mainloop.c 86 KB

12345678910111213141516171819202122232425262728293031323334353637383940414243444546474849505152535455565758596061626364656667686970717273747576777879808182838485868788899091929394959697989910010110210310410510610710810911011111211311411511611711811912012112212312412512612712812913013113213313413513613713813914014114214314414514614714814915015115215315415515615715815916016116216316416516616716816917017117217317417517617717817918018118218318418518618718818919019119219319419519619719819920020120220320420520620720820921021121221321421521621721821922022122222322422522622722822923023123223323423523623723823924024124224324424524624724824925025125225325425525625725825926026126226326426526626726826927027127227327427527627727827928028128228328428528628728828929029129229329429529629729829930030130230330430530630730830931031131231331431531631731831932032132232332432532632732832933033133233333433533633733833934034134234334434534634734834935035135235335435535635735835936036136236336436536636736836937037137237337437537637737837938038138238338438538638738838939039139239339439539639739839940040140240340440540640740840941041141241341441541641741841942042142242342442542642742842943043143243343443543643743843944044144244344444544644744844945045145245345445545645745845946046146246346446546646746846947047147247347447547647747847948048148248348448548648748848949049149249349449549649749849950050150250350450550650750850951051151251351451551651751851952052152252352452552652752852953053153253353453553653753853954054154254354454554654754854955055155255355455555655755855956056156256356456556656756856957057157257357457557657757857958058158258358458558658758858959059159259359459559659759859960060160260360460560660760860961061161261361461561661761861962062162262362462562662762862963063163263363463563663763863964064164264364464564664764864965065165265365465565665765865966066166266366466566666766866967067167267367467567667767867968068168268368468568668768868969069169269369469569669769869970070170270370470570670770870971071171271371471571671771871972072172272372472572672772872973073173273373473573673773873974074174274374474574674774874975075175275375475575675775875976076176276376476576676776876977077177277377477577677777877978078178278378478578678778878979079179279379479579679779879980080180280380480580680780880981081181281381481581681781881982082182282382482582682782882983083183283383483583683783883984084184284384484584684784884985085185285385485585685785885986086186286386486586686786886987087187287387487587687787887988088188288388488588688788888989089189289389489589689789889990090190290390490590690790890991091191291391491591691791891992092192292392492592692792892993093193293393493593693793893994094194294394494594694794894995095195295395495595695795895996096196296396496596696796896997097197297397497597697797897998098198298398498598698798898999099199299399499599699799899910001001100210031004100510061007100810091010101110121013101410151016101710181019102010211022102310241025102610271028102910301031103210331034103510361037103810391040104110421043104410451046104710481049105010511052105310541055105610571058105910601061106210631064106510661067106810691070107110721073107410751076107710781079108010811082108310841085108610871088108910901091109210931094109510961097109810991100110111021103110411051106110711081109111011111112111311141115111611171118111911201121112211231124112511261127112811291130113111321133113411351136113711381139114011411142114311441145114611471148114911501151115211531154115511561157115811591160116111621163116411651166116711681169117011711172117311741175117611771178117911801181118211831184118511861187118811891190119111921193119411951196119711981199120012011202120312041205120612071208120912101211121212131214121512161217121812191220122112221223122412251226122712281229123012311232123312341235123612371238123912401241124212431244124512461247124812491250125112521253125412551256125712581259126012611262126312641265126612671268126912701271127212731274127512761277127812791280128112821283128412851286128712881289129012911292129312941295129612971298129913001301130213031304130513061307130813091310131113121313131413151316131713181319132013211322132313241325132613271328132913301331133213331334133513361337133813391340134113421343134413451346134713481349135013511352135313541355135613571358135913601361136213631364136513661367136813691370137113721373137413751376137713781379138013811382138313841385138613871388138913901391139213931394139513961397139813991400140114021403140414051406140714081409141014111412141314141415141614171418141914201421142214231424142514261427142814291430143114321433143414351436143714381439144014411442144314441445144614471448144914501451145214531454145514561457145814591460146114621463146414651466146714681469147014711472147314741475147614771478147914801481148214831484148514861487148814891490149114921493149414951496149714981499150015011502150315041505150615071508150915101511151215131514151515161517151815191520152115221523152415251526152715281529153015311532153315341535153615371538153915401541154215431544154515461547154815491550155115521553155415551556155715581559156015611562156315641565156615671568156915701571157215731574157515761577157815791580158115821583158415851586158715881589159015911592159315941595159615971598159916001601160216031604160516061607160816091610161116121613161416151616161716181619162016211622162316241625162616271628162916301631163216331634163516361637163816391640164116421643164416451646164716481649165016511652165316541655165616571658165916601661166216631664166516661667166816691670167116721673167416751676167716781679168016811682168316841685168616871688168916901691169216931694169516961697169816991700170117021703170417051706170717081709171017111712171317141715171617171718171917201721172217231724172517261727172817291730173117321733173417351736173717381739174017411742174317441745174617471748174917501751175217531754175517561757175817591760176117621763176417651766176717681769177017711772177317741775177617771778177917801781178217831784178517861787178817891790179117921793179417951796179717981799180018011802180318041805180618071808180918101811181218131814181518161817181818191820182118221823182418251826182718281829183018311832183318341835183618371838183918401841184218431844184518461847184818491850185118521853185418551856185718581859186018611862186318641865186618671868186918701871187218731874187518761877187818791880188118821883188418851886188718881889189018911892189318941895189618971898189919001901190219031904190519061907190819091910191119121913191419151916191719181919192019211922192319241925192619271928192919301931193219331934193519361937193819391940194119421943194419451946194719481949195019511952195319541955195619571958195919601961196219631964196519661967196819691970197119721973197419751976197719781979198019811982198319841985198619871988198919901991199219931994199519961997199819992000200120022003200420052006200720082009201020112012201320142015201620172018201920202021202220232024202520262027202820292030203120322033203420352036203720382039204020412042204320442045204620472048204920502051205220532054205520562057205820592060206120622063206420652066206720682069207020712072207320742075207620772078207920802081208220832084208520862087208820892090209120922093209420952096209720982099210021012102210321042105210621072108210921102111211221132114211521162117211821192120212121222123212421252126212721282129213021312132213321342135213621372138213921402141214221432144214521462147214821492150215121522153215421552156215721582159216021612162216321642165216621672168216921702171217221732174217521762177217821792180218121822183218421852186218721882189219021912192219321942195219621972198219922002201220222032204220522062207220822092210221122122213221422152216221722182219222022212222222322242225222622272228222922302231223222332234223522362237223822392240224122422243224422452246224722482249225022512252225322542255225622572258225922602261226222632264226522662267226822692270227122722273227422752276227722782279228022812282228322842285228622872288228922902291229222932294229522962297229822992300230123022303230423052306230723082309231023112312231323142315231623172318231923202321232223232324232523262327232823292330233123322333233423352336233723382339234023412342234323442345234623472348234923502351235223532354235523562357235823592360236123622363236423652366236723682369237023712372237323742375237623772378237923802381238223832384238523862387238823892390239123922393239423952396239723982399240024012402240324042405240624072408240924102411241224132414241524162417241824192420242124222423242424252426242724282429243024312432243324342435243624372438243924402441244224432444244524462447244824492450245124522453245424552456245724582459246024612462246324642465246624672468246924702471247224732474247524762477247824792480248124822483248424852486248724882489249024912492249324942495249624972498249925002501250225032504250525062507250825092510251125122513251425152516251725182519252025212522252325242525252625272528252925302531253225332534253525362537253825392540254125422543254425452546254725482549255025512552255325542555255625572558255925602561256225632564256525662567256825692570257125722573257425752576257725782579258025812582258325842585258625872588258925902591259225932594259525962597259825992600260126022603260426052606260726082609261026112612261326142615261626172618261926202621262226232624262526262627262826292630263126322633263426352636263726382639264026412642264326442645264626472648264926502651265226532654265526562657265826592660266126622663266426652666266726682669
  1. /* Copyright (c) 2001 Matej Pfajfar.
  2. * Copyright (c) 2001-2004, Roger Dingledine.
  3. * Copyright (c) 2004-2006, Roger Dingledine, Nick Mathewson.
  4. * Copyright (c) 2007-2019, The Tor Project, Inc. */
  5. /* See LICENSE for licensing information */
  6. /**
  7. * \file mainloop.c
  8. * \brief Toplevel module. Handles signals, multiplexes between
  9. * connections, implements main loop, and drives scheduled events.
  10. *
  11. * For the main loop itself; see run_main_loop_once(). It invokes the rest of
  12. * Tor mostly through Libevent callbacks. Libevent callbacks can happen when
  13. * a timer elapses, a signal is received, a socket is ready to read or write,
  14. * or an event is manually activated.
  15. *
  16. * Most events in Tor are driven from these callbacks:
  17. * <ul>
  18. * <li>conn_read_callback() and conn_write_callback() here, which are
  19. * invoked when a socket is ready to read or write respectively.
  20. * <li>signal_callback(), which handles incoming signals.
  21. * </ul>
  22. * Other events are used for specific purposes, or for building more complex
  23. * control structures. If you search for usage of tor_libevent_new(), you
  24. * will find all the events that we construct in Tor.
  25. *
  26. * Tor has numerous housekeeping operations that need to happen
  27. * regularly. They are handled in different ways:
  28. * <ul>
  29. * <li>The most frequent operations are handled after every read or write
  30. * event, at the end of connection_handle_read() and
  31. * connection_handle_write().
  32. *
  33. * <li>The next most frequent operations happen after each invocation of the
  34. * main loop, in run_main_loop_once().
  35. *
  36. * <li>Once per second, we run all of the operations listed in
  37. * second_elapsed_callback(), and in its child, run_scheduled_events().
  38. *
  39. * <li>Once-a-second operations are handled in second_elapsed_callback().
  40. *
  41. * <li>More infrequent operations take place based on the periodic event
  42. * driver in periodic.c . These are stored in the periodic_events[]
  43. * table.
  44. * </ul>
  45. *
  46. **/
  47. #define MAINLOOP_PRIVATE
  48. #include "core/or/or.h"
  49. #include "app/config/config.h"
  50. #include "app/config/statefile.h"
  51. #include "app/main/ntmain.h"
  52. #include "app/main/tor_threads.h"
  53. #include "core/mainloop/connection.h"
  54. #include "core/mainloop/cpuworker.h"
  55. #include "core/mainloop/mainloop.h"
  56. #include "core/mainloop/netstatus.h"
  57. #include "core/mainloop/periodic.h"
  58. #include "core/or/channel.h"
  59. #include "core/or/channelpadding.h"
  60. #include "core/or/channeltls.h"
  61. #include "core/or/circuitbuild.h"
  62. #include "core/or/circuitlist.h"
  63. #include "core/or/circuituse.h"
  64. #include "core/or/connection_edge.h"
  65. #include "core/or/connection_or.h"
  66. #include "core/or/dos.h"
  67. #include "core/or/status.h"
  68. #include "feature/client/addressmap.h"
  69. #include "feature/client/bridges.h"
  70. #include "feature/client/dnsserv.h"
  71. #include "feature/client/entrynodes.h"
  72. #include "feature/client/transports.h"
  73. #include "feature/control/control.h"
  74. #include "feature/control/control_events.h"
  75. #include "feature/dirauth/authmode.h"
  76. #include "feature/dircache/consdiffmgr.h"
  77. #include "feature/dircache/dirserv.h"
  78. #include "feature/dircommon/directory.h"
  79. #include "feature/hibernate/hibernate.h"
  80. #include "feature/hs/hs_cache.h"
  81. #include "feature/hs/hs_client.h"
  82. #include "feature/hs/hs_service.h"
  83. #include "feature/nodelist/microdesc.h"
  84. #include "feature/nodelist/networkstatus.h"
  85. #include "feature/nodelist/nodelist.h"
  86. #include "feature/nodelist/routerlist.h"
  87. #include "feature/relay/dns.h"
  88. #include "feature/relay/routerkeys.h"
  89. #include "feature/relay/routermode.h"
  90. #include "feature/relay/selftest.h"
  91. #include "feature/rend/rendcache.h"
  92. #include "feature/rend/rendservice.h"
  93. #include "feature/stats/geoip_stats.h"
  94. #include "feature/stats/predict_ports.h"
  95. #include "feature/stats/rephist.h"
  96. #include "lib/buf/buffers.h"
  97. #include "lib/crypt_ops/crypto_rand.h"
  98. #include "lib/err/backtrace.h"
  99. #include "lib/tls/buffers_tls.h"
  100. #include "lib/net/buffers_net.h"
  101. #include "lib/evloop/compat_libevent.h"
  102. #include <event2/event.h>
  103. #include "core/or/cell_st.h"
  104. #include "core/or/entry_connection_st.h"
  105. #include "feature/nodelist/networkstatus_st.h"
  106. #include "core/or/or_connection_st.h"
  107. #include "app/config/or_state_st.h"
  108. #include "feature/nodelist/routerinfo_st.h"
  109. #include "core/or/socks_request_st.h"
  110. #ifdef HAVE_UNISTD_H
  111. #include <unistd.h>
  112. #endif
  113. #ifdef HAVE_SYSTEMD
  114. # if defined(__COVERITY__) && !defined(__INCLUDE_LEVEL__)
  115. /* Systemd's use of gcc's __INCLUDE_LEVEL__ extension macro appears to confuse
  116. * Coverity. Here's a kludge to unconfuse it.
  117. */
  118. # define __INCLUDE_LEVEL__ 2
  119. #endif /* defined(__COVERITY__) && !defined(__INCLUDE_LEVEL__) */
  120. #include <systemd/sd-daemon.h>
  121. #endif /* defined(HAVE_SYSTEMD) */
  122. /* Token bucket for all traffic. */
  123. token_bucket_rw_t global_bucket;
  124. /* Token bucket for relayed traffic. */
  125. token_bucket_rw_t global_relayed_bucket;
  126. /* XXX we might want to keep stats about global_relayed_*_bucket too. Or not.*/
  127. /** How many bytes have we read since we started the process? */
  128. static uint64_t stats_n_bytes_read = 0;
  129. /** How many bytes have we written since we started the process? */
  130. static uint64_t stats_n_bytes_written = 0;
  131. /** What time did this process start up? */
  132. time_t time_of_process_start = 0;
  133. /** How many seconds have we been running? */
  134. static long stats_n_seconds_working = 0;
  135. /** How many times have we returned from the main loop successfully? */
  136. static uint64_t stats_n_main_loop_successes = 0;
  137. /** How many times have we received an error from the main loop? */
  138. static uint64_t stats_n_main_loop_errors = 0;
  139. /** How many times have we returned from the main loop with no events. */
  140. static uint64_t stats_n_main_loop_idle = 0;
  141. /** How often will we honor SIGNEWNYM requests? */
  142. #define MAX_SIGNEWNYM_RATE 10
  143. /** When did we last process a SIGNEWNYM request? */
  144. static time_t time_of_last_signewnym = 0;
  145. /** Is there a signewnym request we're currently waiting to handle? */
  146. static int signewnym_is_pending = 0;
  147. /** Mainloop event for the deferred signewnym call. */
  148. static mainloop_event_t *handle_deferred_signewnym_ev = NULL;
  149. /** How many times have we called newnym? */
  150. static unsigned newnym_epoch = 0;
  151. /** Smartlist of all open connections. */
  152. STATIC smartlist_t *connection_array = NULL;
  153. /** List of connections that have been marked for close and need to be freed
  154. * and removed from connection_array. */
  155. static smartlist_t *closeable_connection_lst = NULL;
  156. /** List of linked connections that are currently reading data into their
  157. * inbuf from their partner's outbuf. */
  158. static smartlist_t *active_linked_connection_lst = NULL;
  159. /** Flag: Set to true iff we entered the current libevent main loop via
  160. * <b>loop_once</b>. If so, there's no need to trigger a loopexit in order
  161. * to handle linked connections. */
  162. static int called_loop_once = 0;
  163. /** Flag: if true, it's time to shut down, so the main loop should exit as
  164. * soon as possible.
  165. */
  166. static int main_loop_should_exit = 0;
  167. /** The return value that the main loop should yield when it exits, if
  168. * main_loop_should_exit is true.
  169. */
  170. static int main_loop_exit_value = 0;
  171. /** Flag: if true, it's time to shut down, so the other eventloops should
  172. * exit as soon as possible.
  173. */
  174. static int other_eventloops_should_exit = 0;
  175. /** We set this to 1 when we've opened a circuit, so we can print a log
  176. * entry to inform the user that Tor is working. We set it to 0 when
  177. * we think the fact that we once opened a circuit doesn't mean we can do so
  178. * any longer (a big time jump happened, when we notice our directory is
  179. * heinously out-of-date, etc.
  180. */
  181. static int can_complete_circuits = 0;
  182. /** How often do we check for router descriptors that we should download
  183. * when we have too little directory info? */
  184. #define GREEDY_DESCRIPTOR_RETRY_INTERVAL (10)
  185. /** How often do we check for router descriptors that we should download
  186. * when we have enough directory info? */
  187. #define LAZY_DESCRIPTOR_RETRY_INTERVAL (60)
  188. static int conn_close_if_marked(int i);
  189. static void connection_start_reading_from_linked_conn(connection_t *conn);
  190. static int connection_should_read_from_linked_conn(connection_t *conn);
  191. static void conn_read_callback(evutil_socket_t fd, short event, void *_conn);
  192. static void conn_write_callback(evutil_socket_t fd, short event, void *_conn);
  193. static void shutdown_did_not_work_callback(evutil_socket_t fd, short event,
  194. void *arg) ATTR_NORETURN;
  195. static event_registry_t *event_registry = NULL;
  196. /****************************************************************************
  197. *
  198. * This section contains accessors and other methods on the connection_array
  199. * variables (which are global within this file and unavailable outside it).
  200. *
  201. ****************************************************************************/
  202. /** Return 1 if we have successfully built a circuit, and nothing has changed
  203. * to make us think that maybe we can't.
  204. */
  205. int
  206. have_completed_a_circuit(void)
  207. {
  208. return can_complete_circuits;
  209. }
  210. /** Note that we have successfully built a circuit, so that reachability
  211. * testing and introduction points and so on may be attempted. */
  212. void
  213. note_that_we_completed_a_circuit(void)
  214. {
  215. can_complete_circuits = 1;
  216. }
  217. /** Note that something has happened (like a clock jump, or DisableNetwork) to
  218. * make us think that maybe we can't complete circuits. */
  219. void
  220. note_that_we_maybe_cant_complete_circuits(void)
  221. {
  222. can_complete_circuits = 0;
  223. }
  224. static unsigned long eventloop_counter = 0;
  225. /** Add <b>conn</b> to the array of connections that we can poll on. The
  226. * connection's socket must be set; the connection starts out
  227. * non-reading and non-writing.
  228. */
  229. int
  230. connection_add_impl(connection_t *conn, int is_connecting)
  231. {
  232. (void) is_connecting;
  233. tor_assert(conn != NULL);
  234. tor_assert(conn->conn_array_index == -1); /* can only connection_add once */
  235. conn->conn_array_index = smartlist_len(connection_array);
  236. smartlist_add(connection_array, conn);
  237. event_listener_attach(conn->event_listener, tor_libevent_get_base());
  238. // TODO: check for error
  239. if (conn->safe_conn == NULL) {
  240. tor_assert(conn->type != CONN_TYPE_OR);
  241. tor_assert(SOCKET_OK(conn->s) ||
  242. conn->linked ||
  243. (conn->type == CONN_TYPE_AP &&
  244. TO_EDGE_CONN(conn)->is_dns_request));
  245. if (SOCKET_OK(conn->s) || conn->linked) {
  246. conn->read_event = tor_event_new(tor_libevent_get_base(),
  247. conn->s, EV_READ|EV_PERSIST, conn_read_callback, conn);
  248. conn->write_event = tor_event_new(tor_libevent_get_base(),
  249. conn->s, EV_WRITE|EV_PERSIST, conn_write_callback, conn);
  250. /* XXXX CHECK FOR NULL RETURN! */
  251. }
  252. log_debug(LD_NET, "new conn type %s, socket %d, address %s, n_conns %d.",
  253. conn_type_to_string(conn->type), (int)conn->s, conn->address,
  254. smartlist_len(connection_array));
  255. } else {
  256. tor_assert(conn->type == CONN_TYPE_OR);
  257. int eventloop_index = 1 + (eventloop_counter%(get_num_eventloops()-1));
  258. struct event_base *base = get_eventloop(eventloop_index);
  259. eventloop_counter += 1;
  260. //struct event_base *base = tor_libevent_get_base();
  261. error_t rv = safe_connection_register_events(conn->safe_conn, base);
  262. if (rv != E_SUCCESS) {
  263. smartlist_remove(connection_array, conn);
  264. conn->conn_array_index = -1;
  265. return 1;
  266. }
  267. log_debug(LD_NET, "new conn type %s, address %s, n_conns %d.",
  268. conn_type_to_string(conn->type), conn->address,
  269. smartlist_len(connection_array));
  270. }
  271. return 0;
  272. }
  273. /** Tell libevent that we don't care about <b>conn</b> any more. */
  274. void
  275. connection_unregister_events(connection_t *conn)
  276. {
  277. tor_assert(conn != NULL);
  278. if (conn->safe_conn == NULL) {
  279. tor_assert(conn->type != CONN_TYPE_OR);
  280. if (conn->read_event) {
  281. if (event_del(conn->read_event))
  282. log_warn(LD_BUG, "Error removing read event for %d", (int)conn->s);
  283. tor_free(conn->read_event);
  284. }
  285. if (conn->write_event) {
  286. if (event_del(conn->write_event))
  287. log_warn(LD_BUG, "Error removing write event for %d", (int)conn->s);
  288. tor_free(conn->write_event);
  289. }
  290. if (conn->type == CONN_TYPE_AP_DNS_LISTENER) {
  291. dnsserv_close_listener(conn);
  292. }
  293. } else {
  294. tor_assert(conn->type == CONN_TYPE_OR);
  295. safe_connection_unregister_events(conn->safe_conn);
  296. }
  297. event_listener_detach(conn->event_listener);
  298. // TODO: do we want to process the remaining events?
  299. }
  300. /** Remove the connection from the global list, and remove the
  301. * corresponding poll entry. Calling this function will shift the last
  302. * connection (if any) into the position occupied by conn.
  303. */
  304. int
  305. connection_remove(connection_t *conn)
  306. {
  307. int current_index;
  308. connection_t *tmp;
  309. tor_assert(conn);
  310. log_debug(LD_NET, "removing socket %d (type %s), n_conns now %d",
  311. (int)conn->s, conn_type_to_string(conn->type),
  312. smartlist_len(connection_array));
  313. if (conn->type == CONN_TYPE_AP && conn->socket_family == AF_UNIX) {
  314. log_info(LD_NET, "Closing SOCKS Unix socket connection");
  315. }
  316. control_event_conn_bandwidth(conn);
  317. tor_assert(conn->conn_array_index >= 0);
  318. current_index = conn->conn_array_index;
  319. connection_unregister_events(conn); /* This is redundant, but cheap. */
  320. if (current_index == smartlist_len(connection_array)-1) { /* at the end */
  321. smartlist_del(connection_array, current_index);
  322. return 0;
  323. }
  324. /* replace this one with the one at the end */
  325. smartlist_del(connection_array, current_index);
  326. tmp = smartlist_get(connection_array, current_index);
  327. tmp->conn_array_index = current_index;
  328. return 0;
  329. }
  330. /** If <b>conn</b> is an edge conn, remove it from the list
  331. * of conn's on this circuit. If it's not on an edge,
  332. * flush and send destroys for all circuits on this conn.
  333. *
  334. * Remove it from connection_array (if applicable) and
  335. * from closeable_connection_list.
  336. *
  337. * Then free it.
  338. */
  339. static void
  340. connection_unlink(connection_t *conn)
  341. {
  342. connection_about_to_close_connection(conn);
  343. if (conn->conn_array_index >= 0) {
  344. connection_remove(conn);
  345. }
  346. if (conn->linked_conn) {
  347. conn->linked_conn->linked_conn = NULL;
  348. if (! conn->linked_conn->marked_for_close &&
  349. conn->linked_conn->reading_from_linked_conn)
  350. connection_start_reading(conn->linked_conn);
  351. conn->linked_conn = NULL;
  352. }
  353. smartlist_remove(closeable_connection_lst, conn);
  354. smartlist_remove(active_linked_connection_lst, conn);
  355. if (conn->type == CONN_TYPE_EXIT) {
  356. assert_connection_edge_not_dns_pending(TO_EDGE_CONN(conn));
  357. }
  358. if (conn->type == CONN_TYPE_OR) {
  359. if (!tor_digest_is_zero(TO_OR_CONN(conn)->identity_digest))
  360. connection_or_clear_identity(TO_OR_CONN(conn));
  361. /* connection_unlink() can only get called if the connection
  362. * was already on the closeable list, and it got there by
  363. * connection_mark_for_close(), which was called from
  364. * connection_or_close_normally() or
  365. * connection_or_close_for_error(), so the channel should
  366. * already be in CHANNEL_STATE_CLOSING, and then the
  367. * connection_about_to_close_connection() goes to
  368. * connection_or_about_to_close(), which calls channel_closed()
  369. * to notify the channel_t layer, and closed the channel, so
  370. * nothing more to do here to deal with the channel associated
  371. * with an orconn.
  372. */
  373. }
  374. connection_free(conn);
  375. }
  376. /** Event that invokes schedule_active_linked_connections_cb. */
  377. static mainloop_event_t *schedule_active_linked_connections_event = NULL;
  378. /**
  379. * Callback: used to activate read events for all linked connections, so
  380. * libevent knows to call their read callbacks. This callback run as a
  381. * postloop event, so that the events _it_ activates don't happen until
  382. * Libevent has a chance to check for other events.
  383. */
  384. static void
  385. schedule_active_linked_connections_cb(mainloop_event_t *event, void *arg)
  386. {
  387. (void)event;
  388. (void)arg;
  389. /* All active linked conns should get their read events activated,
  390. * so that libevent knows to run their callbacks. */
  391. SMARTLIST_FOREACH(active_linked_connection_lst, connection_t *, conn,
  392. event_active(conn->read_event, EV_READ, 1));
  393. /* Reactivate the event if we still have connections in the active list.
  394. *
  395. * A linked connection doesn't get woken up by I/O but rather artificially
  396. * by this event callback. It has directory data spooled in it and it is
  397. * sent incrementally by small chunks unless spool_eagerly is true. For that
  398. * to happen, we need to induce the activation of the read event so it can
  399. * be flushed. */
  400. if (smartlist_len(active_linked_connection_lst)) {
  401. mainloop_event_activate(schedule_active_linked_connections_event);
  402. }
  403. }
  404. /** Initialize the global connection list, closeable connection list,
  405. * and active connection list. */
  406. void
  407. tor_init_connection_lists(void)
  408. {
  409. if (!connection_array)
  410. connection_array = smartlist_new();
  411. if (!closeable_connection_lst)
  412. closeable_connection_lst = smartlist_new();
  413. if (!active_linked_connection_lst)
  414. active_linked_connection_lst = smartlist_new();
  415. }
  416. /** Schedule <b>conn</b> to be closed. **/
  417. void
  418. add_connection_to_closeable_list(connection_t *conn)
  419. {
  420. tor_assert(!smartlist_contains(closeable_connection_lst, conn));
  421. tor_assert(conn->marked_for_close);
  422. assert_connection_ok(conn, time(NULL));
  423. smartlist_add(closeable_connection_lst, conn);
  424. mainloop_schedule_postloop_cleanup();
  425. }
  426. /** Return 1 if conn is on the closeable list, else return 0. */
  427. int
  428. connection_is_on_closeable_list(connection_t *conn)
  429. {
  430. return smartlist_contains(closeable_connection_lst, conn);
  431. }
  432. /** Return true iff conn is in the current poll array. */
  433. int
  434. connection_in_array(connection_t *conn)
  435. {
  436. return smartlist_contains(connection_array, conn);
  437. }
  438. /** Set <b>*array</b> to an array of all connections. <b>*array</b> must not
  439. * be modified.
  440. */
  441. MOCK_IMPL(smartlist_t *,
  442. get_connection_array, (void))
  443. {
  444. if (!connection_array)
  445. connection_array = smartlist_new();
  446. return connection_array;
  447. }
  448. /**
  449. * Return the amount of network traffic read, in bytes, over the life of this
  450. * process.
  451. */
  452. MOCK_IMPL(uint64_t,
  453. get_bytes_read,(void))
  454. {
  455. return stats_n_bytes_read;
  456. }
  457. /**
  458. * Return the amount of network traffic read, in bytes, over the life of this
  459. * process.
  460. */
  461. MOCK_IMPL(uint64_t,
  462. get_bytes_written,(void))
  463. {
  464. return stats_n_bytes_written;
  465. }
  466. /**
  467. * Increment the amount of network traffic read and written, over the life of
  468. * this process.
  469. */
  470. void
  471. stats_increment_bytes_read_and_written(uint64_t r, uint64_t w)
  472. {
  473. stats_n_bytes_read += r;
  474. stats_n_bytes_written += w;
  475. }
  476. /** Set the event mask on <b>conn</b> to <b>events</b>. (The event
  477. * mask is a bitmask whose bits are READ_EVENT and WRITE_EVENT)
  478. */
  479. void
  480. connection_watch_events(connection_t *conn, watchable_events_t events)
  481. {
  482. if (events & READ_EVENT)
  483. connection_start_reading(conn);
  484. else
  485. connection_stop_reading(conn);
  486. if (events & WRITE_EVENT)
  487. connection_start_writing(conn);
  488. else
  489. connection_stop_writing(conn);
  490. }
  491. /** Return true iff <b>conn</b> is listening for read events. */
  492. int
  493. connection_is_reading(connection_t *conn)
  494. {
  495. tor_assert(conn);
  496. return conn->reading_from_linked_conn ||
  497. (conn->read_event && event_pending(conn->read_event, EV_READ, NULL));
  498. }
  499. /** Reset our main loop counters. */
  500. void
  501. reset_main_loop_counters(void)
  502. {
  503. stats_n_main_loop_successes = 0;
  504. stats_n_main_loop_errors = 0;
  505. stats_n_main_loop_idle = 0;
  506. }
  507. /** Increment the main loop success counter. */
  508. static void
  509. increment_main_loop_success_count(void)
  510. {
  511. ++stats_n_main_loop_successes;
  512. }
  513. /** Get the main loop success counter. */
  514. uint64_t
  515. get_main_loop_success_count(void)
  516. {
  517. return stats_n_main_loop_successes;
  518. }
  519. /** Increment the main loop error counter. */
  520. static void
  521. increment_main_loop_error_count(void)
  522. {
  523. ++stats_n_main_loop_errors;
  524. }
  525. /** Get the main loop error counter. */
  526. uint64_t
  527. get_main_loop_error_count(void)
  528. {
  529. return stats_n_main_loop_errors;
  530. }
  531. /** Increment the main loop idle counter. */
  532. static void
  533. increment_main_loop_idle_count(void)
  534. {
  535. ++stats_n_main_loop_idle;
  536. }
  537. /** Get the main loop idle counter. */
  538. uint64_t
  539. get_main_loop_idle_count(void)
  540. {
  541. return stats_n_main_loop_idle;
  542. }
  543. /** Check whether <b>conn</b> is correct in having (or not having) a
  544. * read/write event (passed in <b>ev</b>). On success, return 0. On failure,
  545. * log a warning and return -1. */
  546. static int
  547. connection_check_event(connection_t *conn, struct event *ev)
  548. {
  549. int bad;
  550. if (conn->type == CONN_TYPE_AP && TO_EDGE_CONN(conn)->is_dns_request) {
  551. /* DNS requests which we launch through the dnsserv.c module do not have
  552. * any underlying socket or any underlying linked connection, so they
  553. * shouldn't have any attached events either.
  554. */
  555. bad = ev != NULL;
  556. } else {
  557. /* Everything else should have an underlying socket, or a linked
  558. * connection (which is also tracked with a read_event/write_event pair).
  559. */
  560. bad = ev == NULL;
  561. }
  562. if (bad) {
  563. log_warn(LD_BUG, "Event missing on connection %p [%s;%s]. "
  564. "socket=%d. linked=%d. "
  565. "is_dns_request=%d. Marked_for_close=%s:%d",
  566. conn,
  567. conn_type_to_string(conn->type),
  568. conn_state_to_string(conn->type, conn->state),
  569. (int)conn->s, (int)conn->linked,
  570. (conn->type == CONN_TYPE_AP &&
  571. TO_EDGE_CONN(conn)->is_dns_request),
  572. conn->marked_for_close_file ? conn->marked_for_close_file : "-",
  573. conn->marked_for_close
  574. );
  575. log_backtrace(LOG_WARN, LD_BUG, "Backtrace attached.");
  576. return -1;
  577. }
  578. return 0;
  579. }
  580. /** Tell the main loop to stop notifying <b>conn</b> of any read events. */
  581. MOCK_IMPL(void,
  582. connection_stop_reading,(connection_t *conn))
  583. {
  584. tor_assert(conn);
  585. tor_assert(conn->type != CONN_TYPE_OR);
  586. if (connection_check_event(conn, conn->read_event) < 0) {
  587. return;
  588. }
  589. if (conn->linked) {
  590. conn->reading_from_linked_conn = 0;
  591. connection_stop_reading_from_linked_conn(conn);
  592. } else {
  593. if (event_del(conn->read_event))
  594. log_warn(LD_NET, "Error from libevent setting read event state for %d "
  595. "to unwatched: %s",
  596. (int)conn->s,
  597. tor_socket_strerror(tor_socket_errno(conn->s)));
  598. }
  599. }
  600. /** Tell the main loop to start notifying <b>conn</b> of any read events. */
  601. MOCK_IMPL(void,
  602. connection_start_reading,(connection_t *conn))
  603. {
  604. tor_assert(conn);
  605. tor_assert(conn->type != CONN_TYPE_OR);
  606. if (connection_check_event(conn, conn->read_event) < 0) {
  607. return;
  608. }
  609. if (conn->linked) {
  610. conn->reading_from_linked_conn = 1;
  611. if (connection_should_read_from_linked_conn(conn))
  612. connection_start_reading_from_linked_conn(conn);
  613. } else {
  614. if (event_add(conn->read_event, NULL))
  615. log_warn(LD_NET, "Error from libevent setting read event state for %d "
  616. "to watched: %s",
  617. (int)conn->s,
  618. tor_socket_strerror(tor_socket_errno(conn->s)));
  619. }
  620. }
  621. /** Return true iff <b>conn</b> is listening for write events. */
  622. int
  623. connection_is_writing(connection_t *conn)
  624. {
  625. tor_assert(conn);
  626. tor_assert(conn->type != CONN_TYPE_OR);
  627. return conn->writing_to_linked_conn ||
  628. (conn->write_event && event_pending(conn->write_event, EV_WRITE, NULL));
  629. }
  630. /** Tell the main loop to stop notifying <b>conn</b> of any write events. */
  631. MOCK_IMPL(void,
  632. connection_stop_writing,(connection_t *conn))
  633. {
  634. tor_assert(conn);
  635. tor_assert(conn->type != CONN_TYPE_OR);
  636. if (connection_check_event(conn, conn->write_event) < 0) {
  637. return;
  638. }
  639. if (conn->linked) {
  640. conn->writing_to_linked_conn = 0;
  641. if (conn->linked_conn)
  642. connection_stop_reading_from_linked_conn(conn->linked_conn);
  643. } else {
  644. if (event_del(conn->write_event))
  645. log_warn(LD_NET, "Error from libevent setting write event state for %d "
  646. "to unwatched: %s",
  647. (int)conn->s,
  648. tor_socket_strerror(tor_socket_errno(conn->s)));
  649. }
  650. }
  651. /** Tell the main loop to start notifying <b>conn</b> of any write events. */
  652. MOCK_IMPL(void,
  653. connection_start_writing,(connection_t *conn))
  654. {
  655. tor_assert(conn);
  656. tor_assert(conn->type != CONN_TYPE_OR);
  657. if (connection_check_event(conn, conn->write_event) < 0) {
  658. return;
  659. }
  660. if (conn->linked) {
  661. conn->writing_to_linked_conn = 1;
  662. if (conn->linked_conn &&
  663. connection_should_read_from_linked_conn(conn->linked_conn))
  664. connection_start_reading_from_linked_conn(conn->linked_conn);
  665. } else {
  666. if (event_add(conn->write_event, NULL))
  667. log_warn(LD_NET, "Error from libevent setting write event state for %d "
  668. "to watched: %s",
  669. (int)conn->s,
  670. tor_socket_strerror(tor_socket_errno(conn->s)));
  671. }
  672. }
  673. /** Return true iff <b>conn</b> is linked conn, and reading from the conn
  674. * linked to it would be good and feasible. (Reading is "feasible" if the
  675. * other conn exists and has data in its outbuf, and is "good" if we have our
  676. * reading_from_linked_conn flag set and the other conn has its
  677. * writing_to_linked_conn flag set.)*/
  678. static int
  679. connection_should_read_from_linked_conn(connection_t *conn)
  680. {
  681. if (conn->linked && conn->reading_from_linked_conn) {
  682. if (! conn->linked_conn ||
  683. (conn->linked_conn->writing_to_linked_conn &&
  684. buf_datalen(conn->linked_conn->outbuf)))
  685. return 1;
  686. }
  687. return 0;
  688. }
  689. /** Event to run 'shutdown did not work callback'. */
  690. static struct event *shutdown_did_not_work_event = NULL;
  691. /** Failsafe measure that should never actually be necessary: If
  692. * tor_shutdown_event_loop_and_exit() somehow doesn't successfully exit the
  693. * event loop, then this callback will kill Tor with an assertion failure
  694. * seconds later
  695. */
  696. static void
  697. shutdown_did_not_work_callback(evutil_socket_t fd, short event, void *arg)
  698. {
  699. // LCOV_EXCL_START
  700. (void) fd;
  701. (void) event;
  702. (void) arg;
  703. tor_assert_unreached();
  704. // LCOV_EXCL_STOP
  705. }
  706. #ifdef ENABLE_RESTART_DEBUGGING
  707. static struct event *tor_shutdown_event_loop_for_restart_event = NULL;
  708. static void
  709. tor_shutdown_event_loop_for_restart_cb(
  710. evutil_socket_t fd, short event, void *arg)
  711. {
  712. (void)fd;
  713. (void)event;
  714. (void)arg;
  715. tor_event_free(tor_shutdown_event_loop_for_restart_event);
  716. tor_shutdown_event_loop_and_exit(0);
  717. }
  718. #endif /* defined(ENABLE_RESTART_DEBUGGING) */
  719. /**
  720. * After finishing the current callback (if any), shut down the main loop,
  721. * clean up the process, and exit with <b>exitcode</b>.
  722. */
  723. void
  724. tor_shutdown_event_loop_and_exit(int exitcode)
  725. {
  726. if (main_loop_should_exit)
  727. return; /* Ignore multiple calls to this function. */
  728. main_loop_should_exit = 1;
  729. main_loop_exit_value = exitcode;
  730. if (! tor_libevent_is_initialized()) {
  731. return; /* No event loop to shut down. */
  732. }
  733. /* Die with an assertion failure in ten seconds, if for some reason we don't
  734. * exit normally. */
  735. /* XXXX We should consider this code if it's never used. */
  736. struct timeval ten_seconds = { 10, 0 };
  737. shutdown_did_not_work_event = tor_evtimer_new(
  738. tor_libevent_get_base(),
  739. shutdown_did_not_work_callback, NULL);
  740. event_add(shutdown_did_not_work_event, &ten_seconds);
  741. /* Unlike exit_loop_after_delay(), exit_loop_after_callback
  742. * prevents other callbacks from running. */
  743. tor_libevent_exit_loop_after_callback(tor_libevent_get_base());
  744. }
  745. /** Return true iff tor_shutdown_event_loop_and_exit() has been called. */
  746. int
  747. tor_event_loop_shutdown_is_pending(void)
  748. {
  749. return main_loop_should_exit;
  750. }
  751. /** Helper: Tell the main loop to begin reading bytes into <b>conn</b> from
  752. * its linked connection, if it is not doing so already. Called by
  753. * connection_start_reading and connection_start_writing as appropriate. */
  754. static void
  755. connection_start_reading_from_linked_conn(connection_t *conn)
  756. {
  757. tor_assert(conn);
  758. tor_assert(conn->linked == 1);
  759. if (!conn->active_on_link) {
  760. conn->active_on_link = 1;
  761. smartlist_add(active_linked_connection_lst, conn);
  762. mainloop_event_activate(schedule_active_linked_connections_event);
  763. } else {
  764. tor_assert(smartlist_contains(active_linked_connection_lst, conn));
  765. }
  766. }
  767. /** Tell the main loop to stop reading bytes into <b>conn</b> from its linked
  768. * connection, if is currently doing so. Called by connection_stop_reading,
  769. * connection_stop_writing, and connection_read. */
  770. void
  771. connection_stop_reading_from_linked_conn(connection_t *conn)
  772. {
  773. tor_assert(conn);
  774. tor_assert(conn->linked == 1);
  775. if (conn->active_on_link) {
  776. conn->active_on_link = 0;
  777. /* FFFF We could keep an index here so we can smartlist_del
  778. * cleanly. On the other hand, this doesn't show up on profiles,
  779. * so let's leave it alone for now. */
  780. smartlist_remove(active_linked_connection_lst, conn);
  781. } else {
  782. tor_assert(!smartlist_contains(active_linked_connection_lst, conn));
  783. }
  784. }
  785. /** Close all connections that have been scheduled to get closed. */
  786. STATIC void
  787. close_closeable_connections(void)
  788. {
  789. int i;
  790. for (i = 0; i < smartlist_len(closeable_connection_lst); ) {
  791. connection_t *conn = smartlist_get(closeable_connection_lst, i);
  792. if (conn->conn_array_index < 0) {
  793. connection_unlink(conn); /* blow it away right now */
  794. } else {
  795. if (!conn_close_if_marked(conn->conn_array_index))
  796. ++i;
  797. }
  798. }
  799. }
  800. /** Count moribund connections for the OOS handler */
  801. MOCK_IMPL(int,
  802. connection_count_moribund, (void))
  803. {
  804. int moribund = 0;
  805. /*
  806. * Count things we'll try to kill when close_closeable_connections()
  807. * runs next.
  808. */
  809. SMARTLIST_FOREACH_BEGIN(closeable_connection_lst, connection_t *, conn) {
  810. if (SOCKET_OK(conn->s) && connection_is_moribund(conn)) ++moribund;
  811. } SMARTLIST_FOREACH_END(conn);
  812. return moribund;
  813. }
  814. /** Libevent callback: this gets invoked when (connection_t*)<b>conn</b> has
  815. * some data to read. */
  816. static void
  817. conn_read_callback(evutil_socket_t fd, short event, void *_conn)
  818. {
  819. connection_t *conn = _conn;
  820. (void)fd;
  821. (void)event;
  822. log_debug(LD_NET,"socket %d wants to read.",(int)conn->s);
  823. /* assert_connection_ok(conn, time(NULL)); */
  824. /* Handle marked for close connections early */
  825. if (conn->marked_for_close && connection_is_reading(conn)) {
  826. /* Libevent says we can read, but we are marked for close so we will never
  827. * try to read again. We will try to close the connection below inside of
  828. * close_closeable_connections(), but let's make sure not to cause Libevent
  829. * to spin on conn_read_callback() while we wait for the socket to let us
  830. * flush to it.*/
  831. connection_stop_reading(conn);
  832. }
  833. if (connection_handle_read(conn) < 0) {
  834. if (!conn->marked_for_close) {
  835. #ifndef _WIN32
  836. log_warn(LD_BUG,"Unhandled error on read for %s connection "
  837. "(fd %d); removing",
  838. conn_type_to_string(conn->type), (int)conn->s);
  839. tor_fragile_assert();
  840. #endif /* !defined(_WIN32) */
  841. if (CONN_IS_EDGE(conn))
  842. connection_edge_end_errno(TO_EDGE_CONN(conn));
  843. connection_mark_for_close(conn);
  844. }
  845. }
  846. assert_connection_ok(conn, time(NULL));
  847. if (smartlist_len(closeable_connection_lst))
  848. close_closeable_connections();
  849. }
  850. /** Libevent callback: this gets invoked when (connection_t*)<b>conn</b> has
  851. * some data to write. */
  852. static void
  853. conn_write_callback(evutil_socket_t fd, short events, void *_conn)
  854. {
  855. connection_t *conn = _conn;
  856. (void)fd;
  857. (void)events;
  858. LOG_FN_CONN(conn, (LOG_DEBUG, LD_NET, "socket %d wants to write.",
  859. (int)conn->s));
  860. /* assert_connection_ok(conn, time(NULL)); */
  861. if (connection_handle_write(conn, 0) < 0) {
  862. if (!conn->marked_for_close) {
  863. /* this connection is broken. remove it. */
  864. log_fn(LOG_WARN,LD_BUG,
  865. "unhandled error on write for %s connection (fd %d); removing",
  866. conn_type_to_string(conn->type), (int)conn->s);
  867. tor_fragile_assert();
  868. if (CONN_IS_EDGE(conn)) {
  869. /* otherwise we cry wolf about duplicate close */
  870. edge_connection_t *edge_conn = TO_EDGE_CONN(conn);
  871. if (!edge_conn->end_reason)
  872. edge_conn->end_reason = END_STREAM_REASON_INTERNAL;
  873. edge_conn->edge_has_sent_end = 1;
  874. }
  875. connection_close_immediate(conn); /* So we don't try to flush. */
  876. connection_mark_for_close(conn);
  877. }
  878. }
  879. assert_connection_ok(conn, time(NULL));
  880. if (smartlist_len(closeable_connection_lst))
  881. close_closeable_connections();
  882. }
  883. /** If the connection at connection_array[i] is marked for close, then:
  884. * - If it has data that it wants to flush, try to flush it.
  885. * - If it _still_ has data to flush, and conn->hold_open_until_flushed is
  886. * true, then leave the connection open and return.
  887. * - Otherwise, remove the connection from connection_array and from
  888. * all other lists, close it, and free it.
  889. * Returns 1 if the connection was closed, 0 otherwise.
  890. */
  891. static int
  892. conn_close_if_marked(int i)
  893. {
  894. connection_t *conn;
  895. int retval;
  896. time_t now;
  897. conn = smartlist_get(connection_array, i);
  898. if (!conn->marked_for_close)
  899. return 0; /* nothing to see here, move along */
  900. now = time(NULL);
  901. assert_connection_ok(conn, now);
  902. /* assert_all_pending_dns_resolves_ok(); */
  903. log_debug(LD_NET,"Cleaning up connection (fd "TOR_SOCKET_T_FORMAT").",
  904. conn->s);
  905. /* If the connection we are about to close was trying to connect to
  906. a proxy server and failed, the client won't be able to use that
  907. proxy. We should warn the user about this. */
  908. if (conn->proxy_state == PROXY_INFANT)
  909. log_failed_proxy_connection(conn);
  910. if ((SOCKET_OK(conn->s) || conn->linked_conn) &&
  911. connection_wants_to_flush(conn)) {
  912. /* s == -1 means it's an incomplete edge connection, or that the socket
  913. * has already been closed as unflushable. */
  914. //ssize_t sz = connection_bucket_write_limit(conn, now);
  915. ssize_t sz = 0;
  916. if (!conn->hold_open_until_flushed)
  917. log_info(LD_NET,
  918. "Conn (addr %s, fd %d, type %s, state %d) marked, but wants "
  919. "to flush %d bytes. (Marked at %s:%d)",
  920. escaped_safe_str_client(conn->address),
  921. (int)conn->s, conn_type_to_string(conn->type), conn->state,
  922. (int)conn->outbuf_flushlen,
  923. conn->marked_for_close_file, conn->marked_for_close);
  924. if (conn->linked_conn) {
  925. sz = connection_bucket_write_limit(conn, now);
  926. retval = buf_move_to_buf(conn->linked_conn->inbuf, conn->outbuf,
  927. &conn->outbuf_flushlen);
  928. if (retval >= 0) {
  929. /* The linked conn will notice that it has data when it notices that
  930. * we're gone. */
  931. connection_start_reading_from_linked_conn(conn->linked_conn);
  932. }
  933. log_debug(LD_GENERAL, "Flushed last %d bytes from a linked conn; "
  934. "%d left; flushlen %d; wants-to-flush==%d", retval,
  935. (int)connection_get_outbuf_len(conn),
  936. (int)conn->outbuf_flushlen,
  937. connection_wants_to_flush(conn));
  938. } else if (connection_speaks_cells(conn)) {
  939. if (conn->state == OR_CONN_STATE_OPEN) {
  940. // TODO: force TLS flush here, need to communicate to other thread somehow
  941. // really we need to tell the thread to close, and whether it should
  942. // try to flush or not
  943. retval = -1;
  944. //retval = buf_flush_to_tls(conn->outbuf, TO_OR_CONN(conn)->tls, sz,
  945. // &conn->outbuf_flushlen);
  946. } else
  947. retval = -1; /* never flush non-open broken tls connections */
  948. } else {
  949. sz = connection_bucket_write_limit(conn, now);
  950. retval = buf_flush_to_socket(conn->outbuf, conn->s, sz,
  951. &conn->outbuf_flushlen);
  952. }
  953. if (retval >= 0 && /* Technically, we could survive things like
  954. TLS_WANT_WRITE here. But don't bother for now. */
  955. conn->hold_open_until_flushed && connection_wants_to_flush(conn)) {
  956. if (retval > 0) {
  957. LOG_FN_CONN(conn, (LOG_INFO,LD_NET,
  958. "Holding conn (fd %d) open for more flushing.",
  959. (int)conn->s));
  960. conn->timestamp_last_write_allowed = now; /* reset so we can flush
  961. * more */
  962. } else if (sz == 0) {
  963. /* Also, retval==0. If we get here, we didn't want to write anything
  964. * (because of rate-limiting) and we didn't. */
  965. /* Connection must flush before closing, but it's being rate-limited.
  966. * Let's remove from Libevent, and mark it as blocked on bandwidth
  967. * so it will be re-added on next token bucket refill. Prevents
  968. * busy Libevent loops where we keep ending up here and returning
  969. * 0 until we are no longer blocked on bandwidth.
  970. */
  971. connection_consider_empty_write_buckets(conn);
  972. /* Make sure that consider_empty_buckets really disabled the
  973. * connection: */
  974. if (BUG(connection_is_writing(conn))) {
  975. connection_write_bw_exhausted(conn, true);
  976. }
  977. /* The connection is being held due to write rate limit and thus will
  978. * flush its data later. We need to stop reading because this
  979. * connection is about to be closed once flushed. It should not
  980. * process anything more coming in at this stage. */
  981. connection_stop_reading(conn);
  982. }
  983. return 0;
  984. }
  985. if (connection_wants_to_flush(conn)) {
  986. //log_fn(LOG_INFO, LD_NET, "We stalled too much while trying to write %d "
  987. log_fn(LOG_INFO, LD_NET, "We stalled too much while trying to write <?> "
  988. "bytes to address %s. If this happens a lot, either "
  989. "something is wrong with your network connection, or "
  990. "something is wrong with theirs. "
  991. "(fd %d, type %s, state %d, marked at %s:%d).",
  992. //(int)connection_get_outbuf_len(conn),
  993. escaped_safe_str_client(conn->address),
  994. (int)conn->s, conn_type_to_string(conn->type), conn->state,
  995. conn->marked_for_close_file,
  996. conn->marked_for_close);
  997. }
  998. }
  999. connection_unlink(conn); /* unlink, remove, free */
  1000. return 1;
  1001. }
  1002. /** Implementation for directory_all_unreachable. This is done in a callback,
  1003. * since otherwise it would complicate Tor's control-flow graph beyond all
  1004. * reason.
  1005. */
  1006. static void
  1007. directory_all_unreachable_cb(mainloop_event_t *event, void *arg)
  1008. {
  1009. (void)event;
  1010. (void)arg;
  1011. connection_t *conn;
  1012. while ((conn = connection_get_by_type_state(CONN_TYPE_AP,
  1013. AP_CONN_STATE_CIRCUIT_WAIT))) {
  1014. entry_connection_t *entry_conn = TO_ENTRY_CONN(conn);
  1015. log_notice(LD_NET,
  1016. "Is your network connection down? "
  1017. "Failing connection to '%s:%d'.",
  1018. safe_str_client(entry_conn->socks_request->address),
  1019. entry_conn->socks_request->port);
  1020. connection_mark_unattached_ap(entry_conn,
  1021. END_STREAM_REASON_NET_UNREACHABLE);
  1022. }
  1023. control_event_general_error("DIR_ALL_UNREACHABLE");
  1024. }
  1025. static mainloop_event_t *directory_all_unreachable_cb_event = NULL;
  1026. /** We've just tried every dirserver we know about, and none of
  1027. * them were reachable. Assume the network is down. Change state
  1028. * so next time an application connection arrives we'll delay it
  1029. * and try another directory fetch. Kill off all the circuit_wait
  1030. * streams that are waiting now, since they will all timeout anyway.
  1031. */
  1032. void
  1033. directory_all_unreachable(time_t now)
  1034. {
  1035. (void)now;
  1036. reset_uptime(); /* reset it */
  1037. if (!directory_all_unreachable_cb_event) {
  1038. directory_all_unreachable_cb_event =
  1039. mainloop_event_new(directory_all_unreachable_cb, NULL);
  1040. tor_assert(directory_all_unreachable_cb_event);
  1041. }
  1042. mainloop_event_activate(directory_all_unreachable_cb_event);
  1043. }
  1044. /** This function is called whenever we successfully pull down some new
  1045. * network statuses or server descriptors. */
  1046. void
  1047. directory_info_has_arrived(time_t now, int from_cache, int suppress_logs)
  1048. {
  1049. const or_options_t *options = get_options();
  1050. /* if we have enough dir info, then update our guard status with
  1051. * whatever we just learned. */
  1052. int invalidate_circs = guards_update_all();
  1053. if (invalidate_circs) {
  1054. circuit_mark_all_unused_circs();
  1055. circuit_mark_all_dirty_circs_as_unusable();
  1056. }
  1057. if (!router_have_minimum_dir_info()) {
  1058. int quiet = suppress_logs || from_cache ||
  1059. directory_too_idle_to_fetch_descriptors(options, now);
  1060. tor_log(quiet ? LOG_INFO : LOG_NOTICE, LD_DIR,
  1061. "I learned some more directory information, but not enough to "
  1062. "build a circuit: %s", get_dir_info_status_string());
  1063. update_all_descriptor_downloads(now);
  1064. return;
  1065. } else {
  1066. if (directory_fetches_from_authorities(options)) {
  1067. update_all_descriptor_downloads(now);
  1068. }
  1069. /* Don't even bother trying to get extrainfo until the rest of our
  1070. * directory info is up-to-date */
  1071. if (options->DownloadExtraInfo)
  1072. update_extrainfo_downloads(now);
  1073. }
  1074. if (server_mode(options) && !net_is_disabled() && !from_cache &&
  1075. (have_completed_a_circuit() || !any_predicted_circuits(now)))
  1076. router_do_reachability_checks(1, 1);
  1077. }
  1078. /** Perform regular maintenance tasks for a single connection. This
  1079. * function gets run once per second per connection by run_scheduled_events.
  1080. */
  1081. static void
  1082. run_connection_housekeeping(int i, time_t now)
  1083. {
  1084. connection_t *conn = smartlist_get(connection_array, i);
  1085. const or_options_t *options = get_options();
  1086. or_connection_t *or_conn;
  1087. channel_t *chan = NULL;
  1088. int have_any_circuits;
  1089. int past_keepalive =
  1090. now >= conn->timestamp_last_write_allowed + options->KeepalivePeriod;
  1091. if (conn->outbuf && !connection_get_outbuf_len(conn) &&
  1092. conn->type == CONN_TYPE_OR)
  1093. TO_OR_CONN(conn)->timestamp_lastempty = now;
  1094. if (conn->marked_for_close) {
  1095. /* nothing to do here */
  1096. return;
  1097. }
  1098. /* Expire any directory connections that haven't been active (sent
  1099. * if a server or received if a client) for 5 min */
  1100. if (conn->type == CONN_TYPE_DIR &&
  1101. ((DIR_CONN_IS_SERVER(conn) &&
  1102. conn->timestamp_last_write_allowed
  1103. + options->TestingDirConnectionMaxStall < now) ||
  1104. (!DIR_CONN_IS_SERVER(conn) &&
  1105. conn->timestamp_last_read_allowed
  1106. + options->TestingDirConnectionMaxStall < now))) {
  1107. log_info(LD_DIR,"Expiring wedged directory conn (fd %d, purpose %d)",
  1108. (int)conn->s, conn->purpose);
  1109. /* This check is temporary; it's to let us know whether we should consider
  1110. * parsing partial serverdesc responses. */
  1111. if (conn->purpose == DIR_PURPOSE_FETCH_SERVERDESC &&
  1112. connection_get_inbuf_len(conn) >= 1024) {
  1113. log_info(LD_DIR,"Trying to extract information from wedged server desc "
  1114. "download.");
  1115. connection_dir_reached_eof(TO_DIR_CONN(conn));
  1116. } else {
  1117. connection_mark_for_close(conn);
  1118. }
  1119. return;
  1120. }
  1121. if (!connection_speaks_cells(conn))
  1122. return; /* we're all done here, the rest is just for OR conns */
  1123. /* If we haven't flushed to an OR connection for a while, then either nuke
  1124. the connection or send a keepalive, depending. */
  1125. or_conn = TO_OR_CONN(conn);
  1126. //tor_assert(conn->outbuf);
  1127. chan = TLS_CHAN_TO_BASE(or_conn->chan);
  1128. tor_assert(chan);
  1129. if (channel_num_circuits(chan) != 0) {
  1130. have_any_circuits = 1;
  1131. chan->timestamp_last_had_circuits = now;
  1132. } else {
  1133. have_any_circuits = 0;
  1134. }
  1135. if (channel_is_bad_for_new_circs(TLS_CHAN_TO_BASE(or_conn->chan)) &&
  1136. ! have_any_circuits) {
  1137. /* It's bad for new circuits, and has no unmarked circuits on it:
  1138. * mark it now. */
  1139. log_info(LD_OR,
  1140. "Expiring non-used OR connection to fd %d (%s:%d) [Too old].",
  1141. (int)conn->s, conn->address, conn->port);
  1142. if (conn->state == OR_CONN_STATE_CONNECTING)
  1143. connection_or_connect_failed(TO_OR_CONN(conn),
  1144. END_OR_CONN_REASON_TIMEOUT,
  1145. "Tor gave up on the connection");
  1146. connection_or_close_normally(TO_OR_CONN(conn), 1);
  1147. } else if (!connection_state_is_open(conn)) {
  1148. if (past_keepalive) {
  1149. /* We never managed to actually get this connection open and happy. */
  1150. log_info(LD_OR,"Expiring non-open OR connection to fd %d (%s:%d).",
  1151. (int)conn->s,conn->address, conn->port);
  1152. connection_or_close_normally(TO_OR_CONN(conn), 0);
  1153. }
  1154. } else if (we_are_hibernating() &&
  1155. ! have_any_circuits &&
  1156. // !connection_get_outbuf_len(conn)) {
  1157. // TODO: we should have a way to make sure the safe connection has no data to send
  1158. 1) {
  1159. /* We're hibernating or shutting down, there's no circuits, and nothing to
  1160. * flush.*/
  1161. log_info(LD_OR,"Expiring non-used OR connection to fd %d (%s:%d) "
  1162. "[Hibernating or exiting].",
  1163. (int)conn->s,conn->address, conn->port);
  1164. connection_or_close_normally(TO_OR_CONN(conn), 1);
  1165. } else if (!have_any_circuits &&
  1166. now - or_conn->idle_timeout >=
  1167. chan->timestamp_last_had_circuits) {
  1168. log_info(LD_OR,"Expiring non-used OR connection %"PRIu64" to fd %d "
  1169. "(%s:%d) [no circuits for %d; timeout %d; %scanonical].",
  1170. (chan->global_identifier),
  1171. (int)conn->s, conn->address, conn->port,
  1172. (int)(now - chan->timestamp_last_had_circuits),
  1173. or_conn->idle_timeout,
  1174. or_conn->is_canonical ? "" : "non");
  1175. connection_or_close_normally(TO_OR_CONN(conn), 0);
  1176. } else if (
  1177. now >= or_conn->timestamp_lastempty + options->KeepalivePeriod*10 &&
  1178. now >=
  1179. conn->timestamp_last_write_allowed + options->KeepalivePeriod*10) {
  1180. log_fn(LOG_PROTOCOL_WARN,LD_PROTOCOL,
  1181. "Expiring stuck OR connection to fd %d (%s:%d). (%d bytes to "
  1182. "flush; %d seconds since last write)",
  1183. (int)conn->s, conn->address, conn->port,
  1184. //(int)connection_get_outbuf_len(conn),
  1185. 12345678, // TODO: do something sensible here
  1186. (int)(now-conn->timestamp_last_write_allowed));
  1187. connection_or_close_normally(TO_OR_CONN(conn), 0);
  1188. } else if (past_keepalive && 1) { //!connection_get_outbuf_len(conn)) {
  1189. // TODO: we should have a way to make sure the safe connection has no data to send
  1190. /* send a padding cell */
  1191. log_fn(LOG_DEBUG,LD_OR,"Sending keepalive to (%s:%d)",
  1192. conn->address, conn->port);
  1193. cell_t *cell = tor_malloc_zero(sizeof(cell_t));
  1194. //memset(&cell,0,sizeof(cell_t));
  1195. cell->command = CELL_PADDING;
  1196. connection_or_write_cell_to_buf(cell, or_conn);
  1197. } else {
  1198. channelpadding_decide_to_pad_channel(chan);
  1199. }
  1200. }
  1201. /** Honor a NEWNYM request: make future requests unlinkable to past
  1202. * requests. */
  1203. static void
  1204. signewnym_impl(time_t now)
  1205. {
  1206. const or_options_t *options = get_options();
  1207. if (!proxy_mode(options)) {
  1208. log_info(LD_CONTROL, "Ignoring SIGNAL NEWNYM because client functionality "
  1209. "is disabled.");
  1210. return;
  1211. }
  1212. circuit_mark_all_dirty_circs_as_unusable();
  1213. addressmap_clear_transient();
  1214. hs_client_purge_state();
  1215. time_of_last_signewnym = now;
  1216. signewnym_is_pending = 0;
  1217. ++newnym_epoch;
  1218. control_event_signal(SIGNEWNYM);
  1219. }
  1220. /** Callback: run a deferred signewnym. */
  1221. static void
  1222. handle_deferred_signewnym_cb(mainloop_event_t *event, void *arg)
  1223. {
  1224. (void)event;
  1225. (void)arg;
  1226. log_info(LD_CONTROL, "Honoring delayed NEWNYM request");
  1227. do_signewnym(time(NULL));
  1228. }
  1229. /** Either perform a signewnym or schedule one, depending on rate limiting. */
  1230. void
  1231. do_signewnym(time_t now)
  1232. {
  1233. if (time_of_last_signewnym + MAX_SIGNEWNYM_RATE > now) {
  1234. const time_t delay_sec =
  1235. time_of_last_signewnym + MAX_SIGNEWNYM_RATE - now;
  1236. if (! signewnym_is_pending) {
  1237. signewnym_is_pending = 1;
  1238. if (!handle_deferred_signewnym_ev) {
  1239. handle_deferred_signewnym_ev =
  1240. mainloop_event_postloop_new(handle_deferred_signewnym_cb, NULL);
  1241. }
  1242. const struct timeval delay_tv = { delay_sec, 0 };
  1243. mainloop_event_schedule(handle_deferred_signewnym_ev, &delay_tv);
  1244. }
  1245. log_notice(LD_CONTROL,
  1246. "Rate limiting NEWNYM request: delaying by %d second(s)",
  1247. (int)(delay_sec));
  1248. } else {
  1249. signewnym_impl(now);
  1250. }
  1251. }
  1252. /** Return the number of times that signewnym has been called. */
  1253. unsigned
  1254. get_signewnym_epoch(void)
  1255. {
  1256. return newnym_epoch;
  1257. }
  1258. /** True iff we have initialized all the members of <b>periodic_events</b>.
  1259. * Used to prevent double-initialization. */
  1260. static int periodic_events_initialized = 0;
  1261. /* Declare all the timer callback functions... */
  1262. #undef CALLBACK
  1263. #define CALLBACK(name) \
  1264. static int name ## _callback(time_t, const or_options_t *)
  1265. CALLBACK(add_entropy);
  1266. CALLBACK(check_expired_networkstatus);
  1267. CALLBACK(clean_caches);
  1268. CALLBACK(clean_consdiffmgr);
  1269. CALLBACK(fetch_networkstatus);
  1270. CALLBACK(heartbeat);
  1271. CALLBACK(hs_service);
  1272. CALLBACK(launch_descriptor_fetches);
  1273. CALLBACK(prune_old_routers);
  1274. CALLBACK(record_bridge_stats);
  1275. CALLBACK(rend_cache_failure_clean);
  1276. CALLBACK(reset_padding_counts);
  1277. CALLBACK(retry_listeners);
  1278. CALLBACK(rotate_x509_certificate);
  1279. CALLBACK(save_state);
  1280. CALLBACK(write_stats_file);
  1281. CALLBACK(control_per_second_events);
  1282. CALLBACK(second_elapsed);
  1283. #undef CALLBACK
  1284. /* Now we declare an array of periodic_event_item_t for each periodic event */
  1285. #define CALLBACK(name, r, f) \
  1286. PERIODIC_EVENT(name, PERIODIC_EVENT_ROLE_ ## r, f)
  1287. #define FL(name) (PERIODIC_EVENT_FLAG_ ## name)
  1288. STATIC periodic_event_item_t mainloop_periodic_events[] = {
  1289. /* Everyone needs to run these. They need to have very long timeouts for
  1290. * that to be safe. */
  1291. CALLBACK(add_entropy, ALL, 0),
  1292. CALLBACK(heartbeat, ALL, 0),
  1293. CALLBACK(reset_padding_counts, ALL, 0),
  1294. /* This is a legacy catch-all callback that runs once per second if
  1295. * we are online and active. */
  1296. CALLBACK(second_elapsed, NET_PARTICIPANT,
  1297. FL(RUN_ON_DISABLE)),
  1298. /* XXXX Do we have a reason to do this on a callback? Does it do any good at
  1299. * all? For now, if we're dormant, we can let our listeners decay. */
  1300. CALLBACK(retry_listeners, NET_PARTICIPANT, FL(NEED_NET)),
  1301. /* We need to do these if we're participating in the Tor network. */
  1302. CALLBACK(check_expired_networkstatus, NET_PARTICIPANT, 0),
  1303. CALLBACK(fetch_networkstatus, NET_PARTICIPANT, 0),
  1304. CALLBACK(launch_descriptor_fetches, NET_PARTICIPANT, FL(NEED_NET)),
  1305. CALLBACK(rotate_x509_certificate, NET_PARTICIPANT, 0),
  1306. CALLBACK(check_network_participation, NET_PARTICIPANT, 0),
  1307. /* We need to do these if we're participating in the Tor network, and
  1308. * immediately before we stop. */
  1309. CALLBACK(clean_caches, NET_PARTICIPANT, FL(RUN_ON_DISABLE)),
  1310. CALLBACK(save_state, NET_PARTICIPANT, FL(RUN_ON_DISABLE)),
  1311. CALLBACK(write_stats_file, NET_PARTICIPANT, FL(RUN_ON_DISABLE)),
  1312. CALLBACK(prune_old_routers, NET_PARTICIPANT, FL(RUN_ON_DISABLE)),
  1313. /* Hidden Service service only. */
  1314. CALLBACK(hs_service, HS_SERVICE, FL(NEED_NET)), // XXXX break this down more
  1315. /* Bridge only. */
  1316. CALLBACK(record_bridge_stats, BRIDGE, 0),
  1317. /* Client only. */
  1318. /* XXXX this could be restricted to CLIENT+NET_PARTICIPANT */
  1319. CALLBACK(rend_cache_failure_clean, NET_PARTICIPANT, FL(RUN_ON_DISABLE)),
  1320. /* Directory server only. */
  1321. CALLBACK(clean_consdiffmgr, DIRSERVER, 0),
  1322. /* Controller with per-second events only. */
  1323. CALLBACK(control_per_second_events, CONTROLEV, 0),
  1324. END_OF_PERIODIC_EVENTS
  1325. };
  1326. #undef CALLBACK
  1327. #undef FL
  1328. /* These are pointers to members of periodic_events[] that are used to
  1329. * implement particular callbacks. We keep them separate here so that we
  1330. * can access them by name. We also keep them inside periodic_events[]
  1331. * so that we can implement "reset all timers" in a reasonable way. */
  1332. static periodic_event_item_t *fetch_networkstatus_event=NULL;
  1333. static periodic_event_item_t *launch_descriptor_fetches_event=NULL;
  1334. static periodic_event_item_t *check_dns_honesty_event=NULL;
  1335. static periodic_event_item_t *save_state_event=NULL;
  1336. static periodic_event_item_t *prune_old_routers_event=NULL;
  1337. /** Reset all the periodic events so we'll do all our actions again as if we
  1338. * just started up.
  1339. * Useful if our clock just moved back a long time from the future,
  1340. * so we don't wait until that future arrives again before acting.
  1341. */
  1342. void
  1343. reset_all_main_loop_timers(void)
  1344. {
  1345. periodic_events_reset_all();
  1346. }
  1347. /** Return a bitmask of the roles this tor instance is configured for using
  1348. * the given options. */
  1349. STATIC int
  1350. get_my_roles(const or_options_t *options)
  1351. {
  1352. tor_assert(options);
  1353. int roles = PERIODIC_EVENT_ROLE_ALL;
  1354. int is_bridge = options->BridgeRelay;
  1355. int is_relay = server_mode(options);
  1356. int is_dirauth = authdir_mode_v3(options);
  1357. int is_bridgeauth = authdir_mode_bridge(options);
  1358. int is_hidden_service = !!hs_service_get_num_services() ||
  1359. !!rend_num_services();
  1360. int is_dirserver = dir_server_mode(options);
  1361. int sending_control_events = control_any_per_second_event_enabled();
  1362. /* We also consider tor to have the role of a client if the ControlPort is
  1363. * set because a lot of things can be done over the control port which
  1364. * requires tor to have basic functionnalities. */
  1365. int is_client = options_any_client_port_set(options) ||
  1366. options->ControlPort_set ||
  1367. options->OwningControllerFD != UINT64_MAX;
  1368. int is_net_participant = is_participating_on_network() ||
  1369. is_relay || is_hidden_service;
  1370. if (is_bridge) roles |= PERIODIC_EVENT_ROLE_BRIDGE;
  1371. if (is_client) roles |= PERIODIC_EVENT_ROLE_CLIENT;
  1372. if (is_relay) roles |= PERIODIC_EVENT_ROLE_RELAY;
  1373. if (is_dirauth) roles |= PERIODIC_EVENT_ROLE_DIRAUTH;
  1374. if (is_bridgeauth) roles |= PERIODIC_EVENT_ROLE_BRIDGEAUTH;
  1375. if (is_hidden_service) roles |= PERIODIC_EVENT_ROLE_HS_SERVICE;
  1376. if (is_dirserver) roles |= PERIODIC_EVENT_ROLE_DIRSERVER;
  1377. if (is_net_participant) roles |= PERIODIC_EVENT_ROLE_NET_PARTICIPANT;
  1378. if (sending_control_events) roles |= PERIODIC_EVENT_ROLE_CONTROLEV;
  1379. return roles;
  1380. }
  1381. /** Event to run initialize_periodic_events_cb */
  1382. static struct event *initialize_periodic_events_event = NULL;
  1383. /** Helper, run one second after setup:
  1384. * Initializes all members of periodic_events and starts them running.
  1385. *
  1386. * (We do this one second after setup for backward-compatibility reasons;
  1387. * it might not actually be necessary.) */
  1388. static void
  1389. initialize_periodic_events_cb(evutil_socket_t fd, short events, void *data)
  1390. {
  1391. (void) fd;
  1392. (void) events;
  1393. (void) data;
  1394. tor_event_free(initialize_periodic_events_event);
  1395. rescan_periodic_events(get_options());
  1396. }
  1397. /** Set up all the members of mainloop_periodic_events[], and configure them
  1398. * all to be launched from a callback. */
  1399. void
  1400. initialize_periodic_events(void)
  1401. {
  1402. if (periodic_events_initialized)
  1403. return;
  1404. periodic_events_initialized = 1;
  1405. for (int i = 0; mainloop_periodic_events[i].name; ++i) {
  1406. periodic_events_register(&mainloop_periodic_events[i]);
  1407. }
  1408. /* Set up all periodic events. We'll launch them by roles. */
  1409. #define NAMED_CALLBACK(name) \
  1410. STMT_BEGIN name ## _event = periodic_events_find( #name ); STMT_END
  1411. NAMED_CALLBACK(prune_old_routers);
  1412. NAMED_CALLBACK(fetch_networkstatus);
  1413. NAMED_CALLBACK(launch_descriptor_fetches);
  1414. NAMED_CALLBACK(check_dns_honesty);
  1415. NAMED_CALLBACK(save_state);
  1416. }
  1417. STATIC void
  1418. teardown_periodic_events(void)
  1419. {
  1420. periodic_events_disconnect_all();
  1421. fetch_networkstatus_event = NULL;
  1422. launch_descriptor_fetches_event = NULL;
  1423. check_dns_honesty_event = NULL;
  1424. save_state_event = NULL;
  1425. prune_old_routers_event = NULL;
  1426. periodic_events_initialized = 0;
  1427. }
  1428. static mainloop_event_t *rescan_periodic_events_ev = NULL;
  1429. /** Callback: rescan the periodic event list. */
  1430. static void
  1431. rescan_periodic_events_cb(mainloop_event_t *event, void *arg)
  1432. {
  1433. (void)event;
  1434. (void)arg;
  1435. rescan_periodic_events(get_options());
  1436. }
  1437. /**
  1438. * Schedule an event that will rescan which periodic events should run.
  1439. **/
  1440. MOCK_IMPL(void,
  1441. schedule_rescan_periodic_events,(void))
  1442. {
  1443. if (!rescan_periodic_events_ev) {
  1444. rescan_periodic_events_ev =
  1445. mainloop_event_new(rescan_periodic_events_cb, NULL);
  1446. }
  1447. mainloop_event_activate(rescan_periodic_events_ev);
  1448. }
  1449. /** Do a pass at all our periodic events, disable those we don't need anymore
  1450. * and enable those we need now using the given options. */
  1451. void
  1452. rescan_periodic_events(const or_options_t *options)
  1453. {
  1454. tor_assert(options);
  1455. periodic_events_rescan_by_roles(get_my_roles(options), net_is_disabled());
  1456. }
  1457. /* We just got new options globally set, see if we need to enabled or disable
  1458. * periodic events. */
  1459. void
  1460. periodic_events_on_new_options(const or_options_t *options)
  1461. {
  1462. rescan_periodic_events(options);
  1463. }
  1464. /**
  1465. * Update our schedule so that we'll check whether we need to fetch directory
  1466. * info immediately.
  1467. */
  1468. void
  1469. reschedule_directory_downloads(void)
  1470. {
  1471. tor_assert(fetch_networkstatus_event);
  1472. tor_assert(launch_descriptor_fetches_event);
  1473. periodic_event_reschedule(fetch_networkstatus_event);
  1474. periodic_event_reschedule(launch_descriptor_fetches_event);
  1475. }
  1476. /** Mainloop callback: clean up circuits, channels, and connections
  1477. * that are pending close. */
  1478. static void
  1479. postloop_cleanup_cb(mainloop_event_t *ev, void *arg)
  1480. {
  1481. (void)ev;
  1482. (void)arg;
  1483. circuit_close_all_marked();
  1484. close_closeable_connections();
  1485. channel_run_cleanup();
  1486. channel_listener_run_cleanup();
  1487. }
  1488. /** Event to run postloop_cleanup_cb */
  1489. static mainloop_event_t *postloop_cleanup_ev=NULL;
  1490. /** Schedule a post-loop event to clean up marked channels, connections, and
  1491. * circuits. */
  1492. void
  1493. mainloop_schedule_postloop_cleanup(void)
  1494. {
  1495. if (PREDICT_UNLIKELY(postloop_cleanup_ev == NULL)) {
  1496. // (It's possible that we can get here if we decide to close a connection
  1497. // in the earliest stages of our configuration, before we create events.)
  1498. return;
  1499. }
  1500. mainloop_event_activate(postloop_cleanup_ev);
  1501. }
  1502. /** Event to run 'scheduled_shutdown_cb' */
  1503. static mainloop_event_t *scheduled_shutdown_ev=NULL;
  1504. /** Callback: run a scheduled shutdown */
  1505. static void
  1506. scheduled_shutdown_cb(mainloop_event_t *ev, void *arg)
  1507. {
  1508. (void)ev;
  1509. (void)arg;
  1510. log_notice(LD_GENERAL, "Clean shutdown finished. Exiting.");
  1511. tor_shutdown_event_loop_and_exit(0);
  1512. }
  1513. /** Schedule the mainloop to exit after <b>delay_sec</b> seconds. */
  1514. void
  1515. mainloop_schedule_shutdown(int delay_sec)
  1516. {
  1517. const struct timeval delay_tv = { delay_sec, 0 };
  1518. if (! scheduled_shutdown_ev) {
  1519. scheduled_shutdown_ev = mainloop_event_new(scheduled_shutdown_cb, NULL);
  1520. }
  1521. mainloop_event_schedule(scheduled_shutdown_ev, &delay_tv);
  1522. }
  1523. /** Perform regular maintenance tasks. This function gets run once per
  1524. * second.
  1525. */
  1526. static int
  1527. second_elapsed_callback(time_t now, const or_options_t *options)
  1528. {
  1529. /* 0. See if our bandwidth limits are exhausted and we should hibernate
  1530. *
  1531. * Note: we have redundant mechanisms to handle the case where it's
  1532. * time to wake up from hibernation; or where we have a scheduled
  1533. * shutdown and it's time to run it, but this will also handle those.
  1534. */
  1535. consider_hibernation(now);
  1536. /* Maybe enough time elapsed for us to reconsider a circuit. */
  1537. circuit_upgrade_circuits_from_guard_wait();
  1538. if (options->UseBridges && !net_is_disabled()) {
  1539. /* Note: this check uses net_is_disabled(), not should_delay_dir_fetches()
  1540. * -- the latter is only for fetching consensus-derived directory info. */
  1541. // TODO: client
  1542. // Also, schedule this rather than probing 1x / sec
  1543. fetch_bridge_descriptors(options, now);
  1544. }
  1545. if (accounting_is_enabled(options)) {
  1546. // TODO: refactor or rewrite?
  1547. accounting_run_housekeeping(now);
  1548. }
  1549. /* 3a. Every second, we examine pending circuits and prune the
  1550. * ones which have been pending for more than a few seconds.
  1551. * We do this before step 4, so it can try building more if
  1552. * it's not comfortable with the number of available circuits.
  1553. */
  1554. /* (If our circuit build timeout can ever become lower than a second (which
  1555. * it can't, currently), we should do this more often.) */
  1556. // TODO: All expire stuff can become NET_PARTICIPANT, RUN_ON_DISABLE
  1557. circuit_expire_building();
  1558. circuit_expire_waiting_for_better_guard();
  1559. /* 3b. Also look at pending streams and prune the ones that 'began'
  1560. * a long time ago but haven't gotten a 'connected' yet.
  1561. * Do this before step 4, so we can put them back into pending
  1562. * state to be picked up by the new circuit.
  1563. */
  1564. connection_ap_expire_beginning();
  1565. /* 3c. And expire connections that we've held open for too long.
  1566. */
  1567. connection_expire_held_open();
  1568. /* 4. Every second, we try a new circuit if there are no valid
  1569. * circuits. Every NewCircuitPeriod seconds, we expire circuits
  1570. * that became dirty more than MaxCircuitDirtiness seconds ago,
  1571. * and we make a new circ if there are no clean circuits.
  1572. */
  1573. const int have_dir_info = router_have_minimum_dir_info();
  1574. if (have_dir_info && !net_is_disabled()) {
  1575. circuit_build_needed_circs(now);
  1576. } else {
  1577. circuit_expire_old_circs_as_needed(now);
  1578. }
  1579. /* 5. We do housekeeping for each connection... */
  1580. channel_update_bad_for_new_circs(NULL, 0);
  1581. int i;
  1582. for (i=0;i<smartlist_len(connection_array);i++) {
  1583. run_connection_housekeeping(i, now);
  1584. }
  1585. /* Run again in a second. */
  1586. return 1;
  1587. }
  1588. /**
  1589. * Periodic callback: Every {LAZY,GREEDY}_DESCRIPTOR_RETRY_INTERVAL,
  1590. * see about fetching descriptors, microdescriptors, and extrainfo
  1591. * documents.
  1592. */
  1593. static int
  1594. launch_descriptor_fetches_callback(time_t now, const or_options_t *options)
  1595. {
  1596. if (should_delay_dir_fetches(options, NULL))
  1597. return PERIODIC_EVENT_NO_UPDATE;
  1598. update_all_descriptor_downloads(now);
  1599. update_extrainfo_downloads(now);
  1600. if (router_have_minimum_dir_info())
  1601. return LAZY_DESCRIPTOR_RETRY_INTERVAL;
  1602. else
  1603. return GREEDY_DESCRIPTOR_RETRY_INTERVAL;
  1604. }
  1605. /**
  1606. * Periodic event: Rotate our X.509 certificates and TLS keys once every
  1607. * MAX_SSL_KEY_LIFETIME_INTERNAL.
  1608. */
  1609. static int
  1610. rotate_x509_certificate_callback(time_t now, const or_options_t *options)
  1611. {
  1612. static int first = 1;
  1613. (void)now;
  1614. (void)options;
  1615. if (first) {
  1616. first = 0;
  1617. return MAX_SSL_KEY_LIFETIME_INTERNAL;
  1618. }
  1619. /* 1b. Every MAX_SSL_KEY_LIFETIME_INTERNAL seconds, we change our
  1620. * TLS context. */
  1621. log_info(LD_GENERAL,"Rotating tls context.");
  1622. if (router_initialize_tls_context() < 0) {
  1623. log_err(LD_BUG, "Error reinitializing TLS context");
  1624. tor_assert_unreached();
  1625. }
  1626. if (generate_ed_link_cert(options, now, 1)) {
  1627. log_err(LD_OR, "Unable to update Ed25519->TLS link certificate for "
  1628. "new TLS context.");
  1629. tor_assert_unreached();
  1630. }
  1631. /* We also make sure to rotate the TLS connections themselves if they've
  1632. * been up for too long -- but that's done via is_bad_for_new_circs in
  1633. * run_connection_housekeeping() above. */
  1634. return MAX_SSL_KEY_LIFETIME_INTERNAL;
  1635. }
  1636. /**
  1637. * Periodic callback: once an hour, grab some more entropy from the
  1638. * kernel and feed it to our CSPRNG.
  1639. **/
  1640. static int
  1641. add_entropy_callback(time_t now, const or_options_t *options)
  1642. {
  1643. (void)now;
  1644. (void)options;
  1645. /* We already seeded once, so don't die on failure. */
  1646. if (crypto_seed_rng() < 0) {
  1647. log_warn(LD_GENERAL, "Tried to re-seed RNG, but failed. We already "
  1648. "seeded once, though, so we won't exit here.");
  1649. }
  1650. /** How often do we add more entropy to OpenSSL's RNG pool? */
  1651. #define ENTROPY_INTERVAL (60*60)
  1652. return ENTROPY_INTERVAL;
  1653. }
  1654. /** Periodic callback: if there has been no network usage in a while,
  1655. * enter a dormant state. */
  1656. STATIC int
  1657. check_network_participation_callback(time_t now, const or_options_t *options)
  1658. {
  1659. /* If we're a server, we can't become dormant. */
  1660. if (server_mode(options)) {
  1661. goto found_activity;
  1662. }
  1663. /* If we're running an onion service, we can't become dormant. */
  1664. /* XXXX this would be nice to change, so that we can be dormant with a
  1665. * service. */
  1666. if (hs_service_get_num_services() || rend_num_services()) {
  1667. goto found_activity;
  1668. }
  1669. /* If we have any currently open entry streams other than "linked"
  1670. * connections used for directory requests, those count as user activity.
  1671. */
  1672. if (options->DormantTimeoutDisabledByIdleStreams) {
  1673. if (connection_get_by_type_nonlinked(CONN_TYPE_AP) != NULL) {
  1674. goto found_activity;
  1675. }
  1676. }
  1677. /* XXXX Make this configurable? */
  1678. /** How often do we check whether we have had network activity? */
  1679. #define CHECK_PARTICIPATION_INTERVAL (5*60)
  1680. /* Become dormant if there has been no user activity in a long time.
  1681. * (The funny checks below are in order to prevent overflow.) */
  1682. time_t time_since_last_activity = 0;
  1683. if (get_last_user_activity_time() < now)
  1684. time_since_last_activity = now - get_last_user_activity_time();
  1685. if (time_since_last_activity >= options->DormantClientTimeout) {
  1686. log_notice(LD_GENERAL, "No user activity in a long time: becoming"
  1687. " dormant.");
  1688. set_network_participation(false);
  1689. rescan_periodic_events(options);
  1690. }
  1691. return CHECK_PARTICIPATION_INTERVAL;
  1692. found_activity:
  1693. note_user_activity(now);
  1694. return CHECK_PARTICIPATION_INTERVAL;
  1695. }
  1696. /**
  1697. * Periodic callback: If our consensus is too old, recalculate whether
  1698. * we can actually use it.
  1699. */
  1700. static int
  1701. check_expired_networkstatus_callback(time_t now, const or_options_t *options)
  1702. {
  1703. (void)options;
  1704. /* Check whether our networkstatus has expired. */
  1705. networkstatus_t *ns = networkstatus_get_latest_consensus();
  1706. /* Use reasonably live consensuses until they are no longer reasonably live.
  1707. */
  1708. if (ns && !networkstatus_consensus_reasonably_live(ns, now) &&
  1709. router_have_minimum_dir_info()) {
  1710. router_dir_info_changed();
  1711. }
  1712. #define CHECK_EXPIRED_NS_INTERVAL (2*60)
  1713. return CHECK_EXPIRED_NS_INTERVAL;
  1714. }
  1715. /**
  1716. * Scheduled callback: Save the state file to disk if appropriate.
  1717. */
  1718. static int
  1719. save_state_callback(time_t now, const or_options_t *options)
  1720. {
  1721. (void) options;
  1722. (void) or_state_save(now); // only saves if appropriate
  1723. const time_t next_write = get_or_state()->next_write;
  1724. if (next_write == TIME_MAX) {
  1725. return 86400;
  1726. }
  1727. return safe_timer_diff(now, next_write);
  1728. }
  1729. /** Reschedule the event for saving the state file.
  1730. *
  1731. * Run this when the state becomes dirty. */
  1732. void
  1733. reschedule_or_state_save(void)
  1734. {
  1735. if (save_state_event == NULL) {
  1736. /* This can happen early on during startup. */
  1737. return;
  1738. }
  1739. periodic_event_reschedule(save_state_event);
  1740. }
  1741. /**
  1742. * Periodic callback: Write statistics to disk if appropriate.
  1743. */
  1744. static int
  1745. write_stats_file_callback(time_t now, const or_options_t *options)
  1746. {
  1747. /* 1g. Check whether we should write statistics to disk.
  1748. */
  1749. #define CHECK_WRITE_STATS_INTERVAL (60*60)
  1750. time_t next_time_to_write_stats_files = now + CHECK_WRITE_STATS_INTERVAL;
  1751. if (options->CellStatistics) {
  1752. time_t next_write =
  1753. rep_hist_buffer_stats_write(now);
  1754. if (next_write && next_write < next_time_to_write_stats_files)
  1755. next_time_to_write_stats_files = next_write;
  1756. }
  1757. if (options->DirReqStatistics) {
  1758. time_t next_write = geoip_dirreq_stats_write(now);
  1759. if (next_write && next_write < next_time_to_write_stats_files)
  1760. next_time_to_write_stats_files = next_write;
  1761. }
  1762. if (options->EntryStatistics) {
  1763. time_t next_write = geoip_entry_stats_write(now);
  1764. if (next_write && next_write < next_time_to_write_stats_files)
  1765. next_time_to_write_stats_files = next_write;
  1766. }
  1767. if (options->HiddenServiceStatistics) {
  1768. time_t next_write = rep_hist_hs_stats_write(now);
  1769. if (next_write && next_write < next_time_to_write_stats_files)
  1770. next_time_to_write_stats_files = next_write;
  1771. }
  1772. if (options->ExitPortStatistics) {
  1773. time_t next_write = rep_hist_exit_stats_write(now);
  1774. if (next_write && next_write < next_time_to_write_stats_files)
  1775. next_time_to_write_stats_files = next_write;
  1776. }
  1777. if (options->ConnDirectionStatistics) {
  1778. time_t next_write = rep_hist_conn_stats_write(now);
  1779. if (next_write && next_write < next_time_to_write_stats_files)
  1780. next_time_to_write_stats_files = next_write;
  1781. }
  1782. if (options->BridgeAuthoritativeDir) {
  1783. time_t next_write = rep_hist_desc_stats_write(now);
  1784. if (next_write && next_write < next_time_to_write_stats_files)
  1785. next_time_to_write_stats_files = next_write;
  1786. }
  1787. return safe_timer_diff(now, next_time_to_write_stats_files);
  1788. }
  1789. static int
  1790. reset_padding_counts_callback(time_t now, const or_options_t *options)
  1791. {
  1792. if (options->PaddingStatistics) {
  1793. rep_hist_prep_published_padding_counts(now);
  1794. }
  1795. rep_hist_reset_padding_counts();
  1796. return REPHIST_CELL_PADDING_COUNTS_INTERVAL;
  1797. }
  1798. static int should_init_bridge_stats = 1;
  1799. /**
  1800. * Periodic callback: Write bridge statistics to disk if appropriate.
  1801. */
  1802. static int
  1803. record_bridge_stats_callback(time_t now, const or_options_t *options)
  1804. {
  1805. /* 1h. Check whether we should write bridge statistics to disk.
  1806. */
  1807. if (should_record_bridge_info(options)) {
  1808. if (should_init_bridge_stats) {
  1809. /* (Re-)initialize bridge statistics. */
  1810. geoip_bridge_stats_init(now);
  1811. should_init_bridge_stats = 0;
  1812. return WRITE_STATS_INTERVAL;
  1813. } else {
  1814. /* Possibly write bridge statistics to disk and ask when to write
  1815. * them next time. */
  1816. time_t next = geoip_bridge_stats_write(now);
  1817. return safe_timer_diff(now, next);
  1818. }
  1819. } else if (!should_init_bridge_stats) {
  1820. /* Bridge mode was turned off. Ensure that stats are re-initialized
  1821. * next time bridge mode is turned on. */
  1822. should_init_bridge_stats = 1;
  1823. }
  1824. return PERIODIC_EVENT_NO_UPDATE;
  1825. }
  1826. /**
  1827. * Periodic callback: Clean in-memory caches every once in a while
  1828. */
  1829. static int
  1830. clean_caches_callback(time_t now, const or_options_t *options)
  1831. {
  1832. /* Remove old information from rephist and the rend cache. */
  1833. rep_history_clean(now - options->RephistTrackTime);
  1834. rend_cache_clean(now, REND_CACHE_TYPE_SERVICE);
  1835. hs_cache_clean_as_client(now);
  1836. hs_cache_clean_as_dir(now);
  1837. microdesc_cache_rebuild(NULL, 0);
  1838. #define CLEAN_CACHES_INTERVAL (30*60)
  1839. return CLEAN_CACHES_INTERVAL;
  1840. }
  1841. /**
  1842. * Periodic callback: Clean the cache of failed hidden service lookups
  1843. * frequently.
  1844. */
  1845. static int
  1846. rend_cache_failure_clean_callback(time_t now, const or_options_t *options)
  1847. {
  1848. (void)options;
  1849. /* We don't keep entries that are more than five minutes old so we try to
  1850. * clean it as soon as we can since we want to make sure the client waits
  1851. * as little as possible for reachability reasons. */
  1852. rend_cache_failure_clean(now);
  1853. hs_cache_client_intro_state_clean(now);
  1854. return 30;
  1855. }
  1856. /**
  1857. * Periodic callback: prune routerlist of old information about Tor network.
  1858. */
  1859. static int
  1860. prune_old_routers_callback(time_t now, const or_options_t *options)
  1861. {
  1862. #define ROUTERLIST_PRUNING_INTERVAL (60*60) // 1 hour.
  1863. (void)now;
  1864. (void)options;
  1865. if (!net_is_disabled()) {
  1866. /* If any networkstatus documents are no longer recent, we need to
  1867. * update all the descriptors' running status. */
  1868. /* Remove dead routers. */
  1869. log_debug(LD_GENERAL, "Pruning routerlist...");
  1870. routerlist_remove_old_routers();
  1871. }
  1872. return ROUTERLIST_PRUNING_INTERVAL;
  1873. }
  1874. /**
  1875. * Periodic event: once a minute, (or every second if TestingTorNetwork, or
  1876. * during client bootstrap), check whether we want to download any
  1877. * networkstatus documents. */
  1878. static int
  1879. fetch_networkstatus_callback(time_t now, const or_options_t *options)
  1880. {
  1881. /* How often do we check whether we should download network status
  1882. * documents? */
  1883. const int we_are_bootstrapping = networkstatus_consensus_is_bootstrapping(
  1884. now);
  1885. const int prefer_mirrors = !directory_fetches_from_authorities(
  1886. get_options());
  1887. int networkstatus_dl_check_interval = 60;
  1888. /* check more often when testing, or when bootstrapping from mirrors
  1889. * (connection limits prevent too many connections being made) */
  1890. if (options->TestingTorNetwork
  1891. || (we_are_bootstrapping && prefer_mirrors)) {
  1892. networkstatus_dl_check_interval = 1;
  1893. }
  1894. if (should_delay_dir_fetches(options, NULL))
  1895. return PERIODIC_EVENT_NO_UPDATE;
  1896. update_networkstatus_downloads(now);
  1897. return networkstatus_dl_check_interval;
  1898. }
  1899. /**
  1900. * Periodic callback: Every 60 seconds, we relaunch listeners if any died. */
  1901. static int
  1902. retry_listeners_callback(time_t now, const or_options_t *options)
  1903. {
  1904. (void)now;
  1905. (void)options;
  1906. if (!net_is_disabled()) {
  1907. retry_all_listeners(NULL, 0);
  1908. return 60;
  1909. }
  1910. return PERIODIC_EVENT_NO_UPDATE;
  1911. }
  1912. static int heartbeat_callback_first_time = 1;
  1913. /**
  1914. * Periodic callback: write the heartbeat message in the logs.
  1915. *
  1916. * If writing the heartbeat message to the logs fails for some reason, retry
  1917. * again after <b>MIN_HEARTBEAT_PERIOD</b> seconds.
  1918. */
  1919. static int
  1920. heartbeat_callback(time_t now, const or_options_t *options)
  1921. {
  1922. /* Check if heartbeat is disabled */
  1923. if (!options->HeartbeatPeriod) {
  1924. return PERIODIC_EVENT_NO_UPDATE;
  1925. }
  1926. /* Skip the first one. */
  1927. if (heartbeat_callback_first_time) {
  1928. heartbeat_callback_first_time = 0;
  1929. return options->HeartbeatPeriod;
  1930. }
  1931. /* Write the heartbeat message */
  1932. if (log_heartbeat(now) == 0) {
  1933. return options->HeartbeatPeriod;
  1934. } else {
  1935. /* If we couldn't write the heartbeat log message, try again in the minimum
  1936. * interval of time. */
  1937. return MIN_HEARTBEAT_PERIOD;
  1938. }
  1939. }
  1940. #define CDM_CLEAN_CALLBACK_INTERVAL 600
  1941. static int
  1942. clean_consdiffmgr_callback(time_t now, const or_options_t *options)
  1943. {
  1944. (void)now;
  1945. if (dir_server_mode(options)) {
  1946. consdiffmgr_cleanup();
  1947. }
  1948. return CDM_CLEAN_CALLBACK_INTERVAL;
  1949. }
  1950. /*
  1951. * Periodic callback: Run scheduled events for HS service. This is called
  1952. * every second.
  1953. */
  1954. static int
  1955. hs_service_callback(time_t now, const or_options_t *options)
  1956. {
  1957. (void) options;
  1958. /* We need to at least be able to build circuits and that we actually have
  1959. * a working network. */
  1960. if (!have_completed_a_circuit() || net_is_disabled() ||
  1961. networkstatus_get_live_consensus(now) == NULL) {
  1962. goto end;
  1963. }
  1964. hs_service_run_scheduled_events(now);
  1965. end:
  1966. /* Every 1 second. */
  1967. return 1;
  1968. }
  1969. /*
  1970. * Periodic callback: Send once-per-second events to the controller(s).
  1971. * This is called every second.
  1972. */
  1973. static int
  1974. control_per_second_events_callback(time_t now, const or_options_t *options)
  1975. {
  1976. (void) options;
  1977. (void) now;
  1978. control_per_second_events();
  1979. return 1;
  1980. }
  1981. /** Last time that update_current_time was called. */
  1982. static time_t current_second = 0;
  1983. /** Last time that update_current_time updated current_second. */
  1984. static monotime_coarse_t current_second_last_changed;
  1985. /**
  1986. * Set the current time to "now", which should be the value returned by
  1987. * time(). Check for clock jumps and track the total number of seconds we
  1988. * have been running.
  1989. */
  1990. void
  1991. update_current_time(time_t now)
  1992. {
  1993. if (PREDICT_LIKELY(now == current_second)) {
  1994. /* We call this function a lot. Most frequently, the current second
  1995. * will not have changed, so we just return. */
  1996. return;
  1997. }
  1998. const time_t seconds_elapsed = current_second ? (now - current_second) : 0;
  1999. /* Check the wall clock against the monotonic clock, so we can
  2000. * better tell idleness from clock jumps and/or other shenanigans. */
  2001. monotime_coarse_t last_updated;
  2002. memcpy(&last_updated, &current_second_last_changed, sizeof(last_updated));
  2003. monotime_coarse_get(&current_second_last_changed);
  2004. /** How much clock jumping means that we should adjust our idea of when
  2005. * to go dormant? */
  2006. #define NUM_JUMPED_SECONDS_BEFORE_NETSTATUS_UPDATE 20
  2007. /* Don't go dormant early or late just because we jumped in time. */
  2008. if (ABS(seconds_elapsed) >= NUM_JUMPED_SECONDS_BEFORE_NETSTATUS_UPDATE) {
  2009. if (is_participating_on_network()) {
  2010. netstatus_note_clock_jumped(seconds_elapsed);
  2011. }
  2012. }
  2013. /** How much clock jumping do we tolerate? */
  2014. #define NUM_JUMPED_SECONDS_BEFORE_WARN 100
  2015. /** How much idleness do we tolerate? */
  2016. #define NUM_IDLE_SECONDS_BEFORE_WARN 3600
  2017. if (seconds_elapsed < -NUM_JUMPED_SECONDS_BEFORE_WARN) {
  2018. // moving back in time is always a bad sign.
  2019. circuit_note_clock_jumped(seconds_elapsed, false);
  2020. } else if (seconds_elapsed >= NUM_JUMPED_SECONDS_BEFORE_WARN) {
  2021. /* Compare the monotonic clock to the result of time(). */
  2022. const int32_t monotime_msec_passed =
  2023. monotime_coarse_diff_msec32(&last_updated,
  2024. &current_second_last_changed);
  2025. const int monotime_sec_passed = monotime_msec_passed / 1000;
  2026. const int discrepancy = monotime_sec_passed - (int)seconds_elapsed;
  2027. /* If the monotonic clock deviates from time(NULL), we have a couple of
  2028. * possibilities. On some systems, this means we have been suspended or
  2029. * sleeping. Everywhere, it can mean that the wall-clock time has
  2030. * been changed -- for example, with settimeofday().
  2031. *
  2032. * On the other hand, if the monotonic time matches with the wall-clock
  2033. * time, we've probably just been idle for a while, with no events firing.
  2034. * we tolerate much more of that.
  2035. */
  2036. const bool clock_jumped = abs(discrepancy) > 2;
  2037. if (clock_jumped || seconds_elapsed >= NUM_IDLE_SECONDS_BEFORE_WARN) {
  2038. circuit_note_clock_jumped(seconds_elapsed, ! clock_jumped);
  2039. }
  2040. } else if (seconds_elapsed > 0) {
  2041. stats_n_seconds_working += seconds_elapsed;
  2042. }
  2043. update_approx_time(now);
  2044. current_second = now;
  2045. }
  2046. #ifdef HAVE_SYSTEMD_209
  2047. static periodic_timer_t *systemd_watchdog_timer = NULL;
  2048. /** Libevent callback: invoked to reset systemd watchdog. */
  2049. static void
  2050. systemd_watchdog_callback(periodic_timer_t *timer, void *arg)
  2051. {
  2052. (void)timer;
  2053. (void)arg;
  2054. sd_notify(0, "WATCHDOG=1");
  2055. }
  2056. #endif /* defined(HAVE_SYSTEMD_209) */
  2057. #define UPTIME_CUTOFF_FOR_NEW_BANDWIDTH_TEST (6*60*60)
  2058. /** Called when our IP address seems to have changed. <b>at_interface</b>
  2059. * should be true if we detected a change in our interface, and false if we
  2060. * detected a change in our published address. */
  2061. void
  2062. ip_address_changed(int at_interface)
  2063. {
  2064. const or_options_t *options = get_options();
  2065. int server = server_mode(options);
  2066. int exit_reject_interfaces = (server && options->ExitRelay
  2067. && options->ExitPolicyRejectLocalInterfaces);
  2068. if (at_interface) {
  2069. if (! server) {
  2070. /* Okay, change our keys. */
  2071. if (init_keys_client() < 0)
  2072. log_warn(LD_GENERAL, "Unable to rotate keys after IP change!");
  2073. }
  2074. } else {
  2075. if (server) {
  2076. if (get_uptime() > UPTIME_CUTOFF_FOR_NEW_BANDWIDTH_TEST)
  2077. reset_bandwidth_test();
  2078. reset_uptime();
  2079. router_reset_reachability();
  2080. }
  2081. }
  2082. /* Exit relays incorporate interface addresses in their exit policies when
  2083. * ExitPolicyRejectLocalInterfaces is set */
  2084. if (exit_reject_interfaces || (server && !at_interface)) {
  2085. mark_my_descriptor_dirty("IP address changed");
  2086. }
  2087. dns_servers_relaunch_checks();
  2088. }
  2089. /** Forget what we've learned about the correctness of our DNS servers, and
  2090. * start learning again. */
  2091. void
  2092. dns_servers_relaunch_checks(void)
  2093. {
  2094. if (server_mode(get_options())) {
  2095. dns_reset_correctness_checks();
  2096. if (check_dns_honesty_event) {
  2097. periodic_event_reschedule(check_dns_honesty_event);
  2098. }
  2099. }
  2100. }
  2101. /** Initialize some mainloop_event_t objects that we require. */
  2102. void
  2103. initialize_mainloop_events(void)
  2104. {
  2105. if (!schedule_active_linked_connections_event) {
  2106. schedule_active_linked_connections_event =
  2107. mainloop_event_postloop_new(schedule_active_linked_connections_cb, NULL);
  2108. }
  2109. if (!postloop_cleanup_ev) {
  2110. postloop_cleanup_ev =
  2111. mainloop_event_postloop_new(postloop_cleanup_cb, NULL);
  2112. }
  2113. }
  2114. static void
  2115. init_event_registry(void)
  2116. {
  2117. tor_assert(event_registry == NULL);
  2118. event_registry = event_registry_new();
  2119. safe_or_conn_register_events(event_registry);
  2120. or_conn_register_events(event_registry);
  2121. }
  2122. static void
  2123. tor_eventloop_thread(void)
  2124. {
  2125. log_debug(LD_GENERAL, "Starting eventloop thread.");
  2126. struct event_base *base = tor_libevent_get_base();
  2127. local_replyqueue_init(base);
  2128. int loop_result = 0;
  2129. while (loop_result == 0 && other_eventloops_should_exit == 0) {
  2130. loop_result = event_base_loop(base, EVLOOP_NO_EXIT_ON_EMPTY);
  2131. }
  2132. log_debug(LD_GENERAL, "Done eventloop thread.");
  2133. if (other_eventloops_should_exit == 0) {
  2134. log_err(LD_GENERAL, "Eventloop thread stopped unexpectedly. (loop_result:%d, \
  2135. other_eventloops_should_exit:%d)",
  2136. loop_result, other_eventloops_should_exit);
  2137. }
  2138. }
  2139. /** Tor main loop. */
  2140. int
  2141. do_main_loop(void)
  2142. {
  2143. /* initialize the periodic events first, so that code that depends on the
  2144. * events being present does not assert.
  2145. */
  2146. tor_assert(periodic_events_initialized);
  2147. initialize_mainloop_events();
  2148. init_event_registry();
  2149. periodic_events_connect_all();
  2150. struct timeval one_second = { 1, 0 };
  2151. initialize_periodic_events_event = tor_evtimer_new(
  2152. tor_libevent_get_base(),
  2153. initialize_periodic_events_cb, NULL);
  2154. event_add(initialize_periodic_events_event, &one_second);
  2155. #ifdef HAVE_SYSTEMD_209
  2156. uint64_t watchdog_delay;
  2157. /* set up systemd watchdog notification. */
  2158. if (sd_watchdog_enabled(1, &watchdog_delay) > 0) {
  2159. if (! systemd_watchdog_timer) {
  2160. struct timeval watchdog;
  2161. /* The manager will "act on" us if we don't send them a notification
  2162. * every 'watchdog_delay' microseconds. So, send notifications twice
  2163. * that often. */
  2164. watchdog_delay /= 2;
  2165. watchdog.tv_sec = watchdog_delay / 1000000;
  2166. watchdog.tv_usec = watchdog_delay % 1000000;
  2167. systemd_watchdog_timer = periodic_timer_new(tor_libevent_get_base(),
  2168. &watchdog,
  2169. systemd_watchdog_callback,
  2170. NULL);
  2171. tor_assert(systemd_watchdog_timer);
  2172. }
  2173. }
  2174. #endif /* defined(HAVE_SYSTEMD_209) */
  2175. #ifdef ENABLE_RESTART_DEBUGGING
  2176. {
  2177. static int first_time = 1;
  2178. if (first_time && getenv("TOR_DEBUG_RESTART")) {
  2179. first_time = 0;
  2180. const char *sec_str = getenv("TOR_DEBUG_RESTART_AFTER_SECONDS");
  2181. long sec;
  2182. int sec_ok=0;
  2183. if (sec_str &&
  2184. (sec = tor_parse_long(sec_str, 10, 0, INT_MAX, &sec_ok, NULL)) &&
  2185. sec_ok) {
  2186. /* Okay, we parsed the seconds. */
  2187. } else {
  2188. sec = 5;
  2189. }
  2190. struct timeval restart_after = { (time_t) sec, 0 };
  2191. tor_shutdown_event_loop_for_restart_event =
  2192. tor_evtimer_new(tor_libevent_get_base(),
  2193. tor_shutdown_event_loop_for_restart_cb, NULL);
  2194. event_add(tor_shutdown_event_loop_for_restart_event, &restart_after);
  2195. }
  2196. }
  2197. #endif /* defined(ENABLE_RESTART_DEBUGGING) */
  2198. /* Start our eventloop threads, then start the main eventloop. */
  2199. other_eventloops_should_exit = 0;
  2200. start_eventloop_threads(tor_eventloop_thread, start_tor_thread);
  2201. int rv = run_main_loop_until_done();
  2202. /* Stop our eventloops in other threads. */
  2203. other_eventloops_should_exit = 1;
  2204. rescan_eventloops();
  2205. join_eventloop_threads();
  2206. return rv;
  2207. }
  2208. #ifndef _WIN32
  2209. /** Rate-limiter for EINVAL-type libevent warnings. */
  2210. static ratelim_t libevent_error_ratelim = RATELIM_INIT(10);
  2211. #endif
  2212. /**
  2213. * Run the main loop a single time. Return 0 for "exit"; -1 for "exit with
  2214. * error", and 1 for "run this again."
  2215. */
  2216. static int
  2217. run_main_loop_once(void)
  2218. {
  2219. int loop_result;
  2220. if (nt_service_is_stopping())
  2221. return 0;
  2222. if (main_loop_should_exit)
  2223. return 0;
  2224. #ifndef _WIN32
  2225. /* Make it easier to tell whether libevent failure is our fault or not. */
  2226. errno = 0;
  2227. #endif
  2228. if (get_options()->MainloopStats) {
  2229. /* We always enforce that EVLOOP_ONCE is passed to event_base_loop() if we
  2230. * are collecting main loop statistics. */
  2231. called_loop_once = 1;
  2232. } else {
  2233. called_loop_once = 0;
  2234. }
  2235. /* Make sure we know (about) what time it is. */
  2236. update_approx_time(time(NULL));
  2237. /* Here it is: the main loop. Here we tell Libevent to poll until we have
  2238. * an event, or the second ends, or until we have some active linked
  2239. * connections to trigger events for. Libevent will wait till one
  2240. * of these happens, then run all the appropriate callbacks. */
  2241. loop_result = tor_libevent_run_event_loop(tor_libevent_get_base(),
  2242. called_loop_once);
  2243. if (get_options()->MainloopStats) {
  2244. /* Update our main loop counters. */
  2245. if (loop_result == 0) {
  2246. // The call was successful.
  2247. increment_main_loop_success_count();
  2248. } else if (loop_result == -1) {
  2249. // The call was erroneous.
  2250. increment_main_loop_error_count();
  2251. } else if (loop_result == 1) {
  2252. // The call didn't have any active or pending events
  2253. // to handle.
  2254. increment_main_loop_idle_count();
  2255. }
  2256. }
  2257. /* Oh, the loop failed. That might be an error that we need to
  2258. * catch, but more likely, it's just an interrupted poll() call or something,
  2259. * and we should try again. */
  2260. if (loop_result < 0) {
  2261. int e = tor_socket_errno(-1);
  2262. /* let the program survive things like ^z */
  2263. if (e != EINTR && !ERRNO_IS_EINPROGRESS(e)) {
  2264. log_err(LD_NET,"libevent call with %s failed: %s [%d]",
  2265. tor_libevent_get_method(), tor_socket_strerror(e), e);
  2266. return -1;
  2267. #ifndef _WIN32
  2268. } else if (e == EINVAL) {
  2269. log_fn_ratelim(&libevent_error_ratelim, LOG_WARN, LD_NET,
  2270. "EINVAL from libevent: should you upgrade libevent?");
  2271. if (libevent_error_ratelim.n_calls_since_last_time > 8) {
  2272. log_err(LD_NET, "Too many libevent errors, too fast: dying");
  2273. return -1;
  2274. }
  2275. #endif /* !defined(_WIN32) */
  2276. } else {
  2277. tor_assert_nonfatal_once(! ERRNO_IS_EINPROGRESS(e));
  2278. log_debug(LD_NET,"libevent call interrupted.");
  2279. /* You can't trust the results of this poll(). Go back to the
  2280. * top of the big for loop. */
  2281. return 1;
  2282. }
  2283. }
  2284. if (main_loop_should_exit)
  2285. return 0;
  2286. return 1;
  2287. }
  2288. /** Run the run_main_loop_once() function until it declares itself done,
  2289. * and return its final return value.
  2290. *
  2291. * Shadow won't invoke this function, so don't fill it up with things.
  2292. */
  2293. STATIC int
  2294. run_main_loop_until_done(void)
  2295. {
  2296. int loop_result = 1;
  2297. main_loop_should_exit = 0;
  2298. main_loop_exit_value = 0;
  2299. do {
  2300. loop_result = run_main_loop_once();
  2301. } while (loop_result == 1);
  2302. if (main_loop_should_exit)
  2303. return main_loop_exit_value;
  2304. else
  2305. return loop_result;
  2306. }
  2307. /** Returns Tor's uptime. */
  2308. MOCK_IMPL(long,
  2309. get_uptime,(void))
  2310. {
  2311. return stats_n_seconds_working;
  2312. }
  2313. /** Reset Tor's uptime. */
  2314. MOCK_IMPL(void,
  2315. reset_uptime,(void))
  2316. {
  2317. stats_n_seconds_working = 0;
  2318. }
  2319. void
  2320. tor_mainloop_free_all(void)
  2321. {
  2322. smartlist_free(connection_array);
  2323. smartlist_free(closeable_connection_lst);
  2324. smartlist_free(active_linked_connection_lst);
  2325. teardown_periodic_events();
  2326. tor_event_free(shutdown_did_not_work_event);
  2327. tor_event_free(initialize_periodic_events_event);
  2328. mainloop_event_free(directory_all_unreachable_cb_event);
  2329. mainloop_event_free(schedule_active_linked_connections_event);
  2330. mainloop_event_free(postloop_cleanup_ev);
  2331. mainloop_event_free(handle_deferred_signewnym_ev);
  2332. mainloop_event_free(scheduled_shutdown_ev);
  2333. mainloop_event_free(rescan_periodic_events_ev);
  2334. #ifdef HAVE_SYSTEMD_209
  2335. periodic_timer_free(systemd_watchdog_timer);
  2336. #endif
  2337. stats_n_bytes_read = stats_n_bytes_written = 0;
  2338. memset(&global_bucket, 0, sizeof(global_bucket));
  2339. memset(&global_relayed_bucket, 0, sizeof(global_relayed_bucket));
  2340. time_of_process_start = 0;
  2341. time_of_last_signewnym = 0;
  2342. signewnym_is_pending = 0;
  2343. newnym_epoch = 0;
  2344. called_loop_once = 0;
  2345. main_loop_should_exit = 0;
  2346. main_loop_exit_value = 0;
  2347. other_eventloops_should_exit = 0;
  2348. can_complete_circuits = 0;
  2349. quiet_level = 0;
  2350. should_init_bridge_stats = 1;
  2351. heartbeat_callback_first_time = 1;
  2352. current_second = 0;
  2353. memset(&current_second_last_changed, 0,
  2354. sizeof(current_second_last_changed));
  2355. }