lib/cluster/membership.c

/* */
This source file includes following definitions.
crm_remote_peer_cache_size
crm_remote_peer_get
crm_remote_peer_cache_remove
remote_state_from_cib
remote_cache_refresh_helper
mark_dirty
is_dirty
crm_remote_peer_cache_refresh
crm_is_peer_active
crm_reap_dead_member
reap_crm_member
count_peer
crm_active_peers
destroy_crm_node
crm_peer_init
crm_peer_destroy
crm_set_status_callback
crm_set_autoreap
dump_peer_hash
hash_find_by_data
pcmk__search_node_caches
crm_get_peer_full
pcmk__search_cluster_node_cache
remove_conflicting_peer
crm_get_peer
update_peer_uname
proc2text
crm_update_peer_proc
pcmk__update_peer_expected
update_peer_state_iter
pcmk__update_peer_state
pcmk__reap_unseen_nodes
find_known_node
known_node_cache_refresh_helper
refresh_known_node_cache
pcmk__refresh_node_caches_from_cib
pcmk__search_known_node_cache
crm_terminate_member
crm_terminate_member_no_mainloop
   1 /*
   2  * Copyright 2004-2022 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 Lesser General Public License
   7  * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY.
   8  */
   9 
  10 #include <crm_internal.h>
  11 
  12 #ifndef _GNU_SOURCE
  13 #  define _GNU_SOURCE
  14 #endif
  15 
  16 #include <sys/param.h>
  17 #include <sys/types.h>
  18 #include <stdio.h>
  19 #include <unistd.h>
  20 #include <string.h>
  21 #include <glib.h>
  22 #include <crm/common/ipc.h>
  23 #include <crm/common/xml_internal.h>
  24 #include <crm/cluster/internal.h>
  25 #include <crm/msg_xml.h>
  26 #include <crm/stonith-ng.h>
  27 #include "crmcluster_private.h"
  28 
  29 /* The peer cache remembers cluster nodes that have been seen.
  30  * This is managed mostly automatically by libcluster, based on
  31  * cluster membership events.
  32  *
  33  * Because cluster nodes can have conflicting names or UUIDs,
  34  * the hash table key is a uniquely generated ID.
  35  */
  36 GHashTable *crm_peer_cache = NULL;
  37 
  38 /*
  39  * The remote peer cache tracks pacemaker_remote nodes. While the
  40  * value has the same type as the peer cache's, it is tracked separately for
  41  * three reasons: pacemaker_remote nodes can't have conflicting names or UUIDs,
  42  * so the name (which is also the UUID) is used as the hash table key; there
  43  * is no equivalent of membership events, so management is not automatic; and
  44  * most users of the peer cache need to exclude pacemaker_remote nodes.
  45  *
  46  * That said, using a single cache would be more logical and less error-prone,
  47  * so it would be a good idea to merge them one day.
  48  *
  49  * libcluster provides two avenues for populating the cache:
  50  * crm_remote_peer_get() and crm_remote_peer_cache_remove() directly manage it,
  51  * while crm_remote_peer_cache_refresh() populates it via the CIB.
  52  */
  53 GHashTable *crm_remote_peer_cache = NULL;
  54 
  55 /*
  56  * The known node cache tracks cluster and remote nodes that have been seen in
  57  * the CIB. It is useful mainly when a caller needs to know about a node that
  58  * may no longer be in the membership, but doesn't want to add the node to the
  59  * main peer cache tables.
  60  */
  61 static GHashTable *known_node_cache = NULL;
  62 
  63 unsigned long long crm_peer_seq = 0;
  64 gboolean crm_have_quorum = FALSE;
  65 static gboolean crm_autoreap  = TRUE;
  66 
  67 // Flag setting and clearing for crm_node_t:flags
  68 
  69 #define set_peer_flags(peer, flags_to_set) do {                               \
  70         (peer)->flags = pcmk__set_flags_as(__func__, __LINE__, LOG_TRACE,     \
  71                                            "Peer", (peer)->uname,             \
  72                                            (peer)->flags, (flags_to_set),     \
  73                                            #flags_to_set);                    \
  74     } while (0)
  75 
  76 #define clear_peer_flags(peer, flags_to_clear) do {                           \
  77         (peer)->flags = pcmk__clear_flags_as(__func__, __LINE__,              \
  78                                              LOG_TRACE,                       \
  79                                              "Peer", (peer)->uname,           \
  80                                              (peer)->flags, (flags_to_clear), \
  81                                              #flags_to_clear);                \
  82     } while (0)
  83 
  84 static void update_peer_uname(crm_node_t *node, const char *uname);
  85 
  86 int
  87 crm_remote_peer_cache_size(void)
     /*  */
  88 {
  89     if (crm_remote_peer_cache == NULL) {
  90         return 0;
  91     }
  92     return g_hash_table_size(crm_remote_peer_cache);
  93 }
  94 
  95 /*!
  96  * \brief Get a remote node peer cache entry, creating it if necessary
  97  *
  98  * \param[in] node_name  Name of remote node
  99  *
 100  * \return Cache entry for node on success, NULL (and set errno) otherwise
 101  *
 102  * \note When creating a new entry, this will leave the node state undetermined,
 103  *       so the caller should also call pcmk__update_peer_state() if the state
 104  *       is known.
 105  */
 106 crm_node_t *
 107 crm_remote_peer_get(const char *node_name)
     /*  */
 108 {
 109     crm_node_t *node;
 110 
 111     if (node_name == NULL) {
 112         errno = -EINVAL;
 113         return NULL;
 114     }
 115 
 116     /* Return existing cache entry if one exists */
 117     node = g_hash_table_lookup(crm_remote_peer_cache, node_name);
 118     if (node) {
 119         return node;
 120     }
 121 
 122     /* Allocate a new entry */
 123     node = calloc(1, sizeof(crm_node_t));
 124     if (node == NULL) {
 125         return NULL;
 126     }
 127 
 128     /* Populate the essential information */
 129     set_peer_flags(node, crm_remote_node);
 130     node->uuid = strdup(node_name);
 131     if (node->uuid == NULL) {
 132         free(node);
 133         errno = -ENOMEM;
 134         return NULL;
 135     }
 136 
 137     /* Add the new entry to the cache */
 138     g_hash_table_replace(crm_remote_peer_cache, node->uuid, node);
 139     crm_trace("added %s to remote cache", node_name);
 140 
 141     /* Update the entry's uname, ensuring peer status callbacks are called */
 142     update_peer_uname(node, node_name);
 143     return node;
 144 }
 145 
 146 void
 147 crm_remote_peer_cache_remove(const char *node_name)
     /*  */
 148 {
 149     if (g_hash_table_remove(crm_remote_peer_cache, node_name)) {
 150         crm_trace("removed %s from remote peer cache", node_name);
 151     }
 152 }
 153 
 154 /*!
 155  * \internal
 156  * \brief Return node status based on a CIB status entry
 157  *
 158  * \param[in] node_state  XML of node state
 159  *
 160  * \return CRM_NODE_LOST if XML_NODE_IN_CLUSTER is false in node_state,
 161  *         CRM_NODE_MEMBER otherwise
 162  * \note Unlike most boolean XML attributes, this one defaults to true, for
 163  *       backward compatibility with older controllers that don't set it.
 164  */
 165 static const char *
 166 remote_state_from_cib(xmlNode *node_state)
     /*  */
 167 {
 168     bool status = false;
 169 
 170     if (pcmk__xe_get_bool_attr(node_state, XML_NODE_IN_CLUSTER, &status) == pcmk_rc_ok && !status) {
 171         return CRM_NODE_LOST;
 172     } else {
 173         return CRM_NODE_MEMBER;
 174     }
 175 }
 176 
 177 /* user data for looping through remote node xpath searches */
 178 struct refresh_data {
 179     const char *field;  /* XML attribute to check for node name */
 180     gboolean has_state; /* whether to update node state based on XML */
 181 };
 182 
 183 /*!
 184  * \internal
 185  * \brief Process one pacemaker_remote node xpath search result
 186  *
 187  * \param[in] result     XML search result
 188  * \param[in] user_data  what to look for in the XML
 189  */
 190 static void
 191 remote_cache_refresh_helper(xmlNode *result, void *user_data)
     /*  */
 192 {
 193     struct refresh_data *data = user_data;
 194     const char *remote = crm_element_value(result, data->field);
 195     const char *state = NULL;
 196     crm_node_t *node;
 197 
 198     CRM_CHECK(remote != NULL, return);
 199 
 200     /* Determine node's state, if the result has it */
 201     if (data->has_state) {
 202         state = remote_state_from_cib(result);
 203     }
 204 
 205     /* Check whether cache already has entry for node */
 206     node = g_hash_table_lookup(crm_remote_peer_cache, remote);
 207 
 208     if (node == NULL) {
 209         /* Node is not in cache, so add a new entry for it */
 210         node = crm_remote_peer_get(remote);
 211         CRM_ASSERT(node);
 212         if (state) {
 213             pcmk__update_peer_state(__func__, node, state, 0);
 214         }
 215 
 216     } else if (pcmk_is_set(node->flags, crm_node_dirty)) {
 217         /* Node is in cache and hasn't been updated already, so mark it clean */
 218         clear_peer_flags(node, crm_node_dirty);
 219         if (state) {
 220             pcmk__update_peer_state(__func__, node, state, 0);
 221         }
 222     }
 223 }
 224 
 225 static void
 226 mark_dirty(gpointer key, gpointer value, gpointer user_data)
     /*  */
 227 {
 228     set_peer_flags((crm_node_t *) value, crm_node_dirty);
 229 }
 230 
 231 static gboolean
 232 is_dirty(gpointer key, gpointer value, gpointer user_data)
     /*  */
 233 {
 234     return pcmk_is_set(((crm_node_t*)value)->flags, crm_node_dirty);
 235 }
 236 
 237 /*!
 238  * \brief Repopulate the remote peer cache based on CIB XML
 239  *
 240  * \param[in] xmlNode  CIB XML to parse
 241  */
 242 void
 243 crm_remote_peer_cache_refresh(xmlNode *cib)
     /*  */
 244 {
 245     struct refresh_data data;
 246 
 247     crm_peer_init();
 248 
 249     /* First, we mark all existing cache entries as dirty,
 250      * so that later we can remove any that weren't in the CIB.
 251      * We don't empty the cache, because we need to detect changes in state.
 252      */
 253     g_hash_table_foreach(crm_remote_peer_cache, mark_dirty, NULL);
 254 
 255     /* Look for guest nodes and remote nodes in the status section */
 256     data.field = "id";
 257     data.has_state = TRUE;
 258     crm_foreach_xpath_result(cib, PCMK__XP_REMOTE_NODE_STATUS,
 259                              remote_cache_refresh_helper, &data);
 260 
 261     /* Look for guest nodes and remote nodes in the configuration section,
 262      * because they may have just been added and not have a status entry yet.
 263      * In that case, the cached node state will be left NULL, so that the
 264      * peer status callback isn't called until we're sure the node started
 265      * successfully.
 266      */
 267     data.field = "value";
 268     data.has_state = FALSE;
 269     crm_foreach_xpath_result(cib, PCMK__XP_GUEST_NODE_CONFIG,
 270                              remote_cache_refresh_helper, &data);
 271     data.field = "id";
 272     data.has_state = FALSE;
 273     crm_foreach_xpath_result(cib, PCMK__XP_REMOTE_NODE_CONFIG,
 274                              remote_cache_refresh_helper, &data);
 275 
 276     /* Remove all old cache entries that weren't seen in the CIB */
 277     g_hash_table_foreach_remove(crm_remote_peer_cache, is_dirty, NULL);
 278 }
 279 
 280 gboolean
 281 crm_is_peer_active(const crm_node_t * node)
     /*  */
 282 {
 283     if(node == NULL) {
 284         return FALSE;
 285     }
 286 
 287     if (pcmk_is_set(node->flags, crm_remote_node)) {
 288         /* remote nodes are never considered active members. This
 289          * guarantees they will never be considered for DC membership.*/
 290         return FALSE;
 291     }
 292 #if SUPPORT_COROSYNC
 293     if (is_corosync_cluster()) {
 294         return crm_is_corosync_peer_active(node);
 295     }
 296 #endif
 297     crm_err("Unhandled cluster type: %s", name_for_cluster_type(get_cluster_type()));
 298     return FALSE;
 299 }
 300 
 301 static gboolean
 302 crm_reap_dead_member(gpointer key, gpointer value, gpointer user_data)
     /*  */
 303 {
 304     crm_node_t *node = value;
 305     crm_node_t *search = user_data;
 306 
 307     if (search == NULL) {
 308         return FALSE;
 309 
 310     } else if (search->id && node->id != search->id) {
 311         return FALSE;
 312 
 313     } else if (search->id == 0 && !pcmk__str_eq(node->uname, search->uname, pcmk__str_casei)) {
 314         return FALSE;
 315 
 316     } else if (crm_is_peer_active(value) == FALSE) {
 317         crm_info("Removing node with name %s and id %u from membership cache",
 318                  (node->uname? node->uname : "unknown"), node->id);
 319         return TRUE;
 320     }
 321     return FALSE;
 322 }
 323 
 324 /*!
 325  * \brief Remove all peer cache entries matching a node ID and/or uname
 326  *
 327  * \param[in] id    ID of node to remove (or 0 to ignore)
 328  * \param[in] name  Uname of node to remove (or NULL to ignore)
 329  *
 330  * \return Number of cache entries removed
 331  */
 332 guint
 333 reap_crm_member(uint32_t id, const char *name)
     /*  */
 334 {
 335     int matches = 0;
 336     crm_node_t search = { 0, };
 337 
 338     if (crm_peer_cache == NULL) {
 339         crm_trace("Membership cache not initialized, ignoring purge request");
 340         return 0;
 341     }
 342 
 343     search.id = id;
 344     pcmk__str_update(&search.uname, name);
 345     matches = g_hash_table_foreach_remove(crm_peer_cache, crm_reap_dead_member, &search);
 346     if(matches) {
 347         crm_notice("Purged %d peer%s with id=%u%s%s from the membership cache",
 348                    matches, pcmk__plural_s(matches), search.id,
 349                    (search.uname? " and/or uname=" : ""),
 350                    (search.uname? search.uname : ""));
 351 
 352     } else {
 353         crm_info("No peers with id=%u%s%s to purge from the membership cache",
 354                  search.id, (search.uname? " and/or uname=" : ""),
 355                  (search.uname? search.uname : ""));
 356     }
 357 
 358     free(search.uname);
 359     return matches;
 360 }
 361 
 362 static void
 363 count_peer(gpointer key, gpointer value, gpointer user_data)
     /*  */
 364 {
 365     guint *count = user_data;
 366     crm_node_t *node = value;
 367 
 368     if (crm_is_peer_active(node)) {
 369         *count = *count + 1;
 370     }
 371 }
 372 
 373 guint
 374 crm_active_peers(void)
     /*  */
 375 {
 376     guint count = 0;
 377 
 378     if (crm_peer_cache) {
 379         g_hash_table_foreach(crm_peer_cache, count_peer, &count);
 380     }
 381     return count;
 382 }
 383 
 384 static void
 385 destroy_crm_node(gpointer data)
     /*  */
 386 {
 387     crm_node_t *node = data;
 388 
 389     crm_trace("Destroying entry for node %u: %s", node->id, node->uname);
 390 
 391     free(node->uname);
 392     free(node->state);
 393     free(node->uuid);
 394     free(node->expected);
 395     free(node);
 396 }
 397 
 398 void
 399 crm_peer_init(void)
     /*  */
 400 {
 401     if (crm_peer_cache == NULL) {
 402         crm_peer_cache = pcmk__strikey_table(free, destroy_crm_node);
 403     }
 404 
 405     if (crm_remote_peer_cache == NULL) {
 406         crm_remote_peer_cache = pcmk__strikey_table(NULL, destroy_crm_node);
 407     }
 408 
 409     if (known_node_cache == NULL) {
 410         known_node_cache = pcmk__strikey_table(free, destroy_crm_node);
 411     }
 412 }
 413 
 414 void
 415 crm_peer_destroy(void)
     /*  */
 416 {
 417     if (crm_peer_cache != NULL) {
 418         crm_trace("Destroying peer cache with %d members", g_hash_table_size(crm_peer_cache));
 419         g_hash_table_destroy(crm_peer_cache);
 420         crm_peer_cache = NULL;
 421     }
 422 
 423     if (crm_remote_peer_cache != NULL) {
 424         crm_trace("Destroying remote peer cache with %d members", g_hash_table_size(crm_remote_peer_cache));
 425         g_hash_table_destroy(crm_remote_peer_cache);
 426         crm_remote_peer_cache = NULL;
 427     }
 428 
 429     if (known_node_cache != NULL) {
 430         crm_trace("Destroying known node cache with %d members",
 431                   g_hash_table_size(known_node_cache));
 432         g_hash_table_destroy(known_node_cache);
 433         known_node_cache = NULL;
 434     }
 435 
 436 }
 437 
 438 static void (*peer_status_callback)(enum crm_status_type, crm_node_t *,
 439                                     const void *) = NULL;
 440 
 441 /*!
 442  * \brief Set a client function that will be called after peer status changes
 443  *
 444  * \param[in] dispatch  Pointer to function to use as callback
 445  *
 446  * \note Previously, client callbacks were responsible for peer cache
 447  *       management. This is no longer the case, and client callbacks should do
 448  *       only client-specific handling. Callbacks MUST NOT add or remove entries
 449  *       in the peer caches.
 450  */
 451 void
 452 crm_set_status_callback(void (*dispatch) (enum crm_status_type, crm_node_t *, const void *))
     /*  */
 453 {
 454     peer_status_callback = dispatch;
 455 }
 456 
 457 /*!
 458  * \brief Tell the library whether to automatically reap lost nodes
 459  *
 460  * If TRUE (the default), calling crm_update_peer_proc() will also update the
 461  * peer state to CRM_NODE_MEMBER or CRM_NODE_LOST, and pcmk__update_peer_state()
 462  * will reap peers whose state changes to anything other than CRM_NODE_MEMBER.
 463  * Callers should leave this enabled unless they plan to manage the cache
 464  * separately on their own.
 465  *
 466  * \param[in] autoreap  TRUE to enable automatic reaping, FALSE to disable
 467  */
 468 void
 469 crm_set_autoreap(gboolean autoreap)
     /*  */
 470 {
 471     crm_autoreap = autoreap;
 472 }
 473 
 474 static void
 475 dump_peer_hash(int level, const char *caller)
     /*  */
 476 {
 477     GHashTableIter iter;
 478     const char *id = NULL;
 479     crm_node_t *node = NULL;
 480 
 481     g_hash_table_iter_init(&iter, crm_peer_cache);
 482     while (g_hash_table_iter_next(&iter, (gpointer *) &id, (gpointer *) &node)) {
 483         do_crm_log(level, "%s: Node %u/%s = %p - %s", caller, node->id, node->uname, node, id);
 484     }
 485 }
 486 
 487 static gboolean
 488 hash_find_by_data(gpointer key, gpointer value, gpointer user_data)
     /*  */
 489 {
 490     return value == user_data;
 491 }
 492 
 493 /*!
 494  * \internal
 495  * \brief Search caches for a node (cluster or Pacemaker Remote)
 496  *
 497  * \param[in] id     If not 0, cluster node ID to search for
 498  * \param[in] uname  If not NULL, node name to search for
 499  * \param[in] flags  Bitmask of enum crm_get_peer_flags
 500  *
 501  * \return Node cache entry if found, otherwise NULL
 502  */
 503 crm_node_t *
 504 pcmk__search_node_caches(unsigned int id, const char *uname, uint32_t flags)
     /*  */
 505 {
 506     crm_node_t *node = NULL;
 507 
 508     CRM_ASSERT(id > 0 || uname != NULL);
 509 
 510     crm_peer_init();
 511 
 512     if ((uname != NULL) && pcmk_is_set(flags, CRM_GET_PEER_REMOTE)) {
 513         node = g_hash_table_lookup(crm_remote_peer_cache, uname);
 514     }
 515 
 516     if ((node == NULL) && pcmk_is_set(flags, CRM_GET_PEER_CLUSTER)) {
 517         node = pcmk__search_cluster_node_cache(id, uname);
 518     }
 519     return node;
 520 }
 521 
 522 /*!
 523  * \brief Get a node cache entry (cluster or Pacemaker Remote)
 524  *
 525  * \param[in] id     If not 0, cluster node ID to search for
 526  * \param[in] uname  If not NULL, node name to search for
 527  * \param[in] flags  Bitmask of enum crm_get_peer_flags
 528  *
 529  * \return (Possibly newly created) node cache entry
 530  */
 531 crm_node_t *
 532 crm_get_peer_full(unsigned int id, const char *uname, int flags)
     /*  */
 533 {
 534     crm_node_t *node = NULL;
 535 
 536     CRM_ASSERT(id > 0 || uname != NULL);
 537 
 538     crm_peer_init();
 539 
 540     if (pcmk_is_set(flags, CRM_GET_PEER_REMOTE)) {
 541         node = g_hash_table_lookup(crm_remote_peer_cache, uname);
 542     }
 543 
 544     if ((node == NULL) && pcmk_is_set(flags, CRM_GET_PEER_CLUSTER)) {
 545         node = crm_get_peer(id, uname);
 546     }
 547     return node;
 548 }
 549 
 550 /*!
 551  * \internal
 552  * \brief Search cluster node cache
 553  *
 554  * \param[in] id     If not 0, cluster node ID to search for
 555  * \param[in] uname  If not NULL, node name to search for
 556  *
 557  * \return Cluster node cache entry if found, otherwise NULL
 558  */
 559 crm_node_t *
 560 pcmk__search_cluster_node_cache(unsigned int id, const char *uname)
     /*  */
 561 {
 562     GHashTableIter iter;
 563     crm_node_t *node = NULL;
 564     crm_node_t *by_id = NULL;
 565     crm_node_t *by_name = NULL;
 566 
 567     CRM_ASSERT(id > 0 || uname != NULL);
 568 
 569     crm_peer_init();
 570 
 571     if (uname != NULL) {
 572         g_hash_table_iter_init(&iter, crm_peer_cache);
 573         while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &node)) {
 574             if(node->uname && strcasecmp(node->uname, uname) == 0) {
 575                 crm_trace("Name match: %s = %p", node->uname, node);
 576                 by_name = node;
 577                 break;
 578             }
 579         }
 580     }
 581 
 582     if (id > 0) {
 583         g_hash_table_iter_init(&iter, crm_peer_cache);
 584         while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &node)) {
 585             if(node->id == id) {
 586                 crm_trace("ID match: %u = %p", node->id, node);
 587                 by_id = node;
 588                 break;
 589             }
 590         }
 591     }
 592 
 593     node = by_id; /* Good default */
 594     if(by_id == by_name) {
 595         /* Nothing to do if they match (both NULL counts) */
 596         crm_trace("Consistent: %p for %u/%s", by_id, id, uname);
 597 
 598     } else if(by_id == NULL && by_name) {
 599         crm_trace("Only one: %p for %u/%s", by_name, id, uname);
 600 
 601         if(id && by_name->id) {
 602             dump_peer_hash(LOG_WARNING, __func__);
 603             crm_crit("Node %u and %u share the same name '%s'",
 604                      id, by_name->id, uname);
 605             node = NULL; /* Create a new one */
 606 
 607         } else {
 608             node = by_name;
 609         }
 610 
 611     } else if(by_name == NULL && by_id) {
 612         crm_trace("Only one: %p for %u/%s", by_id, id, uname);
 613 
 614         if(uname && by_id->uname) {
 615             dump_peer_hash(LOG_WARNING, __func__);
 616             crm_crit("Node '%s' and '%s' share the same cluster nodeid %u: assuming '%s' is correct",
 617                      uname, by_id->uname, id, uname);
 618         }
 619 
 620     } else if(uname && by_id->uname) {
 621         if(pcmk__str_eq(uname, by_id->uname, pcmk__str_casei)) {
 622             crm_notice("Node '%s' has changed its ID from %u to %u", by_id->uname, by_name->id, by_id->id);
 623             g_hash_table_foreach_remove(crm_peer_cache, hash_find_by_data, by_name);
 624 
 625         } else {
 626             crm_warn("Node '%s' and '%s' share the same cluster nodeid: %u %s", by_id->uname, by_name->uname, id, uname);
 627             dump_peer_hash(LOG_INFO, __func__);
 628             crm_abort(__FILE__, __func__, __LINE__, "member weirdness", TRUE,
 629                       TRUE);
 630         }
 631 
 632     } else if(id && by_name->id) {
 633         crm_warn("Node %u and %u share the same name: '%s'", by_id->id, by_name->id, uname);
 634 
 635     } else {
 636         /* Simple merge */
 637 
 638         /* Only corosync-based clusters use node IDs. The functions that call
 639          * pcmk__update_peer_state() and crm_update_peer_proc() only know
 640          * nodeid, so 'by_id' is authoritative when merging.
 641          */
 642         dump_peer_hash(LOG_DEBUG, __func__);
 643 
 644         crm_info("Merging %p into %p", by_name, by_id);
 645         g_hash_table_foreach_remove(crm_peer_cache, hash_find_by_data, by_name);
 646     }
 647 
 648     return node;
 649 }
 650 
 651 #if SUPPORT_COROSYNC
 652 static guint
 653 remove_conflicting_peer(crm_node_t *node)
     /*  */
 654 {
 655     int matches = 0;
 656     GHashTableIter iter;
 657     crm_node_t *existing_node = NULL;
 658 
 659     if (node->id == 0 || node->uname == NULL) {
 660         return 0;
 661     }
 662 
 663     if (!pcmk__corosync_has_nodelist()) {
 664         return 0;
 665     }
 666 
 667     g_hash_table_iter_init(&iter, crm_peer_cache);
 668     while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &existing_node)) {
 669         if (existing_node->id > 0
 670             && existing_node->id != node->id
 671             && existing_node->uname != NULL
 672             && strcasecmp(existing_node->uname, node->uname) == 0) {
 673 
 674             if (crm_is_peer_active(existing_node)) {
 675                 continue;
 676             }
 677 
 678             crm_warn("Removing cached offline node %u/%s which has conflicting uname with %u",
 679                      existing_node->id, existing_node->uname, node->id);
 680 
 681             g_hash_table_iter_remove(&iter);
 682             matches++;
 683         }
 684     }
 685 
 686     return matches;
 687 }
 688 #endif
 689 
 690 /*!
 691  * \brief Get a cluster node cache entry
 692  *
 693  * \param[in] id     If not 0, cluster node ID to search for
 694  * \param[in] uname  If not NULL, node name to search for
 695  *
 696  * \return (Possibly newly created) cluster node cache entry
 697  */
 698 /* coverity[-alloc] Memory is referenced in one or both hashtables */
 699 crm_node_t *
 700 crm_get_peer(unsigned int id, const char *uname)
     /*  */
 701 {
 702     crm_node_t *node = NULL;
 703     char *uname_lookup = NULL;
 704 
 705     CRM_ASSERT(id > 0 || uname != NULL);
 706 
 707     crm_peer_init();
 708 
 709     node = pcmk__search_cluster_node_cache(id, uname);
 710 
 711     /* if uname wasn't provided, and find_peer did not turn up a uname based on id.
 712      * we need to do a lookup of the node name using the id in the cluster membership. */
 713     if ((node == NULL || node->uname == NULL) && (uname == NULL)) { 
 714         uname_lookup = get_node_name(id);
 715     }
 716 
 717     if (uname_lookup) {
 718         uname = uname_lookup;
 719         crm_trace("Inferred a name of '%s' for node %u", uname, id);
 720 
 721         /* try to turn up the node one more time now that we know the uname. */
 722         if (node == NULL) {
 723             node = pcmk__search_cluster_node_cache(id, uname);
 724         }
 725     }
 726 
 727 
 728     if (node == NULL) {
 729         char *uniqueid = crm_generate_uuid();
 730 
 731         node = calloc(1, sizeof(crm_node_t));
 732         CRM_ASSERT(node);
 733 
 734         crm_info("Created entry %s/%p for node %s/%u (%d total)",
 735                  uniqueid, node, uname, id, 1 + g_hash_table_size(crm_peer_cache));
 736         g_hash_table_replace(crm_peer_cache, uniqueid, node);
 737     }
 738 
 739     if(id > 0 && uname && (node->id == 0 || node->uname == NULL)) {
 740         crm_info("Node %u is now known as %s", id, uname);
 741     }
 742 
 743     if(id > 0 && node->id == 0) {
 744         node->id = id;
 745     }
 746 
 747     if (uname && (node->uname == NULL)) {
 748         update_peer_uname(node, uname);
 749     }
 750 
 751     if(node->uuid == NULL) {
 752         const char *uuid = crm_peer_uuid(node);
 753 
 754         if (uuid) {
 755             crm_info("Node %u has uuid %s", id, uuid);
 756 
 757         } else {
 758             crm_info("Cannot obtain a UUID for node %u/%s", id, node->uname);
 759         }
 760     }
 761 
 762     free(uname_lookup);
 763 
 764     return node;
 765 }
 766 
 767 /*!
 768  * \internal
 769  * \brief Update a node's uname
 770  *
 771  * \param[in] node        Node object to update
 772  * \param[in] uname       New name to set
 773  *
 774  * \note This function should not be called within a peer cache iteration,
 775  *       because in some cases it can remove conflicting cache entries,
 776  *       which would invalidate the iterator.
 777  */
 778 static void
 779 update_peer_uname(crm_node_t *node, const char *uname)
     /*  */
 780 {
 781     CRM_CHECK(uname != NULL,
 782               crm_err("Bug: can't update node name without name"); return);
 783     CRM_CHECK(node != NULL,
 784               crm_err("Bug: can't update node name to %s without node", uname);
 785               return);
 786 
 787     if (pcmk__str_eq(uname, node->uname, pcmk__str_casei)) {
 788         crm_debug("Node uname '%s' did not change", uname);
 789         return;
 790     }
 791 
 792     for (const char *c = uname; *c; ++c) {
 793         if ((*c >= 'A') && (*c <= 'Z')) {
 794             crm_warn("Node names with capitals are discouraged, consider changing '%s'",
 795                      uname);
 796             break;
 797         }
 798     }
 799 
 800     pcmk__str_update(&node->uname, uname);
 801 
 802     if (peer_status_callback != NULL) {
 803         peer_status_callback(crm_status_uname, node, NULL);
 804     }
 805 
 806 #if SUPPORT_COROSYNC
 807     if (is_corosync_cluster() && !pcmk_is_set(node->flags, crm_remote_node)) {
 808         remove_conflicting_peer(node);
 809     }
 810 #endif
 811 }
 812 
 813 /*!
 814  * \internal
 815  * \brief Get log-friendly string equivalent of a process flag
 816  *
 817  * \param[in] proc  Process flag
 818  *
 819  * \return Log-friendly string equivalent of \p proc
 820  */
 821 static inline const char *
 822 proc2text(enum crm_proc_flag proc)
     /*  */
 823 {
 824     const char *text = "unknown";
 825 
 826     switch (proc) {
 827         case crm_proc_none:
 828             text = "none";
 829             break;
 830         case crm_proc_based:
 831             text = "pacemaker-based";
 832             break;
 833         case crm_proc_controld:
 834             text = "pacemaker-controld";
 835             break;
 836         case crm_proc_schedulerd:
 837             text = "pacemaker-schedulerd";
 838             break;
 839         case crm_proc_execd:
 840             text = "pacemaker-execd";
 841             break;
 842         case crm_proc_attrd:
 843             text = "pacemaker-attrd";
 844             break;
 845         case crm_proc_fenced:
 846             text = "pacemaker-fenced";
 847             break;
 848         case crm_proc_cpg:
 849             text = "corosync-cpg";
 850             break;
 851     }
 852     return text;
 853 }
 854 
 855 /*!
 856  * \internal
 857  * \brief Update a node's process information (and potentially state)
 858  *
 859  * \param[in] source      Caller's function name (for log messages)
 860  * \param[in] node        Node object to update
 861  * \param[in] flag        Bitmask of new process information
 862  * \param[in] status      node status (online, offline, etc.)
 863  *
 864  * \return NULL if any node was reaped from peer caches, value of node otherwise
 865  *
 866  * \note If this function returns NULL, the supplied node object was likely
 867  *       freed and should not be used again. This function should not be
 868  *       called within a cache iteration if reaping is possible, otherwise
 869  *       reaping could invalidate the iterator.
 870  */
 871 crm_node_t *
 872 crm_update_peer_proc(const char *source, crm_node_t * node, uint32_t flag, const char *status)
     /*  */
 873 {
 874     uint32_t last = 0;
 875     gboolean changed = FALSE;
 876 
 877     CRM_CHECK(node != NULL, crm_err("%s: Could not set %s to %s for NULL",
 878                                     source, proc2text(flag), status);
 879                             return NULL);
 880 
 881     /* Pacemaker doesn't spawn processes on remote nodes */
 882     if (pcmk_is_set(node->flags, crm_remote_node)) {
 883         return node;
 884     }
 885 
 886     last = node->processes;
 887     if (status == NULL) {
 888         node->processes = flag;
 889         if (node->processes != last) {
 890             changed = TRUE;
 891         }
 892 
 893     } else if (pcmk__str_eq(status, ONLINESTATUS, pcmk__str_casei)) {
 894         if ((node->processes & flag) != flag) {
 895             node->processes = pcmk__set_flags_as(__func__, __LINE__,
 896                                                  LOG_TRACE, "Peer process",
 897                                                  node->uname, node->processes,
 898                                                  flag, "processes");
 899             changed = TRUE;
 900         }
 901 
 902     } else if (node->processes & flag) {
 903         node->processes = pcmk__clear_flags_as(__func__, __LINE__,
 904                                                LOG_TRACE, "Peer process",
 905                                                node->uname, node->processes,
 906                                                flag, "processes");
 907         changed = TRUE;
 908     }
 909 
 910     if (changed) {
 911         if (status == NULL && flag <= crm_proc_none) {
 912             crm_info("%s: Node %s[%u] - all processes are now offline", source, node->uname,
 913                      node->id);
 914         } else {
 915             crm_info("%s: Node %s[%u] - %s is now %s", source, node->uname, node->id,
 916                      proc2text(flag), status);
 917         }
 918 
 919         /* Call the client callback first, then update the peer state,
 920          * in case the node will be reaped
 921          */
 922         if (peer_status_callback != NULL) {
 923             peer_status_callback(crm_status_processes, node, &last);
 924         }
 925 
 926         /* The client callback shouldn't touch the peer caches,
 927          * but as a safety net, bail if the peer cache was destroyed.
 928          */
 929         if (crm_peer_cache == NULL) {
 930             return NULL;
 931         }
 932 
 933         if (crm_autoreap) {
 934             const char *peer_state = NULL;
 935 
 936             if (pcmk_is_set(node->processes, crm_get_cluster_proc())) {
 937                 peer_state = CRM_NODE_MEMBER;
 938             } else {
 939                 peer_state = CRM_NODE_LOST;
 940             }
 941             node = pcmk__update_peer_state(__func__, node, peer_state, 0);
 942         }
 943     } else {
 944         crm_trace("%s: Node %s[%u] - %s is unchanged (%s)", source, node->uname, node->id,
 945                   proc2text(flag), status);
 946     }
 947     return node;
 948 }
 949 
 950 /*!
 951  * \internal
 952  * \brief Update a cluster node cache entry's expected join state
 953  *
 954  * \param[in]     source    Caller's function name (for logging)
 955  * \param[in,out] node      Node to update
 956  * \param[in]     expected  Node's new join state
 957  */
 958 void
 959 pcmk__update_peer_expected(const char *source, crm_node_t *node,
     /*  */
 960                            const char *expected)
 961 {
 962     char *last = NULL;
 963     gboolean changed = FALSE;
 964 
 965     CRM_CHECK(node != NULL, crm_err("%s: Could not set 'expected' to %s", source, expected);
 966               return);
 967 
 968     /* Remote nodes don't participate in joins */
 969     if (pcmk_is_set(node->flags, crm_remote_node)) {
 970         return;
 971     }
 972 
 973     last = node->expected;
 974     if (expected != NULL && !pcmk__str_eq(node->expected, expected, pcmk__str_casei)) {
 975         node->expected = strdup(expected);
 976         changed = TRUE;
 977     }
 978 
 979     if (changed) {
 980         crm_info("%s: Node %s[%u] - expected state is now %s (was %s)", source, node->uname, node->id,
 981                  expected, last);
 982         free(last);
 983     } else {
 984         crm_trace("%s: Node %s[%u] - expected state is unchanged (%s)", source, node->uname,
 985                   node->id, expected);
 986     }
 987 }
 988 
 989 /*!
 990  * \internal
 991  * \brief Update a node's state and membership information
 992  *
 993  * \param[in] source      Caller's function name (for log messages)
 994  * \param[in] node        Node object to update
 995  * \param[in] state       Node's new state
 996  * \param[in] membership  Node's new membership ID
 997  * \param[in] iter        If not NULL, pointer to node's peer cache iterator
 998  *
 999  * \return NULL if any node was reaped, value of node otherwise
1000  *
1001  * \note If this function returns NULL, the supplied node object was likely
1002  *       freed and should not be used again. This function may be called from
1003  *       within a peer cache iteration if the iterator is supplied.
1004  */
1005 static crm_node_t *
1006 update_peer_state_iter(const char *source, crm_node_t *node, const char *state,
     /*  */
1007                        uint64_t membership, GHashTableIter *iter)
1008 {
1009     gboolean is_member;
1010 
1011     CRM_CHECK(node != NULL,
1012               crm_err("Could not set state for unknown host to %s"
1013                       CRM_XS " source=%s", state, source);
1014               return NULL);
1015 
1016     is_member = pcmk__str_eq(state, CRM_NODE_MEMBER, pcmk__str_casei);
1017     if (is_member) {
1018         node->when_lost = 0;
1019         if (membership) {
1020             node->last_seen = membership;
1021         }
1022     }
1023 
1024     if (state && !pcmk__str_eq(node->state, state, pcmk__str_casei)) {
1025         char *last = node->state;
1026 
1027         node->state = strdup(state);
1028         crm_notice("Node %s state is now %s " CRM_XS
1029                    " nodeid=%u previous=%s source=%s", node->uname, state,
1030                    node->id, (last? last : "unknown"), source);
1031         if (peer_status_callback != NULL) {
1032             peer_status_callback(crm_status_nstate, node, last);
1033         }
1034         free(last);
1035 
1036         if (crm_autoreap && !is_member
1037             && !pcmk_is_set(node->flags, crm_remote_node)) {
1038             /* We only autoreap from the peer cache, not the remote peer cache,
1039              * because the latter should be managed only by
1040              * crm_remote_peer_cache_refresh().
1041              */
1042             if(iter) {
1043                 crm_notice("Purged 1 peer with id=%u and/or uname=%s from the membership cache", node->id, node->uname);
1044                 g_hash_table_iter_remove(iter);
1045 
1046             } else {
1047                 reap_crm_member(node->id, node->uname);
1048             }
1049             node = NULL;
1050         }
1051 
1052     } else {
1053         crm_trace("Node %s state is unchanged (%s) " CRM_XS
1054                   " nodeid=%u source=%s", node->uname, state, node->id, source);
1055     }
1056     return node;
1057 }
1058 
1059 /*!
1060  * \brief Update a node's state and membership information
1061  *
1062  * \param[in] source      Caller's function name (for log messages)
1063  * \param[in] node        Node object to update
1064  * \param[in] state       Node's new state
1065  * \param[in] membership  Node's new membership ID
1066  *
1067  * \return NULL if any node was reaped, value of node otherwise
1068  *
1069  * \note If this function returns NULL, the supplied node object was likely
1070  *       freed and should not be used again. This function should not be
1071  *       called within a cache iteration if reaping is possible,
1072  *       otherwise reaping could invalidate the iterator.
1073  */
1074 crm_node_t *
1075 pcmk__update_peer_state(const char *source, crm_node_t *node,
     /*  */
1076                         const char *state, uint64_t membership)
1077 {
1078     return update_peer_state_iter(source, node, state, membership, NULL);
1079 }
1080 
1081 /*!
1082  * \internal
1083  * \brief Reap all nodes from cache whose membership information does not match
1084  *
1085  * \param[in] membership  Membership ID of nodes to keep
1086  */
1087 void
1088 pcmk__reap_unseen_nodes(uint64_t membership)
     /*  */
1089 {
1090     GHashTableIter iter;
1091     crm_node_t *node = NULL;
1092 
1093     crm_trace("Reaping unseen nodes...");
1094     g_hash_table_iter_init(&iter, crm_peer_cache);
1095     while (g_hash_table_iter_next(&iter, NULL, (gpointer *)&node)) {
1096         if (node->last_seen != membership) {
1097             if (node->state) {
1098                 /*
1099                  * Calling update_peer_state_iter() allows us to
1100                  * remove the node from crm_peer_cache without
1101                  * invalidating our iterator
1102                  */
1103                 update_peer_state_iter(__func__, node, CRM_NODE_LOST,
1104                                            membership, &iter);
1105 
1106             } else {
1107                 crm_info("State of node %s[%u] is still unknown",
1108                          node->uname, node->id);
1109             }
1110         }
1111     }
1112 }
1113 
1114 static crm_node_t *
1115 find_known_node(const char *id, const char *uname)
     /*  */
1116 {
1117     GHashTableIter iter;
1118     crm_node_t *node = NULL;
1119     crm_node_t *by_id = NULL;
1120     crm_node_t *by_name = NULL;
1121 
1122     if (uname) {
1123         g_hash_table_iter_init(&iter, known_node_cache);
1124         while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &node)) {
1125             if (node->uname && strcasecmp(node->uname, uname) == 0) {
1126                 crm_trace("Name match: %s = %p", node->uname, node);
1127                 by_name = node;
1128                 break;
1129             }
1130         }
1131     }
1132 
1133     if (id) {
1134         g_hash_table_iter_init(&iter, known_node_cache);
1135         while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &node)) {
1136             if(strcasecmp(node->uuid, id) == 0) {
1137                 crm_trace("ID match: %s= %p", id, node);
1138                 by_id = node;
1139                 break;
1140             }
1141         }
1142     }
1143 
1144     node = by_id; /* Good default */
1145     if (by_id == by_name) {
1146         /* Nothing to do if they match (both NULL counts) */
1147         crm_trace("Consistent: %p for %s/%s", by_id, id, uname);
1148 
1149     } else if (by_id == NULL && by_name) {
1150         crm_trace("Only one: %p for %s/%s", by_name, id, uname);
1151 
1152         if (id) {
1153             node = NULL;
1154 
1155         } else {
1156             node = by_name;
1157         }
1158 
1159     } else if (by_name == NULL && by_id) {
1160         crm_trace("Only one: %p for %s/%s", by_id, id, uname);
1161 
1162         if (uname) {
1163             node = NULL;
1164         }
1165 
1166     } else if (uname && by_id->uname
1167                && pcmk__str_eq(uname, by_id->uname, pcmk__str_casei)) {
1168         /* Multiple nodes have the same uname in the CIB.
1169          * Return by_id. */
1170 
1171     } else if (id && by_name->uuid
1172                && pcmk__str_eq(id, by_name->uuid, pcmk__str_casei)) {
1173         /* Multiple nodes have the same id in the CIB.
1174          * Return by_name. */
1175         node = by_name;
1176 
1177     } else {
1178         node = NULL;
1179     }
1180 
1181     if (node == NULL) {
1182         crm_debug("Couldn't find node%s%s%s%s",
1183                    id? " " : "",
1184                    id? id : "",
1185                    uname? " with name " : "",
1186                    uname? uname : "");
1187     }
1188 
1189     return node;
1190 }
1191 
1192 static void
1193 known_node_cache_refresh_helper(xmlNode *xml_node, void *user_data)
     /*  */
1194 {
1195     const char *id = crm_element_value(xml_node, XML_ATTR_ID);
1196     const char *uname = crm_element_value(xml_node, XML_ATTR_UNAME);
1197     crm_node_t * node =  NULL;
1198 
1199     CRM_CHECK(id != NULL && uname !=NULL, return);
1200     node = find_known_node(id, uname);
1201 
1202     if (node == NULL) {
1203         char *uniqueid = crm_generate_uuid();
1204 
1205         node = calloc(1, sizeof(crm_node_t));
1206         CRM_ASSERT(node != NULL);
1207 
1208         node->uname = strdup(uname);
1209         CRM_ASSERT(node->uname != NULL);
1210 
1211         node->uuid = strdup(id);
1212         CRM_ASSERT(node->uuid != NULL);
1213 
1214         g_hash_table_replace(known_node_cache, uniqueid, node);
1215 
1216     } else if (pcmk_is_set(node->flags, crm_node_dirty)) {
1217         pcmk__str_update(&node->uname, uname);
1218 
1219         /* Node is in cache and hasn't been updated already, so mark it clean */
1220         clear_peer_flags(node, crm_node_dirty);
1221     }
1222 
1223 }
1224 
1225 static void
1226 refresh_known_node_cache(xmlNode *cib)
     /*  */
1227 {
1228     crm_peer_init();
1229 
1230     g_hash_table_foreach(known_node_cache, mark_dirty, NULL);
1231 
1232     crm_foreach_xpath_result(cib, PCMK__XP_MEMBER_NODE_CONFIG,
1233                              known_node_cache_refresh_helper, NULL);
1234 
1235     /* Remove all old cache entries that weren't seen in the CIB */
1236     g_hash_table_foreach_remove(known_node_cache, is_dirty, NULL);
1237 }
1238 
1239 void
1240 pcmk__refresh_node_caches_from_cib(xmlNode *cib)
     /*  */
1241 {
1242     crm_remote_peer_cache_refresh(cib);
1243     refresh_known_node_cache(cib);
1244 }
1245 
1246 /*!
1247  * \internal
1248  * \brief Search known node cache
1249  *
1250  * \param[in] id     If not 0, cluster node ID to search for
1251  * \param[in] uname  If not NULL, node name to search for
1252  * \param[in] flags  Bitmask of enum crm_get_peer_flags
1253  *
1254  * \return Known node cache entry if found, otherwise NULL
1255  */
1256 crm_node_t *
1257 pcmk__search_known_node_cache(unsigned int id, const char *uname,
     /*  */
1258                               uint32_t flags)
1259 {
1260     crm_node_t *node = NULL;
1261     char *id_str = NULL;
1262 
1263     CRM_ASSERT(id > 0 || uname != NULL);
1264 
1265     node = pcmk__search_node_caches(id, uname, flags);
1266 
1267     if (node || !(flags & CRM_GET_PEER_CLUSTER)) {
1268         return node;
1269     }
1270 
1271     if (id > 0) {
1272         id_str = crm_strdup_printf("%u", id);
1273     }
1274 
1275     node = find_known_node(id_str, uname);
1276 
1277     free(id_str);
1278     return node;
1279 }
1280 
1281 
1282 // Deprecated functions kept only for backward API compatibility
1283 // LCOV_EXCL_START
1284 
1285 #include <crm/cluster/compat.h>
1286 
1287 int
1288 crm_terminate_member(int nodeid, const char *uname, void *unused)
     /*  */
1289 {
1290     return stonith_api_kick(nodeid, uname, 120, TRUE);
1291 }
1292 
1293 int
1294 crm_terminate_member_no_mainloop(int nodeid, const char *uname, int *connection)
     /*  */
1295 {
1296     return stonith_api_kick(nodeid, uname, 120, TRUE);
1297 }
1298 
1299 // LCOV_EXCL_STOP
1300 // End deprecated API
/* */
root/lib/cluster/membership.c

DEFINITIONS
