145314541455145614571458145914601461146214631464146514661467146814691470147114721473147414751476147714781479148014811482148314841485148614871488148914901491149214931494149514961497149814991500150115021503150415051506150715081509151015111512151315141515151615171518151915201521152215231524152515261527152815291530153115321533153415351536153715381539154015411542154315441545154615471548154915501551155215531554155515561557155815591560156115621563156415651566156715681569157015711572157315741575157615771578157915801581158215831584158515861587158815891590159115921593159415951596159715981599160016011602160316041605160616071608160916101611161216131614161516161617161816191620162116221623162416251626162716281629163016311632163316341635163616371638163916401641164216431644164516461647164816491650165116521653165416551656165716581659166016611662166316641665166616671668166916701671167216731674167516761677167816791680168116821683168416851686168716881689169016911692169316941695169616971698169917001701170217031704170517061707170817091710171117121713171417151716171717181719172017211722172317241725172617271728172917301731173217331734173517361737173817391740174117421743174417451746174717481749175017511752175317541755175617571758175917601761176217631764176517661767176817691770177117721773177417751776177717781779178017811782178317841785178617871788178917901791179217931794179517961797179817991800180118021803180418051806180718081809181018111812181318141815181618171818181918201821182218231824182518261827182818291830183118321833183418351836183718381839184018411842184318441845184618471848184918501851185218531854185518561857185818591860186118621863186418651866186718681869187018711872187318741875187618771878187918801881188218831884188518861887188818891890189118921893189418951896189718981899190019011902190319041905190619071908190919101911191219131914191519161917191819191920192119221923192419251926192719281929193019311932193319341935193619371938193919401941194219431944194519461947194819491950195119521953195419551956195719581959196019611962196319641965196619671968196919701971197219731974197519761977197819791980198119821983198419851986198719881989199019911992199319941995199619971998199920002001200220032004200520062007200820092010201120122013201420152016201720182019202020212022202320242025202620272028202920302031203220332034203520362037203820392040204120422043204420452046204720482049205020512052205320542055205620572058205920602061206220632064206520662067206820692070207120722073207420752076207720782079208020812082208320842085208620872088208920902091209220932094209520962097209820992100210121022103210421052106210721082109211021112112211321142115211621172118211921202121212221232124212521262127212821292130213121322133213421352136213721382139214021412142214321442145214621472148214921502151215221532154215521562157215821592160216121622163216421652166216721682169217021712172217321742175217621772178217921802181218221832184218521862187
  1. /* Copyright (c) 2004-2006, Roger Dingledine, Nick Mathewson.
  2. * Copyright (c) 2007-2008, The Tor Project, Inc. */
  3. /* See LICENSE for licensing information */
  4. /* $Id$ */
  5. const char rephist_c_id[] =
  6. "$Id$";
  7. /**
  8. * \file rephist.c
  9. * \brief Basic history and "reputation" functionality to remember
  10. * which servers have worked in the past, how much bandwidth we've
  11. * been using, which ports we tend to want, and so on.
  12. **/
  13. #include "or.h"
  14. #include "ht.h"
  15. static void bw_arrays_init(void);
  16. static void predicted_ports_init(void);
  17. static void hs_usage_init(void);
  18. /** Total number of bytes currently allocated in fields used by rephist.c. */
  19. uint64_t rephist_total_alloc=0;
  20. /** Number of or_history_t objects currently allocated. */
  21. uint32_t rephist_total_num=0;
  22. /** If the total weighted run count of all runs for a router ever falls
  23. * below this amount, the router can be treated as having 0 MTBF. */
  24. #define STABILITY_EPSILON 0.0001
  25. /** Value by which to discount all old intervals for MTBF purposses. This
  26. * is compounded every STABILITY_INTERVAL. */
  27. #define STABILITY_ALPHA 0.95
  28. /** Interval at which to discount all old intervals for MTBF purposes. */
  29. #define STABILITY_INTERVAL (12*60*60)
  30. /* (This combination of ALPHA, INTERVAL, and EPSILON makes it so that an
  31. * interval that just ended counts twice as much as one that ended a week ago,
  32. * 20X as much as one that ended a month ago, and routers that have had no
  33. * uptime data for about half a year will get forgotten.) */
  34. /** History of an OR-\>OR link. */
  35. typedef struct link_history_t {
  36. /** When did we start tracking this list? */
  37. time_t since;
  38. /** When did we most recently note a change to this link */
  39. time_t changed;
  40. /** How many times did extending from OR1 to OR2 succeed? */
  41. unsigned long n_extend_ok;
  42. /** How many times did extending from OR1 to OR2 fail? */
  43. unsigned long n_extend_fail;
  44. } link_history_t;
  45. /** History of an OR. */
  46. typedef struct or_history_t {
  47. /** When did we start tracking this OR? */
  48. time_t since;
  49. /** When did we most recently note a change to this OR? */
  50. time_t changed;
  51. /** How many times did we successfully connect? */
  52. unsigned long n_conn_ok;
  53. /** How many times did we try to connect and fail?*/
  54. unsigned long n_conn_fail;
  55. /** How many seconds have we been connected to this OR before
  56. * 'up_since'? */
  57. unsigned long uptime;
  58. /** How many seconds have we been unable to connect to this OR before
  59. * 'down_since'? */
  60. unsigned long downtime;
  61. /** If nonzero, we have been connected since this time. */
  62. time_t up_since;
  63. /** If nonzero, we have been unable to connect since this time. */
  64. time_t down_since;
  65. /* === For MTBF tracking: */
  66. /** Weighted sum total of all times that this router has been online.
  67. */
  68. unsigned long weighted_run_length;
  69. /** If the router is now online (according to stability-checking rules),
  70. * when did it come online? */
  71. time_t start_of_run;
  72. /** Sum of weights for runs in weighted_run_length. */
  73. double total_run_weights;
  74. /* === For fractional uptime tracking: */
  75. time_t start_of_downtime;
  76. unsigned long weighted_uptime;
  77. unsigned long total_weighted_time;
  78. /** Map from hex OR2 identity digest to a link_history_t for the link
  79. * from this OR to OR2. */
  80. digestmap_t *link_history_map;
  81. } or_history_t;
  82. /** When did we last multiply all routers' weighted_run_length and
  83. * total_run_weights by STABILITY_ALPHA? */
  84. static time_t stability_last_downrated = 0;
  85. /** */
  86. static time_t started_tracking_stability = 0;
  87. /** Map from hex OR identity digest to or_history_t. */
  88. static digestmap_t *history_map = NULL;
  89. /** Return the or_history_t for the OR with identity digest <b>id</b>,
  90. * creating it if necessary. */
  91. static or_history_t *
  92. get_or_history(const char* id)
  93. {
  94. or_history_t *hist;
  95. if (tor_mem_is_zero(id, DIGEST_LEN))
  96. return NULL;
  97. hist = digestmap_get(history_map, id);
  98. if (!hist) {
  99. hist = tor_malloc_zero(sizeof(or_history_t));
  100. rephist_total_alloc += sizeof(or_history_t);
  101. rephist_total_num++;
  102. hist->link_history_map = digestmap_new();
  103. hist->since = hist->changed = time(NULL);
  104. digestmap_set(history_map, id, hist);
  105. }
  106. return hist;
  107. }
  108. /** Return the link_history_t for the link from the first named OR to
  109. * the second, creating it if necessary. (ORs are identified by
  110. * identity digest.)
  111. */
  112. static link_history_t *
  113. get_link_history(const char *from_id, const char *to_id)
  114. {
  115. or_history_t *orhist;
  116. link_history_t *lhist;
  117. orhist = get_or_history(from_id);
  118. if (!orhist)
  119. return NULL;
  120. if (tor_mem_is_zero(to_id, DIGEST_LEN))
  121. return NULL;
  122. lhist = (link_history_t*) digestmap_get(orhist->link_history_map, to_id);
  123. if (!lhist) {
  124. lhist = tor_malloc_zero(sizeof(link_history_t));
  125. rephist_total_alloc += sizeof(link_history_t);
  126. lhist->since = lhist->changed = time(NULL);
  127. digestmap_set(orhist->link_history_map, to_id, lhist);
  128. }
  129. return lhist;
  130. }
  131. /** Helper: free storage held by a single link history entry. */
  132. static void
  133. _free_link_history(void *val)
  134. {
  135. rephist_total_alloc -= sizeof(link_history_t);
  136. tor_free(val);
  137. }
  138. /** Helper: free storage held by a single OR history entry. */
  139. static void
  140. free_or_history(void *_hist)
  141. {
  142. or_history_t *hist = _hist;
  143. digestmap_free(hist->link_history_map, _free_link_history);
  144. rephist_total_alloc -= sizeof(or_history_t);
  145. rephist_total_num--;
  146. tor_free(hist);
  147. }
  148. /** Update an or_history_t object <b>hist</b> so that its uptime/downtime
  149. * count is up-to-date as of <b>when</b>.
  150. */
  151. static void
  152. update_or_history(or_history_t *hist, time_t when)
  153. {
  154. tor_assert(hist);
  155. if (hist->up_since) {
  156. tor_assert(!hist->down_since);
  157. hist->uptime += (when - hist->up_since);
  158. hist->up_since = when;
  159. } else if (hist->down_since) {
  160. hist->downtime += (when - hist->down_since);
  161. hist->down_since = when;
  162. }
  163. }
  164. /** Initialize the static data structures for tracking history. */
  165. void
  166. rep_hist_init(void)
  167. {
  168. history_map = digestmap_new();
  169. bw_arrays_init();
  170. predicted_ports_init();
  171. hs_usage_init();
  172. }
  173. /** Helper: note that we are no longer connected to the router with history
  174. * <b>hist</b>. If <b>failed</b>, the connection failed; otherwise, it was
  175. * closed correctly. */
  176. static void
  177. mark_or_down(or_history_t *hist, time_t when, int failed)
  178. {
  179. if (hist->up_since) {
  180. hist->uptime += (when - hist->up_since);
  181. hist->up_since = 0;
  182. }
  183. if (failed && !hist->down_since) {
  184. hist->down_since = when;
  185. }
  186. }
  187. /** Helper: note that we are connected to the router with history
  188. * <b>hist</b>. */
  189. static void
  190. mark_or_up(or_history_t *hist, time_t when)
  191. {
  192. if (hist->down_since) {
  193. hist->downtime += (when - hist->down_since);
  194. hist->down_since = 0;
  195. }
  196. if (!hist->up_since) {
  197. hist->up_since = when;
  198. }
  199. }
  200. /** Remember that an attempt to connect to the OR with identity digest
  201. * <b>id</b> failed at <b>when</b>.
  202. */
  203. void
  204. rep_hist_note_connect_failed(const char* id, time_t when)
  205. {
  206. or_history_t *hist;
  207. hist = get_or_history(id);
  208. if (!hist)
  209. return;
  210. ++hist->n_conn_fail;
  211. mark_or_down(hist, when, 1);
  212. hist->changed = when;
  213. }
  214. /** Remember that an attempt to connect to the OR with identity digest
  215. * <b>id</b> succeeded at <b>when</b>.
  216. */
  217. void
  218. rep_hist_note_connect_succeeded(const char* id, time_t when)
  219. {
  220. or_history_t *hist;
  221. hist = get_or_history(id);
  222. if (!hist)
  223. return;
  224. ++hist->n_conn_ok;
  225. mark_or_up(hist, when);
  226. hist->changed = when;
  227. }
  228. /** Remember that we intentionally closed our connection to the OR
  229. * with identity digest <b>id</b> at <b>when</b>.
  230. */
  231. void
  232. rep_hist_note_disconnect(const char* id, time_t when)
  233. {
  234. or_history_t *hist;
  235. hist = get_or_history(id);
  236. if (!hist)
  237. return;
  238. mark_or_down(hist, when, 0);
  239. hist->changed = when;
  240. }
  241. /** Remember that our connection to the OR with identity digest
  242. * <b>id</b> had an error and stopped working at <b>when</b>.
  243. */
  244. void
  245. rep_hist_note_connection_died(const char* id, time_t when)
  246. {
  247. or_history_t *hist;
  248. if (!id) {
  249. /* If conn has no identity, it didn't complete its handshake, or something
  250. * went wrong. Ignore it.
  251. */
  252. return;
  253. }
  254. hist = get_or_history(id);
  255. if (!hist)
  256. return;
  257. mark_or_down(hist, when, 1);
  258. hist->changed = when;
  259. }
  260. /** We have just decided that this router is reachable, meaning
  261. * we will give it a "Running" flag for the next while. */
  262. void
  263. rep_hist_note_router_reachable(const char *id, time_t when)
  264. {
  265. or_history_t *hist = get_or_history(id);
  266. if (!started_tracking_stability)
  267. started_tracking_stability = time(NULL);
  268. if (hist && !hist->start_of_run) {
  269. hist->start_of_run = when;
  270. }
  271. if (hist && hist->start_of_downtime) {
  272. long down_length = when - hist->start_of_downtime;
  273. hist->total_weighted_time += down_length;
  274. hist->start_of_downtime = 0;
  275. }
  276. }
  277. /** We have just decided that this router is unreachable, meaning
  278. * we are taking away its "Running" flag. */
  279. void
  280. rep_hist_note_router_unreachable(const char *id, time_t when)
  281. {
  282. or_history_t *hist = get_or_history(id);
  283. if (!started_tracking_stability)
  284. started_tracking_stability = time(NULL);
  285. if (hist && hist->start_of_run) {
  286. /*XXXX We could treat failed connections differently from failed
  287. * conect attempts. */
  288. long run_length = when - hist->start_of_run;
  289. hist->weighted_run_length += run_length;
  290. hist->total_run_weights += 1.0;
  291. hist->start_of_run = 0;
  292. hist->weighted_uptime += run_length;
  293. hist->total_weighted_time += run_length;
  294. }
  295. if (hist && !hist->start_of_downtime) {
  296. hist->start_of_downtime = when;
  297. }
  298. }
  299. /** Helper: Discount all old MTBF data, if it is time to do so. Return
  300. * the time at which we should next discount MTBF data. */
  301. time_t
  302. rep_hist_downrate_old_runs(time_t now)
  303. {
  304. digestmap_iter_t *orhist_it;
  305. const char *digest1;
  306. or_history_t *hist;
  307. void *hist_p;
  308. double alpha = 1.0;
  309. if (!history_map)
  310. history_map = digestmap_new();
  311. if (!stability_last_downrated)
  312. stability_last_downrated = now;
  313. if (stability_last_downrated + STABILITY_INTERVAL > now)
  314. return stability_last_downrated + STABILITY_INTERVAL;
  315. /* Okay, we should downrate the data. By how much? */
  316. while (stability_last_downrated + STABILITY_INTERVAL < now) {
  317. stability_last_downrated += STABILITY_INTERVAL;
  318. alpha *= STABILITY_ALPHA;
  319. }
  320. /* Multiply every w_r_l, t_r_w pair by alpha. */
  321. for (orhist_it = digestmap_iter_init(history_map);
  322. !digestmap_iter_done(orhist_it);
  323. orhist_it = digestmap_iter_next(history_map,orhist_it)) {
  324. digestmap_iter_get(orhist_it, &digest1, &hist_p);
  325. hist = hist_p;
  326. hist->weighted_run_length =
  327. (unsigned long)(hist->weighted_run_length * alpha);
  328. hist->total_run_weights *= alpha;
  329. hist->weighted_uptime *= alpha;
  330. hist->total_weighted_time *= alpha;
  331. }
  332. return stability_last_downrated + STABILITY_INTERVAL;
  333. }
  334. /** Helper: Return the weighted MTBF of the router with history <b>hist</b>. */
  335. static double
  336. get_stability(or_history_t *hist, time_t when)
  337. {
  338. unsigned long total = hist->weighted_run_length;
  339. double total_weights = hist->total_run_weights;
  340. if (hist->start_of_run) {
  341. /* We're currently in a run. Let total and total_weights hold the values
  342. * they would hold if the current run were to end now. */
  343. total += (when-hist->start_of_run);
  344. total_weights += 1.0;
  345. }
  346. if (total_weights < STABILITY_EPSILON) {
  347. /* Round down to zero, and avoid divide-by-zero. */
  348. return 0.0;
  349. }
  350. return total / total_weights;
  351. }
  352. /** Return the total amount of time we've been observing, with each run of
  353. * time downrated by the appropriate factor. */
  354. static long
  355. get_total_weighted_time(or_history_t *hist, time_t when)
  356. {
  357. long total = hist->total_weighted_time;
  358. if (hist->start_of_run) {
  359. total += (when - hist->start_of_run);
  360. } else if (hist->start_of_downtime) {
  361. total += (when - hist->start_of_downtime);
  362. }
  363. return total;
  364. }
  365. /** Helper: Return the weighted percent-of-time-online of the router with
  366. * history <b>hist</b>. */
  367. static double
  368. get_weighted_fractional_uptime(or_history_t *hist, time_t when)
  369. {
  370. unsigned long total = hist->total_weighted_time;
  371. unsigned long up = hist->weighted_uptime;
  372. if (hist->start_of_run) {
  373. long run_length = (when - hist->start_of_run);
  374. up += run_length;
  375. total += run_length;
  376. } else if (hist->start_of_downtime) {
  377. total += (when - hist->start_of_downtime);
  378. }
  379. return ((double) up) / total;
  380. }
  381. /** Return an estimated MTBF for the router whose identity digest is
  382. * <b>id</b>. Return 0 if the router is unknown. */
  383. double
  384. rep_hist_get_stability(const char *id, time_t when)
  385. {
  386. or_history_t *hist = get_or_history(id);
  387. if (!hist)
  388. return 0.0;
  389. return get_stability(hist, when);
  390. }
  391. /** Return an estimated percent-of-time-online for the router whose identity
  392. * digest is <b>id</b>. Return 0 if the router is unknown. */
  393. double
  394. rep_hist_get_weighted_fractional_uptime(const char *id, time_t when)
  395. {
  396. or_history_t *hist = get_or_history(id);
  397. if (!hist)
  398. return 0.0;
  399. return get_weighted_fractional_uptime(hist, when);
  400. }
  401. /** Return a number representing how long we've known about the router whose
  402. * digest is <b>id</b>. Return 0 if the router is unknown.
  403. *
  404. * Be careful: this measure incresases monotonically as we know the router for
  405. * longer and longer, but it doesn't increase linearly.
  406. */
  407. long
  408. rep_hist_get_weighted_time_known(const char *id, time_t when)
  409. {
  410. or_history_t *hist = get_or_history(id);
  411. if (!hist)
  412. return 0;
  413. return get_total_weighted_time(hist, when);
  414. }
  415. /** Return true if we've been measuring MTBFs for long enough to
  416. * pronounce on Stability. */
  417. int
  418. rep_hist_have_measured_enough_stability(void)
  419. {
  420. /* XXXX021 This doesn't do so well when we change our opinion
  421. * as to whether we're tracking router stability. */
  422. return started_tracking_stability < time(NULL) - 4*60*60;
  423. }
  424. /** Remember that we successfully extended from the OR with identity
  425. * digest <b>from_id</b> to the OR with identity digest
  426. * <b>to_name</b>.
  427. */
  428. void
  429. rep_hist_note_extend_succeeded(const char *from_id, const char *to_id)
  430. {
  431. link_history_t *hist;
  432. /* log_fn(LOG_WARN, "EXTEND SUCCEEDED: %s->%s",from_name,to_name); */
  433. hist = get_link_history(from_id, to_id);
  434. if (!hist)
  435. return;
  436. ++hist->n_extend_ok;
  437. hist->changed = time(NULL);
  438. }
  439. /** Remember that we tried to extend from the OR with identity digest
  440. * <b>from_id</b> to the OR with identity digest <b>to_name</b>, but
  441. * failed.
  442. */
  443. void
  444. rep_hist_note_extend_failed(const char *from_id, const char *to_id)
  445. {
  446. link_history_t *hist;
  447. /* log_fn(LOG_WARN, "EXTEND FAILED: %s->%s",from_name,to_name); */
  448. hist = get_link_history(from_id, to_id);
  449. if (!hist)
  450. return;
  451. ++hist->n_extend_fail;
  452. hist->changed = time(NULL);
  453. }
  454. /** Log all the reliability data we have remembered, with the chosen
  455. * severity.
  456. */
  457. void
  458. rep_hist_dump_stats(time_t now, int severity)
  459. {
  460. digestmap_iter_t *lhist_it;
  461. digestmap_iter_t *orhist_it;
  462. const char *name1, *name2, *digest1, *digest2;
  463. char hexdigest1[HEX_DIGEST_LEN+1];
  464. or_history_t *or_history;
  465. link_history_t *link_history;
  466. void *or_history_p, *link_history_p;
  467. double uptime;
  468. char buffer[2048];
  469. size_t len;
  470. int ret;
  471. unsigned long upt, downt;
  472. routerinfo_t *r;
  473. rep_history_clean(now - get_options()->RephistTrackTime);
  474. log(severity, LD_GENERAL, "--------------- Dumping history information:");
  475. for (orhist_it = digestmap_iter_init(history_map);
  476. !digestmap_iter_done(orhist_it);
  477. orhist_it = digestmap_iter_next(history_map,orhist_it)) {
  478. double s;
  479. long stability;
  480. digestmap_iter_get(orhist_it, &digest1, &or_history_p);
  481. or_history = (or_history_t*) or_history_p;
  482. if ((r = router_get_by_digest(digest1)))
  483. name1 = r->nickname;
  484. else
  485. name1 = "(unknown)";
  486. base16_encode(hexdigest1, sizeof(hexdigest1), digest1, DIGEST_LEN);
  487. update_or_history(or_history, now);
  488. upt = or_history->uptime;
  489. downt = or_history->downtime;
  490. s = get_stability(or_history, now);
  491. stability = (long)s;
  492. if (upt+downt) {
  493. uptime = ((double)upt) / (upt+downt);
  494. } else {
  495. uptime=1.0;
  496. }
  497. log(severity, LD_GENERAL,
  498. "OR %s [%s]: %ld/%ld good connections; uptime %ld/%ld sec (%.2f%%); "
  499. "wmtbf %lu:%02lu:%02lu",
  500. name1, hexdigest1,
  501. or_history->n_conn_ok, or_history->n_conn_fail+or_history->n_conn_ok,
  502. upt, upt+downt, uptime*100.0,
  503. stability/3600, (stability/60)%60, stability%60);
  504. if (!digestmap_isempty(or_history->link_history_map)) {
  505. strlcpy(buffer, " Extend attempts: ", sizeof(buffer));
  506. len = strlen(buffer);
  507. for (lhist_it = digestmap_iter_init(or_history->link_history_map);
  508. !digestmap_iter_done(lhist_it);
  509. lhist_it = digestmap_iter_next(or_history->link_history_map,
  510. lhist_it)) {
  511. digestmap_iter_get(lhist_it, &digest2, &link_history_p);
  512. if ((r = router_get_by_digest(digest2)))
  513. name2 = r->nickname;
  514. else
  515. name2 = "(unknown)";
  516. link_history = (link_history_t*) link_history_p;
  517. ret = tor_snprintf(buffer+len, 2048-len, "%s(%ld/%ld); ", name2,
  518. link_history->n_extend_ok,
  519. link_history->n_extend_ok+link_history->n_extend_fail);
  520. if (ret<0)
  521. break;
  522. else
  523. len += ret;
  524. }
  525. log(severity, LD_GENERAL, "%s", buffer);
  526. }
  527. }
  528. }
  529. /** Remove history info for routers/links that haven't changed since
  530. * <b>before</b>.
  531. */
  532. void
  533. rep_history_clean(time_t before)
  534. {
  535. int authority = authdir_mode(get_options());
  536. or_history_t *or_history;
  537. link_history_t *link_history;
  538. void *or_history_p, *link_history_p;
  539. digestmap_iter_t *orhist_it, *lhist_it;
  540. const char *d1, *d2;
  541. orhist_it = digestmap_iter_init(history_map);
  542. while (!digestmap_iter_done(orhist_it)) {
  543. int remove;
  544. digestmap_iter_get(orhist_it, &d1, &or_history_p);
  545. or_history = or_history_p;
  546. remove = authority ? (or_history->total_run_weights < STABILITY_EPSILON &&
  547. !or_history->start_of_run)
  548. : (or_history->changed < before);
  549. if (remove) {
  550. orhist_it = digestmap_iter_next_rmv(history_map, orhist_it);
  551. free_or_history(or_history);
  552. continue;
  553. }
  554. for (lhist_it = digestmap_iter_init(or_history->link_history_map);
  555. !digestmap_iter_done(lhist_it); ) {
  556. digestmap_iter_get(lhist_it, &d2, &link_history_p);
  557. link_history = link_history_p;
  558. if (link_history->changed < before) {
  559. lhist_it = digestmap_iter_next_rmv(or_history->link_history_map,
  560. lhist_it);
  561. rephist_total_alloc -= sizeof(link_history_t);
  562. tor_free(link_history);
  563. continue;
  564. }
  565. lhist_it = digestmap_iter_next(or_history->link_history_map,lhist_it);
  566. }
  567. orhist_it = digestmap_iter_next(history_map, orhist_it);
  568. }
  569. }
  570. /** Write MTBF data to disk. Returns 0 on success, negative on failure. */
  571. int
  572. rep_hist_record_mtbf_data(void)
  573. {
  574. char time_buf[ISO_TIME_LEN+1];
  575. digestmap_iter_t *orhist_it;
  576. const char *digest;
  577. void *or_history_p;
  578. or_history_t *hist;
  579. open_file_t *open_file = NULL;
  580. FILE *f;
  581. {
  582. char *filename = get_datadir_fname("router-stability");
  583. f = start_writing_to_stdio_file(filename, OPEN_FLAGS_REPLACE|O_TEXT, 0600,
  584. &open_file);
  585. tor_free(filename);
  586. if (!f)
  587. return -1;
  588. }
  589. /* File format is:
  590. * FormatLine *KeywordLine Data
  591. *
  592. * FormatLine = "format 1" NL
  593. * KeywordLine = Keyword SP Arguments NL
  594. * Data = "data" NL *RouterMTBFLine "." NL
  595. * RouterMTBFLine = Fingerprint SP WeightedRunLen SP
  596. * TotalRunWeights [SP S=StartRunTime] NL
  597. */
  598. #define PUT(s) STMT_BEGIN if (fputs((s),f)<0) goto err; STMT_END
  599. #define PRINTF(args) STMT_BEGIN if (fprintf args <0) goto err; STMT_END
  600. PUT("format 2\n");
  601. format_iso_time(time_buf, time(NULL));
  602. PRINTF((f, "stored-at %s\n", time_buf));
  603. if (started_tracking_stability) {
  604. format_iso_time(time_buf, started_tracking_stability);
  605. PRINTF((f, "tracked-since %s\n", time_buf));
  606. }
  607. if (stability_last_downrated) {
  608. format_iso_time(time_buf, stability_last_downrated);
  609. PRINTF((f, "last-downrated %s\n", time_buf));
  610. }
  611. PUT("data\n");
  612. /* XXX021 Nick: now bridge auths record this for all routers too.
  613. * Should we make them record it only for bridge routers? -RD
  614. * Not for 0.2.0. -NM */
  615. for (orhist_it = digestmap_iter_init(history_map);
  616. !digestmap_iter_done(orhist_it);
  617. orhist_it = digestmap_iter_next(history_map,orhist_it)) {
  618. char dbuf[HEX_DIGEST_LEN+1];
  619. const char *t = NULL;
  620. digestmap_iter_get(orhist_it, &digest, &or_history_p);
  621. hist = (or_history_t*) or_history_p;
  622. base16_encode(dbuf, sizeof(dbuf), digest, DIGEST_LEN);
  623. PRINTF((f, "R %s\n", dbuf));
  624. if (hist->start_of_run > 0) {
  625. format_iso_time(time_buf, hist->start_of_run);
  626. t = time_buf;
  627. }
  628. PRINTF((f, "+MTBF %lu %.5lf%s%s\n",
  629. hist->weighted_run_length, hist->total_run_weights,
  630. t ? " S=" : "", t ? t : ""));
  631. t = NULL;
  632. if (hist->start_of_downtime > 0) {
  633. format_iso_time(time_buf, hist->start_of_downtime);
  634. t = time_buf;
  635. }
  636. PRINTF((f, "+WFU %lu %lu%s%s\n",
  637. hist->weighted_uptime, hist->total_weighted_time,
  638. t ? " S=" : "", t ? t : ""));
  639. }
  640. PUT(".\n");
  641. #undef PUT
  642. #undef PRINTF
  643. return finish_writing_to_file(open_file);
  644. err:
  645. abort_writing_to_file(open_file);
  646. return -1;
  647. }
  648. /** Helper: return the first j >= i such that !strcmpstart(sl[j], prefix) and
  649. * such that no line sl[k] with i <= k < j starts with "R ". Return -1 if no
  650. * such line exists. */
  651. static int
  652. find_next_with(smartlist_t *sl, int i, const char *prefix)
  653. {
  654. for ( ; i < smartlist_len(sl); ++i) {
  655. const char *line = smartlist_get(sl, i);
  656. if (!strcmpstart(line, prefix))
  657. return i;
  658. if (!strcmpstart(line, "R "))
  659. return -1;
  660. }
  661. return -1;
  662. }
  663. /** How many bad times has parse_possibly_bad_iso_time parsed? */
  664. static int n_bogus_times = 0;
  665. /** Parse the ISO-formatted time in <b>s</b> into *<b>time_out</b>, but
  666. * rounds any pre-1970 date to Jan 1, 1970. */
  667. static int
  668. parse_possibly_bad_iso_time(const char *s, time_t *time_out)
  669. {
  670. int year;
  671. char b[5];
  672. strlcpy(b, s, sizeof(b));
  673. b[4] = '\0';
  674. year = (int)tor_parse_long(b, 10, 0, INT_MAX, NULL, NULL);
  675. if (year < 1970) {
  676. *time_out = 0;
  677. ++n_bogus_times;
  678. return 0;
  679. } else
  680. return parse_iso_time(s, time_out);
  681. }
  682. /** We've read a time <b>t</b> from a file stored at <b>stored_at</b>, which
  683. * says we started measuring at <b>started_measuring</b>. Return a new number
  684. * that's about as much before <b>now</b> as <b>t</b> was before
  685. * <b>stored_at</b>.
  686. */
  687. static INLINE time_t
  688. correct_time(time_t t, time_t now, time_t stored_at, time_t started_measuring)
  689. {
  690. if (t < started_measuring - 24*60*60*365)
  691. return 0;
  692. else if (t < started_measuring)
  693. return started_measuring;
  694. else if (t > stored_at)
  695. return 0;
  696. else {
  697. long run_length = stored_at - t;
  698. t = now - run_length;
  699. if (t < started_measuring)
  700. t = started_measuring;
  701. return t;
  702. }
  703. }
  704. /** Load MTBF data from disk. Returns 0 on success or recoverable error, -1
  705. * on failure. */
  706. int
  707. rep_hist_load_mtbf_data(time_t now)
  708. {
  709. /* XXXX won't handle being called while history is already populated. */
  710. smartlist_t *lines;
  711. const char *line = NULL;
  712. int r=0, i;
  713. time_t last_downrated = 0, stored_at = 0, tracked_since = 0;
  714. time_t latest_possible_start = now;
  715. long format = -1;
  716. {
  717. char *filename = get_datadir_fname("router-stability");
  718. char *d = read_file_to_str(filename, RFTS_IGNORE_MISSING, NULL);
  719. tor_free(filename);
  720. if (!d)
  721. return -1;
  722. lines = smartlist_create();
  723. smartlist_split_string(lines, d, "\n", SPLIT_SKIP_SPACE, 0);
  724. tor_free(d);
  725. }
  726. {
  727. const char *firstline;
  728. if (smartlist_len(lines)>4) {
  729. firstline = smartlist_get(lines, 0);
  730. if (!strcmpstart(firstline, "format "))
  731. format = tor_parse_long(firstline+strlen("format "),
  732. 10, -1, LONG_MAX, NULL, NULL);
  733. }
  734. }
  735. if (format != 1 && format != 2) {
  736. log_warn(LD_GENERAL,
  737. "Unrecognized format in mtbf history file. Skipping.");
  738. goto err;
  739. }
  740. for (i = 1; i < smartlist_len(lines); ++i) {
  741. line = smartlist_get(lines, i);
  742. if (!strcmp(line, "data"))
  743. break;
  744. if (!strcmpstart(line, "last-downrated ")) {
  745. if (parse_iso_time(line+strlen("last-downrated "), &last_downrated)<0)
  746. log_warn(LD_GENERAL,"Couldn't parse downrate time in mtbf "
  747. "history file.");
  748. }
  749. if (!strcmpstart(line, "stored-at ")) {
  750. if (parse_iso_time(line+strlen("stored-at "), &stored_at)<0)
  751. log_warn(LD_GENERAL,"Couldn't parse stored time in mtbf "
  752. "history file.");
  753. }
  754. if (!strcmpstart(line, "tracked-since ")) {
  755. if (parse_iso_time(line+strlen("tracked-since "), &tracked_since)<0)
  756. log_warn(LD_GENERAL,"Couldn't parse started-tracking time in mtbf "
  757. "history file.");
  758. }
  759. }
  760. if (last_downrated > now)
  761. last_downrated = now;
  762. if (tracked_since > now)
  763. tracked_since = now;
  764. if (!stored_at) {
  765. log_warn(LD_GENERAL, "No stored time recorded.");
  766. goto err;
  767. }
  768. if (line && !strcmp(line, "data"))
  769. ++i;
  770. n_bogus_times = 0;
  771. for (; i < smartlist_len(lines); ++i) {
  772. char digest[DIGEST_LEN];
  773. char hexbuf[HEX_DIGEST_LEN+1];
  774. char mtbf_timebuf[ISO_TIME_LEN+1];
  775. char wfu_timebuf[ISO_TIME_LEN+1];
  776. time_t start_of_run = 0;
  777. time_t start_of_downtime = 0;
  778. int have_mtbf = 0, have_wfu = 0;
  779. long wrl = 0;
  780. double trw = 0;
  781. long wt_uptime = 0, total_wt_time = 0;
  782. int n;
  783. or_history_t *hist;
  784. line = smartlist_get(lines, i);
  785. if (!strcmp(line, "."))
  786. break;
  787. mtbf_timebuf[0] = '\0';
  788. wfu_timebuf[0] = '\0';
  789. if (format == 1) {
  790. n = sscanf(line, "%40s %ld %lf S=%10s %8s",
  791. hexbuf, &wrl, &trw, mtbf_timebuf, mtbf_timebuf+11);
  792. if (n != 3 && n != 5) {
  793. log_warn(LD_GENERAL, "Couldn't scan line %s", escaped(line));
  794. continue;
  795. }
  796. have_mtbf = 1;
  797. } else {
  798. // format == 2.
  799. int mtbf_idx, wfu_idx;
  800. if (strcmpstart(line, "R ") || strlen(line) < 2+HEX_DIGEST_LEN)
  801. continue;
  802. strlcpy(hexbuf, line+2, sizeof(hexbuf));
  803. mtbf_idx = find_next_with(lines, i+1, "+MTBF ");
  804. wfu_idx = find_next_with(lines, i+1, "+WFU ");
  805. if (mtbf_idx >= 0) {
  806. const char *mtbfline = smartlist_get(lines, mtbf_idx);
  807. n = sscanf(mtbfline, "+MTBF %lu %lf S=%10s %8s",
  808. &wrl, &trw, mtbf_timebuf, mtbf_timebuf+11);
  809. if (n == 2 || n == 4) {
  810. have_mtbf = 1;
  811. } else {
  812. log_warn(LD_GENERAL, "Couldn't scan +MTBF line %s",
  813. escaped(mtbfline));
  814. }
  815. }
  816. if (wfu_idx >= 0) {
  817. const char *wfuline = smartlist_get(lines, wfu_idx);
  818. n = sscanf(wfuline, "+WFU %lu %lu S=%10s %8s",
  819. &wt_uptime, &total_wt_time,
  820. wfu_timebuf, wfu_timebuf+11);
  821. if (n == 2 || n == 4) {
  822. have_wfu = 1;
  823. } else {
  824. log_warn(LD_GENERAL, "Couldn't scan +WFU line %s", escaped(wfuline));
  825. }
  826. }
  827. if (wfu_idx > i)
  828. i = wfu_idx;
  829. if (mtbf_idx > i)
  830. i = mtbf_idx;
  831. }
  832. if (base16_decode(digest, DIGEST_LEN, hexbuf, HEX_DIGEST_LEN) < 0) {
  833. log_warn(LD_GENERAL, "Couldn't hex string %s", escaped(hexbuf));
  834. continue;
  835. }
  836. hist = get_or_history(digest);
  837. if (!hist)
  838. continue;
  839. if (have_mtbf) {
  840. if (mtbf_timebuf[0]) {
  841. mtbf_timebuf[10] = ' ';
  842. if (parse_possibly_bad_iso_time(mtbf_timebuf, &start_of_run)<0)
  843. log_warn(LD_GENERAL, "Couldn't parse time %s",
  844. escaped(mtbf_timebuf));
  845. }
  846. hist->start_of_run = correct_time(start_of_run, now, stored_at,
  847. tracked_since);
  848. if (hist->start_of_run < latest_possible_start + wrl)
  849. latest_possible_start = hist->start_of_run - wrl;
  850. hist->weighted_run_length = wrl;
  851. hist->total_run_weights = trw;
  852. }
  853. if (have_wfu) {
  854. if (wfu_timebuf[0]) {
  855. wfu_timebuf[10] = ' ';
  856. if (parse_possibly_bad_iso_time(wfu_timebuf, &start_of_downtime)<0)
  857. log_warn(LD_GENERAL, "Couldn't parse time %s", escaped(wfu_timebuf));
  858. }
  859. }
  860. hist->start_of_downtime = correct_time(start_of_downtime, now, stored_at,
  861. tracked_since);
  862. hist->weighted_uptime = wt_uptime;
  863. hist->total_weighted_time = total_wt_time;
  864. }
  865. if (strcmp(line, "."))
  866. log_warn(LD_GENERAL, "Truncated MTBF file.");
  867. if (!tracked_since)
  868. tracked_since = latest_possible_start;
  869. stability_last_downrated = last_downrated;
  870. started_tracking_stability = tracked_since;
  871. goto done;
  872. err:
  873. r = -1;
  874. done:
  875. SMARTLIST_FOREACH(lines, char *, cp, tor_free(cp));
  876. smartlist_free(lines);
  877. return r;
  878. }
  879. /** For how many seconds do we keep track of individual per-second bandwidth
  880. * totals? */
  881. #define NUM_SECS_ROLLING_MEASURE 10
  882. /** How large are the intervals for which we track and report bandwidth use? */
  883. #define NUM_SECS_BW_SUM_INTERVAL (15*60)
  884. /** How far in the past do we remember and publish bandwidth use? */
  885. #define NUM_SECS_BW_SUM_IS_VALID (24*60*60)
  886. /** How many bandwidth usage intervals do we remember? (derived) */
  887. #define NUM_TOTALS (NUM_SECS_BW_SUM_IS_VALID/NUM_SECS_BW_SUM_INTERVAL)
  888. /** Structure to track bandwidth use, and remember the maxima for a given
  889. * time period.
  890. */
  891. typedef struct bw_array_t {
  892. /** Observation array: Total number of bytes transferred in each of the last
  893. * NUM_SECS_ROLLING_MEASURE seconds. This is used as a circular array. */
  894. uint64_t obs[NUM_SECS_ROLLING_MEASURE];
  895. int cur_obs_idx; /**< Current position in obs. */
  896. time_t cur_obs_time; /**< Time represented in obs[cur_obs_idx] */
  897. uint64_t total_obs; /**< Total for all members of obs except
  898. * obs[cur_obs_idx] */
  899. uint64_t max_total; /**< Largest value that total_obs has taken on in the
  900. * current period. */
  901. uint64_t total_in_period; /**< Total bytes transferred in the current
  902. * period. */
  903. /** When does the next period begin? */
  904. time_t next_period;
  905. /** Where in 'maxima' should the maximum bandwidth usage for the current
  906. * period be stored? */
  907. int next_max_idx;
  908. /** How many values in maxima/totals have been set ever? */
  909. int num_maxes_set;
  910. /** Circular array of the maximum
  911. * bandwidth-per-NUM_SECS_ROLLING_MEASURE usage for the last
  912. * NUM_TOTALS periods */
  913. uint64_t maxima[NUM_TOTALS];
  914. /** Circular array of the total bandwidth usage for the last NUM_TOTALS
  915. * periods */
  916. uint64_t totals[NUM_TOTALS];
  917. } bw_array_t;
  918. /** Shift the current period of b forward by one. */
  919. static void
  920. commit_max(bw_array_t *b)
  921. {
  922. /* Store total from current period. */
  923. b->totals[b->next_max_idx] = b->total_in_period;
  924. /* Store maximum from current period. */
  925. b->maxima[b->next_max_idx++] = b->max_total;
  926. /* Advance next_period and next_max_idx */
  927. b->next_period += NUM_SECS_BW_SUM_INTERVAL;
  928. if (b->next_max_idx == NUM_TOTALS)
  929. b->next_max_idx = 0;
  930. if (b->num_maxes_set < NUM_TOTALS)
  931. ++b->num_maxes_set;
  932. /* Reset max_total. */
  933. b->max_total = 0;
  934. /* Reset total_in_period. */
  935. b->total_in_period = 0;
  936. }
  937. /** Shift the current observation time of 'b' forward by one second. */
  938. static INLINE void
  939. advance_obs(bw_array_t *b)
  940. {
  941. int nextidx;
  942. uint64_t total;
  943. /* Calculate the total bandwidth for the last NUM_SECS_ROLLING_MEASURE
  944. * seconds; adjust max_total as needed.*/
  945. total = b->total_obs + b->obs[b->cur_obs_idx];
  946. if (total > b->max_total)
  947. b->max_total = total;
  948. nextidx = b->cur_obs_idx+1;
  949. if (nextidx == NUM_SECS_ROLLING_MEASURE)
  950. nextidx = 0;
  951. b->total_obs = total - b->obs[nextidx];
  952. b->obs[nextidx]=0;
  953. b->cur_obs_idx = nextidx;
  954. if (++b->cur_obs_time >= b->next_period)
  955. commit_max(b);
  956. }
  957. /** Add <b>n</b> bytes to the number of bytes in <b>b</b> for second
  958. * <b>when</b>. */
  959. static INLINE void
  960. add_obs(bw_array_t *b, time_t when, uint64_t n)
  961. {
  962. /* Don't record data in the past. */
  963. if (when<b->cur_obs_time)
  964. return;
  965. /* If we're currently adding observations for an earlier second than
  966. * 'when', advance b->cur_obs_time and b->cur_obs_idx by an
  967. * appropriate number of seconds, and do all the other housekeeping */
  968. while (when>b->cur_obs_time)
  969. advance_obs(b);
  970. b->obs[b->cur_obs_idx] += n;
  971. b->total_in_period += n;
  972. }
  973. /** Allocate, initialize, and return a new bw_array. */
  974. static bw_array_t *
  975. bw_array_new(void)
  976. {
  977. bw_array_t *b;
  978. time_t start;
  979. b = tor_malloc_zero(sizeof(bw_array_t));
  980. rephist_total_alloc += sizeof(bw_array_t);
  981. start = time(NULL);
  982. b->cur_obs_time = start;
  983. b->next_period = start + NUM_SECS_BW_SUM_INTERVAL;
  984. return b;
  985. }
  986. static bw_array_t *read_array = NULL;
  987. static bw_array_t *write_array = NULL;
  988. /** Set up read_array and write_array. */
  989. static void
  990. bw_arrays_init(void)
  991. {
  992. read_array = bw_array_new();
  993. write_array = bw_array_new();
  994. }
  995. /** We read <b>num_bytes</b> more bytes in second <b>when</b>.
  996. *
  997. * Add num_bytes to the current running total for <b>when</b>.
  998. *
  999. * <b>when</b> can go back to time, but it's safe to ignore calls
  1000. * earlier than the latest <b>when</b> you've heard of.
  1001. */
  1002. void
  1003. rep_hist_note_bytes_written(size_t num_bytes, time_t when)
  1004. {
  1005. /* Maybe a circular array for recent seconds, and step to a new point
  1006. * every time a new second shows up. Or simpler is to just to have
  1007. * a normal array and push down each item every second; it's short.
  1008. */
  1009. /* When a new second has rolled over, compute the sum of the bytes we've
  1010. * seen over when-1 to when-1-NUM_SECS_ROLLING_MEASURE, and stick it
  1011. * somewhere. See rep_hist_bandwidth_assess() below.
  1012. */
  1013. add_obs(write_array, when, num_bytes);
  1014. }
  1015. /** We wrote <b>num_bytes</b> more bytes in second <b>when</b>.
  1016. * (like rep_hist_note_bytes_written() above)
  1017. */
  1018. void
  1019. rep_hist_note_bytes_read(size_t num_bytes, time_t when)
  1020. {
  1021. /* if we're smart, we can make this func and the one above share code */
  1022. add_obs(read_array, when, num_bytes);
  1023. }
  1024. /** Helper: Return the largest value in b->maxima. (This is equal to the
  1025. * most bandwidth used in any NUM_SECS_ROLLING_MEASURE period for the last
  1026. * NUM_SECS_BW_SUM_IS_VALID seconds.)
  1027. */
  1028. static uint64_t
  1029. find_largest_max(bw_array_t *b)
  1030. {
  1031. int i;
  1032. uint64_t max;
  1033. max=0;
  1034. for (i=0; i<NUM_TOTALS; ++i) {
  1035. if (b->maxima[i]>max)
  1036. max = b->maxima[i];
  1037. }
  1038. return max;
  1039. }
  1040. /** Find the largest sums in the past NUM_SECS_BW_SUM_IS_VALID (roughly)
  1041. * seconds. Find one sum for reading and one for writing. They don't have
  1042. * to be at the same time.
  1043. *
  1044. * Return the smaller of these sums, divided by NUM_SECS_ROLLING_MEASURE.
  1045. */
  1046. int
  1047. rep_hist_bandwidth_assess(void)
  1048. {
  1049. uint64_t w,r;
  1050. r = find_largest_max(read_array);
  1051. w = find_largest_max(write_array);
  1052. if (r>w)
  1053. return (int)(U64_TO_DBL(w)/NUM_SECS_ROLLING_MEASURE);
  1054. else
  1055. return (int)(U64_TO_DBL(r)/NUM_SECS_ROLLING_MEASURE);
  1056. }
  1057. /** Print the bandwidth history of b (either read_array or write_array)
  1058. * into the buffer pointed to by buf. The format is simply comma
  1059. * separated numbers, from oldest to newest.
  1060. *
  1061. * It returns the number of bytes written.
  1062. */
  1063. static size_t
  1064. rep_hist_fill_bandwidth_history(char *buf, size_t len, bw_array_t *b)
  1065. {
  1066. char *cp = buf;
  1067. int i, n;
  1068. or_options_t *options = get_options();
  1069. uint64_t cutoff;
  1070. if (b->num_maxes_set <= b->next_max_idx) {
  1071. /* We haven't been through the circular array yet; time starts at i=0.*/
  1072. i = 0;
  1073. } else {
  1074. /* We've been around the array at least once. The next i to be
  1075. overwritten is the oldest. */
  1076. i = b->next_max_idx;
  1077. }
  1078. if (options->RelayBandwidthRate) {
  1079. /* We don't want to report that we used more bandwidth than the max we're
  1080. * willing to relay; otherwise everybody will know how much traffic
  1081. * we used ourself. */
  1082. cutoff = options->RelayBandwidthRate * NUM_SECS_BW_SUM_INTERVAL;
  1083. } else {
  1084. cutoff = UINT64_MAX;
  1085. }
  1086. for (n=0; n<b->num_maxes_set; ++n,++i) {
  1087. uint64_t total;
  1088. if (i >= NUM_TOTALS)
  1089. i -= NUM_TOTALS;
  1090. tor_assert(i < NUM_TOTALS);
  1091. /* Round the bandwidth used down to the nearest 1k. */
  1092. total = b->totals[i] & ~0x3ff;
  1093. if (total > cutoff)
  1094. total = cutoff;
  1095. if (n==(b->num_maxes_set-1))
  1096. tor_snprintf(cp, len-(cp-buf), U64_FORMAT, U64_PRINTF_ARG(total));
  1097. else
  1098. tor_snprintf(cp, len-(cp-buf), U64_FORMAT",", U64_PRINTF_ARG(total));
  1099. cp += strlen(cp);
  1100. }
  1101. return cp-buf;
  1102. }
  1103. /** Allocate and return lines for representing this server's bandwidth
  1104. * history in its descriptor.
  1105. */
  1106. char *
  1107. rep_hist_get_bandwidth_lines(int for_extrainfo)
  1108. {
  1109. char *buf, *cp;
  1110. char t[ISO_TIME_LEN+1];
  1111. int r;
  1112. bw_array_t *b;
  1113. size_t len;
  1114. /* opt (read|write)-history yyyy-mm-dd HH:MM:SS (n s) n,n,n,n,n... */
  1115. len = (60+20*NUM_TOTALS)*2;
  1116. buf = tor_malloc_zero(len);
  1117. cp = buf;
  1118. for (r=0;r<2;++r) {
  1119. b = r?read_array:write_array;
  1120. tor_assert(b);
  1121. format_iso_time(t, b->next_period-NUM_SECS_BW_SUM_INTERVAL);
  1122. tor_snprintf(cp, len-(cp-buf), "%s%s %s (%d s) ",
  1123. for_extrainfo ? "" : "opt ",
  1124. r ? "read-history" : "write-history", t,
  1125. NUM_SECS_BW_SUM_INTERVAL);
  1126. cp += strlen(cp);
  1127. cp += rep_hist_fill_bandwidth_history(cp, len-(cp-buf), b);
  1128. strlcat(cp, "\n", len-(cp-buf));
  1129. ++cp;
  1130. }
  1131. return buf;
  1132. }
  1133. /** Update <b>state</b> with the newest bandwidth history. */
  1134. void
  1135. rep_hist_update_state(or_state_t *state)
  1136. {
  1137. int len, r;
  1138. char *buf, *cp;
  1139. smartlist_t **s_values;
  1140. time_t *s_begins;
  1141. int *s_interval;
  1142. bw_array_t *b;
  1143. len = 20*NUM_TOTALS+1;
  1144. buf = tor_malloc_zero(len);
  1145. for (r=0;r<2;++r) {
  1146. b = r?read_array:write_array;
  1147. s_begins = r?&state->BWHistoryReadEnds :&state->BWHistoryWriteEnds;
  1148. s_interval= r?&state->BWHistoryReadInterval:&state->BWHistoryWriteInterval;
  1149. s_values = r?&state->BWHistoryReadValues :&state->BWHistoryWriteValues;
  1150. if (*s_values) {
  1151. SMARTLIST_FOREACH(*s_values, char *, val, tor_free(val));
  1152. smartlist_free(*s_values);
  1153. }
  1154. if (! server_mode(get_options())) {
  1155. /* Clients don't need to store bandwidth history persistently;
  1156. * force these values to the defaults. */
  1157. /* FFFF we should pull the default out of config.c's state table,
  1158. * so we don't have two defaults. */
  1159. if (*s_begins != 0 || *s_interval != 900) {
  1160. time_t now = time(NULL);
  1161. time_t save_at = get_options()->AvoidDiskWrites ? now+3600 : now+600;
  1162. or_state_mark_dirty(state, save_at);
  1163. }
  1164. *s_begins = 0;
  1165. *s_interval = 900;
  1166. *s_values = smartlist_create();
  1167. continue;
  1168. }
  1169. *s_begins = b->next_period;
  1170. *s_interval = NUM_SECS_BW_SUM_INTERVAL;
  1171. cp = buf;
  1172. cp += rep_hist_fill_bandwidth_history(cp, len, b);
  1173. tor_snprintf(cp, len-(cp-buf), cp == buf ? U64_FORMAT : ","U64_FORMAT,
  1174. U64_PRINTF_ARG(b->total_in_period));
  1175. *s_values = smartlist_create();
  1176. if (server_mode(get_options()))
  1177. smartlist_split_string(*s_values, buf, ",", SPLIT_SKIP_SPACE, 0);
  1178. }
  1179. tor_free(buf);
  1180. if (server_mode(get_options())) {
  1181. or_state_mark_dirty(get_or_state(), time(NULL)+(2*3600));
  1182. }
  1183. }
  1184. /** Set bandwidth history from our saved state. */
  1185. int
  1186. rep_hist_load_state(or_state_t *state, char **err)
  1187. {
  1188. time_t s_begins, start;
  1189. time_t now = time(NULL);
  1190. uint64_t v;
  1191. int r,i,ok;
  1192. int all_ok = 1;
  1193. int s_interval;
  1194. smartlist_t *s_values;
  1195. bw_array_t *b;
  1196. /* Assert they already have been malloced */
  1197. tor_assert(read_array && write_array);
  1198. for (r=0;r<2;++r) {
  1199. b = r?read_array:write_array;
  1200. s_begins = r?state->BWHistoryReadEnds:state->BWHistoryWriteEnds;
  1201. s_interval = r?state->BWHistoryReadInterval:state->BWHistoryWriteInterval;
  1202. s_values = r?state->BWHistoryReadValues:state->BWHistoryWriteValues;
  1203. if (s_values && s_begins >= now - NUM_SECS_BW_SUM_INTERVAL*NUM_TOTALS) {
  1204. start = s_begins - s_interval*(smartlist_len(s_values));
  1205. if (start > now)
  1206. continue;
  1207. b->cur_obs_time = start;
  1208. b->next_period = start + NUM_SECS_BW_SUM_INTERVAL;
  1209. SMARTLIST_FOREACH(s_values, char *, cp, {
  1210. v = tor_parse_uint64(cp, 10, 0, UINT64_MAX, &ok, NULL);
  1211. if (!ok) {
  1212. all_ok=0;
  1213. log_notice(LD_GENERAL, "Could not parse '%s' into a number.'", cp);
  1214. }
  1215. if (start < now) {
  1216. add_obs(b, start, v);
  1217. start += NUM_SECS_BW_SUM_INTERVAL;
  1218. }
  1219. });
  1220. }
  1221. /* Clean up maxima and observed */
  1222. /* Do we really want to zero this for the purpose of max capacity? */
  1223. for (i=0; i<NUM_SECS_ROLLING_MEASURE; ++i) {
  1224. b->obs[i] = 0;
  1225. }
  1226. b->total_obs = 0;
  1227. for (i=0; i<NUM_TOTALS; ++i) {
  1228. b->maxima[i] = 0;
  1229. }
  1230. b->max_total = 0;
  1231. }
  1232. if (!all_ok) {
  1233. *err = tor_strdup("Parsing of bandwidth history values failed");
  1234. /* and create fresh arrays */
  1235. tor_free(read_array);
  1236. tor_free(write_array);
  1237. read_array = bw_array_new();
  1238. write_array = bw_array_new();
  1239. return -1;
  1240. }
  1241. return 0;
  1242. }
  1243. /*********************************************************************/
  1244. /** A list of port numbers that have been used recently. */
  1245. static smartlist_t *predicted_ports_list=NULL;
  1246. /** The corresponding most recently used time for each port. */
  1247. static smartlist_t *predicted_ports_times=NULL;
  1248. /** We just got an application request for a connection with
  1249. * port <b>port</b>. Remember it for the future, so we can keep
  1250. * some circuits open that will exit to this port.
  1251. */
  1252. static void
  1253. add_predicted_port(uint16_t port, time_t now)
  1254. {
  1255. /* XXXX we could just use uintptr_t here, I think. */
  1256. uint16_t *tmp_port = tor_malloc(sizeof(uint16_t));
  1257. time_t *tmp_time = tor_malloc(sizeof(time_t));
  1258. *tmp_port = port;
  1259. *tmp_time = now;
  1260. rephist_total_alloc += sizeof(uint16_t) + sizeof(time_t);
  1261. smartlist_add(predicted_ports_list, tmp_port);
  1262. smartlist_add(predicted_ports_times, tmp_time);
  1263. }
  1264. /** Initialize whatever memory and structs are needed for predicting
  1265. * which ports will be used. Also seed it with port 80, so we'll build
  1266. * circuits on start-up.
  1267. */
  1268. static void
  1269. predicted_ports_init(void)
  1270. {
  1271. predicted_ports_list = smartlist_create();
  1272. predicted_ports_times = smartlist_create();
  1273. add_predicted_port(80, time(NULL)); /* add one to kickstart us */
  1274. }
  1275. /** Free whatever memory is needed for predicting which ports will
  1276. * be used.
  1277. */
  1278. static void
  1279. predicted_ports_free(void)
  1280. {
  1281. rephist_total_alloc -= smartlist_len(predicted_ports_list)*sizeof(uint16_t);
  1282. SMARTLIST_FOREACH(predicted_ports_list, char *, cp, tor_free(cp));
  1283. smartlist_free(predicted_ports_list);
  1284. rephist_total_alloc -= smartlist_len(predicted_ports_times)*sizeof(time_t);
  1285. SMARTLIST_FOREACH(predicted_ports_times, char *, cp, tor_free(cp));
  1286. smartlist_free(predicted_ports_times);
  1287. }
  1288. /** Remember that <b>port</b> has been asked for as of time <b>now</b>.
  1289. * This is used for predicting what sorts of streams we'll make in the
  1290. * future and making exit circuits to anticipate that.
  1291. */
  1292. void
  1293. rep_hist_note_used_port(uint16_t port, time_t now)
  1294. {
  1295. int i;
  1296. uint16_t *tmp_port;
  1297. time_t *tmp_time;
  1298. tor_assert(predicted_ports_list);
  1299. tor_assert(predicted_ports_times);
  1300. if (!port) /* record nothing */
  1301. return;
  1302. for (i = 0; i < smartlist_len(predicted_ports_list); ++i) {
  1303. tmp_port = smartlist_get(predicted_ports_list, i);
  1304. tmp_time = smartlist_get(predicted_ports_times, i);
  1305. if (*tmp_port == port) {
  1306. *tmp_time = now;
  1307. return;
  1308. }
  1309. }
  1310. /* it's not there yet; we need to add it */
  1311. add_predicted_port(port, now);
  1312. }
  1313. /** For this long after we've seen a request for a given port, assume that
  1314. * we'll want to make connections to the same port in the future. */
  1315. #define PREDICTED_CIRCS_RELEVANCE_TIME (60*60)
  1316. /** Return a pointer to the list of port numbers that
  1317. * are likely to be asked for in the near future.
  1318. *
  1319. * The caller promises not to mess with it.
  1320. */
  1321. smartlist_t *
  1322. rep_hist_get_predicted_ports(time_t now)
  1323. {
  1324. int i;
  1325. uint16_t *tmp_port;
  1326. time_t *tmp_time;
  1327. tor_assert(predicted_ports_list);
  1328. tor_assert(predicted_ports_times);
  1329. /* clean out obsolete entries */
  1330. for (i = 0; i < smartlist_len(predicted_ports_list); ++i) {
  1331. tmp_time = smartlist_get(predicted_ports_times, i);
  1332. if (*tmp_time + PREDICTED_CIRCS_RELEVANCE_TIME < now) {
  1333. tmp_port = smartlist_get(predicted_ports_list, i);
  1334. log_debug(LD_CIRC, "Expiring predicted port %d", *tmp_port);
  1335. smartlist_del(predicted_ports_list, i);
  1336. smartlist_del(predicted_ports_times, i);
  1337. rephist_total_alloc -= sizeof(uint16_t)+sizeof(time_t);
  1338. tor_free(tmp_port);
  1339. tor_free(tmp_time);
  1340. i--;
  1341. }
  1342. }
  1343. return predicted_ports_list;
  1344. }
  1345. /** The user asked us to do a resolve. Rather than keeping track of
  1346. * timings and such of resolves, we fake it for now by making treating
  1347. * it the same way as a connection to port 80. This way we will continue
  1348. * to have circuits lying around if the user only uses Tor for resolves.
  1349. */
  1350. void
  1351. rep_hist_note_used_resolve(time_t now)
  1352. {
  1353. rep_hist_note_used_port(80, now);
  1354. }
  1355. /** The last time at which we needed an internal circ. */
  1356. static time_t predicted_internal_time = 0;
  1357. /** The last time we needed an internal circ with good uptime. */
  1358. static time_t predicted_internal_uptime_time = 0;
  1359. /** The last time we needed an internal circ with good capacity. */
  1360. static time_t predicted_internal_capacity_time = 0;
  1361. /** Remember that we used an internal circ at time <b>now</b>. */
  1362. void
  1363. rep_hist_note_used_internal(time_t now, int need_uptime, int need_capacity)
  1364. {
  1365. predicted_internal_time = now;
  1366. if (need_uptime)
  1367. predicted_internal_uptime_time = now;
  1368. if (need_capacity)
  1369. predicted_internal_capacity_time = now;
  1370. }
  1371. /** Return 1 if we've used an internal circ recently; else return 0. */
  1372. int
  1373. rep_hist_get_predicted_internal(time_t now, int *need_uptime,
  1374. int *need_capacity)
  1375. {
  1376. if (!predicted_internal_time) { /* initialize it */
  1377. predicted_internal_time = now;
  1378. predicted_internal_uptime_time = now;
  1379. predicted_internal_capacity_time = now;
  1380. }
  1381. if (predicted_internal_time + PREDICTED_CIRCS_RELEVANCE_TIME < now)
  1382. return 0; /* too long ago */
  1383. if (predicted_internal_uptime_time + PREDICTED_CIRCS_RELEVANCE_TIME >= now)
  1384. *need_uptime = 1;
  1385. if (predicted_internal_capacity_time + PREDICTED_CIRCS_RELEVANCE_TIME >= now)
  1386. *need_capacity = 1;
  1387. return 1;
  1388. }
  1389. /** Any ports used lately? These are pre-seeded if we just started
  1390. * up or if we're running a hidden service. */
  1391. int
  1392. any_predicted_circuits(time_t now)
  1393. {
  1394. return smartlist_len(predicted_ports_list) ||
  1395. predicted_internal_time + PREDICTED_CIRCS_RELEVANCE_TIME >= now;
  1396. }
  1397. /** Return 1 if we have no need for circuits currently, else return 0. */
  1398. int
  1399. rep_hist_circbuilding_dormant(time_t now)
  1400. {
  1401. if (any_predicted_circuits(now))
  1402. return 0;
  1403. /* see if we'll still need to build testing circuits */
  1404. if (server_mode(get_options()) &&
  1405. (!check_whether_orport_reachable() || !circuit_enough_testing_circs()))
  1406. return 0;
  1407. if (!check_whether_dirport_reachable())
  1408. return 0;
  1409. return 1;
  1410. }
  1411. static uint32_t n_signed_dir_objs = 0;
  1412. static uint32_t n_signed_routerdescs = 0;
  1413. static uint32_t n_verified_dir_objs = 0;
  1414. static uint32_t n_verified_routerdescs = 0;
  1415. static uint32_t n_onionskins_encrypted = 0;
  1416. static uint32_t n_onionskins_decrypted = 0;
  1417. static uint32_t n_tls_client_handshakes = 0;
  1418. static uint32_t n_tls_server_handshakes = 0;
  1419. static uint32_t n_rend_client_ops = 0;
  1420. static uint32_t n_rend_mid_ops = 0;
  1421. static uint32_t n_rend_server_ops = 0;
  1422. /** Increment the count of the number of times we've done <b>operation</b>. */
  1423. void
  1424. note_crypto_pk_op(pk_op_t operation)
  1425. {
  1426. switch (operation)
  1427. {
  1428. case SIGN_DIR:
  1429. n_signed_dir_objs++;
  1430. break;
  1431. case SIGN_RTR:
  1432. n_signed_routerdescs++;
  1433. break;
  1434. case VERIFY_DIR:
  1435. n_verified_dir_objs++;
  1436. break;
  1437. case VERIFY_RTR:
  1438. n_verified_routerdescs++;
  1439. break;
  1440. case ENC_ONIONSKIN:
  1441. n_onionskins_encrypted++;
  1442. break;
  1443. case DEC_ONIONSKIN:
  1444. n_onionskins_decrypted++;
  1445. break;
  1446. case TLS_HANDSHAKE_C:
  1447. n_tls_client_handshakes++;
  1448. break;
  1449. case TLS_HANDSHAKE_S:
  1450. n_tls_server_handshakes++;
  1451. break;
  1452. case REND_CLIENT:
  1453. n_rend_client_ops++;
  1454. break;
  1455. case REND_MID:
  1456. n_rend_mid_ops++;
  1457. break;
  1458. case REND_SERVER:
  1459. n_rend_server_ops++;
  1460. break;
  1461. default:
  1462. log_warn(LD_BUG, "Unknown pk operation %d", operation);
  1463. }
  1464. }
  1465. /** Log the number of times we've done each public/private-key operation. */
  1466. void
  1467. dump_pk_ops(int severity)
  1468. {
  1469. log(severity, LD_GENERAL,
  1470. "PK operations: %lu directory objects signed, "
  1471. "%lu directory objects verified, "
  1472. "%lu routerdescs signed, "
  1473. "%lu routerdescs verified, "
  1474. "%lu onionskins encrypted, "
  1475. "%lu onionskins decrypted, "
  1476. "%lu client-side TLS handshakes, "
  1477. "%lu server-side TLS handshakes, "
  1478. "%lu rendezvous client operations, "
  1479. "%lu rendezvous middle operations, "
  1480. "%lu rendezvous server operations.",
  1481. (unsigned long) n_signed_dir_objs,
  1482. (unsigned long) n_verified_dir_objs,
  1483. (unsigned long) n_signed_routerdescs,
  1484. (unsigned long) n_verified_routerdescs,
  1485. (unsigned long) n_onionskins_encrypted,
  1486. (unsigned long) n_onionskins_decrypted,
  1487. (unsigned long) n_tls_client_handshakes,
  1488. (unsigned long) n_tls_server_handshakes,
  1489. (unsigned long) n_rend_client_ops,
  1490. (unsigned long) n_rend_mid_ops,
  1491. (unsigned long) n_rend_server_ops);
  1492. }
  1493. /** Free all storage held by the OR/link history caches, by the
  1494. * bandwidth history arrays, or by the port history. */
  1495. void
  1496. rep_hist_free_all(void)
  1497. {
  1498. digestmap_free(history_map, free_or_history);
  1499. tor_free(read_array);
  1500. tor_free(write_array);
  1501. predicted_ports_free();
  1502. }
  1503. /****************** hidden service usage statistics ******************/
  1504. /** How large are the intervals for which we track and report hidden service
  1505. * use? */
  1506. #define NUM_SECS_HS_USAGE_SUM_INTERVAL (15*60)
  1507. /** How far in the past do we remember and publish hidden service use? */
  1508. #define NUM_SECS_HS_USAGE_SUM_IS_VALID (24*60*60)
  1509. /** How many hidden service usage intervals do we remember? (derived) */
  1510. #define NUM_TOTALS_HS_USAGE (NUM_SECS_HS_USAGE_SUM_IS_VALID/ \
  1511. NUM_SECS_HS_USAGE_SUM_INTERVAL)
  1512. /** List element containing a service id and the count. */
  1513. typedef struct hs_usage_list_elem_t {
  1514. /** Service id of this elem. */
  1515. char service_id[REND_SERVICE_ID_LEN_BASE32+1];
  1516. /** Number of occurrences for the given service id. */
  1517. uint32_t count;
  1518. /* Pointer to next list elem */
  1519. struct hs_usage_list_elem_t *next;
  1520. } hs_usage_list_elem_t;
  1521. /** Ordered list that stores service ids and the number of observations. It is
  1522. * ordered by the number of occurrences in descending order. Its purpose is to
  1523. * calculate the frequency distribution when the period is over. */
  1524. typedef struct hs_usage_list_t {
  1525. /* Pointer to the first element in the list. */
  1526. hs_usage_list_elem_t *start;
  1527. /* Number of total occurrences for all list elements. */
  1528. uint32_t total_count;
  1529. /* Number of service ids, i.e. number of list elements. */
  1530. uint32_t total_service_ids;
  1531. } hs_usage_list_t;
  1532. /** Tracks service-related observations in the current period and their
  1533. * history. */
  1534. typedef struct hs_usage_service_related_observation_t {
  1535. /** Ordered list that stores service ids and the number of observations in
  1536. * the current period. It is ordered by the number of occurrences in
  1537. * descending order. Its purpose is to calculate the frequency distribution
  1538. * when the period is over. */
  1539. hs_usage_list_t *list;
  1540. /** Circular arrays that store the history of observations. totals stores all
  1541. * observations, twenty (ten, five) the number of observations related to a
  1542. * service id being accounted for the top 20 (10, 5) percent of all
  1543. * observations. */
  1544. uint32_t totals[NUM_TOTALS_HS_USAGE];
  1545. uint32_t five[NUM_TOTALS_HS_USAGE];
  1546. uint32_t ten[NUM_TOTALS_HS_USAGE];
  1547. uint32_t twenty[NUM_TOTALS_HS_USAGE];
  1548. } hs_usage_service_related_observation_t;
  1549. /** Tracks the history of general period-related observations, i.e. those that
  1550. * cannot be related to a specific service id. */
  1551. typedef struct hs_usage_general_period_related_observations_t {
  1552. /** Circular array that stores the history of observations. */
  1553. uint32_t totals[NUM_TOTALS_HS_USAGE];
  1554. } hs_usage_general_period_related_observations_t;
  1555. /** Keeps information about the current observation period and its relation to
  1556. * the histories of observations. */
  1557. typedef struct hs_usage_current_observation_period_t {
  1558. /** Where do we write the next history entry? */
  1559. int next_idx;
  1560. /** How many values in history have been set ever? (upper bound!) */
  1561. int num_set;
  1562. /** When did this period begin? */
  1563. time_t start_of_current_period;
  1564. /** When does the next period begin? */
  1565. time_t start_of_next_period;
  1566. } hs_usage_current_observation_period_t;
  1567. static hs_usage_current_observation_period_t *current_period = NULL;
  1568. static hs_usage_service_related_observation_t *publish_total = NULL;
  1569. static hs_usage_service_related_observation_t *publish_novel = NULL;
  1570. static hs_usage_service_related_observation_t *fetch_total = NULL;
  1571. static hs_usage_service_related_observation_t *fetch_successful = NULL;
  1572. static hs_usage_general_period_related_observations_t *descs = NULL;
  1573. /** Creates an empty ordered list element. */
  1574. static hs_usage_list_elem_t *
  1575. hs_usage_list_elem_new(void)
  1576. {
  1577. hs_usage_list_elem_t *e;
  1578. e = tor_malloc_zero(sizeof(hs_usage_list_elem_t));
  1579. rephist_total_alloc += sizeof(hs_usage_list_elem_t);
  1580. e->count = 1;
  1581. e->next = NULL;
  1582. return e;
  1583. }
  1584. /** Creates an empty ordered list. */
  1585. static hs_usage_list_t *
  1586. hs_usage_list_new(void)
  1587. {
  1588. hs_usage_list_t *l;
  1589. l = tor_malloc_zero(sizeof(hs_usage_list_t));
  1590. rephist_total_alloc += sizeof(hs_usage_list_t);
  1591. l->start = NULL;
  1592. l->total_count = 0;
  1593. l->total_service_ids = 0;
  1594. return l;
  1595. }
  1596. /** Creates an empty structure for storing service-related observations. */
  1597. static hs_usage_service_related_observation_t *
  1598. hs_usage_service_related_observation_new(void)
  1599. {
  1600. hs_usage_service_related_observation_t *h;
  1601. h = tor_malloc_zero(sizeof(hs_usage_service_related_observation_t));
  1602. rephist_total_alloc += sizeof(hs_usage_service_related_observation_t);
  1603. h->list = hs_usage_list_new();
  1604. return h;
  1605. }
  1606. /** Creates an empty structure for storing general period-related
  1607. * observations. */
  1608. static hs_usage_general_period_related_observations_t *
  1609. hs_usage_general_period_related_observations_new(void)
  1610. {
  1611. hs_usage_general_period_related_observations_t *p;
  1612. p = tor_malloc_zero(sizeof(hs_usage_general_period_related_observations_t));
  1613. rephist_total_alloc+= sizeof(hs_usage_general_period_related_observations_t);
  1614. return p;
  1615. }
  1616. /** Creates an empty structure for storing period-specific information. */
  1617. static hs_usage_current_observation_period_t *
  1618. hs_usage_current_observation_period_new(void)
  1619. {
  1620. hs_usage_current_observation_period_t *c;
  1621. time_t now;
  1622. c = tor_malloc_zero(sizeof(hs_usage_current_observation_period_t));
  1623. rephist_total_alloc += sizeof(hs_usage_current_observation_period_t);
  1624. now = time(NULL);
  1625. c->start_of_current_period = now;
  1626. c->start_of_next_period = now + NUM_SECS_HS_USAGE_SUM_INTERVAL;
  1627. return c;
  1628. }
  1629. /** Initializes the structures for collecting hidden service usage data. */
  1630. static void
  1631. hs_usage_init(void)
  1632. {
  1633. current_period = hs_usage_current_observation_period_new();
  1634. publish_total = hs_usage_service_related_observation_new();
  1635. publish_novel = hs_usage_service_related_observation_new();
  1636. fetch_total = hs_usage_service_related_observation_new();
  1637. fetch_successful = hs_usage_service_related_observation_new();
  1638. descs = hs_usage_general_period_related_observations_new();
  1639. }
  1640. /** Clears the given ordered list by resetting its attributes and releasing
  1641. * the memory allocated by its elements. */
  1642. static void
  1643. hs_usage_list_clear(hs_usage_list_t *lst)
  1644. {
  1645. /* walk through elements and free memory */
  1646. hs_usage_list_elem_t *current = lst->start;
  1647. hs_usage_list_elem_t *tmp;
  1648. while (current != NULL) {
  1649. tmp = current->next;
  1650. rephist_total_alloc -= sizeof(hs_usage_list_elem_t);
  1651. tor_free(current);
  1652. current = tmp;
  1653. }
  1654. /* reset attributes */
  1655. lst->start = NULL;
  1656. lst->total_count = 0;
  1657. lst->total_service_ids = 0;
  1658. return;
  1659. }
  1660. /** Frees the memory used by the given list. */
  1661. static void
  1662. hs_usage_list_free(hs_usage_list_t *lst)
  1663. {
  1664. if (!lst)
  1665. return;
  1666. hs_usage_list_clear(lst);
  1667. rephist_total_alloc -= sizeof(hs_usage_list_t);
  1668. tor_free(lst);
  1669. }
  1670. /** Frees the memory used by the given service-related observations. */
  1671. static void
  1672. hs_usage_service_related_observation_free(
  1673. hs_usage_service_related_observation_t *s)
  1674. {
  1675. if (!s)
  1676. return;
  1677. hs_usage_list_free(s->list);
  1678. rephist_total_alloc -= sizeof(hs_usage_service_related_observation_t);
  1679. tor_free(s);
  1680. }
  1681. /** Frees the memory used by the given period-specific observations. */
  1682. static void
  1683. hs_usage_general_period_related_observations_free(
  1684. hs_usage_general_period_related_observations_t *s)
  1685. {
  1686. rephist_total_alloc-=sizeof(hs_usage_general_period_related_observations_t);
  1687. tor_free(s);
  1688. }
  1689. /** Frees the memory used by period-specific information. */
  1690. static void
  1691. hs_usage_current_observation_period_free(
  1692. hs_usage_current_observation_period_t *s)
  1693. {
  1694. rephist_total_alloc -= sizeof(hs_usage_current_observation_period_t);
  1695. tor_free(s);
  1696. }
  1697. /** Frees all memory that was used for collecting hidden service usage data. */
  1698. void
  1699. hs_usage_free_all(void)
  1700. {
  1701. hs_usage_general_period_related_observations_free(descs);
  1702. descs = NULL;
  1703. hs_usage_service_related_observation_free(fetch_successful);
  1704. hs_usage_service_related_observation_free(fetch_total);
  1705. hs_usage_service_related_observation_free(publish_novel);
  1706. hs_usage_service_related_observation_free(publish_total);
  1707. fetch_successful = fetch_total = publish_novel = publish_total = NULL;
  1708. hs_usage_current_observation_period_free(current_period);
  1709. current_period = NULL;
  1710. }
  1711. /** Inserts a new occurrence for the given service id to the given ordered
  1712. * list. */
  1713. static void
  1714. hs_usage_insert_value(hs_usage_list_t *lst, const char *service_id)
  1715. {
  1716. /* search if there is already an elem with same service_id in list */
  1717. hs_usage_list_elem_t *current = lst->start;
  1718. hs_usage_list_elem_t *previous = NULL;
  1719. while (current != NULL && strcasecmp(current->service_id,service_id)) {
  1720. previous = current;
  1721. current = current->next;
  1722. }
  1723. /* found an element with same service_id? */
  1724. if (current == NULL) {
  1725. /* not found! append to end (which could also be the end of a zero-length
  1726. * list), don't need to sort (1 is smallest value). */
  1727. /* create elem */
  1728. hs_usage_list_elem_t *e = hs_usage_list_elem_new();
  1729. /* update list attributes (one new elem, one new occurrence) */
  1730. lst->total_count++;
  1731. lst->total_service_ids++;
  1732. /* copy service id to elem */
  1733. strlcpy(e->service_id,service_id,sizeof(e->service_id));
  1734. /* let either l->start or previously last elem point to new elem */
  1735. if (lst->start == NULL) {
  1736. /* this is the first elem */
  1737. lst->start = e;
  1738. } else {
  1739. /* there were elems in the list before */
  1740. previous->next = e;
  1741. }
  1742. } else {
  1743. /* found! add occurrence to elem and consider resorting */
  1744. /* update list attributes (no new elem, but one new occurrence) */
  1745. lst->total_count++;
  1746. /* add occurrence to elem */
  1747. current->count++;
  1748. /* is it another than the first list elem? and has previous elem fewer
  1749. * count than current? then we need to resort */
  1750. if (previous != NULL && previous->count < current->count) {
  1751. /* yes! we need to resort */
  1752. /* remove current elem first */
  1753. previous->next = current->next;
  1754. /* can we prepend elem to all other elements? */
  1755. if (lst->start->count <= current->count) {
  1756. /* yes! prepend elem */
  1757. current->next = lst->start;
  1758. lst->start = current;
  1759. } else {
  1760. /* no! walk through list a second time and insert at correct place */
  1761. hs_usage_list_elem_t *insert_current = lst->start->next;
  1762. hs_usage_list_elem_t *insert_previous = lst->start;
  1763. while (insert_current != NULL &&
  1764. insert_current->count > current->count) {
  1765. insert_previous = insert_current;
  1766. insert_current = insert_current->next;
  1767. }
  1768. /* insert here */
  1769. current->next = insert_current;
  1770. insert_previous->next = current;
  1771. }
  1772. }
  1773. }
  1774. }
  1775. /** Writes the current service-related observations to the history array and
  1776. * clears the observations of the current period. */
  1777. static void
  1778. hs_usage_write_service_related_observations_to_history(
  1779. hs_usage_current_observation_period_t *p,
  1780. hs_usage_service_related_observation_t *h)
  1781. {
  1782. /* walk through the first 20 % of list elements and calculate frequency
  1783. * distributions */
  1784. /* maximum indices for the three frequencies */
  1785. int five_percent_idx = h->list->total_service_ids/20;
  1786. int ten_percent_idx = h->list->total_service_ids/10;
  1787. int twenty_percent_idx = h->list->total_service_ids/5;
  1788. /* temp values */
  1789. uint32_t five_percent = 0;
  1790. uint32_t ten_percent = 0;
  1791. uint32_t twenty_percent = 0;
  1792. /* walk through list */
  1793. hs_usage_list_elem_t *current = h->list->start;
  1794. int i=0;
  1795. while (current != NULL && i <= twenty_percent_idx) {
  1796. twenty_percent += current->count;
  1797. if (i <= ten_percent_idx)
  1798. ten_percent += current->count;
  1799. if (i <= five_percent_idx)
  1800. five_percent += current->count;
  1801. current = current->next;
  1802. i++;
  1803. }
  1804. /* copy frequencies */
  1805. h->twenty[p->next_idx] = twenty_percent;
  1806. h->ten[p->next_idx] = ten_percent;
  1807. h->five[p->next_idx] = five_percent;
  1808. /* copy total number of observations */
  1809. h->totals[p->next_idx] = h->list->total_count;
  1810. /* free memory of old list */
  1811. hs_usage_list_clear(h->list);
  1812. }
  1813. /** Advances to next observation period. */
  1814. static void
  1815. hs_usage_advance_current_observation_period(void)
  1816. {
  1817. /* aggregate observations to history, including frequency distribution
  1818. * arrays */
  1819. hs_usage_write_service_related_observations_to_history(
  1820. current_period, publish_total);
  1821. hs_usage_write_service_related_observations_to_history(
  1822. current_period, publish_novel);
  1823. hs_usage_write_service_related_observations_to_history(
  1824. current_period, fetch_total);
  1825. hs_usage_write_service_related_observations_to_history(
  1826. current_period, fetch_successful);
  1827. /* write current number of descriptors to descs history */
  1828. descs->totals[current_period->next_idx] = rend_cache_size();
  1829. /* advance to next period */
  1830. current_period->next_idx++;
  1831. if (current_period->next_idx == NUM_TOTALS_HS_USAGE)
  1832. current_period->next_idx = 0;
  1833. if (current_period->num_set < NUM_TOTALS_HS_USAGE)
  1834. ++current_period->num_set;
  1835. current_period->start_of_current_period=current_period->start_of_next_period;
  1836. current_period->start_of_next_period += NUM_SECS_HS_USAGE_SUM_INTERVAL;
  1837. }
  1838. /** Checks if the current period is up to date, and if not, advances it. */
  1839. static void
  1840. hs_usage_check_if_current_period_is_up_to_date(time_t now)
  1841. {
  1842. while (now > current_period->start_of_next_period) {
  1843. hs_usage_advance_current_observation_period();
  1844. }
  1845. }
  1846. /** Adds a service-related observation, maybe after advancing to next
  1847. * observation period. */
  1848. static void
  1849. hs_usage_add_service_related_observation(
  1850. hs_usage_service_related_observation_t *h,
  1851. time_t now,
  1852. const char *service_id)
  1853. {
  1854. if (now < current_period->start_of_current_period) {
  1855. /* don't record old data */
  1856. return;
  1857. }
  1858. /* check if we are up-to-date */
  1859. hs_usage_check_if_current_period_is_up_to_date(now);
  1860. /* add observation */
  1861. hs_usage_insert_value(h->list, service_id);
  1862. }
  1863. /** Adds the observation of storing a rendezvous service descriptor to our
  1864. * cache in our role as HS authoritative directory. */
  1865. void
  1866. hs_usage_note_publish_total(const char *service_id, time_t now)
  1867. {
  1868. hs_usage_add_service_related_observation(publish_total, now, service_id);
  1869. }
  1870. /** Adds the observation of storing a novel rendezvous service descriptor to
  1871. * our cache in our role as HS authoritative directory. */
  1872. void
  1873. hs_usage_note_publish_novel(const char *service_id, time_t now)
  1874. {
  1875. hs_usage_add_service_related_observation(publish_novel, now, service_id);
  1876. }
  1877. /** Adds the observation of being requested for a rendezvous service descriptor
  1878. * in our role as HS authoritative directory. */
  1879. void
  1880. hs_usage_note_fetch_total(const char *service_id, time_t now)
  1881. {
  1882. hs_usage_add_service_related_observation(fetch_total, now, service_id);
  1883. }
  1884. /** Adds the observation of being requested for a rendezvous service descriptor
  1885. * in our role as HS authoritative directory and being able to answer that
  1886. * request successfully. */
  1887. void
  1888. hs_usage_note_fetch_successful(const char *service_id, time_t now)
  1889. {
  1890. hs_usage_add_service_related_observation(fetch_successful, now, service_id);
  1891. }
  1892. /** Writes the given circular array to a string. */
  1893. static size_t
  1894. hs_usage_format_history(char *buf, size_t len, uint32_t *data)
  1895. {
  1896. char *cp = buf; /* pointer where we are in the buffer */
  1897. int i, n;
  1898. if (current_period->num_set <= current_period->next_idx) {
  1899. i = 0; /* not been through circular array */
  1900. } else {
  1901. i = current_period->next_idx;
  1902. }
  1903. for (n = 0; n < current_period->num_set; ++n,++i) {
  1904. if (i >= NUM_TOTALS_HS_USAGE)
  1905. i -= NUM_TOTALS_HS_USAGE;
  1906. tor_assert(i < NUM_TOTALS_HS_USAGE);
  1907. if (n == (current_period->num_set-1))
  1908. tor_snprintf(cp, len-(cp-buf), "%d", data[i]);
  1909. else
  1910. tor_snprintf(cp, len-(cp-buf), "%d,", data[i]);
  1911. cp += strlen(cp);
  1912. }
  1913. return cp-buf;
  1914. }
  1915. /** Writes the complete usage history as hidden service authoritative directory
  1916. * to a string. */
  1917. static char *
  1918. hs_usage_format_statistics(void)
  1919. {
  1920. char *buf, *cp, *s = NULL;
  1921. char t[ISO_TIME_LEN+1];
  1922. int r;
  1923. uint32_t *data = NULL;
  1924. size_t len;
  1925. len = (70+20*NUM_TOTALS_HS_USAGE)*11;
  1926. buf = tor_malloc_zero(len);
  1927. cp = buf;
  1928. for (r = 0; r < 11; ++r) {
  1929. switch (r) {
  1930. case 0:
  1931. s = (char*) "publish-total-history";
  1932. data = publish_total->totals;
  1933. break;
  1934. case 1:
  1935. s = (char*) "publish-novel-history";
  1936. data = publish_novel->totals;
  1937. break;
  1938. case 2:
  1939. s = (char*) "publish-top-5-percent-history";
  1940. data = publish_total->five;
  1941. break;
  1942. case 3:
  1943. s = (char*) "publish-top-10-percent-history";
  1944. data = publish_total->ten;
  1945. break;
  1946. case 4:
  1947. s = (char*) "publish-top-20-percent-history";
  1948. data = publish_total->twenty;
  1949. break;
  1950. case 5:
  1951. s = (char*) "fetch-total-history";
  1952. data = fetch_total->totals;
  1953. break;
  1954. case 6:
  1955. s = (char*) "fetch-successful-history";
  1956. data = fetch_successful->totals;
  1957. break;
  1958. case 7:
  1959. s = (char*) "fetch-top-5-percent-history";
  1960. data = fetch_total->five;
  1961. break;
  1962. case 8:
  1963. s = (char*) "fetch-top-10-percent-history";
  1964. data = fetch_total->ten;
  1965. break;
  1966. case 9:
  1967. s = (char*) "fetch-top-20-percent-history";
  1968. data = fetch_total->twenty;
  1969. break;
  1970. case 10:
  1971. s = (char*) "desc-total-history";
  1972. data = descs->totals;
  1973. break;
  1974. }
  1975. format_iso_time(t, current_period->start_of_current_period);
  1976. tor_snprintf(cp, len-(cp-buf), "%s %s (%d s) ", s, t,
  1977. NUM_SECS_HS_USAGE_SUM_INTERVAL);
  1978. cp += strlen(cp);
  1979. cp += hs_usage_format_history(cp, len-(cp-buf), data);
  1980. strlcat(cp, "\n", len-(cp-buf));
  1981. ++cp;
  1982. }
  1983. return buf;
  1984. }
  1985. /** Write current statistics about hidden service usage to file. */
  1986. void
  1987. hs_usage_write_statistics_to_file(time_t now)
  1988. {
  1989. char *buf;
  1990. size_t len;
  1991. char *fname;
  1992. or_options_t *options = get_options();
  1993. /* check if we are up-to-date */
  1994. hs_usage_check_if_current_period_is_up_to_date(now);
  1995. buf = hs_usage_format_statistics();
  1996. len = strlen(options->DataDirectory) + 16;
  1997. fname = tor_malloc(len);
  1998. tor_snprintf(fname, len, "%s"PATH_SEPARATOR"hsusage",
  1999. options->DataDirectory);
  2000. write_str_to_file(fname,buf,0);
  2001. tor_free(buf);
  2002. tor_free(fname);
  2003. }