votequorum.c 83 KB

12345678910111213141516171819202122232425262728293031323334353637383940414243444546474849505152535455565758596061626364656667686970717273747576777879808182838485868788899091929394959697989910010110210310410510610710810911011111211311411511611711811912012112212312412512612712812913013113213313413513613713813914014114214314414514614714814915015115215315415515615715815916016116216316416516616716816917017117217317417517617717817918018118218318418518618718818919019119219319419519619719819920020120220320420520620720820921021121221321421521621721821922022122222322422522622722822923023123223323423523623723823924024124224324424524624724824925025125225325425525625725825926026126226326426526626726826927027127227327427527627727827928028128228328428528628728828929029129229329429529629729829930030130230330430530630730830931031131231331431531631731831932032132232332432532632732832933033133233333433533633733833934034134234334434534634734834935035135235335435535635735835936036136236336436536636736836937037137237337437537637737837938038138238338438538638738838939039139239339439539639739839940040140240340440540640740840941041141241341441541641741841942042142242342442542642742842943043143243343443543643743843944044144244344444544644744844945045145245345445545645745845946046146246346446546646746846947047147247347447547647747847948048148248348448548648748848949049149249349449549649749849950050150250350450550650750850951051151251351451551651751851952052152252352452552652752852953053153253353453553653753853954054154254354454554654754854955055155255355455555655755855956056156256356456556656756856957057157257357457557657757857958058158258358458558658758858959059159259359459559659759859960060160260360460560660760860961061161261361461561661761861962062162262362462562662762862963063163263363463563663763863964064164264364464564664764864965065165265365465565665765865966066166266366466566666766866967067167267367467567667767867968068168268368468568668768868969069169269369469569669769869970070170270370470570670770870971071171271371471571671771871972072172272372472572672772872973073173273373473573673773873974074174274374474574674774874975075175275375475575675775875976076176276376476576676776876977077177277377477577677777877978078178278378478578678778878979079179279379479579679779879980080180280380480580680780880981081181281381481581681781881982082182282382482582682782882983083183283383483583683783883984084184284384484584684784884985085185285385485585685785885986086186286386486586686786886987087187287387487587687787887988088188288388488588688788888989089189289389489589689789889990090190290390490590690790890991091191291391491591691791891992092192292392492592692792892993093193293393493593693793893994094194294394494594694794894995095195295395495595695795895996096196296396496596696796896997097197297397497597697797897998098198298398498598698798898999099199299399499599699799899910001001100210031004100510061007100810091010101110121013101410151016101710181019102010211022102310241025102610271028102910301031103210331034103510361037103810391040104110421043104410451046104710481049105010511052105310541055105610571058105910601061106210631064106510661067106810691070107110721073107410751076107710781079108010811082108310841085108610871088108910901091109210931094109510961097109810991100110111021103110411051106110711081109111011111112111311141115111611171118111911201121112211231124112511261127112811291130113111321133113411351136113711381139114011411142114311441145114611471148114911501151115211531154115511561157115811591160116111621163116411651166116711681169117011711172117311741175117611771178117911801181118211831184118511861187118811891190119111921193119411951196119711981199120012011202120312041205120612071208120912101211121212131214121512161217121812191220122112221223122412251226122712281229123012311232123312341235123612371238123912401241124212431244124512461247124812491250125112521253125412551256125712581259126012611262126312641265126612671268126912701271127212731274127512761277127812791280128112821283128412851286128712881289129012911292129312941295129612971298129913001301130213031304130513061307130813091310131113121313131413151316131713181319132013211322132313241325132613271328132913301331133213331334133513361337133813391340134113421343134413451346134713481349135013511352135313541355135613571358135913601361136213631364136513661367136813691370137113721373137413751376137713781379138013811382138313841385138613871388138913901391139213931394139513961397139813991400140114021403140414051406140714081409141014111412141314141415141614171418141914201421142214231424142514261427142814291430143114321433143414351436143714381439144014411442144314441445144614471448144914501451145214531454145514561457145814591460146114621463146414651466146714681469147014711472147314741475147614771478147914801481148214831484148514861487148814891490149114921493149414951496149714981499150015011502150315041505150615071508150915101511151215131514151515161517151815191520152115221523152415251526152715281529153015311532153315341535153615371538153915401541154215431544154515461547154815491550155115521553155415551556155715581559156015611562156315641565156615671568156915701571157215731574157515761577157815791580158115821583158415851586158715881589159015911592159315941595159615971598159916001601160216031604160516061607160816091610161116121613161416151616161716181619162016211622162316241625162616271628162916301631163216331634163516361637163816391640164116421643164416451646164716481649165016511652165316541655165616571658165916601661166216631664166516661667166816691670167116721673167416751676167716781679168016811682168316841685168616871688168916901691169216931694169516961697169816991700170117021703170417051706170717081709171017111712171317141715171617171718171917201721172217231724172517261727172817291730173117321733173417351736173717381739174017411742174317441745174617471748174917501751175217531754175517561757175817591760176117621763176417651766176717681769177017711772177317741775177617771778177917801781178217831784178517861787178817891790179117921793179417951796179717981799180018011802180318041805180618071808180918101811181218131814181518161817181818191820182118221823182418251826182718281829183018311832183318341835183618371838183918401841184218431844184518461847184818491850185118521853185418551856185718581859186018611862186318641865186618671868186918701871187218731874187518761877187818791880188118821883188418851886188718881889189018911892189318941895189618971898189919001901190219031904190519061907190819091910191119121913191419151916191719181919192019211922192319241925192619271928192919301931193219331934193519361937193819391940194119421943194419451946194719481949195019511952195319541955195619571958195919601961196219631964196519661967196819691970197119721973197419751976197719781979198019811982198319841985198619871988198919901991199219931994199519961997199819992000200120022003200420052006200720082009201020112012201320142015201620172018201920202021202220232024202520262027202820292030203120322033203420352036203720382039204020412042204320442045204620472048204920502051205220532054205520562057205820592060206120622063206420652066206720682069207020712072207320742075207620772078207920802081208220832084208520862087208820892090209120922093209420952096209720982099210021012102210321042105210621072108210921102111211221132114211521162117211821192120212121222123212421252126212721282129213021312132213321342135213621372138213921402141214221432144214521462147214821492150215121522153215421552156215721582159216021612162216321642165216621672168216921702171217221732174217521762177217821792180218121822183218421852186218721882189219021912192219321942195219621972198219922002201220222032204220522062207220822092210221122122213221422152216221722182219222022212222222322242225222622272228222922302231223222332234223522362237223822392240224122422243224422452246224722482249225022512252225322542255225622572258225922602261226222632264226522662267226822692270227122722273227422752276227722782279228022812282228322842285228622872288228922902291229222932294229522962297229822992300230123022303230423052306230723082309231023112312231323142315231623172318231923202321232223232324232523262327232823292330233123322333233423352336233723382339234023412342234323442345234623472348234923502351235223532354235523562357235823592360236123622363236423652366236723682369237023712372237323742375237623772378237923802381238223832384238523862387238823892390239123922393239423952396239723982399240024012402240324042405240624072408240924102411241224132414241524162417241824192420242124222423242424252426242724282429243024312432243324342435243624372438243924402441244224432444244524462447244824492450245124522453245424552456245724582459246024612462246324642465246624672468246924702471247224732474247524762477247824792480248124822483248424852486248724882489249024912492249324942495249624972498249925002501250225032504250525062507250825092510251125122513251425152516251725182519252025212522252325242525252625272528252925302531253225332534253525362537253825392540254125422543254425452546254725482549255025512552255325542555255625572558255925602561256225632564256525662567256825692570257125722573257425752576257725782579258025812582258325842585258625872588258925902591259225932594259525962597259825992600260126022603260426052606260726082609261026112612261326142615261626172618261926202621262226232624262526262627262826292630263126322633263426352636263726382639264026412642264326442645264626472648264926502651265226532654265526562657265826592660266126622663266426652666266726682669267026712672267326742675267626772678267926802681268226832684268526862687268826892690269126922693269426952696269726982699270027012702270327042705270627072708270927102711271227132714271527162717271827192720272127222723272427252726272727282729273027312732273327342735273627372738273927402741274227432744274527462747274827492750275127522753275427552756275727582759276027612762276327642765276627672768276927702771277227732774277527762777277827792780278127822783278427852786278727882789279027912792279327942795279627972798279928002801280228032804280528062807280828092810281128122813281428152816281728182819282028212822282328242825282628272828282928302831283228332834283528362837283828392840284128422843284428452846284728482849285028512852285328542855285628572858285928602861286228632864286528662867286828692870287128722873287428752876287728782879288028812882288328842885288628872888288928902891289228932894289528962897289828992900290129022903290429052906290729082909291029112912291329142915291629172918291929202921292229232924292529262927292829292930293129322933293429352936293729382939294029412942294329442945294629472948294929502951295229532954295529562957295829592960296129622963296429652966296729682969297029712972297329742975297629772978297929802981298229832984298529862987298829892990299129922993299429952996299729982999300030013002300330043005300630073008300930103011301230133014301530163017301830193020302130223023302430253026302730283029
  1. /*
  2. * Copyright (c) 2009-2015 Red Hat, Inc.
  3. *
  4. * All rights reserved.
  5. *
  6. * Authors: Christine Caulfield (ccaulfie@redhat.com)
  7. * Fabio M. Di Nitto (fdinitto@redhat.com)
  8. *
  9. * This software licensed under BSD license, the text of which follows:
  10. *
  11. * Redistribution and use in source and binary forms, with or without
  12. * modification, are permitted provided that the following conditions are met:
  13. *
  14. * - Redistributions of source code must retain the above copyright notice,
  15. * this list of conditions and the following disclaimer.
  16. * - Redistributions in binary form must reproduce the above copyright notice,
  17. * this list of conditions and the following disclaimer in the documentation
  18. * and/or other materials provided with the distribution.
  19. * - Neither the name of the MontaVista Software, Inc. nor the names of its
  20. * contributors may be used to endorse or promote products derived from this
  21. * software without specific prior written permission.
  22. *
  23. * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTIBUTORS "AS IS"
  24. * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  25. * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  26. * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
  27. * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
  28. * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
  29. * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
  30. * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
  31. * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
  32. * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
  33. * THE POSSIBILITY OF SUCH DAMAGE.
  34. */
  35. #include <config.h>
  36. #include <sys/types.h>
  37. #include <sys/stat.h>
  38. #include <fcntl.h>
  39. #include <stdint.h>
  40. #include <unistd.h>
  41. #include <qb/qbipc_common.h>
  42. #include "quorum.h"
  43. #include <corosync/corodefs.h>
  44. #include <corosync/list.h>
  45. #include <corosync/logsys.h>
  46. #include <corosync/coroapi.h>
  47. #include <corosync/icmap.h>
  48. #include <corosync/votequorum.h>
  49. #include <corosync/ipc_votequorum.h>
  50. #include "service.h"
  51. #include "util.h"
  52. LOGSYS_DECLARE_SUBSYS ("VOTEQ");
  53. /*
  54. * interface with corosync
  55. */
  56. static struct corosync_api_v1 *corosync_api;
  57. /*
  58. * votequorum global config vars
  59. */
  60. static char qdevice_name[VOTEQUORUM_QDEVICE_MAX_NAME_LEN];
  61. static struct cluster_node *qdevice = NULL;
  62. static unsigned int qdevice_timeout = VOTEQUORUM_QDEVICE_DEFAULT_TIMEOUT;
  63. static unsigned int qdevice_sync_timeout = VOTEQUORUM_QDEVICE_DEFAULT_SYNC_TIMEOUT;
  64. static uint8_t qdevice_can_operate = 1;
  65. static void *qdevice_reg_conn = NULL;
  66. static uint8_t qdevice_master_wins = 0;
  67. static uint8_t two_node = 0;
  68. static uint8_t wait_for_all = 0;
  69. static uint8_t wait_for_all_status = 0;
  70. static enum {ATB_NONE, ATB_LOWEST, ATB_HIGHEST, ATB_LIST} auto_tie_breaker = ATB_NONE;
  71. static int lowest_node_id = -1;
  72. static int highest_node_id = -1;
  73. #define DEFAULT_LMS_WIN 10000
  74. static uint8_t last_man_standing = 0;
  75. static uint32_t last_man_standing_window = DEFAULT_LMS_WIN;
  76. static uint8_t allow_downscale = 0;
  77. static uint32_t ev_barrier = 0;
  78. static uint8_t ev_tracking = 0;
  79. static uint32_t ev_tracking_barrier = 0;
  80. static int ev_tracking_fd = -1;
  81. /*
  82. * votequorum_exec defines/structs/forward definitions
  83. */
  84. struct req_exec_quorum_nodeinfo {
  85. struct qb_ipc_request_header header __attribute__((aligned(8)));
  86. uint32_t nodeid;
  87. uint32_t votes;
  88. uint32_t expected_votes;
  89. uint32_t flags;
  90. } __attribute__((packed));
  91. struct req_exec_quorum_reconfigure {
  92. struct qb_ipc_request_header header __attribute__((aligned(8)));
  93. uint32_t nodeid;
  94. uint32_t value;
  95. uint8_t param;
  96. uint8_t _pad0;
  97. uint8_t _pad1;
  98. uint8_t _pad2;
  99. } __attribute__((packed));
  100. struct req_exec_quorum_qdevice_reg {
  101. struct qb_ipc_request_header header __attribute__((aligned(8)));
  102. uint32_t operation;
  103. char qdevice_name[VOTEQUORUM_QDEVICE_MAX_NAME_LEN];
  104. } __attribute__((packed));
  105. struct req_exec_quorum_qdevice_reconfigure {
  106. struct qb_ipc_request_header header __attribute__((aligned(8)));
  107. char oldname[VOTEQUORUM_QDEVICE_MAX_NAME_LEN];
  108. char newname[VOTEQUORUM_QDEVICE_MAX_NAME_LEN];
  109. } __attribute__((packed));
  110. /*
  111. * votequorum_exec onwire version (via totem)
  112. */
  113. #include "votequorum.h"
  114. /*
  115. * votequorum_exec onwire messages (via totem)
  116. */
  117. #define MESSAGE_REQ_EXEC_VOTEQUORUM_NODEINFO 0
  118. #define MESSAGE_REQ_EXEC_VOTEQUORUM_RECONFIGURE 1
  119. #define MESSAGE_REQ_EXEC_VOTEQUORUM_QDEVICE_REG 2
  120. #define MESSAGE_REQ_EXEC_VOTEQUORUM_QDEVICE_RECONFIGURE 3
  121. static void votequorum_exec_send_expectedvotes_notification(void);
  122. static int votequorum_exec_send_quorum_notification(void *conn, uint64_t context);
  123. static int votequorum_exec_send_nodelist_notification(void *conn, uint64_t context);
  124. #define VOTEQUORUM_RECONFIG_PARAM_EXPECTED_VOTES 1
  125. #define VOTEQUORUM_RECONFIG_PARAM_NODE_VOTES 2
  126. #define VOTEQUORUM_RECONFIG_PARAM_CANCEL_WFA 3
  127. static int votequorum_exec_send_reconfigure(uint8_t param, unsigned int nodeid, uint32_t value);
  128. /*
  129. * used by req_exec_quorum_qdevice_reg
  130. */
  131. #define VOTEQUORUM_QDEVICE_OPERATION_UNREGISTER 0
  132. #define VOTEQUORUM_QDEVICE_OPERATION_REGISTER 1
  133. /*
  134. * votequorum internal node status/view
  135. */
  136. #define NODE_FLAGS_QUORATE 1
  137. #define NODE_FLAGS_LEAVING 2
  138. #define NODE_FLAGS_WFASTATUS 4
  139. #define NODE_FLAGS_FIRST 8
  140. #define NODE_FLAGS_QDEVICE_REGISTERED 16
  141. #define NODE_FLAGS_QDEVICE_ALIVE 32
  142. #define NODE_FLAGS_QDEVICE_CAST_VOTE 64
  143. #define NODE_FLAGS_QDEVICE_MASTER_WINS 128
  144. typedef enum {
  145. NODESTATE_MEMBER=1,
  146. NODESTATE_DEAD,
  147. NODESTATE_LEAVING
  148. } nodestate_t;
  149. struct cluster_node {
  150. int node_id;
  151. nodestate_t state;
  152. uint32_t votes;
  153. uint32_t expected_votes;
  154. uint32_t flags;
  155. struct list_head list;
  156. };
  157. /*
  158. * votequorum internal quorum status
  159. */
  160. static uint8_t quorum;
  161. static uint8_t cluster_is_quorate;
  162. /*
  163. * votequorum membership data
  164. */
  165. static struct cluster_node *us;
  166. static struct list_head cluster_members_list;
  167. static unsigned int quorum_members[PROCESSOR_COUNT_MAX+1];
  168. static unsigned int previous_quorum_members[PROCESSOR_COUNT_MAX];
  169. static unsigned int atb_nodelist[PROCESSOR_COUNT_MAX];
  170. static int quorum_members_entries = 0;
  171. static int previous_quorum_members_entries = 0;
  172. static int atb_nodelist_entries = 0;
  173. static struct memb_ring_id quorum_ringid;
  174. /*
  175. * pre allocate all cluster_nodes + one for qdevice
  176. */
  177. static struct cluster_node cluster_nodes[PROCESSOR_COUNT_MAX+2];
  178. static int cluster_nodes_entries = 0;
  179. /*
  180. * votequorum tracking
  181. */
  182. struct quorum_pd {
  183. unsigned char track_flags;
  184. int tracking_enabled;
  185. uint64_t tracking_context;
  186. struct list_head list;
  187. void *conn;
  188. };
  189. static struct list_head trackers_list;
  190. /*
  191. * votequorum timers
  192. */
  193. static corosync_timer_handle_t qdevice_timer;
  194. static int qdevice_timer_set = 0;
  195. static corosync_timer_handle_t last_man_standing_timer;
  196. static int last_man_standing_timer_set = 0;
  197. static int sync_nodeinfo_sent = 0;
  198. static int sync_wait_for_poll_or_timeout = 0;
  199. /*
  200. * Service Interfaces required by service_message_handler struct
  201. */
  202. static int sync_in_progress = 0;
  203. static void votequorum_sync_init (
  204. const unsigned int *trans_list,
  205. size_t trans_list_entries,
  206. const unsigned int *member_list,
  207. size_t member_list_entries,
  208. const struct memb_ring_id *ring_id);
  209. static int votequorum_sync_process (void);
  210. static void votequorum_sync_activate (void);
  211. static void votequorum_sync_abort (void);
  212. static quorum_set_quorate_fn_t quorum_callback;
  213. /*
  214. * votequorum_exec handler and definitions
  215. */
  216. static char *votequorum_exec_init_fn (struct corosync_api_v1 *api);
  217. static int votequorum_exec_exit_fn (void);
  218. static int votequorum_exec_send_nodeinfo(uint32_t nodeid);
  219. static void message_handler_req_exec_votequorum_nodeinfo (
  220. const void *message,
  221. unsigned int nodeid);
  222. static void exec_votequorum_nodeinfo_endian_convert (void *message);
  223. static void message_handler_req_exec_votequorum_reconfigure (
  224. const void *message,
  225. unsigned int nodeid);
  226. static void exec_votequorum_reconfigure_endian_convert (void *message);
  227. static void message_handler_req_exec_votequorum_qdevice_reg (
  228. const void *message,
  229. unsigned int nodeid);
  230. static void exec_votequorum_qdevice_reg_endian_convert (void *message);
  231. static void message_handler_req_exec_votequorum_qdevice_reconfigure (
  232. const void *message,
  233. unsigned int nodeid);
  234. static void exec_votequorum_qdevice_reconfigure_endian_convert (void *message);
  235. static struct corosync_exec_handler votequorum_exec_engine[] =
  236. {
  237. { /* 0 */
  238. .exec_handler_fn = message_handler_req_exec_votequorum_nodeinfo,
  239. .exec_endian_convert_fn = exec_votequorum_nodeinfo_endian_convert
  240. },
  241. { /* 1 */
  242. .exec_handler_fn = message_handler_req_exec_votequorum_reconfigure,
  243. .exec_endian_convert_fn = exec_votequorum_reconfigure_endian_convert
  244. },
  245. { /* 2 */
  246. .exec_handler_fn = message_handler_req_exec_votequorum_qdevice_reg,
  247. .exec_endian_convert_fn = exec_votequorum_qdevice_reg_endian_convert
  248. },
  249. { /* 3 */
  250. .exec_handler_fn = message_handler_req_exec_votequorum_qdevice_reconfigure,
  251. .exec_endian_convert_fn = exec_votequorum_qdevice_reconfigure_endian_convert
  252. },
  253. };
  254. /*
  255. * Library Handler and Functions Definitions
  256. */
  257. static int quorum_lib_init_fn (void *conn);
  258. static int quorum_lib_exit_fn (void *conn);
  259. static void qdevice_timer_fn(void *arg);
  260. static void message_handler_req_lib_votequorum_getinfo (void *conn,
  261. const void *message);
  262. static void message_handler_req_lib_votequorum_setexpected (void *conn,
  263. const void *message);
  264. static void message_handler_req_lib_votequorum_setvotes (void *conn,
  265. const void *message);
  266. static void message_handler_req_lib_votequorum_trackstart (void *conn,
  267. const void *message);
  268. static void message_handler_req_lib_votequorum_trackstop (void *conn,
  269. const void *message);
  270. static void message_handler_req_lib_votequorum_qdevice_register (void *conn,
  271. const void *message);
  272. static void message_handler_req_lib_votequorum_qdevice_unregister (void *conn,
  273. const void *message);
  274. static void message_handler_req_lib_votequorum_qdevice_update (void *conn,
  275. const void *message);
  276. static void message_handler_req_lib_votequorum_qdevice_poll (void *conn,
  277. const void *message);
  278. static void message_handler_req_lib_votequorum_qdevice_master_wins (void *conn,
  279. const void *message);
  280. static struct corosync_lib_handler quorum_lib_service[] =
  281. {
  282. { /* 0 */
  283. .lib_handler_fn = message_handler_req_lib_votequorum_getinfo,
  284. .flow_control = COROSYNC_LIB_FLOW_CONTROL_NOT_REQUIRED
  285. },
  286. { /* 1 */
  287. .lib_handler_fn = message_handler_req_lib_votequorum_setexpected,
  288. .flow_control = COROSYNC_LIB_FLOW_CONTROL_NOT_REQUIRED
  289. },
  290. { /* 2 */
  291. .lib_handler_fn = message_handler_req_lib_votequorum_setvotes,
  292. .flow_control = COROSYNC_LIB_FLOW_CONTROL_NOT_REQUIRED
  293. },
  294. { /* 3 */
  295. .lib_handler_fn = message_handler_req_lib_votequorum_trackstart,
  296. .flow_control = COROSYNC_LIB_FLOW_CONTROL_NOT_REQUIRED
  297. },
  298. { /* 4 */
  299. .lib_handler_fn = message_handler_req_lib_votequorum_trackstop,
  300. .flow_control = COROSYNC_LIB_FLOW_CONTROL_NOT_REQUIRED
  301. },
  302. { /* 5 */
  303. .lib_handler_fn = message_handler_req_lib_votequorum_qdevice_register,
  304. .flow_control = COROSYNC_LIB_FLOW_CONTROL_NOT_REQUIRED
  305. },
  306. { /* 6 */
  307. .lib_handler_fn = message_handler_req_lib_votequorum_qdevice_unregister,
  308. .flow_control = COROSYNC_LIB_FLOW_CONTROL_NOT_REQUIRED
  309. },
  310. { /* 7 */
  311. .lib_handler_fn = message_handler_req_lib_votequorum_qdevice_update,
  312. .flow_control = COROSYNC_LIB_FLOW_CONTROL_NOT_REQUIRED
  313. },
  314. { /* 8 */
  315. .lib_handler_fn = message_handler_req_lib_votequorum_qdevice_poll,
  316. .flow_control = COROSYNC_LIB_FLOW_CONTROL_NOT_REQUIRED
  317. },
  318. { /* 9 */
  319. .lib_handler_fn = message_handler_req_lib_votequorum_qdevice_master_wins,
  320. .flow_control = COROSYNC_LIB_FLOW_CONTROL_NOT_REQUIRED
  321. }
  322. };
  323. static struct corosync_service_engine votequorum_service_engine = {
  324. .name = "corosync vote quorum service v1.0",
  325. .id = VOTEQUORUM_SERVICE,
  326. .priority = 2,
  327. .private_data_size = sizeof (struct quorum_pd),
  328. .allow_inquorate = CS_LIB_ALLOW_INQUORATE,
  329. .flow_control = COROSYNC_LIB_FLOW_CONTROL_REQUIRED,
  330. .lib_init_fn = quorum_lib_init_fn,
  331. .lib_exit_fn = quorum_lib_exit_fn,
  332. .lib_engine = quorum_lib_service,
  333. .lib_engine_count = sizeof (quorum_lib_service) / sizeof (struct corosync_lib_handler),
  334. .exec_init_fn = votequorum_exec_init_fn,
  335. .exec_exit_fn = votequorum_exec_exit_fn,
  336. .exec_engine = votequorum_exec_engine,
  337. .exec_engine_count = sizeof (votequorum_exec_engine) / sizeof (struct corosync_exec_handler),
  338. .sync_init = votequorum_sync_init,
  339. .sync_process = votequorum_sync_process,
  340. .sync_activate = votequorum_sync_activate,
  341. .sync_abort = votequorum_sync_abort
  342. };
  343. struct corosync_service_engine *votequorum_get_service_engine_ver0 (void)
  344. {
  345. return (&votequorum_service_engine);
  346. }
  347. static struct default_service votequorum_service[] = {
  348. {
  349. .name = "corosync_votequorum",
  350. .ver = 0,
  351. .loader = votequorum_get_service_engine_ver0
  352. },
  353. };
  354. /*
  355. * common/utility macros/functions
  356. */
  357. #define max(a,b) (((a) > (b)) ? (a) : (b))
  358. #define list_iterate(v, head) \
  359. for (v = (head)->next; v != head; v = v->next)
  360. static void node_add_ordered(struct cluster_node *newnode)
  361. {
  362. struct cluster_node *node = NULL;
  363. struct list_head *tmp;
  364. struct list_head *newlist = &newnode->list;
  365. ENTER();
  366. list_iterate(tmp, &cluster_members_list) {
  367. node = list_entry(tmp, struct cluster_node, list);
  368. if (newnode->node_id < node->node_id) {
  369. break;
  370. }
  371. }
  372. if (!node) {
  373. list_add(&newnode->list, &cluster_members_list);
  374. } else {
  375. newlist->prev = tmp->prev;
  376. newlist->next = tmp;
  377. tmp->prev->next = newlist;
  378. tmp->prev = newlist;
  379. }
  380. LEAVE();
  381. }
  382. static struct cluster_node *allocate_node(unsigned int nodeid)
  383. {
  384. struct cluster_node *cl = NULL;
  385. struct list_head *tmp;
  386. ENTER();
  387. if (cluster_nodes_entries <= PROCESSOR_COUNT_MAX + 1) {
  388. cl = (struct cluster_node *)&cluster_nodes[cluster_nodes_entries];
  389. cluster_nodes_entries++;
  390. } else {
  391. list_iterate(tmp, &cluster_members_list) {
  392. cl = list_entry(tmp, struct cluster_node, list);
  393. if (cl->state == NODESTATE_DEAD) {
  394. break;
  395. }
  396. }
  397. /*
  398. * this should never happen
  399. */
  400. if (!cl) {
  401. log_printf(LOGSYS_LEVEL_CRIT, "Unable to find memory for node %u data!!", nodeid);
  402. goto out;
  403. }
  404. list_del(tmp);
  405. }
  406. memset(cl, 0, sizeof(struct cluster_node));
  407. cl->node_id = nodeid;
  408. if (nodeid != VOTEQUORUM_QDEVICE_NODEID) {
  409. node_add_ordered(cl);
  410. }
  411. out:
  412. LEAVE();
  413. return cl;
  414. }
  415. static struct cluster_node *find_node_by_nodeid(unsigned int nodeid)
  416. {
  417. struct cluster_node *node;
  418. struct list_head *tmp;
  419. ENTER();
  420. if (nodeid == us->node_id) {
  421. LEAVE();
  422. return us;
  423. }
  424. if (nodeid == VOTEQUORUM_QDEVICE_NODEID) {
  425. LEAVE();
  426. return qdevice;
  427. }
  428. list_iterate(tmp, &cluster_members_list) {
  429. node = list_entry(tmp, struct cluster_node, list);
  430. if (node->node_id == nodeid) {
  431. LEAVE();
  432. return node;
  433. }
  434. }
  435. LEAVE();
  436. return NULL;
  437. }
  438. static void get_lowest_node_id(void)
  439. {
  440. struct cluster_node *node = NULL;
  441. struct list_head *tmp;
  442. ENTER();
  443. lowest_node_id = us->node_id;
  444. list_iterate(tmp, &cluster_members_list) {
  445. node = list_entry(tmp, struct cluster_node, list);
  446. if ((node->state == NODESTATE_MEMBER) &&
  447. (node->node_id < lowest_node_id)) {
  448. lowest_node_id = node->node_id;
  449. }
  450. }
  451. log_printf(LOGSYS_LEVEL_DEBUG, "lowest node id: %d us: %d", lowest_node_id, us->node_id);
  452. icmap_set_uint32("runtime.votequorum.lowest_node_id", lowest_node_id);
  453. LEAVE();
  454. }
  455. static void get_highest_node_id(void)
  456. {
  457. struct cluster_node *node = NULL;
  458. struct list_head *tmp;
  459. ENTER();
  460. highest_node_id = us->node_id;
  461. list_iterate(tmp, &cluster_members_list) {
  462. node = list_entry(tmp, struct cluster_node, list);
  463. if ((node->state == NODESTATE_MEMBER) &&
  464. (node->node_id > highest_node_id)) {
  465. highest_node_id = node->node_id;
  466. }
  467. }
  468. log_printf(LOGSYS_LEVEL_DEBUG, "highest node id: %d us: %d", highest_node_id, us->node_id);
  469. icmap_set_uint32("runtime.votequorum.highest_node_id", highest_node_id);
  470. LEAVE();
  471. }
  472. static int check_low_node_id_partition(void)
  473. {
  474. struct cluster_node *node = NULL;
  475. struct list_head *tmp;
  476. int found = 0;
  477. ENTER();
  478. list_iterate(tmp, &cluster_members_list) {
  479. node = list_entry(tmp, struct cluster_node, list);
  480. if ((node->state == NODESTATE_MEMBER) &&
  481. (node->node_id == lowest_node_id)) {
  482. found = 1;
  483. }
  484. }
  485. LEAVE();
  486. return found;
  487. }
  488. static int check_high_node_id_partition(void)
  489. {
  490. struct cluster_node *node = NULL;
  491. struct list_head *tmp;
  492. int found = 0;
  493. ENTER();
  494. list_iterate(tmp, &cluster_members_list) {
  495. node = list_entry(tmp, struct cluster_node, list);
  496. if ((node->state == NODESTATE_MEMBER) &&
  497. (node->node_id == highest_node_id)) {
  498. found = 1;
  499. }
  500. }
  501. LEAVE();
  502. return found;
  503. }
  504. static int is_in_nodelist(int nodeid, unsigned int *members, int entries)
  505. {
  506. int i;
  507. ENTER();
  508. for (i=0; i<entries; i++) {
  509. if (nodeid == members[i]) {
  510. LEAVE();
  511. return 1;
  512. }
  513. }
  514. LEAVE();
  515. return 0;
  516. }
  517. /*
  518. * The algorithm for a list of tie-breaker nodes is:
  519. * travel the list of nodes in the auto_tie_breaker list,
  520. * if the node IS in our current partition, check if the
  521. * nodes earlier in the atb list are in the 'previous' partition;
  522. * If none are found then we are safe to be quorate, if any are
  523. * then we cannot be as we don't know if that node is up or down.
  524. * If we don't have a node in the current list we are NOT quorate.
  525. * Obviously if we find the first node in the atb list in our
  526. * partition then we are quorate.
  527. *
  528. * Special cases lowest nodeid, and highest nodeid are handled separately.
  529. */
  530. static int check_auto_tie_breaker(void)
  531. {
  532. int i, j;
  533. int res;
  534. ENTER();
  535. if (auto_tie_breaker == ATB_LOWEST) {
  536. res = check_low_node_id_partition();
  537. log_printf(LOGSYS_LEVEL_DEBUG, "ATB_LOWEST decision: %d", res);
  538. LEAVE();
  539. return res;
  540. }
  541. if (auto_tie_breaker == ATB_HIGHEST) {
  542. res = check_high_node_id_partition();
  543. log_printf(LOGSYS_LEVEL_DEBUG, "ATB_HIGHEST decision: %d", res);
  544. LEAVE();
  545. return res;
  546. }
  547. /* Assume ATB_LIST, we should never be called for ATB_NONE */
  548. for (i=0; i < atb_nodelist_entries; i++) {
  549. if (is_in_nodelist(atb_nodelist[i], quorum_members, quorum_members_entries)) {
  550. /*
  551. * Node is in our partition, if any of its predecessors are
  552. * in the previous quorum partition then it might be in the
  553. * 'other half' (as we've got this far without seeing it here)
  554. * and so we can't be quorate.
  555. */
  556. for (j=0; j<i; j++) {
  557. if (is_in_nodelist(atb_nodelist[j], previous_quorum_members, previous_quorum_members_entries)) {
  558. log_printf(LOGSYS_LEVEL_DEBUG, "ATB_LIST found node %d in previous partition but not here, quorum denied", atb_nodelist[j]);
  559. LEAVE();
  560. return 0;
  561. }
  562. }
  563. /*
  564. * None of the other list nodes were in the previous partition, if there
  565. * are enough votes, we can be quorate
  566. */
  567. log_printf(LOGSYS_LEVEL_DEBUG, "ATB_LIST found node %d in current partition, we can be quorate", atb_nodelist[i]);
  568. LEAVE();
  569. return 1;
  570. }
  571. }
  572. log_printf(LOGSYS_LEVEL_DEBUG, "ATB_LIST found no list nodes in current partition, we cannot be quorate");
  573. LEAVE();
  574. return 0;
  575. }
  576. /*
  577. * atb_string can be either:
  578. * 'lowest'
  579. * 'highest'
  580. * a list of nodeids
  581. */
  582. static void parse_atb_string(char *atb_string)
  583. {
  584. char *ptr;
  585. long num;
  586. ENTER();
  587. auto_tie_breaker = ATB_NONE;
  588. if (!strcmp(atb_string, "lowest"))
  589. auto_tie_breaker = ATB_LOWEST;
  590. if (!strcmp(atb_string, "highest"))
  591. auto_tie_breaker = ATB_HIGHEST;
  592. if (atoi(atb_string)) {
  593. atb_nodelist_entries = 0;
  594. ptr = atb_string;
  595. do {
  596. num = strtol(ptr, &ptr, 10);
  597. if (num) {
  598. log_printf(LOGSYS_LEVEL_DEBUG, "ATB nodelist[%d] = %d", atb_nodelist_entries, num);
  599. atb_nodelist[atb_nodelist_entries++] = num;
  600. }
  601. } while (num);
  602. if (atb_nodelist_entries) {
  603. auto_tie_breaker = ATB_LIST;
  604. }
  605. }
  606. icmap_set_uint32("runtime.votequorum.atb_type", auto_tie_breaker);
  607. log_printf(LOGSYS_LEVEL_DEBUG, "ATB type = %d", auto_tie_breaker);
  608. /* Make sure we got something */
  609. if (auto_tie_breaker == ATB_NONE) {
  610. log_printf(LOGSYS_LEVEL_WARNING, "auto_tie_breaker_nodes is not valid. It must be 'lowest', 'highest' or a space-separated list of node IDs. auto_tie_breaker is disabled");
  611. auto_tie_breaker = ATB_NONE;
  612. }
  613. LEAVE();
  614. }
  615. static int check_qdevice_master(void)
  616. {
  617. struct cluster_node *node = NULL;
  618. struct list_head *tmp;
  619. int found = 0;
  620. ENTER();
  621. list_iterate(tmp, &cluster_members_list) {
  622. node = list_entry(tmp, struct cluster_node, list);
  623. if ((node->state == NODESTATE_MEMBER) &&
  624. (node->flags & NODE_FLAGS_QDEVICE_MASTER_WINS) &&
  625. (node->flags & NODE_FLAGS_QDEVICE_CAST_VOTE)) {
  626. found = 1;
  627. }
  628. }
  629. LEAVE();
  630. return found;
  631. }
  632. static void decode_flags(uint32_t flags)
  633. {
  634. ENTER();
  635. log_printf(LOGSYS_LEVEL_DEBUG,
  636. "flags: quorate: %s Leaving: %s WFA Status: %s First: %s Qdevice: %s QdeviceAlive: %s QdeviceCastVote: %s QdeviceMasterWins: %s",
  637. (flags & NODE_FLAGS_QUORATE)?"Yes":"No",
  638. (flags & NODE_FLAGS_LEAVING)?"Yes":"No",
  639. (flags & NODE_FLAGS_WFASTATUS)?"Yes":"No",
  640. (flags & NODE_FLAGS_FIRST)?"Yes":"No",
  641. (flags & NODE_FLAGS_QDEVICE_REGISTERED)?"Yes":"No",
  642. (flags & NODE_FLAGS_QDEVICE_ALIVE)?"Yes":"No",
  643. (flags & NODE_FLAGS_QDEVICE_CAST_VOTE)?"Yes":"No",
  644. (flags & NODE_FLAGS_QDEVICE_MASTER_WINS)?"Yes":"No");
  645. LEAVE();
  646. }
  647. /*
  648. * load/save are copied almost pristine from totemsrp,c
  649. */
  650. static int load_ev_tracking_barrier(void)
  651. {
  652. int res = 0;
  653. char filename[PATH_MAX];
  654. ENTER();
  655. snprintf(filename, sizeof(filename) - 1, "%s/ev_tracking", get_run_dir());
  656. ev_tracking_fd = open(filename, O_RDWR, 0700);
  657. if (ev_tracking_fd != -1) {
  658. res = read (ev_tracking_fd, &ev_tracking_barrier, sizeof(uint32_t));
  659. if (res == sizeof (uint32_t)) {
  660. LEAVE();
  661. return 0;
  662. }
  663. }
  664. ev_tracking_barrier = 0;
  665. umask(0);
  666. ev_tracking_fd = open (filename, O_CREAT|O_RDWR, 0700);
  667. if (ev_tracking_fd != -1) {
  668. res = write (ev_tracking_fd, &ev_tracking_barrier, sizeof (uint32_t));
  669. if ((res == -1) || (res != sizeof (uint32_t))) {
  670. log_printf(LOGSYS_LEVEL_WARNING,
  671. "Unable to write to %s", filename);
  672. }
  673. LEAVE();
  674. return 0;
  675. }
  676. log_printf(LOGSYS_LEVEL_WARNING,
  677. "Unable to create %s file", filename);
  678. LEAVE();
  679. return -1;
  680. }
  681. static void update_wait_for_all_status(uint8_t wfa_status)
  682. {
  683. ENTER();
  684. wait_for_all_status = wfa_status;
  685. if (wait_for_all_status) {
  686. us->flags |= NODE_FLAGS_WFASTATUS;
  687. } else {
  688. us->flags &= ~NODE_FLAGS_WFASTATUS;
  689. }
  690. icmap_set_uint8("runtime.votequorum.wait_for_all_status",
  691. wait_for_all_status);
  692. LEAVE();
  693. }
  694. static void update_two_node(void)
  695. {
  696. ENTER();
  697. icmap_set_uint8("runtime.votequorum.two_node", two_node);
  698. LEAVE();
  699. }
  700. static void update_ev_barrier(uint32_t expected_votes)
  701. {
  702. ENTER();
  703. ev_barrier = expected_votes;
  704. icmap_set_uint32("runtime.votequorum.ev_barrier", ev_barrier);
  705. LEAVE();
  706. }
  707. static void update_qdevice_can_operate(uint8_t status)
  708. {
  709. ENTER();
  710. qdevice_can_operate = status;
  711. icmap_set_uint8("runtime.votequorum.qdevice_can_operate", qdevice_can_operate);
  712. LEAVE();
  713. }
  714. static void update_qdevice_master_wins(uint8_t allow)
  715. {
  716. ENTER();
  717. qdevice_master_wins = allow;
  718. icmap_set_uint8("runtime.votequorum.qdevice_master_wins", qdevice_master_wins);
  719. LEAVE();
  720. }
  721. static void update_ev_tracking_barrier(uint32_t ev_t_barrier)
  722. {
  723. int res;
  724. ENTER();
  725. ev_tracking_barrier = ev_t_barrier;
  726. icmap_set_uint32("runtime.votequorum.ev_tracking_barrier", ev_tracking_barrier);
  727. if (lseek (ev_tracking_fd, 0, SEEK_SET) != 0) {
  728. log_printf(LOGSYS_LEVEL_WARNING,
  729. "Unable to update ev_tracking_barrier on disk data!!!");
  730. LEAVE();
  731. return;
  732. }
  733. res = write (ev_tracking_fd, &ev_tracking_barrier, sizeof (uint32_t));
  734. if (res != sizeof (uint32_t)) {
  735. log_printf(LOGSYS_LEVEL_WARNING,
  736. "Unable to update ev_tracking_barrier on disk data!!!");
  737. }
  738. #ifdef HAVE_FDATASYNC
  739. fdatasync(ev_tracking_fd);
  740. #else
  741. fsync(ev_tracking_fd);
  742. #endif
  743. LEAVE();
  744. }
  745. /*
  746. * quorum calculation core bits
  747. */
  748. static int calculate_quorum(int allow_decrease, unsigned int max_expected, unsigned int *ret_total_votes)
  749. {
  750. struct list_head *nodelist;
  751. struct cluster_node *node;
  752. unsigned int total_votes = 0;
  753. unsigned int highest_expected = 0;
  754. unsigned int newquorum, q1, q2;
  755. unsigned int total_nodes = 0;
  756. ENTER();
  757. if ((allow_downscale) && (allow_decrease) && (max_expected)) {
  758. max_expected = max(ev_barrier, max_expected);
  759. }
  760. list_iterate(nodelist, &cluster_members_list) {
  761. node = list_entry(nodelist, struct cluster_node, list);
  762. log_printf(LOGSYS_LEVEL_DEBUG, "node %u state=%d, votes=%u, expected=%u",
  763. node->node_id, node->state, node->votes, node->expected_votes);
  764. if (node->state == NODESTATE_MEMBER) {
  765. if (max_expected) {
  766. node->expected_votes = max_expected;
  767. } else {
  768. highest_expected = max(highest_expected, node->expected_votes);
  769. }
  770. total_votes += node->votes;
  771. total_nodes++;
  772. }
  773. }
  774. if (us->flags & NODE_FLAGS_QDEVICE_CAST_VOTE) {
  775. log_printf(LOGSYS_LEVEL_DEBUG, "node 0 state=1, votes=%u", qdevice->votes);
  776. total_votes += qdevice->votes;
  777. total_nodes++;
  778. }
  779. if (max_expected > 0) {
  780. highest_expected = max_expected;
  781. }
  782. /*
  783. * This quorum calculation is taken from the OpenVMS Cluster Systems
  784. * manual, but, then, you guessed that didn't you
  785. */
  786. q1 = (highest_expected + 2) / 2;
  787. q2 = (total_votes + 2) / 2;
  788. newquorum = max(q1, q2);
  789. /*
  790. * Normally quorum never decreases but the system administrator can
  791. * force it down by setting expected votes to a maximum value
  792. */
  793. if (!allow_decrease) {
  794. newquorum = max(quorum, newquorum);
  795. }
  796. /*
  797. * The special two_node mode allows each of the two nodes to retain
  798. * quorum if the other fails. Only one of the two should live past
  799. * fencing (as both nodes try to fence each other in split-brain.)
  800. * Also: if there are more than two nodes, force us inquorate to avoid
  801. * any damage or confusion.
  802. */
  803. if (two_node && total_nodes <= 2) {
  804. newquorum = 1;
  805. }
  806. if (ret_total_votes) {
  807. *ret_total_votes = total_votes;
  808. }
  809. LEAVE();
  810. return newquorum;
  811. }
  812. static void do_quorum_callback(void)
  813. {
  814. int real_num_entries = quorum_members_entries;
  815. if (us->flags & NODE_FLAGS_QDEVICE_ALIVE) {
  816. quorum_members[quorum_members_entries] = VOTEQUORUM_QDEVICE_NODEID;
  817. real_num_entries += 1;
  818. }
  819. quorum_callback(quorum_members, real_num_entries,
  820. cluster_is_quorate, &quorum_ringid);
  821. }
  822. static void are_we_quorate(unsigned int total_votes)
  823. {
  824. int quorate;
  825. int quorum_change = 0;
  826. ENTER();
  827. /*
  828. * wait for all nodes to show up before granting quorum
  829. */
  830. if ((wait_for_all) && (wait_for_all_status)) {
  831. if (total_votes != us->expected_votes) {
  832. log_printf(LOGSYS_LEVEL_NOTICE,
  833. "Waiting for all cluster members. "
  834. "Current votes: %d expected_votes: %d",
  835. total_votes, us->expected_votes);
  836. cluster_is_quorate = 0;
  837. return;
  838. }
  839. update_wait_for_all_status(0);
  840. }
  841. if (quorum > total_votes) {
  842. quorate = 0;
  843. } else {
  844. quorate = 1;
  845. get_lowest_node_id();
  846. get_highest_node_id();
  847. }
  848. if ((auto_tie_breaker != ATB_NONE) &&
  849. /* Must be a half (or half-1) split */
  850. (total_votes == (us->expected_votes / 2)) &&
  851. /* If the 'other' partition in a split might have quorum then we can't run ATB */
  852. (previous_quorum_members_entries - quorum_members_entries < quorum) &&
  853. (check_auto_tie_breaker() == 1)) {
  854. quorate = 1;
  855. }
  856. if ((qdevice_master_wins) &&
  857. (!quorate) &&
  858. (check_qdevice_master() == 1)) {
  859. log_printf(LOGSYS_LEVEL_DEBUG, "node is quorate as part of master_wins partition");
  860. quorate = 1;
  861. }
  862. if (cluster_is_quorate && !quorate) {
  863. quorum_change = 1;
  864. log_printf(LOGSYS_LEVEL_DEBUG, "quorum lost, blocking activity");
  865. }
  866. if (!cluster_is_quorate && quorate) {
  867. quorum_change = 1;
  868. log_printf(LOGSYS_LEVEL_DEBUG, "quorum regained, resuming activity");
  869. }
  870. cluster_is_quorate = quorate;
  871. if (cluster_is_quorate) {
  872. us->flags |= NODE_FLAGS_QUORATE;
  873. } else {
  874. us->flags &= ~NODE_FLAGS_QUORATE;
  875. }
  876. if (wait_for_all) {
  877. if (quorate) {
  878. update_wait_for_all_status(0);
  879. } else {
  880. update_wait_for_all_status(1);
  881. }
  882. }
  883. if ((quorum_change) &&
  884. (sync_in_progress == 0)) {
  885. do_quorum_callback();
  886. votequorum_exec_send_quorum_notification(NULL, 0L);
  887. }
  888. LEAVE();
  889. }
  890. static void get_total_votes(unsigned int *totalvotes, unsigned int *current_members)
  891. {
  892. unsigned int total_votes = 0;
  893. unsigned int cluster_members = 0;
  894. struct list_head *nodelist;
  895. struct cluster_node *node;
  896. ENTER();
  897. list_iterate(nodelist, &cluster_members_list) {
  898. node = list_entry(nodelist, struct cluster_node, list);
  899. if (node->state == NODESTATE_MEMBER) {
  900. cluster_members++;
  901. total_votes += node->votes;
  902. }
  903. }
  904. if (qdevice->votes) {
  905. total_votes += qdevice->votes;
  906. cluster_members++;
  907. }
  908. *totalvotes = total_votes;
  909. *current_members = cluster_members;
  910. LEAVE();
  911. }
  912. /*
  913. * Recalculate cluster quorum, set quorate and notify changes
  914. */
  915. static void recalculate_quorum(int allow_decrease, int by_current_nodes)
  916. {
  917. unsigned int total_votes = 0;
  918. unsigned int cluster_members = 0;
  919. ENTER();
  920. get_total_votes(&total_votes, &cluster_members);
  921. if (!by_current_nodes) {
  922. cluster_members = 0;
  923. }
  924. /*
  925. * Keep expected_votes at the highest number of votes in the cluster
  926. */
  927. log_printf(LOGSYS_LEVEL_DEBUG, "total_votes=%d, expected_votes=%d", total_votes, us->expected_votes);
  928. if (total_votes > us->expected_votes) {
  929. us->expected_votes = total_votes;
  930. votequorum_exec_send_expectedvotes_notification();
  931. }
  932. if ((ev_tracking) &&
  933. (us->expected_votes > ev_tracking_barrier)) {
  934. update_ev_tracking_barrier(us->expected_votes);
  935. }
  936. quorum = calculate_quorum(allow_decrease, cluster_members, &total_votes);
  937. are_we_quorate(total_votes);
  938. LEAVE();
  939. }
  940. /*
  941. * configuration bits and pieces
  942. */
  943. static int votequorum_read_nodelist_configuration(uint32_t *votes,
  944. uint32_t *nodes,
  945. uint32_t *expected_votes)
  946. {
  947. icmap_iter_t iter;
  948. const char *iter_key;
  949. char tmp_key[ICMAP_KEYNAME_MAXLEN];
  950. uint32_t our_pos, node_pos;
  951. uint32_t nodecount = 0;
  952. uint32_t nodelist_expected_votes = 0;
  953. uint32_t node_votes = 0;
  954. int res = 0;
  955. ENTER();
  956. if (icmap_get_uint32("nodelist.local_node_pos", &our_pos) != CS_OK) {
  957. log_printf(LOGSYS_LEVEL_DEBUG,
  958. "No nodelist defined or our node is not in the nodelist");
  959. return 0;
  960. }
  961. iter = icmap_iter_init("nodelist.node.");
  962. while ((iter_key = icmap_iter_next(iter, NULL, NULL)) != NULL) {
  963. res = sscanf(iter_key, "nodelist.node.%u.%s", &node_pos, tmp_key);
  964. if (res != 2) {
  965. continue;
  966. }
  967. if (strcmp(tmp_key, "ring0_addr") != 0) {
  968. continue;
  969. }
  970. nodecount++;
  971. snprintf(tmp_key, ICMAP_KEYNAME_MAXLEN, "nodelist.node.%u.quorum_votes", node_pos);
  972. if (icmap_get_uint32(tmp_key, &node_votes) != CS_OK) {
  973. node_votes = 1;
  974. }
  975. nodelist_expected_votes = nodelist_expected_votes + node_votes;
  976. if (node_pos == our_pos) {
  977. *votes = node_votes;
  978. }
  979. }
  980. *expected_votes = nodelist_expected_votes;
  981. *nodes = nodecount;
  982. icmap_iter_finalize(iter);
  983. LEAVE();
  984. return 1;
  985. }
  986. static int votequorum_qdevice_is_configured(uint32_t *qdevice_votes)
  987. {
  988. char *qdevice_model = NULL;
  989. int ret = 0;
  990. ENTER();
  991. if (icmap_get_string("quorum.device.model", &qdevice_model) == CS_OK) {
  992. if (strlen(qdevice_model)) {
  993. if (icmap_get_uint32("quorum.device.votes", qdevice_votes) != CS_OK) {
  994. *qdevice_votes = -1;
  995. }
  996. if (icmap_get_uint32("quorum.device.timeout", &qdevice_timeout) != CS_OK) {
  997. qdevice_timeout = VOTEQUORUM_QDEVICE_DEFAULT_TIMEOUT;
  998. }
  999. if (icmap_get_uint32("quorum.device.sync_timeout", &qdevice_sync_timeout) != CS_OK) {
  1000. qdevice_sync_timeout = VOTEQUORUM_QDEVICE_DEFAULT_SYNC_TIMEOUT;
  1001. }
  1002. update_qdevice_can_operate(1);
  1003. ret = 1;
  1004. }
  1005. free(qdevice_model);
  1006. }
  1007. LEAVE();
  1008. return ret;
  1009. }
  1010. #define VOTEQUORUM_READCONFIG_STARTUP 0
  1011. #define VOTEQUORUM_READCONFIG_RUNTIME 1
  1012. static char *votequorum_readconfig(int runtime)
  1013. {
  1014. uint32_t node_votes = 0, qdevice_votes = 0;
  1015. uint32_t node_expected_votes = 0, expected_votes = 0;
  1016. uint32_t node_count = 0;
  1017. uint8_t atb = 0;
  1018. int have_nodelist, have_qdevice;
  1019. char *atb_string = NULL;
  1020. char *error = NULL;
  1021. ENTER();
  1022. log_printf(LOGSYS_LEVEL_DEBUG, "Reading configuration (runtime: %d)", runtime);
  1023. /*
  1024. * Set the few things we re-read at runtime back to their defaults
  1025. */
  1026. if (runtime) {
  1027. two_node = 0;
  1028. expected_votes = 0;
  1029. }
  1030. /*
  1031. * gather basic data here
  1032. */
  1033. icmap_get_uint32("quorum.expected_votes", &expected_votes);
  1034. have_nodelist = votequorum_read_nodelist_configuration(&node_votes, &node_count, &node_expected_votes);
  1035. have_qdevice = votequorum_qdevice_is_configured(&qdevice_votes);
  1036. icmap_get_uint8("quorum.two_node", &two_node);
  1037. /*
  1038. * do config verification and enablement
  1039. */
  1040. if ((!have_nodelist) && (!expected_votes)) {
  1041. if (!runtime) {
  1042. error = (char *)"configuration error: nodelist or quorum.expected_votes must be configured!";
  1043. } else {
  1044. log_printf(LOGSYS_LEVEL_CRIT, "configuration error: nodelist or quorum.expected_votes must be configured!");
  1045. log_printf(LOGSYS_LEVEL_CRIT, "will continue with current runtime data");
  1046. }
  1047. goto out;
  1048. }
  1049. /*
  1050. * two_node and qdevice are not compatible in the same config.
  1051. * try to make an educated guess of what to do
  1052. */
  1053. if ((two_node) && (have_qdevice)) {
  1054. if (!runtime) {
  1055. error = (char *)"configuration error: two_node and quorum device cannot be configured at the same time!";
  1056. goto out;
  1057. } else {
  1058. log_printf(LOGSYS_LEVEL_CRIT, "configuration error: two_node and quorum device cannot be configured at the same time!");
  1059. if (us->flags & NODE_FLAGS_QDEVICE_REGISTERED) {
  1060. log_printf(LOGSYS_LEVEL_CRIT, "quorum device is registered, disabling two_node");
  1061. two_node = 0;
  1062. } else {
  1063. log_printf(LOGSYS_LEVEL_CRIT, "quorum device is not registered, allowing two_node");
  1064. update_qdevice_can_operate(0);
  1065. }
  1066. }
  1067. }
  1068. /*
  1069. * Enable special features
  1070. */
  1071. if (!runtime) {
  1072. if (two_node) {
  1073. wait_for_all = 1;
  1074. }
  1075. icmap_get_uint8("quorum.allow_downscale", &allow_downscale);
  1076. icmap_get_uint8("quorum.wait_for_all", &wait_for_all);
  1077. icmap_get_uint8("quorum.last_man_standing", &last_man_standing);
  1078. icmap_get_uint32("quorum.last_man_standing_window", &last_man_standing_window);
  1079. icmap_get_uint8("quorum.expected_votes_tracking", &ev_tracking);
  1080. icmap_get_uint8("quorum.auto_tie_breaker", &atb);
  1081. icmap_get_string("quorum.auto_tie_breaker_node", &atb_string);
  1082. /* auto_tie_breaker defaults to LOWEST */
  1083. if (atb) {
  1084. auto_tie_breaker = ATB_LOWEST;
  1085. icmap_set_uint32("runtime.votequorum.atb_type", auto_tie_breaker);
  1086. }
  1087. else {
  1088. auto_tie_breaker = ATB_NONE;
  1089. if (atb_string) {
  1090. log_printf(LOGSYS_LEVEL_WARNING,
  1091. "auto_tie_breaker_node: is meaningless if auto_tie_breaker is set to 0");
  1092. }
  1093. }
  1094. if (atb && atb_string) {
  1095. parse_atb_string(atb_string);
  1096. }
  1097. free(atb_string);
  1098. /* allow_downscale requires ev_tracking */
  1099. if (allow_downscale) {
  1100. ev_tracking = 1;
  1101. }
  1102. if (ev_tracking) {
  1103. if (load_ev_tracking_barrier() < 0) {
  1104. LEAVE();
  1105. return ((char *)"Unable to load ev_tracking file!");
  1106. }
  1107. update_ev_tracking_barrier(ev_tracking_barrier);
  1108. }
  1109. }
  1110. /* two_node and auto_tie_breaker are not compatible as two_node uses
  1111. * a fence race to decide quorum whereas ATB decides based on node id
  1112. */
  1113. if (two_node && auto_tie_breaker != ATB_NONE) {
  1114. log_printf(LOGSYS_LEVEL_CRIT, "two_node and auto_tie_breaker are both specified but are not compatible.");
  1115. log_printf(LOGSYS_LEVEL_CRIT, "two_node has been disabled, please fix your corosync.conf");
  1116. two_node = 0;
  1117. }
  1118. /* If ATB is set and the cluster has an odd number of nodes then wait_for_all needs
  1119. * to be set so that an isolated half+1 without the tie breaker node
  1120. * does not have quorum on reboot.
  1121. */
  1122. if ((auto_tie_breaker != ATB_NONE) && (node_expected_votes % 2) &&
  1123. (!wait_for_all)) {
  1124. if (last_man_standing) {
  1125. /* if LMS is set too, it's a fatal configuration error. We can't dictate to the user what
  1126. * they might want so we'll just quit.
  1127. */
  1128. log_printf(LOGSYS_LEVEL_CRIT, "auto_tie_breaker is set, the cluster has an odd number of nodes\n");
  1129. log_printf(LOGSYS_LEVEL_CRIT, "and last_man_standing is also set. With this situation a better\n");
  1130. log_printf(LOGSYS_LEVEL_CRIT, "solution would be to disable LMS, leave ATB enabled, and also\n");
  1131. log_printf(LOGSYS_LEVEL_CRIT, "enable wait_for_all (mandatory for ATB in odd-numbered clusters).\n");
  1132. log_printf(LOGSYS_LEVEL_CRIT, "Due to this ambiguity, corosync will fail to start. Please fix your corosync.conf\n");
  1133. error = (char *)"configuration error: auto_tie_breaker & last_man_standing not available in odd sized cluster";
  1134. goto out;
  1135. }
  1136. else {
  1137. log_printf(LOGSYS_LEVEL_CRIT, "auto_tie_breaker is set and the cluster has an odd number of nodes.\n");
  1138. log_printf(LOGSYS_LEVEL_CRIT, "wait_for_all needs to be set for this configuration but it is missing\n");
  1139. log_printf(LOGSYS_LEVEL_CRIT, "Therefore auto_tie_breaker has been disabled. Please fix your corosync.conf\n");
  1140. auto_tie_breaker = ATB_NONE;
  1141. icmap_set_uint32("runtime.votequorum.atb_type", auto_tie_breaker);
  1142. }
  1143. }
  1144. /*
  1145. * quorum device is not compatible with last_man_standing and auto_tie_breaker
  1146. * neither lms or atb can be set at runtime, so there is no need to check for
  1147. * runtime incompatibilities, but qdevice can be configured _after_ LMS and ATB have
  1148. * been enabled at startup.
  1149. */
  1150. if ((have_qdevice) && (last_man_standing)) {
  1151. if (!runtime) {
  1152. error = (char *)"configuration error: quorum.device is not compatible with last_man_standing";
  1153. goto out;
  1154. } else {
  1155. log_printf(LOGSYS_LEVEL_CRIT, "configuration error: quorum.device is not compatible with last_man_standing");
  1156. log_printf(LOGSYS_LEVEL_CRIT, "disabling quorum device operations");
  1157. update_qdevice_can_operate(0);
  1158. }
  1159. }
  1160. if ((have_qdevice) && (auto_tie_breaker != ATB_NONE)) {
  1161. if (!runtime) {
  1162. error = (char *)"configuration error: quorum.device is not compatible with auto_tie_breaker";
  1163. goto out;
  1164. } else {
  1165. log_printf(LOGSYS_LEVEL_CRIT, "configuration error: quorum.device is not compatible with auto_tie_breaker");
  1166. log_printf(LOGSYS_LEVEL_CRIT, "disabling quorum device operations");
  1167. update_qdevice_can_operate(0);
  1168. }
  1169. }
  1170. if ((have_qdevice) && (allow_downscale)) {
  1171. if (!runtime) {
  1172. error = (char *)"configuration error: quorum.device is not compatible with allow_downscale";
  1173. goto out;
  1174. } else {
  1175. log_printf(LOGSYS_LEVEL_CRIT, "configuration error: quorum.device is not compatible with allow_downscale");
  1176. log_printf(LOGSYS_LEVEL_CRIT, "disabling quorum device operations");
  1177. update_qdevice_can_operate(0);
  1178. }
  1179. }
  1180. /*
  1181. * if user specifies quorum.expected_votes + quorum.device but NOT the device.votes
  1182. * we don't know what the quorum device should vote.
  1183. */
  1184. if ((expected_votes) && (have_qdevice) && (qdevice_votes == -1)) {
  1185. if (!runtime) {
  1186. error = (char *)"configuration error: quorum.device.votes must be specified when quorum.expected_votes is set";
  1187. goto out;
  1188. } else {
  1189. log_printf(LOGSYS_LEVEL_CRIT, "configuration error: quorum.device.votes must be specified when quorum.expected_votes is set");
  1190. log_printf(LOGSYS_LEVEL_CRIT, "disabling quorum device operations");
  1191. update_qdevice_can_operate(0);
  1192. }
  1193. }
  1194. /*
  1195. * if user specifies a node list with uneven votes and no device.votes
  1196. * we cannot autocalculate the votes
  1197. */
  1198. if ((have_qdevice) &&
  1199. (qdevice_votes == -1) &&
  1200. (have_nodelist) &&
  1201. (node_count != node_expected_votes)) {
  1202. if (!runtime) {
  1203. error = (char *)"configuration error: quorum.device.votes must be specified when not all nodes votes 1";
  1204. goto out;
  1205. } else {
  1206. log_printf(LOGSYS_LEVEL_CRIT, "configuration error: quorum.device.votes must be specified when not all nodes votes 1");
  1207. log_printf(LOGSYS_LEVEL_CRIT, "disabling quorum device operations");
  1208. update_qdevice_can_operate(0);
  1209. }
  1210. }
  1211. /*
  1212. * validate quorum device votes vs expected_votes
  1213. */
  1214. if ((qdevice_votes > 0) && (expected_votes)) {
  1215. int delta = expected_votes - qdevice_votes;
  1216. if (delta < 2) {
  1217. if (!runtime) {
  1218. error = (char *)"configuration error: quorum.device.votes is too high or expected_votes is too low";
  1219. goto out;
  1220. } else {
  1221. log_printf(LOGSYS_LEVEL_CRIT, "configuration error: quorum.device.votes is too high or expected_votes is too low");
  1222. log_printf(LOGSYS_LEVEL_CRIT, "disabling quorum device operations");
  1223. update_qdevice_can_operate(0);
  1224. }
  1225. }
  1226. }
  1227. /*
  1228. * automatically calculate device votes and adjust expected_votes from nodelist
  1229. */
  1230. if ((have_qdevice) &&
  1231. (qdevice_votes == -1) &&
  1232. (!expected_votes) &&
  1233. (have_nodelist) &&
  1234. (node_count == node_expected_votes)) {
  1235. qdevice_votes = node_expected_votes - 1;
  1236. node_expected_votes = node_expected_votes + qdevice_votes;
  1237. }
  1238. /*
  1239. * set this node votes and expected_votes
  1240. */
  1241. log_printf(LOGSYS_LEVEL_DEBUG, "ev_tracking=%d, ev_tracking_barrier = %d: expected_votes = %d\n", ev_tracking, ev_tracking_barrier, expected_votes);
  1242. if (ev_tracking) {
  1243. expected_votes = ev_tracking_barrier;
  1244. }
  1245. if (have_nodelist) {
  1246. us->votes = node_votes;
  1247. us->expected_votes = node_expected_votes;
  1248. } else {
  1249. us->votes = 1;
  1250. icmap_get_uint32("quorum.votes", &us->votes);
  1251. }
  1252. if (expected_votes) {
  1253. us->expected_votes = expected_votes;
  1254. }
  1255. /*
  1256. * set qdevice votes
  1257. */
  1258. if (!have_qdevice) {
  1259. qdevice->votes = 0;
  1260. }
  1261. if (qdevice_votes != -1) {
  1262. qdevice->votes = qdevice_votes;
  1263. }
  1264. update_ev_barrier(us->expected_votes);
  1265. update_two_node();
  1266. if (wait_for_all) {
  1267. update_wait_for_all_status(1);
  1268. }
  1269. out:
  1270. LEAVE();
  1271. return error;
  1272. }
  1273. static void votequorum_refresh_config(
  1274. int32_t event,
  1275. const char *key_name,
  1276. struct icmap_notify_value new_val,
  1277. struct icmap_notify_value old_val,
  1278. void *user_data)
  1279. {
  1280. int old_votes, old_expected_votes;
  1281. uint8_t reloading;
  1282. uint8_t cancel_wfa;
  1283. ENTER();
  1284. /*
  1285. * If a full reload is in progress then don't do anything until it's done and
  1286. * can reconfigure it all atomically
  1287. */
  1288. if (icmap_get_uint8("config.totemconfig_reload_in_progress", &reloading) == CS_OK && reloading) {
  1289. return ;
  1290. }
  1291. icmap_get_uint8("quorum.cancel_wait_for_all", &cancel_wfa);
  1292. if (strcmp(key_name, "quorum.cancel_wait_for_all") == 0 &&
  1293. cancel_wfa >= 1) {
  1294. icmap_set_uint8("quorum.cancel_wait_for_all", 0);
  1295. votequorum_exec_send_reconfigure(VOTEQUORUM_RECONFIG_PARAM_CANCEL_WFA,
  1296. us->node_id, 0);
  1297. return;
  1298. }
  1299. old_votes = us->votes;
  1300. old_expected_votes = us->expected_votes;
  1301. /*
  1302. * Reload the configuration
  1303. */
  1304. votequorum_readconfig(VOTEQUORUM_READCONFIG_RUNTIME);
  1305. /*
  1306. * activate new config
  1307. */
  1308. votequorum_exec_send_nodeinfo(us->node_id);
  1309. votequorum_exec_send_nodeinfo(VOTEQUORUM_QDEVICE_NODEID);
  1310. if (us->votes != old_votes) {
  1311. votequorum_exec_send_reconfigure(VOTEQUORUM_RECONFIG_PARAM_NODE_VOTES,
  1312. us->node_id, us->votes);
  1313. }
  1314. if (us->expected_votes != old_expected_votes) {
  1315. votequorum_exec_send_reconfigure(VOTEQUORUM_RECONFIG_PARAM_EXPECTED_VOTES,
  1316. us->node_id, us->expected_votes);
  1317. }
  1318. LEAVE();
  1319. }
  1320. static void votequorum_exec_add_config_notification(void)
  1321. {
  1322. icmap_track_t icmap_track_nodelist = NULL;
  1323. icmap_track_t icmap_track_quorum = NULL;
  1324. icmap_track_t icmap_track_reload = NULL;
  1325. ENTER();
  1326. icmap_track_add("nodelist.",
  1327. ICMAP_TRACK_ADD | ICMAP_TRACK_DELETE | ICMAP_TRACK_MODIFY | ICMAP_TRACK_PREFIX,
  1328. votequorum_refresh_config,
  1329. NULL,
  1330. &icmap_track_nodelist);
  1331. icmap_track_add("quorum.",
  1332. ICMAP_TRACK_ADD | ICMAP_TRACK_DELETE | ICMAP_TRACK_MODIFY | ICMAP_TRACK_PREFIX,
  1333. votequorum_refresh_config,
  1334. NULL,
  1335. &icmap_track_quorum);
  1336. icmap_track_add("config.totemconfig_reload_in_progress",
  1337. ICMAP_TRACK_ADD | ICMAP_TRACK_MODIFY,
  1338. votequorum_refresh_config,
  1339. NULL,
  1340. &icmap_track_reload);
  1341. LEAVE();
  1342. }
  1343. /*
  1344. * votequorum_exec core
  1345. */
  1346. static int votequorum_exec_send_reconfigure(uint8_t param, unsigned int nodeid, uint32_t value)
  1347. {
  1348. struct req_exec_quorum_reconfigure req_exec_quorum_reconfigure;
  1349. struct iovec iov[1];
  1350. int ret;
  1351. ENTER();
  1352. req_exec_quorum_reconfigure.nodeid = nodeid;
  1353. req_exec_quorum_reconfigure.value = value;
  1354. req_exec_quorum_reconfigure.param = param;
  1355. req_exec_quorum_reconfigure._pad0 = 0;
  1356. req_exec_quorum_reconfigure._pad1 = 0;
  1357. req_exec_quorum_reconfigure._pad2 = 0;
  1358. req_exec_quorum_reconfigure.header.id = SERVICE_ID_MAKE(VOTEQUORUM_SERVICE, MESSAGE_REQ_EXEC_VOTEQUORUM_RECONFIGURE);
  1359. req_exec_quorum_reconfigure.header.size = sizeof(req_exec_quorum_reconfigure);
  1360. iov[0].iov_base = (void *)&req_exec_quorum_reconfigure;
  1361. iov[0].iov_len = sizeof(req_exec_quorum_reconfigure);
  1362. ret = corosync_api->totem_mcast (iov, 1, TOTEM_AGREED);
  1363. LEAVE();
  1364. return ret;
  1365. }
  1366. static int votequorum_exec_send_nodeinfo(uint32_t nodeid)
  1367. {
  1368. struct req_exec_quorum_nodeinfo req_exec_quorum_nodeinfo;
  1369. struct iovec iov[1];
  1370. struct cluster_node *node;
  1371. int ret;
  1372. ENTER();
  1373. node = find_node_by_nodeid(nodeid);
  1374. if (!node) {
  1375. return -1;
  1376. }
  1377. req_exec_quorum_nodeinfo.nodeid = nodeid;
  1378. req_exec_quorum_nodeinfo.votes = node->votes;
  1379. req_exec_quorum_nodeinfo.expected_votes = node->expected_votes;
  1380. req_exec_quorum_nodeinfo.flags = node->flags;
  1381. if (nodeid != VOTEQUORUM_QDEVICE_NODEID) {
  1382. decode_flags(node->flags);
  1383. }
  1384. req_exec_quorum_nodeinfo.header.id = SERVICE_ID_MAKE(VOTEQUORUM_SERVICE, MESSAGE_REQ_EXEC_VOTEQUORUM_NODEINFO);
  1385. req_exec_quorum_nodeinfo.header.size = sizeof(req_exec_quorum_nodeinfo);
  1386. iov[0].iov_base = (void *)&req_exec_quorum_nodeinfo;
  1387. iov[0].iov_len = sizeof(req_exec_quorum_nodeinfo);
  1388. ret = corosync_api->totem_mcast (iov, 1, TOTEM_AGREED);
  1389. LEAVE();
  1390. return ret;
  1391. }
  1392. static int votequorum_exec_send_qdevice_reconfigure(const char *oldname, const char *newname)
  1393. {
  1394. struct req_exec_quorum_qdevice_reconfigure req_exec_quorum_qdevice_reconfigure;
  1395. struct iovec iov[1];
  1396. int ret;
  1397. ENTER();
  1398. req_exec_quorum_qdevice_reconfigure.header.id = SERVICE_ID_MAKE(VOTEQUORUM_SERVICE, MESSAGE_REQ_EXEC_VOTEQUORUM_QDEVICE_RECONFIGURE);
  1399. req_exec_quorum_qdevice_reconfigure.header.size = sizeof(req_exec_quorum_qdevice_reconfigure);
  1400. strcpy(req_exec_quorum_qdevice_reconfigure.oldname, oldname);
  1401. strcpy(req_exec_quorum_qdevice_reconfigure.newname, newname);
  1402. iov[0].iov_base = (void *)&req_exec_quorum_qdevice_reconfigure;
  1403. iov[0].iov_len = sizeof(req_exec_quorum_qdevice_reconfigure);
  1404. ret = corosync_api->totem_mcast (iov, 1, TOTEM_AGREED);
  1405. LEAVE();
  1406. return ret;
  1407. }
  1408. static int votequorum_exec_send_qdevice_reg(uint32_t operation, const char *qdevice_name_req)
  1409. {
  1410. struct req_exec_quorum_qdevice_reg req_exec_quorum_qdevice_reg;
  1411. struct iovec iov[1];
  1412. int ret;
  1413. ENTER();
  1414. req_exec_quorum_qdevice_reg.header.id = SERVICE_ID_MAKE(VOTEQUORUM_SERVICE, MESSAGE_REQ_EXEC_VOTEQUORUM_QDEVICE_REG);
  1415. req_exec_quorum_qdevice_reg.header.size = sizeof(req_exec_quorum_qdevice_reg);
  1416. req_exec_quorum_qdevice_reg.operation = operation;
  1417. strcpy(req_exec_quorum_qdevice_reg.qdevice_name, qdevice_name_req);
  1418. iov[0].iov_base = (void *)&req_exec_quorum_qdevice_reg;
  1419. iov[0].iov_len = sizeof(req_exec_quorum_qdevice_reg);
  1420. ret = corosync_api->totem_mcast (iov, 1, TOTEM_AGREED);
  1421. LEAVE();
  1422. return ret;
  1423. }
  1424. static int votequorum_exec_send_quorum_notification(void *conn, uint64_t context)
  1425. {
  1426. struct res_lib_votequorum_quorum_notification *res_lib_votequorum_notification;
  1427. struct list_head *tmp;
  1428. struct cluster_node *node;
  1429. int i = 0;
  1430. int cluster_members = 0;
  1431. int size;
  1432. char buf[sizeof(struct res_lib_votequorum_quorum_notification) + sizeof(struct votequorum_node) * (PROCESSOR_COUNT_MAX + 2)];
  1433. ENTER();
  1434. log_printf(LOGSYS_LEVEL_DEBUG, "Sending quorum callback, quorate = %d", cluster_is_quorate);
  1435. list_iterate(tmp, &cluster_members_list) {
  1436. node = list_entry(tmp, struct cluster_node, list);
  1437. cluster_members++;
  1438. }
  1439. if (us->flags & NODE_FLAGS_QDEVICE_REGISTERED) {
  1440. cluster_members++;
  1441. }
  1442. size = sizeof(struct res_lib_votequorum_quorum_notification) + sizeof(struct votequorum_node) * cluster_members;
  1443. res_lib_votequorum_notification = (struct res_lib_votequorum_quorum_notification *)&buf;
  1444. res_lib_votequorum_notification->quorate = cluster_is_quorate;
  1445. res_lib_votequorum_notification->context = context;
  1446. res_lib_votequorum_notification->node_list_entries = cluster_members;
  1447. res_lib_votequorum_notification->header.id = MESSAGE_RES_VOTEQUORUM_QUORUM_NOTIFICATION;
  1448. res_lib_votequorum_notification->header.size = size;
  1449. res_lib_votequorum_notification->header.error = CS_OK;
  1450. /* Send all known nodes and their states */
  1451. list_iterate(tmp, &cluster_members_list) {
  1452. node = list_entry(tmp, struct cluster_node, list);
  1453. res_lib_votequorum_notification->node_list[i].nodeid = node->node_id;
  1454. res_lib_votequorum_notification->node_list[i++].state = node->state;
  1455. }
  1456. if (us->flags & NODE_FLAGS_QDEVICE_REGISTERED) {
  1457. res_lib_votequorum_notification->node_list[i].nodeid = VOTEQUORUM_QDEVICE_NODEID;
  1458. res_lib_votequorum_notification->node_list[i++].state = qdevice->state;
  1459. }
  1460. /* Send it to all interested parties */
  1461. if (conn) {
  1462. int ret = corosync_api->ipc_dispatch_send(conn, &buf, size);
  1463. LEAVE();
  1464. return ret;
  1465. } else {
  1466. struct quorum_pd *qpd;
  1467. list_iterate(tmp, &trackers_list) {
  1468. qpd = list_entry(tmp, struct quorum_pd, list);
  1469. res_lib_votequorum_notification->context = qpd->tracking_context;
  1470. corosync_api->ipc_dispatch_send(qpd->conn, &buf, size);
  1471. }
  1472. }
  1473. LEAVE();
  1474. return 0;
  1475. }
  1476. static int votequorum_exec_send_nodelist_notification(void *conn, uint64_t context)
  1477. {
  1478. struct res_lib_votequorum_nodelist_notification *res_lib_votequorum_notification;
  1479. int i = 0;
  1480. int size;
  1481. struct list_head *tmp;
  1482. char buf[sizeof(struct res_lib_votequorum_nodelist_notification) + sizeof(uint32_t) * quorum_members_entries];
  1483. ENTER();
  1484. log_printf(LOGSYS_LEVEL_DEBUG, "Sending nodelist callback. ring_id = %d/%lld", quorum_ringid.rep.nodeid, quorum_ringid.seq);
  1485. size = sizeof(struct res_lib_votequorum_nodelist_notification) + sizeof(uint32_t) * quorum_members_entries;
  1486. res_lib_votequorum_notification = (struct res_lib_votequorum_nodelist_notification *)&buf;
  1487. res_lib_votequorum_notification->node_list_entries = quorum_members_entries;
  1488. res_lib_votequorum_notification->ring_id.nodeid = quorum_ringid.rep.nodeid;
  1489. res_lib_votequorum_notification->ring_id.seq = quorum_ringid.seq;
  1490. res_lib_votequorum_notification->context = context;
  1491. for (i=0; i<quorum_members_entries; i++) {
  1492. res_lib_votequorum_notification->node_list[i] = quorum_members[i];
  1493. }
  1494. res_lib_votequorum_notification->header.id = MESSAGE_RES_VOTEQUORUM_NODELIST_NOTIFICATION;
  1495. res_lib_votequorum_notification->header.size = size;
  1496. res_lib_votequorum_notification->header.error = CS_OK;
  1497. /* Send it to all interested parties */
  1498. if (conn) {
  1499. int ret = corosync_api->ipc_dispatch_send(conn, &buf, size);
  1500. LEAVE();
  1501. return ret;
  1502. } else {
  1503. struct quorum_pd *qpd;
  1504. list_iterate(tmp, &trackers_list) {
  1505. qpd = list_entry(tmp, struct quorum_pd, list);
  1506. res_lib_votequorum_notification->context = qpd->tracking_context;
  1507. corosync_api->ipc_dispatch_send(qpd->conn, &buf, size);
  1508. }
  1509. }
  1510. LEAVE();
  1511. return 0;
  1512. }
  1513. static void votequorum_exec_send_expectedvotes_notification(void)
  1514. {
  1515. struct res_lib_votequorum_expectedvotes_notification res_lib_votequorum_expectedvotes_notification;
  1516. struct quorum_pd *qpd;
  1517. struct list_head *tmp;
  1518. ENTER();
  1519. log_printf(LOGSYS_LEVEL_DEBUG, "Sending expected votes callback");
  1520. res_lib_votequorum_expectedvotes_notification.header.id = MESSAGE_RES_VOTEQUORUM_EXPECTEDVOTES_NOTIFICATION;
  1521. res_lib_votequorum_expectedvotes_notification.header.size = sizeof(res_lib_votequorum_expectedvotes_notification);
  1522. res_lib_votequorum_expectedvotes_notification.header.error = CS_OK;
  1523. res_lib_votequorum_expectedvotes_notification.expected_votes = us->expected_votes;
  1524. list_iterate(tmp, &trackers_list) {
  1525. qpd = list_entry(tmp, struct quorum_pd, list);
  1526. res_lib_votequorum_expectedvotes_notification.context = qpd->tracking_context;
  1527. corosync_api->ipc_dispatch_send(qpd->conn, &res_lib_votequorum_expectedvotes_notification,
  1528. sizeof(struct res_lib_votequorum_expectedvotes_notification));
  1529. }
  1530. LEAVE();
  1531. }
  1532. static void exec_votequorum_qdevice_reconfigure_endian_convert (void *message)
  1533. {
  1534. ENTER();
  1535. LEAVE();
  1536. }
  1537. static void message_handler_req_exec_votequorum_qdevice_reconfigure (
  1538. const void *message,
  1539. unsigned int nodeid)
  1540. {
  1541. const struct req_exec_quorum_qdevice_reconfigure *req_exec_quorum_qdevice_reconfigure = message;
  1542. ENTER();
  1543. log_printf(LOGSYS_LEVEL_DEBUG, "Received qdevice name change req from node %u [from: %s to: %s]",
  1544. nodeid,
  1545. req_exec_quorum_qdevice_reconfigure->oldname,
  1546. req_exec_quorum_qdevice_reconfigure->newname);
  1547. if (!strcmp(req_exec_quorum_qdevice_reconfigure->oldname, qdevice_name)) {
  1548. log_printf(LOGSYS_LEVEL_DEBUG, "Allowing qdevice rename");
  1549. memset(qdevice_name, 0, VOTEQUORUM_QDEVICE_MAX_NAME_LEN);
  1550. strcpy(qdevice_name, req_exec_quorum_qdevice_reconfigure->newname);
  1551. /*
  1552. * TODO: notify qdevices about name change?
  1553. * this is not relevant for now and can wait later on since
  1554. * qdevices are local only and libvotequorum is not final
  1555. */
  1556. }
  1557. LEAVE();
  1558. }
  1559. static void exec_votequorum_qdevice_reg_endian_convert (void *message)
  1560. {
  1561. struct req_exec_quorum_qdevice_reg *req_exec_quorum_qdevice_reg = message;
  1562. ENTER();
  1563. req_exec_quorum_qdevice_reg->operation = swab32(req_exec_quorum_qdevice_reg->operation);
  1564. LEAVE();
  1565. }
  1566. static void message_handler_req_exec_votequorum_qdevice_reg (
  1567. const void *message,
  1568. unsigned int nodeid)
  1569. {
  1570. const struct req_exec_quorum_qdevice_reg *req_exec_quorum_qdevice_reg = message;
  1571. struct res_lib_votequorum_status res_lib_votequorum_status;
  1572. int wipe_qdevice_name = 1;
  1573. struct cluster_node *node = NULL;
  1574. struct list_head *tmp;
  1575. cs_error_t error = CS_OK;
  1576. ENTER();
  1577. log_printf(LOGSYS_LEVEL_DEBUG, "Received qdevice op %u req from node %u [%s]",
  1578. req_exec_quorum_qdevice_reg->operation,
  1579. nodeid, req_exec_quorum_qdevice_reg->qdevice_name);
  1580. switch(req_exec_quorum_qdevice_reg->operation)
  1581. {
  1582. case VOTEQUORUM_QDEVICE_OPERATION_REGISTER:
  1583. if (nodeid != us->node_id) {
  1584. if (!strlen(qdevice_name)) {
  1585. log_printf(LOGSYS_LEVEL_DEBUG, "Remote qdevice name recorded");
  1586. strcpy(qdevice_name, req_exec_quorum_qdevice_reg->qdevice_name);
  1587. }
  1588. LEAVE();
  1589. return;
  1590. }
  1591. /*
  1592. * protect against the case where we broadcast qdevice registration
  1593. * to new memebers, we receive the message back, but there is no registration
  1594. * connection in progress
  1595. */
  1596. if (us->flags & NODE_FLAGS_QDEVICE_REGISTERED) {
  1597. LEAVE();
  1598. return;
  1599. }
  1600. /*
  1601. * this should NEVER happen
  1602. */
  1603. if (!qdevice_reg_conn) {
  1604. log_printf(LOGSYS_LEVEL_WARNING, "Unable to determine origin of the qdevice register call!");
  1605. LEAVE();
  1606. return;
  1607. }
  1608. /*
  1609. * registering our own device in this case
  1610. */
  1611. if (!strlen(qdevice_name)) {
  1612. strcpy(qdevice_name, req_exec_quorum_qdevice_reg->qdevice_name);
  1613. }
  1614. /*
  1615. * check if it is our device or something else
  1616. */
  1617. if ((!strncmp(req_exec_quorum_qdevice_reg->qdevice_name,
  1618. qdevice_name, VOTEQUORUM_QDEVICE_MAX_NAME_LEN))) {
  1619. us->flags |= NODE_FLAGS_QDEVICE_REGISTERED;
  1620. votequorum_exec_send_nodeinfo(VOTEQUORUM_QDEVICE_NODEID);
  1621. votequorum_exec_send_nodeinfo(us->node_id);
  1622. } else {
  1623. log_printf(LOGSYS_LEVEL_WARNING,
  1624. "A new qdevice with different name (new: %s old: %s) is trying to register!",
  1625. req_exec_quorum_qdevice_reg->qdevice_name, qdevice_name);
  1626. error = CS_ERR_EXIST;
  1627. }
  1628. res_lib_votequorum_status.header.size = sizeof(res_lib_votequorum_status);
  1629. res_lib_votequorum_status.header.id = MESSAGE_RES_VOTEQUORUM_STATUS;
  1630. res_lib_votequorum_status.header.error = error;
  1631. corosync_api->ipc_response_send(qdevice_reg_conn, &res_lib_votequorum_status, sizeof(res_lib_votequorum_status));
  1632. qdevice_reg_conn = NULL;
  1633. break;
  1634. case VOTEQUORUM_QDEVICE_OPERATION_UNREGISTER:
  1635. list_iterate(tmp, &cluster_members_list) {
  1636. node = list_entry(tmp, struct cluster_node, list);
  1637. if ((node->state == NODESTATE_MEMBER) &&
  1638. (node->flags & NODE_FLAGS_QDEVICE_REGISTERED)) {
  1639. wipe_qdevice_name = 0;
  1640. }
  1641. }
  1642. if (wipe_qdevice_name) {
  1643. memset(qdevice_name, 0, VOTEQUORUM_QDEVICE_MAX_NAME_LEN);
  1644. }
  1645. break;
  1646. }
  1647. LEAVE();
  1648. }
  1649. static void exec_votequorum_nodeinfo_endian_convert (void *message)
  1650. {
  1651. struct req_exec_quorum_nodeinfo *nodeinfo = message;
  1652. ENTER();
  1653. nodeinfo->nodeid = swab32(nodeinfo->nodeid);
  1654. nodeinfo->votes = swab32(nodeinfo->votes);
  1655. nodeinfo->expected_votes = swab32(nodeinfo->expected_votes);
  1656. nodeinfo->flags = swab32(nodeinfo->flags);
  1657. LEAVE();
  1658. }
  1659. static void message_handler_req_exec_votequorum_nodeinfo (
  1660. const void *message,
  1661. unsigned int sender_nodeid)
  1662. {
  1663. const struct req_exec_quorum_nodeinfo *req_exec_quorum_nodeinfo = message;
  1664. struct cluster_node *node = NULL;
  1665. int old_votes;
  1666. int old_expected;
  1667. uint32_t old_flags;
  1668. nodestate_t old_state;
  1669. int new_node = 0;
  1670. int allow_downgrade = 0;
  1671. int by_node = 0;
  1672. unsigned int nodeid = req_exec_quorum_nodeinfo->nodeid;
  1673. ENTER();
  1674. log_printf(LOGSYS_LEVEL_DEBUG, "got nodeinfo message from cluster node %u", sender_nodeid);
  1675. log_printf(LOGSYS_LEVEL_DEBUG, "nodeinfo message[%u]: votes: %d, expected: %d flags: %d",
  1676. nodeid,
  1677. req_exec_quorum_nodeinfo->votes,
  1678. req_exec_quorum_nodeinfo->expected_votes,
  1679. req_exec_quorum_nodeinfo->flags);
  1680. if (nodeid != VOTEQUORUM_QDEVICE_NODEID) {
  1681. decode_flags(req_exec_quorum_nodeinfo->flags);
  1682. }
  1683. node = find_node_by_nodeid(nodeid);
  1684. if (!node) {
  1685. node = allocate_node(nodeid);
  1686. new_node = 1;
  1687. }
  1688. if (!node) {
  1689. corosync_api->error_memory_failure();
  1690. LEAVE();
  1691. return;
  1692. }
  1693. if (new_node) {
  1694. old_votes = 0;
  1695. old_expected = 0;
  1696. old_state = NODESTATE_DEAD;
  1697. old_flags = 0;
  1698. } else {
  1699. old_votes = node->votes;
  1700. old_expected = node->expected_votes;
  1701. old_state = node->state;
  1702. old_flags = node->flags;
  1703. }
  1704. if (nodeid == VOTEQUORUM_QDEVICE_NODEID) {
  1705. struct cluster_node *sender_node = find_node_by_nodeid(sender_nodeid);
  1706. assert(sender_node != NULL);
  1707. if ((!cluster_is_quorate) &&
  1708. (sender_node->flags & NODE_FLAGS_QUORATE)) {
  1709. node->votes = req_exec_quorum_nodeinfo->votes;
  1710. } else {
  1711. node->votes = max(node->votes, req_exec_quorum_nodeinfo->votes);
  1712. }
  1713. goto recalculate;
  1714. }
  1715. /* Update node state */
  1716. node->flags = req_exec_quorum_nodeinfo->flags;
  1717. node->votes = req_exec_quorum_nodeinfo->votes;
  1718. node->state = NODESTATE_MEMBER;
  1719. if (node->flags & NODE_FLAGS_LEAVING) {
  1720. node->state = NODESTATE_LEAVING;
  1721. allow_downgrade = 1;
  1722. by_node = 1;
  1723. }
  1724. if ((!cluster_is_quorate) &&
  1725. (node->flags & NODE_FLAGS_QUORATE)) {
  1726. allow_downgrade = 1;
  1727. us->expected_votes = req_exec_quorum_nodeinfo->expected_votes;
  1728. }
  1729. if (node->flags & NODE_FLAGS_QUORATE || (ev_tracking)) {
  1730. node->expected_votes = req_exec_quorum_nodeinfo->expected_votes;
  1731. } else {
  1732. node->expected_votes = us->expected_votes;
  1733. }
  1734. if ((last_man_standing) && (node->votes > 1)) {
  1735. log_printf(LOGSYS_LEVEL_WARNING, "Last Man Standing feature is supported only when all"
  1736. "cluster nodes votes are set to 1. Disabling LMS.");
  1737. last_man_standing = 0;
  1738. if (last_man_standing_timer_set) {
  1739. corosync_api->timer_delete(last_man_standing_timer);
  1740. last_man_standing_timer_set = 0;
  1741. }
  1742. }
  1743. recalculate:
  1744. if ((new_node) ||
  1745. (nodeid == us->node_id) ||
  1746. (node->flags & NODE_FLAGS_FIRST) ||
  1747. (old_votes != node->votes) ||
  1748. (old_expected != node->expected_votes) ||
  1749. (old_flags != node->flags) ||
  1750. (old_state != node->state)) {
  1751. recalculate_quorum(allow_downgrade, by_node);
  1752. }
  1753. if ((wait_for_all) &&
  1754. (!(node->flags & NODE_FLAGS_WFASTATUS)) &&
  1755. (node->flags & NODE_FLAGS_QUORATE)) {
  1756. update_wait_for_all_status(0);
  1757. }
  1758. LEAVE();
  1759. }
  1760. static void exec_votequorum_reconfigure_endian_convert (void *message)
  1761. {
  1762. struct req_exec_quorum_reconfigure *reconfigure = message;
  1763. ENTER();
  1764. reconfigure->nodeid = swab32(reconfigure->nodeid);
  1765. reconfigure->value = swab32(reconfigure->value);
  1766. LEAVE();
  1767. }
  1768. static void message_handler_req_exec_votequorum_reconfigure (
  1769. const void *message,
  1770. unsigned int nodeid)
  1771. {
  1772. const struct req_exec_quorum_reconfigure *req_exec_quorum_reconfigure = message;
  1773. struct cluster_node *node;
  1774. struct list_head *nodelist;
  1775. ENTER();
  1776. log_printf(LOGSYS_LEVEL_DEBUG, "got reconfigure message from cluster node %u for %u",
  1777. nodeid, req_exec_quorum_reconfigure->nodeid);
  1778. switch(req_exec_quorum_reconfigure->param)
  1779. {
  1780. case VOTEQUORUM_RECONFIG_PARAM_EXPECTED_VOTES:
  1781. list_iterate(nodelist, &cluster_members_list) {
  1782. node = list_entry(nodelist, struct cluster_node, list);
  1783. if (node->state == NODESTATE_MEMBER) {
  1784. node->expected_votes = req_exec_quorum_reconfigure->value;
  1785. }
  1786. }
  1787. votequorum_exec_send_expectedvotes_notification();
  1788. update_ev_barrier(req_exec_quorum_reconfigure->value);
  1789. if (ev_tracking) {
  1790. us->expected_votes = max(us->expected_votes, ev_tracking_barrier);
  1791. }
  1792. recalculate_quorum(1, 0); /* Allow decrease */
  1793. break;
  1794. case VOTEQUORUM_RECONFIG_PARAM_NODE_VOTES:
  1795. node = find_node_by_nodeid(req_exec_quorum_reconfigure->nodeid);
  1796. if (!node) {
  1797. LEAVE();
  1798. return;
  1799. }
  1800. node->votes = req_exec_quorum_reconfigure->value;
  1801. recalculate_quorum(1, 0); /* Allow decrease */
  1802. break;
  1803. case VOTEQUORUM_RECONFIG_PARAM_CANCEL_WFA:
  1804. update_wait_for_all_status(0);
  1805. log_printf(LOGSYS_LEVEL_INFO, "wait_for_all_status reset by user on node %d.",
  1806. req_exec_quorum_reconfigure->nodeid);
  1807. recalculate_quorum(0, 0);
  1808. break;
  1809. }
  1810. LEAVE();
  1811. }
  1812. static int votequorum_exec_exit_fn (void)
  1813. {
  1814. int ret = 0;
  1815. ENTER();
  1816. /*
  1817. * tell the other nodes we are leaving
  1818. */
  1819. if (allow_downscale) {
  1820. us->flags |= NODE_FLAGS_LEAVING;
  1821. ret = votequorum_exec_send_nodeinfo(us->node_id);
  1822. }
  1823. if ((ev_tracking) && (ev_tracking_fd != -1)) {
  1824. close(ev_tracking_fd);
  1825. }
  1826. LEAVE();
  1827. return ret;
  1828. }
  1829. static void votequorum_set_icmap_ro_keys(void)
  1830. {
  1831. icmap_set_ro_access("quorum.allow_downscale", CS_FALSE, CS_TRUE);
  1832. icmap_set_ro_access("quorum.wait_for_all", CS_FALSE, CS_TRUE);
  1833. icmap_set_ro_access("quorum.last_man_standing", CS_FALSE, CS_TRUE);
  1834. icmap_set_ro_access("quorum.last_man_standing_window", CS_FALSE, CS_TRUE);
  1835. icmap_set_ro_access("quorum.expected_votes_tracking", CS_FALSE, CS_TRUE);
  1836. icmap_set_ro_access("quorum.auto_tie_breaker", CS_FALSE, CS_TRUE);
  1837. icmap_set_ro_access("quorum.auto_tie_breaker_node", CS_FALSE, CS_TRUE);
  1838. }
  1839. static char *votequorum_exec_init_fn (struct corosync_api_v1 *api)
  1840. {
  1841. char *error = NULL;
  1842. ENTER();
  1843. /*
  1844. * make sure we start clean
  1845. */
  1846. list_init(&cluster_members_list);
  1847. list_init(&trackers_list);
  1848. qdevice = NULL;
  1849. us = NULL;
  1850. memset(cluster_nodes, 0, sizeof(cluster_nodes));
  1851. /*
  1852. * Allocate a cluster_node for qdevice
  1853. */
  1854. qdevice = allocate_node(VOTEQUORUM_QDEVICE_NODEID);
  1855. if (!qdevice) {
  1856. LEAVE();
  1857. return ((char *)"Could not allocate node.");
  1858. }
  1859. qdevice->votes = 0;
  1860. memset(qdevice_name, 0, VOTEQUORUM_QDEVICE_MAX_NAME_LEN);
  1861. /*
  1862. * Allocate a cluster_node for us
  1863. */
  1864. us = allocate_node(corosync_api->totem_nodeid_get());
  1865. if (!us) {
  1866. LEAVE();
  1867. return ((char *)"Could not allocate node.");
  1868. }
  1869. icmap_set_uint32("runtime.votequorum.this_node_id", us->node_id);
  1870. us->state = NODESTATE_MEMBER;
  1871. us->votes = 1;
  1872. us->flags |= NODE_FLAGS_FIRST;
  1873. error = votequorum_readconfig(VOTEQUORUM_READCONFIG_STARTUP);
  1874. if (error) {
  1875. return error;
  1876. }
  1877. recalculate_quorum(0, 0);
  1878. /*
  1879. * Set RO keys in icmap
  1880. */
  1881. votequorum_set_icmap_ro_keys();
  1882. /*
  1883. * Listen for changes
  1884. */
  1885. votequorum_exec_add_config_notification();
  1886. /*
  1887. * Start us off with one node
  1888. */
  1889. votequorum_exec_send_nodeinfo(us->node_id);
  1890. LEAVE();
  1891. return (NULL);
  1892. }
  1893. /*
  1894. * votequorum service core
  1895. */
  1896. static void votequorum_last_man_standing_timer_fn(void *arg)
  1897. {
  1898. ENTER();
  1899. last_man_standing_timer_set = 0;
  1900. if (cluster_is_quorate) {
  1901. recalculate_quorum(1,1);
  1902. }
  1903. LEAVE();
  1904. }
  1905. static void votequorum_sync_init (
  1906. const unsigned int *trans_list, size_t trans_list_entries,
  1907. const unsigned int *member_list, size_t member_list_entries,
  1908. const struct memb_ring_id *ring_id)
  1909. {
  1910. int i, j;
  1911. int found;
  1912. int left_nodes;
  1913. struct cluster_node *node;
  1914. ENTER();
  1915. sync_in_progress = 1;
  1916. sync_nodeinfo_sent = 0;
  1917. sync_wait_for_poll_or_timeout = 0;
  1918. if (member_list_entries > 1) {
  1919. us->flags &= ~NODE_FLAGS_FIRST;
  1920. }
  1921. /*
  1922. * we don't need to track which nodes have left directly,
  1923. * since that info is in the node db, but we need to know
  1924. * if somebody has left for last_man_standing
  1925. */
  1926. left_nodes = 0;
  1927. for (i = 0; i < quorum_members_entries; i++) {
  1928. found = 0;
  1929. for (j = 0; j < member_list_entries; j++) {
  1930. if (quorum_members[i] == member_list[j]) {
  1931. found = 1;
  1932. break;
  1933. }
  1934. }
  1935. if (found == 0) {
  1936. left_nodes = 1;
  1937. node = find_node_by_nodeid(quorum_members[i]);
  1938. if (node) {
  1939. node->state = NODESTATE_DEAD;
  1940. }
  1941. }
  1942. }
  1943. if (last_man_standing) {
  1944. if (((member_list_entries >= quorum) && (left_nodes)) ||
  1945. ((member_list_entries <= quorum) && (auto_tie_breaker != ATB_NONE) && (check_low_node_id_partition() == 1))) {
  1946. if (last_man_standing_timer_set) {
  1947. corosync_api->timer_delete(last_man_standing_timer);
  1948. last_man_standing_timer_set = 0;
  1949. }
  1950. corosync_api->timer_add_duration((unsigned long long)last_man_standing_window*1000000,
  1951. NULL, votequorum_last_man_standing_timer_fn,
  1952. &last_man_standing_timer);
  1953. last_man_standing_timer_set = 1;
  1954. }
  1955. }
  1956. memcpy(previous_quorum_members, quorum_members, sizeof(unsigned int) * quorum_members_entries);
  1957. previous_quorum_members_entries = quorum_members_entries;
  1958. memcpy(quorum_members, member_list, sizeof(unsigned int) * member_list_entries);
  1959. quorum_members_entries = member_list_entries;
  1960. memcpy(&quorum_ringid, ring_id, sizeof(*ring_id));
  1961. if (us->flags & NODE_FLAGS_QDEVICE_REGISTERED && us->flags & NODE_FLAGS_QDEVICE_ALIVE) {
  1962. /*
  1963. * Reset poll timer. Sync waiting is interrupted on valid qdevice poll or after timeout
  1964. */
  1965. if (qdevice_timer_set) {
  1966. corosync_api->timer_delete(qdevice_timer);
  1967. }
  1968. corosync_api->timer_add_duration((unsigned long long)qdevice_sync_timeout*1000000, qdevice,
  1969. qdevice_timer_fn, &qdevice_timer);
  1970. qdevice_timer_set = 1;
  1971. sync_wait_for_poll_or_timeout = 1;
  1972. log_printf(LOGSYS_LEVEL_INFO, "waiting for quorum device %s poll (but maximum for %u ms)",
  1973. qdevice_name, qdevice_sync_timeout);
  1974. }
  1975. LEAVE();
  1976. }
  1977. static int votequorum_sync_process (void)
  1978. {
  1979. if (!sync_nodeinfo_sent) {
  1980. votequorum_exec_send_nodeinfo(us->node_id);
  1981. votequorum_exec_send_nodeinfo(VOTEQUORUM_QDEVICE_NODEID);
  1982. if (strlen(qdevice_name)) {
  1983. votequorum_exec_send_qdevice_reg(VOTEQUORUM_QDEVICE_OPERATION_REGISTER,
  1984. qdevice_name);
  1985. }
  1986. votequorum_exec_send_nodelist_notification(NULL, 0LL);
  1987. sync_nodeinfo_sent = 1;
  1988. }
  1989. if (us->flags & NODE_FLAGS_QDEVICE_REGISTERED && sync_wait_for_poll_or_timeout) {
  1990. /*
  1991. * Waiting for qdevice to poll with new ringid or timeout
  1992. */
  1993. return (-1);
  1994. }
  1995. return 0;
  1996. }
  1997. static void votequorum_sync_activate (void)
  1998. {
  1999. recalculate_quorum(0, 0);
  2000. do_quorum_callback();
  2001. votequorum_exec_send_quorum_notification(NULL, 0L);
  2002. sync_in_progress = 0;
  2003. }
  2004. static void votequorum_sync_abort (void)
  2005. {
  2006. }
  2007. char *votequorum_init(struct corosync_api_v1 *api,
  2008. quorum_set_quorate_fn_t q_set_quorate_fn)
  2009. {
  2010. char *error;
  2011. ENTER();
  2012. if (q_set_quorate_fn == NULL) {
  2013. return ((char *)"Quorate function not set");
  2014. }
  2015. corosync_api = api;
  2016. quorum_callback = q_set_quorate_fn;
  2017. error = corosync_service_link_and_init(corosync_api,
  2018. &votequorum_service[0]);
  2019. if (error) {
  2020. return (error);
  2021. }
  2022. LEAVE();
  2023. return (NULL);
  2024. }
  2025. /*
  2026. * Library Handler init/fini
  2027. */
  2028. static int quorum_lib_init_fn (void *conn)
  2029. {
  2030. struct quorum_pd *pd = (struct quorum_pd *)corosync_api->ipc_private_data_get (conn);
  2031. ENTER();
  2032. list_init (&pd->list);
  2033. pd->conn = conn;
  2034. LEAVE();
  2035. return (0);
  2036. }
  2037. static int quorum_lib_exit_fn (void *conn)
  2038. {
  2039. struct quorum_pd *quorum_pd = (struct quorum_pd *)corosync_api->ipc_private_data_get (conn);
  2040. ENTER();
  2041. if (quorum_pd->tracking_enabled) {
  2042. list_del (&quorum_pd->list);
  2043. list_init (&quorum_pd->list);
  2044. }
  2045. LEAVE();
  2046. return (0);
  2047. }
  2048. /*
  2049. * library internal functions
  2050. */
  2051. static void qdevice_timer_fn(void *arg)
  2052. {
  2053. ENTER();
  2054. if ((!(us->flags & NODE_FLAGS_QDEVICE_ALIVE)) ||
  2055. (!qdevice_timer_set)) {
  2056. LEAVE();
  2057. return;
  2058. }
  2059. us->flags &= ~NODE_FLAGS_QDEVICE_ALIVE;
  2060. us->flags &= ~NODE_FLAGS_QDEVICE_CAST_VOTE;
  2061. log_printf(LOGSYS_LEVEL_INFO, "lost contact with quorum device %s", qdevice_name);
  2062. votequorum_exec_send_nodeinfo(us->node_id);
  2063. qdevice_timer_set = 0;
  2064. sync_wait_for_poll_or_timeout = 0;
  2065. LEAVE();
  2066. }
  2067. /*
  2068. * Library Handler Functions
  2069. */
  2070. static void message_handler_req_lib_votequorum_getinfo (void *conn, const void *message)
  2071. {
  2072. const struct req_lib_votequorum_getinfo *req_lib_votequorum_getinfo = message;
  2073. struct res_lib_votequorum_getinfo res_lib_votequorum_getinfo;
  2074. struct cluster_node *node;
  2075. unsigned int highest_expected = 0;
  2076. unsigned int total_votes = 0;
  2077. cs_error_t error = CS_OK;
  2078. uint32_t nodeid = req_lib_votequorum_getinfo->nodeid;
  2079. ENTER();
  2080. log_printf(LOGSYS_LEVEL_DEBUG, "got getinfo request on %p for node %u", conn, req_lib_votequorum_getinfo->nodeid);
  2081. if (nodeid == VOTEQUORUM_QDEVICE_NODEID) {
  2082. nodeid = us->node_id;
  2083. }
  2084. node = find_node_by_nodeid(nodeid);
  2085. if (node) {
  2086. struct cluster_node *iternode;
  2087. struct list_head *nodelist;
  2088. list_iterate(nodelist, &cluster_members_list) {
  2089. iternode = list_entry(nodelist, struct cluster_node, list);
  2090. if (iternode->state == NODESTATE_MEMBER) {
  2091. highest_expected =
  2092. max(highest_expected, iternode->expected_votes);
  2093. total_votes += iternode->votes;
  2094. }
  2095. }
  2096. if (node->flags & NODE_FLAGS_QDEVICE_CAST_VOTE) {
  2097. total_votes += qdevice->votes;
  2098. }
  2099. switch(node->state) {
  2100. case NODESTATE_MEMBER:
  2101. res_lib_votequorum_getinfo.state = VOTEQUORUM_NODESTATE_MEMBER;
  2102. break;
  2103. case NODESTATE_DEAD:
  2104. res_lib_votequorum_getinfo.state = VOTEQUORUM_NODESTATE_DEAD;
  2105. break;
  2106. case NODESTATE_LEAVING:
  2107. res_lib_votequorum_getinfo.state = VOTEQUORUM_NODESTATE_LEAVING;
  2108. break;
  2109. default:
  2110. res_lib_votequorum_getinfo.state = node->state;
  2111. break;
  2112. }
  2113. res_lib_votequorum_getinfo.state = node->state;
  2114. res_lib_votequorum_getinfo.votes = node->votes;
  2115. res_lib_votequorum_getinfo.expected_votes = node->expected_votes;
  2116. res_lib_votequorum_getinfo.highest_expected = highest_expected;
  2117. res_lib_votequorum_getinfo.quorum = quorum;
  2118. res_lib_votequorum_getinfo.total_votes = total_votes;
  2119. res_lib_votequorum_getinfo.flags = 0;
  2120. res_lib_votequorum_getinfo.nodeid = node->node_id;
  2121. if (two_node) {
  2122. res_lib_votequorum_getinfo.flags |= VOTEQUORUM_INFO_TWONODE;
  2123. }
  2124. if (cluster_is_quorate) {
  2125. res_lib_votequorum_getinfo.flags |= VOTEQUORUM_INFO_QUORATE;
  2126. }
  2127. if (wait_for_all) {
  2128. res_lib_votequorum_getinfo.flags |= VOTEQUORUM_INFO_WAIT_FOR_ALL;
  2129. }
  2130. if (last_man_standing) {
  2131. res_lib_votequorum_getinfo.flags |= VOTEQUORUM_INFO_LAST_MAN_STANDING;
  2132. }
  2133. if (auto_tie_breaker != ATB_NONE) {
  2134. res_lib_votequorum_getinfo.flags |= VOTEQUORUM_INFO_AUTO_TIE_BREAKER;
  2135. }
  2136. if (allow_downscale) {
  2137. res_lib_votequorum_getinfo.flags |= VOTEQUORUM_INFO_ALLOW_DOWNSCALE;
  2138. }
  2139. memset(res_lib_votequorum_getinfo.qdevice_name, 0, VOTEQUORUM_QDEVICE_MAX_NAME_LEN);
  2140. strcpy(res_lib_votequorum_getinfo.qdevice_name, qdevice_name);
  2141. res_lib_votequorum_getinfo.qdevice_votes = qdevice->votes;
  2142. if (node->flags & NODE_FLAGS_QDEVICE_REGISTERED) {
  2143. res_lib_votequorum_getinfo.flags |= VOTEQUORUM_INFO_QDEVICE_REGISTERED;
  2144. }
  2145. if (node->flags & NODE_FLAGS_QDEVICE_ALIVE) {
  2146. res_lib_votequorum_getinfo.flags |= VOTEQUORUM_INFO_QDEVICE_ALIVE;
  2147. }
  2148. if (node->flags & NODE_FLAGS_QDEVICE_CAST_VOTE) {
  2149. res_lib_votequorum_getinfo.flags |= VOTEQUORUM_INFO_QDEVICE_CAST_VOTE;
  2150. }
  2151. if (node->flags & NODE_FLAGS_QDEVICE_MASTER_WINS) {
  2152. res_lib_votequorum_getinfo.flags |= VOTEQUORUM_INFO_QDEVICE_MASTER_WINS;
  2153. }
  2154. } else {
  2155. error = CS_ERR_NOT_EXIST;
  2156. }
  2157. res_lib_votequorum_getinfo.header.size = sizeof(res_lib_votequorum_getinfo);
  2158. res_lib_votequorum_getinfo.header.id = MESSAGE_RES_VOTEQUORUM_GETINFO;
  2159. res_lib_votequorum_getinfo.header.error = error;
  2160. corosync_api->ipc_response_send(conn, &res_lib_votequorum_getinfo, sizeof(res_lib_votequorum_getinfo));
  2161. log_printf(LOGSYS_LEVEL_DEBUG, "getinfo response error: %d", error);
  2162. LEAVE();
  2163. }
  2164. static void message_handler_req_lib_votequorum_setexpected (void *conn, const void *message)
  2165. {
  2166. const struct req_lib_votequorum_setexpected *req_lib_votequorum_setexpected = message;
  2167. struct res_lib_votequorum_status res_lib_votequorum_status;
  2168. cs_error_t error = CS_OK;
  2169. unsigned int newquorum;
  2170. unsigned int total_votes;
  2171. uint8_t allow_downscale_status = 0;
  2172. ENTER();
  2173. allow_downscale_status = allow_downscale;
  2174. allow_downscale = 0;
  2175. /*
  2176. * Validate new expected votes
  2177. */
  2178. newquorum = calculate_quorum(1, req_lib_votequorum_setexpected->expected_votes, &total_votes);
  2179. allow_downscale = allow_downscale_status;
  2180. if (newquorum < total_votes / 2 ||
  2181. newquorum > total_votes) {
  2182. error = CS_ERR_INVALID_PARAM;
  2183. goto error_exit;
  2184. }
  2185. votequorum_exec_send_reconfigure(VOTEQUORUM_RECONFIG_PARAM_EXPECTED_VOTES, us->node_id,
  2186. req_lib_votequorum_setexpected->expected_votes);
  2187. error_exit:
  2188. res_lib_votequorum_status.header.size = sizeof(res_lib_votequorum_status);
  2189. res_lib_votequorum_status.header.id = MESSAGE_RES_VOTEQUORUM_STATUS;
  2190. res_lib_votequorum_status.header.error = error;
  2191. corosync_api->ipc_response_send(conn, &res_lib_votequorum_status, sizeof(res_lib_votequorum_status));
  2192. LEAVE();
  2193. }
  2194. static void message_handler_req_lib_votequorum_setvotes (void *conn, const void *message)
  2195. {
  2196. const struct req_lib_votequorum_setvotes *req_lib_votequorum_setvotes = message;
  2197. struct res_lib_votequorum_status res_lib_votequorum_status;
  2198. struct cluster_node *node;
  2199. unsigned int newquorum;
  2200. unsigned int total_votes;
  2201. unsigned int saved_votes;
  2202. cs_error_t error = CS_OK;
  2203. unsigned int nodeid;
  2204. ENTER();
  2205. nodeid = req_lib_votequorum_setvotes->nodeid;
  2206. node = find_node_by_nodeid(nodeid);
  2207. if (!node) {
  2208. error = CS_ERR_NAME_NOT_FOUND;
  2209. goto error_exit;
  2210. }
  2211. /*
  2212. * Check votes is valid
  2213. */
  2214. saved_votes = node->votes;
  2215. node->votes = req_lib_votequorum_setvotes->votes;
  2216. newquorum = calculate_quorum(1, 0, &total_votes);
  2217. if (newquorum < total_votes / 2 ||
  2218. newquorum > total_votes) {
  2219. node->votes = saved_votes;
  2220. error = CS_ERR_INVALID_PARAM;
  2221. goto error_exit;
  2222. }
  2223. votequorum_exec_send_reconfigure(VOTEQUORUM_RECONFIG_PARAM_NODE_VOTES, nodeid,
  2224. req_lib_votequorum_setvotes->votes);
  2225. error_exit:
  2226. res_lib_votequorum_status.header.size = sizeof(res_lib_votequorum_status);
  2227. res_lib_votequorum_status.header.id = MESSAGE_RES_VOTEQUORUM_STATUS;
  2228. res_lib_votequorum_status.header.error = error;
  2229. corosync_api->ipc_response_send(conn, &res_lib_votequorum_status, sizeof(res_lib_votequorum_status));
  2230. LEAVE();
  2231. }
  2232. static void message_handler_req_lib_votequorum_trackstart (void *conn,
  2233. const void *message)
  2234. {
  2235. const struct req_lib_votequorum_trackstart *req_lib_votequorum_trackstart = message;
  2236. struct res_lib_votequorum_status res_lib_votequorum_status;
  2237. struct quorum_pd *quorum_pd = (struct quorum_pd *)corosync_api->ipc_private_data_get (conn);
  2238. cs_error_t error = CS_OK;
  2239. ENTER();
  2240. /*
  2241. * If an immediate listing of the current cluster membership
  2242. * is requested, generate membership list
  2243. */
  2244. if (req_lib_votequorum_trackstart->track_flags & CS_TRACK_CURRENT ||
  2245. req_lib_votequorum_trackstart->track_flags & CS_TRACK_CHANGES) {
  2246. log_printf(LOGSYS_LEVEL_DEBUG, "sending initial status to %p", conn);
  2247. votequorum_exec_send_nodelist_notification(conn, req_lib_votequorum_trackstart->context);
  2248. votequorum_exec_send_quorum_notification(conn, req_lib_votequorum_trackstart->context);
  2249. }
  2250. if (quorum_pd->tracking_enabled) {
  2251. error = CS_ERR_EXIST;
  2252. goto response_send;
  2253. }
  2254. /*
  2255. * Record requests for tracking
  2256. */
  2257. if (req_lib_votequorum_trackstart->track_flags & CS_TRACK_CHANGES ||
  2258. req_lib_votequorum_trackstart->track_flags & CS_TRACK_CHANGES_ONLY) {
  2259. quorum_pd->track_flags = req_lib_votequorum_trackstart->track_flags;
  2260. quorum_pd->tracking_enabled = 1;
  2261. quorum_pd->tracking_context = req_lib_votequorum_trackstart->context;
  2262. list_add (&quorum_pd->list, &trackers_list);
  2263. }
  2264. response_send:
  2265. res_lib_votequorum_status.header.size = sizeof(res_lib_votequorum_status);
  2266. res_lib_votequorum_status.header.id = MESSAGE_RES_VOTEQUORUM_STATUS;
  2267. res_lib_votequorum_status.header.error = error;
  2268. corosync_api->ipc_response_send(conn, &res_lib_votequorum_status, sizeof(res_lib_votequorum_status));
  2269. LEAVE();
  2270. }
  2271. static void message_handler_req_lib_votequorum_trackstop (void *conn,
  2272. const void *message)
  2273. {
  2274. struct res_lib_votequorum_status res_lib_votequorum_status;
  2275. struct quorum_pd *quorum_pd = (struct quorum_pd *)corosync_api->ipc_private_data_get (conn);
  2276. int error = CS_OK;
  2277. ENTER();
  2278. if (quorum_pd->tracking_enabled) {
  2279. error = CS_OK;
  2280. quorum_pd->tracking_enabled = 0;
  2281. list_del (&quorum_pd->list);
  2282. list_init (&quorum_pd->list);
  2283. } else {
  2284. error = CS_ERR_NOT_EXIST;
  2285. }
  2286. res_lib_votequorum_status.header.size = sizeof(res_lib_votequorum_status);
  2287. res_lib_votequorum_status.header.id = MESSAGE_RES_VOTEQUORUM_STATUS;
  2288. res_lib_votequorum_status.header.error = error;
  2289. corosync_api->ipc_response_send(conn, &res_lib_votequorum_status, sizeof(res_lib_votequorum_status));
  2290. LEAVE();
  2291. }
  2292. static void message_handler_req_lib_votequorum_qdevice_register (void *conn,
  2293. const void *message)
  2294. {
  2295. const struct req_lib_votequorum_qdevice_register *req_lib_votequorum_qdevice_register = message;
  2296. struct res_lib_votequorum_status res_lib_votequorum_status;
  2297. cs_error_t error = CS_OK;
  2298. ENTER();
  2299. if (!qdevice_can_operate) {
  2300. log_printf(LOGSYS_LEVEL_INFO, "Registration of quorum device is disabled by incorrect corosync.conf. See logs for more information");
  2301. error = CS_ERR_ACCESS;
  2302. goto out;
  2303. }
  2304. if (us->flags & NODE_FLAGS_QDEVICE_REGISTERED) {
  2305. if ((!strncmp(req_lib_votequorum_qdevice_register->name,
  2306. qdevice_name, VOTEQUORUM_QDEVICE_MAX_NAME_LEN))) {
  2307. goto out;
  2308. } else {
  2309. log_printf(LOGSYS_LEVEL_WARNING,
  2310. "A new qdevice with different name (new: %s old: %s) is trying to re-register!",
  2311. req_lib_votequorum_qdevice_register->name, qdevice_name);
  2312. error = CS_ERR_EXIST;
  2313. goto out;
  2314. }
  2315. } else {
  2316. if (qdevice_reg_conn != NULL) {
  2317. log_printf(LOGSYS_LEVEL_WARNING,
  2318. "Registration request already in progress");
  2319. error = CS_ERR_TRY_AGAIN;
  2320. goto out;
  2321. }
  2322. qdevice_reg_conn = conn;
  2323. if (votequorum_exec_send_qdevice_reg(VOTEQUORUM_QDEVICE_OPERATION_REGISTER,
  2324. req_lib_votequorum_qdevice_register->name) != 0) {
  2325. log_printf(LOGSYS_LEVEL_WARNING,
  2326. "Unable to send qdevice registration request to cluster");
  2327. error = CS_ERR_TRY_AGAIN;
  2328. qdevice_reg_conn = NULL;
  2329. } else {
  2330. LEAVE();
  2331. return;
  2332. }
  2333. }
  2334. out:
  2335. res_lib_votequorum_status.header.size = sizeof(res_lib_votequorum_status);
  2336. res_lib_votequorum_status.header.id = MESSAGE_RES_VOTEQUORUM_STATUS;
  2337. res_lib_votequorum_status.header.error = error;
  2338. corosync_api->ipc_response_send(conn, &res_lib_votequorum_status, sizeof(res_lib_votequorum_status));
  2339. LEAVE();
  2340. }
  2341. static void message_handler_req_lib_votequorum_qdevice_unregister (void *conn,
  2342. const void *message)
  2343. {
  2344. const struct req_lib_votequorum_qdevice_unregister *req_lib_votequorum_qdevice_unregister = message;
  2345. struct res_lib_votequorum_status res_lib_votequorum_status;
  2346. cs_error_t error = CS_OK;
  2347. ENTER();
  2348. if (us->flags & NODE_FLAGS_QDEVICE_REGISTERED) {
  2349. if (strncmp(req_lib_votequorum_qdevice_unregister->name, qdevice_name, VOTEQUORUM_QDEVICE_MAX_NAME_LEN)) {
  2350. error = CS_ERR_INVALID_PARAM;
  2351. goto out;
  2352. }
  2353. if (qdevice_timer_set) {
  2354. corosync_api->timer_delete(qdevice_timer);
  2355. qdevice_timer_set = 0;
  2356. sync_wait_for_poll_or_timeout = 0;
  2357. }
  2358. us->flags &= ~NODE_FLAGS_QDEVICE_REGISTERED;
  2359. us->flags &= ~NODE_FLAGS_QDEVICE_ALIVE;
  2360. us->flags &= ~NODE_FLAGS_QDEVICE_CAST_VOTE;
  2361. us->flags &= ~NODE_FLAGS_QDEVICE_MASTER_WINS;
  2362. votequorum_exec_send_nodeinfo(us->node_id);
  2363. votequorum_exec_send_qdevice_reg(VOTEQUORUM_QDEVICE_OPERATION_UNREGISTER,
  2364. req_lib_votequorum_qdevice_unregister->name);
  2365. } else {
  2366. error = CS_ERR_NOT_EXIST;
  2367. }
  2368. out:
  2369. res_lib_votequorum_status.header.size = sizeof(res_lib_votequorum_status);
  2370. res_lib_votequorum_status.header.id = MESSAGE_RES_VOTEQUORUM_STATUS;
  2371. res_lib_votequorum_status.header.error = error;
  2372. corosync_api->ipc_response_send(conn, &res_lib_votequorum_status, sizeof(res_lib_votequorum_status));
  2373. LEAVE();
  2374. }
  2375. static void message_handler_req_lib_votequorum_qdevice_update (void *conn,
  2376. const void *message)
  2377. {
  2378. const struct req_lib_votequorum_qdevice_update *req_lib_votequorum_qdevice_update = message;
  2379. struct res_lib_votequorum_status res_lib_votequorum_status;
  2380. cs_error_t error = CS_OK;
  2381. ENTER();
  2382. if (us->flags & NODE_FLAGS_QDEVICE_REGISTERED) {
  2383. if (strncmp(req_lib_votequorum_qdevice_update->oldname, qdevice_name, VOTEQUORUM_QDEVICE_MAX_NAME_LEN)) {
  2384. error = CS_ERR_INVALID_PARAM;
  2385. goto out;
  2386. }
  2387. votequorum_exec_send_qdevice_reconfigure(req_lib_votequorum_qdevice_update->oldname,
  2388. req_lib_votequorum_qdevice_update->newname);
  2389. } else {
  2390. error = CS_ERR_NOT_EXIST;
  2391. }
  2392. out:
  2393. res_lib_votequorum_status.header.size = sizeof(res_lib_votequorum_status);
  2394. res_lib_votequorum_status.header.id = MESSAGE_RES_VOTEQUORUM_STATUS;
  2395. res_lib_votequorum_status.header.error = error;
  2396. corosync_api->ipc_response_send(conn, &res_lib_votequorum_status, sizeof(res_lib_votequorum_status));
  2397. LEAVE();
  2398. }
  2399. static void message_handler_req_lib_votequorum_qdevice_poll (void *conn,
  2400. const void *message)
  2401. {
  2402. const struct req_lib_votequorum_qdevice_poll *req_lib_votequorum_qdevice_poll = message;
  2403. struct res_lib_votequorum_status res_lib_votequorum_status;
  2404. cs_error_t error = CS_OK;
  2405. uint32_t oldflags;
  2406. ENTER();
  2407. if (!qdevice_can_operate) {
  2408. error = CS_ERR_ACCESS;
  2409. goto out;
  2410. }
  2411. if (us->flags & NODE_FLAGS_QDEVICE_REGISTERED) {
  2412. if (!(req_lib_votequorum_qdevice_poll->ring_id.nodeid == quorum_ringid.rep.nodeid &&
  2413. req_lib_votequorum_qdevice_poll->ring_id.seq == quorum_ringid.seq)) {
  2414. log_printf(LOGSYS_LEVEL_DEBUG, "Received poll ring id (%u.%"PRIu64") != last sync "
  2415. "ring id (%u.%"PRIu64"). Ignoring poll call.",
  2416. req_lib_votequorum_qdevice_poll->ring_id.nodeid, req_lib_votequorum_qdevice_poll->ring_id.seq,
  2417. quorum_ringid.rep.nodeid, quorum_ringid.seq);
  2418. error = CS_ERR_MESSAGE_ERROR;
  2419. goto out;
  2420. }
  2421. if (strncmp(req_lib_votequorum_qdevice_poll->name, qdevice_name, VOTEQUORUM_QDEVICE_MAX_NAME_LEN)) {
  2422. error = CS_ERR_INVALID_PARAM;
  2423. goto out;
  2424. }
  2425. if (qdevice_timer_set) {
  2426. corosync_api->timer_delete(qdevice_timer);
  2427. qdevice_timer_set = 0;
  2428. }
  2429. oldflags = us->flags;
  2430. us->flags |= NODE_FLAGS_QDEVICE_ALIVE;
  2431. if (req_lib_votequorum_qdevice_poll->cast_vote) {
  2432. us->flags |= NODE_FLAGS_QDEVICE_CAST_VOTE;
  2433. } else {
  2434. us->flags &= ~NODE_FLAGS_QDEVICE_CAST_VOTE;
  2435. }
  2436. if (us->flags != oldflags) {
  2437. votequorum_exec_send_nodeinfo(us->node_id);
  2438. }
  2439. corosync_api->timer_add_duration((unsigned long long)qdevice_timeout*1000000, qdevice,
  2440. qdevice_timer_fn, &qdevice_timer);
  2441. qdevice_timer_set = 1;
  2442. sync_wait_for_poll_or_timeout = 0;
  2443. } else {
  2444. error = CS_ERR_NOT_EXIST;
  2445. }
  2446. out:
  2447. res_lib_votequorum_status.header.size = sizeof(res_lib_votequorum_status);
  2448. res_lib_votequorum_status.header.id = MESSAGE_RES_VOTEQUORUM_STATUS;
  2449. res_lib_votequorum_status.header.error = error;
  2450. corosync_api->ipc_response_send(conn, &res_lib_votequorum_status, sizeof(res_lib_votequorum_status));
  2451. LEAVE();
  2452. }
  2453. static void message_handler_req_lib_votequorum_qdevice_master_wins (void *conn,
  2454. const void *message)
  2455. {
  2456. const struct req_lib_votequorum_qdevice_master_wins *req_lib_votequorum_qdevice_master_wins = message;
  2457. struct res_lib_votequorum_status res_lib_votequorum_status;
  2458. cs_error_t error = CS_OK;
  2459. uint32_t oldflags = us->flags;
  2460. ENTER();
  2461. if (!qdevice_can_operate) {
  2462. error = CS_ERR_ACCESS;
  2463. goto out;
  2464. }
  2465. if (us->flags & NODE_FLAGS_QDEVICE_REGISTERED) {
  2466. if (strncmp(req_lib_votequorum_qdevice_master_wins->name, qdevice_name, VOTEQUORUM_QDEVICE_MAX_NAME_LEN)) {
  2467. error = CS_ERR_INVALID_PARAM;
  2468. goto out;
  2469. }
  2470. if (req_lib_votequorum_qdevice_master_wins->allow) {
  2471. us->flags |= NODE_FLAGS_QDEVICE_MASTER_WINS;
  2472. } else {
  2473. us->flags &= ~NODE_FLAGS_QDEVICE_MASTER_WINS;
  2474. }
  2475. if (us->flags != oldflags) {
  2476. votequorum_exec_send_nodeinfo(us->node_id);
  2477. }
  2478. update_qdevice_master_wins(req_lib_votequorum_qdevice_master_wins->allow);
  2479. } else {
  2480. error = CS_ERR_NOT_EXIST;
  2481. }
  2482. out:
  2483. res_lib_votequorum_status.header.size = sizeof(res_lib_votequorum_status);
  2484. res_lib_votequorum_status.header.id = MESSAGE_RES_VOTEQUORUM_STATUS;
  2485. res_lib_votequorum_status.header.error = error;
  2486. corosync_api->ipc_response_send(conn, &res_lib_votequorum_status, sizeof(res_lib_votequorum_status));
  2487. LEAVE();
  2488. }