votequorum.c 81 KB

1234567891011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859606162636465666768697071727374757677787980818283848586878889909192939495969798991001011021031041051061071081091101111121131141151161171181191201211221231241251261271281291301311321331341351361371381391401411421431441451461471481491501511521531541551561571581591601611621631641651661671681691701711721731741751761771781791801811821831841851861871881891901911921931941951961971981992002012022032042052062072082092102112122132142152162172182192202212222232242252262272282292302312322332342352362372382392402412422432442452462472482492502512522532542552562572582592602612622632642652662672682692702712722732742752762772782792802812822832842852862872882892902912922932942952962972982993003013023033043053063073083093103113123133143153163173183193203213223233243253263273283293303313323333343353363373383393403413423433443453463473483493503513523533543553563573583593603613623633643653663673683693703713723733743753763773783793803813823833843853863873883893903913923933943953963973983994004014024034044054064074084094104114124134144154164174184194204214224234244254264274284294304314324334344354364374384394404414424434444454464474484494504514524534544554564574584594604614624634644654664674684694704714724734744754764774784794804814824834844854864874884894904914924934944954964974984995005015025035045055065075085095105115125135145155165175185195205215225235245255265275285295305315325335345355365375385395405415425435445455465475485495505515525535545555565575585595605615625635645655665675685695705715725735745755765775785795805815825835845855865875885895905915925935945955965975985996006016026036046056066076086096106116126136146156166176186196206216226236246256266276286296306316326336346356366376386396406416426436446456466476486496506516526536546556566576586596606616626636646656666676686696706716726736746756766776786796806816826836846856866876886896906916926936946956966976986997007017027037047057067077087097107117127137147157167177187197207217227237247257267277287297307317327337347357367377387397407417427437447457467477487497507517527537547557567577587597607617627637647657667677687697707717727737747757767777787797807817827837847857867877887897907917927937947957967977987998008018028038048058068078088098108118128138148158168178188198208218228238248258268278288298308318328338348358368378388398408418428438448458468478488498508518528538548558568578588598608618628638648658668678688698708718728738748758768778788798808818828838848858868878888898908918928938948958968978988999009019029039049059069079089099109119129139149159169179189199209219229239249259269279289299309319329339349359369379389399409419429439449459469479489499509519529539549559569579589599609619629639649659669679689699709719729739749759769779789799809819829839849859869879889899909919929939949959969979989991000100110021003100410051006100710081009101010111012101310141015101610171018101910201021102210231024102510261027102810291030103110321033103410351036103710381039104010411042104310441045104610471048104910501051105210531054105510561057105810591060106110621063106410651066106710681069107010711072107310741075107610771078107910801081108210831084108510861087108810891090109110921093109410951096109710981099110011011102110311041105110611071108110911101111111211131114111511161117111811191120112111221123112411251126112711281129113011311132113311341135113611371138113911401141114211431144114511461147114811491150115111521153115411551156115711581159116011611162116311641165116611671168116911701171117211731174117511761177117811791180118111821183118411851186118711881189119011911192119311941195119611971198119912001201120212031204120512061207120812091210121112121213121412151216121712181219122012211222122312241225122612271228122912301231123212331234123512361237123812391240124112421243124412451246124712481249125012511252125312541255125612571258125912601261126212631264126512661267126812691270127112721273127412751276127712781279128012811282128312841285128612871288128912901291129212931294129512961297129812991300130113021303130413051306130713081309131013111312131313141315131613171318131913201321132213231324132513261327132813291330133113321333133413351336133713381339134013411342134313441345134613471348134913501351135213531354135513561357135813591360136113621363136413651366136713681369137013711372137313741375137613771378137913801381138213831384138513861387138813891390139113921393139413951396139713981399140014011402140314041405140614071408140914101411141214131414141514161417141814191420142114221423142414251426142714281429143014311432143314341435143614371438143914401441144214431444144514461447144814491450145114521453145414551456145714581459146014611462146314641465146614671468146914701471147214731474147514761477147814791480148114821483148414851486148714881489149014911492149314941495149614971498149915001501150215031504150515061507150815091510151115121513151415151516151715181519152015211522152315241525152615271528152915301531153215331534153515361537153815391540154115421543154415451546154715481549155015511552155315541555155615571558155915601561156215631564156515661567156815691570157115721573157415751576157715781579158015811582158315841585158615871588158915901591159215931594159515961597159815991600160116021603160416051606160716081609161016111612161316141615161616171618161916201621162216231624162516261627162816291630163116321633163416351636163716381639164016411642164316441645164616471648164916501651165216531654165516561657165816591660166116621663166416651666166716681669167016711672167316741675167616771678167916801681168216831684168516861687168816891690169116921693169416951696169716981699170017011702170317041705170617071708170917101711171217131714171517161717171817191720172117221723172417251726172717281729173017311732173317341735173617371738173917401741174217431744174517461747174817491750175117521753175417551756175717581759176017611762176317641765176617671768176917701771177217731774177517761777177817791780178117821783178417851786178717881789179017911792179317941795179617971798179918001801180218031804180518061807180818091810181118121813181418151816181718181819182018211822182318241825182618271828182918301831183218331834183518361837183818391840184118421843184418451846184718481849185018511852185318541855185618571858185918601861186218631864186518661867186818691870187118721873187418751876187718781879188018811882188318841885188618871888188918901891189218931894189518961897189818991900190119021903190419051906190719081909191019111912191319141915191619171918191919201921192219231924192519261927192819291930193119321933193419351936193719381939194019411942194319441945194619471948194919501951195219531954195519561957195819591960196119621963196419651966196719681969197019711972197319741975197619771978197919801981198219831984198519861987198819891990199119921993199419951996199719981999200020012002200320042005200620072008200920102011201220132014201520162017201820192020202120222023202420252026202720282029203020312032203320342035203620372038203920402041204220432044204520462047204820492050205120522053205420552056205720582059206020612062206320642065206620672068206920702071207220732074207520762077207820792080208120822083208420852086208720882089209020912092209320942095209620972098209921002101210221032104210521062107210821092110211121122113211421152116211721182119212021212122212321242125212621272128212921302131213221332134213521362137213821392140214121422143214421452146214721482149215021512152215321542155215621572158215921602161216221632164216521662167216821692170217121722173217421752176217721782179218021812182218321842185218621872188218921902191219221932194219521962197219821992200220122022203220422052206220722082209221022112212221322142215221622172218221922202221222222232224222522262227222822292230223122322233223422352236223722382239224022412242224322442245224622472248224922502251225222532254225522562257225822592260226122622263226422652266226722682269227022712272227322742275227622772278227922802281228222832284228522862287228822892290229122922293229422952296229722982299230023012302230323042305230623072308230923102311231223132314231523162317231823192320232123222323232423252326232723282329233023312332233323342335233623372338233923402341234223432344234523462347234823492350235123522353235423552356235723582359236023612362236323642365236623672368236923702371237223732374237523762377237823792380238123822383238423852386238723882389239023912392239323942395239623972398239924002401240224032404240524062407240824092410241124122413241424152416241724182419242024212422242324242425242624272428242924302431243224332434243524362437243824392440244124422443244424452446244724482449245024512452245324542455245624572458245924602461246224632464246524662467246824692470247124722473247424752476247724782479248024812482248324842485248624872488248924902491249224932494249524962497249824992500250125022503250425052506250725082509251025112512251325142515251625172518251925202521252225232524252525262527252825292530253125322533253425352536253725382539254025412542254325442545254625472548254925502551255225532554255525562557255825592560256125622563256425652566256725682569257025712572257325742575257625772578257925802581258225832584258525862587258825892590259125922593259425952596259725982599260026012602260326042605260626072608260926102611261226132614261526162617261826192620262126222623262426252626262726282629263026312632263326342635263626372638263926402641264226432644264526462647264826492650265126522653265426552656265726582659266026612662266326642665266626672668266926702671267226732674267526762677267826792680268126822683268426852686268726882689269026912692269326942695269626972698269927002701270227032704270527062707270827092710271127122713271427152716271727182719272027212722272327242725272627272728272927302731273227332734273527362737273827392740274127422743274427452746274727482749275027512752275327542755275627572758275927602761276227632764276527662767276827692770277127722773277427752776277727782779278027812782278327842785278627872788278927902791279227932794279527962797279827992800280128022803280428052806280728082809281028112812281328142815281628172818281928202821282228232824282528262827282828292830283128322833283428352836283728382839284028412842284328442845284628472848284928502851285228532854285528562857285828592860286128622863286428652866286728682869287028712872287328742875287628772878287928802881288228832884288528862887288828892890289128922893289428952896289728982899290029012902290329042905290629072908290929102911291229132914291529162917291829192920292129222923292429252926292729282929293029312932293329342935293629372938293929402941294229432944294529462947294829492950295129522953295429552956295729582959296029612962
  1. /*
  2. * Copyright (c) 2009-2015 Red Hat, Inc.
  3. *
  4. * All rights reserved.
  5. *
  6. * Authors: Christine Caulfield (ccaulfie@redhat.com)
  7. * Fabio M. Di Nitto (fdinitto@redhat.com)
  8. *
  9. * This software licensed under BSD license, the text of which follows:
  10. *
  11. * Redistribution and use in source and binary forms, with or without
  12. * modification, are permitted provided that the following conditions are met:
  13. *
  14. * - Redistributions of source code must retain the above copyright notice,
  15. * this list of conditions and the following disclaimer.
  16. * - Redistributions in binary form must reproduce the above copyright notice,
  17. * this list of conditions and the following disclaimer in the documentation
  18. * and/or other materials provided with the distribution.
  19. * - Neither the name of the MontaVista Software, Inc. nor the names of its
  20. * contributors may be used to endorse or promote products derived from this
  21. * software without specific prior written permission.
  22. *
  23. * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTIBUTORS "AS IS"
  24. * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  25. * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  26. * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
  27. * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
  28. * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
  29. * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
  30. * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
  31. * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
  32. * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
  33. * THE POSSIBILITY OF SUCH DAMAGE.
  34. */
  35. #include <config.h>
  36. #include <sys/types.h>
  37. #include <sys/stat.h>
  38. #include <fcntl.h>
  39. #include <stdint.h>
  40. #include <unistd.h>
  41. #include <qb/qbipc_common.h>
  42. #include "quorum.h"
  43. #include <corosync/corodefs.h>
  44. #include <corosync/list.h>
  45. #include <corosync/logsys.h>
  46. #include <corosync/coroapi.h>
  47. #include <corosync/icmap.h>
  48. #include <corosync/votequorum.h>
  49. #include <corosync/ipc_votequorum.h>
  50. #include "service.h"
  51. #include "util.h"
  52. LOGSYS_DECLARE_SUBSYS ("VOTEQ");
  53. /*
  54. * interface with corosync
  55. */
  56. static struct corosync_api_v1 *corosync_api;
  57. /*
  58. * votequorum global config vars
  59. */
  60. static char qdevice_name[VOTEQUORUM_QDEVICE_MAX_NAME_LEN];
  61. static struct cluster_node *qdevice = NULL;
  62. static unsigned int qdevice_timeout = VOTEQUORUM_QDEVICE_DEFAULT_TIMEOUT;
  63. static unsigned int qdevice_sync_timeout = VOTEQUORUM_QDEVICE_DEFAULT_SYNC_TIMEOUT;
  64. static uint8_t qdevice_can_operate = 1;
  65. static void *qdevice_reg_conn = NULL;
  66. static uint8_t qdevice_master_wins = 0;
  67. static uint8_t two_node = 0;
  68. static uint8_t wait_for_all = 0;
  69. static uint8_t wait_for_all_status = 0;
  70. static enum {ATB_NONE, ATB_LOWEST, ATB_HIGHEST, ATB_LIST} auto_tie_breaker = ATB_NONE;
  71. static int lowest_node_id = -1;
  72. static int highest_node_id = -1;
  73. #define DEFAULT_LMS_WIN 10000
  74. static uint8_t last_man_standing = 0;
  75. static uint32_t last_man_standing_window = DEFAULT_LMS_WIN;
  76. static uint8_t allow_downscale = 0;
  77. static uint32_t ev_barrier = 0;
  78. static uint8_t ev_tracking = 0;
  79. static uint32_t ev_tracking_barrier = 0;
  80. static int ev_tracking_fd = -1;
  81. /*
  82. * votequorum_exec defines/structs/forward definitions
  83. */
  84. struct req_exec_quorum_nodeinfo {
  85. struct qb_ipc_request_header header __attribute__((aligned(8)));
  86. uint32_t nodeid;
  87. uint32_t votes;
  88. uint32_t expected_votes;
  89. uint32_t flags;
  90. } __attribute__((packed));
  91. struct req_exec_quorum_reconfigure {
  92. struct qb_ipc_request_header header __attribute__((aligned(8)));
  93. uint32_t nodeid;
  94. uint32_t value;
  95. uint8_t param;
  96. uint8_t _pad0;
  97. uint8_t _pad1;
  98. uint8_t _pad2;
  99. } __attribute__((packed));
  100. struct req_exec_quorum_qdevice_reg {
  101. struct qb_ipc_request_header header __attribute__((aligned(8)));
  102. uint32_t operation;
  103. char qdevice_name[VOTEQUORUM_QDEVICE_MAX_NAME_LEN];
  104. } __attribute__((packed));
  105. struct req_exec_quorum_qdevice_reconfigure {
  106. struct qb_ipc_request_header header __attribute__((aligned(8)));
  107. char oldname[VOTEQUORUM_QDEVICE_MAX_NAME_LEN];
  108. char newname[VOTEQUORUM_QDEVICE_MAX_NAME_LEN];
  109. } __attribute__((packed));
  110. /*
  111. * votequorum_exec onwire version (via totem)
  112. */
  113. #include "votequorum.h"
  114. /*
  115. * votequorum_exec onwire messages (via totem)
  116. */
  117. #define MESSAGE_REQ_EXEC_VOTEQUORUM_NODEINFO 0
  118. #define MESSAGE_REQ_EXEC_VOTEQUORUM_RECONFIGURE 1
  119. #define MESSAGE_REQ_EXEC_VOTEQUORUM_QDEVICE_REG 2
  120. #define MESSAGE_REQ_EXEC_VOTEQUORUM_QDEVICE_RECONFIGURE 3
  121. static void votequorum_exec_send_expectedvotes_notification(void);
  122. static int votequorum_exec_send_quorum_notification(void *conn, uint64_t context);
  123. #define VOTEQUORUM_RECONFIG_PARAM_EXPECTED_VOTES 1
  124. #define VOTEQUORUM_RECONFIG_PARAM_NODE_VOTES 2
  125. #define VOTEQUORUM_RECONFIG_PARAM_CANCEL_WFA 3
  126. static int votequorum_exec_send_reconfigure(uint8_t param, unsigned int nodeid, uint32_t value);
  127. /*
  128. * used by req_exec_quorum_qdevice_reg
  129. */
  130. #define VOTEQUORUM_QDEVICE_OPERATION_UNREGISTER 0
  131. #define VOTEQUORUM_QDEVICE_OPERATION_REGISTER 1
  132. /*
  133. * votequorum internal node status/view
  134. */
  135. #define NODE_FLAGS_QUORATE 1
  136. #define NODE_FLAGS_LEAVING 2
  137. #define NODE_FLAGS_WFASTATUS 4
  138. #define NODE_FLAGS_FIRST 8
  139. #define NODE_FLAGS_QDEVICE_REGISTERED 16
  140. #define NODE_FLAGS_QDEVICE_ALIVE 32
  141. #define NODE_FLAGS_QDEVICE_CAST_VOTE 64
  142. #define NODE_FLAGS_QDEVICE_MASTER_WINS 128
  143. typedef enum {
  144. NODESTATE_MEMBER=1,
  145. NODESTATE_DEAD,
  146. NODESTATE_LEAVING
  147. } nodestate_t;
  148. struct cluster_node {
  149. int node_id;
  150. nodestate_t state;
  151. uint32_t votes;
  152. uint32_t expected_votes;
  153. uint32_t flags;
  154. struct list_head list;
  155. };
  156. /*
  157. * votequorum internal quorum status
  158. */
  159. static uint8_t quorum;
  160. static uint8_t cluster_is_quorate;
  161. /*
  162. * votequorum membership data
  163. */
  164. static struct cluster_node *us;
  165. static struct list_head cluster_members_list;
  166. static unsigned int quorum_members[PROCESSOR_COUNT_MAX];
  167. static unsigned int previous_quorum_members[PROCESSOR_COUNT_MAX];
  168. static unsigned int atb_nodelist[PROCESSOR_COUNT_MAX];
  169. static int quorum_members_entries = 0;
  170. static int previous_quorum_members_entries = 0;
  171. static int atb_nodelist_entries = 0;
  172. static struct memb_ring_id quorum_ringid;
  173. /*
  174. * pre allocate all cluster_nodes + one for qdevice
  175. */
  176. static struct cluster_node cluster_nodes[PROCESSOR_COUNT_MAX+2];
  177. static int cluster_nodes_entries = 0;
  178. /*
  179. * votequorum tracking
  180. */
  181. struct quorum_pd {
  182. unsigned char track_flags;
  183. int tracking_enabled;
  184. uint64_t tracking_context;
  185. struct list_head list;
  186. void *conn;
  187. };
  188. static struct list_head trackers_list;
  189. /*
  190. * votequorum timers
  191. */
  192. static corosync_timer_handle_t qdevice_timer;
  193. static int qdevice_timer_set = 0;
  194. static corosync_timer_handle_t last_man_standing_timer;
  195. static int last_man_standing_timer_set = 0;
  196. static int sync_nodeinfo_sent = 0;
  197. static int sync_wait_for_poll_or_timeout = 0;
  198. /*
  199. * Service Interfaces required by service_message_handler struct
  200. */
  201. static int sync_in_progress = 0;
  202. static void votequorum_sync_init (
  203. const unsigned int *trans_list,
  204. size_t trans_list_entries,
  205. const unsigned int *member_list,
  206. size_t member_list_entries,
  207. const struct memb_ring_id *ring_id);
  208. static int votequorum_sync_process (void);
  209. static void votequorum_sync_activate (void);
  210. static void votequorum_sync_abort (void);
  211. static quorum_set_quorate_fn_t quorum_callback;
  212. /*
  213. * votequorum_exec handler and definitions
  214. */
  215. static char *votequorum_exec_init_fn (struct corosync_api_v1 *api);
  216. static int votequorum_exec_exit_fn (void);
  217. static int votequorum_exec_send_nodeinfo(uint32_t nodeid);
  218. static void message_handler_req_exec_votequorum_nodeinfo (
  219. const void *message,
  220. unsigned int nodeid);
  221. static void exec_votequorum_nodeinfo_endian_convert (void *message);
  222. static void message_handler_req_exec_votequorum_reconfigure (
  223. const void *message,
  224. unsigned int nodeid);
  225. static void exec_votequorum_reconfigure_endian_convert (void *message);
  226. static void message_handler_req_exec_votequorum_qdevice_reg (
  227. const void *message,
  228. unsigned int nodeid);
  229. static void exec_votequorum_qdevice_reg_endian_convert (void *message);
  230. static void message_handler_req_exec_votequorum_qdevice_reconfigure (
  231. const void *message,
  232. unsigned int nodeid);
  233. static void exec_votequorum_qdevice_reconfigure_endian_convert (void *message);
  234. static struct corosync_exec_handler votequorum_exec_engine[] =
  235. {
  236. { /* 0 */
  237. .exec_handler_fn = message_handler_req_exec_votequorum_nodeinfo,
  238. .exec_endian_convert_fn = exec_votequorum_nodeinfo_endian_convert
  239. },
  240. { /* 1 */
  241. .exec_handler_fn = message_handler_req_exec_votequorum_reconfigure,
  242. .exec_endian_convert_fn = exec_votequorum_reconfigure_endian_convert
  243. },
  244. { /* 2 */
  245. .exec_handler_fn = message_handler_req_exec_votequorum_qdevice_reg,
  246. .exec_endian_convert_fn = exec_votequorum_qdevice_reg_endian_convert
  247. },
  248. { /* 3 */
  249. .exec_handler_fn = message_handler_req_exec_votequorum_qdevice_reconfigure,
  250. .exec_endian_convert_fn = exec_votequorum_qdevice_reconfigure_endian_convert
  251. },
  252. };
  253. /*
  254. * Library Handler and Functions Definitions
  255. */
  256. static int quorum_lib_init_fn (void *conn);
  257. static int quorum_lib_exit_fn (void *conn);
  258. static void qdevice_timer_fn(void *arg);
  259. static void message_handler_req_lib_votequorum_getinfo (void *conn,
  260. const void *message);
  261. static void message_handler_req_lib_votequorum_setexpected (void *conn,
  262. const void *message);
  263. static void message_handler_req_lib_votequorum_setvotes (void *conn,
  264. const void *message);
  265. static void message_handler_req_lib_votequorum_trackstart (void *conn,
  266. const void *message);
  267. static void message_handler_req_lib_votequorum_trackstop (void *conn,
  268. const void *message);
  269. static void message_handler_req_lib_votequorum_qdevice_register (void *conn,
  270. const void *message);
  271. static void message_handler_req_lib_votequorum_qdevice_unregister (void *conn,
  272. const void *message);
  273. static void message_handler_req_lib_votequorum_qdevice_update (void *conn,
  274. const void *message);
  275. static void message_handler_req_lib_votequorum_qdevice_poll (void *conn,
  276. const void *message);
  277. static void message_handler_req_lib_votequorum_qdevice_master_wins (void *conn,
  278. const void *message);
  279. static struct corosync_lib_handler quorum_lib_service[] =
  280. {
  281. { /* 0 */
  282. .lib_handler_fn = message_handler_req_lib_votequorum_getinfo,
  283. .flow_control = COROSYNC_LIB_FLOW_CONTROL_NOT_REQUIRED
  284. },
  285. { /* 1 */
  286. .lib_handler_fn = message_handler_req_lib_votequorum_setexpected,
  287. .flow_control = COROSYNC_LIB_FLOW_CONTROL_NOT_REQUIRED
  288. },
  289. { /* 2 */
  290. .lib_handler_fn = message_handler_req_lib_votequorum_setvotes,
  291. .flow_control = COROSYNC_LIB_FLOW_CONTROL_NOT_REQUIRED
  292. },
  293. { /* 3 */
  294. .lib_handler_fn = message_handler_req_lib_votequorum_trackstart,
  295. .flow_control = COROSYNC_LIB_FLOW_CONTROL_NOT_REQUIRED
  296. },
  297. { /* 4 */
  298. .lib_handler_fn = message_handler_req_lib_votequorum_trackstop,
  299. .flow_control = COROSYNC_LIB_FLOW_CONTROL_NOT_REQUIRED
  300. },
  301. { /* 5 */
  302. .lib_handler_fn = message_handler_req_lib_votequorum_qdevice_register,
  303. .flow_control = COROSYNC_LIB_FLOW_CONTROL_NOT_REQUIRED
  304. },
  305. { /* 6 */
  306. .lib_handler_fn = message_handler_req_lib_votequorum_qdevice_unregister,
  307. .flow_control = COROSYNC_LIB_FLOW_CONTROL_NOT_REQUIRED
  308. },
  309. { /* 7 */
  310. .lib_handler_fn = message_handler_req_lib_votequorum_qdevice_update,
  311. .flow_control = COROSYNC_LIB_FLOW_CONTROL_NOT_REQUIRED
  312. },
  313. { /* 8 */
  314. .lib_handler_fn = message_handler_req_lib_votequorum_qdevice_poll,
  315. .flow_control = COROSYNC_LIB_FLOW_CONTROL_NOT_REQUIRED
  316. },
  317. { /* 9 */
  318. .lib_handler_fn = message_handler_req_lib_votequorum_qdevice_master_wins,
  319. .flow_control = COROSYNC_LIB_FLOW_CONTROL_NOT_REQUIRED
  320. }
  321. };
  322. static struct corosync_service_engine votequorum_service_engine = {
  323. .name = "corosync vote quorum service v1.0",
  324. .id = VOTEQUORUM_SERVICE,
  325. .priority = 2,
  326. .private_data_size = sizeof (struct quorum_pd),
  327. .allow_inquorate = CS_LIB_ALLOW_INQUORATE,
  328. .flow_control = COROSYNC_LIB_FLOW_CONTROL_REQUIRED,
  329. .lib_init_fn = quorum_lib_init_fn,
  330. .lib_exit_fn = quorum_lib_exit_fn,
  331. .lib_engine = quorum_lib_service,
  332. .lib_engine_count = sizeof (quorum_lib_service) / sizeof (struct corosync_lib_handler),
  333. .exec_init_fn = votequorum_exec_init_fn,
  334. .exec_exit_fn = votequorum_exec_exit_fn,
  335. .exec_engine = votequorum_exec_engine,
  336. .exec_engine_count = sizeof (votequorum_exec_engine) / sizeof (struct corosync_exec_handler),
  337. .sync_init = votequorum_sync_init,
  338. .sync_process = votequorum_sync_process,
  339. .sync_activate = votequorum_sync_activate,
  340. .sync_abort = votequorum_sync_abort
  341. };
  342. struct corosync_service_engine *votequorum_get_service_engine_ver0 (void)
  343. {
  344. return (&votequorum_service_engine);
  345. }
  346. static struct default_service votequorum_service[] = {
  347. {
  348. .name = "corosync_votequorum",
  349. .ver = 0,
  350. .loader = votequorum_get_service_engine_ver0
  351. },
  352. };
  353. /*
  354. * common/utility macros/functions
  355. */
  356. #define max(a,b) (((a) > (b)) ? (a) : (b))
  357. #define list_iterate(v, head) \
  358. for (v = (head)->next; v != head; v = v->next)
  359. static void node_add_ordered(struct cluster_node *newnode)
  360. {
  361. struct cluster_node *node = NULL;
  362. struct list_head *tmp;
  363. struct list_head *newlist = &newnode->list;
  364. ENTER();
  365. list_iterate(tmp, &cluster_members_list) {
  366. node = list_entry(tmp, struct cluster_node, list);
  367. if (newnode->node_id < node->node_id) {
  368. break;
  369. }
  370. }
  371. if (!node) {
  372. list_add(&newnode->list, &cluster_members_list);
  373. } else {
  374. newlist->prev = tmp->prev;
  375. newlist->next = tmp;
  376. tmp->prev->next = newlist;
  377. tmp->prev = newlist;
  378. }
  379. LEAVE();
  380. }
  381. static struct cluster_node *allocate_node(unsigned int nodeid)
  382. {
  383. struct cluster_node *cl = NULL;
  384. struct list_head *tmp;
  385. ENTER();
  386. if (cluster_nodes_entries <= PROCESSOR_COUNT_MAX + 1) {
  387. cl = (struct cluster_node *)&cluster_nodes[cluster_nodes_entries];
  388. cluster_nodes_entries++;
  389. } else {
  390. list_iterate(tmp, &cluster_members_list) {
  391. cl = list_entry(tmp, struct cluster_node, list);
  392. if (cl->state == NODESTATE_DEAD) {
  393. break;
  394. }
  395. }
  396. /*
  397. * this should never happen
  398. */
  399. if (!cl) {
  400. log_printf(LOGSYS_LEVEL_CRIT, "Unable to find memory for node %u data!!", nodeid);
  401. goto out;
  402. }
  403. list_del(tmp);
  404. }
  405. memset(cl, 0, sizeof(struct cluster_node));
  406. cl->node_id = nodeid;
  407. if (nodeid != VOTEQUORUM_QDEVICE_NODEID) {
  408. node_add_ordered(cl);
  409. }
  410. out:
  411. LEAVE();
  412. return cl;
  413. }
  414. static struct cluster_node *find_node_by_nodeid(unsigned int nodeid)
  415. {
  416. struct cluster_node *node;
  417. struct list_head *tmp;
  418. ENTER();
  419. if (nodeid == us->node_id) {
  420. LEAVE();
  421. return us;
  422. }
  423. if (nodeid == VOTEQUORUM_QDEVICE_NODEID) {
  424. LEAVE();
  425. return qdevice;
  426. }
  427. list_iterate(tmp, &cluster_members_list) {
  428. node = list_entry(tmp, struct cluster_node, list);
  429. if (node->node_id == nodeid) {
  430. LEAVE();
  431. return node;
  432. }
  433. }
  434. LEAVE();
  435. return NULL;
  436. }
  437. static void get_lowest_node_id(void)
  438. {
  439. struct cluster_node *node = NULL;
  440. struct list_head *tmp;
  441. ENTER();
  442. lowest_node_id = us->node_id;
  443. list_iterate(tmp, &cluster_members_list) {
  444. node = list_entry(tmp, struct cluster_node, list);
  445. if ((node->state == NODESTATE_MEMBER) &&
  446. (node->node_id < lowest_node_id)) {
  447. lowest_node_id = node->node_id;
  448. }
  449. }
  450. log_printf(LOGSYS_LEVEL_DEBUG, "lowest node id: %d us: %d", lowest_node_id, us->node_id);
  451. icmap_set_uint32("runtime.votequorum.lowest_node_id", lowest_node_id);
  452. LEAVE();
  453. }
  454. static void get_highest_node_id(void)
  455. {
  456. struct cluster_node *node = NULL;
  457. struct list_head *tmp;
  458. ENTER();
  459. highest_node_id = us->node_id;
  460. list_iterate(tmp, &cluster_members_list) {
  461. node = list_entry(tmp, struct cluster_node, list);
  462. if ((node->state == NODESTATE_MEMBER) &&
  463. (node->node_id > highest_node_id)) {
  464. highest_node_id = node->node_id;
  465. }
  466. }
  467. log_printf(LOGSYS_LEVEL_DEBUG, "highest node id: %d us: %d", highest_node_id, us->node_id);
  468. icmap_set_uint32("runtime.votequorum.highest_node_id", highest_node_id);
  469. LEAVE();
  470. }
  471. static int check_low_node_id_partition(void)
  472. {
  473. struct cluster_node *node = NULL;
  474. struct list_head *tmp;
  475. int found = 0;
  476. ENTER();
  477. list_iterate(tmp, &cluster_members_list) {
  478. node = list_entry(tmp, struct cluster_node, list);
  479. if ((node->state == NODESTATE_MEMBER) &&
  480. (node->node_id == lowest_node_id)) {
  481. found = 1;
  482. }
  483. }
  484. LEAVE();
  485. return found;
  486. }
  487. static int check_high_node_id_partition(void)
  488. {
  489. struct cluster_node *node = NULL;
  490. struct list_head *tmp;
  491. int found = 0;
  492. ENTER();
  493. list_iterate(tmp, &cluster_members_list) {
  494. node = list_entry(tmp, struct cluster_node, list);
  495. if ((node->state == NODESTATE_MEMBER) &&
  496. (node->node_id == highest_node_id)) {
  497. found = 1;
  498. }
  499. }
  500. LEAVE();
  501. return found;
  502. }
  503. static int is_in_nodelist(int nodeid, unsigned int *members, int entries)
  504. {
  505. int i;
  506. ENTER();
  507. for (i=0; i<entries; i++) {
  508. if (nodeid == members[i]) {
  509. LEAVE();
  510. return 1;
  511. }
  512. }
  513. LEAVE();
  514. return 0;
  515. }
  516. /*
  517. * The algorithm for a list of tie-breaker nodes is:
  518. * travel the list of nodes in the auto_tie_breaker list,
  519. * if the node IS in our current partition, check if the
  520. * nodes earlier in the atb list are in the 'previous' partition;
  521. * If none are found then we are safe to be quorate, if any are
  522. * then we cannot be as we don't know if that node is up or down.
  523. * If we don't have a node in the current list we are NOT quorate.
  524. * Obviously if we find the first node in the atb list in our
  525. * partition then we are quorate.
  526. *
  527. * Special cases lowest nodeid, and highest nodeid are handled separately.
  528. */
  529. static int check_auto_tie_breaker(void)
  530. {
  531. int i, j;
  532. int res;
  533. ENTER();
  534. if (auto_tie_breaker == ATB_LOWEST) {
  535. res = check_low_node_id_partition();
  536. log_printf(LOGSYS_LEVEL_DEBUG, "ATB_LOWEST decision: %d", res);
  537. LEAVE();
  538. return res;
  539. }
  540. if (auto_tie_breaker == ATB_HIGHEST) {
  541. res = check_high_node_id_partition();
  542. log_printf(LOGSYS_LEVEL_DEBUG, "ATB_HIGHEST decision: %d", res);
  543. LEAVE();
  544. return res;
  545. }
  546. /* Assume ATB_LIST, we should never be called for ATB_NONE */
  547. for (i=0; i < atb_nodelist_entries; i++) {
  548. if (is_in_nodelist(atb_nodelist[i], quorum_members, quorum_members_entries)) {
  549. /*
  550. * Node is in our partition, if any of its predecessors are
  551. * in the previous quorum partition then it might be in the
  552. * 'other half' (as we've got this far without seeing it here)
  553. * and so we can't be quorate.
  554. */
  555. for (j=0; j<i; j++) {
  556. if (is_in_nodelist(atb_nodelist[j], previous_quorum_members, previous_quorum_members_entries)) {
  557. log_printf(LOGSYS_LEVEL_DEBUG, "ATB_LIST found node %d in previous partition but not here, quorum denied", atb_nodelist[j]);
  558. LEAVE();
  559. return 0;
  560. }
  561. }
  562. /*
  563. * None of the other list nodes were in the previous partition, if there
  564. * are enough votes, we can be quorate
  565. */
  566. log_printf(LOGSYS_LEVEL_DEBUG, "ATB_LIST found node %d in current partition, we can be quorate", atb_nodelist[i]);
  567. LEAVE();
  568. return 1;
  569. }
  570. }
  571. log_printf(LOGSYS_LEVEL_DEBUG, "ATB_LIST found no list nodes in current partition, we cannot be quorate");
  572. LEAVE();
  573. return 0;
  574. }
  575. /*
  576. * atb_string can be either:
  577. * 'lowest'
  578. * 'highest'
  579. * a list of nodeids
  580. */
  581. static void parse_atb_string(char *atb_string)
  582. {
  583. char *ptr;
  584. long num;
  585. ENTER();
  586. auto_tie_breaker = ATB_NONE;
  587. if (!strcmp(atb_string, "lowest"))
  588. auto_tie_breaker = ATB_LOWEST;
  589. if (!strcmp(atb_string, "highest"))
  590. auto_tie_breaker = ATB_HIGHEST;
  591. if (atoi(atb_string)) {
  592. atb_nodelist_entries = 0;
  593. ptr = atb_string;
  594. do {
  595. num = strtol(ptr, &ptr, 10);
  596. if (num) {
  597. log_printf(LOGSYS_LEVEL_DEBUG, "ATB nodelist[%d] = %d", atb_nodelist_entries, num);
  598. atb_nodelist[atb_nodelist_entries++] = num;
  599. }
  600. } while (num);
  601. if (atb_nodelist_entries) {
  602. auto_tie_breaker = ATB_LIST;
  603. }
  604. }
  605. icmap_set_uint32("runtime.votequorum.atb_type", auto_tie_breaker);
  606. log_printf(LOGSYS_LEVEL_DEBUG, "ATB type = %d", auto_tie_breaker);
  607. /* Make sure we got something */
  608. if (auto_tie_breaker == ATB_NONE) {
  609. 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");
  610. auto_tie_breaker = ATB_NONE;
  611. }
  612. LEAVE();
  613. }
  614. static int check_qdevice_master(void)
  615. {
  616. struct cluster_node *node = NULL;
  617. struct list_head *tmp;
  618. int found = 0;
  619. ENTER();
  620. list_iterate(tmp, &cluster_members_list) {
  621. node = list_entry(tmp, struct cluster_node, list);
  622. if ((node->state == NODESTATE_MEMBER) &&
  623. (node->flags & NODE_FLAGS_QDEVICE_MASTER_WINS) &&
  624. (node->flags & NODE_FLAGS_QDEVICE_CAST_VOTE)) {
  625. found = 1;
  626. }
  627. }
  628. LEAVE();
  629. return found;
  630. }
  631. static void decode_flags(uint32_t flags)
  632. {
  633. ENTER();
  634. log_printf(LOGSYS_LEVEL_DEBUG,
  635. "flags: quorate: %s Leaving: %s WFA Status: %s First: %s Qdevice: %s QdeviceAlive: %s QdeviceCastVote: %s QdeviceMasterWins: %s",
  636. (flags & NODE_FLAGS_QUORATE)?"Yes":"No",
  637. (flags & NODE_FLAGS_LEAVING)?"Yes":"No",
  638. (flags & NODE_FLAGS_WFASTATUS)?"Yes":"No",
  639. (flags & NODE_FLAGS_FIRST)?"Yes":"No",
  640. (flags & NODE_FLAGS_QDEVICE_REGISTERED)?"Yes":"No",
  641. (flags & NODE_FLAGS_QDEVICE_ALIVE)?"Yes":"No",
  642. (flags & NODE_FLAGS_QDEVICE_CAST_VOTE)?"Yes":"No",
  643. (flags & NODE_FLAGS_QDEVICE_MASTER_WINS)?"Yes":"No");
  644. LEAVE();
  645. }
  646. /*
  647. * load/save are copied almost pristine from totemsrp,c
  648. */
  649. static int load_ev_tracking_barrier(void)
  650. {
  651. int res = 0;
  652. char filename[PATH_MAX];
  653. ENTER();
  654. snprintf(filename, sizeof(filename) - 1, "%s/ev_tracking", get_run_dir());
  655. ev_tracking_fd = open(filename, O_RDWR, 0700);
  656. if (ev_tracking_fd != -1) {
  657. res = read (ev_tracking_fd, &ev_tracking_barrier, sizeof(uint32_t));
  658. if (res == sizeof (uint32_t)) {
  659. LEAVE();
  660. return 0;
  661. }
  662. }
  663. ev_tracking_barrier = 0;
  664. umask(0);
  665. ev_tracking_fd = open (filename, O_CREAT|O_RDWR, 0700);
  666. if (ev_tracking_fd != -1) {
  667. res = write (ev_tracking_fd, &ev_tracking_barrier, sizeof (uint32_t));
  668. if ((res == -1) || (res != sizeof (uint32_t))) {
  669. log_printf(LOGSYS_LEVEL_WARNING,
  670. "Unable to write to %s", filename);
  671. }
  672. LEAVE();
  673. return 0;
  674. }
  675. log_printf(LOGSYS_LEVEL_WARNING,
  676. "Unable to create %s file", filename);
  677. LEAVE();
  678. return -1;
  679. }
  680. static void update_wait_for_all_status(uint8_t wfa_status)
  681. {
  682. ENTER();
  683. wait_for_all_status = wfa_status;
  684. if (wait_for_all_status) {
  685. us->flags |= NODE_FLAGS_WFASTATUS;
  686. } else {
  687. us->flags &= ~NODE_FLAGS_WFASTATUS;
  688. }
  689. icmap_set_uint8("runtime.votequorum.wait_for_all_status",
  690. wait_for_all_status);
  691. LEAVE();
  692. }
  693. static void update_two_node(void)
  694. {
  695. ENTER();
  696. icmap_set_uint8("runtime.votequorum.two_node", two_node);
  697. LEAVE();
  698. }
  699. static void update_ev_barrier(uint32_t expected_votes)
  700. {
  701. ENTER();
  702. ev_barrier = expected_votes;
  703. icmap_set_uint32("runtime.votequorum.ev_barrier", ev_barrier);
  704. LEAVE();
  705. }
  706. static void update_qdevice_can_operate(uint8_t status)
  707. {
  708. ENTER();
  709. qdevice_can_operate = status;
  710. icmap_set_uint8("runtime.votequorum.qdevice_can_operate", qdevice_can_operate);
  711. LEAVE();
  712. }
  713. static void update_qdevice_master_wins(uint8_t allow)
  714. {
  715. ENTER();
  716. qdevice_master_wins = allow;
  717. icmap_set_uint8("runtime.votequorum.qdevice_master_wins", qdevice_master_wins);
  718. LEAVE();
  719. }
  720. static void update_ev_tracking_barrier(uint32_t ev_t_barrier)
  721. {
  722. int res;
  723. ENTER();
  724. ev_tracking_barrier = ev_t_barrier;
  725. icmap_set_uint32("runtime.votequorum.ev_tracking_barrier", ev_tracking_barrier);
  726. if (lseek (ev_tracking_fd, 0, SEEK_SET) != 0) {
  727. log_printf(LOGSYS_LEVEL_WARNING,
  728. "Unable to update ev_tracking_barrier on disk data!!!");
  729. LEAVE();
  730. return;
  731. }
  732. res = write (ev_tracking_fd, &ev_tracking_barrier, sizeof (uint32_t));
  733. if (res != sizeof (uint32_t)) {
  734. log_printf(LOGSYS_LEVEL_WARNING,
  735. "Unable to update ev_tracking_barrier on disk data!!!");
  736. }
  737. #ifdef HAVE_FDATASYNC
  738. fdatasync(ev_tracking_fd);
  739. #else
  740. fsync(ev_tracking_fd);
  741. #endif
  742. LEAVE();
  743. }
  744. /*
  745. * quorum calculation core bits
  746. */
  747. static int calculate_quorum(int allow_decrease, unsigned int max_expected, unsigned int *ret_total_votes)
  748. {
  749. struct list_head *nodelist;
  750. struct cluster_node *node;
  751. unsigned int total_votes = 0;
  752. unsigned int highest_expected = 0;
  753. unsigned int newquorum, q1, q2;
  754. unsigned int total_nodes = 0;
  755. ENTER();
  756. if ((allow_downscale) && (allow_decrease) && (max_expected)) {
  757. max_expected = max(ev_barrier, max_expected);
  758. }
  759. list_iterate(nodelist, &cluster_members_list) {
  760. node = list_entry(nodelist, struct cluster_node, list);
  761. log_printf(LOGSYS_LEVEL_DEBUG, "node %u state=%d, votes=%u, expected=%u",
  762. node->node_id, node->state, node->votes, node->expected_votes);
  763. if (node->state == NODESTATE_MEMBER) {
  764. if (max_expected) {
  765. node->expected_votes = max_expected;
  766. } else {
  767. highest_expected = max(highest_expected, node->expected_votes);
  768. }
  769. total_votes += node->votes;
  770. total_nodes++;
  771. }
  772. }
  773. if (us->flags & NODE_FLAGS_QDEVICE_CAST_VOTE) {
  774. log_printf(LOGSYS_LEVEL_DEBUG, "node 0 state=1, votes=%u", qdevice->votes);
  775. total_votes += qdevice->votes;
  776. total_nodes++;
  777. }
  778. if (max_expected > 0) {
  779. highest_expected = max_expected;
  780. }
  781. /*
  782. * This quorum calculation is taken from the OpenVMS Cluster Systems
  783. * manual, but, then, you guessed that didn't you
  784. */
  785. q1 = (highest_expected + 2) / 2;
  786. q2 = (total_votes + 2) / 2;
  787. newquorum = max(q1, q2);
  788. /*
  789. * Normally quorum never decreases but the system administrator can
  790. * force it down by setting expected votes to a maximum value
  791. */
  792. if (!allow_decrease) {
  793. newquorum = max(quorum, newquorum);
  794. }
  795. /*
  796. * The special two_node mode allows each of the two nodes to retain
  797. * quorum if the other fails. Only one of the two should live past
  798. * fencing (as both nodes try to fence each other in split-brain.)
  799. * Also: if there are more than two nodes, force us inquorate to avoid
  800. * any damage or confusion.
  801. */
  802. if (two_node && total_nodes <= 2) {
  803. newquorum = 1;
  804. }
  805. if (ret_total_votes) {
  806. *ret_total_votes = total_votes;
  807. }
  808. LEAVE();
  809. return newquorum;
  810. }
  811. static void are_we_quorate(unsigned int total_votes)
  812. {
  813. int quorate;
  814. int quorum_change = 0;
  815. ENTER();
  816. /*
  817. * wait for all nodes to show up before granting quorum
  818. */
  819. if ((wait_for_all) && (wait_for_all_status)) {
  820. if (total_votes != us->expected_votes) {
  821. log_printf(LOGSYS_LEVEL_NOTICE,
  822. "Waiting for all cluster members. "
  823. "Current votes: %d expected_votes: %d",
  824. total_votes, us->expected_votes);
  825. cluster_is_quorate = 0;
  826. return;
  827. }
  828. update_wait_for_all_status(0);
  829. }
  830. if (quorum > total_votes) {
  831. quorate = 0;
  832. } else {
  833. quorate = 1;
  834. get_lowest_node_id();
  835. get_highest_node_id();
  836. }
  837. if ((auto_tie_breaker != ATB_NONE) &&
  838. /* Must be a half (or half-1) split */
  839. (total_votes == (us->expected_votes / 2)) &&
  840. /* If the 'other' partition in a split might have quorum then we can't run ATB */
  841. (previous_quorum_members_entries - quorum_members_entries < quorum) &&
  842. (check_auto_tie_breaker() == 1)) {
  843. quorate = 1;
  844. }
  845. if ((qdevice_master_wins) &&
  846. (!quorate) &&
  847. (check_qdevice_master() == 1)) {
  848. log_printf(LOGSYS_LEVEL_DEBUG, "node is quorate as part of master_wins partition");
  849. quorate = 1;
  850. }
  851. if (cluster_is_quorate && !quorate) {
  852. quorum_change = 1;
  853. log_printf(LOGSYS_LEVEL_DEBUG, "quorum lost, blocking activity");
  854. }
  855. if (!cluster_is_quorate && quorate) {
  856. quorum_change = 1;
  857. log_printf(LOGSYS_LEVEL_DEBUG, "quorum regained, resuming activity");
  858. }
  859. cluster_is_quorate = quorate;
  860. if (cluster_is_quorate) {
  861. us->flags |= NODE_FLAGS_QUORATE;
  862. } else {
  863. us->flags &= ~NODE_FLAGS_QUORATE;
  864. }
  865. if (wait_for_all) {
  866. if (quorate) {
  867. update_wait_for_all_status(0);
  868. } else {
  869. update_wait_for_all_status(1);
  870. }
  871. }
  872. if ((quorum_change) &&
  873. (sync_in_progress == 0)) {
  874. quorum_callback(quorum_members, quorum_members_entries,
  875. cluster_is_quorate, &quorum_ringid);
  876. votequorum_exec_send_quorum_notification(NULL, 0L);
  877. }
  878. LEAVE();
  879. }
  880. static void get_total_votes(unsigned int *totalvotes, unsigned int *current_members)
  881. {
  882. unsigned int total_votes = 0;
  883. unsigned int cluster_members = 0;
  884. struct list_head *nodelist;
  885. struct cluster_node *node;
  886. ENTER();
  887. list_iterate(nodelist, &cluster_members_list) {
  888. node = list_entry(nodelist, struct cluster_node, list);
  889. if (node->state == NODESTATE_MEMBER) {
  890. cluster_members++;
  891. total_votes += node->votes;
  892. }
  893. }
  894. if (qdevice->votes) {
  895. total_votes += qdevice->votes;
  896. cluster_members++;
  897. }
  898. *totalvotes = total_votes;
  899. *current_members = cluster_members;
  900. LEAVE();
  901. }
  902. /*
  903. * Recalculate cluster quorum, set quorate and notify changes
  904. */
  905. static void recalculate_quorum(int allow_decrease, int by_current_nodes)
  906. {
  907. unsigned int total_votes = 0;
  908. unsigned int cluster_members = 0;
  909. ENTER();
  910. get_total_votes(&total_votes, &cluster_members);
  911. if (!by_current_nodes) {
  912. cluster_members = 0;
  913. }
  914. /*
  915. * Keep expected_votes at the highest number of votes in the cluster
  916. */
  917. log_printf(LOGSYS_LEVEL_DEBUG, "total_votes=%d, expected_votes=%d", total_votes, us->expected_votes);
  918. if (total_votes > us->expected_votes) {
  919. us->expected_votes = total_votes;
  920. votequorum_exec_send_expectedvotes_notification();
  921. }
  922. if ((ev_tracking) &&
  923. (us->expected_votes > ev_tracking_barrier)) {
  924. update_ev_tracking_barrier(us->expected_votes);
  925. }
  926. quorum = calculate_quorum(allow_decrease, cluster_members, &total_votes);
  927. are_we_quorate(total_votes);
  928. LEAVE();
  929. }
  930. /*
  931. * configuration bits and pieces
  932. */
  933. static int votequorum_read_nodelist_configuration(uint32_t *votes,
  934. uint32_t *nodes,
  935. uint32_t *expected_votes)
  936. {
  937. icmap_iter_t iter;
  938. const char *iter_key;
  939. char tmp_key[ICMAP_KEYNAME_MAXLEN];
  940. uint32_t our_pos, node_pos;
  941. uint32_t nodecount = 0;
  942. uint32_t nodelist_expected_votes = 0;
  943. uint32_t node_votes = 0;
  944. int res = 0;
  945. ENTER();
  946. if (icmap_get_uint32("nodelist.local_node_pos", &our_pos) != CS_OK) {
  947. log_printf(LOGSYS_LEVEL_DEBUG,
  948. "No nodelist defined or our node is not in the nodelist");
  949. return 0;
  950. }
  951. iter = icmap_iter_init("nodelist.node.");
  952. while ((iter_key = icmap_iter_next(iter, NULL, NULL)) != NULL) {
  953. res = sscanf(iter_key, "nodelist.node.%u.%s", &node_pos, tmp_key);
  954. if (res != 2) {
  955. continue;
  956. }
  957. if (strcmp(tmp_key, "ring0_addr") != 0) {
  958. continue;
  959. }
  960. nodecount++;
  961. snprintf(tmp_key, ICMAP_KEYNAME_MAXLEN, "nodelist.node.%u.quorum_votes", node_pos);
  962. if (icmap_get_uint32(tmp_key, &node_votes) != CS_OK) {
  963. node_votes = 1;
  964. }
  965. nodelist_expected_votes = nodelist_expected_votes + node_votes;
  966. if (node_pos == our_pos) {
  967. *votes = node_votes;
  968. }
  969. }
  970. *expected_votes = nodelist_expected_votes;
  971. *nodes = nodecount;
  972. icmap_iter_finalize(iter);
  973. LEAVE();
  974. return 1;
  975. }
  976. static int votequorum_qdevice_is_configured(uint32_t *qdevice_votes)
  977. {
  978. char *qdevice_model = NULL;
  979. int ret = 0;
  980. ENTER();
  981. if (icmap_get_string("quorum.device.model", &qdevice_model) == CS_OK) {
  982. if (strlen(qdevice_model)) {
  983. if (icmap_get_uint32("quorum.device.votes", qdevice_votes) != CS_OK) {
  984. *qdevice_votes = -1;
  985. }
  986. if (icmap_get_uint32("quorum.device.timeout", &qdevice_timeout) != CS_OK) {
  987. qdevice_timeout = VOTEQUORUM_QDEVICE_DEFAULT_TIMEOUT;
  988. }
  989. if (icmap_get_uint32("quorum.device.sync_timeout", &qdevice_sync_timeout) != CS_OK) {
  990. qdevice_sync_timeout = VOTEQUORUM_QDEVICE_DEFAULT_SYNC_TIMEOUT;
  991. }
  992. update_qdevice_can_operate(1);
  993. ret = 1;
  994. }
  995. free(qdevice_model);
  996. }
  997. LEAVE();
  998. return ret;
  999. }
  1000. #define VOTEQUORUM_READCONFIG_STARTUP 0
  1001. #define VOTEQUORUM_READCONFIG_RUNTIME 1
  1002. static char *votequorum_readconfig(int runtime)
  1003. {
  1004. uint32_t node_votes = 0, qdevice_votes = 0;
  1005. uint32_t node_expected_votes = 0, expected_votes = 0;
  1006. uint32_t node_count = 0;
  1007. uint8_t atb = 0;
  1008. int have_nodelist, have_qdevice;
  1009. char *atb_string = NULL;
  1010. char *error = NULL;
  1011. ENTER();
  1012. log_printf(LOGSYS_LEVEL_DEBUG, "Reading configuration (runtime: %d)", runtime);
  1013. /*
  1014. * Set the few things we re-read at runtime back to their defaults
  1015. */
  1016. if (runtime) {
  1017. two_node = 0;
  1018. expected_votes = 0;
  1019. }
  1020. /*
  1021. * gather basic data here
  1022. */
  1023. icmap_get_uint32("quorum.expected_votes", &expected_votes);
  1024. have_nodelist = votequorum_read_nodelist_configuration(&node_votes, &node_count, &node_expected_votes);
  1025. have_qdevice = votequorum_qdevice_is_configured(&qdevice_votes);
  1026. icmap_get_uint8("quorum.two_node", &two_node);
  1027. /*
  1028. * do config verification and enablement
  1029. */
  1030. if ((!have_nodelist) && (!expected_votes)) {
  1031. if (!runtime) {
  1032. error = (char *)"configuration error: nodelist or quorum.expected_votes must be configured!";
  1033. } else {
  1034. log_printf(LOGSYS_LEVEL_CRIT, "configuration error: nodelist or quorum.expected_votes must be configured!");
  1035. log_printf(LOGSYS_LEVEL_CRIT, "will continue with current runtime data");
  1036. }
  1037. goto out;
  1038. }
  1039. /*
  1040. * two_node and qdevice are not compatible in the same config.
  1041. * try to make an educated guess of what to do
  1042. */
  1043. if ((two_node) && (have_qdevice)) {
  1044. if (!runtime) {
  1045. error = (char *)"configuration error: two_node and quorum device cannot be configured at the same time!";
  1046. goto out;
  1047. } else {
  1048. log_printf(LOGSYS_LEVEL_CRIT, "configuration error: two_node and quorum device cannot be configured at the same time!");
  1049. if (us->flags & NODE_FLAGS_QDEVICE_REGISTERED) {
  1050. log_printf(LOGSYS_LEVEL_CRIT, "quorum device is registered, disabling two_node");
  1051. two_node = 0;
  1052. } else {
  1053. log_printf(LOGSYS_LEVEL_CRIT, "quorum device is not registered, allowing two_node");
  1054. update_qdevice_can_operate(0);
  1055. }
  1056. }
  1057. }
  1058. /*
  1059. * Enable special features
  1060. */
  1061. if (!runtime) {
  1062. if (two_node) {
  1063. wait_for_all = 1;
  1064. }
  1065. icmap_get_uint8("quorum.allow_downscale", &allow_downscale);
  1066. icmap_get_uint8("quorum.wait_for_all", &wait_for_all);
  1067. icmap_get_uint8("quorum.last_man_standing", &last_man_standing);
  1068. icmap_get_uint32("quorum.last_man_standing_window", &last_man_standing_window);
  1069. icmap_get_uint8("quorum.expected_votes_tracking", &ev_tracking);
  1070. icmap_get_uint8("quorum.auto_tie_breaker", &atb);
  1071. icmap_get_string("quorum.auto_tie_breaker_node", &atb_string);
  1072. /* auto_tie_breaker defaults to LOWEST */
  1073. if (atb) {
  1074. auto_tie_breaker = ATB_LOWEST;
  1075. icmap_set_uint32("runtime.votequorum.atb_type", auto_tie_breaker);
  1076. }
  1077. else {
  1078. auto_tie_breaker = ATB_NONE;
  1079. if (atb_string) {
  1080. log_printf(LOGSYS_LEVEL_WARNING,
  1081. "auto_tie_breaker_node: is meaningless if auto_tie_breaker is set to 0");
  1082. }
  1083. }
  1084. if (atb && atb_string) {
  1085. parse_atb_string(atb_string);
  1086. }
  1087. free(atb_string);
  1088. /* allow_downscale requires ev_tracking */
  1089. if (allow_downscale) {
  1090. ev_tracking = 1;
  1091. }
  1092. if (ev_tracking) {
  1093. if (load_ev_tracking_barrier() < 0) {
  1094. LEAVE();
  1095. return ((char *)"Unable to load ev_tracking file!");
  1096. }
  1097. update_ev_tracking_barrier(ev_tracking_barrier);
  1098. }
  1099. }
  1100. /* two_node and auto_tie_breaker are not compatible as two_node uses
  1101. * a fence race to decide quorum whereas ATB decides based on node id
  1102. */
  1103. if (two_node && auto_tie_breaker != ATB_NONE) {
  1104. log_printf(LOGSYS_LEVEL_CRIT, "two_node and auto_tie_breaker are both specified but are not compatible.");
  1105. log_printf(LOGSYS_LEVEL_CRIT, "two_node has been disabled, please fix your corosync.conf");
  1106. two_node = 0;
  1107. }
  1108. /* If ATB is set and the cluster has an odd number of nodes then wait_for_all needs
  1109. * to be set so that an isolated half+1 without the tie breaker node
  1110. * does not have quorum on reboot.
  1111. */
  1112. if ((auto_tie_breaker != ATB_NONE) && (node_expected_votes % 2) &&
  1113. (!wait_for_all)) {
  1114. if (last_man_standing) {
  1115. /* if LMS is set too, it's a fatal configuration error. We can't dictate to the user what
  1116. * they might want so we'll just quit.
  1117. */
  1118. log_printf(LOGSYS_LEVEL_CRIT, "auto_tie_breaker is set, the cluster has an odd number of nodes\n");
  1119. log_printf(LOGSYS_LEVEL_CRIT, "and last_man_standing is also set. With this situation a better\n");
  1120. log_printf(LOGSYS_LEVEL_CRIT, "solution would be to disable LMS, leave ATB enabled, and also\n");
  1121. log_printf(LOGSYS_LEVEL_CRIT, "enable wait_for_all (mandatory for ATB in odd-numbered clusters).\n");
  1122. log_printf(LOGSYS_LEVEL_CRIT, "Due to this ambiguity, corosync will fail to start. Please fix your corosync.conf\n");
  1123. error = (char *)"configuration error: auto_tie_breaker & last_man_standing not available in odd sized cluster";
  1124. goto out;
  1125. }
  1126. else {
  1127. log_printf(LOGSYS_LEVEL_CRIT, "auto_tie_breaker is set and the cluster has an odd number of nodes.\n");
  1128. log_printf(LOGSYS_LEVEL_CRIT, "wait_for_all needs to be set for this configuration but it is missing\n");
  1129. log_printf(LOGSYS_LEVEL_CRIT, "Therefore auto_tie_breaker has been disabled. Please fix your corosync.conf\n");
  1130. auto_tie_breaker = ATB_NONE;
  1131. icmap_set_uint32("runtime.votequorum.atb_type", auto_tie_breaker);
  1132. }
  1133. }
  1134. /*
  1135. * quorum device is not compatible with last_man_standing and auto_tie_breaker
  1136. * neither lms or atb can be set at runtime, so there is no need to check for
  1137. * runtime incompatibilities, but qdevice can be configured _after_ LMS and ATB have
  1138. * been enabled at startup.
  1139. */
  1140. if ((have_qdevice) && (last_man_standing)) {
  1141. if (!runtime) {
  1142. error = (char *)"configuration error: quorum.device is not compatible with last_man_standing";
  1143. goto out;
  1144. } else {
  1145. log_printf(LOGSYS_LEVEL_CRIT, "configuration error: quorum.device is not compatible with last_man_standing");
  1146. log_printf(LOGSYS_LEVEL_CRIT, "disabling quorum device operations");
  1147. update_qdevice_can_operate(0);
  1148. }
  1149. }
  1150. if ((have_qdevice) && (auto_tie_breaker != ATB_NONE)) {
  1151. if (!runtime) {
  1152. error = (char *)"configuration error: quorum.device is not compatible with auto_tie_breaker";
  1153. goto out;
  1154. } else {
  1155. log_printf(LOGSYS_LEVEL_CRIT, "configuration error: quorum.device is not compatible with auto_tie_breaker");
  1156. log_printf(LOGSYS_LEVEL_CRIT, "disabling quorum device operations");
  1157. update_qdevice_can_operate(0);
  1158. }
  1159. }
  1160. if ((have_qdevice) && (wait_for_all)) {
  1161. if (!runtime) {
  1162. error = (char *)"configuration error: quorum.device is not compatible with wait_for_all";
  1163. goto out;
  1164. } else {
  1165. log_printf(LOGSYS_LEVEL_CRIT, "configuration error: quorum.device is not compatible with wait_for_all");
  1166. log_printf(LOGSYS_LEVEL_CRIT, "disabling quorum device operations");
  1167. update_qdevice_can_operate(0);
  1168. }
  1169. }
  1170. if ((have_qdevice) && (allow_downscale)) {
  1171. if (!runtime) {
  1172. error = (char *)"configuration error: quorum.device is not compatible with allow_downscale";
  1173. goto out;
  1174. } else {
  1175. log_printf(LOGSYS_LEVEL_CRIT, "configuration error: quorum.device is not compatible with allow_downscale");
  1176. log_printf(LOGSYS_LEVEL_CRIT, "disabling quorum device operations");
  1177. update_qdevice_can_operate(0);
  1178. }
  1179. }
  1180. /*
  1181. * if user specifies quorum.expected_votes + quorum.device but NOT the device.votes
  1182. * we don't know what the quorum device should vote.
  1183. */
  1184. if ((expected_votes) && (have_qdevice) && (qdevice_votes == -1)) {
  1185. if (!runtime) {
  1186. error = (char *)"configuration error: quorum.device.votes must be specified when quorum.expected_votes is set";
  1187. goto out;
  1188. } else {
  1189. log_printf(LOGSYS_LEVEL_CRIT, "configuration error: quorum.device.votes must be specified when quorum.expected_votes is set");
  1190. log_printf(LOGSYS_LEVEL_CRIT, "disabling quorum device operations");
  1191. update_qdevice_can_operate(0);
  1192. }
  1193. }
  1194. /*
  1195. * if user specifies a node list with uneven votes and no device.votes
  1196. * we cannot autocalculate the votes
  1197. */
  1198. if ((have_qdevice) &&
  1199. (qdevice_votes == -1) &&
  1200. (have_nodelist) &&
  1201. (node_count != node_expected_votes)) {
  1202. if (!runtime) {
  1203. error = (char *)"configuration error: quorum.device.votes must be specified when not all nodes votes 1";
  1204. goto out;
  1205. } else {
  1206. log_printf(LOGSYS_LEVEL_CRIT, "configuration error: quorum.device.votes must be specified when not all nodes votes 1");
  1207. log_printf(LOGSYS_LEVEL_CRIT, "disabling quorum device operations");
  1208. update_qdevice_can_operate(0);
  1209. }
  1210. }
  1211. /*
  1212. * validate quorum device votes vs expected_votes
  1213. */
  1214. if ((qdevice_votes > 0) && (expected_votes)) {
  1215. int delta = expected_votes - qdevice_votes;
  1216. if (delta < 2) {
  1217. if (!runtime) {
  1218. error = (char *)"configuration error: quorum.device.votes is too high or expected_votes is too low";
  1219. goto out;
  1220. } else {
  1221. log_printf(LOGSYS_LEVEL_CRIT, "configuration error: quorum.device.votes is too high or expected_votes is too low");
  1222. log_printf(LOGSYS_LEVEL_CRIT, "disabling quorum device operations");
  1223. update_qdevice_can_operate(0);
  1224. }
  1225. }
  1226. }
  1227. /*
  1228. * automatically calculate device votes and adjust expected_votes from nodelist
  1229. */
  1230. if ((have_qdevice) &&
  1231. (qdevice_votes == -1) &&
  1232. (!expected_votes) &&
  1233. (have_nodelist) &&
  1234. (node_count == node_expected_votes)) {
  1235. qdevice_votes = node_expected_votes - 1;
  1236. node_expected_votes = node_expected_votes + qdevice_votes;
  1237. }
  1238. /*
  1239. * set this node votes and expected_votes
  1240. */
  1241. log_printf(LOGSYS_LEVEL_DEBUG, "ev_tracking=%d, ev_tracking_barrier = %d: expected_votes = %d\n", ev_tracking, ev_tracking_barrier, expected_votes);
  1242. if (ev_tracking) {
  1243. expected_votes = ev_tracking_barrier;
  1244. }
  1245. if (have_nodelist) {
  1246. us->votes = node_votes;
  1247. us->expected_votes = node_expected_votes;
  1248. } else {
  1249. us->votes = 1;
  1250. icmap_get_uint32("quorum.votes", &us->votes);
  1251. }
  1252. if (expected_votes) {
  1253. us->expected_votes = expected_votes;
  1254. }
  1255. /*
  1256. * set qdevice votes
  1257. */
  1258. if (!have_qdevice) {
  1259. qdevice->votes = 0;
  1260. }
  1261. if (qdevice_votes != -1) {
  1262. qdevice->votes = qdevice_votes;
  1263. }
  1264. update_ev_barrier(us->expected_votes);
  1265. update_two_node();
  1266. if (wait_for_all) {
  1267. update_wait_for_all_status(1);
  1268. }
  1269. out:
  1270. LEAVE();
  1271. return error;
  1272. }
  1273. static void votequorum_refresh_config(
  1274. int32_t event,
  1275. const char *key_name,
  1276. struct icmap_notify_value new_val,
  1277. struct icmap_notify_value old_val,
  1278. void *user_data)
  1279. {
  1280. int old_votes, old_expected_votes;
  1281. uint8_t reloading;
  1282. uint8_t cancel_wfa;
  1283. ENTER();
  1284. /*
  1285. * If a full reload is in progress then don't do anything until it's done and
  1286. * can reconfigure it all atomically
  1287. */
  1288. if (icmap_get_uint8("config.totemconfig_reload_in_progress", &reloading) == CS_OK && reloading) {
  1289. return ;
  1290. }
  1291. icmap_get_uint8("quorum.cancel_wait_for_all", &cancel_wfa);
  1292. if (strcmp(key_name, "quorum.cancel_wait_for_all") == 0 &&
  1293. cancel_wfa >= 1) {
  1294. icmap_set_uint8("quorum.cancel_wait_for_all", 0);
  1295. votequorum_exec_send_reconfigure(VOTEQUORUM_RECONFIG_PARAM_CANCEL_WFA,
  1296. us->node_id, 0);
  1297. return;
  1298. }
  1299. old_votes = us->votes;
  1300. old_expected_votes = us->expected_votes;
  1301. /*
  1302. * Reload the configuration
  1303. */
  1304. votequorum_readconfig(VOTEQUORUM_READCONFIG_RUNTIME);
  1305. /*
  1306. * activate new config
  1307. */
  1308. votequorum_exec_send_nodeinfo(us->node_id);
  1309. votequorum_exec_send_nodeinfo(VOTEQUORUM_QDEVICE_NODEID);
  1310. if (us->votes != old_votes) {
  1311. votequorum_exec_send_reconfigure(VOTEQUORUM_RECONFIG_PARAM_NODE_VOTES,
  1312. us->node_id, us->votes);
  1313. }
  1314. if (us->expected_votes != old_expected_votes) {
  1315. votequorum_exec_send_reconfigure(VOTEQUORUM_RECONFIG_PARAM_EXPECTED_VOTES,
  1316. us->node_id, us->expected_votes);
  1317. }
  1318. LEAVE();
  1319. }
  1320. static void votequorum_exec_add_config_notification(void)
  1321. {
  1322. icmap_track_t icmap_track_nodelist = NULL;
  1323. icmap_track_t icmap_track_quorum = NULL;
  1324. icmap_track_t icmap_track_reload = NULL;
  1325. ENTER();
  1326. icmap_track_add("nodelist.",
  1327. ICMAP_TRACK_ADD | ICMAP_TRACK_DELETE | ICMAP_TRACK_MODIFY | ICMAP_TRACK_PREFIX,
  1328. votequorum_refresh_config,
  1329. NULL,
  1330. &icmap_track_nodelist);
  1331. icmap_track_add("quorum.",
  1332. ICMAP_TRACK_ADD | ICMAP_TRACK_DELETE | ICMAP_TRACK_MODIFY | ICMAP_TRACK_PREFIX,
  1333. votequorum_refresh_config,
  1334. NULL,
  1335. &icmap_track_quorum);
  1336. icmap_track_add("config.totemconfig_reload_in_progress",
  1337. ICMAP_TRACK_ADD | ICMAP_TRACK_MODIFY,
  1338. votequorum_refresh_config,
  1339. NULL,
  1340. &icmap_track_reload);
  1341. LEAVE();
  1342. }
  1343. /*
  1344. * votequorum_exec core
  1345. */
  1346. static int votequorum_exec_send_reconfigure(uint8_t param, unsigned int nodeid, uint32_t value)
  1347. {
  1348. struct req_exec_quorum_reconfigure req_exec_quorum_reconfigure;
  1349. struct iovec iov[1];
  1350. int ret;
  1351. ENTER();
  1352. req_exec_quorum_reconfigure.nodeid = nodeid;
  1353. req_exec_quorum_reconfigure.value = value;
  1354. req_exec_quorum_reconfigure.param = param;
  1355. req_exec_quorum_reconfigure._pad0 = 0;
  1356. req_exec_quorum_reconfigure._pad1 = 0;
  1357. req_exec_quorum_reconfigure._pad2 = 0;
  1358. req_exec_quorum_reconfigure.header.id = SERVICE_ID_MAKE(VOTEQUORUM_SERVICE, MESSAGE_REQ_EXEC_VOTEQUORUM_RECONFIGURE);
  1359. req_exec_quorum_reconfigure.header.size = sizeof(req_exec_quorum_reconfigure);
  1360. iov[0].iov_base = (void *)&req_exec_quorum_reconfigure;
  1361. iov[0].iov_len = sizeof(req_exec_quorum_reconfigure);
  1362. ret = corosync_api->totem_mcast (iov, 1, TOTEM_AGREED);
  1363. LEAVE();
  1364. return ret;
  1365. }
  1366. static int votequorum_exec_send_nodeinfo(uint32_t nodeid)
  1367. {
  1368. struct req_exec_quorum_nodeinfo req_exec_quorum_nodeinfo;
  1369. struct iovec iov[1];
  1370. struct cluster_node *node;
  1371. int ret;
  1372. ENTER();
  1373. node = find_node_by_nodeid(nodeid);
  1374. if (!node) {
  1375. return -1;
  1376. }
  1377. req_exec_quorum_nodeinfo.nodeid = nodeid;
  1378. req_exec_quorum_nodeinfo.votes = node->votes;
  1379. req_exec_quorum_nodeinfo.expected_votes = node->expected_votes;
  1380. req_exec_quorum_nodeinfo.flags = node->flags;
  1381. if (nodeid != VOTEQUORUM_QDEVICE_NODEID) {
  1382. decode_flags(node->flags);
  1383. }
  1384. req_exec_quorum_nodeinfo.header.id = SERVICE_ID_MAKE(VOTEQUORUM_SERVICE, MESSAGE_REQ_EXEC_VOTEQUORUM_NODEINFO);
  1385. req_exec_quorum_nodeinfo.header.size = sizeof(req_exec_quorum_nodeinfo);
  1386. iov[0].iov_base = (void *)&req_exec_quorum_nodeinfo;
  1387. iov[0].iov_len = sizeof(req_exec_quorum_nodeinfo);
  1388. ret = corosync_api->totem_mcast (iov, 1, TOTEM_AGREED);
  1389. LEAVE();
  1390. return ret;
  1391. }
  1392. static int votequorum_exec_send_qdevice_reconfigure(const char *oldname, const char *newname)
  1393. {
  1394. struct req_exec_quorum_qdevice_reconfigure req_exec_quorum_qdevice_reconfigure;
  1395. struct iovec iov[1];
  1396. int ret;
  1397. ENTER();
  1398. req_exec_quorum_qdevice_reconfigure.header.id = SERVICE_ID_MAKE(VOTEQUORUM_SERVICE, MESSAGE_REQ_EXEC_VOTEQUORUM_QDEVICE_RECONFIGURE);
  1399. req_exec_quorum_qdevice_reconfigure.header.size = sizeof(req_exec_quorum_qdevice_reconfigure);
  1400. strcpy(req_exec_quorum_qdevice_reconfigure.oldname, oldname);
  1401. strcpy(req_exec_quorum_qdevice_reconfigure.newname, newname);
  1402. iov[0].iov_base = (void *)&req_exec_quorum_qdevice_reconfigure;
  1403. iov[0].iov_len = sizeof(req_exec_quorum_qdevice_reconfigure);
  1404. ret = corosync_api->totem_mcast (iov, 1, TOTEM_AGREED);
  1405. LEAVE();
  1406. return ret;
  1407. }
  1408. static int votequorum_exec_send_qdevice_reg(uint32_t operation, const char *qdevice_name_req)
  1409. {
  1410. struct req_exec_quorum_qdevice_reg req_exec_quorum_qdevice_reg;
  1411. struct iovec iov[1];
  1412. int ret;
  1413. ENTER();
  1414. req_exec_quorum_qdevice_reg.header.id = SERVICE_ID_MAKE(VOTEQUORUM_SERVICE, MESSAGE_REQ_EXEC_VOTEQUORUM_QDEVICE_REG);
  1415. req_exec_quorum_qdevice_reg.header.size = sizeof(req_exec_quorum_qdevice_reg);
  1416. req_exec_quorum_qdevice_reg.operation = operation;
  1417. strcpy(req_exec_quorum_qdevice_reg.qdevice_name, qdevice_name_req);
  1418. iov[0].iov_base = (void *)&req_exec_quorum_qdevice_reg;
  1419. iov[0].iov_len = sizeof(req_exec_quorum_qdevice_reg);
  1420. ret = corosync_api->totem_mcast (iov, 1, TOTEM_AGREED);
  1421. LEAVE();
  1422. return ret;
  1423. }
  1424. static int votequorum_exec_send_quorum_notification(void *conn, uint64_t context)
  1425. {
  1426. struct res_lib_votequorum_notification *res_lib_votequorum_notification;
  1427. struct list_head *tmp;
  1428. struct cluster_node *node;
  1429. int cluster_members = 0;
  1430. int i = 0;
  1431. int size;
  1432. char buf[sizeof(struct res_lib_votequorum_notification) + sizeof(struct votequorum_node) * (PROCESSOR_COUNT_MAX + 2)];
  1433. ENTER();
  1434. list_iterate(tmp, &cluster_members_list) {
  1435. node = list_entry(tmp, struct cluster_node, list);
  1436. cluster_members++;
  1437. }
  1438. if (us->flags & NODE_FLAGS_QDEVICE_REGISTERED) {
  1439. cluster_members++;
  1440. }
  1441. size = sizeof(struct res_lib_votequorum_notification) + sizeof(struct votequorum_node) * cluster_members;
  1442. res_lib_votequorum_notification = (struct res_lib_votequorum_notification *)&buf;
  1443. res_lib_votequorum_notification->quorate = cluster_is_quorate;
  1444. res_lib_votequorum_notification->node_list_entries = cluster_members;
  1445. res_lib_votequorum_notification->ring_id.nodeid = quorum_ringid.rep.nodeid;
  1446. res_lib_votequorum_notification->ring_id.seq = quorum_ringid.seq;
  1447. res_lib_votequorum_notification->context = context;
  1448. list_iterate(tmp, &cluster_members_list) {
  1449. node = list_entry(tmp, struct cluster_node, list);
  1450. res_lib_votequorum_notification->node_list[i].nodeid = node->node_id;
  1451. res_lib_votequorum_notification->node_list[i++].state = node->state;
  1452. }
  1453. if (us->flags & NODE_FLAGS_QDEVICE_REGISTERED) {
  1454. res_lib_votequorum_notification->node_list[i].nodeid = VOTEQUORUM_QDEVICE_NODEID;
  1455. res_lib_votequorum_notification->node_list[i++].state = qdevice->state;
  1456. }
  1457. res_lib_votequorum_notification->header.id = MESSAGE_RES_VOTEQUORUM_NOTIFICATION;
  1458. res_lib_votequorum_notification->header.size = size;
  1459. res_lib_votequorum_notification->header.error = CS_OK;
  1460. /* Send it to all interested parties */
  1461. if (conn) {
  1462. int ret = corosync_api->ipc_dispatch_send(conn, &buf, size);
  1463. LEAVE();
  1464. return ret;
  1465. } else {
  1466. struct quorum_pd *qpd;
  1467. list_iterate(tmp, &trackers_list) {
  1468. qpd = list_entry(tmp, struct quorum_pd, list);
  1469. res_lib_votequorum_notification->context = qpd->tracking_context;
  1470. corosync_api->ipc_dispatch_send(qpd->conn, &buf, size);
  1471. }
  1472. }
  1473. LEAVE();
  1474. return 0;
  1475. }
  1476. static void votequorum_exec_send_expectedvotes_notification(void)
  1477. {
  1478. struct res_lib_votequorum_expectedvotes_notification res_lib_votequorum_expectedvotes_notification;
  1479. struct quorum_pd *qpd;
  1480. struct list_head *tmp;
  1481. ENTER();
  1482. log_printf(LOGSYS_LEVEL_DEBUG, "Sending expected votes callback");
  1483. res_lib_votequorum_expectedvotes_notification.header.id = MESSAGE_RES_VOTEQUORUM_EXPECTEDVOTES_NOTIFICATION;
  1484. res_lib_votequorum_expectedvotes_notification.header.size = sizeof(res_lib_votequorum_expectedvotes_notification);
  1485. res_lib_votequorum_expectedvotes_notification.header.error = CS_OK;
  1486. res_lib_votequorum_expectedvotes_notification.expected_votes = us->expected_votes;
  1487. list_iterate(tmp, &trackers_list) {
  1488. qpd = list_entry(tmp, struct quorum_pd, list);
  1489. res_lib_votequorum_expectedvotes_notification.context = qpd->tracking_context;
  1490. corosync_api->ipc_dispatch_send(qpd->conn, &res_lib_votequorum_expectedvotes_notification,
  1491. sizeof(struct res_lib_votequorum_expectedvotes_notification));
  1492. }
  1493. LEAVE();
  1494. }
  1495. static void exec_votequorum_qdevice_reconfigure_endian_convert (void *message)
  1496. {
  1497. ENTER();
  1498. LEAVE();
  1499. }
  1500. static void message_handler_req_exec_votequorum_qdevice_reconfigure (
  1501. const void *message,
  1502. unsigned int nodeid)
  1503. {
  1504. const struct req_exec_quorum_qdevice_reconfigure *req_exec_quorum_qdevice_reconfigure = message;
  1505. ENTER();
  1506. log_printf(LOGSYS_LEVEL_DEBUG, "Received qdevice name change req from node %u [from: %s to: %s]",
  1507. nodeid,
  1508. req_exec_quorum_qdevice_reconfigure->oldname,
  1509. req_exec_quorum_qdevice_reconfigure->newname);
  1510. if (!strcmp(req_exec_quorum_qdevice_reconfigure->oldname, qdevice_name)) {
  1511. log_printf(LOGSYS_LEVEL_DEBUG, "Allowing qdevice rename");
  1512. memset(qdevice_name, 0, VOTEQUORUM_QDEVICE_MAX_NAME_LEN);
  1513. strcpy(qdevice_name, req_exec_quorum_qdevice_reconfigure->newname);
  1514. /*
  1515. * TODO: notify qdevices about name change?
  1516. * this is not relevant for now and can wait later on since
  1517. * qdevices are local only and libvotequorum is not final
  1518. */
  1519. }
  1520. LEAVE();
  1521. }
  1522. static void exec_votequorum_qdevice_reg_endian_convert (void *message)
  1523. {
  1524. struct req_exec_quorum_qdevice_reg *req_exec_quorum_qdevice_reg = message;
  1525. ENTER();
  1526. req_exec_quorum_qdevice_reg->operation = swab32(req_exec_quorum_qdevice_reg->operation);
  1527. LEAVE();
  1528. }
  1529. static void message_handler_req_exec_votequorum_qdevice_reg (
  1530. const void *message,
  1531. unsigned int nodeid)
  1532. {
  1533. const struct req_exec_quorum_qdevice_reg *req_exec_quorum_qdevice_reg = message;
  1534. struct res_lib_votequorum_status res_lib_votequorum_status;
  1535. int wipe_qdevice_name = 1;
  1536. struct cluster_node *node = NULL;
  1537. struct list_head *tmp;
  1538. cs_error_t error = CS_OK;
  1539. ENTER();
  1540. log_printf(LOGSYS_LEVEL_DEBUG, "Received qdevice op %u req from node %u [%s]",
  1541. req_exec_quorum_qdevice_reg->operation,
  1542. nodeid, req_exec_quorum_qdevice_reg->qdevice_name);
  1543. switch(req_exec_quorum_qdevice_reg->operation)
  1544. {
  1545. case VOTEQUORUM_QDEVICE_OPERATION_REGISTER:
  1546. if (nodeid != us->node_id) {
  1547. if (!strlen(qdevice_name)) {
  1548. log_printf(LOGSYS_LEVEL_DEBUG, "Remote qdevice name recorded");
  1549. strcpy(qdevice_name, req_exec_quorum_qdevice_reg->qdevice_name);
  1550. }
  1551. LEAVE();
  1552. return;
  1553. }
  1554. /*
  1555. * protect against the case where we broadcast qdevice registration
  1556. * to new memebers, we receive the message back, but there is no registration
  1557. * connection in progress
  1558. */
  1559. if (us->flags & NODE_FLAGS_QDEVICE_REGISTERED) {
  1560. LEAVE();
  1561. return;
  1562. }
  1563. /*
  1564. * this should NEVER happen
  1565. */
  1566. if (!qdevice_reg_conn) {
  1567. log_printf(LOGSYS_LEVEL_WARNING, "Unable to determine origin of the qdevice register call!");
  1568. LEAVE();
  1569. return;
  1570. }
  1571. /*
  1572. * registering our own device in this case
  1573. */
  1574. if (!strlen(qdevice_name)) {
  1575. strcpy(qdevice_name, req_exec_quorum_qdevice_reg->qdevice_name);
  1576. }
  1577. /*
  1578. * check if it is our device or something else
  1579. */
  1580. if ((!strncmp(req_exec_quorum_qdevice_reg->qdevice_name,
  1581. qdevice_name, VOTEQUORUM_QDEVICE_MAX_NAME_LEN))) {
  1582. us->flags |= NODE_FLAGS_QDEVICE_REGISTERED;
  1583. votequorum_exec_send_nodeinfo(VOTEQUORUM_QDEVICE_NODEID);
  1584. votequorum_exec_send_nodeinfo(us->node_id);
  1585. } else {
  1586. log_printf(LOGSYS_LEVEL_WARNING,
  1587. "A new qdevice with different name (new: %s old: %s) is trying to register!",
  1588. req_exec_quorum_qdevice_reg->qdevice_name, qdevice_name);
  1589. error = CS_ERR_EXIST;
  1590. }
  1591. res_lib_votequorum_status.header.size = sizeof(res_lib_votequorum_status);
  1592. res_lib_votequorum_status.header.id = MESSAGE_RES_VOTEQUORUM_STATUS;
  1593. res_lib_votequorum_status.header.error = error;
  1594. corosync_api->ipc_response_send(qdevice_reg_conn, &res_lib_votequorum_status, sizeof(res_lib_votequorum_status));
  1595. qdevice_reg_conn = NULL;
  1596. break;
  1597. case VOTEQUORUM_QDEVICE_OPERATION_UNREGISTER:
  1598. list_iterate(tmp, &cluster_members_list) {
  1599. node = list_entry(tmp, struct cluster_node, list);
  1600. if ((node->state == NODESTATE_MEMBER) &&
  1601. (node->flags & NODE_FLAGS_QDEVICE_REGISTERED)) {
  1602. wipe_qdevice_name = 0;
  1603. }
  1604. }
  1605. if (wipe_qdevice_name) {
  1606. memset(qdevice_name, 0, VOTEQUORUM_QDEVICE_MAX_NAME_LEN);
  1607. }
  1608. break;
  1609. }
  1610. LEAVE();
  1611. }
  1612. static void exec_votequorum_nodeinfo_endian_convert (void *message)
  1613. {
  1614. struct req_exec_quorum_nodeinfo *nodeinfo = message;
  1615. ENTER();
  1616. nodeinfo->nodeid = swab32(nodeinfo->nodeid);
  1617. nodeinfo->votes = swab32(nodeinfo->votes);
  1618. nodeinfo->expected_votes = swab32(nodeinfo->expected_votes);
  1619. nodeinfo->flags = swab32(nodeinfo->flags);
  1620. LEAVE();
  1621. }
  1622. static void message_handler_req_exec_votequorum_nodeinfo (
  1623. const void *message,
  1624. unsigned int sender_nodeid)
  1625. {
  1626. const struct req_exec_quorum_nodeinfo *req_exec_quorum_nodeinfo = message;
  1627. struct cluster_node *node = NULL;
  1628. int old_votes;
  1629. int old_expected;
  1630. uint32_t old_flags;
  1631. nodestate_t old_state;
  1632. int new_node = 0;
  1633. int allow_downgrade = 0;
  1634. int by_node = 0;
  1635. unsigned int nodeid = req_exec_quorum_nodeinfo->nodeid;
  1636. ENTER();
  1637. log_printf(LOGSYS_LEVEL_DEBUG, "got nodeinfo message from cluster node %u", sender_nodeid);
  1638. log_printf(LOGSYS_LEVEL_DEBUG, "nodeinfo message[%u]: votes: %d, expected: %d flags: %d",
  1639. nodeid,
  1640. req_exec_quorum_nodeinfo->votes,
  1641. req_exec_quorum_nodeinfo->expected_votes,
  1642. req_exec_quorum_nodeinfo->flags);
  1643. if (nodeid != VOTEQUORUM_QDEVICE_NODEID) {
  1644. decode_flags(req_exec_quorum_nodeinfo->flags);
  1645. }
  1646. node = find_node_by_nodeid(nodeid);
  1647. if (!node) {
  1648. node = allocate_node(nodeid);
  1649. new_node = 1;
  1650. }
  1651. if (!node) {
  1652. corosync_api->error_memory_failure();
  1653. LEAVE();
  1654. return;
  1655. }
  1656. if (new_node) {
  1657. old_votes = 0;
  1658. old_expected = 0;
  1659. old_state = NODESTATE_DEAD;
  1660. old_flags = 0;
  1661. } else {
  1662. old_votes = node->votes;
  1663. old_expected = node->expected_votes;
  1664. old_state = node->state;
  1665. old_flags = node->flags;
  1666. }
  1667. if (nodeid == VOTEQUORUM_QDEVICE_NODEID) {
  1668. struct cluster_node *sender_node = find_node_by_nodeid(sender_nodeid);
  1669. assert(sender_node != NULL);
  1670. if ((!cluster_is_quorate) &&
  1671. (sender_node->flags & NODE_FLAGS_QUORATE)) {
  1672. node->votes = req_exec_quorum_nodeinfo->votes;
  1673. } else {
  1674. node->votes = max(node->votes, req_exec_quorum_nodeinfo->votes);
  1675. }
  1676. goto recalculate;
  1677. }
  1678. /* Update node state */
  1679. node->flags = req_exec_quorum_nodeinfo->flags;
  1680. node->votes = req_exec_quorum_nodeinfo->votes;
  1681. node->state = NODESTATE_MEMBER;
  1682. if (node->flags & NODE_FLAGS_LEAVING) {
  1683. node->state = NODESTATE_LEAVING;
  1684. allow_downgrade = 1;
  1685. by_node = 1;
  1686. }
  1687. if ((!cluster_is_quorate) &&
  1688. (node->flags & NODE_FLAGS_QUORATE)) {
  1689. allow_downgrade = 1;
  1690. us->expected_votes = req_exec_quorum_nodeinfo->expected_votes;
  1691. }
  1692. if (node->flags & NODE_FLAGS_QUORATE || (ev_tracking)) {
  1693. node->expected_votes = req_exec_quorum_nodeinfo->expected_votes;
  1694. } else {
  1695. node->expected_votes = us->expected_votes;
  1696. }
  1697. if ((last_man_standing) && (node->votes > 1)) {
  1698. log_printf(LOGSYS_LEVEL_WARNING, "Last Man Standing feature is supported only when all"
  1699. "cluster nodes votes are set to 1. Disabling LMS.");
  1700. last_man_standing = 0;
  1701. if (last_man_standing_timer_set) {
  1702. corosync_api->timer_delete(last_man_standing_timer);
  1703. last_man_standing_timer_set = 0;
  1704. }
  1705. }
  1706. recalculate:
  1707. if ((new_node) ||
  1708. (nodeid == us->node_id) ||
  1709. (node->flags & NODE_FLAGS_FIRST) ||
  1710. (old_votes != node->votes) ||
  1711. (old_expected != node->expected_votes) ||
  1712. (old_flags != node->flags) ||
  1713. (old_state != node->state)) {
  1714. recalculate_quorum(allow_downgrade, by_node);
  1715. }
  1716. if ((wait_for_all) &&
  1717. (!(node->flags & NODE_FLAGS_WFASTATUS)) &&
  1718. (node->flags & NODE_FLAGS_QUORATE)) {
  1719. update_wait_for_all_status(0);
  1720. }
  1721. LEAVE();
  1722. }
  1723. static void exec_votequorum_reconfigure_endian_convert (void *message)
  1724. {
  1725. struct req_exec_quorum_reconfigure *reconfigure = message;
  1726. ENTER();
  1727. reconfigure->nodeid = swab32(reconfigure->nodeid);
  1728. reconfigure->value = swab32(reconfigure->value);
  1729. LEAVE();
  1730. }
  1731. static void message_handler_req_exec_votequorum_reconfigure (
  1732. const void *message,
  1733. unsigned int nodeid)
  1734. {
  1735. const struct req_exec_quorum_reconfigure *req_exec_quorum_reconfigure = message;
  1736. struct cluster_node *node;
  1737. struct list_head *nodelist;
  1738. ENTER();
  1739. log_printf(LOGSYS_LEVEL_DEBUG, "got reconfigure message from cluster node %u for %u",
  1740. nodeid, req_exec_quorum_reconfigure->nodeid);
  1741. switch(req_exec_quorum_reconfigure->param)
  1742. {
  1743. case VOTEQUORUM_RECONFIG_PARAM_EXPECTED_VOTES:
  1744. list_iterate(nodelist, &cluster_members_list) {
  1745. node = list_entry(nodelist, struct cluster_node, list);
  1746. if (node->state == NODESTATE_MEMBER) {
  1747. node->expected_votes = req_exec_quorum_reconfigure->value;
  1748. }
  1749. }
  1750. votequorum_exec_send_expectedvotes_notification();
  1751. update_ev_barrier(req_exec_quorum_reconfigure->value);
  1752. if (ev_tracking) {
  1753. us->expected_votes = max(us->expected_votes, ev_tracking_barrier);
  1754. }
  1755. recalculate_quorum(1, 0); /* Allow decrease */
  1756. break;
  1757. case VOTEQUORUM_RECONFIG_PARAM_NODE_VOTES:
  1758. node = find_node_by_nodeid(req_exec_quorum_reconfigure->nodeid);
  1759. if (!node) {
  1760. LEAVE();
  1761. return;
  1762. }
  1763. node->votes = req_exec_quorum_reconfigure->value;
  1764. recalculate_quorum(1, 0); /* Allow decrease */
  1765. break;
  1766. case VOTEQUORUM_RECONFIG_PARAM_CANCEL_WFA:
  1767. update_wait_for_all_status(0);
  1768. log_printf(LOGSYS_LEVEL_INFO, "wait_for_all_status reset by user on node %d.",
  1769. req_exec_quorum_reconfigure->nodeid);
  1770. recalculate_quorum(0, 0);
  1771. break;
  1772. }
  1773. LEAVE();
  1774. }
  1775. static int votequorum_exec_exit_fn (void)
  1776. {
  1777. int ret = 0;
  1778. ENTER();
  1779. /*
  1780. * tell the other nodes we are leaving
  1781. */
  1782. if (allow_downscale) {
  1783. us->flags |= NODE_FLAGS_LEAVING;
  1784. ret = votequorum_exec_send_nodeinfo(us->node_id);
  1785. }
  1786. if ((ev_tracking) && (ev_tracking_fd != -1)) {
  1787. close(ev_tracking_fd);
  1788. }
  1789. LEAVE();
  1790. return ret;
  1791. }
  1792. static char *votequorum_exec_init_fn (struct corosync_api_v1 *api)
  1793. {
  1794. char *error = NULL;
  1795. ENTER();
  1796. /*
  1797. * make sure we start clean
  1798. */
  1799. list_init(&cluster_members_list);
  1800. list_init(&trackers_list);
  1801. qdevice = NULL;
  1802. us = NULL;
  1803. memset(cluster_nodes, 0, sizeof(cluster_nodes));
  1804. /*
  1805. * Allocate a cluster_node for qdevice
  1806. */
  1807. qdevice = allocate_node(VOTEQUORUM_QDEVICE_NODEID);
  1808. if (!qdevice) {
  1809. LEAVE();
  1810. return ((char *)"Could not allocate node.");
  1811. }
  1812. qdevice->votes = 0;
  1813. memset(qdevice_name, 0, VOTEQUORUM_QDEVICE_MAX_NAME_LEN);
  1814. /*
  1815. * Allocate a cluster_node for us
  1816. */
  1817. us = allocate_node(corosync_api->totem_nodeid_get());
  1818. if (!us) {
  1819. LEAVE();
  1820. return ((char *)"Could not allocate node.");
  1821. }
  1822. icmap_set_uint32("runtime.votequorum.this_node_id", us->node_id);
  1823. us->state = NODESTATE_MEMBER;
  1824. us->votes = 1;
  1825. us->flags |= NODE_FLAGS_FIRST;
  1826. error = votequorum_readconfig(VOTEQUORUM_READCONFIG_STARTUP);
  1827. if (error) {
  1828. return error;
  1829. }
  1830. recalculate_quorum(0, 0);
  1831. /*
  1832. * Listen for changes
  1833. */
  1834. votequorum_exec_add_config_notification();
  1835. /*
  1836. * Start us off with one node
  1837. */
  1838. votequorum_exec_send_nodeinfo(us->node_id);
  1839. LEAVE();
  1840. return (NULL);
  1841. }
  1842. /*
  1843. * votequorum service core
  1844. */
  1845. static void votequorum_last_man_standing_timer_fn(void *arg)
  1846. {
  1847. ENTER();
  1848. last_man_standing_timer_set = 0;
  1849. if (cluster_is_quorate) {
  1850. recalculate_quorum(1,1);
  1851. }
  1852. LEAVE();
  1853. }
  1854. static void votequorum_sync_init (
  1855. const unsigned int *trans_list, size_t trans_list_entries,
  1856. const unsigned int *member_list, size_t member_list_entries,
  1857. const struct memb_ring_id *ring_id)
  1858. {
  1859. int i, j;
  1860. int found;
  1861. int left_nodes;
  1862. struct cluster_node *node;
  1863. ENTER();
  1864. sync_in_progress = 1;
  1865. sync_nodeinfo_sent = 0;
  1866. sync_wait_for_poll_or_timeout = 0;
  1867. if (member_list_entries > 1) {
  1868. us->flags &= ~NODE_FLAGS_FIRST;
  1869. }
  1870. /*
  1871. * we don't need to track which nodes have left directly,
  1872. * since that info is in the node db, but we need to know
  1873. * if somebody has left for last_man_standing
  1874. */
  1875. left_nodes = 0;
  1876. for (i = 0; i < quorum_members_entries; i++) {
  1877. found = 0;
  1878. for (j = 0; j < member_list_entries; j++) {
  1879. if (quorum_members[i] == member_list[j]) {
  1880. found = 1;
  1881. break;
  1882. }
  1883. }
  1884. if (found == 0) {
  1885. left_nodes = 1;
  1886. node = find_node_by_nodeid(quorum_members[i]);
  1887. if (node) {
  1888. node->state = NODESTATE_DEAD;
  1889. }
  1890. }
  1891. }
  1892. if (last_man_standing) {
  1893. if (((member_list_entries >= quorum) && (left_nodes)) ||
  1894. ((member_list_entries <= quorum) && (auto_tie_breaker != ATB_NONE) && (check_low_node_id_partition() == 1))) {
  1895. if (last_man_standing_timer_set) {
  1896. corosync_api->timer_delete(last_man_standing_timer);
  1897. last_man_standing_timer_set = 0;
  1898. }
  1899. corosync_api->timer_add_duration((unsigned long long)last_man_standing_window*1000000,
  1900. NULL, votequorum_last_man_standing_timer_fn,
  1901. &last_man_standing_timer);
  1902. last_man_standing_timer_set = 1;
  1903. }
  1904. }
  1905. memcpy(previous_quorum_members, quorum_members, sizeof(unsigned int) * quorum_members_entries);
  1906. previous_quorum_members_entries = quorum_members_entries;
  1907. memcpy(quorum_members, member_list, sizeof(unsigned int) * member_list_entries);
  1908. quorum_members_entries = member_list_entries;
  1909. memcpy(&quorum_ringid, ring_id, sizeof(*ring_id));
  1910. if (us->flags & NODE_FLAGS_QDEVICE_REGISTERED && us->flags & NODE_FLAGS_QDEVICE_ALIVE) {
  1911. /*
  1912. * Reset poll timer. Sync waiting is interrupted on valid qdevice poll or after timeout
  1913. */
  1914. if (qdevice_timer_set) {
  1915. corosync_api->timer_delete(qdevice_timer);
  1916. }
  1917. corosync_api->timer_add_duration((unsigned long long)qdevice_sync_timeout*1000000, qdevice,
  1918. qdevice_timer_fn, &qdevice_timer);
  1919. qdevice_timer_set = 1;
  1920. sync_wait_for_poll_or_timeout = 1;
  1921. log_printf(LOGSYS_LEVEL_INFO, "waiting for quorum device %s poll (but maximum for %u ms)",
  1922. qdevice_name, qdevice_sync_timeout);
  1923. }
  1924. LEAVE();
  1925. }
  1926. static int votequorum_sync_process (void)
  1927. {
  1928. if (!sync_nodeinfo_sent) {
  1929. votequorum_exec_send_nodeinfo(us->node_id);
  1930. votequorum_exec_send_nodeinfo(VOTEQUORUM_QDEVICE_NODEID);
  1931. if (strlen(qdevice_name)) {
  1932. votequorum_exec_send_qdevice_reg(VOTEQUORUM_QDEVICE_OPERATION_REGISTER,
  1933. qdevice_name);
  1934. }
  1935. sync_nodeinfo_sent = 1;
  1936. }
  1937. if (us->flags & NODE_FLAGS_QDEVICE_REGISTERED && sync_wait_for_poll_or_timeout) {
  1938. /*
  1939. * Waiting for qdevice to poll with new ringid or timeout
  1940. */
  1941. return (-1);
  1942. }
  1943. return 0;
  1944. }
  1945. static void votequorum_sync_activate (void)
  1946. {
  1947. recalculate_quorum(0, 0);
  1948. quorum_callback(quorum_members, quorum_members_entries,
  1949. cluster_is_quorate, &quorum_ringid);
  1950. votequorum_exec_send_quorum_notification(NULL, 0L);
  1951. sync_in_progress = 0;
  1952. }
  1953. static void votequorum_sync_abort (void)
  1954. {
  1955. }
  1956. char *votequorum_init(struct corosync_api_v1 *api,
  1957. quorum_set_quorate_fn_t q_set_quorate_fn)
  1958. {
  1959. char *error;
  1960. ENTER();
  1961. if (q_set_quorate_fn == NULL) {
  1962. return ((char *)"Quorate function not set");
  1963. }
  1964. corosync_api = api;
  1965. quorum_callback = q_set_quorate_fn;
  1966. error = corosync_service_link_and_init(corosync_api,
  1967. &votequorum_service[0]);
  1968. if (error) {
  1969. return (error);
  1970. }
  1971. LEAVE();
  1972. return (NULL);
  1973. }
  1974. /*
  1975. * Library Handler init/fini
  1976. */
  1977. static int quorum_lib_init_fn (void *conn)
  1978. {
  1979. struct quorum_pd *pd = (struct quorum_pd *)corosync_api->ipc_private_data_get (conn);
  1980. ENTER();
  1981. list_init (&pd->list);
  1982. pd->conn = conn;
  1983. LEAVE();
  1984. return (0);
  1985. }
  1986. static int quorum_lib_exit_fn (void *conn)
  1987. {
  1988. struct quorum_pd *quorum_pd = (struct quorum_pd *)corosync_api->ipc_private_data_get (conn);
  1989. ENTER();
  1990. if (quorum_pd->tracking_enabled) {
  1991. list_del (&quorum_pd->list);
  1992. list_init (&quorum_pd->list);
  1993. }
  1994. LEAVE();
  1995. return (0);
  1996. }
  1997. /*
  1998. * library internal functions
  1999. */
  2000. static void qdevice_timer_fn(void *arg)
  2001. {
  2002. ENTER();
  2003. if ((!(us->flags & NODE_FLAGS_QDEVICE_ALIVE)) ||
  2004. (!qdevice_timer_set)) {
  2005. LEAVE();
  2006. return;
  2007. }
  2008. us->flags &= ~NODE_FLAGS_QDEVICE_ALIVE;
  2009. us->flags &= ~NODE_FLAGS_QDEVICE_CAST_VOTE;
  2010. log_printf(LOGSYS_LEVEL_INFO, "lost contact with quorum device %s", qdevice_name);
  2011. votequorum_exec_send_nodeinfo(us->node_id);
  2012. qdevice_timer_set = 0;
  2013. sync_wait_for_poll_or_timeout = 0;
  2014. LEAVE();
  2015. }
  2016. /*
  2017. * Library Handler Functions
  2018. */
  2019. static void message_handler_req_lib_votequorum_getinfo (void *conn, const void *message)
  2020. {
  2021. const struct req_lib_votequorum_getinfo *req_lib_votequorum_getinfo = message;
  2022. struct res_lib_votequorum_getinfo res_lib_votequorum_getinfo;
  2023. struct cluster_node *node;
  2024. unsigned int highest_expected = 0;
  2025. unsigned int total_votes = 0;
  2026. cs_error_t error = CS_OK;
  2027. uint32_t nodeid = req_lib_votequorum_getinfo->nodeid;
  2028. ENTER();
  2029. log_printf(LOGSYS_LEVEL_DEBUG, "got getinfo request on %p for node %u", conn, req_lib_votequorum_getinfo->nodeid);
  2030. if (nodeid == VOTEQUORUM_QDEVICE_NODEID) {
  2031. nodeid = us->node_id;
  2032. }
  2033. node = find_node_by_nodeid(nodeid);
  2034. if (node) {
  2035. struct cluster_node *iternode;
  2036. struct list_head *nodelist;
  2037. list_iterate(nodelist, &cluster_members_list) {
  2038. iternode = list_entry(nodelist, struct cluster_node, list);
  2039. if (iternode->state == NODESTATE_MEMBER) {
  2040. highest_expected =
  2041. max(highest_expected, iternode->expected_votes);
  2042. total_votes += iternode->votes;
  2043. }
  2044. }
  2045. if (node->flags & NODE_FLAGS_QDEVICE_CAST_VOTE) {
  2046. total_votes += qdevice->votes;
  2047. }
  2048. switch(node->state) {
  2049. case NODESTATE_MEMBER:
  2050. res_lib_votequorum_getinfo.state = VOTEQUORUM_NODESTATE_MEMBER;
  2051. break;
  2052. case NODESTATE_DEAD:
  2053. res_lib_votequorum_getinfo.state = VOTEQUORUM_NODESTATE_DEAD;
  2054. break;
  2055. case NODESTATE_LEAVING:
  2056. res_lib_votequorum_getinfo.state = VOTEQUORUM_NODESTATE_LEAVING;
  2057. break;
  2058. default:
  2059. res_lib_votequorum_getinfo.state = node->state;
  2060. break;
  2061. }
  2062. res_lib_votequorum_getinfo.state = node->state;
  2063. res_lib_votequorum_getinfo.votes = node->votes;
  2064. res_lib_votequorum_getinfo.expected_votes = node->expected_votes;
  2065. res_lib_votequorum_getinfo.highest_expected = highest_expected;
  2066. res_lib_votequorum_getinfo.quorum = quorum;
  2067. res_lib_votequorum_getinfo.total_votes = total_votes;
  2068. res_lib_votequorum_getinfo.flags = 0;
  2069. res_lib_votequorum_getinfo.nodeid = node->node_id;
  2070. if (two_node) {
  2071. res_lib_votequorum_getinfo.flags |= VOTEQUORUM_INFO_TWONODE;
  2072. }
  2073. if (cluster_is_quorate) {
  2074. res_lib_votequorum_getinfo.flags |= VOTEQUORUM_INFO_QUORATE;
  2075. }
  2076. if (wait_for_all) {
  2077. res_lib_votequorum_getinfo.flags |= VOTEQUORUM_INFO_WAIT_FOR_ALL;
  2078. }
  2079. if (last_man_standing) {
  2080. res_lib_votequorum_getinfo.flags |= VOTEQUORUM_INFO_LAST_MAN_STANDING;
  2081. }
  2082. if (auto_tie_breaker != ATB_NONE) {
  2083. res_lib_votequorum_getinfo.flags |= VOTEQUORUM_INFO_AUTO_TIE_BREAKER;
  2084. }
  2085. if (allow_downscale) {
  2086. res_lib_votequorum_getinfo.flags |= VOTEQUORUM_INFO_ALLOW_DOWNSCALE;
  2087. }
  2088. memset(res_lib_votequorum_getinfo.qdevice_name, 0, VOTEQUORUM_QDEVICE_MAX_NAME_LEN);
  2089. strcpy(res_lib_votequorum_getinfo.qdevice_name, qdevice_name);
  2090. res_lib_votequorum_getinfo.qdevice_votes = qdevice->votes;
  2091. if (node->flags & NODE_FLAGS_QDEVICE_REGISTERED) {
  2092. res_lib_votequorum_getinfo.flags |= VOTEQUORUM_INFO_QDEVICE_REGISTERED;
  2093. }
  2094. if (node->flags & NODE_FLAGS_QDEVICE_ALIVE) {
  2095. res_lib_votequorum_getinfo.flags |= VOTEQUORUM_INFO_QDEVICE_ALIVE;
  2096. }
  2097. if (node->flags & NODE_FLAGS_QDEVICE_CAST_VOTE) {
  2098. res_lib_votequorum_getinfo.flags |= VOTEQUORUM_INFO_QDEVICE_CAST_VOTE;
  2099. }
  2100. if (node->flags & NODE_FLAGS_QDEVICE_MASTER_WINS) {
  2101. res_lib_votequorum_getinfo.flags |= VOTEQUORUM_INFO_QDEVICE_MASTER_WINS;
  2102. }
  2103. } else {
  2104. error = CS_ERR_NOT_EXIST;
  2105. }
  2106. res_lib_votequorum_getinfo.header.size = sizeof(res_lib_votequorum_getinfo);
  2107. res_lib_votequorum_getinfo.header.id = MESSAGE_RES_VOTEQUORUM_GETINFO;
  2108. res_lib_votequorum_getinfo.header.error = error;
  2109. corosync_api->ipc_response_send(conn, &res_lib_votequorum_getinfo, sizeof(res_lib_votequorum_getinfo));
  2110. log_printf(LOGSYS_LEVEL_DEBUG, "getinfo response error: %d", error);
  2111. LEAVE();
  2112. }
  2113. static void message_handler_req_lib_votequorum_setexpected (void *conn, const void *message)
  2114. {
  2115. const struct req_lib_votequorum_setexpected *req_lib_votequorum_setexpected = message;
  2116. struct res_lib_votequorum_status res_lib_votequorum_status;
  2117. cs_error_t error = CS_OK;
  2118. unsigned int newquorum;
  2119. unsigned int total_votes;
  2120. uint8_t allow_downscale_status = 0;
  2121. ENTER();
  2122. allow_downscale_status = allow_downscale;
  2123. allow_downscale = 0;
  2124. /*
  2125. * Validate new expected votes
  2126. */
  2127. newquorum = calculate_quorum(1, req_lib_votequorum_setexpected->expected_votes, &total_votes);
  2128. allow_downscale = allow_downscale_status;
  2129. if (newquorum < total_votes / 2 ||
  2130. newquorum > total_votes) {
  2131. error = CS_ERR_INVALID_PARAM;
  2132. goto error_exit;
  2133. }
  2134. votequorum_exec_send_reconfigure(VOTEQUORUM_RECONFIG_PARAM_EXPECTED_VOTES, us->node_id,
  2135. req_lib_votequorum_setexpected->expected_votes);
  2136. error_exit:
  2137. res_lib_votequorum_status.header.size = sizeof(res_lib_votequorum_status);
  2138. res_lib_votequorum_status.header.id = MESSAGE_RES_VOTEQUORUM_STATUS;
  2139. res_lib_votequorum_status.header.error = error;
  2140. corosync_api->ipc_response_send(conn, &res_lib_votequorum_status, sizeof(res_lib_votequorum_status));
  2141. LEAVE();
  2142. }
  2143. static void message_handler_req_lib_votequorum_setvotes (void *conn, const void *message)
  2144. {
  2145. const struct req_lib_votequorum_setvotes *req_lib_votequorum_setvotes = message;
  2146. struct res_lib_votequorum_status res_lib_votequorum_status;
  2147. struct cluster_node *node;
  2148. unsigned int newquorum;
  2149. unsigned int total_votes;
  2150. unsigned int saved_votes;
  2151. cs_error_t error = CS_OK;
  2152. unsigned int nodeid;
  2153. ENTER();
  2154. nodeid = req_lib_votequorum_setvotes->nodeid;
  2155. node = find_node_by_nodeid(nodeid);
  2156. if (!node) {
  2157. error = CS_ERR_NAME_NOT_FOUND;
  2158. goto error_exit;
  2159. }
  2160. /*
  2161. * Check votes is valid
  2162. */
  2163. saved_votes = node->votes;
  2164. node->votes = req_lib_votequorum_setvotes->votes;
  2165. newquorum = calculate_quorum(1, 0, &total_votes);
  2166. if (newquorum < total_votes / 2 ||
  2167. newquorum > total_votes) {
  2168. node->votes = saved_votes;
  2169. error = CS_ERR_INVALID_PARAM;
  2170. goto error_exit;
  2171. }
  2172. votequorum_exec_send_reconfigure(VOTEQUORUM_RECONFIG_PARAM_NODE_VOTES, nodeid,
  2173. req_lib_votequorum_setvotes->votes);
  2174. error_exit:
  2175. res_lib_votequorum_status.header.size = sizeof(res_lib_votequorum_status);
  2176. res_lib_votequorum_status.header.id = MESSAGE_RES_VOTEQUORUM_STATUS;
  2177. res_lib_votequorum_status.header.error = error;
  2178. corosync_api->ipc_response_send(conn, &res_lib_votequorum_status, sizeof(res_lib_votequorum_status));
  2179. LEAVE();
  2180. }
  2181. static void message_handler_req_lib_votequorum_trackstart (void *conn,
  2182. const void *message)
  2183. {
  2184. const struct req_lib_votequorum_trackstart *req_lib_votequorum_trackstart = message;
  2185. struct res_lib_votequorum_status res_lib_votequorum_status;
  2186. struct quorum_pd *quorum_pd = (struct quorum_pd *)corosync_api->ipc_private_data_get (conn);
  2187. cs_error_t error = CS_OK;
  2188. ENTER();
  2189. /*
  2190. * If an immediate listing of the current cluster membership
  2191. * is requested, generate membership list
  2192. */
  2193. if (req_lib_votequorum_trackstart->track_flags & CS_TRACK_CURRENT ||
  2194. req_lib_votequorum_trackstart->track_flags & CS_TRACK_CHANGES) {
  2195. log_printf(LOGSYS_LEVEL_DEBUG, "sending initial status to %p", conn);
  2196. votequorum_exec_send_quorum_notification(conn, req_lib_votequorum_trackstart->context);
  2197. }
  2198. if (quorum_pd->tracking_enabled) {
  2199. error = CS_ERR_EXIST;
  2200. goto response_send;
  2201. }
  2202. /*
  2203. * Record requests for tracking
  2204. */
  2205. if (req_lib_votequorum_trackstart->track_flags & CS_TRACK_CHANGES ||
  2206. req_lib_votequorum_trackstart->track_flags & CS_TRACK_CHANGES_ONLY) {
  2207. quorum_pd->track_flags = req_lib_votequorum_trackstart->track_flags;
  2208. quorum_pd->tracking_enabled = 1;
  2209. quorum_pd->tracking_context = req_lib_votequorum_trackstart->context;
  2210. list_add (&quorum_pd->list, &trackers_list);
  2211. }
  2212. response_send:
  2213. res_lib_votequorum_status.header.size = sizeof(res_lib_votequorum_status);
  2214. res_lib_votequorum_status.header.id = MESSAGE_RES_VOTEQUORUM_STATUS;
  2215. res_lib_votequorum_status.header.error = error;
  2216. corosync_api->ipc_response_send(conn, &res_lib_votequorum_status, sizeof(res_lib_votequorum_status));
  2217. LEAVE();
  2218. }
  2219. static void message_handler_req_lib_votequorum_trackstop (void *conn,
  2220. const void *message)
  2221. {
  2222. struct res_lib_votequorum_status res_lib_votequorum_status;
  2223. struct quorum_pd *quorum_pd = (struct quorum_pd *)corosync_api->ipc_private_data_get (conn);
  2224. int error = CS_OK;
  2225. ENTER();
  2226. if (quorum_pd->tracking_enabled) {
  2227. error = CS_OK;
  2228. quorum_pd->tracking_enabled = 0;
  2229. list_del (&quorum_pd->list);
  2230. list_init (&quorum_pd->list);
  2231. } else {
  2232. error = CS_ERR_NOT_EXIST;
  2233. }
  2234. res_lib_votequorum_status.header.size = sizeof(res_lib_votequorum_status);
  2235. res_lib_votequorum_status.header.id = MESSAGE_RES_VOTEQUORUM_STATUS;
  2236. res_lib_votequorum_status.header.error = error;
  2237. corosync_api->ipc_response_send(conn, &res_lib_votequorum_status, sizeof(res_lib_votequorum_status));
  2238. LEAVE();
  2239. }
  2240. static void message_handler_req_lib_votequorum_qdevice_register (void *conn,
  2241. const void *message)
  2242. {
  2243. const struct req_lib_votequorum_qdevice_register *req_lib_votequorum_qdevice_register = message;
  2244. struct res_lib_votequorum_status res_lib_votequorum_status;
  2245. cs_error_t error = CS_OK;
  2246. ENTER();
  2247. if (!qdevice_can_operate) {
  2248. log_printf(LOGSYS_LEVEL_INFO, "Registration of quorum device is disabled by incorrect corosync.conf. See logs for more information");
  2249. error = CS_ERR_ACCESS;
  2250. goto out;
  2251. }
  2252. if (us->flags & NODE_FLAGS_QDEVICE_REGISTERED) {
  2253. if ((!strncmp(req_lib_votequorum_qdevice_register->name,
  2254. qdevice_name, VOTEQUORUM_QDEVICE_MAX_NAME_LEN))) {
  2255. goto out;
  2256. } else {
  2257. log_printf(LOGSYS_LEVEL_WARNING,
  2258. "A new qdevice with different name (new: %s old: %s) is trying to re-register!",
  2259. req_lib_votequorum_qdevice_register->name, qdevice_name);
  2260. error = CS_ERR_EXIST;
  2261. goto out;
  2262. }
  2263. } else {
  2264. if (qdevice_reg_conn != NULL) {
  2265. log_printf(LOGSYS_LEVEL_WARNING,
  2266. "Registration request already in progress");
  2267. error = CS_ERR_TRY_AGAIN;
  2268. goto out;
  2269. }
  2270. qdevice_reg_conn = conn;
  2271. if (votequorum_exec_send_qdevice_reg(VOTEQUORUM_QDEVICE_OPERATION_REGISTER,
  2272. req_lib_votequorum_qdevice_register->name) != 0) {
  2273. log_printf(LOGSYS_LEVEL_WARNING,
  2274. "Unable to send qdevice registration request to cluster");
  2275. error = CS_ERR_TRY_AGAIN;
  2276. qdevice_reg_conn = NULL;
  2277. } else {
  2278. LEAVE();
  2279. return;
  2280. }
  2281. }
  2282. out:
  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_qdevice_unregister (void *conn,
  2290. const void *message)
  2291. {
  2292. const struct req_lib_votequorum_qdevice_unregister *req_lib_votequorum_qdevice_unregister = message;
  2293. struct res_lib_votequorum_status res_lib_votequorum_status;
  2294. cs_error_t error = CS_OK;
  2295. ENTER();
  2296. if (us->flags & NODE_FLAGS_QDEVICE_REGISTERED) {
  2297. if (strncmp(req_lib_votequorum_qdevice_unregister->name, qdevice_name, VOTEQUORUM_QDEVICE_MAX_NAME_LEN)) {
  2298. error = CS_ERR_INVALID_PARAM;
  2299. goto out;
  2300. }
  2301. if (qdevice_timer_set) {
  2302. corosync_api->timer_delete(qdevice_timer);
  2303. qdevice_timer_set = 0;
  2304. sync_wait_for_poll_or_timeout = 0;
  2305. }
  2306. us->flags &= ~NODE_FLAGS_QDEVICE_REGISTERED;
  2307. us->flags &= ~NODE_FLAGS_QDEVICE_ALIVE;
  2308. us->flags &= ~NODE_FLAGS_QDEVICE_CAST_VOTE;
  2309. us->flags &= ~NODE_FLAGS_QDEVICE_MASTER_WINS;
  2310. votequorum_exec_send_nodeinfo(us->node_id);
  2311. votequorum_exec_send_qdevice_reg(VOTEQUORUM_QDEVICE_OPERATION_UNREGISTER,
  2312. req_lib_votequorum_qdevice_unregister->name);
  2313. } else {
  2314. error = CS_ERR_NOT_EXIST;
  2315. }
  2316. out:
  2317. res_lib_votequorum_status.header.size = sizeof(res_lib_votequorum_status);
  2318. res_lib_votequorum_status.header.id = MESSAGE_RES_VOTEQUORUM_STATUS;
  2319. res_lib_votequorum_status.header.error = error;
  2320. corosync_api->ipc_response_send(conn, &res_lib_votequorum_status, sizeof(res_lib_votequorum_status));
  2321. LEAVE();
  2322. }
  2323. static void message_handler_req_lib_votequorum_qdevice_update (void *conn,
  2324. const void *message)
  2325. {
  2326. const struct req_lib_votequorum_qdevice_update *req_lib_votequorum_qdevice_update = message;
  2327. struct res_lib_votequorum_status res_lib_votequorum_status;
  2328. cs_error_t error = CS_OK;
  2329. ENTER();
  2330. if (us->flags & NODE_FLAGS_QDEVICE_REGISTERED) {
  2331. if (strncmp(req_lib_votequorum_qdevice_update->oldname, qdevice_name, VOTEQUORUM_QDEVICE_MAX_NAME_LEN)) {
  2332. error = CS_ERR_INVALID_PARAM;
  2333. goto out;
  2334. }
  2335. votequorum_exec_send_qdevice_reconfigure(req_lib_votequorum_qdevice_update->oldname,
  2336. req_lib_votequorum_qdevice_update->newname);
  2337. } else {
  2338. error = CS_ERR_NOT_EXIST;
  2339. }
  2340. out:
  2341. res_lib_votequorum_status.header.size = sizeof(res_lib_votequorum_status);
  2342. res_lib_votequorum_status.header.id = MESSAGE_RES_VOTEQUORUM_STATUS;
  2343. res_lib_votequorum_status.header.error = error;
  2344. corosync_api->ipc_response_send(conn, &res_lib_votequorum_status, sizeof(res_lib_votequorum_status));
  2345. LEAVE();
  2346. }
  2347. static void message_handler_req_lib_votequorum_qdevice_poll (void *conn,
  2348. const void *message)
  2349. {
  2350. const struct req_lib_votequorum_qdevice_poll *req_lib_votequorum_qdevice_poll = message;
  2351. struct res_lib_votequorum_status res_lib_votequorum_status;
  2352. cs_error_t error = CS_OK;
  2353. uint32_t oldflags;
  2354. ENTER();
  2355. if (!qdevice_can_operate) {
  2356. error = CS_ERR_ACCESS;
  2357. goto out;
  2358. }
  2359. if (us->flags & NODE_FLAGS_QDEVICE_REGISTERED) {
  2360. if (!(req_lib_votequorum_qdevice_poll->ring_id.nodeid == quorum_ringid.rep.nodeid &&
  2361. req_lib_votequorum_qdevice_poll->ring_id.seq == quorum_ringid.seq)) {
  2362. log_printf(LOGSYS_LEVEL_DEBUG, "Received poll ring id (%u.%"PRIu64") != last sync "
  2363. "ring id (%u.%"PRIu64"). Ignoring poll call.",
  2364. req_lib_votequorum_qdevice_poll->ring_id.nodeid, req_lib_votequorum_qdevice_poll->ring_id.seq,
  2365. quorum_ringid.rep.nodeid, quorum_ringid.seq);
  2366. error = CS_ERR_MESSAGE_ERROR;
  2367. goto out;
  2368. }
  2369. if (strncmp(req_lib_votequorum_qdevice_poll->name, qdevice_name, VOTEQUORUM_QDEVICE_MAX_NAME_LEN)) {
  2370. error = CS_ERR_INVALID_PARAM;
  2371. goto out;
  2372. }
  2373. if (qdevice_timer_set) {
  2374. corosync_api->timer_delete(qdevice_timer);
  2375. qdevice_timer_set = 0;
  2376. }
  2377. oldflags = us->flags;
  2378. us->flags |= NODE_FLAGS_QDEVICE_ALIVE;
  2379. if (req_lib_votequorum_qdevice_poll->cast_vote) {
  2380. us->flags |= NODE_FLAGS_QDEVICE_CAST_VOTE;
  2381. } else {
  2382. us->flags &= ~NODE_FLAGS_QDEVICE_CAST_VOTE;
  2383. }
  2384. if (us->flags != oldflags) {
  2385. votequorum_exec_send_nodeinfo(us->node_id);
  2386. }
  2387. corosync_api->timer_add_duration((unsigned long long)qdevice_timeout*1000000, qdevice,
  2388. qdevice_timer_fn, &qdevice_timer);
  2389. qdevice_timer_set = 1;
  2390. sync_wait_for_poll_or_timeout = 0;
  2391. } else {
  2392. error = CS_ERR_NOT_EXIST;
  2393. }
  2394. out:
  2395. res_lib_votequorum_status.header.size = sizeof(res_lib_votequorum_status);
  2396. res_lib_votequorum_status.header.id = MESSAGE_RES_VOTEQUORUM_STATUS;
  2397. res_lib_votequorum_status.header.error = error;
  2398. corosync_api->ipc_response_send(conn, &res_lib_votequorum_status, sizeof(res_lib_votequorum_status));
  2399. LEAVE();
  2400. }
  2401. static void message_handler_req_lib_votequorum_qdevice_master_wins (void *conn,
  2402. const void *message)
  2403. {
  2404. const struct req_lib_votequorum_qdevice_master_wins *req_lib_votequorum_qdevice_master_wins = message;
  2405. struct res_lib_votequorum_status res_lib_votequorum_status;
  2406. cs_error_t error = CS_OK;
  2407. uint32_t oldflags = us->flags;
  2408. ENTER();
  2409. if (!qdevice_can_operate) {
  2410. error = CS_ERR_ACCESS;
  2411. goto out;
  2412. }
  2413. if (us->flags & NODE_FLAGS_QDEVICE_REGISTERED) {
  2414. if (strncmp(req_lib_votequorum_qdevice_master_wins->name, qdevice_name, VOTEQUORUM_QDEVICE_MAX_NAME_LEN)) {
  2415. error = CS_ERR_INVALID_PARAM;
  2416. goto out;
  2417. }
  2418. if (req_lib_votequorum_qdevice_master_wins->allow) {
  2419. us->flags |= NODE_FLAGS_QDEVICE_MASTER_WINS;
  2420. } else {
  2421. us->flags &= ~NODE_FLAGS_QDEVICE_MASTER_WINS;
  2422. }
  2423. if (us->flags != oldflags) {
  2424. votequorum_exec_send_nodeinfo(us->node_id);
  2425. }
  2426. update_qdevice_master_wins(req_lib_votequorum_qdevice_master_wins->allow);
  2427. } else {
  2428. error = CS_ERR_NOT_EXIST;
  2429. }
  2430. out:
  2431. res_lib_votequorum_status.header.size = sizeof(res_lib_votequorum_status);
  2432. res_lib_votequorum_status.header.id = MESSAGE_RES_VOTEQUORUM_STATUS;
  2433. res_lib_votequorum_status.header.error = error;
  2434. corosync_api->ipc_response_send(conn, &res_lib_votequorum_status, sizeof(res_lib_votequorum_status));
  2435. LEAVE();
  2436. }