daemons/controld/controld

/* */
This source file includes following definitions.
update_stonith_max_attempts
set_fence_reaction
controld_configure_fencing
too_many_st_failures
st_fail_count_reset
st_fail_count_increment
cib_fencing_updated
send_stonith_update
abort_for_stonith_failure
add_stonith_cleanup
remove_stonith_cleanup
purge_stonith_cleanup
execute_stonith_cleanup
fail_incompletable_stonith
tengine_stonith_connection_destroy
handle_fence_notification
te_connect_stonith
controld_trigger_fencer_connect
controld_disconnect_fencer
do_stonith_history_sync
tengine_stonith_callback
fence_with_delay
controld_execute_fence_action
controld_verify_stonith_watchdog_timeout
te_cleanup_stonith_history_sync
tengine_stonith_history_synced
stonith_history_sync_set_trigger
te_trigger_stonith_history_sync
   1 /*
   2  * Copyright 2004-2023 the Pacemaker project contributors
   3  *
   4  * The version control history for this file may have further details.
   5  *
   6  * This source code is licensed under the GNU General Public License version 2
   7  * or later (GPLv2+) WITHOUT ANY WARRANTY.
   8  */
   9 
  10 #include <crm_internal.h>
  11 #include <crm/crm.h>
  12 #include <crm/msg_xml.h>
  13 #include <crm/common/xml.h>
  14 #include <crm/stonith-ng.h>
  15 #include <crm/fencing/internal.h>
  16 
  17 #include <pacemaker-controld.h>
  18 
  19 static void
  20 tengine_stonith_history_synced(stonith_t *st, stonith_event_t *st_event);
  21 
  22 /*
  23  * stonith failure counting
  24  *
  25  * We don't want to get stuck in a permanent fencing loop. Keep track of the
  26  * number of fencing failures for each target node, and the most we'll restart a
  27  * transition for.
  28  */
  29 
  30 struct st_fail_rec {
  31     int count;
  32 };
  33 
  34 static bool fence_reaction_panic = false;
  35 static unsigned long int stonith_max_attempts = 10;
  36 static GHashTable *stonith_failures = NULL;
  37 
  38 /*!
  39  * \internal
  40  * \brief Update max fencing attempts before giving up
  41  *
  42  * \param[in] value  New max fencing attempts
  43  */
  44 static void
  45 update_stonith_max_attempts(const char *value)
     /*  */
  46 {
  47     stonith_max_attempts = char2score(value);
  48     if (stonith_max_attempts < 1UL) {
  49         stonith_max_attempts = 10UL;
  50     }
  51 }
  52 
  53 /*!
  54  * \internal
  55  * \brief Configure reaction to notification of local node being fenced
  56  *
  57  * \param[in] reaction_s  Reaction type
  58  */
  59 static void
  60 set_fence_reaction(const char *reaction_s)
     /*  */
  61 {
  62     if (pcmk__str_eq(reaction_s, "panic", pcmk__str_casei)) {
  63         fence_reaction_panic = true;
  64 
  65     } else {
  66         if (!pcmk__str_eq(reaction_s, "stop", pcmk__str_casei)) {
  67             crm_warn("Invalid value '%s' for %s, using 'stop'",
  68                      reaction_s, XML_CONFIG_ATTR_FENCE_REACTION);
  69         }
  70         fence_reaction_panic = false;
  71     }
  72 }
  73 
  74 /*!
  75  * \internal
  76  * \brief Configure fencing options based on the CIB
  77  *
  78  * \param[in,out] options  Name/value pairs for configured options
  79  */
  80 void
  81 controld_configure_fencing(GHashTable *options)
     /*  */
  82 {
  83     const char *value = NULL;
  84 
  85     value = g_hash_table_lookup(options, XML_CONFIG_ATTR_FENCE_REACTION);
  86     set_fence_reaction(value);
  87 
  88     value = g_hash_table_lookup(options, "stonith-max-attempts");
  89     update_stonith_max_attempts(value);
  90 }
  91 
  92 static gboolean
  93 too_many_st_failures(const char *target)
     /*  */
  94 {
  95     GHashTableIter iter;
  96     const char *key = NULL;
  97     struct st_fail_rec *value = NULL;
  98 
  99     if (stonith_failures == NULL) {
 100         return FALSE;
 101     }
 102 
 103     if (target == NULL) {
 104         g_hash_table_iter_init(&iter, stonith_failures);
 105         while (g_hash_table_iter_next(&iter, (gpointer *) &key,
 106                (gpointer *) &value)) {
 107 
 108             if (value->count >= stonith_max_attempts) {
 109                 target = (const char*)key;
 110                 goto too_many;
 111             }
 112         }
 113     } else {
 114         value = g_hash_table_lookup(stonith_failures, target);
 115         if ((value != NULL) && (value->count >= stonith_max_attempts)) {
 116             goto too_many;
 117         }
 118     }
 119     return FALSE;
 120 
 121 too_many:
 122     crm_warn("Too many failures (%d) to fence %s, giving up",
 123              value->count, target);
 124     return TRUE;
 125 }
 126 
 127 /*!
 128  * \internal
 129  * \brief Reset a stonith fail count
 130  *
 131  * \param[in] target  Name of node to reset, or NULL for all
 132  */
 133 void
 134 st_fail_count_reset(const char *target)
     /*  */
 135 {
 136     if (stonith_failures == NULL) {
 137         return;
 138     }
 139 
 140     if (target) {
 141         struct st_fail_rec *rec = NULL;
 142 
 143         rec = g_hash_table_lookup(stonith_failures, target);
 144         if (rec) {
 145             rec->count = 0;
 146         }
 147     } else {
 148         GHashTableIter iter;
 149         const char *key = NULL;
 150         struct st_fail_rec *rec = NULL;
 151 
 152         g_hash_table_iter_init(&iter, stonith_failures);
 153         while (g_hash_table_iter_next(&iter, (gpointer *) &key,
 154                                       (gpointer *) &rec)) {
 155             rec->count = 0;
 156         }
 157     }
 158 }
 159 
 160 static void
 161 st_fail_count_increment(const char *target)
     /*  */
 162 {
 163     struct st_fail_rec *rec = NULL;
 164 
 165     if (stonith_failures == NULL) {
 166         stonith_failures = pcmk__strkey_table(free, free);
 167     }
 168 
 169     rec = g_hash_table_lookup(stonith_failures, target);
 170     if (rec) {
 171         rec->count++;
 172     } else {
 173         rec = malloc(sizeof(struct st_fail_rec));
 174         if(rec == NULL) {
 175             return;
 176         }
 177 
 178         rec->count = 1;
 179         g_hash_table_insert(stonith_failures, strdup(target), rec);
 180     }
 181 }
 182 
 183 /* end stonith fail count functions */
 184 
 185 
 186 static void
 187 cib_fencing_updated(xmlNode *msg, int call_id, int rc, xmlNode *output,
     /*  */
 188                     void *user_data)
 189 {
 190     if (rc < pcmk_ok) {
 191         crm_err("Fencing update %d for %s: failed - %s (%d)",
 192                 call_id, (char *)user_data, pcmk_strerror(rc), rc);
 193         crm_log_xml_warn(msg, "Failed update");
 194         abort_transition(INFINITY, pcmk__graph_shutdown, "CIB update failed",
 195                          NULL);
 196 
 197     } else {
 198         crm_info("Fencing update %d for %s: complete", call_id, (char *)user_data);
 199     }
 200 }
 201 
 202 static void
 203 send_stonith_update(pcmk__graph_action_t *action, const char *target,
     /*  */
 204                     const char *uuid)
 205 {
 206     int rc = pcmk_ok;
 207     crm_node_t *peer = NULL;
 208 
 209     /* We (usually) rely on the membership layer to do node_update_cluster,
 210      * and the peer status callback to do node_update_peer, because the node
 211      * might have already rejoined before we get the stonith result here.
 212      */
 213     int flags = node_update_join | node_update_expected;
 214 
 215     /* zero out the node-status & remove all LRM status info */
 216     xmlNode *node_state = NULL;
 217 
 218     CRM_CHECK(target != NULL, return);
 219     CRM_CHECK(uuid != NULL, return);
 220 
 221     /* Make sure the membership and join caches are accurate */
 222     peer = crm_get_peer_full(0, target, CRM_GET_PEER_ANY);
 223 
 224     CRM_CHECK(peer != NULL, return);
 225 
 226     if (peer->state == NULL) {
 227         /* Usually, we rely on the membership layer to update the cluster state
 228          * in the CIB. However, if the node has never been seen, do it here, so
 229          * the node is not considered unclean.
 230          */
 231         flags |= node_update_cluster;
 232     }
 233 
 234     if (peer->uuid == NULL) {
 235         crm_info("Recording uuid '%s' for node '%s'", uuid, target);
 236         peer->uuid = strdup(uuid);
 237     }
 238 
 239     crmd_peer_down(peer, TRUE);
 240 
 241     /* Generate a node state update for the CIB */
 242     node_state = create_node_state_update(peer, flags, NULL, __func__);
 243 
 244     /* we have to mark whether or not remote nodes have already been fenced */
 245     if (peer->flags & crm_remote_node) {
 246         char *now_s = pcmk__ttoa(time(NULL));
 247 
 248         crm_xml_add(node_state, XML_NODE_IS_FENCED, now_s);
 249         free(now_s);
 250     }
 251 
 252     /* Force our known ID */
 253     crm_xml_add(node_state, XML_ATTR_ID, uuid);
 254 
 255     rc = controld_globals.cib_conn->cmds->modify(controld_globals.cib_conn,
 256                                                  XML_CIB_TAG_STATUS, node_state,
 257                                                  cib_scope_local
 258                                                  |cib_can_create);
 259 
 260     /* Delay processing the trigger until the update completes */
 261     crm_debug("Sending fencing update %d for %s", rc, target);
 262     fsa_register_cib_callback(rc, strdup(target), cib_fencing_updated);
 263 
 264     // Make sure it sticks
 265     /* controld_globals.cib_conn->cmds->bump_epoch(controld_globals.cib_conn,
 266      *                                             cib_scope_local);
 267      */
 268 
 269     controld_delete_node_state(peer->uname, controld_section_all,
 270                                cib_scope_local);
 271     free_xml(node_state);
 272     return;
 273 }
 274 
 275 /*!
 276  * \internal
 277  * \brief Abort transition due to stonith failure
 278  *
 279  * \param[in] abort_action  Whether to restart or stop transition
 280  * \param[in] target  Don't restart if this (NULL for any) has too many failures
 281  * \param[in] reason  Log this stonith action XML as abort reason (or NULL)
 282  */
 283 static void
 284 abort_for_stonith_failure(enum pcmk__graph_next abort_action,
     /*  */
 285                           const char *target, const xmlNode *reason)
 286 {
 287     /* If stonith repeatedly fails, we eventually give up on starting a new
 288      * transition for that reason.
 289      */
 290     if ((abort_action != pcmk__graph_wait) && too_many_st_failures(target)) {
 291         abort_action = pcmk__graph_wait;
 292     }
 293     abort_transition(INFINITY, abort_action, "Stonith failed", reason);
 294 }
 295 
 296 
 297 /*
 298  * stonith cleanup list
 299  *
 300  * If the DC is shot, proper notifications might not go out.
 301  * The stonith cleanup list allows the cluster to (re-)send
 302  * notifications once a new DC is elected.
 303  */
 304 
 305 static GList *stonith_cleanup_list = NULL;
 306 
 307 /*!
 308  * \internal
 309  * \brief Add a node to the stonith cleanup list
 310  *
 311  * \param[in] target  Name of node to add
 312  */
 313 void
 314 add_stonith_cleanup(const char *target) {
     /*  */
 315     stonith_cleanup_list = g_list_append(stonith_cleanup_list, strdup(target));
 316 }
 317 
 318 /*!
 319  * \internal
 320  * \brief Remove a node from the stonith cleanup list
 321  *
 322  * \param[in] Name of node to remove
 323  */
 324 void
 325 remove_stonith_cleanup(const char *target)
     /*  */
 326 {
 327     GList *iter = stonith_cleanup_list;
 328 
 329     while (iter != NULL) {
 330         GList *tmp = iter;
 331         char *iter_name = tmp->data;
 332 
 333         iter = iter->next;
 334         if (pcmk__str_eq(target, iter_name, pcmk__str_casei)) {
 335             crm_trace("Removing %s from the cleanup list", iter_name);
 336             stonith_cleanup_list = g_list_delete_link(stonith_cleanup_list, tmp);
 337             free(iter_name);
 338         }
 339     }
 340 }
 341 
 342 /*!
 343  * \internal
 344  * \brief Purge all entries from the stonith cleanup list
 345  */
 346 void
 347 purge_stonith_cleanup(void)
     /*  */
 348 {
 349     if (stonith_cleanup_list) {
 350         GList *iter = NULL;
 351 
 352         for (iter = stonith_cleanup_list; iter != NULL; iter = iter->next) {
 353             char *target = iter->data;
 354 
 355             crm_info("Purging %s from stonith cleanup list", target);
 356             free(target);
 357         }
 358         g_list_free(stonith_cleanup_list);
 359         stonith_cleanup_list = NULL;
 360     }
 361 }
 362 
 363 /*!
 364  * \internal
 365  * \brief Send stonith updates for all entries in cleanup list, then purge it
 366  */
 367 void
 368 execute_stonith_cleanup(void)
     /*  */
 369 {
 370     GList *iter;
 371 
 372     for (iter = stonith_cleanup_list; iter != NULL; iter = iter->next) {
 373         char *target = iter->data;
 374         crm_node_t *target_node = crm_get_peer(0, target);
 375         const char *uuid = crm_peer_uuid(target_node);
 376 
 377         crm_notice("Marking %s, target of a previous stonith action, as clean", target);
 378         send_stonith_update(NULL, target, uuid);
 379         free(target);
 380     }
 381     g_list_free(stonith_cleanup_list);
 382     stonith_cleanup_list = NULL;
 383 }
 384 
 385 /* end stonith cleanup list functions */
 386 
 387 
 388 /* stonith API client
 389  *
 390  * Functions that need to interact directly with the fencer via its API
 391  */
 392 
 393 static stonith_t *stonith_api = NULL;
 394 static crm_trigger_t *stonith_reconnect = NULL;
 395 static char *te_client_id = NULL;
 396 
 397 static gboolean
 398 fail_incompletable_stonith(pcmk__graph_t *graph)
     /*  */
 399 {
 400     GList *lpc = NULL;
 401     const char *task = NULL;
 402     xmlNode *last_action = NULL;
 403 
 404     if (graph == NULL) {
 405         return FALSE;
 406     }
 407 
 408     for (lpc = graph->synapses; lpc != NULL; lpc = lpc->next) {
 409         GList *lpc2 = NULL;
 410         pcmk__graph_synapse_t *synapse = (pcmk__graph_synapse_t *) lpc->data;
 411 
 412         if (pcmk_is_set(synapse->flags, pcmk__synapse_confirmed)) {
 413             continue;
 414         }
 415 
 416         for (lpc2 = synapse->actions; lpc2 != NULL; lpc2 = lpc2->next) {
 417             pcmk__graph_action_t *action = (pcmk__graph_action_t *) lpc2->data;
 418 
 419             if ((action->type != pcmk__cluster_graph_action)
 420                 || pcmk_is_set(action->flags, pcmk__graph_action_confirmed)) {
 421                 continue;
 422             }
 423 
 424             task = crm_element_value(action->xml, XML_LRM_ATTR_TASK);
 425             if (task && pcmk__str_eq(task, CRM_OP_FENCE, pcmk__str_casei)) {
 426                 pcmk__set_graph_action_flags(action, pcmk__graph_action_failed);
 427                 last_action = action->xml;
 428                 pcmk__update_graph(graph, action);
 429                 crm_notice("Failing action %d (%s): fencer terminated",
 430                            action->id, ID(action->xml));
 431             }
 432         }
 433     }
 434 
 435     if (last_action != NULL) {
 436         crm_warn("Fencer failure resulted in unrunnable actions");
 437         abort_for_stonith_failure(pcmk__graph_restart, NULL, last_action);
 438         return TRUE;
 439     }
 440 
 441     return FALSE;
 442 }
 443 
 444 static void
 445 tengine_stonith_connection_destroy(stonith_t *st, stonith_event_t *e)
     /*  */
 446 {
 447     te_cleanup_stonith_history_sync(st, FALSE);
 448 
 449     if (pcmk_is_set(controld_globals.fsa_input_register, R_ST_REQUIRED)) {
 450         crm_crit("Fencing daemon connection failed");
 451         mainloop_set_trigger(stonith_reconnect);
 452 
 453     } else {
 454         crm_info("Fencing daemon disconnected");
 455     }
 456 
 457     if (stonith_api) {
 458         /* the client API won't properly reconnect notifications
 459          * if they are still in the table - so remove them
 460          */
 461         if (stonith_api->state != stonith_disconnected) {
 462             stonith_api->cmds->disconnect(st);
 463         }
 464         stonith_api->cmds->remove_notification(stonith_api, NULL);
 465     }
 466 
 467     if (AM_I_DC) {
 468         fail_incompletable_stonith(controld_globals.transition_graph);
 469         trigger_graph();
 470     }
 471 }
 472 
 473 /*!
 474  * \internal
 475  * \brief Handle an event notification from the fencing API
 476  *
 477  * \param[in] st     Fencing API connection (ignored)
 478  * \param[in] event  Fencing API event notification
 479  */
 480 static void
 481 handle_fence_notification(stonith_t *st, stonith_event_t *event)
     /*  */
 482 {
 483     bool succeeded = true;
 484     const char *executioner = "the cluster";
 485     const char *client = "a client";
 486     const char *reason = NULL;
 487     int exec_status;
 488 
 489     if (te_client_id == NULL) {
 490         te_client_id = crm_strdup_printf("%s.%lu", crm_system_name,
 491                                          (unsigned long) getpid());
 492     }
 493 
 494     if (event == NULL) {
 495         crm_err("Notify data not found");
 496         return;
 497     }
 498 
 499     if (event->executioner != NULL) {
 500         executioner = event->executioner;
 501     }
 502     if (event->client_origin != NULL) {
 503         client = event->client_origin;
 504     }
 505 
 506     exec_status = stonith__event_execution_status(event);
 507     if ((stonith__event_exit_status(event) != CRM_EX_OK)
 508         || (exec_status != PCMK_EXEC_DONE)) {
 509         succeeded = false;
 510         if (exec_status == PCMK_EXEC_DONE) {
 511             exec_status = PCMK_EXEC_ERROR;
 512         }
 513     }
 514     reason = stonith__event_exit_reason(event);
 515 
 516     crmd_alert_fencing_op(event);
 517 
 518     if (pcmk__str_eq("on", event->action, pcmk__str_none)) {
 519         // Unfencing doesn't need special handling, just a log message
 520         if (succeeded) {
 521             crm_notice("%s was unfenced by %s at the request of %s@%s",
 522                        event->target, executioner, client, event->origin);
 523         } else {
 524             crm_err("Unfencing of %s by %s failed (%s%s%s) with exit status %d",
 525                     event->target, executioner,
 526                     pcmk_exec_status_str(exec_status),
 527                     ((reason == NULL)? "" : ": "),
 528                     ((reason == NULL)? "" : reason),
 529                     stonith__event_exit_status(event));
 530         }
 531         return;
 532     }
 533 
 534     if (succeeded
 535         && pcmk__str_eq(event->target, controld_globals.our_nodename,
 536                         pcmk__str_casei)) {
 537         /* We were notified of our own fencing. Most likely, either fencing was
 538          * misconfigured, or fabric fencing that doesn't cut cluster
 539          * communication is in use.
 540          *
 541          * Either way, shutting down the local host is a good idea, to require
 542          * administrator intervention. Also, other nodes would otherwise likely
 543          * set our status to lost because of the fencing callback and discard
 544          * our subsequent election votes as "not part of our cluster".
 545          */
 546         crm_crit("We were allegedly just fenced by %s for %s!",
 547                  executioner, event->origin); // Dumps blackbox if enabled
 548         if (fence_reaction_panic) {
 549             pcmk__panic(__func__);
 550         } else {
 551             crm_exit(CRM_EX_FATAL);
 552         }
 553         return; // Should never get here
 554     }
 555 
 556     /* Update the count of fencing failures for this target, in case we become
 557      * DC later. The current DC has already updated its fail count in
 558      * tengine_stonith_callback().
 559      */
 560     if (!AM_I_DC) {
 561         if (succeeded) {
 562             st_fail_count_reset(event->target);
 563         } else {
 564             st_fail_count_increment(event->target);
 565         }
 566     }
 567 
 568     crm_notice("Peer %s was%s terminated (%s) by %s on behalf of %s@%s: "
 569                "%s%s%s%s " CRM_XS " event=%s",
 570                event->target, (succeeded? "" : " not"),
 571                event->action, executioner, client, event->origin,
 572                (succeeded? "OK" : pcmk_exec_status_str(exec_status)),
 573                ((reason == NULL)? "" : " ("),
 574                ((reason == NULL)? "" : reason),
 575                ((reason == NULL)? "" : ")"),
 576                event->id);
 577 
 578     if (succeeded) {
 579         crm_node_t *peer = pcmk__search_known_node_cache(0, event->target,
 580                                                          CRM_GET_PEER_ANY);
 581         const char *uuid = NULL;
 582 
 583         if (peer == NULL) {
 584             return;
 585         }
 586 
 587         uuid = crm_peer_uuid(peer);
 588 
 589         if (AM_I_DC) {
 590             /* The DC always sends updates */
 591             send_stonith_update(NULL, event->target, uuid);
 592 
 593             /* @TODO Ideally, at this point, we'd check whether the fenced node
 594              * hosted any guest nodes, and call remote_node_down() for them.
 595              * Unfortunately, the controller doesn't have a simple, reliable way
 596              * to map hosts to guests. It might be possible to track this in the
 597              * peer cache via crm_remote_peer_cache_refresh(). For now, we rely
 598              * on the scheduler creating fence pseudo-events for the guests.
 599              */
 600 
 601             if (!pcmk__str_eq(client, te_client_id, pcmk__str_casei)) {
 602                 /* Abort the current transition if it wasn't the cluster that
 603                  * initiated fencing.
 604                  */
 605                 crm_info("External fencing operation from %s fenced %s",
 606                          client, event->target);
 607                 abort_transition(INFINITY, pcmk__graph_restart,
 608                                  "External Fencing Operation", NULL);
 609             }
 610 
 611         } else if (pcmk__str_eq(controld_globals.dc_name, event->target,
 612                                 pcmk__str_null_matches|pcmk__str_casei)
 613                    && !pcmk_is_set(peer->flags, crm_remote_node)) {
 614             // Assume the target was our DC if we don't currently have one
 615 
 616             if (controld_globals.dc_name != NULL) {
 617                 crm_notice("Fencing target %s was our DC", event->target);
 618             } else {
 619                 crm_notice("Fencing target %s may have been our DC",
 620                            event->target);
 621             }
 622 
 623             /* Given the CIB resyncing that occurs around elections,
 624              * have one node update the CIB now and, if the new DC is different,
 625              * have them do so too after the election
 626              */
 627             if (pcmk__str_eq(event->executioner, controld_globals.our_nodename,
 628                              pcmk__str_casei)) {
 629                 send_stonith_update(NULL, event->target, uuid);
 630             }
 631             add_stonith_cleanup(event->target);
 632         }
 633 
 634         /* If the target is a remote node, and we host its connection,
 635          * immediately fail all monitors so it can be recovered quickly.
 636          * The connection won't necessarily drop when a remote node is fenced,
 637          * so the failure might not otherwise be detected until the next poke.
 638          */
 639         if (pcmk_is_set(peer->flags, crm_remote_node)) {
 640             remote_ra_fail(event->target);
 641         }
 642 
 643         crmd_peer_down(peer, TRUE);
 644      }
 645 }
 646 
 647 /*!
 648  * \brief Connect to fencer
 649  *
 650  * \param[in] user_data  If NULL, retry failures now, otherwise retry in main loop
 651  *
 652  * \return TRUE
 653  * \note If user_data is NULL, this will wait 2s between attempts, for up to
 654  *       30 attempts, meaning the controller could be blocked as long as 58s.
 655  */
 656 static gboolean
 657 te_connect_stonith(gpointer user_data)
     /*  */
 658 {
 659     int rc = pcmk_ok;
 660 
 661     if (stonith_api == NULL) {
 662         stonith_api = stonith_api_new();
 663         if (stonith_api == NULL) {
 664             crm_err("Could not connect to fencer: API memory allocation failed");
 665             return TRUE;
 666         }
 667     }
 668 
 669     if (stonith_api->state != stonith_disconnected) {
 670         crm_trace("Already connected to fencer, no need to retry");
 671         return TRUE;
 672     }
 673 
 674     if (user_data == NULL) {
 675         // Blocking (retry failures now until successful)
 676         rc = stonith_api_connect_retry(stonith_api, crm_system_name, 30);
 677         if (rc != pcmk_ok) {
 678             crm_err("Could not connect to fencer in 30 attempts: %s "
 679                     CRM_XS " rc=%d", pcmk_strerror(rc), rc);
 680         }
 681     } else {
 682         // Non-blocking (retry failures later in main loop)
 683         rc = stonith_api->cmds->connect(stonith_api, crm_system_name, NULL);
 684         if (rc != pcmk_ok) {
 685             if (pcmk_is_set(controld_globals.fsa_input_register,
 686                             R_ST_REQUIRED)) {
 687                 crm_notice("Fencer connection failed (will retry): %s "
 688                            CRM_XS " rc=%d", pcmk_strerror(rc), rc);
 689                 mainloop_set_trigger(stonith_reconnect);
 690             } else {
 691                 crm_info("Fencer connection failed (ignoring because no longer required): %s "
 692                          CRM_XS " rc=%d", pcmk_strerror(rc), rc);
 693             }
 694             return TRUE;
 695         }
 696     }
 697 
 698     if (rc == pcmk_ok) {
 699         stonith_api->cmds->register_notification(stonith_api,
 700                                                  T_STONITH_NOTIFY_DISCONNECT,
 701                                                  tengine_stonith_connection_destroy);
 702         stonith_api->cmds->register_notification(stonith_api,
 703                                                  T_STONITH_NOTIFY_FENCE,
 704                                                  handle_fence_notification);
 705         stonith_api->cmds->register_notification(stonith_api,
 706                                                  T_STONITH_NOTIFY_HISTORY_SYNCED,
 707                                                  tengine_stonith_history_synced);
 708         te_trigger_stonith_history_sync(TRUE);
 709         crm_notice("Fencer successfully connected");
 710     }
 711 
 712     return TRUE;
 713 }
 714 
 715 /*!
 716     \internal
 717     \brief Schedule fencer connection attempt in main loop
 718 */
 719 void
 720 controld_trigger_fencer_connect(void)
     /*  */
 721 {
 722     if (stonith_reconnect == NULL) {
 723         stonith_reconnect = mainloop_add_trigger(G_PRIORITY_LOW,
 724                                                  te_connect_stonith,
 725                                                  GINT_TO_POINTER(TRUE));
 726     }
 727     controld_set_fsa_input_flags(R_ST_REQUIRED);
 728     mainloop_set_trigger(stonith_reconnect);
 729 }
 730 
 731 void
 732 controld_disconnect_fencer(bool destroy)
     /*  */
 733 {
 734     if (stonith_api) {
 735         // Prevent fencer connection from coming up again
 736         controld_clear_fsa_input_flags(R_ST_REQUIRED);
 737 
 738         if (stonith_api->state != stonith_disconnected) {
 739             stonith_api->cmds->disconnect(stonith_api);
 740         }
 741         stonith_api->cmds->remove_notification(stonith_api, NULL);
 742     }
 743     if (destroy) {
 744         if (stonith_api) {
 745             stonith_api->cmds->free(stonith_api);
 746             stonith_api = NULL;
 747         }
 748         if (stonith_reconnect) {
 749             mainloop_destroy_trigger(stonith_reconnect);
 750             stonith_reconnect = NULL;
 751         }
 752         if (te_client_id) {
 753             free(te_client_id);
 754             te_client_id = NULL;
 755         }
 756     }
 757 }
 758 
 759 static gboolean
 760 do_stonith_history_sync(gpointer user_data)
     /*  */
 761 {
 762     if (stonith_api && (stonith_api->state != stonith_disconnected)) {
 763         stonith_history_t *history = NULL;
 764 
 765         te_cleanup_stonith_history_sync(stonith_api, FALSE);
 766         stonith_api->cmds->history(stonith_api,
 767                                    st_opt_sync_call | st_opt_broadcast,
 768                                    NULL, &history, 5);
 769         stonith_history_free(history);
 770         return TRUE;
 771     } else {
 772         crm_info("Skip triggering stonith history-sync as stonith is disconnected");
 773         return FALSE;
 774     }
 775 }
 776 
 777 static void
 778 tengine_stonith_callback(stonith_t *stonith, stonith_callback_data_t *data)
     /*  */
 779 {
 780     char *uuid = NULL;
 781     int stonith_id = -1;
 782     int transition_id = -1;
 783     pcmk__graph_action_t *action = NULL;
 784     const char *target = NULL;
 785 
 786     if ((data == NULL) || (data->userdata == NULL)) {
 787         crm_err("Ignoring fence operation %d result: "
 788                 "No transition key given (bug?)",
 789                 ((data == NULL)? -1 : data->call_id));
 790         return;
 791     }
 792 
 793     if (!AM_I_DC) {
 794         const char *reason = stonith__exit_reason(data);
 795 
 796         if (reason == NULL) {
 797            reason = pcmk_exec_status_str(stonith__execution_status(data));
 798         }
 799         crm_notice("Result of fence operation %d: %d (%s) " CRM_XS " key=%s",
 800                    data->call_id, stonith__exit_status(data), reason,
 801                    (const char *) data->userdata);
 802         return;
 803     }
 804 
 805     CRM_CHECK(decode_transition_key(data->userdata, &uuid, &transition_id,
 806                                     &stonith_id, NULL),
 807               goto bail);
 808 
 809     if (controld_globals.transition_graph->complete || (stonith_id < 0)
 810         || !pcmk__str_eq(uuid, controld_globals.te_uuid, pcmk__str_none)
 811         || (controld_globals.transition_graph->id != transition_id)) {
 812         crm_info("Ignoring fence operation %d result: "
 813                  "Not from current transition " CRM_XS
 814                  " complete=%s action=%d uuid=%s (vs %s) transition=%d (vs %d)",
 815                  data->call_id,
 816                  pcmk__btoa(controld_globals.transition_graph->complete),
 817                  stonith_id, uuid, controld_globals.te_uuid, transition_id,
 818                  controld_globals.transition_graph->id);
 819         goto bail;
 820     }
 821 
 822     action = controld_get_action(stonith_id);
 823     if (action == NULL) {
 824         crm_err("Ignoring fence operation %d result: "
 825                 "Action %d not found in transition graph (bug?) "
 826                 CRM_XS " uuid=%s transition=%d",
 827                 data->call_id, stonith_id, uuid, transition_id);
 828         goto bail;
 829     }
 830 
 831     target = crm_element_value(action->xml, XML_LRM_ATTR_TARGET);
 832     if (target == NULL) {
 833         crm_err("Ignoring fence operation %d result: No target given (bug?)",
 834                 data->call_id);
 835         goto bail;
 836     }
 837 
 838     stop_te_timer(action);
 839     if (stonith__exit_status(data) == CRM_EX_OK) {
 840         const char *uuid = crm_element_value(action->xml, XML_LRM_ATTR_TARGET_UUID);
 841         const char *op = crm_meta_value(action->params, "stonith_action");
 842 
 843         crm_info("Fence operation %d for %s succeeded", data->call_id, target);
 844         if (!(pcmk_is_set(action->flags, pcmk__graph_action_confirmed))) {
 845             te_action_confirmed(action, NULL);
 846             if (pcmk__str_eq("on", op, pcmk__str_casei)) {
 847                 const char *value = NULL;
 848                 char *now = pcmk__ttoa(time(NULL));
 849                 gboolean is_remote_node = FALSE;
 850 
 851                 /* This check is not 100% reliable, since this node is not
 852                  * guaranteed to have the remote node cached. However, it
 853                  * doesn't have to be reliable, since the attribute manager can
 854                  * learn a node's "remoteness" by other means sooner or later.
 855                  * This allows it to learn more quickly if this node does have
 856                  * the information.
 857                  */
 858                 if (g_hash_table_lookup(crm_remote_peer_cache, uuid) != NULL) {
 859                     is_remote_node = TRUE;
 860                 }
 861 
 862                 update_attrd(target, CRM_ATTR_UNFENCED, now, NULL,
 863                              is_remote_node);
 864                 free(now);
 865 
 866                 value = crm_meta_value(action->params, XML_OP_ATTR_DIGESTS_ALL);
 867                 update_attrd(target, CRM_ATTR_DIGESTS_ALL, value, NULL,
 868                              is_remote_node);
 869 
 870                 value = crm_meta_value(action->params, XML_OP_ATTR_DIGESTS_SECURE);
 871                 update_attrd(target, CRM_ATTR_DIGESTS_SECURE, value, NULL,
 872                              is_remote_node);
 873 
 874             } else if (!(pcmk_is_set(action->flags, pcmk__graph_action_sent_update))) {
 875                 send_stonith_update(action, target, uuid);
 876                 pcmk__set_graph_action_flags(action,
 877                                              pcmk__graph_action_sent_update);
 878             }
 879         }
 880         st_fail_count_reset(target);
 881 
 882     } else {
 883         enum pcmk__graph_next abort_action = pcmk__graph_restart;
 884         int status = stonith__execution_status(data);
 885         const char *reason = stonith__exit_reason(data);
 886 
 887         if (reason == NULL) {
 888             if (status == PCMK_EXEC_DONE) {
 889                 reason = "Agent returned error";
 890             } else {
 891                 reason = pcmk_exec_status_str(status);
 892             }
 893         }
 894         pcmk__set_graph_action_flags(action, pcmk__graph_action_failed);
 895 
 896         /* If no fence devices were available, there's no use in immediately
 897          * checking again, so don't start a new transition in that case.
 898          */
 899         if (status == PCMK_EXEC_NO_FENCE_DEVICE) {
 900             crm_warn("Fence operation %d for %s failed: %s "
 901                      "(aborting transition and giving up for now)",
 902                      data->call_id, target, reason);
 903             abort_action = pcmk__graph_wait;
 904         } else {
 905             crm_notice("Fence operation %d for %s failed: %s "
 906                        "(aborting transition)", data->call_id, target, reason);
 907         }
 908 
 909         /* Increment the fail count now, so abort_for_stonith_failure() can
 910          * check it. Non-DC nodes will increment it in
 911          * handle_fence_notification().
 912          */
 913         st_fail_count_increment(target);
 914         abort_for_stonith_failure(abort_action, target, NULL);
 915     }
 916 
 917     pcmk__update_graph(controld_globals.transition_graph, action);
 918     trigger_graph();
 919 
 920   bail:
 921     free(data->userdata);
 922     free(uuid);
 923     return;
 924 }
 925 
 926 static int
 927 fence_with_delay(const char *target, const char *type, int delay)
     /*  */
 928 {
 929     uint32_t options = st_opt_none; // Group of enum stonith_call_options
 930     int timeout_sec = (int) (controld_globals.transition_graph->stonith_timeout
 931                              / 1000);
 932 
 933     if (crmd_join_phase_count(crm_join_confirmed) == 1) {
 934         stonith__set_call_options(options, target, st_opt_allow_suicide);
 935     }
 936     return stonith_api->cmds->fence_with_delay(stonith_api, options, target,
 937                                                type, timeout_sec, 0, delay);
 938 }
 939 
 940 /*!
 941  * \internal
 942  * \brief Execute a fencing action from a transition graph
 943  *
 944  * \param[in] graph   Transition graph being executed (ignored)
 945  * \param[in] action  Fencing action to execute
 946  *
 947  * \return Standard Pacemaker return code
 948  */
 949 int
 950 controld_execute_fence_action(pcmk__graph_t *graph,
     /*  */
 951                               pcmk__graph_action_t *action)
 952 {
 953     int rc = 0;
 954     const char *id = ID(action->xml);
 955     const char *uuid = crm_element_value(action->xml, XML_LRM_ATTR_TARGET_UUID);
 956     const char *target = crm_element_value(action->xml, XML_LRM_ATTR_TARGET);
 957     const char *type = crm_meta_value(action->params, "stonith_action");
 958     char *transition_key = NULL;
 959     const char *priority_delay = NULL;
 960     int delay_i = 0;
 961     gboolean invalid_action = FALSE;
 962     int stonith_timeout = (int) (controld_globals.transition_graph->stonith_timeout
 963                                  / 1000);
 964 
 965     CRM_CHECK(id != NULL, invalid_action = TRUE);
 966     CRM_CHECK(uuid != NULL, invalid_action = TRUE);
 967     CRM_CHECK(type != NULL, invalid_action = TRUE);
 968     CRM_CHECK(target != NULL, invalid_action = TRUE);
 969 
 970     if (invalid_action) {
 971         crm_log_xml_warn(action->xml, "BadAction");
 972         return EPROTO;
 973     }
 974 
 975     priority_delay = crm_meta_value(action->params, XML_CONFIG_ATTR_PRIORITY_FENCING_DELAY);
 976 
 977     crm_notice("Requesting fencing (%s) targeting node %s "
 978                CRM_XS " action=%s timeout=%i%s%s",
 979                type, target, id, stonith_timeout,
 980                priority_delay ? " priority_delay=" : "",
 981                priority_delay ? priority_delay : "");
 982 
 983     /* Passing NULL means block until we can connect... */
 984     te_connect_stonith(NULL);
 985 
 986     pcmk__scan_min_int(priority_delay, &delay_i, 0);
 987     rc = fence_with_delay(target, type, delay_i);
 988     transition_key = pcmk__transition_key(controld_globals.transition_graph->id,
 989                                           action->id, 0,
 990                                           controld_globals.te_uuid),
 991     stonith_api->cmds->register_callback(stonith_api, rc,
 992                                          (stonith_timeout
 993                                           + (delay_i > 0 ? delay_i : 0)),
 994                                          st_opt_timeout_updates, transition_key,
 995                                          "tengine_stonith_callback",
 996                                          tengine_stonith_callback);
 997     return pcmk_rc_ok;
 998 }
 999 
1000 bool
1001 controld_verify_stonith_watchdog_timeout(const char *value)
     /*  */
1002 {
1003     const char *our_nodename = controld_globals.our_nodename;
1004     gboolean rv = TRUE;
1005 
1006     if (stonith_api && (stonith_api->state != stonith_disconnected) &&
1007         stonith__watchdog_fencing_enabled_for_node_api(stonith_api,
1008                                                        our_nodename)) {
1009         rv = pcmk__valid_sbd_timeout(value);
1010     }
1011     return rv;
1012 }
1013 
1014 /* end stonith API client functions */
1015 
1016 
1017 /*
1018  * stonith history synchronization
1019  *
1020  * Each node's fencer keeps track of a cluster-wide fencing history. When a node
1021  * joins or leaves, we need to synchronize the history across all nodes.
1022  */
1023 
1024 static crm_trigger_t *stonith_history_sync_trigger = NULL;
1025 static mainloop_timer_t *stonith_history_sync_timer_short = NULL;
1026 static mainloop_timer_t *stonith_history_sync_timer_long = NULL;
1027 
1028 void
1029 te_cleanup_stonith_history_sync(stonith_t *st, bool free_timers)
     /*  */
1030 {
1031     if (free_timers) {
1032         mainloop_timer_del(stonith_history_sync_timer_short);
1033         stonith_history_sync_timer_short = NULL;
1034         mainloop_timer_del(stonith_history_sync_timer_long);
1035         stonith_history_sync_timer_long = NULL;
1036     } else {
1037         mainloop_timer_stop(stonith_history_sync_timer_short);
1038         mainloop_timer_stop(stonith_history_sync_timer_long);
1039     }
1040 
1041     if (st) {
1042         st->cmds->remove_notification(st, T_STONITH_NOTIFY_HISTORY_SYNCED);
1043     }
1044 }
1045 
1046 static void
1047 tengine_stonith_history_synced(stonith_t *st, stonith_event_t *st_event)
     /*  */
1048 {
1049     te_cleanup_stonith_history_sync(st, FALSE);
1050     crm_debug("Fence-history synced - cancel all timers");
1051 }
1052 
1053 static gboolean
1054 stonith_history_sync_set_trigger(gpointer user_data)
     /*  */
1055 {
1056     mainloop_set_trigger(stonith_history_sync_trigger);
1057     return FALSE;
1058 }
1059 
1060 void
1061 te_trigger_stonith_history_sync(bool long_timeout)
     /*  */
1062 {
1063     /* trigger a sync in 5s to give more nodes the
1064      * chance to show up so that we don't create
1065      * unnecessary stonith-history-sync traffic
1066      *
1067      * the long timeout of 30s is there as a fallback
1068      * so that after a successful connection to fenced
1069      * we will wait for 30s for the DC to trigger a
1070      * history-sync
1071      * if this doesn't happen we trigger a sync locally
1072      * (e.g. fenced segfaults and is restarted by pacemakerd)
1073      */
1074 
1075     /* as we are finally checking the stonith-connection
1076      * in do_stonith_history_sync we should be fine
1077      * leaving stonith_history_sync_time & stonith_history_sync_trigger
1078      * around
1079      */
1080     if (stonith_history_sync_trigger == NULL) {
1081         stonith_history_sync_trigger =
1082             mainloop_add_trigger(G_PRIORITY_LOW,
1083                                  do_stonith_history_sync, NULL);
1084     }
1085 
1086     if (long_timeout) {
1087         if(stonith_history_sync_timer_long == NULL) {
1088             stonith_history_sync_timer_long =
1089                 mainloop_timer_add("history_sync_long", 30000,
1090                                    FALSE, stonith_history_sync_set_trigger,
1091                                    NULL);
1092         }
1093         crm_info("Fence history will be synchronized cluster-wide within 30 seconds");
1094         mainloop_timer_start(stonith_history_sync_timer_long);
1095     } else {
1096         if(stonith_history_sync_timer_short == NULL) {
1097             stonith_history_sync_timer_short =
1098                 mainloop_timer_add("history_sync_short", 5000,
1099                                    FALSE, stonith_history_sync_set_trigger,
1100                                    NULL);
1101         }
1102         crm_info("Fence history will be synchronized cluster-wide within 5 seconds");
1103         mainloop_timer_start(stonith_history_sync_timer_short);
1104     }
1105 
1106 }
1107 
1108 /* end stonith history synchronization functions */
/* */
root/daemons/controld/controld_fencing.c

DEFINITIONS
