root/daemons/controld/controld_execd.c

/* [previous][next][first][last][top][bottom][index][help] */

DEFINITIONS

This source file includes following definitions.
  1. lrm_connection_destroy
  2. make_stop_id
  3. copy_instance_keys
  4. copy_meta_keys
  5. history_remove_recurring_op
  6. history_free_recurring_ops
  7. history_free
  8. update_history_cache
  9. send_task_ok_ack
  10. op_node_name
  11. lrm_op_callback
  12. try_local_executor_connect
  13. do_lrm_control
  14. lrm_state_verify_stopped
  15. build_parameter_list
  16. append_restart_list
  17. append_secure_list
  18. build_operation_update
  19. is_rsc_active
  20. build_active_RAs
  21. do_lrm_query_internal
  22. controld_query_executor_state
  23. controld_rc2event
  24. controld_trigger_delete_refresh
  25. notify_deleted
  26. lrm_remove_deleted_rsc
  27. lrm_remove_deleted_op
  28. delete_rsc_entry
  29. erase_lrm_history_by_op
  30. erase_lrm_history_by_id
  31. last_failed_matches_op
  32. lrm_clear_last_failure
  33. cancel_op
  34. cancel_action_by_key
  35. cancel_op_key
  36. get_lrm_resource
  37. delete_resource
  38. get_fake_call_id
  39. fake_op_status
  40. force_reprobe
  41. synthesize_lrmd_failure
  42. lrm_op_target
  43. fail_lrm_resource
  44. handle_refresh_op
  45. handle_query_op
  46. handle_reprobe_op
  47. do_lrm_cancel
  48. do_lrm_delete
  49. do_lrm_invoke
  50. resolve_versioned_parameters
  51. construct_op
  52. controld_ack_event_directly
  53. verify_stopped
  54. stop_recurring_action_by_rsc
  55. stop_recurring_actions
  56. record_pending_op
  57. do_lrm_rsc_op
  58. cib_rsc_callback
  59. should_preserve_lock
  60. do_update_resource
  61. do_lrm_event
  62. unescape_newlines
  63. did_lrm_rsc_op_fail
  64. log_executor_event
  65. process_lrm_event

   1 /*
   2  * Copyright 2004-2021 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 
  12 #include <regex.h>
  13 #include <sys/param.h>
  14 #include <sys/types.h>
  15 #include <sys/wait.h>
  16 
  17 #include <crm/crm.h>
  18 #include <crm/lrmd.h>           // lrmd_event_data_t, lrmd_rsc_info_t, etc.
  19 #include <crm/services.h>
  20 #include <crm/msg_xml.h>
  21 #include <crm/common/xml.h>
  22 #include <crm/pengine/rules.h>
  23 #include <crm/lrmd_internal.h>
  24 
  25 #include <pacemaker-internal.h>
  26 #include <pacemaker-controld.h>
  27 
  28 #define START_DELAY_THRESHOLD 5 * 60 * 1000
  29 #define MAX_LRM_REG_FAILS 30
  30 
  31 struct delete_event_s {
  32     int rc;
  33     const char *rsc;
  34     lrm_state_t *lrm_state;
  35 };
  36 
  37 static gboolean is_rsc_active(lrm_state_t * lrm_state, const char *rsc_id);
  38 static gboolean build_active_RAs(lrm_state_t * lrm_state, xmlNode * rsc_list);
  39 static gboolean stop_recurring_actions(gpointer key, gpointer value, gpointer user_data);
  40 
  41 static lrmd_event_data_t *construct_op(lrm_state_t * lrm_state, xmlNode * rsc_op,
  42                                        const char *rsc_id, const char *operation);
  43 static void do_lrm_rsc_op(lrm_state_t *lrm_state, lrmd_rsc_info_t *rsc,
  44                           const char *operation, xmlNode *msg);
  45 
  46 static gboolean lrm_state_verify_stopped(lrm_state_t * lrm_state, enum crmd_fsa_state cur_state,
  47                                          int log_level);
  48 static int do_update_resource(const char *node_name, lrmd_rsc_info_t *rsc,
  49                               lrmd_event_data_t *op, time_t lock_time);
  50 
  51 static void
  52 lrm_connection_destroy(void)
     /* [previous][next][first][last][top][bottom][index][help] */
  53 {
  54     if (pcmk_is_set(fsa_input_register, R_LRM_CONNECTED)) {
  55         crm_crit("Connection to executor failed");
  56         register_fsa_input(C_FSA_INTERNAL, I_ERROR, NULL);
  57         controld_clear_fsa_input_flags(R_LRM_CONNECTED);
  58 
  59     } else {
  60         crm_info("Disconnected from executor");
  61     }
  62 
  63 }
  64 
  65 static char *
  66 make_stop_id(const char *rsc, int call_id)
     /* [previous][next][first][last][top][bottom][index][help] */
  67 {
  68     return crm_strdup_printf("%s:%d", rsc, call_id);
  69 }
  70 
  71 static void
  72 copy_instance_keys(gpointer key, gpointer value, gpointer user_data)
     /* [previous][next][first][last][top][bottom][index][help] */
  73 {
  74     if (strstr(key, CRM_META "_") == NULL) {
  75         g_hash_table_replace(user_data, strdup((const char *)key), strdup((const char *)value));
  76     }
  77 }
  78 
  79 static void
  80 copy_meta_keys(gpointer key, gpointer value, gpointer user_data)
     /* [previous][next][first][last][top][bottom][index][help] */
  81 {
  82     if (strstr(key, CRM_META "_") != NULL) {
  83         g_hash_table_replace(user_data, strdup((const char *)key), strdup((const char *)value));
  84     }
  85 }
  86 
  87 /*!
  88  * \internal
  89  * \brief Remove a recurring operation from a resource's history
  90  *
  91  * \param[in,out] history  Resource history to modify
  92  * \param[in]     op       Operation to remove
  93  *
  94  * \return TRUE if the operation was found and removed, FALSE otherwise
  95  */
  96 static gboolean
  97 history_remove_recurring_op(rsc_history_t *history, const lrmd_event_data_t *op)
     /* [previous][next][first][last][top][bottom][index][help] */
  98 {
  99     GList *iter;
 100 
 101     for (iter = history->recurring_op_list; iter != NULL; iter = iter->next) {
 102         lrmd_event_data_t *existing = iter->data;
 103 
 104         if ((op->interval_ms == existing->interval_ms)
 105             && pcmk__str_eq(op->rsc_id, existing->rsc_id, pcmk__str_none)
 106             && pcmk__str_eq(op->op_type, existing->op_type, pcmk__str_casei)) {
 107 
 108             history->recurring_op_list = g_list_delete_link(history->recurring_op_list, iter);
 109             lrmd_free_event(existing);
 110             return TRUE;
 111         }
 112     }
 113     return FALSE;
 114 }
 115 
 116 /*!
 117  * \internal
 118  * \brief Free all recurring operations in resource history
 119  *
 120  * \param[in,out] history  Resource history to modify
 121  */
 122 static void
 123 history_free_recurring_ops(rsc_history_t *history)
     /* [previous][next][first][last][top][bottom][index][help] */
 124 {
 125     GList *iter;
 126 
 127     for (iter = history->recurring_op_list; iter != NULL; iter = iter->next) {
 128         lrmd_free_event(iter->data);
 129     }
 130     g_list_free(history->recurring_op_list);
 131     history->recurring_op_list = NULL;
 132 }
 133 
 134 /*!
 135  * \internal
 136  * \brief Free resource history
 137  *
 138  * \param[in,out] history  Resource history to free
 139  */
 140 void
 141 history_free(gpointer data)
     /* [previous][next][first][last][top][bottom][index][help] */
 142 {
 143     rsc_history_t *history = (rsc_history_t*)data;
 144 
 145     if (history->stop_params) {
 146         g_hash_table_destroy(history->stop_params);
 147     }
 148 
 149     /* Don't need to free history->rsc.id because it's set to history->id */
 150     free(history->rsc.type);
 151     free(history->rsc.standard);
 152     free(history->rsc.provider);
 153 
 154     lrmd_free_event(history->failed);
 155     lrmd_free_event(history->last);
 156     free(history->id);
 157     history_free_recurring_ops(history);
 158     free(history);
 159 }
 160 
 161 static void
 162 update_history_cache(lrm_state_t * lrm_state, lrmd_rsc_info_t * rsc, lrmd_event_data_t * op)
     /* [previous][next][first][last][top][bottom][index][help] */
 163 {
 164     int target_rc = 0;
 165     rsc_history_t *entry = NULL;
 166 
 167     if (op->rsc_deleted) {
 168         crm_debug("Purged history for '%s' after %s", op->rsc_id, op->op_type);
 169         controld_delete_resource_history(op->rsc_id, lrm_state->node_name,
 170                                          NULL, crmd_cib_smart_opt());
 171         return;
 172     }
 173 
 174     if (pcmk__str_eq(op->op_type, RSC_NOTIFY, pcmk__str_casei)) {
 175         return;
 176     }
 177 
 178     crm_debug("Updating history for '%s' with %s op", op->rsc_id, op->op_type);
 179 
 180     entry = g_hash_table_lookup(lrm_state->resource_history, op->rsc_id);
 181     if (entry == NULL && rsc) {
 182         entry = calloc(1, sizeof(rsc_history_t));
 183         entry->id = strdup(op->rsc_id);
 184         g_hash_table_insert(lrm_state->resource_history, entry->id, entry);
 185 
 186         entry->rsc.id = entry->id;
 187         entry->rsc.type = strdup(rsc->type);
 188         entry->rsc.standard = strdup(rsc->standard);
 189         if (rsc->provider) {
 190             entry->rsc.provider = strdup(rsc->provider);
 191         } else {
 192             entry->rsc.provider = NULL;
 193         }
 194 
 195     } else if (entry == NULL) {
 196         crm_info("Resource %s no longer exists, not updating cache", op->rsc_id);
 197         return;
 198     }
 199 
 200     entry->last_callid = op->call_id;
 201     target_rc = rsc_op_expected_rc(op);
 202     if (op->op_status == PCMK_EXEC_CANCELLED) {
 203         if (op->interval_ms > 0) {
 204             crm_trace("Removing cancelled recurring op: " PCMK__OP_FMT,
 205                       op->rsc_id, op->op_type, op->interval_ms);
 206             history_remove_recurring_op(entry, op);
 207             return;
 208         } else {
 209             crm_trace("Skipping " PCMK__OP_FMT " rc=%d, status=%d",
 210                       op->rsc_id, op->op_type, op->interval_ms, op->rc,
 211                       op->op_status);
 212         }
 213 
 214     } else if (did_rsc_op_fail(op, target_rc)) {
 215         /* Store failed monitors here, otherwise the block below will cause them
 216          * to be forgotten when a stop happens.
 217          */
 218         if (entry->failed) {
 219             lrmd_free_event(entry->failed);
 220         }
 221         entry->failed = lrmd_copy_event(op);
 222 
 223     } else if (op->interval_ms == 0) {
 224         if (entry->last) {
 225             lrmd_free_event(entry->last);
 226         }
 227         entry->last = lrmd_copy_event(op);
 228 
 229         if (op->params && pcmk__strcase_any_of(op->op_type, CRMD_ACTION_START,
 230                                                CRMD_ACTION_RELOAD,
 231                                                CRMD_ACTION_RELOAD_AGENT,
 232                                                CRMD_ACTION_STATUS, NULL)) {
 233             if (entry->stop_params) {
 234                 g_hash_table_destroy(entry->stop_params);
 235             }
 236             entry->stop_params = pcmk__strkey_table(free, free);
 237 
 238             g_hash_table_foreach(op->params, copy_instance_keys, entry->stop_params);
 239         }
 240     }
 241 
 242     if (op->interval_ms > 0) {
 243         /* Ensure there are no duplicates */
 244         history_remove_recurring_op(entry, op);
 245 
 246         crm_trace("Adding recurring op: " PCMK__OP_FMT,
 247                   op->rsc_id, op->op_type, op->interval_ms);
 248         entry->recurring_op_list = g_list_prepend(entry->recurring_op_list, lrmd_copy_event(op));
 249 
 250     } else if (entry->recurring_op_list && !pcmk__str_eq(op->op_type, RSC_STATUS, pcmk__str_casei)) {
 251         crm_trace("Dropping %d recurring ops because of: " PCMK__OP_FMT,
 252                   g_list_length(entry->recurring_op_list), op->rsc_id,
 253                   op->op_type, op->interval_ms);
 254         history_free_recurring_ops(entry);
 255     }
 256 }
 257 
 258 /*!
 259  * \internal
 260  * \brief Send a direct OK ack for a resource task
 261  *
 262  * \param[in] lrm_state  LRM connection
 263  * \param[in] input      Input message being ack'ed
 264  * \param[in] rsc_id     ID of affected resource
 265  * \param[in] rsc        Affected resource (if available)
 266  * \param[in] task       Operation task being ack'ed
 267  * \param[in] ack_host   Name of host to send ack to
 268  * \param[in] ack_sys    IPC system name to ack
 269  */
 270 static void
 271 send_task_ok_ack(lrm_state_t *lrm_state, ha_msg_input_t *input,
     /* [previous][next][first][last][top][bottom][index][help] */
 272                  const char *rsc_id, lrmd_rsc_info_t *rsc, const char *task,
 273                  const char *ack_host, const char *ack_sys)
 274 {
 275     lrmd_event_data_t *op = construct_op(lrm_state, input->xml, rsc_id, task);
 276 
 277     lrmd__set_result(op, PCMK_OCF_OK, PCMK_EXEC_DONE, NULL);
 278     controld_ack_event_directly(ack_host, ack_sys, rsc, op, rsc_id);
 279     lrmd_free_event(op);
 280 }
 281 
 282 static inline const char *
 283 op_node_name(lrmd_event_data_t *op)
     /* [previous][next][first][last][top][bottom][index][help] */
 284 {
 285     return op->remote_nodename? op->remote_nodename : fsa_our_uname;
 286 }
 287 
 288 void
 289 lrm_op_callback(lrmd_event_data_t * op)
     /* [previous][next][first][last][top][bottom][index][help] */
 290 {
 291     CRM_CHECK(op != NULL, return);
 292     switch (op->type) {
 293         case lrmd_event_disconnect:
 294             if (op->remote_nodename == NULL) {
 295                 /* If this is the local executor IPC connection, set the right
 296                  * bits in the controller when the connection goes down.
 297                  */
 298                 lrm_connection_destroy();
 299             }
 300             break;
 301 
 302         case lrmd_event_exec_complete:
 303             {
 304                 lrm_state_t *lrm_state = lrm_state_find(op_node_name(op));
 305 
 306                 CRM_ASSERT(lrm_state != NULL);
 307                 process_lrm_event(lrm_state, op, NULL, NULL);
 308             }
 309             break;
 310 
 311         default:
 312             break;
 313     }
 314 }
 315 
 316 static void
 317 try_local_executor_connect(long long action, fsa_data_t *msg_data,
     /* [previous][next][first][last][top][bottom][index][help] */
 318                            lrm_state_t *lrm_state)
 319 {
 320     int rc = pcmk_rc_ok;
 321 
 322     crm_debug("Connecting to the local executor");
 323 
 324     // If we can connect, great
 325     rc = controld_connect_local_executor(lrm_state);
 326     if (rc == pcmk_rc_ok) {
 327         controld_set_fsa_input_flags(R_LRM_CONNECTED);
 328         crm_info("Connection to the local executor established");
 329         return;
 330     }
 331 
 332     // Otherwise, if we can try again, set a timer to do so
 333     if (lrm_state->num_lrm_register_fails < MAX_LRM_REG_FAILS) {
 334         crm_warn("Failed to connect to the local executor %d time%s "
 335                  "(%d max): %s", lrm_state->num_lrm_register_fails,
 336                  pcmk__plural_s(lrm_state->num_lrm_register_fails),
 337                  MAX_LRM_REG_FAILS, pcmk_rc_str(rc));
 338         controld_start_timer(wait_timer);
 339         crmd_fsa_stall(FALSE);
 340         return;
 341     }
 342 
 343     // Otherwise give up
 344     crm_err("Failed to connect to the executor the max allowed "
 345             "%d time%s: %s", lrm_state->num_lrm_register_fails,
 346             pcmk__plural_s(lrm_state->num_lrm_register_fails),
 347             pcmk_rc_str(rc));
 348     register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL);
 349 }
 350 
 351 /*       A_LRM_CONNECT  */
 352 void
 353 do_lrm_control(long long action,
     /* [previous][next][first][last][top][bottom][index][help] */
 354                enum crmd_fsa_cause cause,
 355                enum crmd_fsa_state cur_state,
 356                enum crmd_fsa_input current_input, fsa_data_t * msg_data)
 357 {
 358     /* This only pertains to local executor connections. Remote connections are
 359      * handled as resources within the scheduler. Connecting and disconnecting
 360      * from remote executor instances is handled differently.
 361      */
 362 
 363     lrm_state_t *lrm_state = NULL;
 364 
 365     if(fsa_our_uname == NULL) {
 366         return; /* Nothing to do */
 367     }
 368     lrm_state = lrm_state_find_or_create(fsa_our_uname);
 369     if (lrm_state == NULL) {
 370         register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL);
 371         return;
 372     }
 373 
 374     if (action & A_LRM_DISCONNECT) {
 375         if (lrm_state_verify_stopped(lrm_state, cur_state, LOG_INFO) == FALSE) {
 376             if (action == A_LRM_DISCONNECT) {
 377                 crmd_fsa_stall(FALSE);
 378                 return;
 379             }
 380         }
 381 
 382         controld_clear_fsa_input_flags(R_LRM_CONNECTED);
 383         crm_info("Disconnecting from the executor");
 384         lrm_state_disconnect(lrm_state);
 385         lrm_state_reset_tables(lrm_state, FALSE);
 386         crm_notice("Disconnected from the executor");
 387     }
 388 
 389     if (action & A_LRM_CONNECT) {
 390         try_local_executor_connect(action, msg_data, lrm_state);
 391     }
 392 
 393     if (action & ~(A_LRM_CONNECT | A_LRM_DISCONNECT)) {
 394         crm_err("Unexpected action %s in %s", fsa_action2string(action),
 395                 __func__);
 396     }
 397 }
 398 
 399 static gboolean
 400 lrm_state_verify_stopped(lrm_state_t * lrm_state, enum crmd_fsa_state cur_state, int log_level)
     /* [previous][next][first][last][top][bottom][index][help] */
 401 {
 402     int counter = 0;
 403     gboolean rc = TRUE;
 404     const char *when = "lrm disconnect";
 405 
 406     GHashTableIter gIter;
 407     const char *key = NULL;
 408     rsc_history_t *entry = NULL;
 409     active_op_t *pending = NULL;
 410 
 411     crm_debug("Checking for active resources before exit");
 412 
 413     if (cur_state == S_TERMINATE) {
 414         log_level = LOG_ERR;
 415         when = "shutdown";
 416 
 417     } else if (pcmk_is_set(fsa_input_register, R_SHUTDOWN)) {
 418         when = "shutdown... waiting";
 419     }
 420 
 421     if (lrm_state->pending_ops && lrm_state_is_connected(lrm_state) == TRUE) {
 422         guint removed = g_hash_table_foreach_remove(
 423             lrm_state->pending_ops, stop_recurring_actions, lrm_state);
 424         guint nremaining = g_hash_table_size(lrm_state->pending_ops);
 425 
 426         if (removed || nremaining) {
 427             crm_notice("Stopped %u recurring operation%s at %s (%u remaining)",
 428                        removed, pcmk__plural_s(removed), when, nremaining);
 429         }
 430     }
 431 
 432     if (lrm_state->pending_ops) {
 433         g_hash_table_iter_init(&gIter, lrm_state->pending_ops);
 434         while (g_hash_table_iter_next(&gIter, NULL, (void **)&pending)) {
 435             /* Ignore recurring actions in the shutdown calculations */
 436             if (pending->interval_ms == 0) {
 437                 counter++;
 438             }
 439         }
 440     }
 441 
 442     if (counter > 0) {
 443         do_crm_log(log_level, "%d pending executor operation%s at %s",
 444                    counter, pcmk__plural_s(counter), when);
 445 
 446         if ((cur_state == S_TERMINATE)
 447             || !pcmk_is_set(fsa_input_register, R_SENT_RSC_STOP)) {
 448             g_hash_table_iter_init(&gIter, lrm_state->pending_ops);
 449             while (g_hash_table_iter_next(&gIter, (gpointer*)&key, (gpointer*)&pending)) {
 450                 do_crm_log(log_level, "Pending action: %s (%s)", key, pending->op_key);
 451             }
 452 
 453         } else {
 454             rc = FALSE;
 455         }
 456         return rc;
 457     }
 458 
 459     if (lrm_state->resource_history == NULL) {
 460         return rc;
 461     }
 462 
 463     if (pcmk_is_set(fsa_input_register, R_SHUTDOWN)) {
 464         /* At this point we're not waiting, we're just shutting down */
 465         when = "shutdown";
 466     }
 467 
 468     counter = 0;
 469     g_hash_table_iter_init(&gIter, lrm_state->resource_history);
 470     while (g_hash_table_iter_next(&gIter, NULL, (gpointer*)&entry)) {
 471         if (is_rsc_active(lrm_state, entry->id) == FALSE) {
 472             continue;
 473         }
 474 
 475         counter++;
 476         if (log_level == LOG_ERR) {
 477             crm_info("Found %s active at %s", entry->id, when);
 478         } else {
 479             crm_trace("Found %s active at %s", entry->id, when);
 480         }
 481         if (lrm_state->pending_ops) {
 482             GHashTableIter hIter;
 483 
 484             g_hash_table_iter_init(&hIter, lrm_state->pending_ops);
 485             while (g_hash_table_iter_next(&hIter, (gpointer*)&key, (gpointer*)&pending)) {
 486                 if (pcmk__str_eq(entry->id, pending->rsc_id, pcmk__str_none)) {
 487                     crm_notice("%sction %s (%s) incomplete at %s",
 488                                pending->interval_ms == 0 ? "A" : "Recurring a",
 489                                key, pending->op_key, when);
 490                 }
 491             }
 492         }
 493     }
 494 
 495     if (counter) {
 496         crm_err("%d resource%s active at %s",
 497                 counter, (counter == 1)? " was" : "s were", when);
 498     }
 499 
 500     return rc;
 501 }
 502 
 503 /*!
 504  * \internal
 505  * \brief Build XML and string of parameters meeting some criteria, for digest
 506  *
 507  * \param[in]  op              Executor event with parameter table to use
 508  * \param[in]  metadata        Parsed meta-data for executed resource agent
 509  * \param[in]  param_type      Flag used for selection criteria
 510  * \param[out] result          Will be set to newly created XML with selected
 511  *                             parameters as attributes
 512  *
 513  * \return Newly allocated space-separated string of parameter names
 514  * \note Selection criteria varies by param_type: for the restart digest, we
 515  *       want parameters that are *not* marked reloadable (OCF 1.1) or that
 516  *       *are* marked unique (pre-1.1), for both string and XML results; for the
 517  *       secure digest, we want parameters that *are* marked private for the
 518  *       string, but parameters that are *not* marked private for the XML.
 519  * \note It is the caller's responsibility to free the string return value with
 520  *       free() and the XML result with free_xml().
 521  */
 522 static char *
 523 build_parameter_list(const lrmd_event_data_t *op,
     /* [previous][next][first][last][top][bottom][index][help] */
 524                      const struct ra_metadata_s *metadata,
 525                      enum ra_param_flags_e param_type, xmlNode **result)
 526 {
 527     char *list = NULL;
 528     size_t len = 0;
 529 
 530     *result = create_xml_node(NULL, XML_TAG_PARAMS);
 531 
 532     for (GList *iter = metadata->ra_params; iter != NULL; iter = iter->next) {
 533         struct ra_param_s *param = (struct ra_param_s *) iter->data;
 534 
 535         bool accept_for_list = false;
 536         bool accept_for_xml = false;
 537 
 538         switch (param_type) {
 539             case ra_param_reloadable:
 540                 accept_for_list = !pcmk_is_set(param->rap_flags, param_type);
 541                 accept_for_xml = accept_for_list;
 542                 break;
 543 
 544             case ra_param_unique:
 545                 accept_for_list = pcmk_is_set(param->rap_flags, param_type);
 546                 accept_for_xml = accept_for_list;
 547                 break;
 548 
 549             case ra_param_private:
 550                 accept_for_list = pcmk_is_set(param->rap_flags, param_type);
 551                 accept_for_xml = !accept_for_list;
 552                 break;
 553         }
 554 
 555         if (accept_for_list) {
 556             crm_trace("Attr %s is %s", param->rap_name, ra_param_flag2text(param_type));
 557 
 558             if (list == NULL) {
 559                 // We will later search for " WORD ", so start list with a space
 560                 pcmk__add_word(&list, &len, " ");
 561             }
 562             pcmk__add_word(&list, &len, param->rap_name);
 563 
 564         } else {
 565             crm_trace("Rejecting %s for %s", param->rap_name, ra_param_flag2text(param_type));
 566         }
 567 
 568         if (accept_for_xml) {
 569             const char *v = g_hash_table_lookup(op->params, param->rap_name);
 570 
 571             if (v != NULL) {
 572                 crm_trace("Adding attr %s=%s to the xml result", param->rap_name, v);
 573                 crm_xml_add(*result, param->rap_name, v);
 574             }
 575         }
 576     }
 577 
 578     if (list != NULL) {
 579         // We will later search for " WORD ", so end list with a space
 580         pcmk__add_word(&list, &len, " ");
 581     }
 582     return list;
 583 }
 584 
 585 static void
 586 append_restart_list(lrmd_event_data_t *op, struct ra_metadata_s *metadata,
     /* [previous][next][first][last][top][bottom][index][help] */
 587                     xmlNode *update, const char *version)
 588 {
 589     char *list = NULL;
 590     char *digest = NULL;
 591     xmlNode *restart = NULL;
 592 
 593     CRM_LOG_ASSERT(op->params != NULL);
 594 
 595     if (op->interval_ms > 0) {
 596         /* monitors are not reloadable */
 597         return;
 598     }
 599 
 600     if (pcmk_is_set(metadata->ra_flags, ra_supports_reload_agent)) {
 601         // Add parameters not marked reloadable to the "op-force-restart" list
 602         list = build_parameter_list(op, metadata, ra_param_reloadable,
 603                                     &restart);
 604 
 605     } else if (pcmk_is_set(metadata->ra_flags, ra_supports_legacy_reload)) {
 606         /* @COMPAT pre-OCF-1.1 resource agents
 607          *
 608          * Before OCF 1.1, Pacemaker abused "unique=0" to indicate
 609          * reloadability. Add any parameters with unique="1" to the
 610          * "op-force-restart" list.
 611          */
 612         list = build_parameter_list(op, metadata, ra_param_unique, &restart);
 613 
 614     } else {
 615         // Resource does not support agent reloads
 616         return;
 617     }
 618 
 619     digest = calculate_operation_digest(restart, version);
 620     /* Add "op-force-restart" and "op-restart-digest" to indicate the resource supports reload,
 621      * no matter if it actually supports any parameters with unique="1"). */
 622     crm_xml_add(update, XML_LRM_ATTR_OP_RESTART, list? list: "");
 623     crm_xml_add(update, XML_LRM_ATTR_RESTART_DIGEST, digest);
 624 
 625     crm_trace("%s: %s, %s", op->rsc_id, digest, list);
 626     crm_log_xml_trace(restart, "restart digest source");
 627 
 628     free_xml(restart);
 629     free(digest);
 630     free(list);
 631 }
 632 
 633 static void
 634 append_secure_list(lrmd_event_data_t *op, struct ra_metadata_s *metadata,
     /* [previous][next][first][last][top][bottom][index][help] */
 635                    xmlNode *update, const char *version)
 636 {
 637     char *list = NULL;
 638     char *digest = NULL;
 639     xmlNode *secure = NULL;
 640 
 641     CRM_LOG_ASSERT(op->params != NULL);
 642 
 643     /*
 644      * To keep XML_LRM_ATTR_OP_SECURE short, we want it to contain the
 645      * secure parameters but XML_LRM_ATTR_SECURE_DIGEST to be based on
 646      * the insecure ones
 647      */
 648     list = build_parameter_list(op, metadata, ra_param_private, &secure);
 649 
 650     if (list != NULL) {
 651         digest = calculate_operation_digest(secure, version);
 652         crm_xml_add(update, XML_LRM_ATTR_OP_SECURE, list);
 653         crm_xml_add(update, XML_LRM_ATTR_SECURE_DIGEST, digest);
 654 
 655         crm_trace("%s: %s, %s", op->rsc_id, digest, list);
 656         crm_log_xml_trace(secure, "secure digest source");
 657     } else {
 658         crm_trace("%s: no secure parameters", op->rsc_id);
 659     }
 660 
 661     free_xml(secure);
 662     free(digest);
 663     free(list);
 664 }
 665 
 666 static gboolean
 667 build_operation_update(xmlNode * parent, lrmd_rsc_info_t * rsc, lrmd_event_data_t * op,
     /* [previous][next][first][last][top][bottom][index][help] */
 668                        const char *node_name, const char *src)
 669 {
 670     int target_rc = 0;
 671     xmlNode *xml_op = NULL;
 672     struct ra_metadata_s *metadata = NULL;
 673     const char *caller_version = NULL;
 674     lrm_state_t *lrm_state = NULL;
 675     uint32_t metadata_source = controld_metadata_from_agent;
 676 
 677     if (op == NULL) {
 678         return FALSE;
 679     }
 680 
 681     target_rc = rsc_op_expected_rc(op);
 682 
 683     /* there is a small risk in formerly mixed clusters that it will
 684      * be sub-optimal.
 685      *
 686      * however with our upgrade policy, the update we send should
 687      * still be completely supported anyway
 688      */
 689     caller_version = g_hash_table_lookup(op->params, XML_ATTR_CRM_VERSION);
 690     CRM_LOG_ASSERT(caller_version != NULL);
 691 
 692     if(caller_version == NULL) {
 693         caller_version = CRM_FEATURE_SET;
 694     }
 695 
 696     crm_trace("Building %s operation update with originator version: %s", op->rsc_id, caller_version);
 697     xml_op = pcmk__create_history_xml(parent, op, caller_version, target_rc,
 698                                       fsa_our_uname, src, LOG_DEBUG);
 699     if (xml_op == NULL) {
 700         return TRUE;
 701     }
 702 
 703     if ((rsc == NULL) || (op->params == NULL)
 704         || !crm_op_needs_metadata(rsc->standard, op->op_type)) {
 705 
 706         crm_trace("No digests needed for %s action on %s (params=%p rsc=%p)",
 707                   op->op_type, op->rsc_id, op->params, rsc);
 708         return TRUE;
 709     }
 710 
 711     lrm_state = lrm_state_find(node_name);
 712     if (lrm_state == NULL) {
 713         crm_warn("Cannot calculate digests for operation " PCMK__OP_FMT
 714                  " because we have no connection to executor for %s",
 715                  op->rsc_id, op->op_type, op->interval_ms, node_name);
 716         return TRUE;
 717     }
 718 
 719     /* Getting meta-data from cache is OK unless this is a successful start
 720      * action -- always refresh from the agent for those, in case the
 721      * resource agent was updated.
 722      *
 723      * @TODO Only refresh the meta-data after starts if the agent actually
 724      * changed (using something like inotify, or a hash or modification time of
 725      * the agent executable).
 726      */
 727     if ((op->op_status != PCMK_EXEC_DONE) || (op->rc != target_rc)
 728         || !pcmk__str_eq(op->op_type, CRMD_ACTION_START, pcmk__str_none)) {
 729         metadata_source |= controld_metadata_from_cache;
 730     }
 731     metadata = controld_get_rsc_metadata(lrm_state, rsc, metadata_source);
 732     if (metadata == NULL) {
 733         return TRUE;
 734     }
 735 
 736 #if ENABLE_VERSIONED_ATTRS
 737     crm_xml_add(xml_op, XML_ATTR_RA_VERSION, metadata->ra_version);
 738 #endif
 739 
 740     crm_trace("Including additional digests for %s:%s:%s",
 741               rsc->standard, rsc->provider, rsc->type);
 742     append_restart_list(op, metadata, xml_op, caller_version);
 743     append_secure_list(op, metadata, xml_op, caller_version);
 744 
 745     return TRUE;
 746 }
 747 
 748 static gboolean
 749 is_rsc_active(lrm_state_t * lrm_state, const char *rsc_id)
     /* [previous][next][first][last][top][bottom][index][help] */
 750 {
 751     rsc_history_t *entry = NULL;
 752 
 753     entry = g_hash_table_lookup(lrm_state->resource_history, rsc_id);
 754     if (entry == NULL || entry->last == NULL) {
 755         return FALSE;
 756     }
 757 
 758     crm_trace("Processing %s: %s.%d=%d", rsc_id, entry->last->op_type,
 759               entry->last->interval_ms, entry->last->rc);
 760     if (entry->last->rc == PCMK_OCF_OK && pcmk__str_eq(entry->last->op_type, CRMD_ACTION_STOP, pcmk__str_casei)) {
 761         return FALSE;
 762 
 763     } else if (entry->last->rc == PCMK_OCF_OK
 764                && pcmk__str_eq(entry->last->op_type, CRMD_ACTION_MIGRATE, pcmk__str_casei)) {
 765         // A stricter check is too complex ... leave that to the scheduler
 766         return FALSE;
 767 
 768     } else if (entry->last->rc == PCMK_OCF_NOT_RUNNING) {
 769         return FALSE;
 770 
 771     } else if ((entry->last->interval_ms == 0)
 772                && (entry->last->rc == PCMK_OCF_NOT_CONFIGURED)) {
 773         /* Badly configured resources can't be reliably stopped */
 774         return FALSE;
 775     }
 776 
 777     return TRUE;
 778 }
 779 
 780 static gboolean
 781 build_active_RAs(lrm_state_t * lrm_state, xmlNode * rsc_list)
     /* [previous][next][first][last][top][bottom][index][help] */
 782 {
 783     GHashTableIter iter;
 784     rsc_history_t *entry = NULL;
 785 
 786     g_hash_table_iter_init(&iter, lrm_state->resource_history);
 787     while (g_hash_table_iter_next(&iter, NULL, (void **)&entry)) {
 788 
 789         GList *gIter = NULL;
 790         xmlNode *xml_rsc = create_xml_node(rsc_list, XML_LRM_TAG_RESOURCE);
 791 
 792         crm_xml_add(xml_rsc, XML_ATTR_ID, entry->id);
 793         crm_xml_add(xml_rsc, XML_ATTR_TYPE, entry->rsc.type);
 794         crm_xml_add(xml_rsc, XML_AGENT_ATTR_CLASS, entry->rsc.standard);
 795         crm_xml_add(xml_rsc, XML_AGENT_ATTR_PROVIDER, entry->rsc.provider);
 796 
 797         if (entry->last && entry->last->params) {
 798             const char *container = g_hash_table_lookup(entry->last->params, CRM_META"_"XML_RSC_ATTR_CONTAINER);
 799             if (container) {
 800                 crm_trace("Resource %s is a part of container resource %s", entry->id, container);
 801                 crm_xml_add(xml_rsc, XML_RSC_ATTR_CONTAINER, container);
 802             }
 803         }
 804         build_operation_update(xml_rsc, &(entry->rsc), entry->failed, lrm_state->node_name,
 805                                __func__);
 806         build_operation_update(xml_rsc, &(entry->rsc), entry->last, lrm_state->node_name,
 807                                __func__);
 808         for (gIter = entry->recurring_op_list; gIter != NULL; gIter = gIter->next) {
 809             build_operation_update(xml_rsc, &(entry->rsc), gIter->data, lrm_state->node_name,
 810                                    __func__);
 811         }
 812     }
 813 
 814     return FALSE;
 815 }
 816 
 817 static xmlNode *
 818 do_lrm_query_internal(lrm_state_t *lrm_state, int update_flags)
     /* [previous][next][first][last][top][bottom][index][help] */
 819 {
 820     xmlNode *xml_state = NULL;
 821     xmlNode *xml_data = NULL;
 822     xmlNode *rsc_list = NULL;
 823     crm_node_t *peer = NULL;
 824 
 825     peer = crm_get_peer_full(0, lrm_state->node_name, CRM_GET_PEER_ANY);
 826     CRM_CHECK(peer != NULL, return NULL);
 827 
 828     xml_state = create_node_state_update(peer, update_flags, NULL,
 829                                          __func__);
 830     if (xml_state == NULL) {
 831         return NULL;
 832     }
 833 
 834     xml_data = create_xml_node(xml_state, XML_CIB_TAG_LRM);
 835     crm_xml_add(xml_data, XML_ATTR_ID, peer->uuid);
 836     rsc_list = create_xml_node(xml_data, XML_LRM_TAG_RESOURCES);
 837 
 838     /* Build a list of active (not always running) resources */
 839     build_active_RAs(lrm_state, rsc_list);
 840 
 841     crm_log_xml_trace(xml_state, "Current executor state");
 842 
 843     return xml_state;
 844 }
 845 
 846 xmlNode *
 847 controld_query_executor_state(const char *node_name)
     /* [previous][next][first][last][top][bottom][index][help] */
 848 {
 849     lrm_state_t *lrm_state = lrm_state_find(node_name);
 850 
 851     if (!lrm_state) {
 852         crm_err("Could not find executor state for node %s", node_name);
 853         return NULL;
 854     }
 855     return do_lrm_query_internal(lrm_state,
 856                                  node_update_cluster|node_update_peer);
 857 }
 858 
 859 /*!
 860  * \internal
 861  * \brief Map standard Pacemaker return code to operation status and OCF code
 862  *
 863  * \param[out] event  Executor event whose status and return code should be set
 864  * \param[in]  rc     Standard Pacemaker return code
 865  */
 866 void
 867 controld_rc2event(lrmd_event_data_t *event, int rc)
     /* [previous][next][first][last][top][bottom][index][help] */
 868 {
 869     /* This is called for cleanup requests from controller peers/clients, not
 870      * for resource actions, so no exit reason is needed.
 871      */
 872     switch (rc) {
 873         case pcmk_rc_ok:
 874             lrmd__set_result(event, PCMK_OCF_OK, PCMK_EXEC_DONE, NULL);
 875             break;
 876         case EACCES:
 877             lrmd__set_result(event, PCMK_OCF_INSUFFICIENT_PRIV,
 878                              PCMK_EXEC_ERROR, NULL);
 879             break;
 880         default:
 881             lrmd__set_result(event, PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_ERROR,
 882                              NULL);
 883             break;
 884     }
 885 }
 886 
 887 /*!
 888  * \internal
 889  * \brief Trigger a new transition after CIB status was deleted
 890  *
 891  * If a CIB status delete was not expected (as part of the transition graph),
 892  * trigger a new transition by updating the (arbitrary) "last-lrm-refresh"
 893  * cluster property.
 894  *
 895  * \param[in] from_sys  IPC name that requested the delete
 896  * \param[in] rsc_id    Resource whose status was deleted (for logging only)
 897  */
 898 void
 899 controld_trigger_delete_refresh(const char *from_sys, const char *rsc_id)
     /* [previous][next][first][last][top][bottom][index][help] */
 900 {
 901     if (!pcmk__str_eq(from_sys, CRM_SYSTEM_TENGINE, pcmk__str_casei)) {
 902         char *now_s = crm_strdup_printf("%lld", (long long) time(NULL));
 903 
 904         crm_debug("Triggering a refresh after %s cleaned %s", from_sys, rsc_id);
 905         update_attr_delegate(fsa_cib_conn, cib_none, XML_CIB_TAG_CRMCONFIG,
 906                              NULL, NULL, NULL, NULL, "last-lrm-refresh", now_s,
 907                              FALSE, NULL, NULL);
 908         free(now_s);
 909     }
 910 }
 911 
 912 static void
 913 notify_deleted(lrm_state_t * lrm_state, ha_msg_input_t * input, const char *rsc_id, int rc)
     /* [previous][next][first][last][top][bottom][index][help] */
 914 {
 915     lrmd_event_data_t *op = NULL;
 916     const char *from_sys = crm_element_value(input->msg, F_CRM_SYS_FROM);
 917     const char *from_host = crm_element_value(input->msg, F_CRM_HOST_FROM);
 918 
 919     crm_info("Notifying %s on %s that %s was%s deleted",
 920              from_sys, (from_host? from_host : "localhost"), rsc_id,
 921              ((rc == pcmk_ok)? "" : " not"));
 922     op = construct_op(lrm_state, input->xml, rsc_id, CRMD_ACTION_DELETE);
 923     controld_rc2event(op, pcmk_legacy2rc(rc));
 924     controld_ack_event_directly(from_host, from_sys, NULL, op, rsc_id);
 925     lrmd_free_event(op);
 926     controld_trigger_delete_refresh(from_sys, rsc_id);
 927 }
 928 
 929 static gboolean
 930 lrm_remove_deleted_rsc(gpointer key, gpointer value, gpointer user_data)
     /* [previous][next][first][last][top][bottom][index][help] */
 931 {
 932     struct delete_event_s *event = user_data;
 933     struct pending_deletion_op_s *op = value;
 934 
 935     if (pcmk__str_eq(event->rsc, op->rsc, pcmk__str_none)) {
 936         notify_deleted(event->lrm_state, op->input, event->rsc, event->rc);
 937         return TRUE;
 938     }
 939     return FALSE;
 940 }
 941 
 942 static gboolean
 943 lrm_remove_deleted_op(gpointer key, gpointer value, gpointer user_data)
     /* [previous][next][first][last][top][bottom][index][help] */
 944 {
 945     const char *rsc = user_data;
 946     active_op_t *pending = value;
 947 
 948     if (pcmk__str_eq(rsc, pending->rsc_id, pcmk__str_none)) {
 949         crm_info("Removing op %s:%d for deleted resource %s",
 950                  pending->op_key, pending->call_id, rsc);
 951         return TRUE;
 952     }
 953     return FALSE;
 954 }
 955 
 956 static void
 957 delete_rsc_entry(lrm_state_t * lrm_state, ha_msg_input_t * input, const char *rsc_id,
     /* [previous][next][first][last][top][bottom][index][help] */
 958                  GHashTableIter * rsc_gIter, int rc, const char *user_name)
 959 {
 960     struct delete_event_s event;
 961 
 962     CRM_CHECK(rsc_id != NULL, return);
 963 
 964     if (rc == pcmk_ok) {
 965         char *rsc_id_copy = strdup(rsc_id);
 966 
 967         if (rsc_gIter) {
 968             g_hash_table_iter_remove(rsc_gIter);
 969         } else {
 970             g_hash_table_remove(lrm_state->resource_history, rsc_id_copy);
 971         }
 972         controld_delete_resource_history(rsc_id_copy, lrm_state->node_name,
 973                                          user_name, crmd_cib_smart_opt());
 974         g_hash_table_foreach_remove(lrm_state->pending_ops, lrm_remove_deleted_op, rsc_id_copy);
 975         free(rsc_id_copy);
 976     }
 977 
 978     if (input) {
 979         notify_deleted(lrm_state, input, rsc_id, rc);
 980     }
 981 
 982     event.rc = rc;
 983     event.rsc = rsc_id;
 984     event.lrm_state = lrm_state;
 985     g_hash_table_foreach_remove(lrm_state->deletion_ops, lrm_remove_deleted_rsc, &event);
 986 }
 987 
 988 /*!
 989  * \internal
 990  * \brief Erase an LRM history entry from the CIB, given the operation data
 991  *
 992  * \param[in] lrm_state  LRM state of the desired node
 993  * \param[in] op         Operation whose history should be deleted
 994  */
 995 static void
 996 erase_lrm_history_by_op(lrm_state_t *lrm_state, lrmd_event_data_t *op)
     /* [previous][next][first][last][top][bottom][index][help] */
 997 {
 998     xmlNode *xml_top = NULL;
 999 
1000     CRM_CHECK(op != NULL, return);
1001 
1002     xml_top = create_xml_node(NULL, XML_LRM_TAG_RSC_OP);
1003     crm_xml_add_int(xml_top, XML_LRM_ATTR_CALLID, op->call_id);
1004     crm_xml_add(xml_top, XML_ATTR_TRANSITION_KEY, op->user_data);
1005 
1006     if (op->interval_ms > 0) {
1007         char *op_id = pcmk__op_key(op->rsc_id, op->op_type, op->interval_ms);
1008 
1009         /* Avoid deleting last_failure too (if it was a result of this recurring op failing) */
1010         crm_xml_add(xml_top, XML_ATTR_ID, op_id);
1011         free(op_id);
1012     }
1013 
1014     crm_debug("Erasing resource operation history for " PCMK__OP_FMT " (call=%d)",
1015               op->rsc_id, op->op_type, op->interval_ms, op->call_id);
1016 
1017     fsa_cib_conn->cmds->remove(fsa_cib_conn, XML_CIB_TAG_STATUS, xml_top,
1018                                cib_quorum_override);
1019 
1020     crm_log_xml_trace(xml_top, "op:cancel");
1021     free_xml(xml_top);
1022 }
1023 
1024 /* Define xpath to find LRM resource history entry by node and resource */
1025 #define XPATH_HISTORY                                   \
1026     "/" XML_TAG_CIB "/" XML_CIB_TAG_STATUS              \
1027     "/" XML_CIB_TAG_STATE "[@" XML_ATTR_UNAME "='%s']"  \
1028     "/" XML_CIB_TAG_LRM "/" XML_LRM_TAG_RESOURCES       \
1029     "/" XML_LRM_TAG_RESOURCE "[@" XML_ATTR_ID "='%s']"  \
1030     "/" XML_LRM_TAG_RSC_OP
1031 
1032 /* ... and also by operation key */
1033 #define XPATH_HISTORY_ID XPATH_HISTORY \
1034     "[@" XML_ATTR_ID "='%s']"
1035 
1036 /* ... and also by operation key and operation call ID */
1037 #define XPATH_HISTORY_CALL XPATH_HISTORY \
1038     "[@" XML_ATTR_ID "='%s' and @" XML_LRM_ATTR_CALLID "='%d']"
1039 
1040 /* ... and also by operation key and original operation key */
1041 #define XPATH_HISTORY_ORIG XPATH_HISTORY \
1042     "[@" XML_ATTR_ID "='%s' and @" XML_LRM_ATTR_TASK_KEY "='%s']"
1043 
1044 /*!
1045  * \internal
1046  * \brief Erase an LRM history entry from the CIB, given operation identifiers
1047  *
1048  * \param[in] lrm_state  LRM state of the node to clear history for
1049  * \param[in] rsc_id     Name of resource to clear history for
1050  * \param[in] key        Operation key of operation to clear history for
1051  * \param[in] orig_op    If specified, delete only if it has this original op
1052  * \param[in] call_id    If specified, delete entry only if it has this call ID
1053  */
1054 static void
1055 erase_lrm_history_by_id(lrm_state_t *lrm_state, const char *rsc_id,
     /* [previous][next][first][last][top][bottom][index][help] */
1056                         const char *key, const char *orig_op, int call_id)
1057 {
1058     char *op_xpath = NULL;
1059 
1060     CRM_CHECK((rsc_id != NULL) && (key != NULL), return);
1061 
1062     if (call_id > 0) {
1063         op_xpath = crm_strdup_printf(XPATH_HISTORY_CALL,
1064                                      lrm_state->node_name, rsc_id, key,
1065                                      call_id);
1066 
1067     } else if (orig_op) {
1068         op_xpath = crm_strdup_printf(XPATH_HISTORY_ORIG,
1069                                      lrm_state->node_name, rsc_id, key,
1070                                      orig_op);
1071     } else {
1072         op_xpath = crm_strdup_printf(XPATH_HISTORY_ID,
1073                                      lrm_state->node_name, rsc_id, key);
1074     }
1075 
1076     crm_debug("Erasing resource operation history for %s on %s (call=%d)",
1077               key, rsc_id, call_id);
1078     fsa_cib_conn->cmds->remove(fsa_cib_conn, op_xpath, NULL,
1079                                cib_quorum_override | cib_xpath);
1080     free(op_xpath);
1081 }
1082 
1083 static inline gboolean
1084 last_failed_matches_op(rsc_history_t *entry, const char *op, guint interval_ms)
     /* [previous][next][first][last][top][bottom][index][help] */
1085 {
1086     if (entry == NULL) {
1087         return FALSE;
1088     }
1089     if (op == NULL) {
1090         return TRUE;
1091     }
1092     return (pcmk__str_eq(op, entry->failed->op_type, pcmk__str_casei)
1093             && (interval_ms == entry->failed->interval_ms));
1094 }
1095 
1096 /*!
1097  * \internal
1098  * \brief Clear a resource's last failure
1099  *
1100  * Erase a resource's last failure on a particular node from both the
1101  * LRM resource history in the CIB, and the resource history remembered
1102  * for the LRM state.
1103  *
1104  * \param[in] rsc_id      Resource name
1105  * \param[in] node_name   Node name
1106  * \param[in] operation   If specified, only clear if matching this operation
1107  * \param[in] interval_ms If operation is specified, it has this interval
1108  */
1109 void
1110 lrm_clear_last_failure(const char *rsc_id, const char *node_name,
     /* [previous][next][first][last][top][bottom][index][help] */
1111                        const char *operation, guint interval_ms)
1112 {
1113     char *op_key = NULL;
1114     char *orig_op_key = NULL;
1115     lrm_state_t *lrm_state = NULL;
1116 
1117     lrm_state = lrm_state_find(node_name);
1118     if (lrm_state == NULL) {
1119         return;
1120     }
1121 
1122     /* Erase from CIB */
1123     op_key = pcmk__op_key(rsc_id, "last_failure", 0);
1124     if (operation) {
1125         orig_op_key = pcmk__op_key(rsc_id, operation, interval_ms);
1126     }
1127     erase_lrm_history_by_id(lrm_state, rsc_id, op_key, orig_op_key, 0);
1128     free(op_key);
1129     free(orig_op_key);
1130 
1131     /* Remove from memory */
1132     if (lrm_state->resource_history) {
1133         rsc_history_t *entry = g_hash_table_lookup(lrm_state->resource_history,
1134                                                    rsc_id);
1135 
1136         if (last_failed_matches_op(entry, operation, interval_ms)) {
1137             lrmd_free_event(entry->failed);
1138             entry->failed = NULL;
1139         }
1140     }
1141 }
1142 
1143 /* Returns: gboolean - cancellation is in progress */
1144 static gboolean
1145 cancel_op(lrm_state_t * lrm_state, const char *rsc_id, const char *key, int op, gboolean remove)
     /* [previous][next][first][last][top][bottom][index][help] */
1146 {
1147     int rc = pcmk_ok;
1148     char *local_key = NULL;
1149     active_op_t *pending = NULL;
1150 
1151     CRM_CHECK(op != 0, return FALSE);
1152     CRM_CHECK(rsc_id != NULL, return FALSE);
1153     if (key == NULL) {
1154         local_key = make_stop_id(rsc_id, op);
1155         key = local_key;
1156     }
1157     pending = g_hash_table_lookup(lrm_state->pending_ops, key);
1158 
1159     if (pending) {
1160         if (remove && !pcmk_is_set(pending->flags, active_op_remove)) {
1161             controld_set_active_op_flags(pending, active_op_remove);
1162             crm_debug("Scheduling %s for removal", key);
1163         }
1164 
1165         if (pcmk_is_set(pending->flags, active_op_cancelled)) {
1166             crm_debug("Operation %s already cancelled", key);
1167             free(local_key);
1168             return FALSE;
1169         }
1170         controld_set_active_op_flags(pending, active_op_cancelled);
1171 
1172     } else {
1173         crm_info("No pending op found for %s", key);
1174         free(local_key);
1175         return FALSE;
1176     }
1177 
1178     crm_debug("Cancelling op %d for %s (%s)", op, rsc_id, key);
1179     rc = lrm_state_cancel(lrm_state, pending->rsc_id, pending->op_type,
1180                           pending->interval_ms);
1181     if (rc == pcmk_ok) {
1182         crm_debug("Op %d for %s (%s): cancelled", op, rsc_id, key);
1183         free(local_key);
1184         return TRUE;
1185     }
1186 
1187     crm_debug("Op %d for %s (%s): Nothing to cancel", op, rsc_id, key);
1188     /* The caller needs to make sure the entry is
1189      * removed from the pending_ops list
1190      *
1191      * Usually by returning TRUE inside the worker function
1192      * supplied to g_hash_table_foreach_remove()
1193      *
1194      * Not removing the entry from pending_ops will block
1195      * the node from shutting down
1196      */
1197     free(local_key);
1198     return FALSE;
1199 }
1200 
1201 struct cancel_data {
1202     gboolean done;
1203     gboolean remove;
1204     const char *key;
1205     lrmd_rsc_info_t *rsc;
1206     lrm_state_t *lrm_state;
1207 };
1208 
1209 static gboolean
1210 cancel_action_by_key(gpointer key, gpointer value, gpointer user_data)
     /* [previous][next][first][last][top][bottom][index][help] */
1211 {
1212     gboolean remove = FALSE;
1213     struct cancel_data *data = user_data;
1214     active_op_t *op = value;
1215 
1216     if (pcmk__str_eq(op->op_key, data->key, pcmk__str_none)) {
1217         data->done = TRUE;
1218         remove = !cancel_op(data->lrm_state, data->rsc->id, key, op->call_id, data->remove);
1219     }
1220     return remove;
1221 }
1222 
1223 static gboolean
1224 cancel_op_key(lrm_state_t * lrm_state, lrmd_rsc_info_t * rsc, const char *key, gboolean remove)
     /* [previous][next][first][last][top][bottom][index][help] */
1225 {
1226     guint removed = 0;
1227     struct cancel_data data;
1228 
1229     CRM_CHECK(rsc != NULL, return FALSE);
1230     CRM_CHECK(key != NULL, return FALSE);
1231 
1232     data.key = key;
1233     data.rsc = rsc;
1234     data.done = FALSE;
1235     data.remove = remove;
1236     data.lrm_state = lrm_state;
1237 
1238     removed = g_hash_table_foreach_remove(lrm_state->pending_ops, cancel_action_by_key, &data);
1239     crm_trace("Removed %u op cache entries, new size: %u",
1240               removed, g_hash_table_size(lrm_state->pending_ops));
1241     return data.done;
1242 }
1243 
1244 /*!
1245  * \internal
1246  * \brief Retrieve resource information from LRM
1247  *
1248  * \param[in]  lrm_state LRM connection to use
1249  * \param[in]  rsc_xml   XML containing resource configuration
1250  * \param[in]  do_create If true, register resource with LRM if not already
1251  * \param[out] rsc_info  Where to store resource information obtained from LRM
1252  *
1253  * \retval pcmk_ok   Success (and rsc_info holds newly allocated result)
1254  * \retval -EINVAL   Required information is missing from arguments
1255  * \retval -ENOTCONN No active connection to LRM
1256  * \retval -ENODEV   Resource not found
1257  * \retval -errno    Error communicating with executor when registering resource
1258  *
1259  * \note Caller is responsible for freeing result on success.
1260  */
1261 static int
1262 get_lrm_resource(lrm_state_t *lrm_state, xmlNode *rsc_xml, gboolean do_create,
     /* [previous][next][first][last][top][bottom][index][help] */
1263                  lrmd_rsc_info_t **rsc_info)
1264 {
1265     const char *id = ID(rsc_xml);
1266 
1267     CRM_CHECK(lrm_state && rsc_xml && rsc_info, return -EINVAL);
1268     CRM_CHECK(id, return -EINVAL);
1269 
1270     if (lrm_state_is_connected(lrm_state) == FALSE) {
1271         return -ENOTCONN;
1272     }
1273 
1274     crm_trace("Retrieving resource information for %s from the executor", id);
1275     *rsc_info = lrm_state_get_rsc_info(lrm_state, id, 0);
1276 
1277     // If resource isn't known by ID, try clone name, if provided
1278     if (!*rsc_info) {
1279         const char *long_id = crm_element_value(rsc_xml, XML_ATTR_ID_LONG);
1280 
1281         if (long_id) {
1282             *rsc_info = lrm_state_get_rsc_info(lrm_state, long_id, 0);
1283         }
1284     }
1285 
1286     if ((*rsc_info == NULL) && do_create) {
1287         const char *class = crm_element_value(rsc_xml, XML_AGENT_ATTR_CLASS);
1288         const char *provider = crm_element_value(rsc_xml, XML_AGENT_ATTR_PROVIDER);
1289         const char *type = crm_element_value(rsc_xml, XML_ATTR_TYPE);
1290         int rc;
1291 
1292         crm_trace("Registering resource %s with the executor", id);
1293         rc = lrm_state_register_rsc(lrm_state, id, class, provider, type,
1294                                     lrmd_opt_drop_recurring);
1295         if (rc != pcmk_ok) {
1296             fsa_data_t *msg_data = NULL;
1297 
1298             crm_err("Could not register resource %s with the executor on %s: %s "
1299                     CRM_XS " rc=%d",
1300                     id, lrm_state->node_name, pcmk_strerror(rc), rc);
1301 
1302             /* Register this as an internal error if this involves the local
1303              * executor. Otherwise, we're likely dealing with an unresponsive
1304              * remote node, which is not an FSA failure.
1305              */
1306             if (lrm_state_is_local(lrm_state) == TRUE) {
1307                 register_fsa_error(C_FSA_INTERNAL, I_FAIL, NULL);
1308             }
1309             return rc;
1310         }
1311 
1312         *rsc_info = lrm_state_get_rsc_info(lrm_state, id, 0);
1313     }
1314     return *rsc_info? pcmk_ok : -ENODEV;
1315 }
1316 
1317 static void
1318 delete_resource(lrm_state_t * lrm_state,
     /* [previous][next][first][last][top][bottom][index][help] */
1319                 const char *id,
1320                 lrmd_rsc_info_t * rsc,
1321                 GHashTableIter * gIter,
1322                 const char *sys,
1323                 const char *user,
1324                 ha_msg_input_t * request,
1325                 gboolean unregister)
1326 {
1327     int rc = pcmk_ok;
1328 
1329     crm_info("Removing resource %s from executor for %s%s%s",
1330              id, sys, (user? " as " : ""), (user? user : ""));
1331 
1332     if (rsc && unregister) {
1333         rc = lrm_state_unregister_rsc(lrm_state, id, 0);
1334     }
1335 
1336     if (rc == pcmk_ok) {
1337         crm_trace("Resource %s deleted from executor", id);
1338     } else if (rc == -EINPROGRESS) {
1339         crm_info("Deletion of resource '%s' from executor is pending", id);
1340         if (request) {
1341             struct pending_deletion_op_s *op = NULL;
1342             char *ref = crm_element_value_copy(request->msg, XML_ATTR_REFERENCE);
1343 
1344             op = calloc(1, sizeof(struct pending_deletion_op_s));
1345             op->rsc = strdup(rsc->id);
1346             op->input = copy_ha_msg_input(request);
1347             g_hash_table_insert(lrm_state->deletion_ops, ref, op);
1348         }
1349         return;
1350     } else {
1351         crm_warn("Could not delete '%s' from executor for %s%s%s: %s "
1352                  CRM_XS " rc=%d", id, sys, (user? " as " : ""),
1353                  (user? user : ""), pcmk_strerror(rc), rc);
1354     }
1355 
1356     delete_rsc_entry(lrm_state, request, id, gIter, rc, user);
1357 }
1358 
1359 static int
1360 get_fake_call_id(lrm_state_t *lrm_state, const char *rsc_id)
     /* [previous][next][first][last][top][bottom][index][help] */
1361 {
1362     int call_id = 999999999;
1363     rsc_history_t *entry = NULL;
1364 
1365     if(lrm_state) {
1366         entry = g_hash_table_lookup(lrm_state->resource_history, rsc_id);
1367     }
1368 
1369     /* Make sure the call id is greater than the last successful operation,
1370      * otherwise the failure will not result in a possible recovery of the resource
1371      * as it could appear the failure occurred before the successful start */
1372     if (entry) {
1373         call_id = entry->last_callid + 1;
1374     }
1375 
1376     if (call_id < 0) {
1377         call_id = 1;
1378     }
1379     return call_id;
1380 }
1381 
1382 static void
1383 fake_op_status(lrm_state_t *lrm_state, lrmd_event_data_t *op, int op_status,
     /* [previous][next][first][last][top][bottom][index][help] */
1384                enum ocf_exitcode op_exitcode, const char *exit_reason)
1385 {
1386     op->call_id = get_fake_call_id(lrm_state, op->rsc_id);
1387     op->t_run = time(NULL);
1388     op->t_rcchange = op->t_run;
1389     lrmd__set_result(op, op_exitcode, op_status, exit_reason);
1390 }
1391 
1392 static void
1393 force_reprobe(lrm_state_t *lrm_state, const char *from_sys,
     /* [previous][next][first][last][top][bottom][index][help] */
1394               const char *from_host, const char *user_name,
1395               gboolean is_remote_node)
1396 {
1397     GHashTableIter gIter;
1398     rsc_history_t *entry = NULL;
1399 
1400     crm_info("Clearing resource history on node %s", lrm_state->node_name);
1401     g_hash_table_iter_init(&gIter, lrm_state->resource_history);
1402     while (g_hash_table_iter_next(&gIter, NULL, (void **)&entry)) {
1403         /* only unregister the resource during a reprobe if it is not a remote connection
1404          * resource. otherwise unregistering the connection will terminate remote-node
1405          * membership */
1406         gboolean unregister = TRUE;
1407 
1408         if (is_remote_lrmd_ra(NULL, NULL, entry->id)) {
1409             lrm_state_t *remote_lrm_state = lrm_state_find(entry->id);
1410             if (remote_lrm_state) {
1411                 /* when forcing a reprobe, make sure to clear remote node before
1412                  * clearing the remote node's connection resource */ 
1413                 force_reprobe(remote_lrm_state, from_sys, from_host, user_name, TRUE);
1414             }
1415             unregister = FALSE;
1416         }
1417 
1418         delete_resource(lrm_state, entry->id, &entry->rsc, &gIter, from_sys,
1419                         user_name, NULL, unregister);
1420     }
1421 
1422     /* Now delete the copy in the CIB */
1423     controld_delete_node_state(lrm_state->node_name, controld_section_lrm,
1424                                cib_scope_local);
1425 
1426     /* Finally, _delete_ the value in pacemaker-attrd -- setting it to FALSE
1427      * would result in the scheduler sending us back here again
1428      */
1429     update_attrd(lrm_state->node_name, CRM_OP_PROBED, NULL, user_name, is_remote_node);
1430 }
1431 
1432 /*!
1433  * \internal
1434  * \brief Fail a requested action without actually executing it
1435  *
1436  * For an action that can't be executed, process it similarly to an actual
1437  * execution result, with specified error status (except for notify actions,
1438  * which will always be treated as successful).
1439  *
1440  * \param[in] lrm_state  Executor connection that action is for
1441  * \param[in] action     Action XML from request
1442  * \param[in] rc         Desired return code to use
1443  * \param[in] op_status  Desired operation status to use
1444  * \param[in] exit_reason  Human-friendly detail, if error
1445  */
1446 static void
1447 synthesize_lrmd_failure(lrm_state_t *lrm_state, xmlNode *action,
     /* [previous][next][first][last][top][bottom][index][help] */
1448                         int op_status, enum ocf_exitcode rc,
1449                         const char *exit_reason)
1450 {
1451     lrmd_event_data_t *op = NULL;
1452     const char *operation = crm_element_value(action, XML_LRM_ATTR_TASK);
1453     const char *target_node = crm_element_value(action, XML_LRM_ATTR_TARGET);
1454     xmlNode *xml_rsc = find_xml_node(action, XML_CIB_TAG_RESOURCE, TRUE);
1455 
1456     if ((xml_rsc == NULL) || (ID(xml_rsc) == NULL)) {
1457         /* @TODO Should we do something else, like direct ack? */
1458         crm_info("Can't fake %s failure (%d) on %s without resource configuration",
1459                  crm_element_value(action, XML_LRM_ATTR_TASK_KEY), rc,
1460                  target_node);
1461         return;
1462 
1463     } else if(operation == NULL) {
1464         /* This probably came from crm_resource -C, nothing to do */
1465         crm_info("Can't fake %s failure (%d) on %s without operation",
1466                  ID(xml_rsc), rc, target_node);
1467         return;
1468     }
1469 
1470     op = construct_op(lrm_state, action, ID(xml_rsc), operation);
1471 
1472     if (pcmk__str_eq(operation, RSC_NOTIFY, pcmk__str_casei)) { // Notifications can't fail
1473         fake_op_status(lrm_state, op, PCMK_EXEC_DONE, PCMK_OCF_OK, NULL);
1474     } else {
1475         fake_op_status(lrm_state, op, op_status, rc, exit_reason);
1476     }
1477 
1478     crm_info("Faking " PCMK__OP_FMT " result (%d) on %s",
1479              op->rsc_id, op->op_type, op->interval_ms, op->rc, target_node);
1480 
1481     // Process the result as if it came from the LRM
1482     process_lrm_event(lrm_state, op, NULL, action);
1483     lrmd_free_event(op);
1484 }
1485 
1486 /*!
1487  * \internal
1488  * \brief Get target of an LRM operation
1489  *
1490  * \param[in] xml  LRM operation data XML
1491  *
1492  * \return LRM operation target node name (local node or Pacemaker Remote node)
1493  */
1494 static const char *
1495 lrm_op_target(xmlNode *xml)
     /* [previous][next][first][last][top][bottom][index][help] */
1496 {
1497     const char *target = NULL;
1498 
1499     if (xml) {
1500         target = crm_element_value(xml, XML_LRM_ATTR_TARGET);
1501     }
1502     if (target == NULL) {
1503         target = fsa_our_uname;
1504     }
1505     return target;
1506 }
1507 
1508 static void
1509 fail_lrm_resource(xmlNode *xml, lrm_state_t *lrm_state, const char *user_name,
     /* [previous][next][first][last][top][bottom][index][help] */
1510                   const char *from_host, const char *from_sys)
1511 {
1512     lrmd_event_data_t *op = NULL;
1513     lrmd_rsc_info_t *rsc = NULL;
1514     xmlNode *xml_rsc = find_xml_node(xml, XML_CIB_TAG_RESOURCE, TRUE);
1515 
1516     CRM_CHECK(xml_rsc != NULL, return);
1517 
1518     /* The executor simply executes operations and reports the results, without
1519      * any concept of success or failure, so to fail a resource, we must fake
1520      * what a failure looks like.
1521      *
1522      * To do this, we create a fake executor operation event for the resource,
1523      * and pass that event to the executor client callback so it will be
1524      * processed as if it came from the executor.
1525      */
1526     op = construct_op(lrm_state, xml, ID(xml_rsc), "asyncmon");
1527 
1528     free((char*) op->user_data);
1529     op->user_data = NULL;
1530     op->interval_ms = 0;
1531 
1532     if (user_name && !pcmk__is_privileged(user_name)) {
1533         crm_err("%s does not have permission to fail %s", user_name, ID(xml_rsc));
1534         fake_op_status(lrm_state, op, PCMK_EXEC_ERROR,
1535                        PCMK_OCF_INSUFFICIENT_PRIV,
1536                        "Unprivileged user cannot fail resources");
1537         controld_ack_event_directly(from_host, from_sys, NULL, op, ID(xml_rsc));
1538         lrmd_free_event(op);
1539         return;
1540     }
1541 
1542 
1543     if (get_lrm_resource(lrm_state, xml_rsc, TRUE, &rsc) == pcmk_ok) {
1544         crm_info("Failing resource %s...", rsc->id);
1545         fake_op_status(lrm_state, op, PCMK_EXEC_DONE, PCMK_OCF_UNKNOWN_ERROR,
1546                        "Simulated failure");
1547         process_lrm_event(lrm_state, op, NULL, xml);
1548         op->rc = PCMK_OCF_OK; // The request to fail the resource succeeded
1549         lrmd_free_rsc_info(rsc);
1550 
1551     } else {
1552         crm_info("Cannot find/create resource in order to fail it...");
1553         crm_log_xml_warn(xml, "bad input");
1554         fake_op_status(lrm_state, op, PCMK_EXEC_ERROR, PCMK_OCF_UNKNOWN_ERROR,
1555                        "Cannot fail unknown resource");
1556     }
1557 
1558     controld_ack_event_directly(from_host, from_sys, NULL, op, ID(xml_rsc));
1559     lrmd_free_event(op);
1560 }
1561 
1562 static void
1563 handle_refresh_op(lrm_state_t *lrm_state, const char *user_name,
     /* [previous][next][first][last][top][bottom][index][help] */
1564                   const char *from_host, const char *from_sys)
1565 {
1566     int rc = pcmk_ok;
1567     xmlNode *fragment = do_lrm_query_internal(lrm_state, node_update_all);
1568 
1569     fsa_cib_update(XML_CIB_TAG_STATUS, fragment, cib_quorum_override, rc, user_name);
1570     crm_info("Forced a local resource history refresh: call=%d", rc);
1571 
1572     if (!pcmk__str_eq(CRM_SYSTEM_CRMD, from_sys, pcmk__str_casei)) {
1573         xmlNode *reply = create_request(CRM_OP_INVOKE_LRM, fragment, from_host,
1574                                         from_sys, CRM_SYSTEM_LRMD,
1575                                         fsa_our_uuid);
1576 
1577         crm_debug("ACK'ing refresh from %s (%s)", from_sys, from_host);
1578 
1579         if (relay_message(reply, TRUE) == FALSE) {
1580             crm_log_xml_err(reply, "Unable to route reply");
1581         }
1582         free_xml(reply);
1583     }
1584 
1585     free_xml(fragment);
1586 }
1587 
1588 static void
1589 handle_query_op(xmlNode *msg, lrm_state_t *lrm_state)
     /* [previous][next][first][last][top][bottom][index][help] */
1590 {
1591     xmlNode *data = do_lrm_query_internal(lrm_state, node_update_all);
1592     xmlNode *reply = create_reply(msg, data);
1593 
1594     if (relay_message(reply, TRUE) == FALSE) {
1595         crm_err("Unable to route reply");
1596         crm_log_xml_err(reply, "reply");
1597     }
1598     free_xml(reply);
1599     free_xml(data);
1600 }
1601 
1602 static void
1603 handle_reprobe_op(lrm_state_t *lrm_state, const char *from_sys,
     /* [previous][next][first][last][top][bottom][index][help] */
1604                   const char *from_host, const char *user_name,
1605                   gboolean is_remote_node)
1606 {
1607     crm_notice("Forcing the status of all resources to be redetected");
1608     force_reprobe(lrm_state, from_sys, from_host, user_name, is_remote_node);
1609 
1610     if (!pcmk__strcase_any_of(from_sys, CRM_SYSTEM_PENGINE, CRM_SYSTEM_TENGINE, NULL)) {
1611 
1612         xmlNode *reply = create_request(CRM_OP_INVOKE_LRM, NULL, from_host,
1613                                         from_sys, CRM_SYSTEM_LRMD,
1614                                         fsa_our_uuid);
1615 
1616         crm_debug("ACK'ing re-probe from %s (%s)", from_sys, from_host);
1617 
1618         if (relay_message(reply, TRUE) == FALSE) {
1619             crm_log_xml_err(reply, "Unable to route reply");
1620         }
1621         free_xml(reply);
1622     }
1623 }
1624 
1625 static bool do_lrm_cancel(ha_msg_input_t *input, lrm_state_t *lrm_state,
     /* [previous][next][first][last][top][bottom][index][help] */
1626               lrmd_rsc_info_t *rsc, const char *from_host, const char *from_sys)
1627 {
1628     char *op_key = NULL;
1629     char *meta_key = NULL;
1630     int call = 0;
1631     const char *call_id = NULL;
1632     const char *op_task = NULL;
1633     guint interval_ms = 0;
1634     gboolean in_progress = FALSE;
1635     xmlNode *params = find_xml_node(input->xml, XML_TAG_ATTRS, TRUE);
1636 
1637     CRM_CHECK(params != NULL, return FALSE);
1638 
1639     meta_key = crm_meta_name(XML_LRM_ATTR_TASK);
1640     op_task = crm_element_value(params, meta_key);
1641     free(meta_key);
1642     CRM_CHECK(op_task != NULL, return FALSE);
1643 
1644     meta_key = crm_meta_name(XML_LRM_ATTR_INTERVAL_MS);
1645     if (crm_element_value_ms(params, meta_key, &interval_ms) != pcmk_ok) {
1646         free(meta_key);
1647         return FALSE;
1648     }
1649     free(meta_key);
1650 
1651     op_key = pcmk__op_key(rsc->id, op_task, interval_ms);
1652 
1653     meta_key = crm_meta_name(XML_LRM_ATTR_CALLID);
1654     call_id = crm_element_value(params, meta_key);
1655     free(meta_key);
1656 
1657     crm_debug("Scheduler requested op %s (call=%s) be cancelled",
1658               op_key, (call_id? call_id : "NA"));
1659     pcmk__scan_min_int(call_id, &call, 0);
1660     if (call == 0) {
1661         // Normal case when the scheduler cancels a recurring op
1662         in_progress = cancel_op_key(lrm_state, rsc, op_key, TRUE);
1663 
1664     } else {
1665         // Normal case when the scheduler cancels an orphan op
1666         in_progress = cancel_op(lrm_state, rsc->id, NULL, call, TRUE);
1667     }
1668 
1669     // Acknowledge cancellation operation if for a remote connection resource
1670     if (!in_progress || is_remote_lrmd_ra(NULL, NULL, rsc->id)) {
1671         char *op_id = make_stop_id(rsc->id, call);
1672 
1673         if (is_remote_lrmd_ra(NULL, NULL, rsc->id) == FALSE) {
1674             crm_info("Nothing known about operation %d for %s", call, op_key);
1675         }
1676         erase_lrm_history_by_id(lrm_state, rsc->id, op_key, NULL, call);
1677         send_task_ok_ack(lrm_state, input, rsc->id, rsc, op_task,
1678                          from_host, from_sys);
1679 
1680         /* needed at least for cancellation of a remote operation */
1681         g_hash_table_remove(lrm_state->pending_ops, op_id);
1682         free(op_id);
1683 
1684     } else {
1685         /* No ack is needed since abcdaa8, but peers with older versions
1686          * in a rolling upgrade need one. We didn't bump the feature set
1687          * at that commit, so we can only compare against the previous
1688          * CRM version (3.0.8). If any peers have feature set 3.0.9 but
1689          * not abcdaa8, they will time out waiting for the ack (no
1690          * released versions of Pacemaker are affected).
1691          */
1692         const char *peer_version = crm_element_value(params, XML_ATTR_CRM_VERSION);
1693 
1694         if (compare_version(peer_version, "3.0.8") <= 0) {
1695             crm_info("Sending compatibility ack for %s cancellation to %s (CRM version %s)",
1696                      op_key, from_host, peer_version);
1697             send_task_ok_ack(lrm_state, input, rsc->id, rsc, op_task,
1698                              from_host, from_sys);
1699         }
1700     }
1701 
1702     free(op_key);
1703     return TRUE;
1704 }
1705 
1706 static void
1707 do_lrm_delete(ha_msg_input_t *input, lrm_state_t *lrm_state,
     /* [previous][next][first][last][top][bottom][index][help] */
1708               lrmd_rsc_info_t *rsc, const char *from_sys, const char *from_host,
1709               bool crm_rsc_delete, const char *user_name)
1710 {
1711     gboolean unregister = TRUE;
1712     int cib_rc = controld_delete_resource_history(rsc->id, lrm_state->node_name,
1713                                                   user_name,
1714                                                   cib_dryrun|cib_sync_call);
1715 
1716     if (cib_rc != pcmk_rc_ok) {
1717         lrmd_event_data_t *op = NULL;
1718 
1719         op = construct_op(lrm_state, input->xml, rsc->id, CRMD_ACTION_DELETE);
1720 
1721         /* These are resource clean-ups, not actions, so no exit reason is
1722          * needed.
1723          */
1724         lrmd__set_result(op, pcmk_rc2ocf(cib_rc), PCMK_EXEC_ERROR, NULL);
1725         controld_ack_event_directly(from_host, from_sys, NULL, op, rsc->id);
1726         lrmd_free_event(op);
1727         return;
1728     }
1729 
1730     if (crm_rsc_delete && is_remote_lrmd_ra(NULL, NULL, rsc->id)) {
1731         unregister = FALSE;
1732     }
1733 
1734     delete_resource(lrm_state, rsc->id, rsc, NULL, from_sys,
1735                     user_name, input, unregister);
1736 }
1737 
1738 /*       A_LRM_INVOKE   */
1739 void
1740 do_lrm_invoke(long long action,
     /* [previous][next][first][last][top][bottom][index][help] */
1741               enum crmd_fsa_cause cause,
1742               enum crmd_fsa_state cur_state,
1743               enum crmd_fsa_input current_input, fsa_data_t * msg_data)
1744 {
1745     lrm_state_t *lrm_state = NULL;
1746     const char *crm_op = NULL;
1747     const char *from_sys = NULL;
1748     const char *from_host = NULL;
1749     const char *operation = NULL;
1750     ha_msg_input_t *input = fsa_typed_data(fsa_dt_ha_msg);
1751     const char *user_name = NULL;
1752     const char *target_node = NULL;
1753     gboolean is_remote_node = FALSE;
1754     bool crm_rsc_delete = FALSE;
1755 
1756     target_node = lrm_op_target(input->xml);
1757     is_remote_node = !pcmk__str_eq(target_node, fsa_our_uname,
1758                                    pcmk__str_casei);
1759 
1760     lrm_state = lrm_state_find(target_node);
1761     if ((lrm_state == NULL) && is_remote_node) {
1762         crm_err("Failing action because local node has never had connection to remote node %s",
1763                 target_node);
1764         synthesize_lrmd_failure(NULL, input->xml, PCMK_EXEC_NOT_CONNECTED,
1765                                 PCMK_OCF_UNKNOWN_ERROR,
1766                                 "Local node has no connection to remote");
1767         return;
1768     }
1769     CRM_ASSERT(lrm_state != NULL);
1770 
1771     user_name = pcmk__update_acl_user(input->msg, F_CRM_USER, NULL);
1772     crm_op = crm_element_value(input->msg, F_CRM_TASK);
1773     from_sys = crm_element_value(input->msg, F_CRM_SYS_FROM);
1774     if (!pcmk__str_eq(from_sys, CRM_SYSTEM_TENGINE, pcmk__str_casei)) {
1775         from_host = crm_element_value(input->msg, F_CRM_HOST_FROM);
1776     }
1777     crm_trace("Executor %s command from %s as user %s",
1778               crm_op, from_sys, user_name);
1779 
1780     if (pcmk__str_eq(crm_op, CRM_OP_LRM_DELETE, pcmk__str_casei)) {
1781         if (!pcmk__str_eq(from_sys, CRM_SYSTEM_TENGINE, pcmk__str_casei)) {
1782             crm_rsc_delete = TRUE; // from crm_resource
1783         }
1784         operation = CRMD_ACTION_DELETE;
1785 
1786     } else if (pcmk__str_eq(crm_op, CRM_OP_LRM_FAIL, pcmk__str_casei)) {
1787         fail_lrm_resource(input->xml, lrm_state, user_name, from_host,
1788                           from_sys);
1789         return;
1790 
1791     } else if (input->xml != NULL) {
1792         operation = crm_element_value(input->xml, XML_LRM_ATTR_TASK);
1793     }
1794 
1795     if (pcmk__str_eq(crm_op, CRM_OP_LRM_REFRESH, pcmk__str_casei)) {
1796         handle_refresh_op(lrm_state, user_name, from_host, from_sys);
1797 
1798     } else if (pcmk__str_eq(crm_op, CRM_OP_LRM_QUERY, pcmk__str_casei)) {
1799         handle_query_op(input->msg, lrm_state);
1800 
1801     } else if (pcmk__str_eq(operation, CRM_OP_PROBED, pcmk__str_casei)) {
1802         update_attrd(lrm_state->node_name, CRM_OP_PROBED, XML_BOOLEAN_TRUE,
1803                      user_name, is_remote_node);
1804 
1805     } else if (pcmk__str_eq(crm_op, CRM_OP_REPROBE, pcmk__str_casei)
1806                || pcmk__str_eq(operation, CRM_OP_REPROBE, pcmk__str_casei)) {
1807         handle_reprobe_op(lrm_state, from_sys, from_host, user_name,
1808                           is_remote_node);
1809 
1810     } else if (operation != NULL) {
1811         lrmd_rsc_info_t *rsc = NULL;
1812         xmlNode *xml_rsc = find_xml_node(input->xml, XML_CIB_TAG_RESOURCE, TRUE);
1813         gboolean create_rsc = !pcmk__str_eq(operation, CRMD_ACTION_DELETE,
1814                                             pcmk__str_casei);
1815         int rc;
1816 
1817         // We can't return anything meaningful without a resource ID
1818         CRM_CHECK(xml_rsc && ID(xml_rsc), return);
1819 
1820         rc = get_lrm_resource(lrm_state, xml_rsc, create_rsc, &rsc);
1821         if (rc == -ENOTCONN) {
1822             synthesize_lrmd_failure(lrm_state, input->xml,
1823                                     PCMK_EXEC_NOT_CONNECTED,
1824                                     PCMK_OCF_UNKNOWN_ERROR,
1825                                     "Not connected to remote executor");
1826             return;
1827 
1828         } else if ((rc < 0) && !create_rsc) {
1829             /* Delete of malformed or nonexistent resource
1830              * (deleting something that does not exist is a success)
1831              */
1832             crm_notice("Not registering resource '%s' for a %s event "
1833                        CRM_XS " get-rc=%d (%s) transition-key=%s",
1834                        ID(xml_rsc), operation,
1835                        rc, pcmk_strerror(rc), ID(input->xml));
1836             delete_rsc_entry(lrm_state, input, ID(xml_rsc), NULL, pcmk_ok,
1837                              user_name);
1838             return;
1839 
1840         } else if (rc == -EINVAL) {
1841             // Resource operation on malformed resource
1842             crm_err("Invalid resource definition for %s", ID(xml_rsc));
1843             crm_log_xml_warn(input->msg, "invalid resource");
1844             synthesize_lrmd_failure(lrm_state, input->xml, PCMK_EXEC_ERROR,
1845                                     PCMK_OCF_NOT_CONFIGURED, // fatal error
1846                                     "Invalid resource definition");
1847             return;
1848 
1849         } else if (rc < 0) {
1850             // Error communicating with the executor
1851             crm_err("Could not register resource '%s' with executor: %s "
1852                     CRM_XS " rc=%d",
1853                     ID(xml_rsc), pcmk_strerror(rc), rc);
1854             crm_log_xml_warn(input->msg, "failed registration");
1855             synthesize_lrmd_failure(lrm_state, input->xml, PCMK_EXEC_ERROR,
1856                                     PCMK_OCF_INVALID_PARAM, // hard error
1857                                     "Could not register resource with executor");
1858             return;
1859         }
1860 
1861         if (pcmk__str_eq(operation, CRMD_ACTION_CANCEL, pcmk__str_casei)) {
1862             if (!do_lrm_cancel(input, lrm_state, rsc, from_host, from_sys)) {
1863                 crm_log_xml_warn(input->xml, "Bad command");
1864             }
1865 
1866         } else if (pcmk__str_eq(operation, CRMD_ACTION_DELETE, pcmk__str_casei)) {
1867             do_lrm_delete(input, lrm_state, rsc, from_sys, from_host,
1868                           crm_rsc_delete, user_name);
1869 
1870         } else if (pcmk__str_any_of(operation, CRMD_ACTION_RELOAD,
1871                                     CRMD_ACTION_RELOAD_AGENT, NULL)) {
1872             /* Pre-2.1.0 DCs will schedule reload actions only, and 2.1.0+ DCs
1873              * will schedule reload-agent actions only. In either case, we need
1874              * to map that to whatever the resource agent actually supports.
1875              * Default to the OCF 1.1 name.
1876              */
1877             struct ra_metadata_s *md = NULL;
1878             const char *reload_name = CRMD_ACTION_RELOAD_AGENT;
1879 
1880             md = controld_get_rsc_metadata(lrm_state, rsc,
1881                                            controld_metadata_from_cache);
1882             if ((md != NULL)
1883                 && pcmk_is_set(md->ra_flags, ra_supports_legacy_reload)) {
1884                 reload_name = CRMD_ACTION_RELOAD;
1885             }
1886             do_lrm_rsc_op(lrm_state, rsc, reload_name, input->xml);
1887 
1888         } else {
1889             do_lrm_rsc_op(lrm_state, rsc, operation, input->xml);
1890         }
1891 
1892         lrmd_free_rsc_info(rsc);
1893 
1894     } else {
1895         crm_err("Cannot perform operation %s of unknown type", crm_str(crm_op));
1896         register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL);
1897     }
1898 }
1899 
1900 #if ENABLE_VERSIONED_ATTRS
1901 static void
1902 resolve_versioned_parameters(lrm_state_t *lrm_state, const char *rsc_id,
     /* [previous][next][first][last][top][bottom][index][help] */
1903                              const xmlNode *rsc_op, GHashTable *params)
1904 {
1905     /* Resource info *should* already be cached, so we don't get
1906      * executor call */
1907     lrmd_rsc_info_t *rsc = lrm_state_get_rsc_info(lrm_state, rsc_id, 0);
1908     struct ra_metadata_s *metadata;
1909 
1910     metadata = controld_get_rsc_metadata(lrm_state, rsc,
1911                                          controld_metadata_from_cache);
1912     if (metadata) {
1913         xmlNode *versioned_attrs = NULL;
1914         GHashTable *hash = NULL;
1915         char *key = NULL;
1916         char *value = NULL;
1917         GHashTableIter iter;
1918 
1919         versioned_attrs = first_named_child(rsc_op, XML_TAG_OP_VER_ATTRS);
1920         hash = pe_unpack_versioned_parameters(versioned_attrs, metadata->ra_version);
1921         g_hash_table_iter_init(&iter, hash);
1922         while (g_hash_table_iter_next(&iter, (gpointer *) &key, (gpointer *) &value)) {
1923             g_hash_table_iter_steal(&iter);
1924             g_hash_table_replace(params, key, value);
1925         }
1926         g_hash_table_destroy(hash);
1927 
1928         versioned_attrs = first_named_child(rsc_op, XML_TAG_OP_VER_META);
1929         hash = pe_unpack_versioned_parameters(versioned_attrs, metadata->ra_version);
1930         g_hash_table_iter_init(&iter, hash);
1931         while (g_hash_table_iter_next(&iter, (gpointer *) &key, (gpointer *) &value)) {
1932             g_hash_table_replace(params, crm_meta_name(key), strdup(value));
1933 
1934             if (pcmk__str_eq(key, XML_ATTR_TIMEOUT, pcmk__str_casei)) {
1935                 pcmk__scan_min_int(value, &op->timeout, 0);
1936             } else if (pcmk__str_eq(key, XML_OP_ATTR_START_DELAY, pcmk__str_casei)) {
1937                 pcmk__scan_min_int(value, &op->start_delay, 0);
1938             }
1939         }
1940         g_hash_table_destroy(hash);
1941 
1942         versioned_attrs = first_named_child(rsc_op, XML_TAG_RSC_VER_ATTRS);
1943         hash = pe_unpack_versioned_parameters(versioned_attrs, metadata->ra_version);
1944         g_hash_table_iter_init(&iter, hash);
1945         while (g_hash_table_iter_next(&iter, (gpointer *) &key, (gpointer *) &value)) {
1946             g_hash_table_iter_steal(&iter);
1947             g_hash_table_replace(params, key, value);
1948         }
1949         g_hash_table_destroy(hash);
1950     }
1951 
1952     lrmd_free_rsc_info(rsc);
1953 }
1954 #endif
1955 
1956 static lrmd_event_data_t *
1957 construct_op(lrm_state_t *lrm_state, xmlNode *rsc_op, const char *rsc_id,
     /* [previous][next][first][last][top][bottom][index][help] */
1958              const char *operation)
1959 {
1960     lrmd_event_data_t *op = NULL;
1961     const char *op_delay = NULL;
1962     const char *op_timeout = NULL;
1963     GHashTable *params = NULL;
1964 
1965     xmlNode *primitive = NULL;
1966     const char *class = NULL;
1967 
1968     const char *transition = NULL;
1969 
1970     CRM_ASSERT(rsc_id && operation);
1971 
1972     op = lrmd_new_event(rsc_id, operation, 0);
1973     op->type = lrmd_event_exec_complete;
1974     op->timeout = 0;
1975     op->start_delay = 0;
1976     lrmd__set_result(op, PCMK_OCF_UNKNOWN, PCMK_EXEC_PENDING, NULL);
1977 
1978     if (rsc_op == NULL) {
1979         CRM_LOG_ASSERT(pcmk__str_eq(CRMD_ACTION_STOP, operation, pcmk__str_casei));
1980         op->user_data = NULL;
1981         /* the stop_all_resources() case
1982          * by definition there is no DC (or they'd be shutting
1983          *   us down).
1984          * So we should put our version here.
1985          */
1986         op->params = pcmk__strkey_table(free, free);
1987 
1988         g_hash_table_insert(op->params, strdup(XML_ATTR_CRM_VERSION), strdup(CRM_FEATURE_SET));
1989 
1990         crm_trace("Constructed %s op for %s", operation, rsc_id);
1991         return op;
1992     }
1993 
1994     params = xml2list(rsc_op);
1995     g_hash_table_remove(params, CRM_META "_op_target_rc");
1996 
1997     op_delay = crm_meta_value(params, XML_OP_ATTR_START_DELAY);
1998     pcmk__scan_min_int(op_delay, &op->start_delay, 0);
1999 
2000     op_timeout = crm_meta_value(params, XML_ATTR_TIMEOUT);
2001     pcmk__scan_min_int(op_timeout, &op->timeout, 0);
2002 
2003     if (pcmk__guint_from_hash(params, CRM_META "_" XML_LRM_ATTR_INTERVAL_MS, 0,
2004                               &(op->interval_ms)) != pcmk_rc_ok) {
2005         op->interval_ms = 0;
2006     }
2007 
2008     /* Use pcmk_monitor_timeout instead of meta timeout for stonith
2009        recurring monitor, if set */
2010     primitive = find_xml_node(rsc_op, XML_CIB_TAG_RESOURCE, FALSE);
2011     class = crm_element_value(primitive, XML_AGENT_ATTR_CLASS);
2012 
2013     if (pcmk_is_set(pcmk_get_ra_caps(class), pcmk_ra_cap_fence_params)
2014             && pcmk__str_eq(operation, CRMD_ACTION_STATUS, pcmk__str_casei)
2015             && (op->interval_ms > 0)) {
2016 
2017         op_timeout = g_hash_table_lookup(params, "pcmk_monitor_timeout");
2018         if (op_timeout != NULL) {
2019             op->timeout = crm_get_msec(op_timeout);
2020         }
2021     }
2022 
2023 #if ENABLE_VERSIONED_ATTRS
2024     if (lrm_state && !is_remote_lrmd_ra(NULL, NULL, rsc_id)
2025         && !pcmk__strcase_any_of(op_type, CRMD_ACTION_METADATA, CRMD_ACTION_DELETE,
2026                                  NULL)) {
2027         resolve_versioned_parameters(lrm_state, rsc_id, rsc_op, params);
2028     }
2029 #endif
2030 
2031     if (!pcmk__str_eq(operation, RSC_STOP, pcmk__str_casei)) {
2032         op->params = params;
2033 
2034     } else {
2035         rsc_history_t *entry = NULL;
2036 
2037         if (lrm_state) {
2038             entry = g_hash_table_lookup(lrm_state->resource_history, rsc_id);
2039         }
2040 
2041         /* If we do not have stop parameters cached, use
2042          * whatever we are given */
2043         if (!entry || !entry->stop_params) {
2044             op->params = params;
2045         } else {
2046             /* Copy the cached parameter list so that we stop the resource
2047              * with the old attributes, not the new ones */
2048             op->params = pcmk__strkey_table(free, free);
2049 
2050             g_hash_table_foreach(params, copy_meta_keys, op->params);
2051             g_hash_table_foreach(entry->stop_params, copy_instance_keys, op->params);
2052             g_hash_table_destroy(params);
2053             params = NULL;
2054         }
2055     }
2056 
2057     /* sanity */
2058     if (op->timeout <= 0) {
2059         op->timeout = op->interval_ms;
2060     }
2061     if (op->start_delay < 0) {
2062         op->start_delay = 0;
2063     }
2064 
2065     transition = crm_element_value(rsc_op, XML_ATTR_TRANSITION_KEY);
2066     CRM_CHECK(transition != NULL, return op);
2067 
2068     op->user_data = strdup(transition);
2069 
2070     if (op->interval_ms != 0) {
2071         if (pcmk__strcase_any_of(operation, CRMD_ACTION_START, CRMD_ACTION_STOP, NULL)) {
2072             crm_err("Start and Stop actions cannot have an interval: %u",
2073                     op->interval_ms);
2074             op->interval_ms = 0;
2075         }
2076     }
2077 
2078     crm_trace("Constructed %s op for %s: interval=%u",
2079               operation, rsc_id, op->interval_ms);
2080 
2081     return op;
2082 }
2083 
2084 /*!
2085  * \internal
2086  * \brief Send a (synthesized) event result
2087  *
2088  * Reply with a synthesized event result directly, as opposed to going through
2089  * the executor.
2090  *
2091  * \param[in] to_host  Host to send result to
2092  * \param[in] to_sys   IPC name to send result to (NULL for transition engine)
2093  * \param[in] rsc      Type information about resource the result is for
2094  * \param[in] op       Event with result to send
2095  * \param[in] rsc_id   ID of resource the result is for
2096  */
2097 void
2098 controld_ack_event_directly(const char *to_host, const char *to_sys,
     /* [previous][next][first][last][top][bottom][index][help] */
2099                             lrmd_rsc_info_t *rsc, lrmd_event_data_t *op,
2100                             const char *rsc_id)
2101 {
2102     xmlNode *reply = NULL;
2103     xmlNode *update, *iter;
2104     crm_node_t *peer = NULL;
2105 
2106     CRM_CHECK(op != NULL, return);
2107     if (op->rsc_id == NULL) {
2108         CRM_ASSERT(rsc_id != NULL);
2109         op->rsc_id = strdup(rsc_id);
2110     }
2111     if (to_sys == NULL) {
2112         to_sys = CRM_SYSTEM_TENGINE;
2113     }
2114 
2115     peer = crm_get_peer(0, fsa_our_uname);
2116     update = create_node_state_update(peer, node_update_none, NULL,
2117                                       __func__);
2118 
2119     iter = create_xml_node(update, XML_CIB_TAG_LRM);
2120     crm_xml_add(iter, XML_ATTR_ID, fsa_our_uuid);
2121     iter = create_xml_node(iter, XML_LRM_TAG_RESOURCES);
2122     iter = create_xml_node(iter, XML_LRM_TAG_RESOURCE);
2123 
2124     crm_xml_add(iter, XML_ATTR_ID, op->rsc_id);
2125 
2126     build_operation_update(iter, rsc, op, fsa_our_uname, __func__);
2127     reply = create_request(CRM_OP_INVOKE_LRM, update, to_host, to_sys, CRM_SYSTEM_LRMD, NULL);
2128 
2129     crm_log_xml_trace(update, "[direct ACK]");
2130 
2131     crm_debug("ACK'ing resource op " PCMK__OP_FMT " from %s: %s",
2132               op->rsc_id, op->op_type, op->interval_ms, op->user_data,
2133               crm_element_value(reply, XML_ATTR_REFERENCE));
2134 
2135     if (relay_message(reply, TRUE) == FALSE) {
2136         crm_log_xml_err(reply, "Unable to route reply");
2137     }
2138 
2139     free_xml(update);
2140     free_xml(reply);
2141 }
2142 
2143 gboolean
2144 verify_stopped(enum crmd_fsa_state cur_state, int log_level)
     /* [previous][next][first][last][top][bottom][index][help] */
2145 {
2146     gboolean res = TRUE;
2147     GList *lrm_state_list = lrm_state_get_list();
2148     GList *state_entry;
2149 
2150     for (state_entry = lrm_state_list; state_entry != NULL; state_entry = state_entry->next) {
2151         lrm_state_t *lrm_state = state_entry->data;
2152 
2153         if (!lrm_state_verify_stopped(lrm_state, cur_state, log_level)) {
2154             /* keep iterating through all even when false is returned */
2155             res = FALSE;
2156         }
2157     }
2158 
2159     controld_set_fsa_input_flags(R_SENT_RSC_STOP);
2160     g_list_free(lrm_state_list); lrm_state_list = NULL;
2161     return res;
2162 }
2163 
2164 struct stop_recurring_action_s {
2165     lrmd_rsc_info_t *rsc;
2166     lrm_state_t *lrm_state;
2167 };
2168 
2169 static gboolean
2170 stop_recurring_action_by_rsc(gpointer key, gpointer value, gpointer user_data)
     /* [previous][next][first][last][top][bottom][index][help] */
2171 {
2172     gboolean remove = FALSE;
2173     struct stop_recurring_action_s *event = user_data;
2174     active_op_t *op = value;
2175 
2176     if ((op->interval_ms != 0)
2177         && pcmk__str_eq(op->rsc_id, event->rsc->id, pcmk__str_none)) {
2178 
2179         crm_debug("Cancelling op %d for %s (%s)", op->call_id, op->rsc_id, (char*)key);
2180         remove = !cancel_op(event->lrm_state, event->rsc->id, key, op->call_id, FALSE);
2181     }
2182 
2183     return remove;
2184 }
2185 
2186 static gboolean
2187 stop_recurring_actions(gpointer key, gpointer value, gpointer user_data)
     /* [previous][next][first][last][top][bottom][index][help] */
2188 {
2189     gboolean remove = FALSE;
2190     lrm_state_t *lrm_state = user_data;
2191     active_op_t *op = value;
2192 
2193     if (op->interval_ms != 0) {
2194         crm_info("Cancelling op %d for %s (%s)", op->call_id, op->rsc_id,
2195                  (const char *) key);
2196         remove = !cancel_op(lrm_state, op->rsc_id, key, op->call_id, FALSE);
2197     }
2198 
2199     return remove;
2200 }
2201 
2202 static void
2203 record_pending_op(const char *node_name, lrmd_rsc_info_t *rsc, lrmd_event_data_t *op)
     /* [previous][next][first][last][top][bottom][index][help] */
2204 {
2205     const char *record_pending = NULL;
2206 
2207     CRM_CHECK(node_name != NULL, return);
2208     CRM_CHECK(rsc != NULL, return);
2209     CRM_CHECK(op != NULL, return);
2210 
2211     // Never record certain operation types as pending
2212     if ((op->op_type == NULL) || (op->params == NULL)
2213         || !controld_action_is_recordable(op->op_type)) {
2214         return;
2215     }
2216 
2217     // defaults to true
2218     record_pending = crm_meta_value(op->params, XML_OP_ATTR_PENDING);
2219     if (record_pending && !crm_is_true(record_pending)) {
2220         return;
2221     }
2222 
2223     op->call_id = -1;
2224     lrmd__set_result(op, PCMK_OCF_UNKNOWN, PCMK_EXEC_PENDING, NULL);
2225 
2226     op->t_run = time(NULL);
2227     op->t_rcchange = op->t_run;
2228 
2229     /* write a "pending" entry to the CIB, inhibit notification */
2230     crm_debug("Recording pending op " PCMK__OP_FMT " on %s in the CIB",
2231               op->rsc_id, op->op_type, op->interval_ms, node_name);
2232 
2233     do_update_resource(node_name, rsc, op, 0);
2234 }
2235 
2236 static void
2237 do_lrm_rsc_op(lrm_state_t *lrm_state, lrmd_rsc_info_t *rsc,
     /* [previous][next][first][last][top][bottom][index][help] */
2238               const char *operation, xmlNode *msg)
2239 {
2240     int rc;
2241     int call_id = 0;
2242     char *op_id = NULL;
2243     lrmd_event_data_t *op = NULL;
2244     lrmd_key_value_t *params = NULL;
2245     fsa_data_t *msg_data = NULL;
2246     const char *transition = NULL;
2247     gboolean stop_recurring = FALSE;
2248     const char *nack_reason = NULL;
2249 
2250     CRM_CHECK(rsc != NULL, return);
2251     CRM_CHECK(operation != NULL, return);
2252 
2253     if (msg != NULL) {
2254         transition = crm_element_value(msg, XML_ATTR_TRANSITION_KEY);
2255         if (transition == NULL) {
2256             crm_log_xml_err(msg, "Missing transition number");
2257         }
2258     }
2259 
2260     op = construct_op(lrm_state, msg, rsc->id, operation);
2261     CRM_CHECK(op != NULL, return);
2262 
2263     if (is_remote_lrmd_ra(NULL, NULL, rsc->id)
2264         && (op->interval_ms == 0)
2265         && strcmp(operation, CRMD_ACTION_MIGRATE) == 0) {
2266 
2267         /* pcmk remote connections are a special use case.
2268          * We never ever want to stop monitoring a connection resource until
2269          * the entire migration has completed. If the connection is unexpectedly
2270          * severed, even during a migration, this is an event we must detect.*/
2271         stop_recurring = FALSE;
2272 
2273     } else if ((op->interval_ms == 0)
2274         && strcmp(operation, CRMD_ACTION_STATUS) != 0
2275         && strcmp(operation, CRMD_ACTION_NOTIFY) != 0) {
2276 
2277         /* stop any previous monitor operations before changing the resource state */
2278         stop_recurring = TRUE;
2279     }
2280 
2281     if (stop_recurring == TRUE) {
2282         guint removed = 0;
2283         struct stop_recurring_action_s data;
2284 
2285         data.rsc = rsc;
2286         data.lrm_state = lrm_state;
2287         removed = g_hash_table_foreach_remove(
2288             lrm_state->pending_ops, stop_recurring_action_by_rsc, &data);
2289 
2290         if (removed) {
2291             crm_debug("Stopped %u recurring operation%s in preparation for "
2292                       PCMK__OP_FMT, removed, pcmk__plural_s(removed),
2293                       rsc->id, operation, op->interval_ms);
2294         }
2295     }
2296 
2297     /* now do the op */
2298     crm_notice("Requesting local execution of %s operation for %s on %s "
2299                CRM_XS " transition_key=%s op_key=" PCMK__OP_FMT,
2300                crm_action_str(op->op_type, op->interval_ms), rsc->id, lrm_state->node_name,
2301                transition, rsc->id, operation, op->interval_ms);
2302 
2303     if (pcmk_is_set(fsa_input_register, R_SHUTDOWN)
2304         && pcmk__str_eq(operation, RSC_START, pcmk__str_casei)) {
2305 
2306         register_fsa_input(C_SHUTDOWN, I_SHUTDOWN, NULL);
2307         nack_reason = "Not attempting start due to shutdown in progress";
2308 
2309     } else if (fsa_state != S_NOT_DC
2310                && fsa_state != S_POLICY_ENGINE /* Recalculating */
2311                && fsa_state != S_TRANSITION_ENGINE
2312                && !pcmk__str_eq(operation, CRMD_ACTION_STOP, pcmk__str_casei)) {
2313         nack_reason = "Controller cannot attempt actions at this time";
2314     }
2315 
2316     if (nack_reason != NULL) {
2317         crm_notice("Discarding attempt to perform action %s on %s in state %s (shutdown=%s)",
2318                    operation, rsc->id, fsa_state2string(fsa_state),
2319                    pcmk__btoa(pcmk_is_set(fsa_input_register, R_SHUTDOWN)));
2320 
2321         lrmd__set_result(op, PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_INVALID,
2322                          nack_reason);
2323         controld_ack_event_directly(NULL, NULL, rsc, op, rsc->id);
2324         lrmd_free_event(op);
2325         free(op_id);
2326         return;
2327     }
2328 
2329     record_pending_op(lrm_state->node_name, rsc, op);
2330 
2331     op_id = pcmk__op_key(rsc->id, op->op_type, op->interval_ms);
2332 
2333     if (op->interval_ms > 0) {
2334         /* cancel it so we can then restart it without conflict */
2335         cancel_op_key(lrm_state, rsc, op_id, FALSE);
2336     }
2337 
2338     if (op->params) {
2339         char *key = NULL;
2340         char *value = NULL;
2341         GHashTableIter iter;
2342 
2343         g_hash_table_iter_init(&iter, op->params);
2344         while (g_hash_table_iter_next(&iter, (gpointer *) & key, (gpointer *) & value)) {
2345             params = lrmd_key_value_add(params, key, value);
2346         }
2347     }
2348 
2349     rc = controld_execute_resource_agent(lrm_state, rsc->id, op->op_type,
2350                                          op->user_data, op->interval_ms,
2351                                          op->timeout, op->start_delay, params,
2352                                          &call_id);
2353     if (rc == pcmk_rc_ok) {
2354         /* record all operations so we can wait
2355          * for them to complete during shutdown
2356          */
2357         char *call_id_s = make_stop_id(rsc->id, call_id);
2358         active_op_t *pending = NULL;
2359 
2360         pending = calloc(1, sizeof(active_op_t));
2361         crm_trace("Recording pending op: %d - %s %s", call_id, op_id, call_id_s);
2362 
2363         pending->call_id = call_id;
2364         pending->interval_ms = op->interval_ms;
2365         pending->op_type = strdup(operation);
2366         pending->op_key = strdup(op_id);
2367         pending->rsc_id = strdup(rsc->id);
2368         pending->start_time = time(NULL);
2369         pending->user_data = op->user_data? strdup(op->user_data) : NULL;
2370         if (crm_element_value_epoch(msg, XML_CONFIG_ATTR_SHUTDOWN_LOCK,
2371                                     &(pending->lock_time)) != pcmk_ok) {
2372             pending->lock_time = 0;
2373         }
2374         g_hash_table_replace(lrm_state->pending_ops, call_id_s, pending);
2375 
2376         if ((op->interval_ms > 0)
2377             && (op->start_delay > START_DELAY_THRESHOLD)) {
2378             int target_rc = PCMK_OCF_OK;
2379 
2380             crm_info("Faking confirmation of %s: execution postponed for over 5 minutes", op_id);
2381             decode_transition_key(op->user_data, NULL, NULL, NULL, &target_rc);
2382             lrmd__set_result(op, target_rc, PCMK_EXEC_DONE, NULL);
2383             controld_ack_event_directly(NULL, NULL, rsc, op, rsc->id);
2384         }
2385 
2386         pending->params = op->params;
2387         op->params = NULL;
2388 
2389     } else if (lrm_state_is_local(lrm_state)) {
2390         crm_err("Could not initiate %s action for resource %s locally: %s "
2391                 CRM_XS " rc=%d", operation, rsc->id, pcmk_rc_str(rc), rc);
2392         fake_op_status(lrm_state, op, PCMK_EXEC_NOT_CONNECTED,
2393                        PCMK_OCF_UNKNOWN_ERROR, pcmk_rc_str(rc));
2394         process_lrm_event(lrm_state, op, NULL, NULL);
2395         register_fsa_error(C_FSA_INTERNAL, I_FAIL, NULL);
2396 
2397     } else {
2398         crm_err("Could not initiate %s action for resource %s remotely on %s: "
2399                 "%s " CRM_XS " rc=%d",
2400                 operation, rsc->id, lrm_state->node_name, pcmk_rc_str(rc), rc);
2401         fake_op_status(lrm_state, op, PCMK_EXEC_NOT_CONNECTED,
2402                        PCMK_OCF_UNKNOWN_ERROR, pcmk_rc_str(rc));
2403         process_lrm_event(lrm_state, op, NULL, NULL);
2404     }
2405 
2406     free(op_id);
2407     lrmd_free_event(op);
2408 }
2409 
2410 int last_resource_update = 0;
2411 
2412 static void
2413 cib_rsc_callback(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data)
     /* [previous][next][first][last][top][bottom][index][help] */
2414 {
2415     switch (rc) {
2416         case pcmk_ok:
2417         case -pcmk_err_diff_failed:
2418         case -pcmk_err_diff_resync:
2419             crm_trace("Resource update %d complete: rc=%d", call_id, rc);
2420             break;
2421         default:
2422             crm_warn("Resource update %d failed: (rc=%d) %s", call_id, rc, pcmk_strerror(rc));
2423     }
2424 
2425     if (call_id == last_resource_update) {
2426         last_resource_update = 0;
2427         trigger_fsa();
2428     }
2429 }
2430 
2431 /* Only successful stops, and probes that found the resource inactive, get locks
2432  * recorded in the history. This ensures the resource stays locked to the node
2433  * until it is active there again after the node comes back up.
2434  */
2435 static bool
2436 should_preserve_lock(lrmd_event_data_t *op)
     /* [previous][next][first][last][top][bottom][index][help] */
2437 {
2438     if (!controld_shutdown_lock_enabled) {
2439         return false;
2440     }
2441     if (!strcmp(op->op_type, RSC_STOP) && (op->rc == PCMK_OCF_OK)) {
2442         return true;
2443     }
2444     if (!strcmp(op->op_type, RSC_STATUS) && (op->rc == PCMK_OCF_NOT_RUNNING)) {
2445         return true;
2446     }
2447     return false;
2448 }
2449 
2450 static int
2451 do_update_resource(const char *node_name, lrmd_rsc_info_t *rsc,
     /* [previous][next][first][last][top][bottom][index][help] */
2452                    lrmd_event_data_t *op, time_t lock_time)
2453 {
2454 /*
2455   <status>
2456   <nodes_status id=uname>
2457   <lrm>
2458   <lrm_resources>
2459   <lrm_resource id=...>
2460   </...>
2461 */
2462     int rc = pcmk_ok;
2463     xmlNode *update, *iter = NULL;
2464     int call_opt = crmd_cib_smart_opt();
2465     const char *uuid = NULL;
2466 
2467     CRM_CHECK(op != NULL, return 0);
2468 
2469     iter = create_xml_node(iter, XML_CIB_TAG_STATUS);
2470     update = iter;
2471     iter = create_xml_node(iter, XML_CIB_TAG_STATE);
2472 
2473     if (pcmk__str_eq(node_name, fsa_our_uname, pcmk__str_casei)) {
2474         uuid = fsa_our_uuid;
2475 
2476     } else {
2477         /* remote nodes uuid and uname are equal */
2478         uuid = node_name;
2479         crm_xml_add(iter, XML_NODE_IS_REMOTE, "true");
2480     }
2481 
2482     CRM_LOG_ASSERT(uuid != NULL);
2483     if(uuid == NULL) {
2484         rc = -EINVAL;
2485         goto done;
2486     }
2487 
2488     crm_xml_add(iter, XML_ATTR_UUID,  uuid);
2489     crm_xml_add(iter, XML_ATTR_UNAME, node_name);
2490     crm_xml_add(iter, XML_ATTR_ORIGIN, __func__);
2491 
2492     iter = create_xml_node(iter, XML_CIB_TAG_LRM);
2493     crm_xml_add(iter, XML_ATTR_ID, uuid);
2494 
2495     iter = create_xml_node(iter, XML_LRM_TAG_RESOURCES);
2496     iter = create_xml_node(iter, XML_LRM_TAG_RESOURCE);
2497     crm_xml_add(iter, XML_ATTR_ID, op->rsc_id);
2498 
2499     build_operation_update(iter, rsc, op, node_name, __func__);
2500 
2501     if (rsc) {
2502         const char *container = NULL;
2503 
2504         crm_xml_add(iter, XML_ATTR_TYPE, rsc->type);
2505         crm_xml_add(iter, XML_AGENT_ATTR_CLASS, rsc->standard);
2506         crm_xml_add(iter, XML_AGENT_ATTR_PROVIDER, rsc->provider);
2507         if (lock_time != 0) {
2508             /* Actions on a locked resource should either preserve the lock by
2509              * recording it with the action result, or clear it.
2510              */
2511             if (!should_preserve_lock(op)) {
2512                 lock_time = 0;
2513             }
2514             crm_xml_add_ll(iter, XML_CONFIG_ATTR_SHUTDOWN_LOCK,
2515                            (long long) lock_time);
2516         }
2517 
2518         if (op->params) {
2519             container = g_hash_table_lookup(op->params, CRM_META"_"XML_RSC_ATTR_CONTAINER);
2520         }
2521         if (container) {
2522             crm_trace("Resource %s is a part of container resource %s", op->rsc_id, container);
2523             crm_xml_add(iter, XML_RSC_ATTR_CONTAINER, container);
2524         }
2525 
2526     } else {
2527         crm_warn("Resource %s no longer exists in the executor", op->rsc_id);
2528         controld_ack_event_directly(NULL, NULL, rsc, op, op->rsc_id);
2529         goto cleanup;
2530     }
2531 
2532     crm_log_xml_trace(update, __func__);
2533 
2534     /* make it an asynchronous call and be done with it
2535      *
2536      * Best case:
2537      *   the resource state will be discovered during
2538      *   the next signup or election.
2539      *
2540      * Bad case:
2541      *   we are shutting down and there is no DC at the time,
2542      *   but then why were we shutting down then anyway?
2543      *   (probably because of an internal error)
2544      *
2545      * Worst case:
2546      *   we get shot for having resources "running" that really weren't
2547      *
2548      * the alternative however means blocking here for too long, which
2549      * isn't acceptable
2550      */
2551     fsa_cib_update(XML_CIB_TAG_STATUS, update, call_opt, rc, NULL);
2552 
2553     if (rc > 0) {
2554         last_resource_update = rc;
2555     }
2556   done:
2557     /* the return code is a call number, not an error code */
2558     crm_trace("Sent resource state update message: %d for %s=%u on %s",
2559               rc, op->op_type, op->interval_ms, op->rsc_id);
2560     fsa_register_cib_callback(rc, FALSE, NULL, cib_rsc_callback);
2561 
2562   cleanup:
2563     free_xml(update);
2564     return rc;
2565 }
2566 
2567 void
2568 do_lrm_event(long long action,
     /* [previous][next][first][last][top][bottom][index][help] */
2569              enum crmd_fsa_cause cause,
2570              enum crmd_fsa_state cur_state, enum crmd_fsa_input cur_input, fsa_data_t * msg_data)
2571 {
2572     CRM_CHECK(FALSE, return);
2573 }
2574 
2575 static char *
2576 unescape_newlines(const char *string)
     /* [previous][next][first][last][top][bottom][index][help] */
2577 {
2578     char *pch = NULL;
2579     char *ret = NULL;
2580     static const char *escaped_newline = "\\n";
2581 
2582     if (!string) {
2583         return NULL;
2584     }
2585 
2586     ret = strdup(string);
2587     pch = strstr(ret, escaped_newline);
2588     while (pch != NULL) {
2589         /* Replace newline escape pattern with actual newline (and a space so we
2590          * don't have to shuffle the rest of the buffer)
2591          */
2592         pch[0] = '\n';
2593         pch[1] = ' ';
2594         pch = strstr(pch, escaped_newline);
2595     }
2596 
2597     return ret;
2598 }
2599 
2600 static bool
2601 did_lrm_rsc_op_fail(lrm_state_t *lrm_state, const char * rsc_id,
     /* [previous][next][first][last][top][bottom][index][help] */
2602                     const char * op_type, guint interval_ms)
2603 {
2604     rsc_history_t *entry = NULL;
2605 
2606     CRM_CHECK(lrm_state != NULL, return FALSE);
2607     CRM_CHECK(rsc_id != NULL, return FALSE);
2608     CRM_CHECK(op_type != NULL, return FALSE);
2609 
2610     entry = g_hash_table_lookup(lrm_state->resource_history, rsc_id);
2611     if (entry == NULL || entry->failed == NULL) {
2612         return FALSE;
2613     }
2614 
2615     if (pcmk__str_eq(entry->failed->rsc_id, rsc_id, pcmk__str_none)
2616         && pcmk__str_eq(entry->failed->op_type, op_type, pcmk__str_casei)
2617         && entry->failed->interval_ms == interval_ms) {
2618         return TRUE;
2619     }
2620 
2621     return FALSE;
2622 }
2623 
2624 /*!
2625  * \internal
2626  * \brief Log the result of an executor action (actual or synthesized)
2627  *
2628  * \param[in] op         Executor action to log result for
2629  * \param[in] op_key     Operation key for action
2630  * \param[in] node_name  Name of node action was performed on, if known
2631  * \param[in] update_id  Call ID for CIB update (or 0 if none)
2632  * \param[in] confirmed  Whether to log that graph action was confirmed
2633  */
2634 static void
2635 log_executor_event(lrmd_event_data_t *op, const char *op_key,
     /* [previous][next][first][last][top][bottom][index][help] */
2636                    const char *node_name, int update_id, gboolean confirmed)
2637 {
2638     int log_level = LOG_ERR;
2639     GString *str = g_string_sized_new(100); // reasonable starting size
2640 
2641     g_string_printf(str, "Result of %s operation for %s",
2642                     crm_action_str(op->op_type, op->interval_ms), op->rsc_id);
2643 
2644     if (node_name != NULL) {
2645         g_string_append_printf(str, " on %s", node_name);
2646     }
2647 
2648     switch (op->op_status) {
2649         case PCMK_EXEC_DONE:
2650             log_level = LOG_NOTICE;
2651             g_string_append_printf(str, ": %s",
2652                                    services_ocf_exitcode_str(op->rc));
2653             break;
2654 
2655         case PCMK_EXEC_TIMEOUT:
2656             g_string_append_printf(str, ": %s after %s",
2657                                    pcmk_exec_status_str(op->op_status),
2658                                    pcmk__readable_interval(op->timeout));
2659             break;
2660 
2661         case PCMK_EXEC_CANCELLED:
2662             log_level = LOG_INFO;
2663             // Fall through
2664         default:
2665             g_string_append_printf(str, ": %s",
2666                                    pcmk_exec_status_str(op->op_status));
2667     }
2668 
2669     if ((op->exit_reason != NULL)
2670         && ((op->op_status != PCMK_EXEC_DONE) || (op->rc != PCMK_OCF_OK))) {
2671         g_string_append_printf(str, " (%s)", op->exit_reason);
2672     }
2673 
2674     g_string_append(str, " " CRM_XS);
2675     if (update_id != 0) {
2676         g_string_append_printf(str, " CIB update %d,", update_id);
2677     }
2678     g_string_append_printf(str, " graph action %sconfirmed; call=%d key=%s",
2679                            (confirmed? "" : "un"), op->call_id, op_key);
2680     if (op->op_status == PCMK_EXEC_DONE) {
2681         g_string_append_printf(str, " rc=%d", op->rc);
2682     }
2683 
2684     do_crm_log(log_level, "%s", str->str);
2685     g_string_free(str, TRUE);
2686 
2687     if (op->output != NULL) {
2688         char *prefix = crm_strdup_printf("%s-" PCMK__OP_FMT ":%d", node_name,
2689                                          op->rsc_id, op->op_type,
2690                                          op->interval_ms, op->call_id);
2691 
2692         if (op->rc) {
2693             crm_log_output(LOG_NOTICE, prefix, op->output);
2694         } else {
2695             crm_log_output(LOG_DEBUG, prefix, op->output);
2696         }
2697         free(prefix);
2698     }
2699 }
2700 
2701 void
2702 process_lrm_event(lrm_state_t *lrm_state, lrmd_event_data_t *op,
     /* [previous][next][first][last][top][bottom][index][help] */
2703                   active_op_t *pending, xmlNode *action_xml)
2704 {
2705     char *op_id = NULL;
2706     char *op_key = NULL;
2707 
2708     int update_id = 0;
2709     gboolean remove = FALSE;
2710     gboolean removed = FALSE;
2711     bool need_direct_ack = FALSE;
2712     lrmd_rsc_info_t *rsc = NULL;
2713     const char *node_name = NULL;
2714 
2715     CRM_CHECK(op != NULL, return);
2716     CRM_CHECK(op->rsc_id != NULL, return);
2717 
2718     // Remap new status codes for older DCs
2719     if (compare_version(fsa_our_dc_version, "3.2.0") < 0) {
2720         switch (op->op_status) {
2721             case PCMK_EXEC_NOT_CONNECTED:
2722                 lrmd__set_result(op, PCMK_OCF_CONNECTION_DIED,
2723                                  PCMK_EXEC_ERROR, op->exit_reason);
2724                 break;
2725             case PCMK_EXEC_INVALID:
2726                 lrmd__set_result(op, CRM_DIRECT_NACK_RC, PCMK_EXEC_ERROR,
2727                                  op->exit_reason);
2728                 break;
2729             default:
2730                 break;
2731         }
2732     }
2733 
2734     op_id = make_stop_id(op->rsc_id, op->call_id);
2735     op_key = pcmk__op_key(op->rsc_id, op->op_type, op->interval_ms);
2736 
2737     // Get resource info if available (from executor state or action XML)
2738     if (lrm_state) {
2739         rsc = lrm_state_get_rsc_info(lrm_state, op->rsc_id, 0);
2740     }
2741     if ((rsc == NULL) && action_xml) {
2742         xmlNode *xml = find_xml_node(action_xml, XML_CIB_TAG_RESOURCE, TRUE);
2743 
2744         const char *standard = crm_element_value(xml, XML_AGENT_ATTR_CLASS);
2745         const char *provider = crm_element_value(xml, XML_AGENT_ATTR_PROVIDER);
2746         const char *type = crm_element_value(xml, XML_ATTR_TYPE);
2747 
2748         if (standard && type) {
2749             crm_info("%s agent information not cached, using %s%s%s:%s from action XML",
2750                      op->rsc_id, standard,
2751                      (provider? ":" : ""), (provider? provider : ""), type);
2752             rsc = lrmd_new_rsc_info(op->rsc_id, standard, provider, type);
2753         } else {
2754             crm_err("Can't process %s result because %s agent information not cached or in XML",
2755                     op_key, op->rsc_id);
2756         }
2757     }
2758 
2759     // Get node name if available (from executor state or action XML)
2760     if (lrm_state) {
2761         node_name = lrm_state->node_name;
2762     } else if (action_xml) {
2763         node_name = crm_element_value(action_xml, XML_LRM_ATTR_TARGET);
2764     }
2765 
2766     if(pending == NULL) {
2767         remove = TRUE;
2768         if (lrm_state) {
2769             pending = g_hash_table_lookup(lrm_state->pending_ops, op_id);
2770         }
2771     }
2772 
2773     if (op->op_status == PCMK_EXEC_ERROR) {
2774         switch(op->rc) {
2775             case PCMK_OCF_NOT_RUNNING:
2776             case PCMK_OCF_RUNNING_PROMOTED:
2777             case PCMK_OCF_DEGRADED:
2778             case PCMK_OCF_DEGRADED_PROMOTED:
2779                 // Leave it to the TE/scheduler to decide if this is an error
2780                 op->op_status = PCMK_EXEC_DONE;
2781                 break;
2782             default:
2783                 /* Nothing to do */
2784                 break;
2785         }
2786     }
2787 
2788     if (op->op_status != PCMK_EXEC_CANCELLED) {
2789         /* We might not record the result, so directly acknowledge it to the
2790          * originator instead, so it doesn't time out waiting for the result
2791          * (especially important if part of a transition).
2792          */
2793         need_direct_ack = TRUE;
2794 
2795         if (controld_action_is_recordable(op->op_type)) {
2796             if (node_name && rsc) {
2797                 // We should record the result, and happily, we can
2798                 update_id = do_update_resource(node_name, rsc, op,
2799                                                pending? pending->lock_time : 0);
2800                 need_direct_ack = FALSE;
2801 
2802             } else if (op->rsc_deleted) {
2803                 /* We shouldn't record the result (likely the resource was
2804                  * refreshed, cleaned, or removed while this operation was
2805                  * in flight).
2806                  */
2807                 crm_notice("Not recording %s result in CIB because "
2808                            "resource information was removed since it was initiated",
2809                            op_key);
2810             } else {
2811                 /* This shouldn't be possible; the executor didn't consider the
2812                  * resource deleted, but we couldn't find resource or node
2813                  * information.
2814                  */
2815                 crm_err("Unable to record %s result in CIB: %s", op_key,
2816                         (node_name? "No resource information" : "No node name"));
2817             }
2818         }
2819 
2820     } else if (op->interval_ms == 0) {
2821         /* A non-recurring operation was cancelled. Most likely, the
2822          * never-initiated action was removed from the executor's pending
2823          * operations list upon resource removal.
2824          */
2825         need_direct_ack = TRUE;
2826 
2827     } else if (pending == NULL) {
2828         /* This recurring operation was cancelled, but was not pending. No
2829          * transition actions are waiting on it, nothing needs to be done.
2830          */
2831 
2832     } else if (op->user_data == NULL) {
2833         /* This recurring operation was cancelled and pending, but we don't
2834          * have a transition key. This should never happen.
2835          */
2836         crm_err("Recurring operation %s was cancelled without transition information",
2837                 op_key);
2838 
2839     } else if (pcmk_is_set(pending->flags, active_op_remove)) {
2840         /* This recurring operation was cancelled (by us) and pending, and we
2841          * have been waiting for it to finish.
2842          */
2843         if (lrm_state) {
2844             erase_lrm_history_by_op(lrm_state, op);
2845         }
2846 
2847         /* If the recurring operation had failed, the lrm_rsc_op is recorded as
2848          * "last_failure" which won't get erased from the cib given the logic on
2849          * purpose in erase_lrm_history_by_op(). So that the cancel action won't
2850          * have a chance to get confirmed by DC with process_op_deletion().
2851          * Cluster transition would get stuck waiting for the remaining action
2852          * timer to time out.
2853          *
2854          * Directly acknowledge the cancel operation in this case.
2855          */
2856         if (did_lrm_rsc_op_fail(lrm_state, pending->rsc_id,
2857                                 pending->op_type, pending->interval_ms)) {
2858             need_direct_ack = TRUE;
2859         }
2860 
2861     } else if (op->rsc_deleted) {
2862         /* This recurring operation was cancelled (but not by us, and the
2863          * executor does not have resource information, likely due to resource
2864          * cleanup, refresh, or removal) and pending.
2865          */
2866         crm_debug("Recurring op %s was cancelled due to resource deletion",
2867                   op_key);
2868         need_direct_ack = TRUE;
2869 
2870     } else {
2871         /* This recurring operation was cancelled (but not by us, likely by the
2872          * executor before stopping the resource) and pending. We don't need to
2873          * do anything special.
2874          */
2875     }
2876 
2877     if (need_direct_ack) {
2878         controld_ack_event_directly(NULL, NULL, NULL, op, op->rsc_id);
2879     }
2880 
2881     if(remove == FALSE) {
2882         /* The caller will do this afterwards, but keep the logging consistent */
2883         removed = TRUE;
2884 
2885     } else if (lrm_state && ((op->interval_ms == 0)
2886                              || (op->op_status == PCMK_EXEC_CANCELLED))) {
2887 
2888         gboolean found = g_hash_table_remove(lrm_state->pending_ops, op_id);
2889 
2890         if (op->interval_ms != 0) {
2891             removed = TRUE;
2892         } else if (found) {
2893             removed = TRUE;
2894             crm_trace("Op %s (call=%d, stop-id=%s, remaining=%u): Confirmed",
2895                       op_key, op->call_id, op_id,
2896                       g_hash_table_size(lrm_state->pending_ops));
2897         }
2898     }
2899 
2900     log_executor_event(op, op_key, node_name, update_id, removed);
2901 
2902     if (lrm_state) {
2903         if (!pcmk__str_eq(op->op_type, RSC_METADATA, pcmk__str_casei)) {
2904             crmd_alert_resource_op(lrm_state->node_name, op);
2905         } else if (rsc && (op->rc == PCMK_OCF_OK)) {
2906             char *metadata = unescape_newlines(op->output);
2907 
2908             metadata_cache_update(lrm_state->metadata_cache, rsc, metadata);
2909             free(metadata);
2910         }
2911     }
2912 
2913     if (op->rsc_deleted) {
2914         crm_info("Deletion of resource '%s' complete after %s", op->rsc_id, op_key);
2915         if (lrm_state) {
2916             delete_rsc_entry(lrm_state, NULL, op->rsc_id, NULL, pcmk_ok, NULL);
2917         }
2918     }
2919 
2920     /* If a shutdown was escalated while operations were pending,
2921      * then the FSA will be stalled right now... allow it to continue
2922      */
2923     mainloop_set_trigger(fsa_source);
2924     if (lrm_state && rsc) {
2925         update_history_cache(lrm_state, rsc, op);
2926     }
2927 
2928     lrmd_free_rsc_info(rsc);
2929     free(op_key);
2930     free(op_id);
2931 }

/* [previous][next][first][last][top][bottom][index][help] */