hibernate.c 34 KB

12345678910111213141516171819202122232425262728293031323334353637383940414243444546474849505152535455565758596061626364656667686970717273747576777879808182838485868788899091929394959697989910010110210310410510610710810911011111211311411511611711811912012112212312412512612712812913013113213313413513613713813914014114214314414514614714814915015115215315415515615715815916016116216316416516616716816917017117217317417517617717817918018118218318418518618718818919019119219319419519619719819920020120220320420520620720820921021121221321421521621721821922022122222322422522622722822923023123223323423523623723823924024124224324424524624724824925025125225325425525625725825926026126226326426526626726826927027127227327427527627727827928028128228328428528628728828929029129229329429529629729829930030130230330430530630730830931031131231331431531631731831932032132232332432532632732832933033133233333433533633733833934034134234334434534634734834935035135235335435535635735835936036136236336436536636736836937037137237337437537637737837938038138238338438538638738838939039139239339439539639739839940040140240340440540640740840941041141241341441541641741841942042142242342442542642742842943043143243343443543643743843944044144244344444544644744844945045145245345445545645745845946046146246346446546646746846947047147247347447547647747847948048148248348448548648748848949049149249349449549649749849950050150250350450550650750850951051151251351451551651751851952052152252352452552652752852953053153253353453553653753853954054154254354454554654754854955055155255355455555655755855956056156256356456556656756856957057157257357457557657757857958058158258358458558658758858959059159259359459559659759859960060160260360460560660760860961061161261361461561661761861962062162262362462562662762862963063163263363463563663763863964064164264364464564664764864965065165265365465565665765865966066166266366466566666766866967067167267367467567667767867968068168268368468568668768868969069169269369469569669769869970070170270370470570670770870971071171271371471571671771871972072172272372472572672772872973073173273373473573673773873974074174274374474574674774874975075175275375475575675775875976076176276376476576676776876977077177277377477577677777877978078178278378478578678778878979079179279379479579679779879980080180280380480580680780880981081181281381481581681781881982082182282382482582682782882983083183283383483583683783883984084184284384484584684784884985085185285385485585685785885986086186286386486586686786886987087187287387487587687787887988088188288388488588688788888989089189289389489589689789889990090190290390490590690790890991091191291391491591691791891992092192292392492592692792892993093193293393493593693793893994094194294394494594694794894995095195295395495595695795895996096196296396496596696796896997097197297397497597697797897998098198298398498598698798898999099199299399499599699799899910001001100210031004100510061007100810091010101110121013101410151016101710181019102010211022
  1. /* Copyright (c) 2004-2006, Roger Dingledine, Nick Mathewson.
  2. * Copyright (c) 2007-2013, The Tor Project, Inc. */
  3. /* See LICENSE for licensing information */
  4. /**
  5. * \file hibernate.c
  6. * \brief Functions to close listeners, stop allowing new circuits,
  7. * etc in preparation for closing down or going dormant; and to track
  8. * bandwidth and time intervals to know when to hibernate and when to
  9. * stop hibernating.
  10. **/
  11. /*
  12. hibernating, phase 1:
  13. - send destroy in response to create cells
  14. - send end (policy failed) in response to begin cells
  15. - close an OR conn when it has no circuits
  16. hibernating, phase 2:
  17. (entered when bandwidth hard limit reached)
  18. - close all OR/AP/exit conns)
  19. */
  20. #define HIBERNATE_PRIVATE
  21. #include "or.h"
  22. #include "channel.h"
  23. #include "channeltls.h"
  24. #include "config.h"
  25. #include "connection.h"
  26. #include "connection_edge.h"
  27. #include "hibernate.h"
  28. #include "main.h"
  29. #include "router.h"
  30. #include "statefile.h"
  31. extern long stats_n_seconds_working; /* published uptime */
  32. /** Are we currently awake, asleep, running out of bandwidth, or shutting
  33. * down? */
  34. static hibernate_state_t hibernate_state = HIBERNATE_STATE_INITIAL;
  35. /** If are hibernating, when do we plan to wake up? Set to 0 if we
  36. * aren't hibernating. */
  37. static time_t hibernate_end_time = 0;
  38. /** If we are shutting down, when do we plan finally exit? Set to 0 if
  39. * we aren't shutting down. */
  40. static time_t shutdown_time = 0;
  41. /** Possible accounting periods. */
  42. typedef enum {
  43. UNIT_MONTH=1, UNIT_WEEK=2, UNIT_DAY=3,
  44. } time_unit_t;
  45. /* Fields for accounting logic. Accounting overview:
  46. *
  47. * Accounting is designed to ensure that no more than N bytes are sent in
  48. * either direction over a given interval (currently, one month, one week, or
  49. * one day) We could
  50. * try to do this by choking our bandwidth to a trickle, but that
  51. * would make our streams useless. Instead, we estimate what our
  52. * bandwidth usage will be, and guess how long we'll be able to
  53. * provide that much bandwidth before hitting our limit. We then
  54. * choose a random time within the accounting interval to come up (so
  55. * that we don't get 50 Tors running on the 1st of the month and none
  56. * on the 30th).
  57. *
  58. * Each interval runs as follows:
  59. *
  60. * 1. We guess our bandwidth usage, based on how much we used
  61. * last time. We choose a "wakeup time" within the interval to come up.
  62. * 2. Until the chosen wakeup time, we hibernate.
  63. * 3. We come up at the wakeup time, and provide bandwidth until we are
  64. * "very close" to running out.
  65. * 4. Then we go into low-bandwidth mode, and stop accepting new
  66. * connections, but provide bandwidth until we run out.
  67. * 5. Then we hibernate until the end of the interval.
  68. *
  69. * If the interval ends before we run out of bandwidth, we go back to
  70. * step one.
  71. */
  72. /** How many bytes have we read in this accounting interval? */
  73. static uint64_t n_bytes_read_in_interval = 0;
  74. /** How many bytes have we written in this accounting interval? */
  75. static uint64_t n_bytes_written_in_interval = 0;
  76. /** How many seconds have we been running this interval? */
  77. static uint32_t n_seconds_active_in_interval = 0;
  78. /** How many seconds were we active in this interval before we hit our soft
  79. * limit? */
  80. static int n_seconds_to_hit_soft_limit = 0;
  81. /** When in this interval was the soft limit hit. */
  82. static time_t soft_limit_hit_at = 0;
  83. /** How many bytes had we read/written when we hit the soft limit? */
  84. static uint64_t n_bytes_at_soft_limit = 0;
  85. /** When did this accounting interval start? */
  86. static time_t interval_start_time = 0;
  87. /** When will this accounting interval end? */
  88. static time_t interval_end_time = 0;
  89. /** How far into the accounting interval should we hibernate? */
  90. static time_t interval_wakeup_time = 0;
  91. /** How much bandwidth do we 'expect' to use per minute? (0 if we have no
  92. * info from the last period.) */
  93. static uint64_t expected_bandwidth_usage = 0;
  94. /** What unit are we using for our accounting? */
  95. static time_unit_t cfg_unit = UNIT_MONTH;
  96. /** How many days,hours,minutes into each unit does our accounting interval
  97. * start? */
  98. /** @{ */
  99. static int cfg_start_day = 0,
  100. cfg_start_hour = 0,
  101. cfg_start_min = 0;
  102. /** @} */
  103. static void reset_accounting(time_t now);
  104. static int read_bandwidth_usage(void);
  105. static time_t start_of_accounting_period_after(time_t now);
  106. static time_t start_of_accounting_period_containing(time_t now);
  107. static void accounting_set_wakeup_time(void);
  108. /* ************
  109. * Functions for bandwidth accounting.
  110. * ************/
  111. /** Configure accounting start/end time settings based on
  112. * options->AccountingStart. Return 0 on success, -1 on failure. If
  113. * <b>validate_only</b> is true, do not change the current settings. */
  114. int
  115. accounting_parse_options(const or_options_t *options, int validate_only)
  116. {
  117. time_unit_t unit;
  118. int ok, idx;
  119. long d,h,m;
  120. smartlist_t *items;
  121. const char *v = options->AccountingStart;
  122. const char *s;
  123. char *cp;
  124. if (!v) {
  125. if (!validate_only) {
  126. cfg_unit = UNIT_MONTH;
  127. cfg_start_day = 1;
  128. cfg_start_hour = 0;
  129. cfg_start_min = 0;
  130. }
  131. return 0;
  132. }
  133. items = smartlist_new();
  134. smartlist_split_string(items, v, NULL,
  135. SPLIT_SKIP_SPACE|SPLIT_IGNORE_BLANK,0);
  136. if (smartlist_len(items)<2) {
  137. log_warn(LD_CONFIG, "Too few arguments to AccountingStart");
  138. goto err;
  139. }
  140. s = smartlist_get(items,0);
  141. if (0==strcasecmp(s, "month")) {
  142. unit = UNIT_MONTH;
  143. } else if (0==strcasecmp(s, "week")) {
  144. unit = UNIT_WEEK;
  145. } else if (0==strcasecmp(s, "day")) {
  146. unit = UNIT_DAY;
  147. } else {
  148. log_warn(LD_CONFIG,
  149. "Unrecognized accounting unit '%s': only 'month', 'week',"
  150. " and 'day' are supported.", s);
  151. goto err;
  152. }
  153. switch (unit) {
  154. case UNIT_WEEK:
  155. d = tor_parse_long(smartlist_get(items,1), 10, 1, 7, &ok, NULL);
  156. if (!ok) {
  157. log_warn(LD_CONFIG, "Weekly accounting must begin on a day between "
  158. "1 (Monday) and 7 (Sunday)");
  159. goto err;
  160. }
  161. break;
  162. case UNIT_MONTH:
  163. d = tor_parse_long(smartlist_get(items,1), 10, 1, 28, &ok, NULL);
  164. if (!ok) {
  165. log_warn(LD_CONFIG, "Monthly accounting must begin on a day between "
  166. "1 and 28");
  167. goto err;
  168. }
  169. break;
  170. case UNIT_DAY:
  171. d = 0;
  172. break;
  173. /* Coverity dislikes unreachable default cases; some compilers warn on
  174. * switch statements missing a case. Tell Coverity not to worry. */
  175. /* coverity[dead_error_begin] */
  176. default:
  177. tor_assert(0);
  178. }
  179. idx = unit==UNIT_DAY?1:2;
  180. if (smartlist_len(items) != (idx+1)) {
  181. log_warn(LD_CONFIG,"Accounting unit '%s' requires %d argument%s.",
  182. s, idx, (idx>1)?"s":"");
  183. goto err;
  184. }
  185. s = smartlist_get(items, idx);
  186. h = tor_parse_long(s, 10, 0, 23, &ok, &cp);
  187. if (!ok) {
  188. log_warn(LD_CONFIG,"Accounting start time not parseable: bad hour.");
  189. goto err;
  190. }
  191. if (!cp || *cp!=':') {
  192. log_warn(LD_CONFIG,
  193. "Accounting start time not parseable: not in HH:MM format");
  194. goto err;
  195. }
  196. m = tor_parse_long(cp+1, 10, 0, 59, &ok, &cp);
  197. if (!ok) {
  198. log_warn(LD_CONFIG, "Accounting start time not parseable: bad minute");
  199. goto err;
  200. }
  201. if (!cp || *cp!='\0') {
  202. log_warn(LD_CONFIG,
  203. "Accounting start time not parseable: not in HH:MM format");
  204. goto err;
  205. }
  206. if (!validate_only) {
  207. cfg_unit = unit;
  208. cfg_start_day = (int)d;
  209. cfg_start_hour = (int)h;
  210. cfg_start_min = (int)m;
  211. }
  212. SMARTLIST_FOREACH(items, char *, item, tor_free(item));
  213. smartlist_free(items);
  214. return 0;
  215. err:
  216. SMARTLIST_FOREACH(items, char *, item, tor_free(item));
  217. smartlist_free(items);
  218. return -1;
  219. }
  220. /** If we want to manage the accounting system and potentially
  221. * hibernate, return 1, else return 0.
  222. */
  223. int
  224. accounting_is_enabled(const or_options_t *options)
  225. {
  226. if (options->AccountingMax)
  227. return 1;
  228. return 0;
  229. }
  230. /** If accounting is enabled, return how long (in seconds) this
  231. * interval lasts. */
  232. int
  233. accounting_get_interval_length(void)
  234. {
  235. return (int)(interval_end_time - interval_start_time);
  236. }
  237. /** Called from main.c to tell us that <b>seconds</b> seconds have
  238. * passed, <b>n_read</b> bytes have been read, and <b>n_written</b>
  239. * bytes have been written. */
  240. void
  241. accounting_add_bytes(size_t n_read, size_t n_written, int seconds)
  242. {
  243. n_bytes_read_in_interval += n_read;
  244. n_bytes_written_in_interval += n_written;
  245. /* If we haven't been called in 10 seconds, we're probably jumping
  246. * around in time. */
  247. n_seconds_active_in_interval += (seconds < 10) ? seconds : 0;
  248. }
  249. /** If get_end, return the end of the accounting period that contains
  250. * the time <b>now</b>. Else, return the start of the accounting
  251. * period that contains the time <b>now</b> */
  252. static time_t
  253. edge_of_accounting_period_containing(time_t now, int get_end)
  254. {
  255. int before;
  256. struct tm tm;
  257. tor_localtime_r(&now, &tm);
  258. /* Set 'before' to true iff the current time is before the hh:mm
  259. * changeover time for today. */
  260. before = tm.tm_hour < cfg_start_hour ||
  261. (tm.tm_hour == cfg_start_hour && tm.tm_min < cfg_start_min);
  262. /* Dispatch by unit. First, find the start day of the given period;
  263. * then, if get_end is true, increment to the end day. */
  264. switch (cfg_unit)
  265. {
  266. case UNIT_MONTH: {
  267. /* If this is before the Nth, we want the Nth of last month. */
  268. if (tm.tm_mday < cfg_start_day ||
  269. (tm.tm_mday == cfg_start_day && before)) {
  270. --tm.tm_mon;
  271. }
  272. /* Otherwise, the month is correct. */
  273. tm.tm_mday = cfg_start_day;
  274. if (get_end)
  275. ++tm.tm_mon;
  276. break;
  277. }
  278. case UNIT_WEEK: {
  279. /* What is the 'target' day of the week in struct tm format? (We
  280. say Sunday==7; struct tm says Sunday==0.) */
  281. int wday = cfg_start_day % 7;
  282. /* How many days do we subtract from today to get to the right day? */
  283. int delta = (7+tm.tm_wday-wday)%7;
  284. /* If we are on the right day, but the changeover hasn't happened yet,
  285. * then subtract a whole week. */
  286. if (delta == 0 && before)
  287. delta = 7;
  288. tm.tm_mday -= delta;
  289. if (get_end)
  290. tm.tm_mday += 7;
  291. break;
  292. }
  293. case UNIT_DAY:
  294. if (before)
  295. --tm.tm_mday;
  296. if (get_end)
  297. ++tm.tm_mday;
  298. break;
  299. default:
  300. tor_assert(0);
  301. }
  302. tm.tm_hour = cfg_start_hour;
  303. tm.tm_min = cfg_start_min;
  304. tm.tm_sec = 0;
  305. tm.tm_isdst = -1; /* Autodetect DST */
  306. return mktime(&tm);
  307. }
  308. /** Return the start of the accounting period containing the time
  309. * <b>now</b>. */
  310. static time_t
  311. start_of_accounting_period_containing(time_t now)
  312. {
  313. return edge_of_accounting_period_containing(now, 0);
  314. }
  315. /** Return the start of the accounting period that comes after the one
  316. * containing the time <b>now</b>. */
  317. static time_t
  318. start_of_accounting_period_after(time_t now)
  319. {
  320. return edge_of_accounting_period_containing(now, 1);
  321. }
  322. /** Return the length of the accounting period containing the time
  323. * <b>now</b>. */
  324. static long
  325. length_of_accounting_period_containing(time_t now)
  326. {
  327. return edge_of_accounting_period_containing(now, 1) -
  328. edge_of_accounting_period_containing(now, 0);
  329. }
  330. /** Initialize the accounting subsystem. */
  331. void
  332. configure_accounting(time_t now)
  333. {
  334. time_t s_now;
  335. /* Try to remember our recorded usage. */
  336. if (!interval_start_time)
  337. read_bandwidth_usage(); /* If we fail, we'll leave values at zero, and
  338. * reset below.*/
  339. s_now = start_of_accounting_period_containing(now);
  340. if (!interval_start_time) {
  341. /* We didn't have recorded usage; Start a new interval. */
  342. log_info(LD_ACCT, "Starting new accounting interval.");
  343. reset_accounting(now);
  344. } else if (s_now == interval_start_time) {
  345. log_info(LD_ACCT, "Continuing accounting interval.");
  346. /* We are in the interval we thought we were in. Do nothing.*/
  347. interval_end_time = start_of_accounting_period_after(interval_start_time);
  348. } else {
  349. long duration =
  350. length_of_accounting_period_containing(interval_start_time);
  351. double delta = ((double)(s_now - interval_start_time)) / duration;
  352. if (-0.50 <= delta && delta <= 0.50) {
  353. /* The start of the period is now a little later or earlier than we
  354. * remembered. That's fine; we might lose some bytes we could otherwise
  355. * have written, but better to err on the side of obeying people's
  356. * accounting settings. */
  357. log_info(LD_ACCT, "Accounting interval moved by %.02f%%; "
  358. "that's fine.", delta*100);
  359. interval_end_time = start_of_accounting_period_after(now);
  360. } else if (delta >= 0.99) {
  361. /* This is the regular time-moved-forward case; don't be too noisy
  362. * about it or people will complain */
  363. log_info(LD_ACCT, "Accounting interval elapsed; starting a new one");
  364. reset_accounting(now);
  365. } else {
  366. log_warn(LD_ACCT,
  367. "Mismatched accounting interval: moved by %.02f%%. "
  368. "Starting a fresh one.", delta*100);
  369. reset_accounting(now);
  370. }
  371. }
  372. accounting_set_wakeup_time();
  373. }
  374. /** Set expected_bandwidth_usage based on how much we sent/received
  375. * per minute last interval (if we were up for at least 30 minutes),
  376. * or based on our declared bandwidth otherwise. */
  377. static void
  378. update_expected_bandwidth(void)
  379. {
  380. uint64_t expected;
  381. const or_options_t *options= get_options();
  382. uint64_t max_configured = (options->RelayBandwidthRate > 0 ?
  383. options->RelayBandwidthRate :
  384. options->BandwidthRate) * 60;
  385. #define MIN_TIME_FOR_MEASUREMENT (1800)
  386. if (soft_limit_hit_at > interval_start_time && n_bytes_at_soft_limit &&
  387. (soft_limit_hit_at - interval_start_time) > MIN_TIME_FOR_MEASUREMENT) {
  388. /* If we hit our soft limit last time, only count the bytes up to that
  389. * time. This is a better predictor of our actual bandwidth than
  390. * considering the entirety of the last interval, since we likely started
  391. * using bytes very slowly once we hit our soft limit. */
  392. expected = n_bytes_at_soft_limit /
  393. (soft_limit_hit_at - interval_start_time);
  394. expected /= 60;
  395. } else if (n_seconds_active_in_interval >= MIN_TIME_FOR_MEASUREMENT) {
  396. /* Otherwise, we either measured enough time in the last interval but
  397. * never hit our soft limit, or we're using a state file from a Tor that
  398. * doesn't know to store soft-limit info. Just take rate at which
  399. * we were reading/writing in the last interval as our expected rate.
  400. */
  401. uint64_t used = MAX(n_bytes_written_in_interval,
  402. n_bytes_read_in_interval);
  403. expected = used / (n_seconds_active_in_interval / 60);
  404. } else {
  405. /* If we haven't gotten enough data last interval, set 'expected'
  406. * to 0. This will set our wakeup to the start of the interval.
  407. * Next interval, we'll choose our starting time based on how much
  408. * we sent this interval.
  409. */
  410. expected = 0;
  411. }
  412. if (expected > max_configured)
  413. expected = max_configured;
  414. expected_bandwidth_usage = expected;
  415. }
  416. /** Called at the start of a new accounting interval: reset our
  417. * expected bandwidth usage based on what happened last time, set up
  418. * the start and end of the interval, and clear byte/time totals.
  419. */
  420. static void
  421. reset_accounting(time_t now)
  422. {
  423. log_info(LD_ACCT, "Starting new accounting interval.");
  424. update_expected_bandwidth();
  425. interval_start_time = start_of_accounting_period_containing(now);
  426. interval_end_time = start_of_accounting_period_after(interval_start_time);
  427. n_bytes_read_in_interval = 0;
  428. n_bytes_written_in_interval = 0;
  429. n_seconds_active_in_interval = 0;
  430. n_bytes_at_soft_limit = 0;
  431. soft_limit_hit_at = 0;
  432. n_seconds_to_hit_soft_limit = 0;
  433. }
  434. /** Return true iff we should save our bandwidth usage to disk. */
  435. static INLINE int
  436. time_to_record_bandwidth_usage(time_t now)
  437. {
  438. /* Note every 600 sec */
  439. #define NOTE_INTERVAL (600)
  440. /* Or every 20 megabytes */
  441. #define NOTE_BYTES 20*(1024*1024)
  442. static uint64_t last_read_bytes_noted = 0;
  443. static uint64_t last_written_bytes_noted = 0;
  444. static time_t last_time_noted = 0;
  445. if (last_time_noted + NOTE_INTERVAL <= now ||
  446. last_read_bytes_noted + NOTE_BYTES <= n_bytes_read_in_interval ||
  447. last_written_bytes_noted + NOTE_BYTES <= n_bytes_written_in_interval ||
  448. (interval_end_time && interval_end_time <= now)) {
  449. last_time_noted = now;
  450. last_read_bytes_noted = n_bytes_read_in_interval;
  451. last_written_bytes_noted = n_bytes_written_in_interval;
  452. return 1;
  453. }
  454. return 0;
  455. }
  456. /** Invoked once per second. Checks whether it is time to hibernate,
  457. * record bandwidth used, etc. */
  458. void
  459. accounting_run_housekeeping(time_t now)
  460. {
  461. if (now >= interval_end_time) {
  462. configure_accounting(now);
  463. }
  464. if (time_to_record_bandwidth_usage(now)) {
  465. if (accounting_record_bandwidth_usage(now, get_or_state())) {
  466. log_warn(LD_FS, "Couldn't record bandwidth usage to disk.");
  467. }
  468. }
  469. }
  470. /** Based on our interval and our estimated bandwidth, choose a
  471. * deterministic (but random-ish) time to wake up. */
  472. static void
  473. accounting_set_wakeup_time(void)
  474. {
  475. char digest[DIGEST_LEN];
  476. crypto_digest_t *d_env;
  477. uint64_t time_to_exhaust_bw;
  478. int time_to_consider;
  479. if (! server_identity_key_is_set()) {
  480. if (init_keys() < 0) {
  481. log_err(LD_BUG, "Error initializing keys");
  482. tor_assert(0);
  483. }
  484. }
  485. if (server_identity_key_is_set()) {
  486. char buf[ISO_TIME_LEN+1];
  487. format_iso_time(buf, interval_start_time);
  488. crypto_pk_get_digest(get_server_identity_key(), digest);
  489. d_env = crypto_digest_new();
  490. crypto_digest_add_bytes(d_env, buf, ISO_TIME_LEN);
  491. crypto_digest_add_bytes(d_env, digest, DIGEST_LEN);
  492. crypto_digest_get_digest(d_env, digest, DIGEST_LEN);
  493. crypto_digest_free(d_env);
  494. } else {
  495. crypto_rand(digest, DIGEST_LEN);
  496. }
  497. if (!expected_bandwidth_usage) {
  498. char buf1[ISO_TIME_LEN+1];
  499. char buf2[ISO_TIME_LEN+1];
  500. format_local_iso_time(buf1, interval_start_time);
  501. format_local_iso_time(buf2, interval_end_time);
  502. interval_wakeup_time = interval_start_time;
  503. log_notice(LD_ACCT,
  504. "Configured hibernation. This interval begins at %s "
  505. "and ends at %s. We have no prior estimate for bandwidth, so "
  506. "we will start out awake and hibernate when we exhaust our quota.",
  507. buf1, buf2);
  508. return;
  509. }
  510. time_to_exhaust_bw =
  511. (get_options()->AccountingMax/expected_bandwidth_usage)*60;
  512. if (time_to_exhaust_bw > INT_MAX) {
  513. time_to_exhaust_bw = INT_MAX;
  514. time_to_consider = 0;
  515. } else {
  516. time_to_consider = accounting_get_interval_length() -
  517. (int)time_to_exhaust_bw;
  518. }
  519. if (time_to_consider<=0) {
  520. interval_wakeup_time = interval_start_time;
  521. } else {
  522. /* XXX can we simplify this just by picking a random (non-deterministic)
  523. * time to be up? If we go down and come up, then we pick a new one. Is
  524. * that good enough? -RD */
  525. /* This is not a perfectly unbiased conversion, but it is good enough:
  526. * in the worst case, the first half of the day is 0.06 percent likelier
  527. * to be chosen than the last half. */
  528. interval_wakeup_time = interval_start_time +
  529. (get_uint32(digest) % time_to_consider);
  530. }
  531. {
  532. char buf1[ISO_TIME_LEN+1];
  533. char buf2[ISO_TIME_LEN+1];
  534. char buf3[ISO_TIME_LEN+1];
  535. char buf4[ISO_TIME_LEN+1];
  536. time_t down_time;
  537. if (interval_wakeup_time+time_to_exhaust_bw > TIME_MAX)
  538. down_time = TIME_MAX;
  539. else
  540. down_time = (time_t)(interval_wakeup_time+time_to_exhaust_bw);
  541. if (down_time>interval_end_time)
  542. down_time = interval_end_time;
  543. format_local_iso_time(buf1, interval_start_time);
  544. format_local_iso_time(buf2, interval_wakeup_time);
  545. format_local_iso_time(buf3, down_time);
  546. format_local_iso_time(buf4, interval_end_time);
  547. log_notice(LD_ACCT,
  548. "Configured hibernation. This interval began at %s; "
  549. "the scheduled wake-up time %s %s; "
  550. "we expect%s to exhaust our quota for this interval around %s; "
  551. "the next interval begins at %s (all times local)",
  552. buf1,
  553. time(NULL)<interval_wakeup_time?"is":"was", buf2,
  554. time(NULL)<down_time?"":"ed", buf3,
  555. buf4);
  556. }
  557. }
  558. /* This rounds 0 up to 1000, but that's actually a feature. */
  559. #define ROUND_UP(x) (((x) + 0x3ff) & ~0x3ff)
  560. /** Save all our bandwidth tracking information to disk. Return 0 on
  561. * success, -1 on failure. */
  562. int
  563. accounting_record_bandwidth_usage(time_t now, or_state_t *state)
  564. {
  565. /* Just update the state */
  566. state->AccountingIntervalStart = interval_start_time;
  567. state->AccountingBytesReadInInterval = ROUND_UP(n_bytes_read_in_interval);
  568. state->AccountingBytesWrittenInInterval =
  569. ROUND_UP(n_bytes_written_in_interval);
  570. state->AccountingSecondsActive = n_seconds_active_in_interval;
  571. state->AccountingExpectedUsage = expected_bandwidth_usage;
  572. state->AccountingSecondsToReachSoftLimit = n_seconds_to_hit_soft_limit;
  573. state->AccountingSoftLimitHitAt = soft_limit_hit_at;
  574. state->AccountingBytesAtSoftLimit = n_bytes_at_soft_limit;
  575. or_state_mark_dirty(state,
  576. now+(get_options()->AvoidDiskWrites ? 7200 : 60));
  577. return 0;
  578. }
  579. #undef ROUND_UP
  580. /** Read stored accounting information from disk. Return 0 on success;
  581. * return -1 and change nothing on failure. */
  582. static int
  583. read_bandwidth_usage(void)
  584. {
  585. or_state_t *state = get_or_state();
  586. {
  587. char *fname = get_datadir_fname("bw_accounting");
  588. unlink(fname);
  589. tor_free(fname);
  590. }
  591. if (!state)
  592. return -1;
  593. log_info(LD_ACCT, "Reading bandwidth accounting data from state file");
  594. n_bytes_read_in_interval = state->AccountingBytesReadInInterval;
  595. n_bytes_written_in_interval = state->AccountingBytesWrittenInInterval;
  596. n_seconds_active_in_interval = state->AccountingSecondsActive;
  597. interval_start_time = state->AccountingIntervalStart;
  598. expected_bandwidth_usage = state->AccountingExpectedUsage;
  599. /* Older versions of Tor (before 0.2.2.17-alpha or so) didn't generate these
  600. * fields. If you switch back and forth, you might get an
  601. * AccountingSoftLimitHitAt value from long before the most recent
  602. * interval_start_time. If that's so, then ignore the softlimit-related
  603. * values. */
  604. if (state->AccountingSoftLimitHitAt > interval_start_time) {
  605. soft_limit_hit_at = state->AccountingSoftLimitHitAt;
  606. n_bytes_at_soft_limit = state->AccountingBytesAtSoftLimit;
  607. n_seconds_to_hit_soft_limit = state->AccountingSecondsToReachSoftLimit;
  608. } else {
  609. soft_limit_hit_at = 0;
  610. n_bytes_at_soft_limit = 0;
  611. n_seconds_to_hit_soft_limit = 0;
  612. }
  613. {
  614. char tbuf1[ISO_TIME_LEN+1];
  615. char tbuf2[ISO_TIME_LEN+1];
  616. format_iso_time(tbuf1, state->LastWritten);
  617. format_iso_time(tbuf2, state->AccountingIntervalStart);
  618. log_info(LD_ACCT,
  619. "Successfully read bandwidth accounting info from state written at %s "
  620. "for interval starting at %s. We have been active for %lu seconds in "
  621. "this interval. At the start of the interval, we expected to use "
  622. "about %lu KB per second. ("U64_FORMAT" bytes read so far, "
  623. U64_FORMAT" bytes written so far)",
  624. tbuf1, tbuf2,
  625. (unsigned long)n_seconds_active_in_interval,
  626. (unsigned long)(expected_bandwidth_usage*1024/60),
  627. U64_PRINTF_ARG(n_bytes_read_in_interval),
  628. U64_PRINTF_ARG(n_bytes_written_in_interval));
  629. }
  630. return 0;
  631. }
  632. /** Return true iff we have sent/received all the bytes we are willing
  633. * to send/receive this interval. */
  634. static int
  635. hibernate_hard_limit_reached(void)
  636. {
  637. uint64_t hard_limit = get_options()->AccountingMax;
  638. if (!hard_limit)
  639. return 0;
  640. return n_bytes_read_in_interval >= hard_limit
  641. || n_bytes_written_in_interval >= hard_limit;
  642. }
  643. /** Return true iff we have sent/received almost all the bytes we are willing
  644. * to send/receive this interval. */
  645. static int
  646. hibernate_soft_limit_reached(void)
  647. {
  648. const uint64_t acct_max = get_options()->AccountingMax;
  649. #define SOFT_LIM_PCT (.95)
  650. #define SOFT_LIM_BYTES (500*1024*1024)
  651. #define SOFT_LIM_MINUTES (3*60)
  652. /* The 'soft limit' is a fair bit more complicated now than once it was.
  653. * We want to stop accepting connections when ALL of the following are true:
  654. * - We expect to use up the remaining bytes in under 3 hours
  655. * - We have used up 95% of our bytes.
  656. * - We have less than 500MB of bytes left.
  657. */
  658. uint64_t soft_limit = DBL_TO_U64(U64_TO_DBL(acct_max) * SOFT_LIM_PCT);
  659. if (acct_max > SOFT_LIM_BYTES && acct_max - SOFT_LIM_BYTES > soft_limit) {
  660. soft_limit = acct_max - SOFT_LIM_BYTES;
  661. }
  662. if (expected_bandwidth_usage) {
  663. const uint64_t expected_usage =
  664. expected_bandwidth_usage * SOFT_LIM_MINUTES;
  665. if (acct_max > expected_usage && acct_max - expected_usage > soft_limit)
  666. soft_limit = acct_max - expected_usage;
  667. }
  668. if (!soft_limit)
  669. return 0;
  670. return n_bytes_read_in_interval >= soft_limit
  671. || n_bytes_written_in_interval >= soft_limit;
  672. }
  673. /** Called when we get a SIGINT, or when bandwidth soft limit is
  674. * reached. Puts us into "loose hibernation": we don't accept new
  675. * connections, but we continue handling old ones. */
  676. static void
  677. hibernate_begin(hibernate_state_t new_state, time_t now)
  678. {
  679. const or_options_t *options = get_options();
  680. if (new_state == HIBERNATE_STATE_EXITING &&
  681. hibernate_state != HIBERNATE_STATE_LIVE) {
  682. log_notice(LD_GENERAL,"SIGINT received %s; exiting now.",
  683. hibernate_state == HIBERNATE_STATE_EXITING ?
  684. "a second time" : "while hibernating");
  685. tor_cleanup();
  686. exit(0);
  687. }
  688. if (new_state == HIBERNATE_STATE_LOWBANDWIDTH &&
  689. hibernate_state == HIBERNATE_STATE_LIVE) {
  690. soft_limit_hit_at = now;
  691. n_seconds_to_hit_soft_limit = n_seconds_active_in_interval;
  692. n_bytes_at_soft_limit = MAX(n_bytes_read_in_interval,
  693. n_bytes_written_in_interval);
  694. }
  695. /* close listeners. leave control listener(s). */
  696. connection_mark_all_noncontrol_listeners();
  697. /* XXX kill intro point circs */
  698. /* XXX upload rendezvous service descriptors with no intro points */
  699. if (new_state == HIBERNATE_STATE_EXITING) {
  700. log_notice(LD_GENERAL,"Interrupt: we have stopped accepting new "
  701. "connections, and will shut down in %d seconds. Interrupt "
  702. "again to exit now.", options->ShutdownWaitLength);
  703. shutdown_time = time(NULL) + options->ShutdownWaitLength;
  704. } else { /* soft limit reached */
  705. hibernate_end_time = interval_end_time;
  706. }
  707. hibernate_state = new_state;
  708. accounting_record_bandwidth_usage(now, get_or_state());
  709. or_state_mark_dirty(get_or_state(),
  710. get_options()->AvoidDiskWrites ? now+600 : 0);
  711. }
  712. /** Called when we've been hibernating and our timeout is reached. */
  713. static void
  714. hibernate_end(hibernate_state_t new_state)
  715. {
  716. tor_assert(hibernate_state == HIBERNATE_STATE_LOWBANDWIDTH ||
  717. hibernate_state == HIBERNATE_STATE_DORMANT ||
  718. hibernate_state == HIBERNATE_STATE_INITIAL);
  719. /* listeners will be relaunched in run_scheduled_events() in main.c */
  720. if (hibernate_state != HIBERNATE_STATE_INITIAL)
  721. log_notice(LD_ACCT,"Hibernation period ended. Resuming normal activity.");
  722. hibernate_state = new_state;
  723. hibernate_end_time = 0; /* no longer hibernating */
  724. stats_n_seconds_working = 0; /* reset published uptime */
  725. }
  726. /** A wrapper around hibernate_begin, for when we get SIGINT. */
  727. void
  728. hibernate_begin_shutdown(void)
  729. {
  730. hibernate_begin(HIBERNATE_STATE_EXITING, time(NULL));
  731. }
  732. /** Return true iff we are currently hibernating. */
  733. int
  734. we_are_hibernating(void)
  735. {
  736. return hibernate_state != HIBERNATE_STATE_LIVE;
  737. }
  738. /** If we aren't currently dormant, close all connections and become
  739. * dormant. */
  740. static void
  741. hibernate_go_dormant(time_t now)
  742. {
  743. connection_t *conn;
  744. if (hibernate_state == HIBERNATE_STATE_DORMANT)
  745. return;
  746. else if (hibernate_state == HIBERNATE_STATE_LOWBANDWIDTH)
  747. hibernate_state = HIBERNATE_STATE_DORMANT;
  748. else
  749. hibernate_begin(HIBERNATE_STATE_DORMANT, now);
  750. log_notice(LD_ACCT,"Going dormant. Blowing away remaining connections.");
  751. /* Close all OR/AP/exit conns. Leave dir conns because we still want
  752. * to be able to upload server descriptors so people know we're still
  753. * running, and download directories so we can detect if we're obsolete.
  754. * Leave control conns because we still want to be controllable.
  755. */
  756. while ((conn = connection_get_by_type(CONN_TYPE_OR)) ||
  757. (conn = connection_get_by_type(CONN_TYPE_AP)) ||
  758. (conn = connection_get_by_type(CONN_TYPE_EXIT))) {
  759. if (CONN_IS_EDGE(conn))
  760. connection_edge_end(TO_EDGE_CONN(conn), END_STREAM_REASON_HIBERNATING);
  761. log_info(LD_NET,"Closing conn type %d", conn->type);
  762. if (conn->type == CONN_TYPE_AP) /* send socks failure if needed */
  763. connection_mark_unattached_ap(TO_ENTRY_CONN(conn),
  764. END_STREAM_REASON_HIBERNATING);
  765. else if (conn->type == CONN_TYPE_OR) {
  766. if (TO_OR_CONN(conn)->chan) {
  767. channel_mark_for_close(TLS_CHAN_TO_BASE(TO_OR_CONN(conn)->chan));
  768. } else {
  769. connection_mark_for_close(conn);
  770. }
  771. } else
  772. connection_mark_for_close(conn);
  773. }
  774. if (now < interval_wakeup_time)
  775. hibernate_end_time = interval_wakeup_time;
  776. else
  777. hibernate_end_time = interval_end_time;
  778. accounting_record_bandwidth_usage(now, get_or_state());
  779. or_state_mark_dirty(get_or_state(),
  780. get_options()->AvoidDiskWrites ? now+600 : 0);
  781. }
  782. /** Called when hibernate_end_time has arrived. */
  783. static void
  784. hibernate_end_time_elapsed(time_t now)
  785. {
  786. char buf[ISO_TIME_LEN+1];
  787. /* The interval has ended, or it is wakeup time. Find out which. */
  788. accounting_run_housekeeping(now);
  789. if (interval_wakeup_time <= now) {
  790. /* The interval hasn't changed, but interval_wakeup_time has passed.
  791. * It's time to wake up and start being a server. */
  792. hibernate_end(HIBERNATE_STATE_LIVE);
  793. return;
  794. } else {
  795. /* The interval has changed, and it isn't time to wake up yet. */
  796. hibernate_end_time = interval_wakeup_time;
  797. format_iso_time(buf,interval_wakeup_time);
  798. if (hibernate_state != HIBERNATE_STATE_DORMANT) {
  799. /* We weren't sleeping before; we should sleep now. */
  800. log_notice(LD_ACCT,
  801. "Accounting period ended. Commencing hibernation until "
  802. "%s UTC", buf);
  803. hibernate_go_dormant(now);
  804. } else {
  805. log_notice(LD_ACCT,
  806. "Accounting period ended. This period, we will hibernate"
  807. " until %s UTC",buf);
  808. }
  809. }
  810. }
  811. /** Consider our environment and decide if it's time
  812. * to start/stop hibernating.
  813. */
  814. void
  815. consider_hibernation(time_t now)
  816. {
  817. int accounting_enabled = get_options()->AccountingMax != 0;
  818. char buf[ISO_TIME_LEN+1];
  819. /* If we're in 'exiting' mode, then we just shut down after the interval
  820. * elapses. */
  821. if (hibernate_state == HIBERNATE_STATE_EXITING) {
  822. tor_assert(shutdown_time);
  823. if (shutdown_time <= now) {
  824. log_notice(LD_GENERAL, "Clean shutdown finished. Exiting.");
  825. tor_cleanup();
  826. exit(0);
  827. }
  828. return; /* if exiting soon, don't worry about bandwidth limits */
  829. }
  830. if (hibernate_state == HIBERNATE_STATE_DORMANT) {
  831. /* We've been hibernating because of bandwidth accounting. */
  832. tor_assert(hibernate_end_time);
  833. if (hibernate_end_time > now && accounting_enabled) {
  834. /* If we're hibernating, don't wake up until it's time, regardless of
  835. * whether we're in a new interval. */
  836. return ;
  837. } else {
  838. hibernate_end_time_elapsed(now);
  839. }
  840. }
  841. /* Else, we aren't hibernating. See if it's time to start hibernating, or to
  842. * go dormant. */
  843. if (hibernate_state == HIBERNATE_STATE_LIVE ||
  844. hibernate_state == HIBERNATE_STATE_INITIAL) {
  845. if (hibernate_soft_limit_reached()) {
  846. log_notice(LD_ACCT,
  847. "Bandwidth soft limit reached; commencing hibernation. "
  848. "No new connections will be accepted");
  849. hibernate_begin(HIBERNATE_STATE_LOWBANDWIDTH, now);
  850. } else if (accounting_enabled && now < interval_wakeup_time) {
  851. format_local_iso_time(buf,interval_wakeup_time);
  852. log_notice(LD_ACCT,
  853. "Commencing hibernation. We will wake up at %s local time.",
  854. buf);
  855. hibernate_go_dormant(now);
  856. } else if (hibernate_state == HIBERNATE_STATE_INITIAL) {
  857. hibernate_end(HIBERNATE_STATE_LIVE);
  858. }
  859. }
  860. if (hibernate_state == HIBERNATE_STATE_LOWBANDWIDTH) {
  861. if (!accounting_enabled) {
  862. hibernate_end_time_elapsed(now);
  863. } else if (hibernate_hard_limit_reached()) {
  864. hibernate_go_dormant(now);
  865. } else if (hibernate_end_time <= now) {
  866. /* The hibernation period ended while we were still in lowbandwidth.*/
  867. hibernate_end_time_elapsed(now);
  868. }
  869. }
  870. }
  871. /** Helper function: called when we get a GETINFO request for an
  872. * accounting-related key on the control connection <b>conn</b>. If we can
  873. * answer the request for <b>question</b>, then set *<b>answer</b> to a newly
  874. * allocated string holding the result. Otherwise, set *<b>answer</b> to
  875. * NULL. */
  876. int
  877. getinfo_helper_accounting(control_connection_t *conn,
  878. const char *question, char **answer,
  879. const char **errmsg)
  880. {
  881. (void) conn;
  882. (void) errmsg;
  883. if (!strcmp(question, "accounting/enabled")) {
  884. *answer = tor_strdup(accounting_is_enabled(get_options()) ? "1" : "0");
  885. } else if (!strcmp(question, "accounting/hibernating")) {
  886. if (hibernate_state == HIBERNATE_STATE_DORMANT)
  887. *answer = tor_strdup("hard");
  888. else if (hibernate_state == HIBERNATE_STATE_LOWBANDWIDTH)
  889. *answer = tor_strdup("soft");
  890. else
  891. *answer = tor_strdup("awake");
  892. } else if (!strcmp(question, "accounting/bytes")) {
  893. tor_asprintf(answer, U64_FORMAT" "U64_FORMAT,
  894. U64_PRINTF_ARG(n_bytes_read_in_interval),
  895. U64_PRINTF_ARG(n_bytes_written_in_interval));
  896. } else if (!strcmp(question, "accounting/bytes-left")) {
  897. uint64_t limit = get_options()->AccountingMax;
  898. uint64_t read_left = 0, write_left = 0;
  899. if (n_bytes_read_in_interval < limit)
  900. read_left = limit - n_bytes_read_in_interval;
  901. if (n_bytes_written_in_interval < limit)
  902. write_left = limit - n_bytes_written_in_interval;
  903. tor_asprintf(answer, U64_FORMAT" "U64_FORMAT,
  904. U64_PRINTF_ARG(read_left), U64_PRINTF_ARG(write_left));
  905. } else if (!strcmp(question, "accounting/interval-start")) {
  906. *answer = tor_malloc(ISO_TIME_LEN+1);
  907. format_iso_time(*answer, interval_start_time);
  908. } else if (!strcmp(question, "accounting/interval-wake")) {
  909. *answer = tor_malloc(ISO_TIME_LEN+1);
  910. format_iso_time(*answer, interval_wakeup_time);
  911. } else if (!strcmp(question, "accounting/interval-end")) {
  912. *answer = tor_malloc(ISO_TIME_LEN+1);
  913. format_iso_time(*answer, interval_end_time);
  914. } else {
  915. *answer = NULL;
  916. }
  917. return 0;
  918. }
  919. /**
  920. * Manually change the hibernation state. Private; used only by the unit
  921. * tests.
  922. */
  923. void
  924. hibernate_set_state_for_testing_(hibernate_state_t newstate)
  925. {
  926. hibernate_state = newstate;
  927. }