hibernate.c 34 KB

1234567891011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859606162636465666768697071727374757677787980818283848586878889909192939495969798991001011021031041051061071081091101111121131141151161171181191201211221231241251261271281291301311321331341351361371381391401411421431441451461471481491501511521531541551561571581591601611621631641651661671681691701711721731741751761771781791801811821831841851861871881891901911921931941951961971981992002012022032042052062072082092102112122132142152162172182192202212222232242252262272282292302312322332342352362372382392402412422432442452462472482492502512522532542552562572582592602612622632642652662672682692702712722732742752762772782792802812822832842852862872882892902912922932942952962972982993003013023033043053063073083093103113123133143153163173183193203213223233243253263273283293303313323333343353363373383393403413423433443453463473483493503513523533543553563573583593603613623633643653663673683693703713723733743753763773783793803813823833843853863873883893903913923933943953963973983994004014024034044054064074084094104114124134144154164174184194204214224234244254264274284294304314324334344354364374384394404414424434444454464474484494504514524534544554564574584594604614624634644654664674684694704714724734744754764774784794804814824834844854864874884894904914924934944954964974984995005015025035045055065075085095105115125135145155165175185195205215225235245255265275285295305315325335345355365375385395405415425435445455465475485495505515525535545555565575585595605615625635645655665675685695705715725735745755765775785795805815825835845855865875885895905915925935945955965975985996006016026036046056066076086096106116126136146156166176186196206216226236246256266276286296306316326336346356366376386396406416426436446456466476486496506516526536546556566576586596606616626636646656666676686696706716726736746756766776786796806816826836846856866876886896906916926936946956966976986997007017027037047057067077087097107117127137147157167177187197207217227237247257267277287297307317327337347357367377387397407417427437447457467477487497507517527537547557567577587597607617627637647657667677687697707717727737747757767777787797807817827837847857867877887897907917927937947957967977987998008018028038048058068078088098108118128138148158168178188198208218228238248258268278288298308318328338348358368378388398408418428438448458468478488498508518528538548558568578588598608618628638648658668678688698708718728738748758768778788798808818828838848858868878888898908918928938948958968978988999009019029039049059069079089099109119129139149159169179189199209219229239249259269279289299309319329339349359369379389399409419429439449459469479489499509519529539549559569579589599609619629639649659669679689699709719729739749759769779789799809819829839849859869879889899909919929939949959969979989991000100110021003100410051006100710081009101010111012101310141015101610171018
  1. /* Copyright (c) 2004-2006, Roger Dingledine, Nick Mathewson.
  2. * Copyright (c) 2007-2012, 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 "config.h"
  23. #include "connection.h"
  24. #include "connection_edge.h"
  25. #include "hibernate.h"
  26. #include "main.h"
  27. #include "router.h"
  28. #include "statefile.h"
  29. extern long stats_n_seconds_working; /* published uptime */
  30. /** Are we currently awake, asleep, running out of bandwidth, or shutting
  31. * down? */
  32. static hibernate_state_t hibernate_state = HIBERNATE_STATE_INITIAL;
  33. /** If are hibernating, when do we plan to wake up? Set to 0 if we
  34. * aren't hibernating. */
  35. static time_t hibernate_end_time = 0;
  36. /** If we are shutting down, when do we plan finally exit? Set to 0 if
  37. * we aren't shutting down. */
  38. static time_t shutdown_time = 0;
  39. /** Possible accounting periods. */
  40. typedef enum {
  41. UNIT_MONTH=1, UNIT_WEEK=2, UNIT_DAY=3,
  42. } time_unit_t;
  43. /* Fields for accounting logic. Accounting overview:
  44. *
  45. * Accounting is designed to ensure that no more than N bytes are sent in
  46. * either direction over a given interval (currently, one month, one week, or
  47. * one day) We could
  48. * try to do this by choking our bandwidth to a trickle, but that
  49. * would make our streams useless. Instead, we estimate what our
  50. * bandwidth usage will be, and guess how long we'll be able to
  51. * provide that much bandwidth before hitting our limit. We then
  52. * choose a random time within the accounting interval to come up (so
  53. * that we don't get 50 Tors running on the 1st of the month and none
  54. * on the 30th).
  55. *
  56. * Each interval runs as follows:
  57. *
  58. * 1. We guess our bandwidth usage, based on how much we used
  59. * last time. We choose a "wakeup time" within the interval to come up.
  60. * 2. Until the chosen wakeup time, we hibernate.
  61. * 3. We come up at the wakeup time, and provide bandwidth until we are
  62. * "very close" to running out.
  63. * 4. Then we go into low-bandwidth mode, and stop accepting new
  64. * connections, but provide bandwidth until we run out.
  65. * 5. Then we hibernate until the end of the interval.
  66. *
  67. * If the interval ends before we run out of bandwidth, we go back to
  68. * step one.
  69. */
  70. /** How many bytes have we read in this accounting interval? */
  71. static uint64_t n_bytes_read_in_interval = 0;
  72. /** How many bytes have we written in this accounting interval? */
  73. static uint64_t n_bytes_written_in_interval = 0;
  74. /** How many seconds have we been running this interval? */
  75. static uint32_t n_seconds_active_in_interval = 0;
  76. /** How many seconds were we active in this interval before we hit our soft
  77. * limit? */
  78. static int n_seconds_to_hit_soft_limit = 0;
  79. /** When in this interval was the soft limit hit. */
  80. static time_t soft_limit_hit_at = 0;
  81. /** How many bytes had we read/written when we hit the soft limit? */
  82. static uint64_t n_bytes_at_soft_limit = 0;
  83. /** When did this accounting interval start? */
  84. static time_t interval_start_time = 0;
  85. /** When will this accounting interval end? */
  86. static time_t interval_end_time = 0;
  87. /** How far into the accounting interval should we hibernate? */
  88. static time_t interval_wakeup_time = 0;
  89. /** How much bandwidth do we 'expect' to use per minute? (0 if we have no
  90. * info from the last period.) */
  91. static uint64_t expected_bandwidth_usage = 0;
  92. /** What unit are we using for our accounting? */
  93. static time_unit_t cfg_unit = UNIT_MONTH;
  94. /** How many days,hours,minutes into each unit does our accounting interval
  95. * start? */
  96. /** @{ */
  97. static int cfg_start_day = 0,
  98. cfg_start_hour = 0,
  99. cfg_start_min = 0;
  100. /** @} */
  101. static void reset_accounting(time_t now);
  102. static int read_bandwidth_usage(void);
  103. static time_t start_of_accounting_period_after(time_t now);
  104. static time_t start_of_accounting_period_containing(time_t now);
  105. static void accounting_set_wakeup_time(void);
  106. /* ************
  107. * Functions for bandwidth accounting.
  108. * ************/
  109. /** Configure accounting start/end time settings based on
  110. * options->AccountingStart. Return 0 on success, -1 on failure. If
  111. * <b>validate_only</b> is true, do not change the current settings. */
  112. int
  113. accounting_parse_options(const or_options_t *options, int validate_only)
  114. {
  115. time_unit_t unit;
  116. int ok, idx;
  117. long d,h,m;
  118. smartlist_t *items;
  119. const char *v = options->AccountingStart;
  120. const char *s;
  121. char *cp;
  122. if (!v) {
  123. if (!validate_only) {
  124. cfg_unit = UNIT_MONTH;
  125. cfg_start_day = 1;
  126. cfg_start_hour = 0;
  127. cfg_start_min = 0;
  128. }
  129. return 0;
  130. }
  131. items = smartlist_new();
  132. smartlist_split_string(items, v, NULL,
  133. SPLIT_SKIP_SPACE|SPLIT_IGNORE_BLANK,0);
  134. if (smartlist_len(items)<2) {
  135. log_warn(LD_CONFIG, "Too few arguments to AccountingStart");
  136. goto err;
  137. }
  138. s = smartlist_get(items,0);
  139. if (0==strcasecmp(s, "month")) {
  140. unit = UNIT_MONTH;
  141. } else if (0==strcasecmp(s, "week")) {
  142. unit = UNIT_WEEK;
  143. } else if (0==strcasecmp(s, "day")) {
  144. unit = UNIT_DAY;
  145. } else {
  146. log_warn(LD_CONFIG,
  147. "Unrecognized accounting unit '%s': only 'month', 'week',"
  148. " and 'day' are supported.", s);
  149. goto err;
  150. }
  151. switch (unit) {
  152. case UNIT_WEEK:
  153. d = tor_parse_long(smartlist_get(items,1), 10, 1, 7, &ok, NULL);
  154. if (!ok) {
  155. log_warn(LD_CONFIG, "Weekly accounting must begin on a day between "
  156. "1 (Monday) and 7 (Sunday)");
  157. goto err;
  158. }
  159. break;
  160. case UNIT_MONTH:
  161. d = tor_parse_long(smartlist_get(items,1), 10, 1, 28, &ok, NULL);
  162. if (!ok) {
  163. log_warn(LD_CONFIG, "Monthly accounting must begin on a day between "
  164. "1 and 28");
  165. goto err;
  166. }
  167. break;
  168. case UNIT_DAY:
  169. d = 0;
  170. break;
  171. /* Coverity dislikes unreachable default cases; some compilers warn on
  172. * switch statements missing a case. Tell Coverity not to worry. */
  173. /* coverity[dead_error_begin] */
  174. default:
  175. tor_assert(0);
  176. }
  177. idx = unit==UNIT_DAY?1:2;
  178. if (smartlist_len(items) != (idx+1)) {
  179. log_warn(LD_CONFIG,"Accounting unit '%s' requires %d argument%s.",
  180. s, idx, (idx>1)?"s":"");
  181. goto err;
  182. }
  183. s = smartlist_get(items, idx);
  184. h = tor_parse_long(s, 10, 0, 23, &ok, &cp);
  185. if (!ok) {
  186. log_warn(LD_CONFIG,"Accounting start time not parseable: bad hour.");
  187. goto err;
  188. }
  189. if (!cp || *cp!=':') {
  190. log_warn(LD_CONFIG,
  191. "Accounting start time not parseable: not in HH:MM format");
  192. goto err;
  193. }
  194. m = tor_parse_long(cp+1, 10, 0, 59, &ok, &cp);
  195. if (!ok) {
  196. log_warn(LD_CONFIG, "Accounting start time not parseable: bad minute");
  197. goto err;
  198. }
  199. if (!cp || *cp!='\0') {
  200. log_warn(LD_CONFIG,
  201. "Accounting start time not parseable: not in HH:MM format");
  202. goto err;
  203. }
  204. if (!validate_only) {
  205. cfg_unit = unit;
  206. cfg_start_day = (int)d;
  207. cfg_start_hour = (int)h;
  208. cfg_start_min = (int)m;
  209. }
  210. SMARTLIST_FOREACH(items, char *, item, tor_free(item));
  211. smartlist_free(items);
  212. return 0;
  213. err:
  214. SMARTLIST_FOREACH(items, char *, item, tor_free(item));
  215. smartlist_free(items);
  216. return -1;
  217. }
  218. /** If we want to manage the accounting system and potentially
  219. * hibernate, return 1, else return 0.
  220. */
  221. int
  222. accounting_is_enabled(const or_options_t *options)
  223. {
  224. if (options->AccountingMax)
  225. return 1;
  226. return 0;
  227. }
  228. /** If accounting is enabled, return how long (in seconds) this
  229. * interval lasts. */
  230. int
  231. accounting_get_interval_length(void)
  232. {
  233. return (int)(interval_end_time - interval_start_time);
  234. }
  235. /** Called from main.c to tell us that <b>seconds</b> seconds have
  236. * passed, <b>n_read</b> bytes have been read, and <b>n_written</b>
  237. * bytes have been written. */
  238. void
  239. accounting_add_bytes(size_t n_read, size_t n_written, int seconds)
  240. {
  241. n_bytes_read_in_interval += n_read;
  242. n_bytes_written_in_interval += n_written;
  243. /* If we haven't been called in 10 seconds, we're probably jumping
  244. * around in time. */
  245. n_seconds_active_in_interval += (seconds < 10) ? seconds : 0;
  246. }
  247. /** If get_end, return the end of the accounting period that contains
  248. * the time <b>now</b>. Else, return the start of the accounting
  249. * period that contains the time <b>now</b> */
  250. static time_t
  251. edge_of_accounting_period_containing(time_t now, int get_end)
  252. {
  253. int before;
  254. struct tm tm;
  255. tor_localtime_r(&now, &tm);
  256. /* Set 'before' to true iff the current time is before the hh:mm
  257. * changeover time for today. */
  258. before = tm.tm_hour < cfg_start_hour ||
  259. (tm.tm_hour == cfg_start_hour && tm.tm_min < cfg_start_min);
  260. /* Dispatch by unit. First, find the start day of the given period;
  261. * then, if get_end is true, increment to the end day. */
  262. switch (cfg_unit)
  263. {
  264. case UNIT_MONTH: {
  265. /* If this is before the Nth, we want the Nth of last month. */
  266. if (tm.tm_mday < cfg_start_day ||
  267. (tm.tm_mday < cfg_start_day && before)) {
  268. --tm.tm_mon;
  269. }
  270. /* Otherwise, the month is correct. */
  271. tm.tm_mday = cfg_start_day;
  272. if (get_end)
  273. ++tm.tm_mon;
  274. break;
  275. }
  276. case UNIT_WEEK: {
  277. /* What is the 'target' day of the week in struct tm format? (We
  278. say Sunday==7; struct tm says Sunday==0.) */
  279. int wday = cfg_start_day % 7;
  280. /* How many days do we subtract from today to get to the right day? */
  281. int delta = (7+tm.tm_wday-wday)%7;
  282. /* If we are on the right day, but the changeover hasn't happened yet,
  283. * then subtract a whole week. */
  284. if (delta == 0 && before)
  285. delta = 7;
  286. tm.tm_mday -= delta;
  287. if (get_end)
  288. tm.tm_mday += 7;
  289. break;
  290. }
  291. case UNIT_DAY:
  292. if (before)
  293. --tm.tm_mday;
  294. if (get_end)
  295. ++tm.tm_mday;
  296. break;
  297. default:
  298. tor_assert(0);
  299. }
  300. tm.tm_hour = cfg_start_hour;
  301. tm.tm_min = cfg_start_min;
  302. tm.tm_sec = 0;
  303. tm.tm_isdst = -1; /* Autodetect DST */
  304. return mktime(&tm);
  305. }
  306. /** Return the start of the accounting period containing the time
  307. * <b>now</b>. */
  308. static time_t
  309. start_of_accounting_period_containing(time_t now)
  310. {
  311. return edge_of_accounting_period_containing(now, 0);
  312. }
  313. /** Return the start of the accounting period that comes after the one
  314. * containing the time <b>now</b>. */
  315. static time_t
  316. start_of_accounting_period_after(time_t now)
  317. {
  318. return edge_of_accounting_period_containing(now, 1);
  319. }
  320. /** Return the length of the accounting period containing the time
  321. * <b>now</b>. */
  322. static long
  323. length_of_accounting_period_containing(time_t now)
  324. {
  325. return edge_of_accounting_period_containing(now, 1) -
  326. edge_of_accounting_period_containing(now, 0);
  327. }
  328. /** Initialize the accounting subsystem. */
  329. void
  330. configure_accounting(time_t now)
  331. {
  332. time_t s_now;
  333. /* Try to remember our recorded usage. */
  334. if (!interval_start_time)
  335. read_bandwidth_usage(); /* If we fail, we'll leave values at zero, and
  336. * reset below.*/
  337. s_now = start_of_accounting_period_containing(now);
  338. if (!interval_start_time) {
  339. /* We didn't have recorded usage; Start a new interval. */
  340. log_info(LD_ACCT, "Starting new accounting interval.");
  341. reset_accounting(now);
  342. } else if (s_now == interval_start_time) {
  343. log_info(LD_ACCT, "Continuing accounting interval.");
  344. /* We are in the interval we thought we were in. Do nothing.*/
  345. interval_end_time = start_of_accounting_period_after(interval_start_time);
  346. } else {
  347. long duration =
  348. length_of_accounting_period_containing(interval_start_time);
  349. double delta = ((double)(s_now - interval_start_time)) / duration;
  350. if (-0.50 <= delta && delta <= 0.50) {
  351. /* The start of the period is now a little later or earlier than we
  352. * remembered. That's fine; we might lose some bytes we could otherwise
  353. * have written, but better to err on the side of obeying people's
  354. * accounting settings. */
  355. log_info(LD_ACCT, "Accounting interval moved by %.02f%%; "
  356. "that's fine.", delta*100);
  357. interval_end_time = start_of_accounting_period_after(now);
  358. } else if (delta >= 0.99) {
  359. /* This is the regular time-moved-forward case; don't be too noisy
  360. * about it or people will complain */
  361. log_info(LD_ACCT, "Accounting interval elapsed; starting a new one");
  362. reset_accounting(now);
  363. } else {
  364. log_warn(LD_ACCT,
  365. "Mismatched accounting interval: moved by %.02f%%. "
  366. "Starting a fresh one.", delta*100);
  367. reset_accounting(now);
  368. }
  369. }
  370. accounting_set_wakeup_time();
  371. }
  372. /** Set expected_bandwidth_usage based on how much we sent/received
  373. * per minute last interval (if we were up for at least 30 minutes),
  374. * or based on our declared bandwidth otherwise. */
  375. static void
  376. update_expected_bandwidth(void)
  377. {
  378. uint64_t expected;
  379. const or_options_t *options= get_options();
  380. uint64_t max_configured = (options->RelayBandwidthRate > 0 ?
  381. options->RelayBandwidthRate :
  382. options->BandwidthRate) * 60;
  383. #define MIN_TIME_FOR_MEASUREMENT (1800)
  384. if (soft_limit_hit_at > interval_start_time && n_bytes_at_soft_limit &&
  385. (soft_limit_hit_at - interval_start_time) > MIN_TIME_FOR_MEASUREMENT) {
  386. /* If we hit our soft limit last time, only count the bytes up to that
  387. * time. This is a better predictor of our actual bandwidth than
  388. * considering the entirety of the last interval, since we likely started
  389. * using bytes very slowly once we hit our soft limit. */
  390. expected = n_bytes_at_soft_limit /
  391. (soft_limit_hit_at - interval_start_time);
  392. expected /= 60;
  393. } else if (n_seconds_active_in_interval >= MIN_TIME_FOR_MEASUREMENT) {
  394. /* Otherwise, we either measured enough time in the last interval but
  395. * never hit our soft limit, or we're using a state file from a Tor that
  396. * doesn't know to store soft-limit info. Just take rate at which
  397. * we were reading/writing in the last interval as our expected rate.
  398. */
  399. uint64_t used = MAX(n_bytes_written_in_interval,
  400. n_bytes_read_in_interval);
  401. expected = used / (n_seconds_active_in_interval / 60);
  402. } else {
  403. /* If we haven't gotten enough data last interval, set 'expected'
  404. * to 0. This will set our wakeup to the start of the interval.
  405. * Next interval, we'll choose our starting time based on how much
  406. * we sent this interval.
  407. */
  408. expected = 0;
  409. }
  410. if (expected > max_configured)
  411. expected = max_configured;
  412. expected_bandwidth_usage = expected;
  413. }
  414. /** Called at the start of a new accounting interval: reset our
  415. * expected bandwidth usage based on what happened last time, set up
  416. * the start and end of the interval, and clear byte/time totals.
  417. */
  418. static void
  419. reset_accounting(time_t now)
  420. {
  421. log_info(LD_ACCT, "Starting new accounting interval.");
  422. update_expected_bandwidth();
  423. interval_start_time = start_of_accounting_period_containing(now);
  424. interval_end_time = start_of_accounting_period_after(interval_start_time);
  425. n_bytes_read_in_interval = 0;
  426. n_bytes_written_in_interval = 0;
  427. n_seconds_active_in_interval = 0;
  428. n_bytes_at_soft_limit = 0;
  429. soft_limit_hit_at = 0;
  430. n_seconds_to_hit_soft_limit = 0;
  431. }
  432. /** Return true iff we should save our bandwidth usage to disk. */
  433. static INLINE int
  434. time_to_record_bandwidth_usage(time_t now)
  435. {
  436. /* Note every 600 sec */
  437. #define NOTE_INTERVAL (600)
  438. /* Or every 20 megabytes */
  439. #define NOTE_BYTES 20*(1024*1024)
  440. static uint64_t last_read_bytes_noted = 0;
  441. static uint64_t last_written_bytes_noted = 0;
  442. static time_t last_time_noted = 0;
  443. if (last_time_noted + NOTE_INTERVAL <= now ||
  444. last_read_bytes_noted + NOTE_BYTES <= n_bytes_read_in_interval ||
  445. last_written_bytes_noted + NOTE_BYTES <= n_bytes_written_in_interval ||
  446. (interval_end_time && interval_end_time <= now)) {
  447. last_time_noted = now;
  448. last_read_bytes_noted = n_bytes_read_in_interval;
  449. last_written_bytes_noted = n_bytes_written_in_interval;
  450. return 1;
  451. }
  452. return 0;
  453. }
  454. /** Invoked once per second. Checks whether it is time to hibernate,
  455. * record bandwidth used, etc. */
  456. void
  457. accounting_run_housekeeping(time_t now)
  458. {
  459. if (now >= interval_end_time) {
  460. configure_accounting(now);
  461. }
  462. if (time_to_record_bandwidth_usage(now)) {
  463. if (accounting_record_bandwidth_usage(now, get_or_state())) {
  464. log_warn(LD_FS, "Couldn't record bandwidth usage to disk.");
  465. }
  466. }
  467. }
  468. /** When we have no idea how fast we are, how long do we assume it will take
  469. * us to exhaust our bandwidth? */
  470. #define GUESS_TIME_TO_USE_BANDWIDTH (24*60*60)
  471. /** Based on our interval and our estimated bandwidth, choose a
  472. * deterministic (but random-ish) time to wake up. */
  473. static void
  474. accounting_set_wakeup_time(void)
  475. {
  476. char digest[DIGEST_LEN];
  477. crypto_digest_t *d_env;
  478. uint64_t time_to_exhaust_bw;
  479. int time_to_consider;
  480. if (! server_identity_key_is_set()) {
  481. if (init_keys() < 0) {
  482. log_err(LD_BUG, "Error initializing keys");
  483. tor_assert(0);
  484. }
  485. }
  486. if (server_identity_key_is_set()) {
  487. char buf[ISO_TIME_LEN+1];
  488. format_iso_time(buf, interval_start_time);
  489. crypto_pk_get_digest(get_server_identity_key(), digest);
  490. d_env = crypto_digest_new();
  491. crypto_digest_add_bytes(d_env, buf, ISO_TIME_LEN);
  492. crypto_digest_add_bytes(d_env, digest, DIGEST_LEN);
  493. crypto_digest_get_digest(d_env, digest, DIGEST_LEN);
  494. crypto_digest_free(d_env);
  495. } else {
  496. crypto_rand(digest, DIGEST_LEN);
  497. }
  498. if (!expected_bandwidth_usage) {
  499. char buf1[ISO_TIME_LEN+1];
  500. char buf2[ISO_TIME_LEN+1];
  501. format_local_iso_time(buf1, interval_start_time);
  502. format_local_iso_time(buf2, interval_end_time);
  503. interval_wakeup_time = interval_start_time;
  504. log_notice(LD_ACCT,
  505. "Configured hibernation. This interval begins at %s "
  506. "and ends at %s. We have no prior estimate for bandwidth, so "
  507. "we will start out awake and hibernate when we exhaust our quota.",
  508. buf1, buf2);
  509. return;
  510. }
  511. time_to_exhaust_bw =
  512. (get_options()->AccountingMax/expected_bandwidth_usage)*60;
  513. if (time_to_exhaust_bw > INT_MAX) {
  514. time_to_exhaust_bw = INT_MAX;
  515. time_to_consider = 0;
  516. } else {
  517. time_to_consider = accounting_get_interval_length() -
  518. (int)time_to_exhaust_bw;
  519. }
  520. if (time_to_consider<=0) {
  521. interval_wakeup_time = interval_start_time;
  522. } else {
  523. /* XXX can we simplify this just by picking a random (non-deterministic)
  524. * time to be up? If we go down and come up, then we pick a new one. Is
  525. * that good enough? -RD */
  526. /* This is not a perfectly unbiased conversion, but it is good enough:
  527. * in the worst case, the first half of the day is 0.06 percent likelier
  528. * to be chosen than the last half. */
  529. interval_wakeup_time = interval_start_time +
  530. (get_uint32(digest) % time_to_consider);
  531. }
  532. {
  533. char buf1[ISO_TIME_LEN+1];
  534. char buf2[ISO_TIME_LEN+1];
  535. char buf3[ISO_TIME_LEN+1];
  536. char buf4[ISO_TIME_LEN+1];
  537. time_t down_time;
  538. if (interval_wakeup_time+time_to_exhaust_bw > TIME_MAX)
  539. down_time = TIME_MAX;
  540. else
  541. down_time = (time_t)(interval_wakeup_time+time_to_exhaust_bw);
  542. if (down_time>interval_end_time)
  543. down_time = interval_end_time;
  544. format_local_iso_time(buf1, interval_start_time);
  545. format_local_iso_time(buf2, interval_wakeup_time);
  546. format_local_iso_time(buf3, down_time);
  547. format_local_iso_time(buf4, interval_end_time);
  548. log_notice(LD_ACCT,
  549. "Configured hibernation. This interval began at %s; "
  550. "the scheduled wake-up time %s %s; "
  551. "we expect%s to exhaust our quota for this interval around %s; "
  552. "the next interval begins at %s (all times local)",
  553. buf1,
  554. time(NULL)<interval_wakeup_time?"is":"was", buf2,
  555. time(NULL)<down_time?"":"ed", buf3,
  556. buf4);
  557. }
  558. }
  559. /* This rounds 0 up to 1000, but that's actually a feature. */
  560. #define ROUND_UP(x) (((x) + 0x3ff) & ~0x3ff)
  561. /** Save all our bandwidth tracking information to disk. Return 0 on
  562. * success, -1 on failure. */
  563. int
  564. accounting_record_bandwidth_usage(time_t now, or_state_t *state)
  565. {
  566. /* Just update the state */
  567. state->AccountingIntervalStart = interval_start_time;
  568. state->AccountingBytesReadInInterval = ROUND_UP(n_bytes_read_in_interval);
  569. state->AccountingBytesWrittenInInterval =
  570. ROUND_UP(n_bytes_written_in_interval);
  571. state->AccountingSecondsActive = n_seconds_active_in_interval;
  572. state->AccountingExpectedUsage = expected_bandwidth_usage;
  573. state->AccountingSecondsToReachSoftLimit = n_seconds_to_hit_soft_limit;
  574. state->AccountingSoftLimitHitAt = soft_limit_hit_at;
  575. state->AccountingBytesAtSoftLimit = n_bytes_at_soft_limit;
  576. or_state_mark_dirty(state,
  577. now+(get_options()->AvoidDiskWrites ? 7200 : 60));
  578. return 0;
  579. }
  580. #undef ROUND_UP
  581. /** Read stored accounting information from disk. Return 0 on success;
  582. * return -1 and change nothing on failure. */
  583. static int
  584. read_bandwidth_usage(void)
  585. {
  586. or_state_t *state = get_or_state();
  587. {
  588. char *fname = get_datadir_fname("bw_accounting");
  589. unlink(fname);
  590. tor_free(fname);
  591. }
  592. if (!state)
  593. return -1;
  594. log_info(LD_ACCT, "Reading bandwidth accounting data from state file");
  595. n_bytes_read_in_interval = state->AccountingBytesReadInInterval;
  596. n_bytes_written_in_interval = state->AccountingBytesWrittenInInterval;
  597. n_seconds_active_in_interval = state->AccountingSecondsActive;
  598. interval_start_time = state->AccountingIntervalStart;
  599. expected_bandwidth_usage = state->AccountingExpectedUsage;
  600. /* Older versions of Tor (before 0.2.2.17-alpha or so) didn't generate these
  601. * fields. If you switch back and forth, you might get an
  602. * AccountingSoftLimitHitAt value from long before the most recent
  603. * interval_start_time. If that's so, then ignore the softlimit-related
  604. * values. */
  605. if (state->AccountingSoftLimitHitAt > interval_start_time) {
  606. soft_limit_hit_at = state->AccountingSoftLimitHitAt;
  607. n_bytes_at_soft_limit = state->AccountingBytesAtSoftLimit;
  608. n_seconds_to_hit_soft_limit = state->AccountingSecondsToReachSoftLimit;
  609. } else {
  610. soft_limit_hit_at = 0;
  611. n_bytes_at_soft_limit = 0;
  612. n_seconds_to_hit_soft_limit = 0;
  613. }
  614. {
  615. char tbuf1[ISO_TIME_LEN+1];
  616. char tbuf2[ISO_TIME_LEN+1];
  617. format_iso_time(tbuf1, state->LastWritten);
  618. format_iso_time(tbuf2, state->AccountingIntervalStart);
  619. log_info(LD_ACCT,
  620. "Successfully read bandwidth accounting info from state written at %s "
  621. "for interval starting at %s. We have been active for %lu seconds in "
  622. "this interval. At the start of the interval, we expected to use "
  623. "about %lu KB per second. ("U64_FORMAT" bytes read so far, "
  624. U64_FORMAT" bytes written so far)",
  625. tbuf1, tbuf2,
  626. (unsigned long)n_seconds_active_in_interval,
  627. (unsigned long)(expected_bandwidth_usage*1024/60),
  628. U64_PRINTF_ARG(n_bytes_read_in_interval),
  629. U64_PRINTF_ARG(n_bytes_written_in_interval));
  630. }
  631. return 0;
  632. }
  633. /** Return true iff we have sent/received all the bytes we are willing
  634. * to send/receive this interval. */
  635. static int
  636. hibernate_hard_limit_reached(void)
  637. {
  638. uint64_t hard_limit = get_options()->AccountingMax;
  639. if (!hard_limit)
  640. return 0;
  641. return n_bytes_read_in_interval >= hard_limit
  642. || n_bytes_written_in_interval >= hard_limit;
  643. }
  644. /** Return true iff we have sent/received almost all the bytes we are willing
  645. * to send/receive this interval. */
  646. static int
  647. hibernate_soft_limit_reached(void)
  648. {
  649. const uint64_t acct_max = get_options()->AccountingMax;
  650. #define SOFT_LIM_PCT (.95)
  651. #define SOFT_LIM_BYTES (500*1024*1024)
  652. #define SOFT_LIM_MINUTES (3*60)
  653. /* The 'soft limit' is a fair bit more complicated now than once it was.
  654. * We want to stop accepting connections when ALL of the following are true:
  655. * - We expect to use up the remaining bytes in under 3 hours
  656. * - We have used up 95% of our bytes.
  657. * - We have less than 500MB of bytes left.
  658. */
  659. uint64_t soft_limit = DBL_TO_U64(U64_TO_DBL(acct_max) * SOFT_LIM_PCT);
  660. if (acct_max > SOFT_LIM_BYTES && acct_max - SOFT_LIM_BYTES > soft_limit) {
  661. soft_limit = acct_max - SOFT_LIM_BYTES;
  662. }
  663. if (expected_bandwidth_usage) {
  664. const uint64_t expected_usage =
  665. expected_bandwidth_usage * SOFT_LIM_MINUTES;
  666. if (acct_max > expected_usage && acct_max - expected_usage > soft_limit)
  667. soft_limit = acct_max - expected_usage;
  668. }
  669. if (!soft_limit)
  670. return 0;
  671. return n_bytes_read_in_interval >= soft_limit
  672. || n_bytes_written_in_interval >= soft_limit;
  673. }
  674. /** Called when we get a SIGINT, or when bandwidth soft limit is
  675. * reached. Puts us into "loose hibernation": we don't accept new
  676. * connections, but we continue handling old ones. */
  677. static void
  678. hibernate_begin(hibernate_state_t new_state, time_t now)
  679. {
  680. const or_options_t *options = get_options();
  681. if (new_state == HIBERNATE_STATE_EXITING &&
  682. hibernate_state != HIBERNATE_STATE_LIVE) {
  683. log_notice(LD_GENERAL,"SIGINT received %s; exiting now.",
  684. hibernate_state == HIBERNATE_STATE_EXITING ?
  685. "a second time" : "while hibernating");
  686. tor_cleanup();
  687. exit(0);
  688. }
  689. if (new_state == HIBERNATE_STATE_LOWBANDWIDTH &&
  690. hibernate_state == HIBERNATE_STATE_LIVE) {
  691. soft_limit_hit_at = now;
  692. n_seconds_to_hit_soft_limit = n_seconds_active_in_interval;
  693. n_bytes_at_soft_limit = MAX(n_bytes_read_in_interval,
  694. n_bytes_written_in_interval);
  695. }
  696. /* close listeners. leave control listener(s). */
  697. connection_mark_all_noncontrol_listeners();
  698. /* XXX kill intro point circs */
  699. /* XXX upload rendezvous service descriptors with no intro points */
  700. if (new_state == HIBERNATE_STATE_EXITING) {
  701. log_notice(LD_GENERAL,"Interrupt: we have stopped accepting new "
  702. "connections, and will shut down in %d seconds. Interrupt "
  703. "again to exit now.", options->ShutdownWaitLength);
  704. shutdown_time = time(NULL) + options->ShutdownWaitLength;
  705. } else { /* soft limit reached */
  706. hibernate_end_time = interval_end_time;
  707. }
  708. hibernate_state = new_state;
  709. accounting_record_bandwidth_usage(now, get_or_state());
  710. or_state_mark_dirty(get_or_state(),
  711. get_options()->AvoidDiskWrites ? now+600 : 0);
  712. }
  713. /** Called when we've been hibernating and our timeout is reached. */
  714. static void
  715. hibernate_end(hibernate_state_t new_state)
  716. {
  717. tor_assert(hibernate_state == HIBERNATE_STATE_LOWBANDWIDTH ||
  718. hibernate_state == HIBERNATE_STATE_DORMANT ||
  719. hibernate_state == HIBERNATE_STATE_INITIAL);
  720. /* listeners will be relaunched in run_scheduled_events() in main.c */
  721. if (hibernate_state != HIBERNATE_STATE_INITIAL)
  722. log_notice(LD_ACCT,"Hibernation period ended. Resuming normal activity.");
  723. hibernate_state = new_state;
  724. hibernate_end_time = 0; /* no longer hibernating */
  725. stats_n_seconds_working = 0; /* reset published uptime */
  726. }
  727. /** A wrapper around hibernate_begin, for when we get SIGINT. */
  728. void
  729. hibernate_begin_shutdown(void)
  730. {
  731. hibernate_begin(HIBERNATE_STATE_EXITING, time(NULL));
  732. }
  733. /** Return true iff we are currently hibernating. */
  734. int
  735. we_are_hibernating(void)
  736. {
  737. return hibernate_state != HIBERNATE_STATE_LIVE;
  738. }
  739. /** If we aren't currently dormant, close all connections and become
  740. * dormant. */
  741. static void
  742. hibernate_go_dormant(time_t now)
  743. {
  744. connection_t *conn;
  745. if (hibernate_state == HIBERNATE_STATE_DORMANT)
  746. return;
  747. else if (hibernate_state == HIBERNATE_STATE_LOWBANDWIDTH)
  748. hibernate_state = HIBERNATE_STATE_DORMANT;
  749. else
  750. hibernate_begin(HIBERNATE_STATE_DORMANT, now);
  751. log_notice(LD_ACCT,"Going dormant. Blowing away remaining connections.");
  752. /* Close all OR/AP/exit conns. Leave dir conns because we still want
  753. * to be able to upload server descriptors so people know we're still
  754. * running, and download directories so we can detect if we're obsolete.
  755. * Leave control conns because we still want to be controllable.
  756. */
  757. while ((conn = connection_get_by_type(CONN_TYPE_OR)) ||
  758. (conn = connection_get_by_type(CONN_TYPE_AP)) ||
  759. (conn = connection_get_by_type(CONN_TYPE_EXIT))) {
  760. if (CONN_IS_EDGE(conn))
  761. connection_edge_end(TO_EDGE_CONN(conn), END_STREAM_REASON_HIBERNATING);
  762. log_info(LD_NET,"Closing conn type %d", conn->type);
  763. if (conn->type == CONN_TYPE_AP) /* send socks failure if needed */
  764. connection_mark_unattached_ap(TO_ENTRY_CONN(conn),
  765. END_STREAM_REASON_HIBERNATING);
  766. else
  767. connection_mark_for_close(conn);
  768. }
  769. if (now < interval_wakeup_time)
  770. hibernate_end_time = interval_wakeup_time;
  771. else
  772. hibernate_end_time = interval_end_time;
  773. accounting_record_bandwidth_usage(now, get_or_state());
  774. or_state_mark_dirty(get_or_state(),
  775. get_options()->AvoidDiskWrites ? now+600 : 0);
  776. }
  777. /** Called when hibernate_end_time has arrived. */
  778. static void
  779. hibernate_end_time_elapsed(time_t now)
  780. {
  781. char buf[ISO_TIME_LEN+1];
  782. /* The interval has ended, or it is wakeup time. Find out which. */
  783. accounting_run_housekeeping(now);
  784. if (interval_wakeup_time <= now) {
  785. /* The interval hasn't changed, but interval_wakeup_time has passed.
  786. * It's time to wake up and start being a server. */
  787. hibernate_end(HIBERNATE_STATE_LIVE);
  788. return;
  789. } else {
  790. /* The interval has changed, and it isn't time to wake up yet. */
  791. hibernate_end_time = interval_wakeup_time;
  792. format_iso_time(buf,interval_wakeup_time);
  793. if (hibernate_state != HIBERNATE_STATE_DORMANT) {
  794. /* We weren't sleeping before; we should sleep now. */
  795. log_notice(LD_ACCT,
  796. "Accounting period ended. Commencing hibernation until "
  797. "%s GMT", buf);
  798. hibernate_go_dormant(now);
  799. } else {
  800. log_notice(LD_ACCT,
  801. "Accounting period ended. This period, we will hibernate"
  802. " until %s GMT",buf);
  803. }
  804. }
  805. }
  806. /** Consider our environment and decide if it's time
  807. * to start/stop hibernating.
  808. */
  809. void
  810. consider_hibernation(time_t now)
  811. {
  812. int accounting_enabled = get_options()->AccountingMax != 0;
  813. char buf[ISO_TIME_LEN+1];
  814. /* If we're in 'exiting' mode, then we just shut down after the interval
  815. * elapses. */
  816. if (hibernate_state == HIBERNATE_STATE_EXITING) {
  817. tor_assert(shutdown_time);
  818. if (shutdown_time <= now) {
  819. log_notice(LD_GENERAL, "Clean shutdown finished. Exiting.");
  820. tor_cleanup();
  821. exit(0);
  822. }
  823. return; /* if exiting soon, don't worry about bandwidth limits */
  824. }
  825. if (hibernate_state == HIBERNATE_STATE_DORMANT) {
  826. /* We've been hibernating because of bandwidth accounting. */
  827. tor_assert(hibernate_end_time);
  828. if (hibernate_end_time > now && accounting_enabled) {
  829. /* If we're hibernating, don't wake up until it's time, regardless of
  830. * whether we're in a new interval. */
  831. return ;
  832. } else {
  833. hibernate_end_time_elapsed(now);
  834. }
  835. }
  836. /* Else, we aren't hibernating. See if it's time to start hibernating, or to
  837. * go dormant. */
  838. if (hibernate_state == HIBERNATE_STATE_LIVE ||
  839. hibernate_state == HIBERNATE_STATE_INITIAL) {
  840. if (hibernate_soft_limit_reached()) {
  841. log_notice(LD_ACCT,
  842. "Bandwidth soft limit reached; commencing hibernation. "
  843. "No new connections will be accepted");
  844. hibernate_begin(HIBERNATE_STATE_LOWBANDWIDTH, now);
  845. } else if (accounting_enabled && now < interval_wakeup_time) {
  846. format_local_iso_time(buf,interval_wakeup_time);
  847. log_notice(LD_ACCT,
  848. "Commencing hibernation. We will wake up at %s local time.",
  849. buf);
  850. hibernate_go_dormant(now);
  851. } else if (hibernate_state == HIBERNATE_STATE_INITIAL) {
  852. hibernate_end(HIBERNATE_STATE_LIVE);
  853. }
  854. }
  855. if (hibernate_state == HIBERNATE_STATE_LOWBANDWIDTH) {
  856. if (!accounting_enabled) {
  857. hibernate_end_time_elapsed(now);
  858. } else if (hibernate_hard_limit_reached()) {
  859. hibernate_go_dormant(now);
  860. } else if (hibernate_end_time <= now) {
  861. /* The hibernation period ended while we were still in lowbandwidth.*/
  862. hibernate_end_time_elapsed(now);
  863. }
  864. }
  865. }
  866. /** Helper function: called when we get a GETINFO request for an
  867. * accounting-related key on the control connection <b>conn</b>. If we can
  868. * answer the request for <b>question</b>, then set *<b>answer</b> to a newly
  869. * allocated string holding the result. Otherwise, set *<b>answer</b> to
  870. * NULL. */
  871. int
  872. getinfo_helper_accounting(control_connection_t *conn,
  873. const char *question, char **answer,
  874. const char **errmsg)
  875. {
  876. (void) conn;
  877. (void) errmsg;
  878. if (!strcmp(question, "accounting/enabled")) {
  879. *answer = tor_strdup(accounting_is_enabled(get_options()) ? "1" : "0");
  880. } else if (!strcmp(question, "accounting/hibernating")) {
  881. if (hibernate_state == HIBERNATE_STATE_DORMANT)
  882. *answer = tor_strdup("hard");
  883. else if (hibernate_state == HIBERNATE_STATE_LOWBANDWIDTH)
  884. *answer = tor_strdup("soft");
  885. else
  886. *answer = tor_strdup("awake");
  887. } else if (!strcmp(question, "accounting/bytes")) {
  888. tor_asprintf(answer, U64_FORMAT" "U64_FORMAT,
  889. U64_PRINTF_ARG(n_bytes_read_in_interval),
  890. U64_PRINTF_ARG(n_bytes_written_in_interval));
  891. } else if (!strcmp(question, "accounting/bytes-left")) {
  892. uint64_t limit = get_options()->AccountingMax;
  893. uint64_t read_left = 0, write_left = 0;
  894. if (n_bytes_read_in_interval < limit)
  895. read_left = limit - n_bytes_read_in_interval;
  896. if (n_bytes_written_in_interval < limit)
  897. write_left = limit - n_bytes_written_in_interval;
  898. tor_asprintf(answer, U64_FORMAT" "U64_FORMAT,
  899. U64_PRINTF_ARG(read_left), U64_PRINTF_ARG(write_left));
  900. } else if (!strcmp(question, "accounting/interval-start")) {
  901. *answer = tor_malloc(ISO_TIME_LEN+1);
  902. format_iso_time(*answer, interval_start_time);
  903. } else if (!strcmp(question, "accounting/interval-wake")) {
  904. *answer = tor_malloc(ISO_TIME_LEN+1);
  905. format_iso_time(*answer, interval_wakeup_time);
  906. } else if (!strcmp(question, "accounting/interval-end")) {
  907. *answer = tor_malloc(ISO_TIME_LEN+1);
  908. format_iso_time(*answer, interval_end_time);
  909. } else {
  910. *answer = NULL;
  911. }
  912. return 0;
  913. }
  914. /**
  915. * Manually change the hibernation state. Private; used only by the unit
  916. * tests.
  917. */
  918. void
  919. hibernate_set_state_for_testing_(hibernate_state_t newstate)
  920. {
  921. hibernate_state = newstate;
  922. }