votequorum.c 85 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888889890891892893894895896897898899900901902903904905906907908909910911912913914915916917918919920921922923924925926927928929930931932933934935936937938939940941942943944945946947948949950951952953954955956957958959960961962963964965966967968969970971972973974975976977978979980981982983984985986987988989990991992993994995996997998999100010011002100310041005100610071008100910101011101210131014101510161017101810191020102110221023102410251026102710281029103010311032103310341035103610371038103910401041104210431044104510461047104810491050105110521053105410551056105710581059106010611062106310641065106610671068106910701071107210731074107510761077107810791080108110821083108410851086108710881089109010911092109310941095109610971098109911001101110211031104110511061107110811091110111111121113111411151116111711181119112011211122112311241125112611271128112911301131113211331134113511361137113811391140114111421143114411451146114711481149115011511152115311541155115611571158115911601161116211631164116511661167116811691170117111721173117411751176117711781179118011811182118311841185118611871188118911901191119211931194119511961197119811991200120112021203120412051206120712081209121012111212121312141215121612171218121912201221122212231224122512261227122812291230123112321233123412351236123712381239124012411242124312441245124612471248124912501251125212531254125512561257125812591260126112621263126412651266126712681269127012711272127312741275127612771278127912801281128212831284128512861287128812891290129112921293129412951296129712981299130013011302130313041305130613071308130913101311131213131314131513161317131813191320132113221323132413251326132713281329133013311332133313341335133613371338133913401341134213431344134513461347134813491350135113521353135413551356135713581359136013611362136313641365136613671368136913701371137213731374137513761377137813791380138113821383138413851386138713881389139013911392139313941395139613971398139914001401140214031404140514061407140814091410141114121413141414151416141714181419142014211422142314241425142614271428142914301431143214331434143514361437143814391440144114421443144414451446144714481449145014511452145314541455145614571458145914601461146214631464146514661467146814691470147114721473147414751476147714781479148014811482148314841485148614871488148914901491149214931494149514961497149814991500150115021503150415051506150715081509151015111512151315141515151615171518151915201521152215231524152515261527152815291530153115321533153415351536153715381539154015411542154315441545154615471548154915501551155215531554155515561557155815591560156115621563156415651566156715681569157015711572157315741575157615771578157915801581158215831584158515861587158815891590159115921593159415951596159715981599160016011602160316041605160616071608160916101611161216131614161516161617161816191620162116221623162416251626162716281629163016311632163316341635163616371638163916401641164216431644164516461647164816491650165116521653165416551656165716581659166016611662166316641665166616671668166916701671167216731674167516761677167816791680168116821683168416851686168716881689169016911692169316941695169616971698169917001701170217031704170517061707170817091710171117121713171417151716171717181719172017211722172317241725172617271728172917301731173217331734173517361737173817391740174117421743174417451746174717481749175017511752175317541755175617571758175917601761176217631764176517661767176817691770177117721773177417751776177717781779178017811782178317841785178617871788178917901791179217931794179517961797179817991800180118021803180418051806180718081809181018111812181318141815181618171818181918201821182218231824182518261827182818291830183118321833183418351836183718381839184018411842184318441845184618471848184918501851185218531854185518561857185818591860186118621863186418651866186718681869187018711872187318741875187618771878187918801881188218831884188518861887188818891890189118921893189418951896189718981899190019011902190319041905190619071908190919101911191219131914191519161917191819191920192119221923192419251926192719281929193019311932193319341935193619371938193919401941194219431944194519461947194819491950195119521953195419551956195719581959196019611962196319641965196619671968196919701971197219731974197519761977197819791980198119821983198419851986198719881989199019911992199319941995199619971998199920002001200220032004200520062007200820092010201120122013201420152016201720182019202020212022202320242025202620272028202920302031203220332034203520362037203820392040204120422043204420452046204720482049205020512052205320542055205620572058205920602061206220632064206520662067206820692070207120722073207420752076207720782079208020812082208320842085208620872088208920902091209220932094209520962097209820992100210121022103210421052106210721082109211021112112211321142115211621172118211921202121212221232124212521262127212821292130213121322133213421352136213721382139214021412142214321442145214621472148214921502151215221532154215521562157215821592160216121622163216421652166216721682169217021712172217321742175217621772178217921802181218221832184218521862187218821892190219121922193219421952196219721982199220022012202220322042205220622072208220922102211221222132214221522162217221822192220222122222223222422252226222722282229223022312232223322342235223622372238223922402241224222432244224522462247224822492250225122522253225422552256225722582259226022612262226322642265226622672268226922702271227222732274227522762277227822792280228122822283228422852286228722882289229022912292229322942295229622972298229923002301230223032304230523062307230823092310231123122313231423152316231723182319232023212322232323242325232623272328232923302331233223332334233523362337233823392340234123422343234423452346234723482349235023512352235323542355235623572358235923602361236223632364236523662367236823692370237123722373237423752376237723782379238023812382238323842385238623872388238923902391239223932394239523962397239823992400240124022403240424052406240724082409241024112412241324142415241624172418241924202421242224232424242524262427242824292430243124322433243424352436243724382439244024412442244324442445244624472448244924502451245224532454245524562457245824592460246124622463246424652466246724682469247024712472247324742475247624772478247924802481248224832484248524862487248824892490249124922493249424952496249724982499250025012502250325042505250625072508250925102511251225132514251525162517251825192520252125222523252425252526252725282529253025312532253325342535253625372538253925402541254225432544254525462547254825492550255125522553255425552556255725582559256025612562256325642565256625672568256925702571257225732574257525762577257825792580258125822583258425852586258725882589259025912592259325942595259625972598259926002601260226032604260526062607260826092610261126122613261426152616261726182619262026212622262326242625262626272628262926302631263226332634263526362637263826392640264126422643264426452646264726482649265026512652265326542655265626572658265926602661266226632664266526662667266826692670267126722673267426752676267726782679268026812682268326842685268626872688268926902691269226932694269526962697269826992700270127022703270427052706270727082709271027112712271327142715271627172718271927202721272227232724272527262727272827292730273127322733273427352736273727382739274027412742274327442745274627472748274927502751275227532754275527562757275827592760276127622763276427652766276727682769277027712772277327742775277627772778277927802781278227832784278527862787278827892790279127922793279427952796279727982799280028012802280328042805280628072808280928102811281228132814281528162817281828192820282128222823282428252826282728282829283028312832283328342835283628372838283928402841284228432844284528462847284828492850285128522853285428552856285728582859286028612862286328642865286628672868286928702871287228732874287528762877287828792880288128822883288428852886288728882889289028912892289328942895289628972898289929002901290229032904290529062907290829092910291129122913291429152916291729182919292029212922292329242925292629272928292929302931293229332934293529362937293829392940294129422943294429452946294729482949295029512952295329542955295629572958295929602961296229632964296529662967296829692970297129722973297429752976297729782979298029812982298329842985298629872988298929902991299229932994299529962997299829993000300130023003300430053006300730083009301030113012301330143015301630173018301930203021302230233024302530263027302830293030303130323033303430353036303730383039304030413042304330443045304630473048
  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/qblist.h>
  42. #include <qb/qbipc_common.h>
  43. #include "quorum.h"
  44. #include <corosync/corodefs.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, initial_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 qb_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 qb_list_head cluster_members_list;
  167. static unsigned int quorum_members[PROCESSOR_COUNT_MAX];
  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 qb_list_head list;
  187. void *conn;
  188. };
  189. static struct qb_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. static void node_add_ordered(struct cluster_node *newnode)
  359. {
  360. struct cluster_node *node = NULL;
  361. struct qb_list_head *tmp;
  362. ENTER();
  363. qb_list_for_each(tmp, &cluster_members_list) {
  364. node = qb_list_entry(tmp, struct cluster_node, list);
  365. if (newnode->node_id < node->node_id) {
  366. break;
  367. }
  368. }
  369. if (!node) {
  370. qb_list_add(&newnode->list, &cluster_members_list);
  371. } else {
  372. qb_list_add_tail(&newnode->list, &node->list);
  373. }
  374. LEAVE();
  375. }
  376. static struct cluster_node *allocate_node(unsigned int nodeid)
  377. {
  378. struct cluster_node *cl = NULL;
  379. struct qb_list_head *tmp;
  380. ENTER();
  381. if (cluster_nodes_entries <= PROCESSOR_COUNT_MAX + 1) {
  382. cl = (struct cluster_node *)&cluster_nodes[cluster_nodes_entries];
  383. cluster_nodes_entries++;
  384. } else {
  385. qb_list_for_each(tmp, &cluster_members_list) {
  386. cl = qb_list_entry(tmp, struct cluster_node, list);
  387. if (cl->state == NODESTATE_DEAD) {
  388. break;
  389. }
  390. }
  391. /*
  392. * this should never happen
  393. */
  394. if (!cl) {
  395. log_printf(LOGSYS_LEVEL_CRIT, "Unable to find memory for node " CS_PRI_NODE_ID " data!!", nodeid);
  396. goto out;
  397. }
  398. qb_list_del(tmp);
  399. }
  400. memset(cl, 0, sizeof(struct cluster_node));
  401. cl->node_id = nodeid;
  402. if (nodeid != VOTEQUORUM_QDEVICE_NODEID) {
  403. node_add_ordered(cl);
  404. }
  405. out:
  406. LEAVE();
  407. return cl;
  408. }
  409. static struct cluster_node *find_node_by_nodeid(unsigned int nodeid)
  410. {
  411. struct cluster_node *node;
  412. struct qb_list_head *tmp;
  413. ENTER();
  414. if (nodeid == us->node_id) {
  415. LEAVE();
  416. return us;
  417. }
  418. if (nodeid == VOTEQUORUM_QDEVICE_NODEID) {
  419. LEAVE();
  420. return qdevice;
  421. }
  422. qb_list_for_each(tmp, &cluster_members_list) {
  423. node = qb_list_entry(tmp, struct cluster_node, list);
  424. if (node->node_id == nodeid) {
  425. LEAVE();
  426. return node;
  427. }
  428. }
  429. LEAVE();
  430. return NULL;
  431. }
  432. static void get_lowest_node_id(void)
  433. {
  434. struct cluster_node *node = NULL;
  435. struct qb_list_head *tmp;
  436. ENTER();
  437. lowest_node_id = us->node_id;
  438. qb_list_for_each(tmp, &cluster_members_list) {
  439. node = qb_list_entry(tmp, struct cluster_node, list);
  440. if ((node->state == NODESTATE_MEMBER) &&
  441. (node->node_id < lowest_node_id)) {
  442. lowest_node_id = node->node_id;
  443. }
  444. }
  445. log_printf(LOGSYS_LEVEL_DEBUG, "lowest node id: " CS_PRI_NODE_ID " us: " CS_PRI_NODE_ID, lowest_node_id, us->node_id);
  446. icmap_set_uint32("runtime.votequorum.lowest_node_id", lowest_node_id);
  447. LEAVE();
  448. }
  449. static void get_highest_node_id(void)
  450. {
  451. struct cluster_node *node = NULL;
  452. struct qb_list_head *tmp;
  453. ENTER();
  454. highest_node_id = us->node_id;
  455. qb_list_for_each(tmp, &cluster_members_list) {
  456. node = qb_list_entry(tmp, struct cluster_node, list);
  457. if ((node->state == NODESTATE_MEMBER) &&
  458. (node->node_id > highest_node_id)) {
  459. highest_node_id = node->node_id;
  460. }
  461. }
  462. log_printf(LOGSYS_LEVEL_DEBUG, "highest node id: " CS_PRI_NODE_ID " us: " CS_PRI_NODE_ID, highest_node_id, us->node_id);
  463. icmap_set_uint32("runtime.votequorum.highest_node_id", highest_node_id);
  464. LEAVE();
  465. }
  466. static int check_low_node_id_partition(void)
  467. {
  468. struct cluster_node *node = NULL;
  469. struct qb_list_head *tmp;
  470. int found = 0;
  471. ENTER();
  472. qb_list_for_each(tmp, &cluster_members_list) {
  473. node = qb_list_entry(tmp, struct cluster_node, list);
  474. if ((node->state == NODESTATE_MEMBER) &&
  475. (node->node_id == lowest_node_id)) {
  476. found = 1;
  477. }
  478. }
  479. LEAVE();
  480. return found;
  481. }
  482. static int check_high_node_id_partition(void)
  483. {
  484. struct cluster_node *node = NULL;
  485. struct qb_list_head *tmp;
  486. int found = 0;
  487. ENTER();
  488. qb_list_for_each(tmp, &cluster_members_list) {
  489. node = qb_list_entry(tmp, struct cluster_node, list);
  490. if ((node->state == NODESTATE_MEMBER) &&
  491. (node->node_id == highest_node_id)) {
  492. found = 1;
  493. }
  494. }
  495. LEAVE();
  496. return found;
  497. }
  498. static int is_in_nodelist(int nodeid, unsigned int *members, int entries)
  499. {
  500. int i;
  501. ENTER();
  502. for (i=0; i<entries; i++) {
  503. if (nodeid == members[i]) {
  504. LEAVE();
  505. return 1;
  506. }
  507. }
  508. LEAVE();
  509. return 0;
  510. }
  511. /*
  512. * The algorithm for a list of tie-breaker nodes is:
  513. * travel the list of nodes in the auto_tie_breaker list,
  514. * if the node IS in our current partition, check if the
  515. * nodes earlier in the atb list are in the 'previous' partition;
  516. * If none are found then we are safe to be quorate, if any are
  517. * then we cannot be as we don't know if that node is up or down.
  518. * If we don't have a node in the current list we are NOT quorate.
  519. * Obviously if we find the first node in the atb list in our
  520. * partition then we are quorate.
  521. *
  522. * Special cases lowest nodeid, and highest nodeid are handled separately.
  523. */
  524. static int check_auto_tie_breaker(void)
  525. {
  526. int i, j;
  527. int res;
  528. ENTER();
  529. if (auto_tie_breaker == ATB_LOWEST) {
  530. res = check_low_node_id_partition();
  531. log_printf(LOGSYS_LEVEL_DEBUG, "ATB_LOWEST decision: %d", res);
  532. LEAVE();
  533. return res;
  534. }
  535. if (auto_tie_breaker == ATB_HIGHEST) {
  536. res = check_high_node_id_partition();
  537. log_printf(LOGSYS_LEVEL_DEBUG, "ATB_HIGHEST decision: %d", res);
  538. LEAVE();
  539. return res;
  540. }
  541. /* Assume ATB_LIST, we should never be called for ATB_NONE */
  542. for (i=0; i < atb_nodelist_entries; i++) {
  543. if (is_in_nodelist(atb_nodelist[i], quorum_members, quorum_members_entries)) {
  544. /*
  545. * Node is in our partition, if any of its predecessors are
  546. * in the previous quorum partition then it might be in the
  547. * 'other half' (as we've got this far without seeing it here)
  548. * and so we can't be quorate.
  549. */
  550. for (j=0; j<i; j++) {
  551. if (is_in_nodelist(atb_nodelist[j], previous_quorum_members, previous_quorum_members_entries)) {
  552. log_printf(LOGSYS_LEVEL_DEBUG, "ATB_LIST found node " CS_PRI_NODE_ID " in previous partition but not here, quorum denied", atb_nodelist[j]);
  553. LEAVE();
  554. return 0;
  555. }
  556. }
  557. /*
  558. * None of the other list nodes were in the previous partition, if there
  559. * are enough votes, we can be quorate
  560. */
  561. log_printf(LOGSYS_LEVEL_DEBUG, "ATB_LIST found node " CS_PRI_NODE_ID " in current partition, we can be quorate", atb_nodelist[i]);
  562. LEAVE();
  563. return 1;
  564. }
  565. }
  566. log_printf(LOGSYS_LEVEL_DEBUG, "ATB_LIST found no list nodes in current partition, we cannot be quorate");
  567. LEAVE();
  568. return 0;
  569. }
  570. /*
  571. * atb_string can be either:
  572. * 'lowest'
  573. * 'highest'
  574. * a list of nodeids
  575. */
  576. static void parse_atb_string(char *atb_string)
  577. {
  578. char *ptr;
  579. long num;
  580. ENTER();
  581. auto_tie_breaker = ATB_NONE;
  582. if (!strcmp(atb_string, "lowest"))
  583. auto_tie_breaker = ATB_LOWEST;
  584. if (!strcmp(atb_string, "highest"))
  585. auto_tie_breaker = ATB_HIGHEST;
  586. if (atoi(atb_string)) {
  587. atb_nodelist_entries = 0;
  588. ptr = atb_string;
  589. do {
  590. num = strtol(ptr, &ptr, 10);
  591. if (num) {
  592. log_printf(LOGSYS_LEVEL_DEBUG, "ATB nodelist[%d] = %d", atb_nodelist_entries, num);
  593. atb_nodelist[atb_nodelist_entries++] = num;
  594. }
  595. } while (num);
  596. if (atb_nodelist_entries) {
  597. auto_tie_breaker = ATB_LIST;
  598. }
  599. }
  600. icmap_set_uint32("runtime.votequorum.atb_type", auto_tie_breaker);
  601. log_printf(LOGSYS_LEVEL_DEBUG, "ATB type = %d", auto_tie_breaker);
  602. /* Make sure we got something */
  603. if (auto_tie_breaker == ATB_NONE) {
  604. 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");
  605. auto_tie_breaker = ATB_NONE;
  606. }
  607. LEAVE();
  608. }
  609. static int check_qdevice_master(void)
  610. {
  611. struct cluster_node *node = NULL;
  612. struct qb_list_head *tmp;
  613. int found = 0;
  614. ENTER();
  615. qb_list_for_each(tmp, &cluster_members_list) {
  616. node = qb_list_entry(tmp, struct cluster_node, list);
  617. if ((node->state == NODESTATE_MEMBER) &&
  618. (node->flags & NODE_FLAGS_QDEVICE_MASTER_WINS) &&
  619. (node->flags & NODE_FLAGS_QDEVICE_CAST_VOTE)) {
  620. found = 1;
  621. }
  622. }
  623. LEAVE();
  624. return found;
  625. }
  626. static void decode_flags(uint32_t flags)
  627. {
  628. ENTER();
  629. log_printf(LOGSYS_LEVEL_DEBUG,
  630. "flags: quorate: %s Leaving: %s WFA Status: %s First: %s Qdevice: %s QdeviceAlive: %s QdeviceCastVote: %s QdeviceMasterWins: %s",
  631. (flags & NODE_FLAGS_QUORATE)?"Yes":"No",
  632. (flags & NODE_FLAGS_LEAVING)?"Yes":"No",
  633. (flags & NODE_FLAGS_WFASTATUS)?"Yes":"No",
  634. (flags & NODE_FLAGS_FIRST)?"Yes":"No",
  635. (flags & NODE_FLAGS_QDEVICE_REGISTERED)?"Yes":"No",
  636. (flags & NODE_FLAGS_QDEVICE_ALIVE)?"Yes":"No",
  637. (flags & NODE_FLAGS_QDEVICE_CAST_VOTE)?"Yes":"No",
  638. (flags & NODE_FLAGS_QDEVICE_MASTER_WINS)?"Yes":"No");
  639. LEAVE();
  640. }
  641. /*
  642. * load/save are copied almost pristine from totemsrp,c
  643. */
  644. static int load_ev_tracking_barrier(void)
  645. {
  646. int res = 0;
  647. char filename[PATH_MAX];
  648. ENTER();
  649. snprintf(filename, sizeof(filename) - 1, "%s/ev_tracking", get_state_dir());
  650. ev_tracking_fd = open(filename, O_RDWR, 0700);
  651. if (ev_tracking_fd != -1) {
  652. res = read (ev_tracking_fd, &ev_tracking_barrier, sizeof(uint32_t));
  653. close(ev_tracking_fd);
  654. if (res == sizeof (uint32_t)) {
  655. LEAVE();
  656. return 0;
  657. }
  658. }
  659. ev_tracking_barrier = 0;
  660. umask(0);
  661. ev_tracking_fd = open (filename, O_CREAT|O_RDWR, 0700);
  662. if (ev_tracking_fd != -1) {
  663. res = write (ev_tracking_fd, &ev_tracking_barrier, sizeof (uint32_t));
  664. if ((res == -1) || (res != sizeof (uint32_t))) {
  665. log_printf(LOGSYS_LEVEL_WARNING,
  666. "Unable to write to %s", filename);
  667. }
  668. close(ev_tracking_fd);
  669. LEAVE();
  670. return 0;
  671. }
  672. log_printf(LOGSYS_LEVEL_WARNING,
  673. "Unable to create %s file", filename);
  674. LEAVE();
  675. return -1;
  676. }
  677. static void update_wait_for_all_status(uint8_t wfa_status)
  678. {
  679. ENTER();
  680. wait_for_all_status = wfa_status;
  681. if (wait_for_all_status) {
  682. us->flags |= NODE_FLAGS_WFASTATUS;
  683. } else {
  684. us->flags &= ~NODE_FLAGS_WFASTATUS;
  685. }
  686. icmap_set_uint8("runtime.votequorum.wait_for_all_status",
  687. wait_for_all_status);
  688. LEAVE();
  689. }
  690. static void update_two_node(void)
  691. {
  692. ENTER();
  693. icmap_set_uint8("runtime.votequorum.two_node", two_node);
  694. LEAVE();
  695. }
  696. static void update_ev_barrier(uint32_t expected_votes)
  697. {
  698. ENTER();
  699. ev_barrier = expected_votes;
  700. icmap_set_uint32("runtime.votequorum.ev_barrier", ev_barrier);
  701. LEAVE();
  702. }
  703. static void update_qdevice_can_operate(uint8_t status)
  704. {
  705. ENTER();
  706. qdevice_can_operate = status;
  707. icmap_set_uint8("runtime.votequorum.qdevice_can_operate", qdevice_can_operate);
  708. LEAVE();
  709. }
  710. static void update_qdevice_master_wins(uint8_t allow)
  711. {
  712. ENTER();
  713. qdevice_master_wins = allow;
  714. icmap_set_uint8("runtime.votequorum.qdevice_master_wins", qdevice_master_wins);
  715. LEAVE();
  716. }
  717. static void update_ev_tracking_barrier(uint32_t ev_t_barrier)
  718. {
  719. int res;
  720. ENTER();
  721. ev_tracking_barrier = ev_t_barrier;
  722. icmap_set_uint32("runtime.votequorum.ev_tracking_barrier", ev_tracking_barrier);
  723. if (lseek (ev_tracking_fd, 0, SEEK_SET) != 0) {
  724. log_printf(LOGSYS_LEVEL_WARNING,
  725. "Unable to update ev_tracking_barrier on disk data!!!");
  726. LEAVE();
  727. return;
  728. }
  729. res = write (ev_tracking_fd, &ev_tracking_barrier, sizeof (uint32_t));
  730. if (res != sizeof (uint32_t)) {
  731. log_printf(LOGSYS_LEVEL_WARNING,
  732. "Unable to update ev_tracking_barrier on disk data!!!");
  733. }
  734. #ifdef HAVE_FDATASYNC
  735. fdatasync(ev_tracking_fd);
  736. #else
  737. fsync(ev_tracking_fd);
  738. #endif
  739. LEAVE();
  740. }
  741. /*
  742. * quorum calculation core bits
  743. */
  744. static int calculate_quorum(int allow_decrease, unsigned int max_expected, unsigned int *ret_total_votes)
  745. {
  746. struct qb_list_head *nodelist;
  747. struct cluster_node *node;
  748. unsigned int total_votes = 0;
  749. unsigned int highest_expected = 0;
  750. unsigned int newquorum, q1, q2;
  751. unsigned int total_nodes = 0;
  752. ENTER();
  753. if ((allow_downscale) && (allow_decrease) && (max_expected)) {
  754. max_expected = max(ev_barrier, max_expected);
  755. }
  756. qb_list_for_each(nodelist, &cluster_members_list) {
  757. node = qb_list_entry(nodelist, struct cluster_node, list);
  758. log_printf(LOGSYS_LEVEL_DEBUG, "node " CS_PRI_NODE_ID " state=%d, votes=%u, expected=%u",
  759. node->node_id, node->state, node->votes, node->expected_votes);
  760. if (node->state == NODESTATE_MEMBER) {
  761. highest_expected = max(highest_expected, node->expected_votes);
  762. total_votes += node->votes;
  763. total_nodes++;
  764. }
  765. }
  766. if (us->flags & NODE_FLAGS_QDEVICE_CAST_VOTE) {
  767. log_printf(LOGSYS_LEVEL_DEBUG, "node 0 state=1, votes=%u", qdevice->votes);
  768. total_votes += qdevice->votes;
  769. total_nodes++;
  770. }
  771. if (max_expected > 0) {
  772. highest_expected = max_expected;
  773. }
  774. /*
  775. * This quorum calculation is taken from the OpenVMS Cluster Systems
  776. * manual, but, then, you guessed that didn't you
  777. */
  778. q1 = (highest_expected + 2) / 2;
  779. q2 = (total_votes + 2) / 2;
  780. newquorum = max(q1, q2);
  781. /*
  782. * Normally quorum never decreases but the system administrator can
  783. * force it down by setting expected votes to a maximum value
  784. */
  785. if (!allow_decrease) {
  786. newquorum = max(quorum, newquorum);
  787. }
  788. /*
  789. * The special two_node mode allows each of the two nodes to retain
  790. * quorum if the other fails. Only one of the two should live past
  791. * fencing (as both nodes try to fence each other in split-brain.)
  792. * Also: if there are more than two nodes, force us inquorate to avoid
  793. * any damage or confusion.
  794. */
  795. if (two_node && total_nodes <= 2) {
  796. newquorum = 1;
  797. }
  798. if (ret_total_votes) {
  799. *ret_total_votes = total_votes;
  800. }
  801. LEAVE();
  802. return newquorum;
  803. }
  804. static void update_node_expected_votes(int new_expected_votes)
  805. {
  806. struct qb_list_head *nodelist;
  807. struct cluster_node *node;
  808. if (new_expected_votes) {
  809. qb_list_for_each(nodelist, &cluster_members_list) {
  810. node = qb_list_entry(nodelist, struct cluster_node, list);
  811. if (node->state == NODESTATE_MEMBER) {
  812. node->expected_votes = new_expected_votes;
  813. }
  814. }
  815. }
  816. }
  817. static void are_we_quorate(unsigned int total_votes)
  818. {
  819. int quorate;
  820. int quorum_change = 0;
  821. ENTER();
  822. /*
  823. * wait for all nodes to show up before granting quorum
  824. */
  825. if ((wait_for_all) && (wait_for_all_status)) {
  826. if (total_votes != us->expected_votes) {
  827. log_printf(LOGSYS_LEVEL_NOTICE,
  828. "Waiting for all cluster members. "
  829. "Current votes: %d expected_votes: %d",
  830. total_votes, us->expected_votes);
  831. cluster_is_quorate = 0;
  832. return;
  833. }
  834. update_wait_for_all_status(0);
  835. }
  836. if (quorum > total_votes) {
  837. quorate = 0;
  838. } else {
  839. quorate = 1;
  840. get_lowest_node_id();
  841. get_highest_node_id();
  842. }
  843. if ((auto_tie_breaker != ATB_NONE) &&
  844. /* Must be a half (or half-1) split */
  845. (total_votes == (us->expected_votes / 2)) &&
  846. /* If the 'other' partition in a split might have quorum then we can't run ATB */
  847. (previous_quorum_members_entries - quorum_members_entries < quorum) &&
  848. (check_auto_tie_breaker() == 1)) {
  849. quorate = 1;
  850. }
  851. if ((qdevice_master_wins) &&
  852. (!quorate) &&
  853. (check_qdevice_master() == 1)) {
  854. log_printf(LOGSYS_LEVEL_DEBUG, "node is quorate as part of master_wins partition");
  855. quorate = 1;
  856. }
  857. if (cluster_is_quorate && !quorate) {
  858. quorum_change = 1;
  859. log_printf(LOGSYS_LEVEL_DEBUG, "quorum lost, blocking activity");
  860. }
  861. if (!cluster_is_quorate && quorate) {
  862. quorum_change = 1;
  863. log_printf(LOGSYS_LEVEL_DEBUG, "quorum regained, resuming activity");
  864. }
  865. cluster_is_quorate = quorate;
  866. if (cluster_is_quorate) {
  867. us->flags |= NODE_FLAGS_QUORATE;
  868. } else {
  869. us->flags &= ~NODE_FLAGS_QUORATE;
  870. }
  871. if (wait_for_all) {
  872. if (quorate) {
  873. update_wait_for_all_status(0);
  874. } else {
  875. update_wait_for_all_status(1);
  876. }
  877. }
  878. if ((quorum_change) &&
  879. (sync_in_progress == 0)) {
  880. quorum_callback(quorum_members, quorum_members_entries,
  881. cluster_is_quorate, &quorum_ringid);
  882. votequorum_exec_send_quorum_notification(NULL, 0L);
  883. }
  884. LEAVE();
  885. }
  886. static void get_total_votes(unsigned int *totalvotes, unsigned int *current_members)
  887. {
  888. unsigned int total_votes = 0;
  889. unsigned int cluster_members = 0;
  890. struct qb_list_head *nodelist;
  891. struct cluster_node *node;
  892. ENTER();
  893. qb_list_for_each(nodelist, &cluster_members_list) {
  894. node = qb_list_entry(nodelist, struct cluster_node, list);
  895. if (node->state == NODESTATE_MEMBER) {
  896. cluster_members++;
  897. total_votes += node->votes;
  898. }
  899. }
  900. if (qdevice->votes) {
  901. total_votes += qdevice->votes;
  902. cluster_members++;
  903. }
  904. *totalvotes = total_votes;
  905. *current_members = cluster_members;
  906. LEAVE();
  907. }
  908. /*
  909. * Recalculate cluster quorum, set quorate and notify changes
  910. */
  911. static void recalculate_quorum(int allow_decrease, int by_current_nodes)
  912. {
  913. unsigned int total_votes = 0;
  914. unsigned int cluster_members = 0;
  915. ENTER();
  916. get_total_votes(&total_votes, &cluster_members);
  917. if (!by_current_nodes) {
  918. cluster_members = 0;
  919. }
  920. /*
  921. * Keep expected_votes at the highest number of votes in the cluster
  922. */
  923. log_printf(LOGSYS_LEVEL_DEBUG, "total_votes=%d, expected_votes=%d", total_votes, us->expected_votes);
  924. if (total_votes > us->expected_votes) {
  925. us->expected_votes = total_votes;
  926. votequorum_exec_send_expectedvotes_notification();
  927. }
  928. if ((ev_tracking) &&
  929. (us->expected_votes > ev_tracking_barrier)) {
  930. update_ev_tracking_barrier(us->expected_votes);
  931. }
  932. quorum = calculate_quorum(allow_decrease, cluster_members, &total_votes);
  933. update_node_expected_votes(cluster_members);
  934. are_we_quorate(total_votes);
  935. LEAVE();
  936. }
  937. /*
  938. * configuration bits and pieces
  939. */
  940. static int votequorum_read_nodelist_configuration(uint32_t *votes,
  941. uint32_t *nodes,
  942. uint32_t *expected_votes)
  943. {
  944. icmap_iter_t iter;
  945. const char *iter_key;
  946. char tmp_key[ICMAP_KEYNAME_MAXLEN];
  947. uint32_t our_pos, node_pos, last_node_pos=-1;
  948. uint32_t nodecount = 0;
  949. uint32_t nodelist_expected_votes = 0;
  950. uint32_t node_votes = 0;
  951. int res = 0;
  952. ENTER();
  953. if (icmap_get_uint32("nodelist.local_node_pos", &our_pos) != CS_OK) {
  954. log_printf(LOGSYS_LEVEL_DEBUG,
  955. "No nodelist defined or our node is not in the nodelist");
  956. return 0;
  957. }
  958. iter = icmap_iter_init("nodelist.node.");
  959. while ((iter_key = icmap_iter_next(iter, NULL, NULL)) != NULL) {
  960. res = sscanf(iter_key, "nodelist.node.%u.%s", &node_pos, tmp_key);
  961. if (res != 2) {
  962. continue;
  963. }
  964. /*
  965. * If current node_pos is the same as the last_node_pos then skip it
  966. * so we only do the code below once per node.
  967. * (icmap keys are always in order)
  968. */
  969. if (last_node_pos == node_pos) {
  970. continue;
  971. }
  972. last_node_pos = node_pos;
  973. nodecount++;
  974. snprintf(tmp_key, ICMAP_KEYNAME_MAXLEN, "nodelist.node.%u.quorum_votes", node_pos);
  975. if (icmap_get_uint32(tmp_key, &node_votes) != CS_OK) {
  976. node_votes = 1;
  977. }
  978. nodelist_expected_votes = nodelist_expected_votes + node_votes;
  979. if (node_pos == our_pos) {
  980. *votes = node_votes;
  981. }
  982. }
  983. *expected_votes = nodelist_expected_votes;
  984. *nodes = nodecount;
  985. icmap_iter_finalize(iter);
  986. LEAVE();
  987. return 1;
  988. }
  989. static int votequorum_qdevice_is_configured(uint32_t *qdevice_votes)
  990. {
  991. char *qdevice_model = NULL;
  992. int ret = 0;
  993. ENTER();
  994. if (icmap_get_string("quorum.device.model", &qdevice_model) == CS_OK) {
  995. if (strlen(qdevice_model)) {
  996. if (icmap_get_uint32("quorum.device.votes", qdevice_votes) != CS_OK) {
  997. *qdevice_votes = -1;
  998. }
  999. if (icmap_get_uint32("quorum.device.timeout", &qdevice_timeout) != CS_OK) {
  1000. qdevice_timeout = VOTEQUORUM_QDEVICE_DEFAULT_TIMEOUT;
  1001. }
  1002. if (icmap_get_uint32("quorum.device.sync_timeout", &qdevice_sync_timeout) != CS_OK) {
  1003. qdevice_sync_timeout = VOTEQUORUM_QDEVICE_DEFAULT_SYNC_TIMEOUT;
  1004. }
  1005. update_qdevice_can_operate(1);
  1006. ret = 1;
  1007. }
  1008. free(qdevice_model);
  1009. }
  1010. LEAVE();
  1011. return ret;
  1012. }
  1013. #define VOTEQUORUM_READCONFIG_STARTUP 0
  1014. #define VOTEQUORUM_READCONFIG_RUNTIME 1
  1015. static char *votequorum_readconfig(int runtime)
  1016. {
  1017. uint32_t node_votes = 0, qdevice_votes = 0;
  1018. uint32_t node_expected_votes = 0, expected_votes = 0;
  1019. uint32_t node_count = 0;
  1020. uint8_t atb = 0;
  1021. int have_nodelist, have_qdevice;
  1022. char *atb_string = NULL;
  1023. char *error = NULL;
  1024. ENTER();
  1025. log_printf(LOGSYS_LEVEL_DEBUG, "Reading configuration (runtime: %d)", runtime);
  1026. /*
  1027. * Set the few things we re-read at runtime back to their defaults
  1028. */
  1029. if (runtime) {
  1030. two_node = 0;
  1031. expected_votes = 0;
  1032. /* auto_tie_breaker cannot be changed by config reload, but
  1033. * we automatically disable it on odd-sized clusters without
  1034. * wait_for_all.
  1035. * We may need to re-enable it when membership changes to ensure
  1036. * that auto_tie_breaker is consistent across all nodes */
  1037. auto_tie_breaker = initial_auto_tie_breaker;
  1038. icmap_set_uint32("runtime.votequorum.atb_type", auto_tie_breaker);
  1039. }
  1040. /*
  1041. * gather basic data here
  1042. */
  1043. icmap_get_uint32("quorum.expected_votes", &expected_votes);
  1044. have_nodelist = votequorum_read_nodelist_configuration(&node_votes, &node_count, &node_expected_votes);
  1045. have_qdevice = votequorum_qdevice_is_configured(&qdevice_votes);
  1046. icmap_get_uint8("quorum.two_node", &two_node);
  1047. /*
  1048. * do config verification and enablement
  1049. */
  1050. if ((!have_nodelist) && (!expected_votes)) {
  1051. if (!runtime) {
  1052. error = (char *)"configuration error: nodelist or quorum.expected_votes must be configured!";
  1053. } else {
  1054. log_printf(LOGSYS_LEVEL_CRIT, "configuration error: nodelist or quorum.expected_votes must be configured!");
  1055. log_printf(LOGSYS_LEVEL_CRIT, "will continue with current runtime data");
  1056. }
  1057. goto out;
  1058. }
  1059. /*
  1060. * two_node and qdevice are not compatible in the same config.
  1061. * try to make an educated guess of what to do
  1062. */
  1063. if ((two_node) && (have_qdevice)) {
  1064. if (!runtime) {
  1065. error = (char *)"configuration error: two_node and quorum device cannot be configured at the same time!";
  1066. goto out;
  1067. } else {
  1068. log_printf(LOGSYS_LEVEL_CRIT, "configuration error: two_node and quorum device cannot be configured at the same time!");
  1069. if (us->flags & NODE_FLAGS_QDEVICE_REGISTERED) {
  1070. log_printf(LOGSYS_LEVEL_CRIT, "quorum device is registered, disabling two_node");
  1071. two_node = 0;
  1072. } else {
  1073. log_printf(LOGSYS_LEVEL_CRIT, "quorum device is not registered, allowing two_node");
  1074. update_qdevice_can_operate(0);
  1075. }
  1076. }
  1077. }
  1078. /*
  1079. * Enable special features
  1080. */
  1081. if (!runtime) {
  1082. if (two_node) {
  1083. wait_for_all = 1;
  1084. }
  1085. icmap_get_uint8("quorum.allow_downscale", &allow_downscale);
  1086. icmap_get_uint8("quorum.wait_for_all", &wait_for_all);
  1087. icmap_get_uint8("quorum.last_man_standing", &last_man_standing);
  1088. icmap_get_uint32("quorum.last_man_standing_window", &last_man_standing_window);
  1089. icmap_get_uint8("quorum.expected_votes_tracking", &ev_tracking);
  1090. icmap_get_uint8("quorum.auto_tie_breaker", &atb);
  1091. icmap_get_string("quorum.auto_tie_breaker_node", &atb_string);
  1092. /* auto_tie_breaker defaults to LOWEST */
  1093. if (atb) {
  1094. auto_tie_breaker = ATB_LOWEST;
  1095. icmap_set_uint32("runtime.votequorum.atb_type", auto_tie_breaker);
  1096. }
  1097. else {
  1098. auto_tie_breaker = ATB_NONE;
  1099. if (atb_string) {
  1100. log_printf(LOGSYS_LEVEL_WARNING,
  1101. "auto_tie_breaker_node: is meaningless if auto_tie_breaker is set to 0");
  1102. }
  1103. }
  1104. if (atb && atb_string) {
  1105. parse_atb_string(atb_string);
  1106. }
  1107. free(atb_string);
  1108. initial_auto_tie_breaker = auto_tie_breaker;
  1109. /* allow_downscale requires ev_tracking */
  1110. if (allow_downscale) {
  1111. ev_tracking = 1;
  1112. }
  1113. if (ev_tracking) {
  1114. if (load_ev_tracking_barrier() < 0) {
  1115. LEAVE();
  1116. return ((char *)"Unable to load ev_tracking file!");
  1117. }
  1118. update_ev_tracking_barrier(ev_tracking_barrier);
  1119. }
  1120. }
  1121. /* two_node and auto_tie_breaker are not compatible as two_node uses
  1122. * a fence race to decide quorum whereas ATB decides based on node id
  1123. */
  1124. if (two_node && auto_tie_breaker != ATB_NONE) {
  1125. log_printf(LOGSYS_LEVEL_CRIT, "two_node and auto_tie_breaker are both specified but are not compatible.");
  1126. log_printf(LOGSYS_LEVEL_CRIT, "two_node has been disabled, please fix your corosync.conf");
  1127. two_node = 0;
  1128. }
  1129. /* If ATB is set and the cluster has an odd number of nodes then wait_for_all needs
  1130. * to be set so that an isolated half+1 without the tie breaker node
  1131. * does not have quorum on reboot.
  1132. */
  1133. if ((auto_tie_breaker != ATB_NONE) && (node_expected_votes % 2) &&
  1134. (!wait_for_all)) {
  1135. if (last_man_standing) {
  1136. /* if LMS is set too, it's a fatal configuration error. We can't dictate to the user what
  1137. * they might want so we'll just quit.
  1138. */
  1139. log_printf(LOGSYS_LEVEL_CRIT, "auto_tie_breaker is set, the cluster has an odd number of nodes\n");
  1140. log_printf(LOGSYS_LEVEL_CRIT, "and last_man_standing is also set. With this situation a better\n");
  1141. log_printf(LOGSYS_LEVEL_CRIT, "solution would be to disable LMS, leave ATB enabled, and also\n");
  1142. log_printf(LOGSYS_LEVEL_CRIT, "enable wait_for_all (mandatory for ATB in odd-numbered clusters).\n");
  1143. log_printf(LOGSYS_LEVEL_CRIT, "Due to this ambiguity, corosync will fail to start. Please fix your corosync.conf\n");
  1144. error = (char *)"configuration error: auto_tie_breaker & last_man_standing not available in odd sized cluster";
  1145. goto out;
  1146. }
  1147. else {
  1148. log_printf(LOGSYS_LEVEL_CRIT, "auto_tie_breaker is set and the cluster has an odd number of nodes.\n");
  1149. log_printf(LOGSYS_LEVEL_CRIT, "wait_for_all needs to be set for this configuration but it is missing\n");
  1150. log_printf(LOGSYS_LEVEL_CRIT, "Therefore auto_tie_breaker has been disabled. Please fix your corosync.conf\n");
  1151. auto_tie_breaker = ATB_NONE;
  1152. icmap_set_uint32("runtime.votequorum.atb_type", auto_tie_breaker);
  1153. }
  1154. }
  1155. /*
  1156. * quorum device is not compatible with last_man_standing and auto_tie_breaker
  1157. * neither lms or atb can be set at runtime, so there is no need to check for
  1158. * runtime incompatibilities, but qdevice can be configured _after_ LMS and ATB have
  1159. * been enabled at startup.
  1160. */
  1161. if ((have_qdevice) && (last_man_standing)) {
  1162. if (!runtime) {
  1163. error = (char *)"configuration error: quorum.device is not compatible with last_man_standing";
  1164. goto out;
  1165. } else {
  1166. log_printf(LOGSYS_LEVEL_CRIT, "configuration error: quorum.device is not compatible with last_man_standing");
  1167. log_printf(LOGSYS_LEVEL_CRIT, "disabling quorum device operations");
  1168. update_qdevice_can_operate(0);
  1169. }
  1170. }
  1171. if ((have_qdevice) && (auto_tie_breaker != ATB_NONE)) {
  1172. if (!runtime) {
  1173. error = (char *)"configuration error: quorum.device is not compatible with auto_tie_breaker";
  1174. goto out;
  1175. } else {
  1176. log_printf(LOGSYS_LEVEL_CRIT, "configuration error: quorum.device is not compatible with auto_tie_breaker");
  1177. log_printf(LOGSYS_LEVEL_CRIT, "disabling quorum device operations");
  1178. update_qdevice_can_operate(0);
  1179. }
  1180. }
  1181. if ((have_qdevice) && (allow_downscale)) {
  1182. if (!runtime) {
  1183. error = (char *)"configuration error: quorum.device is not compatible with allow_downscale";
  1184. goto out;
  1185. } else {
  1186. log_printf(LOGSYS_LEVEL_CRIT, "configuration error: quorum.device is not compatible with allow_downscale");
  1187. log_printf(LOGSYS_LEVEL_CRIT, "disabling quorum device operations");
  1188. update_qdevice_can_operate(0);
  1189. }
  1190. }
  1191. /*
  1192. * if user specifies quorum.expected_votes + quorum.device but NOT the device.votes
  1193. * we don't know what the quorum device should vote.
  1194. */
  1195. if ((expected_votes) && (have_qdevice) && (qdevice_votes == -1)) {
  1196. if (!runtime) {
  1197. error = (char *)"configuration error: quorum.device.votes must be specified when quorum.expected_votes is set";
  1198. goto out;
  1199. } else {
  1200. log_printf(LOGSYS_LEVEL_CRIT, "configuration error: quorum.device.votes must be specified when quorum.expected_votes is set");
  1201. log_printf(LOGSYS_LEVEL_CRIT, "disabling quorum device operations");
  1202. update_qdevice_can_operate(0);
  1203. }
  1204. }
  1205. /*
  1206. * if user specifies a node list with uneven votes and no device.votes
  1207. * we cannot autocalculate the votes
  1208. */
  1209. if ((have_qdevice) &&
  1210. (qdevice_votes == -1) &&
  1211. (have_nodelist) &&
  1212. (node_count != node_expected_votes)) {
  1213. if (!runtime) {
  1214. error = (char *)"configuration error: quorum.device.votes must be specified when not all nodes votes 1";
  1215. goto out;
  1216. } else {
  1217. log_printf(LOGSYS_LEVEL_CRIT, "configuration error: quorum.device.votes must be specified when not all nodes votes 1");
  1218. log_printf(LOGSYS_LEVEL_CRIT, "disabling quorum device operations");
  1219. update_qdevice_can_operate(0);
  1220. }
  1221. }
  1222. /*
  1223. * validate quorum device votes vs expected_votes
  1224. */
  1225. if ((qdevice_votes > 0) && (expected_votes)) {
  1226. int delta = expected_votes - qdevice_votes;
  1227. if (delta < 2) {
  1228. if (!runtime) {
  1229. error = (char *)"configuration error: quorum.device.votes is too high or expected_votes is too low";
  1230. goto out;
  1231. } else {
  1232. log_printf(LOGSYS_LEVEL_CRIT, "configuration error: quorum.device.votes is too high or expected_votes is too low");
  1233. log_printf(LOGSYS_LEVEL_CRIT, "disabling quorum device operations");
  1234. update_qdevice_can_operate(0);
  1235. }
  1236. }
  1237. }
  1238. /*
  1239. * automatically calculate device votes and adjust expected_votes from nodelist
  1240. */
  1241. if ((have_qdevice) &&
  1242. (qdevice_votes == -1) &&
  1243. (!expected_votes) &&
  1244. (have_nodelist) &&
  1245. (node_count == node_expected_votes)) {
  1246. qdevice_votes = node_expected_votes - 1;
  1247. node_expected_votes = node_expected_votes + qdevice_votes;
  1248. }
  1249. /*
  1250. * set this node votes and expected_votes
  1251. */
  1252. log_printf(LOGSYS_LEVEL_DEBUG, "ev_tracking=%d, ev_tracking_barrier = %d: expected_votes = %d\n", ev_tracking, ev_tracking_barrier, expected_votes);
  1253. if (ev_tracking) {
  1254. expected_votes = ev_tracking_barrier;
  1255. }
  1256. if (have_nodelist) {
  1257. us->votes = node_votes;
  1258. us->expected_votes = node_expected_votes;
  1259. } else {
  1260. us->votes = 1;
  1261. icmap_get_uint32("quorum.votes", &us->votes);
  1262. }
  1263. if (expected_votes) {
  1264. us->expected_votes = expected_votes;
  1265. }
  1266. /*
  1267. * set qdevice votes
  1268. */
  1269. if (!have_qdevice) {
  1270. qdevice->votes = 0;
  1271. }
  1272. if (qdevice_votes != -1) {
  1273. qdevice->votes = qdevice_votes;
  1274. }
  1275. update_ev_barrier(us->expected_votes);
  1276. update_two_node();
  1277. if (wait_for_all) {
  1278. update_wait_for_all_status(1);
  1279. }
  1280. out:
  1281. LEAVE();
  1282. return error;
  1283. }
  1284. static void votequorum_refresh_config(
  1285. int32_t event,
  1286. const char *key_name,
  1287. struct icmap_notify_value new_val,
  1288. struct icmap_notify_value old_val,
  1289. void *user_data)
  1290. {
  1291. int old_votes, old_expected_votes;
  1292. uint8_t reloading;
  1293. uint8_t cancel_wfa;
  1294. ENTER();
  1295. /*
  1296. * If a full reload is in progress then don't do anything until it's done and
  1297. * can reconfigure it all atomically
  1298. */
  1299. if (icmap_get_uint8("config.totemconfig_reload_in_progress", &reloading) == CS_OK && reloading) {
  1300. return ;
  1301. }
  1302. icmap_get_uint8("quorum.cancel_wait_for_all", &cancel_wfa);
  1303. if (strcmp(key_name, "quorum.cancel_wait_for_all") == 0 &&
  1304. cancel_wfa >= 1) {
  1305. icmap_set_uint8("quorum.cancel_wait_for_all", 0);
  1306. if (votequorum_exec_send_reconfigure(VOTEQUORUM_RECONFIG_PARAM_CANCEL_WFA,
  1307. us->node_id, 0)) {
  1308. log_printf(LOGSYS_LEVEL_ERROR, "Failed to send Cancel WFA message to other nodes");
  1309. }
  1310. return;
  1311. }
  1312. old_votes = us->votes;
  1313. old_expected_votes = us->expected_votes;
  1314. /*
  1315. * Reload the configuration
  1316. */
  1317. votequorum_readconfig(VOTEQUORUM_READCONFIG_RUNTIME);
  1318. /*
  1319. * activate new config
  1320. */
  1321. votequorum_exec_send_nodeinfo(us->node_id);
  1322. votequorum_exec_send_nodeinfo(VOTEQUORUM_QDEVICE_NODEID);
  1323. if (us->votes != old_votes) {
  1324. if (votequorum_exec_send_reconfigure(VOTEQUORUM_RECONFIG_PARAM_NODE_VOTES,
  1325. us->node_id, us->votes)) {
  1326. log_printf(LOGSYS_LEVEL_ERROR, "Failed to send new votes message to other nodes");
  1327. }
  1328. }
  1329. if (us->expected_votes != old_expected_votes) {
  1330. if (votequorum_exec_send_reconfigure(VOTEQUORUM_RECONFIG_PARAM_EXPECTED_VOTES,
  1331. us->node_id, us->expected_votes)) {
  1332. log_printf(LOGSYS_LEVEL_ERROR, "Failed to send expected votes message to other nodes");
  1333. }
  1334. }
  1335. LEAVE();
  1336. }
  1337. static void votequorum_exec_add_config_notification(void)
  1338. {
  1339. icmap_track_t icmap_track_nodelist = NULL;
  1340. icmap_track_t icmap_track_quorum = NULL;
  1341. icmap_track_t icmap_track_reload = NULL;
  1342. ENTER();
  1343. icmap_track_add("nodelist.",
  1344. ICMAP_TRACK_ADD | ICMAP_TRACK_DELETE | ICMAP_TRACK_MODIFY | ICMAP_TRACK_PREFIX,
  1345. votequorum_refresh_config,
  1346. NULL,
  1347. &icmap_track_nodelist);
  1348. icmap_track_add("quorum.",
  1349. ICMAP_TRACK_ADD | ICMAP_TRACK_DELETE | ICMAP_TRACK_MODIFY | ICMAP_TRACK_PREFIX,
  1350. votequorum_refresh_config,
  1351. NULL,
  1352. &icmap_track_quorum);
  1353. icmap_track_add("config.totemconfig_reload_in_progress",
  1354. ICMAP_TRACK_ADD | ICMAP_TRACK_MODIFY,
  1355. votequorum_refresh_config,
  1356. NULL,
  1357. &icmap_track_reload);
  1358. LEAVE();
  1359. }
  1360. /*
  1361. * votequorum_exec core
  1362. */
  1363. static int votequorum_exec_send_reconfigure(uint8_t param, unsigned int nodeid, uint32_t value)
  1364. {
  1365. struct req_exec_quorum_reconfigure req_exec_quorum_reconfigure;
  1366. struct iovec iov[1];
  1367. int ret;
  1368. ENTER();
  1369. req_exec_quorum_reconfigure.nodeid = nodeid;
  1370. req_exec_quorum_reconfigure.value = value;
  1371. req_exec_quorum_reconfigure.param = param;
  1372. req_exec_quorum_reconfigure._pad0 = 0;
  1373. req_exec_quorum_reconfigure._pad1 = 0;
  1374. req_exec_quorum_reconfigure._pad2 = 0;
  1375. req_exec_quorum_reconfigure.header.id = SERVICE_ID_MAKE(VOTEQUORUM_SERVICE, MESSAGE_REQ_EXEC_VOTEQUORUM_RECONFIGURE);
  1376. req_exec_quorum_reconfigure.header.size = sizeof(req_exec_quorum_reconfigure);
  1377. iov[0].iov_base = (void *)&req_exec_quorum_reconfigure;
  1378. iov[0].iov_len = sizeof(req_exec_quorum_reconfigure);
  1379. ret = corosync_api->totem_mcast (iov, 1, TOTEM_AGREED);
  1380. LEAVE();
  1381. return ret;
  1382. }
  1383. static int votequorum_exec_send_nodeinfo(uint32_t nodeid)
  1384. {
  1385. struct req_exec_quorum_nodeinfo req_exec_quorum_nodeinfo;
  1386. struct iovec iov[1];
  1387. struct cluster_node *node;
  1388. int ret;
  1389. ENTER();
  1390. node = find_node_by_nodeid(nodeid);
  1391. if (!node) {
  1392. return -1;
  1393. }
  1394. memset(&req_exec_quorum_nodeinfo, 0, sizeof(req_exec_quorum_nodeinfo));
  1395. req_exec_quorum_nodeinfo.nodeid = nodeid;
  1396. req_exec_quorum_nodeinfo.votes = node->votes;
  1397. req_exec_quorum_nodeinfo.expected_votes = node->expected_votes;
  1398. req_exec_quorum_nodeinfo.flags = node->flags;
  1399. if (nodeid != VOTEQUORUM_QDEVICE_NODEID) {
  1400. decode_flags(node->flags);
  1401. }
  1402. req_exec_quorum_nodeinfo.header.id = SERVICE_ID_MAKE(VOTEQUORUM_SERVICE, MESSAGE_REQ_EXEC_VOTEQUORUM_NODEINFO);
  1403. req_exec_quorum_nodeinfo.header.size = sizeof(req_exec_quorum_nodeinfo);
  1404. iov[0].iov_base = (void *)&req_exec_quorum_nodeinfo;
  1405. iov[0].iov_len = sizeof(req_exec_quorum_nodeinfo);
  1406. ret = corosync_api->totem_mcast (iov, 1, TOTEM_AGREED);
  1407. LEAVE();
  1408. return ret;
  1409. }
  1410. static int votequorum_exec_send_qdevice_reconfigure(const char *oldname, const char *newname)
  1411. {
  1412. struct req_exec_quorum_qdevice_reconfigure req_exec_quorum_qdevice_reconfigure;
  1413. struct iovec iov[1];
  1414. int ret;
  1415. ENTER();
  1416. req_exec_quorum_qdevice_reconfigure.header.id = SERVICE_ID_MAKE(VOTEQUORUM_SERVICE, MESSAGE_REQ_EXEC_VOTEQUORUM_QDEVICE_RECONFIGURE);
  1417. req_exec_quorum_qdevice_reconfigure.header.size = sizeof(req_exec_quorum_qdevice_reconfigure);
  1418. strcpy(req_exec_quorum_qdevice_reconfigure.oldname, oldname);
  1419. strcpy(req_exec_quorum_qdevice_reconfigure.newname, newname);
  1420. iov[0].iov_base = (void *)&req_exec_quorum_qdevice_reconfigure;
  1421. iov[0].iov_len = sizeof(req_exec_quorum_qdevice_reconfigure);
  1422. ret = corosync_api->totem_mcast (iov, 1, TOTEM_AGREED);
  1423. LEAVE();
  1424. return ret;
  1425. }
  1426. static int votequorum_exec_send_qdevice_reg(uint32_t operation, const char *qdevice_name_req)
  1427. {
  1428. struct req_exec_quorum_qdevice_reg req_exec_quorum_qdevice_reg;
  1429. struct iovec iov[1];
  1430. int ret;
  1431. ENTER();
  1432. req_exec_quorum_qdevice_reg.header.id = SERVICE_ID_MAKE(VOTEQUORUM_SERVICE, MESSAGE_REQ_EXEC_VOTEQUORUM_QDEVICE_REG);
  1433. req_exec_quorum_qdevice_reg.header.size = sizeof(req_exec_quorum_qdevice_reg);
  1434. req_exec_quorum_qdevice_reg.operation = operation;
  1435. strcpy(req_exec_quorum_qdevice_reg.qdevice_name, qdevice_name_req);
  1436. iov[0].iov_base = (void *)&req_exec_quorum_qdevice_reg;
  1437. iov[0].iov_len = sizeof(req_exec_quorum_qdevice_reg);
  1438. ret = corosync_api->totem_mcast (iov, 1, TOTEM_AGREED);
  1439. LEAVE();
  1440. return ret;
  1441. }
  1442. static int votequorum_exec_send_quorum_notification(void *conn, uint64_t context)
  1443. {
  1444. struct res_lib_votequorum_quorum_notification *res_lib_votequorum_notification;
  1445. struct qb_list_head *tmp;
  1446. struct cluster_node *node;
  1447. int i = 0;
  1448. int cluster_members = 0;
  1449. int size;
  1450. char buf[sizeof(struct res_lib_votequorum_quorum_notification) + sizeof(struct votequorum_node) * (PROCESSOR_COUNT_MAX + 2)];
  1451. ENTER();
  1452. log_printf(LOGSYS_LEVEL_DEBUG, "Sending quorum callback, quorate = %d", cluster_is_quorate);
  1453. qb_list_for_each(tmp, &cluster_members_list) {
  1454. node = qb_list_entry(tmp, struct cluster_node, list);
  1455. cluster_members++;
  1456. }
  1457. if (us->flags & NODE_FLAGS_QDEVICE_REGISTERED) {
  1458. cluster_members++;
  1459. }
  1460. size = sizeof(struct res_lib_votequorum_quorum_notification) + sizeof(struct votequorum_node) * cluster_members;
  1461. res_lib_votequorum_notification = (struct res_lib_votequorum_quorum_notification *)&buf;
  1462. res_lib_votequorum_notification->quorate = cluster_is_quorate;
  1463. res_lib_votequorum_notification->context = context;
  1464. res_lib_votequorum_notification->node_list_entries = cluster_members;
  1465. res_lib_votequorum_notification->header.id = MESSAGE_RES_VOTEQUORUM_QUORUM_NOTIFICATION;
  1466. res_lib_votequorum_notification->header.size = size;
  1467. res_lib_votequorum_notification->header.error = CS_OK;
  1468. /* Send all known nodes and their states */
  1469. qb_list_for_each(tmp, &cluster_members_list) {
  1470. node = qb_list_entry(tmp, struct cluster_node, list);
  1471. res_lib_votequorum_notification->node_list[i].nodeid = node->node_id;
  1472. res_lib_votequorum_notification->node_list[i++].state = node->state;
  1473. }
  1474. if (us->flags & NODE_FLAGS_QDEVICE_REGISTERED) {
  1475. res_lib_votequorum_notification->node_list[i].nodeid = VOTEQUORUM_QDEVICE_NODEID;
  1476. res_lib_votequorum_notification->node_list[i++].state = qdevice->state;
  1477. }
  1478. /* Send it to all interested parties */
  1479. if (conn) {
  1480. int ret = corosync_api->ipc_dispatch_send(conn, &buf, size);
  1481. LEAVE();
  1482. return ret;
  1483. } else {
  1484. struct quorum_pd *qpd;
  1485. qb_list_for_each(tmp, &trackers_list) {
  1486. qpd = qb_list_entry(tmp, struct quorum_pd, list);
  1487. res_lib_votequorum_notification->context = qpd->tracking_context;
  1488. corosync_api->ipc_dispatch_send(qpd->conn, &buf, size);
  1489. }
  1490. }
  1491. LEAVE();
  1492. return 0;
  1493. }
  1494. static int votequorum_exec_send_nodelist_notification(void *conn, uint64_t context)
  1495. {
  1496. struct res_lib_votequorum_nodelist_notification *res_lib_votequorum_notification;
  1497. int i = 0;
  1498. int size;
  1499. struct qb_list_head *tmp;
  1500. char buf[sizeof(struct res_lib_votequorum_nodelist_notification) + sizeof(uint32_t) * quorum_members_entries];
  1501. ENTER();
  1502. log_printf(LOGSYS_LEVEL_DEBUG, "Sending nodelist callback. ring_id = " CS_PRI_RING_ID, quorum_ringid.nodeid, quorum_ringid.seq);
  1503. size = sizeof(struct res_lib_votequorum_nodelist_notification) + sizeof(uint32_t) * quorum_members_entries;
  1504. res_lib_votequorum_notification = (struct res_lib_votequorum_nodelist_notification *)&buf;
  1505. res_lib_votequorum_notification->node_list_entries = quorum_members_entries;
  1506. res_lib_votequorum_notification->ring_id.nodeid = quorum_ringid.nodeid;
  1507. res_lib_votequorum_notification->ring_id.seq = quorum_ringid.seq;
  1508. res_lib_votequorum_notification->context = context;
  1509. for (i=0; i<quorum_members_entries; i++) {
  1510. res_lib_votequorum_notification->node_list[i] = quorum_members[i];
  1511. }
  1512. res_lib_votequorum_notification->header.id = MESSAGE_RES_VOTEQUORUM_NODELIST_NOTIFICATION;
  1513. res_lib_votequorum_notification->header.size = size;
  1514. res_lib_votequorum_notification->header.error = CS_OK;
  1515. /* Send it to all interested parties */
  1516. if (conn) {
  1517. int ret = corosync_api->ipc_dispatch_send(conn, &buf, size);
  1518. LEAVE();
  1519. return ret;
  1520. } else {
  1521. struct quorum_pd *qpd;
  1522. qb_list_for_each(tmp, &trackers_list) {
  1523. qpd = qb_list_entry(tmp, struct quorum_pd, list);
  1524. res_lib_votequorum_notification->context = qpd->tracking_context;
  1525. corosync_api->ipc_dispatch_send(qpd->conn, &buf, size);
  1526. }
  1527. }
  1528. LEAVE();
  1529. return 0;
  1530. }
  1531. static void votequorum_exec_send_expectedvotes_notification(void)
  1532. {
  1533. struct res_lib_votequorum_expectedvotes_notification res_lib_votequorum_expectedvotes_notification;
  1534. struct quorum_pd *qpd;
  1535. struct qb_list_head *tmp;
  1536. ENTER();
  1537. log_printf(LOGSYS_LEVEL_DEBUG, "Sending expected votes callback");
  1538. res_lib_votequorum_expectedvotes_notification.header.id = MESSAGE_RES_VOTEQUORUM_EXPECTEDVOTES_NOTIFICATION;
  1539. res_lib_votequorum_expectedvotes_notification.header.size = sizeof(res_lib_votequorum_expectedvotes_notification);
  1540. res_lib_votequorum_expectedvotes_notification.header.error = CS_OK;
  1541. res_lib_votequorum_expectedvotes_notification.expected_votes = us->expected_votes;
  1542. qb_list_for_each(tmp, &trackers_list) {
  1543. qpd = qb_list_entry(tmp, struct quorum_pd, list);
  1544. res_lib_votequorum_expectedvotes_notification.context = qpd->tracking_context;
  1545. corosync_api->ipc_dispatch_send(qpd->conn, &res_lib_votequorum_expectedvotes_notification,
  1546. sizeof(struct res_lib_votequorum_expectedvotes_notification));
  1547. }
  1548. LEAVE();
  1549. }
  1550. static void exec_votequorum_qdevice_reconfigure_endian_convert (void *message)
  1551. {
  1552. ENTER();
  1553. LEAVE();
  1554. }
  1555. static void message_handler_req_exec_votequorum_qdevice_reconfigure (
  1556. const void *message,
  1557. unsigned int nodeid)
  1558. {
  1559. const struct req_exec_quorum_qdevice_reconfigure *req_exec_quorum_qdevice_reconfigure = message;
  1560. ENTER();
  1561. log_printf(LOGSYS_LEVEL_DEBUG, "Received qdevice name change req from node " CS_PRI_NODE_ID " [from: %s to: %s]",
  1562. nodeid,
  1563. req_exec_quorum_qdevice_reconfigure->oldname,
  1564. req_exec_quorum_qdevice_reconfigure->newname);
  1565. if (!strcmp(req_exec_quorum_qdevice_reconfigure->oldname, qdevice_name)) {
  1566. log_printf(LOGSYS_LEVEL_DEBUG, "Allowing qdevice rename");
  1567. memset(qdevice_name, 0, VOTEQUORUM_QDEVICE_MAX_NAME_LEN);
  1568. strcpy(qdevice_name, req_exec_quorum_qdevice_reconfigure->newname);
  1569. /*
  1570. * TODO: notify qdevices about name change?
  1571. * this is not relevant for now and can wait later on since
  1572. * qdevices are local only and libvotequorum is not final
  1573. */
  1574. }
  1575. LEAVE();
  1576. }
  1577. static void exec_votequorum_qdevice_reg_endian_convert (void *message)
  1578. {
  1579. struct req_exec_quorum_qdevice_reg *req_exec_quorum_qdevice_reg = message;
  1580. ENTER();
  1581. req_exec_quorum_qdevice_reg->operation = swab32(req_exec_quorum_qdevice_reg->operation);
  1582. LEAVE();
  1583. }
  1584. static void message_handler_req_exec_votequorum_qdevice_reg (
  1585. const void *message,
  1586. unsigned int nodeid)
  1587. {
  1588. const struct req_exec_quorum_qdevice_reg *req_exec_quorum_qdevice_reg = message;
  1589. struct res_lib_votequorum_status res_lib_votequorum_status;
  1590. int wipe_qdevice_name = 1;
  1591. struct cluster_node *node = NULL;
  1592. struct qb_list_head *tmp;
  1593. cs_error_t error = CS_OK;
  1594. ENTER();
  1595. log_printf(LOGSYS_LEVEL_DEBUG, "Received qdevice op %u req from node " CS_PRI_NODE_ID " [%s]",
  1596. req_exec_quorum_qdevice_reg->operation,
  1597. nodeid, req_exec_quorum_qdevice_reg->qdevice_name);
  1598. switch(req_exec_quorum_qdevice_reg->operation)
  1599. {
  1600. case VOTEQUORUM_QDEVICE_OPERATION_REGISTER:
  1601. if (nodeid != us->node_id) {
  1602. if (!strlen(qdevice_name)) {
  1603. log_printf(LOGSYS_LEVEL_DEBUG, "Remote qdevice name recorded");
  1604. strcpy(qdevice_name, req_exec_quorum_qdevice_reg->qdevice_name);
  1605. }
  1606. LEAVE();
  1607. return;
  1608. }
  1609. /*
  1610. * protect against the case where we broadcast qdevice registration
  1611. * to new memebers, we receive the message back, but there is no registration
  1612. * connection in progress
  1613. */
  1614. if (us->flags & NODE_FLAGS_QDEVICE_REGISTERED) {
  1615. LEAVE();
  1616. return;
  1617. }
  1618. /*
  1619. * this should NEVER happen
  1620. */
  1621. if (!qdevice_reg_conn) {
  1622. log_printf(LOGSYS_LEVEL_WARNING, "Unable to determine origin of the qdevice register call!");
  1623. LEAVE();
  1624. return;
  1625. }
  1626. /*
  1627. * registering our own device in this case
  1628. */
  1629. if (!strlen(qdevice_name)) {
  1630. strcpy(qdevice_name, req_exec_quorum_qdevice_reg->qdevice_name);
  1631. }
  1632. /*
  1633. * check if it is our device or something else
  1634. */
  1635. if ((!strncmp(req_exec_quorum_qdevice_reg->qdevice_name,
  1636. qdevice_name, VOTEQUORUM_QDEVICE_MAX_NAME_LEN))) {
  1637. us->flags |= NODE_FLAGS_QDEVICE_REGISTERED;
  1638. votequorum_exec_send_nodeinfo(VOTEQUORUM_QDEVICE_NODEID);
  1639. votequorum_exec_send_nodeinfo(us->node_id);
  1640. } else {
  1641. log_printf(LOGSYS_LEVEL_WARNING,
  1642. "A new qdevice with different name (new: %s old: %s) is trying to register!",
  1643. req_exec_quorum_qdevice_reg->qdevice_name, qdevice_name);
  1644. error = CS_ERR_EXIST;
  1645. }
  1646. res_lib_votequorum_status.header.size = sizeof(res_lib_votequorum_status);
  1647. res_lib_votequorum_status.header.id = MESSAGE_RES_VOTEQUORUM_STATUS;
  1648. res_lib_votequorum_status.header.error = error;
  1649. corosync_api->ipc_response_send(qdevice_reg_conn, &res_lib_votequorum_status, sizeof(res_lib_votequorum_status));
  1650. qdevice_reg_conn = NULL;
  1651. break;
  1652. case VOTEQUORUM_QDEVICE_OPERATION_UNREGISTER:
  1653. qb_list_for_each(tmp, &cluster_members_list) {
  1654. node = qb_list_entry(tmp, struct cluster_node, list);
  1655. if ((node->state == NODESTATE_MEMBER) &&
  1656. (node->flags & NODE_FLAGS_QDEVICE_REGISTERED)) {
  1657. wipe_qdevice_name = 0;
  1658. }
  1659. }
  1660. if (wipe_qdevice_name) {
  1661. memset(qdevice_name, 0, VOTEQUORUM_QDEVICE_MAX_NAME_LEN);
  1662. }
  1663. break;
  1664. }
  1665. LEAVE();
  1666. }
  1667. static void exec_votequorum_nodeinfo_endian_convert (void *message)
  1668. {
  1669. struct req_exec_quorum_nodeinfo *nodeinfo = message;
  1670. ENTER();
  1671. nodeinfo->nodeid = swab32(nodeinfo->nodeid);
  1672. nodeinfo->votes = swab32(nodeinfo->votes);
  1673. nodeinfo->expected_votes = swab32(nodeinfo->expected_votes);
  1674. nodeinfo->flags = swab32(nodeinfo->flags);
  1675. LEAVE();
  1676. }
  1677. static void message_handler_req_exec_votequorum_nodeinfo (
  1678. const void *message,
  1679. unsigned int sender_nodeid)
  1680. {
  1681. const struct req_exec_quorum_nodeinfo *req_exec_quorum_nodeinfo = message;
  1682. struct cluster_node *node = NULL;
  1683. int old_votes;
  1684. int old_expected;
  1685. uint32_t old_flags;
  1686. nodestate_t old_state;
  1687. int new_node = 0;
  1688. int allow_downgrade = 0;
  1689. int by_node = 0;
  1690. unsigned int nodeid = req_exec_quorum_nodeinfo->nodeid;
  1691. ENTER();
  1692. log_printf(LOGSYS_LEVEL_DEBUG, "got nodeinfo message from cluster node " CS_PRI_NODE_ID, sender_nodeid);
  1693. log_printf(LOGSYS_LEVEL_DEBUG, "nodeinfo message[" CS_PRI_NODE_ID "]: votes: %d, expected: %d flags: %d",
  1694. nodeid,
  1695. req_exec_quorum_nodeinfo->votes,
  1696. req_exec_quorum_nodeinfo->expected_votes,
  1697. req_exec_quorum_nodeinfo->flags);
  1698. if (nodeid != VOTEQUORUM_QDEVICE_NODEID) {
  1699. decode_flags(req_exec_quorum_nodeinfo->flags);
  1700. }
  1701. node = find_node_by_nodeid(nodeid);
  1702. if (!node) {
  1703. node = allocate_node(nodeid);
  1704. new_node = 1;
  1705. }
  1706. if (!node) {
  1707. corosync_api->error_memory_failure();
  1708. LEAVE();
  1709. return;
  1710. }
  1711. if (new_node) {
  1712. old_votes = 0;
  1713. old_expected = 0;
  1714. old_state = NODESTATE_DEAD;
  1715. old_flags = 0;
  1716. } else {
  1717. old_votes = node->votes;
  1718. old_expected = node->expected_votes;
  1719. old_state = node->state;
  1720. old_flags = node->flags;
  1721. }
  1722. if (nodeid == VOTEQUORUM_QDEVICE_NODEID) {
  1723. struct cluster_node *sender_node = find_node_by_nodeid(sender_nodeid);
  1724. assert(sender_node != NULL);
  1725. if ((!cluster_is_quorate) &&
  1726. (sender_node->flags & NODE_FLAGS_QUORATE)) {
  1727. node->votes = req_exec_quorum_nodeinfo->votes;
  1728. } else {
  1729. node->votes = max(node->votes, req_exec_quorum_nodeinfo->votes);
  1730. }
  1731. goto recalculate;
  1732. }
  1733. /* Update node state */
  1734. node->flags = req_exec_quorum_nodeinfo->flags;
  1735. node->votes = req_exec_quorum_nodeinfo->votes;
  1736. node->state = NODESTATE_MEMBER;
  1737. if (node->flags & NODE_FLAGS_LEAVING) {
  1738. node->state = NODESTATE_LEAVING;
  1739. allow_downgrade = 1;
  1740. by_node = 1;
  1741. }
  1742. if ((!cluster_is_quorate) &&
  1743. (node->flags & NODE_FLAGS_QUORATE)) {
  1744. allow_downgrade = 1;
  1745. us->expected_votes = req_exec_quorum_nodeinfo->expected_votes;
  1746. }
  1747. if (node->flags & NODE_FLAGS_QUORATE || (ev_tracking)) {
  1748. node->expected_votes = req_exec_quorum_nodeinfo->expected_votes;
  1749. } else {
  1750. node->expected_votes = us->expected_votes;
  1751. }
  1752. if ((last_man_standing) && (node->votes > 1)) {
  1753. log_printf(LOGSYS_LEVEL_WARNING, "Last Man Standing feature is supported only when all"
  1754. "cluster nodes votes are set to 1. Disabling LMS.");
  1755. last_man_standing = 0;
  1756. if (last_man_standing_timer_set) {
  1757. corosync_api->timer_delete(last_man_standing_timer);
  1758. last_man_standing_timer_set = 0;
  1759. }
  1760. }
  1761. recalculate:
  1762. if ((new_node) ||
  1763. (nodeid == us->node_id) ||
  1764. (node->flags & NODE_FLAGS_FIRST) ||
  1765. (old_votes != node->votes) ||
  1766. (old_expected != node->expected_votes) ||
  1767. (old_flags != node->flags) ||
  1768. (old_state != node->state)) {
  1769. recalculate_quorum(allow_downgrade, by_node);
  1770. }
  1771. if ((wait_for_all) &&
  1772. (!(node->flags & NODE_FLAGS_WFASTATUS)) &&
  1773. (node->flags & NODE_FLAGS_QUORATE)) {
  1774. update_wait_for_all_status(0);
  1775. }
  1776. LEAVE();
  1777. }
  1778. static void exec_votequorum_reconfigure_endian_convert (void *message)
  1779. {
  1780. struct req_exec_quorum_reconfigure *reconfigure = message;
  1781. ENTER();
  1782. reconfigure->nodeid = swab32(reconfigure->nodeid);
  1783. reconfigure->value = swab32(reconfigure->value);
  1784. LEAVE();
  1785. }
  1786. static void message_handler_req_exec_votequorum_reconfigure (
  1787. const void *message,
  1788. unsigned int nodeid)
  1789. {
  1790. const struct req_exec_quorum_reconfigure *req_exec_quorum_reconfigure = message;
  1791. struct cluster_node *node;
  1792. ENTER();
  1793. log_printf(LOGSYS_LEVEL_DEBUG, "got reconfigure message from cluster node " CS_PRI_NODE_ID " for " CS_PRI_NODE_ID,
  1794. nodeid, req_exec_quorum_reconfigure->nodeid);
  1795. switch(req_exec_quorum_reconfigure->param)
  1796. {
  1797. case VOTEQUORUM_RECONFIG_PARAM_EXPECTED_VOTES:
  1798. update_node_expected_votes(req_exec_quorum_reconfigure->value);
  1799. votequorum_exec_send_expectedvotes_notification();
  1800. update_ev_barrier(req_exec_quorum_reconfigure->value);
  1801. if (ev_tracking) {
  1802. us->expected_votes = max(us->expected_votes, ev_tracking_barrier);
  1803. }
  1804. recalculate_quorum(1, 0); /* Allow decrease */
  1805. break;
  1806. case VOTEQUORUM_RECONFIG_PARAM_NODE_VOTES:
  1807. node = find_node_by_nodeid(req_exec_quorum_reconfigure->nodeid);
  1808. if (!node) {
  1809. LEAVE();
  1810. return;
  1811. }
  1812. node->votes = req_exec_quorum_reconfigure->value;
  1813. recalculate_quorum(1, 0); /* Allow decrease */
  1814. break;
  1815. case VOTEQUORUM_RECONFIG_PARAM_CANCEL_WFA:
  1816. update_wait_for_all_status(0);
  1817. log_printf(LOGSYS_LEVEL_INFO, "wait_for_all_status reset by user on node " CS_PRI_NODE_ID ".",
  1818. req_exec_quorum_reconfigure->nodeid);
  1819. recalculate_quorum(0, 0);
  1820. break;
  1821. }
  1822. LEAVE();
  1823. }
  1824. static int votequorum_exec_exit_fn (void)
  1825. {
  1826. int ret = 0;
  1827. ENTER();
  1828. /*
  1829. * tell the other nodes we are leaving
  1830. */
  1831. if (allow_downscale) {
  1832. us->flags |= NODE_FLAGS_LEAVING;
  1833. ret = votequorum_exec_send_nodeinfo(us->node_id);
  1834. }
  1835. if ((ev_tracking) && (ev_tracking_fd != -1)) {
  1836. close(ev_tracking_fd);
  1837. }
  1838. LEAVE();
  1839. return ret;
  1840. }
  1841. static void votequorum_set_icmap_ro_keys(void)
  1842. {
  1843. icmap_set_ro_access("quorum.allow_downscale", CS_FALSE, CS_TRUE);
  1844. icmap_set_ro_access("quorum.wait_for_all", CS_FALSE, CS_TRUE);
  1845. icmap_set_ro_access("quorum.last_man_standing", CS_FALSE, CS_TRUE);
  1846. icmap_set_ro_access("quorum.last_man_standing_window", CS_FALSE, CS_TRUE);
  1847. icmap_set_ro_access("quorum.expected_votes_tracking", CS_FALSE, CS_TRUE);
  1848. icmap_set_ro_access("quorum.auto_tie_breaker", CS_FALSE, CS_TRUE);
  1849. icmap_set_ro_access("quorum.auto_tie_breaker_node", CS_FALSE, CS_TRUE);
  1850. }
  1851. static char *votequorum_exec_init_fn (struct corosync_api_v1 *api)
  1852. {
  1853. char *error = NULL;
  1854. ENTER();
  1855. /*
  1856. * make sure we start clean
  1857. */
  1858. qb_list_init(&cluster_members_list);
  1859. qb_list_init(&trackers_list);
  1860. qdevice = NULL;
  1861. us = NULL;
  1862. memset(cluster_nodes, 0, sizeof(cluster_nodes));
  1863. /*
  1864. * Allocate a cluster_node for qdevice
  1865. */
  1866. qdevice = allocate_node(VOTEQUORUM_QDEVICE_NODEID);
  1867. if (!qdevice) {
  1868. LEAVE();
  1869. return ((char *)"Could not allocate node.");
  1870. }
  1871. qdevice->votes = 0;
  1872. memset(qdevice_name, 0, VOTEQUORUM_QDEVICE_MAX_NAME_LEN);
  1873. /*
  1874. * Allocate a cluster_node for us
  1875. */
  1876. us = allocate_node(corosync_api->totem_nodeid_get());
  1877. if (!us) {
  1878. LEAVE();
  1879. return ((char *)"Could not allocate node.");
  1880. }
  1881. icmap_set_uint32("runtime.votequorum.this_node_id", us->node_id);
  1882. us->state = NODESTATE_MEMBER;
  1883. us->votes = 1;
  1884. us->flags |= NODE_FLAGS_FIRST;
  1885. error = votequorum_readconfig(VOTEQUORUM_READCONFIG_STARTUP);
  1886. if (error) {
  1887. return error;
  1888. }
  1889. recalculate_quorum(0, 0);
  1890. /*
  1891. * Set RO keys in icmap
  1892. */
  1893. votequorum_set_icmap_ro_keys();
  1894. /*
  1895. * Listen for changes
  1896. */
  1897. votequorum_exec_add_config_notification();
  1898. /*
  1899. * Start us off with one node
  1900. */
  1901. votequorum_exec_send_nodeinfo(us->node_id);
  1902. LEAVE();
  1903. return (NULL);
  1904. }
  1905. /*
  1906. * votequorum service core
  1907. */
  1908. static void votequorum_last_man_standing_timer_fn(void *arg)
  1909. {
  1910. ENTER();
  1911. last_man_standing_timer_set = 0;
  1912. if (cluster_is_quorate) {
  1913. recalculate_quorum(1,1);
  1914. }
  1915. LEAVE();
  1916. }
  1917. static void votequorum_sync_init (
  1918. const unsigned int *trans_list, size_t trans_list_entries,
  1919. const unsigned int *member_list, size_t member_list_entries,
  1920. const struct memb_ring_id *ring_id)
  1921. {
  1922. int i, j;
  1923. int found;
  1924. int left_nodes;
  1925. struct cluster_node *node;
  1926. ENTER();
  1927. sync_in_progress = 1;
  1928. sync_nodeinfo_sent = 0;
  1929. sync_wait_for_poll_or_timeout = 0;
  1930. if (member_list_entries > 1) {
  1931. us->flags &= ~NODE_FLAGS_FIRST;
  1932. }
  1933. /*
  1934. * we don't need to track which nodes have left directly,
  1935. * since that info is in the node db, but we need to know
  1936. * if somebody has left for last_man_standing
  1937. */
  1938. left_nodes = 0;
  1939. for (i = 0; i < quorum_members_entries; i++) {
  1940. found = 0;
  1941. for (j = 0; j < member_list_entries; j++) {
  1942. if (quorum_members[i] == member_list[j]) {
  1943. found = 1;
  1944. break;
  1945. }
  1946. }
  1947. if (found == 0) {
  1948. left_nodes = 1;
  1949. node = find_node_by_nodeid(quorum_members[i]);
  1950. if (node) {
  1951. node->state = NODESTATE_DEAD;
  1952. }
  1953. }
  1954. }
  1955. if (last_man_standing) {
  1956. if (((member_list_entries >= quorum) && (left_nodes)) ||
  1957. ((member_list_entries <= quorum) && (auto_tie_breaker != ATB_NONE) && (check_low_node_id_partition() == 1))) {
  1958. if (last_man_standing_timer_set) {
  1959. corosync_api->timer_delete(last_man_standing_timer);
  1960. last_man_standing_timer_set = 0;
  1961. }
  1962. corosync_api->timer_add_duration((unsigned long long)last_man_standing_window*1000000,
  1963. NULL, votequorum_last_man_standing_timer_fn,
  1964. &last_man_standing_timer);
  1965. last_man_standing_timer_set = 1;
  1966. }
  1967. }
  1968. memcpy(previous_quorum_members, quorum_members, sizeof(unsigned int) * quorum_members_entries);
  1969. previous_quorum_members_entries = quorum_members_entries;
  1970. memcpy(quorum_members, member_list, sizeof(unsigned int) * member_list_entries);
  1971. quorum_members_entries = member_list_entries;
  1972. memcpy(&quorum_ringid, ring_id, sizeof(*ring_id));
  1973. if (us->flags & NODE_FLAGS_QDEVICE_REGISTERED && us->flags & NODE_FLAGS_QDEVICE_ALIVE) {
  1974. /*
  1975. * Reset poll timer. Sync waiting is interrupted on valid qdevice poll or after timeout
  1976. */
  1977. if (qdevice_timer_set) {
  1978. corosync_api->timer_delete(qdevice_timer);
  1979. }
  1980. corosync_api->timer_add_duration((unsigned long long)qdevice_sync_timeout*1000000, qdevice,
  1981. qdevice_timer_fn, &qdevice_timer);
  1982. qdevice_timer_set = 1;
  1983. sync_wait_for_poll_or_timeout = 1;
  1984. log_printf(LOGSYS_LEVEL_INFO, "waiting for quorum device %s poll (but maximum for %u ms)",
  1985. qdevice_name, qdevice_sync_timeout);
  1986. }
  1987. LEAVE();
  1988. }
  1989. static int votequorum_sync_process (void)
  1990. {
  1991. if (!sync_nodeinfo_sent) {
  1992. votequorum_exec_send_nodeinfo(us->node_id);
  1993. votequorum_exec_send_nodeinfo(VOTEQUORUM_QDEVICE_NODEID);
  1994. if (strlen(qdevice_name)) {
  1995. votequorum_exec_send_qdevice_reg(VOTEQUORUM_QDEVICE_OPERATION_REGISTER,
  1996. qdevice_name);
  1997. }
  1998. votequorum_exec_send_nodelist_notification(NULL, 0LL);
  1999. sync_nodeinfo_sent = 1;
  2000. }
  2001. if (us->flags & NODE_FLAGS_QDEVICE_REGISTERED && sync_wait_for_poll_or_timeout) {
  2002. /*
  2003. * Waiting for qdevice to poll with new ringid or timeout
  2004. */
  2005. return (-1);
  2006. }
  2007. return 0;
  2008. }
  2009. static void votequorum_sync_activate (void)
  2010. {
  2011. recalculate_quorum(0, 0);
  2012. quorum_callback(quorum_members, quorum_members_entries,
  2013. cluster_is_quorate, &quorum_ringid);
  2014. votequorum_exec_send_quorum_notification(NULL, 0L);
  2015. sync_in_progress = 0;
  2016. }
  2017. static void votequorum_sync_abort (void)
  2018. {
  2019. }
  2020. char *votequorum_init(struct corosync_api_v1 *api,
  2021. quorum_set_quorate_fn_t q_set_quorate_fn)
  2022. {
  2023. char *error;
  2024. ENTER();
  2025. if (q_set_quorate_fn == NULL) {
  2026. return ((char *)"Quorate function not set");
  2027. }
  2028. corosync_api = api;
  2029. quorum_callback = q_set_quorate_fn;
  2030. error = corosync_service_link_and_init(corosync_api,
  2031. &votequorum_service[0]);
  2032. if (error) {
  2033. return (error);
  2034. }
  2035. LEAVE();
  2036. return (NULL);
  2037. }
  2038. /*
  2039. * Library Handler init/fini
  2040. */
  2041. static int quorum_lib_init_fn (void *conn)
  2042. {
  2043. struct quorum_pd *pd = (struct quorum_pd *)corosync_api->ipc_private_data_get (conn);
  2044. ENTER();
  2045. qb_list_init (&pd->list);
  2046. pd->conn = conn;
  2047. LEAVE();
  2048. return (0);
  2049. }
  2050. static int quorum_lib_exit_fn (void *conn)
  2051. {
  2052. struct quorum_pd *quorum_pd = (struct quorum_pd *)corosync_api->ipc_private_data_get (conn);
  2053. ENTER();
  2054. if (quorum_pd->tracking_enabled) {
  2055. qb_list_del (&quorum_pd->list);
  2056. qb_list_init (&quorum_pd->list);
  2057. }
  2058. LEAVE();
  2059. return (0);
  2060. }
  2061. /*
  2062. * library internal functions
  2063. */
  2064. static void qdevice_timer_fn(void *arg)
  2065. {
  2066. ENTER();
  2067. if ((!(us->flags & NODE_FLAGS_QDEVICE_ALIVE)) ||
  2068. (!qdevice_timer_set)) {
  2069. LEAVE();
  2070. return;
  2071. }
  2072. us->flags &= ~NODE_FLAGS_QDEVICE_ALIVE;
  2073. us->flags &= ~NODE_FLAGS_QDEVICE_CAST_VOTE;
  2074. log_printf(LOGSYS_LEVEL_INFO, "lost contact with quorum device %s", qdevice_name);
  2075. votequorum_exec_send_nodeinfo(us->node_id);
  2076. qdevice_timer_set = 0;
  2077. sync_wait_for_poll_or_timeout = 0;
  2078. LEAVE();
  2079. }
  2080. /*
  2081. * Library Handler Functions
  2082. */
  2083. static void message_handler_req_lib_votequorum_getinfo (void *conn, const void *message)
  2084. {
  2085. const struct req_lib_votequorum_getinfo *req_lib_votequorum_getinfo = message;
  2086. struct res_lib_votequorum_getinfo res_lib_votequorum_getinfo;
  2087. struct cluster_node *node;
  2088. unsigned int highest_expected = 0;
  2089. unsigned int total_votes = 0;
  2090. cs_error_t error = CS_OK;
  2091. uint32_t nodeid = req_lib_votequorum_getinfo->nodeid;
  2092. ENTER();
  2093. log_printf(LOGSYS_LEVEL_DEBUG, "got getinfo request on %p for node " CS_PRI_NODE_ID, conn, req_lib_votequorum_getinfo->nodeid);
  2094. if (nodeid == VOTEQUORUM_QDEVICE_NODEID) {
  2095. nodeid = us->node_id;
  2096. }
  2097. node = find_node_by_nodeid(nodeid);
  2098. if (node) {
  2099. struct cluster_node *iternode;
  2100. struct qb_list_head *nodelist;
  2101. qb_list_for_each(nodelist, &cluster_members_list) {
  2102. iternode = qb_list_entry(nodelist, struct cluster_node, list);
  2103. if (iternode->state == NODESTATE_MEMBER) {
  2104. highest_expected =
  2105. max(highest_expected, iternode->expected_votes);
  2106. total_votes += iternode->votes;
  2107. }
  2108. }
  2109. if (node->flags & NODE_FLAGS_QDEVICE_CAST_VOTE) {
  2110. total_votes += qdevice->votes;
  2111. }
  2112. switch(node->state) {
  2113. case NODESTATE_MEMBER:
  2114. res_lib_votequorum_getinfo.state = VOTEQUORUM_NODESTATE_MEMBER;
  2115. break;
  2116. case NODESTATE_DEAD:
  2117. res_lib_votequorum_getinfo.state = VOTEQUORUM_NODESTATE_DEAD;
  2118. break;
  2119. case NODESTATE_LEAVING:
  2120. res_lib_votequorum_getinfo.state = VOTEQUORUM_NODESTATE_LEAVING;
  2121. break;
  2122. default:
  2123. res_lib_votequorum_getinfo.state = node->state;
  2124. break;
  2125. }
  2126. res_lib_votequorum_getinfo.state = node->state;
  2127. res_lib_votequorum_getinfo.votes = node->votes;
  2128. res_lib_votequorum_getinfo.expected_votes = node->expected_votes;
  2129. res_lib_votequorum_getinfo.highest_expected = highest_expected;
  2130. res_lib_votequorum_getinfo.quorum = quorum;
  2131. res_lib_votequorum_getinfo.total_votes = total_votes;
  2132. res_lib_votequorum_getinfo.flags = 0;
  2133. res_lib_votequorum_getinfo.nodeid = node->node_id;
  2134. if (two_node) {
  2135. res_lib_votequorum_getinfo.flags |= VOTEQUORUM_INFO_TWONODE;
  2136. }
  2137. if (cluster_is_quorate) {
  2138. res_lib_votequorum_getinfo.flags |= VOTEQUORUM_INFO_QUORATE;
  2139. }
  2140. if (wait_for_all) {
  2141. res_lib_votequorum_getinfo.flags |= VOTEQUORUM_INFO_WAIT_FOR_ALL;
  2142. }
  2143. if (last_man_standing) {
  2144. res_lib_votequorum_getinfo.flags |= VOTEQUORUM_INFO_LAST_MAN_STANDING;
  2145. }
  2146. if (auto_tie_breaker != ATB_NONE) {
  2147. res_lib_votequorum_getinfo.flags |= VOTEQUORUM_INFO_AUTO_TIE_BREAKER;
  2148. }
  2149. if (allow_downscale) {
  2150. res_lib_votequorum_getinfo.flags |= VOTEQUORUM_INFO_ALLOW_DOWNSCALE;
  2151. }
  2152. memset(res_lib_votequorum_getinfo.qdevice_name, 0, VOTEQUORUM_QDEVICE_MAX_NAME_LEN);
  2153. strcpy(res_lib_votequorum_getinfo.qdevice_name, qdevice_name);
  2154. res_lib_votequorum_getinfo.qdevice_votes = qdevice->votes;
  2155. if (node->flags & NODE_FLAGS_QDEVICE_REGISTERED) {
  2156. res_lib_votequorum_getinfo.flags |= VOTEQUORUM_INFO_QDEVICE_REGISTERED;
  2157. }
  2158. if (node->flags & NODE_FLAGS_QDEVICE_ALIVE) {
  2159. res_lib_votequorum_getinfo.flags |= VOTEQUORUM_INFO_QDEVICE_ALIVE;
  2160. }
  2161. if (node->flags & NODE_FLAGS_QDEVICE_CAST_VOTE) {
  2162. res_lib_votequorum_getinfo.flags |= VOTEQUORUM_INFO_QDEVICE_CAST_VOTE;
  2163. }
  2164. if (node->flags & NODE_FLAGS_QDEVICE_MASTER_WINS) {
  2165. res_lib_votequorum_getinfo.flags |= VOTEQUORUM_INFO_QDEVICE_MASTER_WINS;
  2166. }
  2167. } else {
  2168. error = CS_ERR_NOT_EXIST;
  2169. }
  2170. res_lib_votequorum_getinfo.header.size = sizeof(res_lib_votequorum_getinfo);
  2171. res_lib_votequorum_getinfo.header.id = MESSAGE_RES_VOTEQUORUM_GETINFO;
  2172. res_lib_votequorum_getinfo.header.error = error;
  2173. corosync_api->ipc_response_send(conn, &res_lib_votequorum_getinfo, sizeof(res_lib_votequorum_getinfo));
  2174. log_printf(LOGSYS_LEVEL_DEBUG, "getinfo response error: %d", error);
  2175. LEAVE();
  2176. }
  2177. static void message_handler_req_lib_votequorum_setexpected (void *conn, const void *message)
  2178. {
  2179. const struct req_lib_votequorum_setexpected *req_lib_votequorum_setexpected = message;
  2180. struct res_lib_votequorum_status res_lib_votequorum_status;
  2181. cs_error_t error = CS_OK;
  2182. unsigned int newquorum;
  2183. unsigned int total_votes;
  2184. uint8_t allow_downscale_status = 0;
  2185. ENTER();
  2186. allow_downscale_status = allow_downscale;
  2187. allow_downscale = 0;
  2188. /*
  2189. * Validate new expected votes
  2190. */
  2191. newquorum = calculate_quorum(1, req_lib_votequorum_setexpected->expected_votes, &total_votes);
  2192. allow_downscale = allow_downscale_status;
  2193. if (newquorum < total_votes / 2 ||
  2194. newquorum > total_votes) {
  2195. error = CS_ERR_INVALID_PARAM;
  2196. goto error_exit;
  2197. }
  2198. update_node_expected_votes(req_lib_votequorum_setexpected->expected_votes);
  2199. if (votequorum_exec_send_reconfigure(VOTEQUORUM_RECONFIG_PARAM_EXPECTED_VOTES, us->node_id,
  2200. req_lib_votequorum_setexpected->expected_votes)) {
  2201. error = CS_ERR_NO_RESOURCES;
  2202. }
  2203. error_exit:
  2204. res_lib_votequorum_status.header.size = sizeof(res_lib_votequorum_status);
  2205. res_lib_votequorum_status.header.id = MESSAGE_RES_VOTEQUORUM_STATUS;
  2206. res_lib_votequorum_status.header.error = error;
  2207. corosync_api->ipc_response_send(conn, &res_lib_votequorum_status, sizeof(res_lib_votequorum_status));
  2208. LEAVE();
  2209. }
  2210. static void message_handler_req_lib_votequorum_setvotes (void *conn, const void *message)
  2211. {
  2212. const struct req_lib_votequorum_setvotes *req_lib_votequorum_setvotes = message;
  2213. struct res_lib_votequorum_status res_lib_votequorum_status;
  2214. struct cluster_node *node;
  2215. unsigned int newquorum;
  2216. unsigned int total_votes;
  2217. unsigned int saved_votes;
  2218. cs_error_t error = CS_OK;
  2219. unsigned int nodeid;
  2220. ENTER();
  2221. nodeid = req_lib_votequorum_setvotes->nodeid;
  2222. node = find_node_by_nodeid(nodeid);
  2223. if (!node) {
  2224. error = CS_ERR_NAME_NOT_FOUND;
  2225. goto error_exit;
  2226. }
  2227. /*
  2228. * Check votes is valid
  2229. */
  2230. saved_votes = node->votes;
  2231. node->votes = req_lib_votequorum_setvotes->votes;
  2232. newquorum = calculate_quorum(1, 0, &total_votes);
  2233. if (newquorum < total_votes / 2 ||
  2234. newquorum > total_votes) {
  2235. node->votes = saved_votes;
  2236. error = CS_ERR_INVALID_PARAM;
  2237. goto error_exit;
  2238. }
  2239. if (votequorum_exec_send_reconfigure(VOTEQUORUM_RECONFIG_PARAM_NODE_VOTES, nodeid,
  2240. req_lib_votequorum_setvotes->votes)) {
  2241. error = CS_ERR_NO_RESOURCES;
  2242. }
  2243. error_exit:
  2244. res_lib_votequorum_status.header.size = sizeof(res_lib_votequorum_status);
  2245. res_lib_votequorum_status.header.id = MESSAGE_RES_VOTEQUORUM_STATUS;
  2246. res_lib_votequorum_status.header.error = error;
  2247. corosync_api->ipc_response_send(conn, &res_lib_votequorum_status, sizeof(res_lib_votequorum_status));
  2248. LEAVE();
  2249. }
  2250. static void message_handler_req_lib_votequorum_trackstart (void *conn,
  2251. const void *message)
  2252. {
  2253. const struct req_lib_votequorum_trackstart *req_lib_votequorum_trackstart = message;
  2254. struct res_lib_votequorum_status res_lib_votequorum_status;
  2255. struct quorum_pd *quorum_pd = (struct quorum_pd *)corosync_api->ipc_private_data_get (conn);
  2256. cs_error_t error = CS_OK;
  2257. ENTER();
  2258. /*
  2259. * If an immediate listing of the current cluster membership
  2260. * is requested, generate membership list
  2261. */
  2262. if (req_lib_votequorum_trackstart->track_flags & CS_TRACK_CURRENT ||
  2263. req_lib_votequorum_trackstart->track_flags & CS_TRACK_CHANGES) {
  2264. log_printf(LOGSYS_LEVEL_DEBUG, "sending initial status to %p", conn);
  2265. votequorum_exec_send_nodelist_notification(conn, req_lib_votequorum_trackstart->context);
  2266. votequorum_exec_send_quorum_notification(conn, req_lib_votequorum_trackstart->context);
  2267. }
  2268. if (quorum_pd->tracking_enabled) {
  2269. error = CS_ERR_EXIST;
  2270. goto response_send;
  2271. }
  2272. /*
  2273. * Record requests for tracking
  2274. */
  2275. if (req_lib_votequorum_trackstart->track_flags & CS_TRACK_CHANGES ||
  2276. req_lib_votequorum_trackstart->track_flags & CS_TRACK_CHANGES_ONLY) {
  2277. quorum_pd->track_flags = req_lib_votequorum_trackstart->track_flags;
  2278. quorum_pd->tracking_enabled = 1;
  2279. quorum_pd->tracking_context = req_lib_votequorum_trackstart->context;
  2280. qb_list_add (&quorum_pd->list, &trackers_list);
  2281. }
  2282. response_send:
  2283. res_lib_votequorum_status.header.size = sizeof(res_lib_votequorum_status);
  2284. res_lib_votequorum_status.header.id = MESSAGE_RES_VOTEQUORUM_STATUS;
  2285. res_lib_votequorum_status.header.error = error;
  2286. corosync_api->ipc_response_send(conn, &res_lib_votequorum_status, sizeof(res_lib_votequorum_status));
  2287. LEAVE();
  2288. }
  2289. static void message_handler_req_lib_votequorum_trackstop (void *conn,
  2290. const void *message)
  2291. {
  2292. struct res_lib_votequorum_status res_lib_votequorum_status;
  2293. struct quorum_pd *quorum_pd = (struct quorum_pd *)corosync_api->ipc_private_data_get (conn);
  2294. int error = CS_OK;
  2295. ENTER();
  2296. if (quorum_pd->tracking_enabled) {
  2297. error = CS_OK;
  2298. quorum_pd->tracking_enabled = 0;
  2299. qb_list_del (&quorum_pd->list);
  2300. qb_list_init (&quorum_pd->list);
  2301. } else {
  2302. error = CS_ERR_NOT_EXIST;
  2303. }
  2304. res_lib_votequorum_status.header.size = sizeof(res_lib_votequorum_status);
  2305. res_lib_votequorum_status.header.id = MESSAGE_RES_VOTEQUORUM_STATUS;
  2306. res_lib_votequorum_status.header.error = error;
  2307. corosync_api->ipc_response_send(conn, &res_lib_votequorum_status, sizeof(res_lib_votequorum_status));
  2308. LEAVE();
  2309. }
  2310. static void message_handler_req_lib_votequorum_qdevice_register (void *conn,
  2311. const void *message)
  2312. {
  2313. const struct req_lib_votequorum_qdevice_register *req_lib_votequorum_qdevice_register = message;
  2314. struct res_lib_votequorum_status res_lib_votequorum_status;
  2315. cs_error_t error = CS_OK;
  2316. ENTER();
  2317. if (!qdevice_can_operate) {
  2318. log_printf(LOGSYS_LEVEL_INFO, "Registration of quorum device is disabled by incorrect corosync.conf. See logs for more information");
  2319. error = CS_ERR_ACCESS;
  2320. goto out;
  2321. }
  2322. if (us->flags & NODE_FLAGS_QDEVICE_REGISTERED) {
  2323. if ((!strncmp(req_lib_votequorum_qdevice_register->name,
  2324. qdevice_name, VOTEQUORUM_QDEVICE_MAX_NAME_LEN))) {
  2325. goto out;
  2326. } else {
  2327. log_printf(LOGSYS_LEVEL_WARNING,
  2328. "A new qdevice with different name (new: %s old: %s) is trying to re-register!",
  2329. req_lib_votequorum_qdevice_register->name, qdevice_name);
  2330. error = CS_ERR_EXIST;
  2331. goto out;
  2332. }
  2333. } else {
  2334. if (qdevice_reg_conn != NULL) {
  2335. log_printf(LOGSYS_LEVEL_WARNING,
  2336. "Registration request already in progress");
  2337. error = CS_ERR_TRY_AGAIN;
  2338. goto out;
  2339. }
  2340. qdevice_reg_conn = conn;
  2341. if (votequorum_exec_send_qdevice_reg(VOTEQUORUM_QDEVICE_OPERATION_REGISTER,
  2342. req_lib_votequorum_qdevice_register->name) != 0) {
  2343. log_printf(LOGSYS_LEVEL_WARNING,
  2344. "Unable to send qdevice registration request to cluster");
  2345. error = CS_ERR_TRY_AGAIN;
  2346. qdevice_reg_conn = NULL;
  2347. } else {
  2348. LEAVE();
  2349. return;
  2350. }
  2351. }
  2352. out:
  2353. res_lib_votequorum_status.header.size = sizeof(res_lib_votequorum_status);
  2354. res_lib_votequorum_status.header.id = MESSAGE_RES_VOTEQUORUM_STATUS;
  2355. res_lib_votequorum_status.header.error = error;
  2356. corosync_api->ipc_response_send(conn, &res_lib_votequorum_status, sizeof(res_lib_votequorum_status));
  2357. LEAVE();
  2358. }
  2359. static void message_handler_req_lib_votequorum_qdevice_unregister (void *conn,
  2360. const void *message)
  2361. {
  2362. const struct req_lib_votequorum_qdevice_unregister *req_lib_votequorum_qdevice_unregister = message;
  2363. struct res_lib_votequorum_status res_lib_votequorum_status;
  2364. cs_error_t error = CS_OK;
  2365. ENTER();
  2366. if (us->flags & NODE_FLAGS_QDEVICE_REGISTERED) {
  2367. if (strncmp(req_lib_votequorum_qdevice_unregister->name, qdevice_name, VOTEQUORUM_QDEVICE_MAX_NAME_LEN)) {
  2368. error = CS_ERR_INVALID_PARAM;
  2369. goto out;
  2370. }
  2371. if (qdevice_timer_set) {
  2372. corosync_api->timer_delete(qdevice_timer);
  2373. qdevice_timer_set = 0;
  2374. sync_wait_for_poll_or_timeout = 0;
  2375. }
  2376. us->flags &= ~NODE_FLAGS_QDEVICE_REGISTERED;
  2377. us->flags &= ~NODE_FLAGS_QDEVICE_ALIVE;
  2378. us->flags &= ~NODE_FLAGS_QDEVICE_CAST_VOTE;
  2379. us->flags &= ~NODE_FLAGS_QDEVICE_MASTER_WINS;
  2380. votequorum_exec_send_nodeinfo(us->node_id);
  2381. votequorum_exec_send_qdevice_reg(VOTEQUORUM_QDEVICE_OPERATION_UNREGISTER,
  2382. req_lib_votequorum_qdevice_unregister->name);
  2383. } else {
  2384. error = CS_ERR_NOT_EXIST;
  2385. }
  2386. out:
  2387. res_lib_votequorum_status.header.size = sizeof(res_lib_votequorum_status);
  2388. res_lib_votequorum_status.header.id = MESSAGE_RES_VOTEQUORUM_STATUS;
  2389. res_lib_votequorum_status.header.error = error;
  2390. corosync_api->ipc_response_send(conn, &res_lib_votequorum_status, sizeof(res_lib_votequorum_status));
  2391. LEAVE();
  2392. }
  2393. static void message_handler_req_lib_votequorum_qdevice_update (void *conn,
  2394. const void *message)
  2395. {
  2396. const struct req_lib_votequorum_qdevice_update *req_lib_votequorum_qdevice_update = message;
  2397. struct res_lib_votequorum_status res_lib_votequorum_status;
  2398. cs_error_t error = CS_OK;
  2399. ENTER();
  2400. if (us->flags & NODE_FLAGS_QDEVICE_REGISTERED) {
  2401. if (strncmp(req_lib_votequorum_qdevice_update->oldname, qdevice_name, VOTEQUORUM_QDEVICE_MAX_NAME_LEN)) {
  2402. error = CS_ERR_INVALID_PARAM;
  2403. goto out;
  2404. }
  2405. votequorum_exec_send_qdevice_reconfigure(req_lib_votequorum_qdevice_update->oldname,
  2406. req_lib_votequorum_qdevice_update->newname);
  2407. } else {
  2408. error = CS_ERR_NOT_EXIST;
  2409. }
  2410. out:
  2411. res_lib_votequorum_status.header.size = sizeof(res_lib_votequorum_status);
  2412. res_lib_votequorum_status.header.id = MESSAGE_RES_VOTEQUORUM_STATUS;
  2413. res_lib_votequorum_status.header.error = error;
  2414. corosync_api->ipc_response_send(conn, &res_lib_votequorum_status, sizeof(res_lib_votequorum_status));
  2415. LEAVE();
  2416. }
  2417. static void message_handler_req_lib_votequorum_qdevice_poll (void *conn,
  2418. const void *message)
  2419. {
  2420. const struct req_lib_votequorum_qdevice_poll *req_lib_votequorum_qdevice_poll = message;
  2421. struct res_lib_votequorum_status res_lib_votequorum_status;
  2422. cs_error_t error = CS_OK;
  2423. uint32_t oldflags;
  2424. ENTER();
  2425. if (!qdevice_can_operate) {
  2426. error = CS_ERR_ACCESS;
  2427. goto out;
  2428. }
  2429. if (us->flags & NODE_FLAGS_QDEVICE_REGISTERED) {
  2430. if (!(req_lib_votequorum_qdevice_poll->ring_id.nodeid == quorum_ringid.nodeid &&
  2431. req_lib_votequorum_qdevice_poll->ring_id.seq == quorum_ringid.seq)) {
  2432. log_printf(LOGSYS_LEVEL_DEBUG, "Received poll ring id (" CS_PRI_RING_ID ") != last sync "
  2433. "ring id (" CS_PRI_RING_ID "). Ignoring poll call.",
  2434. req_lib_votequorum_qdevice_poll->ring_id.nodeid, req_lib_votequorum_qdevice_poll->ring_id.seq,
  2435. quorum_ringid.nodeid, quorum_ringid.seq);
  2436. error = CS_ERR_MESSAGE_ERROR;
  2437. goto out;
  2438. }
  2439. if (strncmp(req_lib_votequorum_qdevice_poll->name, qdevice_name, VOTEQUORUM_QDEVICE_MAX_NAME_LEN)) {
  2440. error = CS_ERR_INVALID_PARAM;
  2441. goto out;
  2442. }
  2443. if (qdevice_timer_set) {
  2444. corosync_api->timer_delete(qdevice_timer);
  2445. qdevice_timer_set = 0;
  2446. }
  2447. oldflags = us->flags;
  2448. us->flags |= NODE_FLAGS_QDEVICE_ALIVE;
  2449. if (req_lib_votequorum_qdevice_poll->cast_vote) {
  2450. us->flags |= NODE_FLAGS_QDEVICE_CAST_VOTE;
  2451. } else {
  2452. us->flags &= ~NODE_FLAGS_QDEVICE_CAST_VOTE;
  2453. }
  2454. if (us->flags != oldflags) {
  2455. votequorum_exec_send_nodeinfo(us->node_id);
  2456. }
  2457. corosync_api->timer_add_duration((unsigned long long)qdevice_timeout*1000000, qdevice,
  2458. qdevice_timer_fn, &qdevice_timer);
  2459. qdevice_timer_set = 1;
  2460. sync_wait_for_poll_or_timeout = 0;
  2461. } else {
  2462. error = CS_ERR_NOT_EXIST;
  2463. }
  2464. out:
  2465. res_lib_votequorum_status.header.size = sizeof(res_lib_votequorum_status);
  2466. res_lib_votequorum_status.header.id = MESSAGE_RES_VOTEQUORUM_STATUS;
  2467. res_lib_votequorum_status.header.error = error;
  2468. corosync_api->ipc_response_send(conn, &res_lib_votequorum_status, sizeof(res_lib_votequorum_status));
  2469. LEAVE();
  2470. }
  2471. static void message_handler_req_lib_votequorum_qdevice_master_wins (void *conn,
  2472. const void *message)
  2473. {
  2474. const struct req_lib_votequorum_qdevice_master_wins *req_lib_votequorum_qdevice_master_wins = message;
  2475. struct res_lib_votequorum_status res_lib_votequorum_status;
  2476. cs_error_t error = CS_OK;
  2477. uint32_t oldflags = us->flags;
  2478. ENTER();
  2479. if (!qdevice_can_operate) {
  2480. error = CS_ERR_ACCESS;
  2481. goto out;
  2482. }
  2483. if (us->flags & NODE_FLAGS_QDEVICE_REGISTERED) {
  2484. if (strncmp(req_lib_votequorum_qdevice_master_wins->name, qdevice_name, VOTEQUORUM_QDEVICE_MAX_NAME_LEN)) {
  2485. error = CS_ERR_INVALID_PARAM;
  2486. goto out;
  2487. }
  2488. if (req_lib_votequorum_qdevice_master_wins->allow) {
  2489. us->flags |= NODE_FLAGS_QDEVICE_MASTER_WINS;
  2490. } else {
  2491. us->flags &= ~NODE_FLAGS_QDEVICE_MASTER_WINS;
  2492. }
  2493. if (us->flags != oldflags) {
  2494. votequorum_exec_send_nodeinfo(us->node_id);
  2495. }
  2496. update_qdevice_master_wins(req_lib_votequorum_qdevice_master_wins->allow);
  2497. } else {
  2498. error = CS_ERR_NOT_EXIST;
  2499. }
  2500. out:
  2501. res_lib_votequorum_status.header.size = sizeof(res_lib_votequorum_status);
  2502. res_lib_votequorum_status.header.id = MESSAGE_RES_VOTEQUORUM_STATUS;
  2503. res_lib_votequorum_status.header.error = error;
  2504. corosync_api->ipc_response_send(conn, &res_lib_votequorum_status, sizeof(res_lib_votequorum_status));
  2505. LEAVE();
  2506. }