daemons/controld/controld

/* */
This source file includes following definitions.
handle_cib_disconnect
do_cib_updated
controld_disconnect_cib_manager
do_cib_control
cib_op_timeout
crmd_cib_smart_opt
cib_delete_callback
controld_node_state_deletion_strings
controld_delete_node_state
controld_delete_resource_history
build_parameter_list
append_restart_list
append_secure_list
controld_add_resource_history_xml_as
controld_record_pending_op
cib_rsc_callback
should_preserve_lock
controld_update_cib
controld_update_resource_history
controld_delete_action_history
controld_cib_delete_last_failure
controld_delete_action_history_by_key
   1 /*
   2  * Copyright 2004-2024 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 <unistd.h>  /* sleep */
  13 
  14 #include <crm/common/alerts_internal.h>
  15 #include <crm/common/xml.h>
  16 #include <crm/crm.h>
  17 #include <crm/lrmd_internal.h>
  18 
  19 #include <pacemaker-controld.h>
  20 
  21 // Call ID of the most recent in-progress CIB resource update (or 0 if none)
  22 static int pending_rsc_update = 0;
  23 
  24 /*!
  25  * \internal
  26  * \brief Respond to a dropped CIB connection
  27  *
  28  * \param[in] user_data  CIB connection that dropped
  29  */
  30 static void
  31 handle_cib_disconnect(gpointer user_data)
     /*  */
  32 {
  33     CRM_LOG_ASSERT(user_data == controld_globals.cib_conn);
  34 
  35     controld_trigger_fsa();
  36     controld_globals.cib_conn->state = cib_disconnected;
  37 
  38     if (pcmk_is_set(controld_globals.fsa_input_register, R_CIB_CONNECTED)) {
  39         // @TODO This should trigger a reconnect, not a shutdown
  40         crm_crit("Lost connection to the CIB manager, shutting down");
  41         register_fsa_input(C_FSA_INTERNAL, I_ERROR, NULL);
  42         controld_clear_fsa_input_flags(R_CIB_CONNECTED);
  43 
  44     } else { // Expected
  45         crm_info("Disconnected from the CIB manager");
  46     }
  47 }
  48 
  49 static void
  50 do_cib_updated(const char *event, xmlNode * msg)
     /*  */
  51 {
  52     const xmlNode *patchset = NULL;
  53     const char *client_name = NULL;
  54 
  55     crm_debug("Received CIB diff notification: DC=%s", pcmk__btoa(AM_I_DC));
  56 
  57     if (cib__get_notify_patchset(msg, &patchset) != pcmk_rc_ok) {
  58         return;
  59     }
  60 
  61     if (cib__element_in_patchset(patchset, PCMK_XE_ALERTS)
  62         || cib__element_in_patchset(patchset, PCMK_XE_CRM_CONFIG)) {
  63 
  64         controld_trigger_config();
  65     }
  66 
  67     if (!AM_I_DC) {
  68         // We're not in control of the join sequence
  69         return;
  70     }
  71 
  72     client_name = crm_element_value(msg, PCMK__XA_CIB_CLIENTNAME);
  73     if (!cib__client_triggers_refresh(client_name)) {
  74         // The CIB is still accurate
  75         return;
  76     }
  77 
  78     if (cib__element_in_patchset(patchset, PCMK_XE_NODES)
  79         || cib__element_in_patchset(patchset, PCMK_XE_STATUS)) {
  80 
  81         /* An unsafe client modified the PCMK_XE_NODES or PCMK_XE_STATUS
  82          * section. Ensure the node list is up-to-date, and start the join
  83          * process again so we get everyone's current resource history.
  84          */
  85         if (client_name == NULL) {
  86             client_name = crm_element_value(msg, PCMK__XA_CIB_CLIENTID);
  87         }
  88         crm_notice("Populating nodes and starting an election after %s event "
  89                    "triggered by %s",
  90                    event, pcmk__s(client_name, "(unidentified client)"));
  91 
  92         populate_cib_nodes(node_update_quick|node_update_all, __func__);
  93         register_fsa_input(C_FSA_INTERNAL, I_ELECTION, NULL);
  94     }
  95 }
  96 
  97 void
  98 controld_disconnect_cib_manager(void)
     /*  */
  99 {
 100     cib_t *cib_conn = controld_globals.cib_conn;
 101 
 102     pcmk__assert(cib_conn != NULL);
 103 
 104     crm_debug("Disconnecting from the CIB manager");
 105 
 106     controld_clear_fsa_input_flags(R_CIB_CONNECTED);
 107 
 108     cib_conn->cmds->del_notify_callback(cib_conn, PCMK__VALUE_CIB_DIFF_NOTIFY,
 109                                         do_cib_updated);
 110     cib_free_callbacks(cib_conn);
 111 
 112     if (cib_conn->state != cib_disconnected) {
 113         cib_conn->cmds->set_secondary(cib_conn,
 114                                       cib_scope_local|cib_discard_reply);
 115         cib_conn->cmds->signoff(cib_conn);
 116     }
 117 }
 118 
 119 /* A_CIB_STOP, A_CIB_START, O_CIB_RESTART */
 120 void
 121 do_cib_control(long long action,
     /*  */
 122                enum crmd_fsa_cause cause,
 123                enum crmd_fsa_state cur_state,
 124                enum crmd_fsa_input current_input, fsa_data_t * msg_data)
 125 {
 126     static int cib_retries = 0;
 127 
 128     cib_t *cib_conn = controld_globals.cib_conn;
 129 
 130     void (*dnotify_fn) (gpointer user_data) = handle_cib_disconnect;
 131     void (*update_cb) (const char *event, xmlNodePtr msg) = do_cib_updated;
 132 
 133     int rc = pcmk_ok;
 134 
 135     pcmk__assert(cib_conn != NULL);
 136 
 137     if (pcmk_is_set(action, A_CIB_STOP)) {
 138         if ((cib_conn->state != cib_disconnected)
 139             && (pending_rsc_update != 0)) {
 140 
 141             crm_info("Waiting for resource update %d to complete",
 142                      pending_rsc_update);
 143             crmd_fsa_stall(FALSE);
 144             return;
 145         }
 146         controld_disconnect_cib_manager();
 147     }
 148 
 149     if (!pcmk_is_set(action, A_CIB_START)) {
 150         return;
 151     }
 152 
 153     if (cur_state == S_STOPPING) {
 154         crm_err("Ignoring request to connect to the CIB manager after "
 155                 "shutdown");
 156         return;
 157     }
 158 
 159     rc = cib_conn->cmds->signon(cib_conn, CRM_SYSTEM_CRMD,
 160                                 cib_command_nonblocking);
 161 
 162     if (rc != pcmk_ok) {
 163         // A short wait that usually avoids stalling the FSA
 164         sleep(1);
 165         rc = cib_conn->cmds->signon(cib_conn, CRM_SYSTEM_CRMD,
 166                                     cib_command_nonblocking);
 167     }
 168 
 169     if (rc != pcmk_ok) {
 170         crm_info("Could not connect to the CIB manager: %s", pcmk_strerror(rc));
 171 
 172     } else if (cib_conn->cmds->set_connection_dnotify(cib_conn,
 173                                                       dnotify_fn) != pcmk_ok) {
 174         crm_err("Could not set dnotify callback");
 175 
 176     } else if (cib_conn->cmds->add_notify_callback(cib_conn,
 177                                                    PCMK__VALUE_CIB_DIFF_NOTIFY,
 178                                                    update_cb) != pcmk_ok) {
 179         crm_err("Could not set CIB notification callback (update)");
 180 
 181     } else {
 182         controld_set_fsa_input_flags(R_CIB_CONNECTED);
 183         cib_retries = 0;
 184     }
 185 
 186     if (!pcmk_is_set(controld_globals.fsa_input_register, R_CIB_CONNECTED)) {
 187         cib_retries++;
 188 
 189         if (cib_retries < 30) {
 190             crm_warn("Couldn't complete CIB registration %d times... "
 191                      "pause and retry", cib_retries);
 192             controld_start_wait_timer();
 193             crmd_fsa_stall(FALSE);
 194 
 195         } else {
 196             crm_err("Could not complete CIB registration %d times... "
 197                     "hard error", cib_retries);
 198             register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL);
 199         }
 200     }
 201 }
 202 
 203 #define MIN_CIB_OP_TIMEOUT (30)
 204 
 205 /*!
 206  * \internal
 207  * \brief Get the timeout (in seconds) that should be used with CIB operations
 208  *
 209  * \return The maximum of 30 seconds, the value of the PCMK_cib_timeout
 210  *         environment variable, or 10 seconds times one more than the number of
 211  *         nodes in the cluster.
 212  */
 213 unsigned int
 214 cib_op_timeout(void)
     /*  */
 215 {
 216     // @COMPAT: Drop env_timeout at 3.0.0
 217     static int env_timeout = -1;
 218     unsigned int calculated_timeout = 0;
 219 
 220     if (env_timeout == -1) {
 221         const char *env = pcmk__env_option(PCMK__ENV_CIB_TIMEOUT);
 222 
 223         pcmk__scan_min_int(env, &env_timeout, MIN_CIB_OP_TIMEOUT);
 224         crm_trace("Minimum CIB op timeout: %ds (environment: %s)",
 225                   env_timeout, (env? env : "none"));
 226     }
 227 
 228     calculated_timeout = 10U * (1U
 229                                 + pcmk__cluster_num_active_nodes()
 230                                 + pcmk__cluster_num_remote_nodes());
 231     calculated_timeout = QB_MAX(calculated_timeout, env_timeout);
 232     crm_trace("Calculated timeout: %us", calculated_timeout);
 233 
 234     if (controld_globals.cib_conn) {
 235         controld_globals.cib_conn->call_timeout = calculated_timeout;
 236     }
 237     return calculated_timeout;
 238 }
 239 
 240 /*!
 241  * \internal
 242  * \brief Get CIB call options to use local scope if primary is unavailable
 243  *
 244  * \return CIB call options
 245  */
 246 int
 247 crmd_cib_smart_opt(void)
     /*  */
 248 {
 249     int call_opt = cib_none;
 250 
 251     if ((controld_globals.fsa_state == S_ELECTION)
 252         || (controld_globals.fsa_state == S_PENDING)) {
 253         crm_info("Sending update to local CIB in state: %s",
 254                  fsa_state2string(controld_globals.fsa_state));
 255         cib__set_call_options(call_opt, "update", cib_scope_local);
 256     }
 257     return call_opt;
 258 }
 259 
 260 static void
 261 cib_delete_callback(xmlNode *msg, int call_id, int rc, xmlNode *output,
     /*  */
 262                     void *user_data)
 263 {
 264     char *desc = user_data;
 265 
 266     if (rc == 0) {
 267         crm_debug("Deletion of %s (via CIB call %d) succeeded", desc, call_id);
 268     } else {
 269         crm_warn("Deletion of %s (via CIB call %d) failed: %s " CRM_XS " rc=%d",
 270                  desc, call_id, pcmk_strerror(rc), rc);
 271     }
 272 }
 273 
 274 // Searches for various portions of PCMK__XE_NODE_STATE to delete
 275 
 276 // Match a particular node's PCMK__XE_NODE_STATE (takes node name 1x)
 277 #define XPATH_NODE_STATE "//" PCMK__XE_NODE_STATE "[@" PCMK_XA_UNAME "='%s']"
 278 
 279 // Node's lrm section (name 1x)
 280 #define XPATH_NODE_LRM XPATH_NODE_STATE "/" PCMK__XE_LRM
 281 
 282 /* Node's PCMK__XE_LRM_RSC_OP entries and PCMK__XE_LRM_RESOURCE entries without
 283  * unexpired lock
 284  * (name 2x, (seconds_since_epoch - PCMK_OPT_SHUTDOWN_LOCK_LIMIT) 1x)
 285  */
 286 #define XPATH_NODE_LRM_UNLOCKED XPATH_NODE_STATE "//" PCMK__XE_LRM_RSC_OP   \
 287                                 "|" XPATH_NODE_STATE                        \
 288                                 "//" PCMK__XE_LRM_RESOURCE                  \
 289                                 "[not(@" PCMK_OPT_SHUTDOWN_LOCK ") "        \
 290                                     "or " PCMK_OPT_SHUTDOWN_LOCK "<%lld]"
 291 
 292 // Node's PCMK__XE_TRANSIENT_ATTRIBUTES section (name 1x)
 293 #define XPATH_NODE_ATTRS XPATH_NODE_STATE "/" PCMK__XE_TRANSIENT_ATTRIBUTES
 294 
 295 // Everything under PCMK__XE_NODE_STATE (name 1x)
 296 #define XPATH_NODE_ALL          XPATH_NODE_STATE "/*"
 297 
 298 /* Unlocked history + transient attributes
 299  * (name 2x, (seconds_since_epoch - PCMK_OPT_SHUTDOWN_LOCK_LIMIT) 1x, name 1x)
 300  */
 301 #define XPATH_NODE_ALL_UNLOCKED XPATH_NODE_LRM_UNLOCKED "|" XPATH_NODE_ATTRS
 302 
 303 /*!
 304  * \internal
 305  * \brief Get the XPath and description of a node state section to be deleted
 306  *
 307  * \param[in]  uname    Desired node
 308  * \param[in]  section  Subsection of \c PCMK__XE_NODE_STATE to be deleted
 309  * \param[out] xpath    Where to store XPath of \p section
 310  * \param[out] desc     If not \c NULL, where to store description of \p section
 311  */
 312 void
 313 controld_node_state_deletion_strings(const char *uname,
     /*  */
 314                                      enum controld_section_e section,
 315                                      char **xpath, char **desc)
 316 {
 317     const char *desc_pre = NULL;
 318 
 319     // Shutdown locks that started before this time are expired
 320     long long expire = (long long) time(NULL)
 321                        - controld_globals.shutdown_lock_limit;
 322 
 323     switch (section) {
 324         case controld_section_lrm:
 325             *xpath = crm_strdup_printf(XPATH_NODE_LRM, uname);
 326             desc_pre = "resource history";
 327             break;
 328         case controld_section_lrm_unlocked:
 329             *xpath = crm_strdup_printf(XPATH_NODE_LRM_UNLOCKED,
 330                                        uname, uname, expire);
 331             desc_pre = "resource history (other than shutdown locks)";
 332             break;
 333         case controld_section_attrs:
 334             *xpath = crm_strdup_printf(XPATH_NODE_ATTRS, uname);
 335             desc_pre = "transient attributes";
 336             break;
 337         case controld_section_all:
 338             *xpath = crm_strdup_printf(XPATH_NODE_ALL, uname);
 339             desc_pre = "all state";
 340             break;
 341         case controld_section_all_unlocked:
 342             *xpath = crm_strdup_printf(XPATH_NODE_ALL_UNLOCKED,
 343                                        uname, uname, expire, uname);
 344             desc_pre = "all state (other than shutdown locks)";
 345             break;
 346         default:
 347             // We called this function incorrectly
 348             pcmk__assert(false);
 349             break;
 350     }
 351 
 352     if (desc != NULL) {
 353         *desc = crm_strdup_printf("%s for node %s", desc_pre, uname);
 354     }
 355 }
 356 
 357 /*!
 358  * \internal
 359  * \brief Delete subsection of a node's CIB \c PCMK__XE_NODE_STATE
 360  *
 361  * \param[in] uname    Desired node
 362  * \param[in] section  Subsection of \c PCMK__XE_NODE_STATE to delete
 363  * \param[in] options  CIB call options to use
 364  */
 365 void
 366 controld_delete_node_state(const char *uname, enum controld_section_e section,
     /*  */
 367                            int options)
 368 {
 369     cib_t *cib = controld_globals.cib_conn;
 370     char *xpath = NULL;
 371     char *desc = NULL;
 372     int cib_rc = pcmk_ok;
 373 
 374     pcmk__assert((uname != NULL) && (cib != NULL));
 375 
 376     controld_node_state_deletion_strings(uname, section, &xpath, &desc);
 377 
 378     cib__set_call_options(options, "node state deletion",
 379                           cib_xpath|cib_multiple);
 380     cib_rc = cib->cmds->remove(cib, xpath, NULL, options);
 381     fsa_register_cib_callback(cib_rc, desc, cib_delete_callback);
 382     crm_info("Deleting %s (via CIB call %d) " CRM_XS " xpath=%s",
 383              desc, cib_rc, xpath);
 384 
 385     // CIB library handles freeing desc
 386     free(xpath);
 387 }
 388 
 389 // Takes node name and resource ID
 390 #define XPATH_RESOURCE_HISTORY "//" PCMK__XE_NODE_STATE                 \
 391                                "[@" PCMK_XA_UNAME "='%s']/"             \
 392                                PCMK__XE_LRM "/" PCMK__XE_LRM_RESOURCES  \
 393                                "/" PCMK__XE_LRM_RESOURCE                \
 394                                "[@" PCMK_XA_ID "='%s']"
 395 // @TODO could add "and @PCMK_OPT_SHUTDOWN_LOCK" to limit to locks
 396 
 397 /*!
 398  * \internal
 399  * \brief Clear resource history from CIB for a given resource and node
 400  *
 401  * \param[in]  rsc_id        ID of resource to be cleared
 402  * \param[in]  node          Node whose resource history should be cleared
 403  * \param[in]  user_name     ACL user name to use
 404  * \param[in]  call_options  CIB call options
 405  *
 406  * \return Standard Pacemaker return code
 407  */
 408 int
 409 controld_delete_resource_history(const char *rsc_id, const char *node,
     /*  */
 410                                  const char *user_name, int call_options)
 411 {
 412     char *desc = NULL;
 413     char *xpath = NULL;
 414     int rc = pcmk_rc_ok;
 415     cib_t *cib = controld_globals.cib_conn;
 416 
 417     CRM_CHECK((rsc_id != NULL) && (node != NULL), return EINVAL);
 418 
 419     desc = crm_strdup_printf("resource history for %s on %s", rsc_id, node);
 420     if (cib == NULL) {
 421         crm_err("Unable to clear %s: no CIB connection", desc);
 422         free(desc);
 423         return ENOTCONN;
 424     }
 425 
 426     // Ask CIB to delete the entry
 427     xpath = crm_strdup_printf(XPATH_RESOURCE_HISTORY, node, rsc_id);
 428 
 429     cib->cmds->set_user(cib, user_name);
 430     rc = cib->cmds->remove(cib, xpath, NULL, call_options|cib_xpath);
 431     cib->cmds->set_user(cib, NULL);
 432 
 433     if (rc < 0) {
 434         rc = pcmk_legacy2rc(rc);
 435         crm_err("Could not delete resource status of %s on %s%s%s: %s "
 436                 CRM_XS " rc=%d", rsc_id, node,
 437                 (user_name? " for user " : ""), (user_name? user_name : ""),
 438                 pcmk_rc_str(rc), rc);
 439         free(desc);
 440         free(xpath);
 441         return rc;
 442     }
 443 
 444     if (pcmk_is_set(call_options, cib_sync_call)) {
 445         if (pcmk_is_set(call_options, cib_dryrun)) {
 446             crm_debug("Deletion of %s would succeed", desc);
 447         } else {
 448             crm_debug("Deletion of %s succeeded", desc);
 449         }
 450         free(desc);
 451 
 452     } else {
 453         crm_info("Clearing %s (via CIB call %d) " CRM_XS " xpath=%s",
 454                  desc, rc, xpath);
 455         fsa_register_cib_callback(rc, desc, cib_delete_callback);
 456         // CIB library handles freeing desc
 457     }
 458 
 459     free(xpath);
 460     return pcmk_rc_ok;
 461 }
 462 
 463 /*!
 464  * \internal
 465  * \brief Build XML and string of parameters meeting some criteria, for digest
 466  *
 467  * \param[in]  op          Executor event with parameter table to use
 468  * \param[in]  metadata    Parsed meta-data for executed resource agent
 469  * \param[in]  param_type  Flag used for selection criteria
 470  * \param[out] result      Will be set to newly created XML with selected
 471  *                         parameters as attributes
 472  *
 473  * \return Newly allocated space-separated string of parameter names
 474  * \note Selection criteria varies by param_type: for the restart digest, we
 475  *       want parameters that are *not* marked reloadable (OCF 1.1) or that
 476  *       *are* marked unique (pre-1.1), for both string and XML results; for the
 477  *       secure digest, we want parameters that *are* marked private for the
 478  *       string, but parameters that are *not* marked private for the XML.
 479  * \note It is the caller's responsibility to free the string return value with
 480  *       \p g_string_free() and the XML result with \p free_xml().
 481  */
 482 static GString *
 483 build_parameter_list(const lrmd_event_data_t *op,
     /*  */
 484                      const struct ra_metadata_s *metadata,
 485                      enum ra_param_flags_e param_type, xmlNode **result)
 486 {
 487     GString *list = NULL;
 488 
 489     *result = pcmk__xe_create(NULL, PCMK_XE_PARAMETERS);
 490 
 491     /* Consider all parameters only except private ones to be consistent with
 492      * what scheduler does with calculate_secure_digest().
 493      */
 494     if (param_type == ra_param_private
 495         && compare_version(controld_globals.dc_version, "3.16.0") >= 0) {
 496         g_hash_table_foreach(op->params, hash2field, *result);
 497         pcmk__filter_op_for_digest(*result);
 498     }
 499 
 500     for (GList *iter = metadata->ra_params; iter != NULL; iter = iter->next) {
 501         struct ra_param_s *param = (struct ra_param_s *) iter->data;
 502 
 503         bool accept_for_list = false;
 504         bool accept_for_xml = false;
 505 
 506         switch (param_type) {
 507             case ra_param_reloadable:
 508                 accept_for_list = !pcmk_is_set(param->rap_flags, param_type);
 509                 accept_for_xml = accept_for_list;
 510                 break;
 511 
 512             case ra_param_unique:
 513                 accept_for_list = pcmk_is_set(param->rap_flags, param_type);
 514                 accept_for_xml = accept_for_list;
 515                 break;
 516 
 517             case ra_param_private:
 518                 accept_for_list = pcmk_is_set(param->rap_flags, param_type);
 519                 accept_for_xml = !accept_for_list;
 520                 break;
 521         }
 522 
 523         if (accept_for_list) {
 524             crm_trace("Attr %s is %s", param->rap_name, ra_param_flag2text(param_type));
 525 
 526             if (list == NULL) {
 527                 // We will later search for " WORD ", so start list with a space
 528                 pcmk__add_word(&list, 256, " ");
 529             }
 530             pcmk__add_word(&list, 0, param->rap_name);
 531 
 532         } else {
 533             crm_trace("Rejecting %s for %s", param->rap_name, ra_param_flag2text(param_type));
 534         }
 535 
 536         if (accept_for_xml) {
 537             const char *v = g_hash_table_lookup(op->params, param->rap_name);
 538 
 539             if (v != NULL) {
 540                 crm_trace("Adding attr %s=%s to the xml result", param->rap_name, v);
 541                 crm_xml_add(*result, param->rap_name, v);
 542             }
 543 
 544         } else {
 545             crm_trace("Removing attr %s from the xml result", param->rap_name);
 546             pcmk__xe_remove_attr(*result, param->rap_name);
 547         }
 548     }
 549 
 550     if (list != NULL) {
 551         // We will later search for " WORD ", so end list with a space
 552         pcmk__add_word(&list, 0, " ");
 553     }
 554     return list;
 555 }
 556 
 557 static void
 558 append_restart_list(lrmd_event_data_t *op, struct ra_metadata_s *metadata,
     /*  */
 559                     xmlNode *update, const char *version)
 560 {
 561     GString *list = NULL;
 562     char *digest = NULL;
 563     xmlNode *restart = NULL;
 564 
 565     CRM_LOG_ASSERT(op->params != NULL);
 566 
 567     if (op->interval_ms > 0) {
 568         /* monitors are not reloadable */
 569         return;
 570     }
 571 
 572     if (pcmk_is_set(metadata->ra_flags, ra_supports_reload_agent)) {
 573         /* Add parameters not marked reloadable to the PCMK__XA_OP_FORCE_RESTART
 574          * list
 575          */
 576         list = build_parameter_list(op, metadata, ra_param_reloadable,
 577                                     &restart);
 578 
 579     } else if (pcmk_is_set(metadata->ra_flags, ra_supports_legacy_reload)) {
 580         /* @COMPAT pre-OCF-1.1 resource agents
 581          *
 582          * Before OCF 1.1, Pacemaker abused "unique=0" to indicate
 583          * reloadability. Add any parameters with unique="1" to the
 584          * PCMK__XA_OP_FORCE_RESTART list.
 585          */
 586         list = build_parameter_list(op, metadata, ra_param_unique, &restart);
 587 
 588     } else {
 589         // Resource does not support agent reloads
 590         return;
 591     }
 592 
 593     digest = calculate_operation_digest(restart, version);
 594     /* Add PCMK__XA_OP_FORCE_RESTART and PCMK__XA_OP_RESTART_DIGEST to indicate
 595      * the resource supports reload, no matter if it actually supports any
 596      * reloadable parameters
 597      */
 598     crm_xml_add(update, PCMK__XA_OP_FORCE_RESTART,
 599                 (list == NULL)? "" : (const char *) list->str);
 600     crm_xml_add(update, PCMK__XA_OP_RESTART_DIGEST, digest);
 601 
 602     if ((list != NULL) && (list->len > 0)) {
 603         crm_trace("%s: %s, %s", op->rsc_id, digest, (const char *) list->str);
 604     } else {
 605         crm_trace("%s: %s", op->rsc_id, digest);
 606     }
 607 
 608     if (list != NULL) {
 609         g_string_free(list, TRUE);
 610     }
 611     free_xml(restart);
 612     free(digest);
 613 }
 614 
 615 static void
 616 append_secure_list(lrmd_event_data_t *op, struct ra_metadata_s *metadata,
     /*  */
 617                    xmlNode *update, const char *version)
 618 {
 619     GString *list = NULL;
 620     char *digest = NULL;
 621     xmlNode *secure = NULL;
 622 
 623     CRM_LOG_ASSERT(op->params != NULL);
 624 
 625     /* To keep PCMK__XA_OP_SECURE_PARAMS short, we want it to contain the secure
 626      * parameters but PCMK__XA_OP_SECURE_DIGEST to be based on the insecure ones
 627      */
 628     list = build_parameter_list(op, metadata, ra_param_private, &secure);
 629 
 630     if (list != NULL) {
 631         digest = calculate_operation_digest(secure, version);
 632         crm_xml_add(update, PCMK__XA_OP_SECURE_PARAMS,
 633                     (const char *) list->str);
 634         crm_xml_add(update, PCMK__XA_OP_SECURE_DIGEST, digest);
 635 
 636         crm_trace("%s: %s, %s", op->rsc_id, digest, (const char *) list->str);
 637         g_string_free(list, TRUE);
 638     } else {
 639         crm_trace("%s: no secure parameters", op->rsc_id);
 640     }
 641 
 642     free_xml(secure);
 643     free(digest);
 644 }
 645 
 646 /*!
 647  * \internal
 648  * \brief Create XML for a resource history entry
 649  *
 650  * \param[in]     func       Function name of caller
 651  * \param[in,out] parent     XML to add entry to
 652  * \param[in]     rsc        Affected resource
 653  * \param[in,out] op         Action to add an entry for (or NULL to do nothing)
 654  * \param[in]     node_name  Node where action occurred
 655  */
 656 void
 657 controld_add_resource_history_xml_as(const char *func, xmlNode *parent,
     /*  */
 658                                      const lrmd_rsc_info_t *rsc,
 659                                      lrmd_event_data_t *op,
 660                                      const char *node_name)
 661 {
 662     int target_rc = 0;
 663     xmlNode *xml_op = NULL;
 664     struct ra_metadata_s *metadata = NULL;
 665     const char *caller_version = NULL;
 666     lrm_state_t *lrm_state = NULL;
 667 
 668     if (op == NULL) {
 669         return;
 670     }
 671 
 672     target_rc = rsc_op_expected_rc(op);
 673 
 674     caller_version = g_hash_table_lookup(op->params, PCMK_XA_CRM_FEATURE_SET);
 675     CRM_CHECK(caller_version != NULL, caller_version = CRM_FEATURE_SET);
 676 
 677     xml_op = pcmk__create_history_xml(parent, op, caller_version, target_rc,
 678                                       controld_globals.our_nodename, func);
 679     if (xml_op == NULL) {
 680         return;
 681     }
 682 
 683     if ((rsc == NULL) || (op->params == NULL)
 684         || !crm_op_needs_metadata(rsc->standard, op->op_type)) {
 685 
 686         crm_trace("No digests needed for %s action on %s (params=%p rsc=%p)",
 687                   op->op_type, op->rsc_id, op->params, rsc);
 688         return;
 689     }
 690 
 691     lrm_state = lrm_state_find(node_name);
 692     if (lrm_state == NULL) {
 693         crm_warn("Cannot calculate digests for operation " PCMK__OP_FMT
 694                  " because we have no connection to executor for %s",
 695                  op->rsc_id, op->op_type, op->interval_ms, node_name);
 696         return;
 697     }
 698 
 699     /* Ideally the metadata is cached, and the agent is just a fallback.
 700      *
 701      * @TODO Go through all callers and ensure they get metadata asynchronously
 702      * first.
 703      */
 704     metadata = controld_get_rsc_metadata(lrm_state, rsc,
 705                                          controld_metadata_from_agent
 706                                          |controld_metadata_from_cache);
 707     if (metadata == NULL) {
 708         return;
 709     }
 710 
 711     crm_trace("Including additional digests for %s:%s:%s",
 712               rsc->standard, rsc->provider, rsc->type);
 713     append_restart_list(op, metadata, xml_op, caller_version);
 714     append_secure_list(op, metadata, xml_op, caller_version);
 715 
 716     return;
 717 }
 718 
 719 /*!
 720  * \internal
 721  * \brief Record an action as pending in the CIB, if appropriate
 722  *
 723  * \param[in]     node_name  Node where the action is pending
 724  * \param[in]     rsc        Resource that action is for
 725  * \param[in,out] op         Pending action
 726  *
 727  * \return true if action was recorded in CIB, otherwise false
 728  */
 729 bool
 730 controld_record_pending_op(const char *node_name, const lrmd_rsc_info_t *rsc,
     /*  */
 731                            lrmd_event_data_t *op)
 732 {
 733     const char *record_pending = NULL;
 734 
 735     CRM_CHECK((node_name != NULL) && (rsc != NULL) && (op != NULL),
 736               return false);
 737 
 738     // Never record certain operation types as pending
 739     if ((op->op_type == NULL) || (op->params == NULL)
 740         || !controld_action_is_recordable(op->op_type)) {
 741         return false;
 742     }
 743 
 744     // Check action's PCMK_META_RECORD_PENDING meta-attribute (defaults to true)
 745     record_pending = crm_meta_value(op->params, PCMK_META_RECORD_PENDING);
 746     if ((record_pending != NULL) && !crm_is_true(record_pending)) {
 747         return false;
 748     }
 749 
 750     op->call_id = -1;
 751     // coverity[store_truncates_time_t]
 752     op->t_run = time(NULL);
 753     op->t_rcchange = op->t_run;
 754 
 755     lrmd__set_result(op, PCMK_OCF_UNKNOWN, PCMK_EXEC_PENDING, NULL);
 756 
 757     crm_debug("Recording pending %s-interval %s for %s on %s in the CIB",
 758               pcmk__readable_interval(op->interval_ms), op->op_type, op->rsc_id,
 759               node_name);
 760     controld_update_resource_history(node_name, rsc, op, 0);
 761     return true;
 762 }
 763 
 764 static void
 765 cib_rsc_callback(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data)
     /*  */
 766 {
 767     switch (rc) {
 768         case pcmk_ok:
 769         case -pcmk_err_diff_failed:
 770         case -pcmk_err_diff_resync:
 771             crm_trace("Resource history update completed (call=%d rc=%d)",
 772                       call_id, rc);
 773             break;
 774         default:
 775             if (call_id > 0) {
 776                 crm_warn("Resource history update %d failed: %s "
 777                          CRM_XS " rc=%d", call_id, pcmk_strerror(rc), rc);
 778             } else {
 779                 crm_warn("Resource history update failed: %s " CRM_XS " rc=%d",
 780                          pcmk_strerror(rc), rc);
 781             }
 782     }
 783 
 784     if (call_id == pending_rsc_update) {
 785         pending_rsc_update = 0;
 786         controld_trigger_fsa();
 787     }
 788 }
 789 
 790 /* Only successful stops, and probes that found the resource inactive, get locks
 791  * recorded in the history. This ensures the resource stays locked to the node
 792  * until it is active there again after the node comes back up.
 793  */
 794 static bool
 795 should_preserve_lock(lrmd_event_data_t *op)
     /*  */
 796 {
 797     if (!pcmk_is_set(controld_globals.flags, controld_shutdown_lock_enabled)) {
 798         return false;
 799     }
 800     if (!strcmp(op->op_type, PCMK_ACTION_STOP) && (op->rc == PCMK_OCF_OK)) {
 801         return true;
 802     }
 803     if (!strcmp(op->op_type, PCMK_ACTION_MONITOR)
 804         && (op->rc == PCMK_OCF_NOT_RUNNING)) {
 805         return true;
 806     }
 807     return false;
 808 }
 809 
 810 /*!
 811  * \internal
 812  * \brief Request a CIB update
 813  *
 814  * \param[in]     section    Section of CIB to update
 815  * \param[in]     data       New XML of CIB section to update
 816  * \param[in]     options    CIB call options
 817  * \param[in]     callback   If not \c NULL, set this as the operation callback
 818  *
 819  * \return Standard Pacemaker return code
 820  *
 821  * \note If \p callback is \p cib_rsc_callback(), the CIB update's call ID is
 822  *       stored in \p pending_rsc_update on success.
 823  */
 824 int
 825 controld_update_cib(const char *section, xmlNode *data, int options,
     /*  */
 826                     void (*callback)(xmlNode *, int, int, xmlNode *, void *))
 827 {
 828     cib_t *cib = controld_globals.cib_conn;
 829     int cib_rc = -ENOTCONN;
 830 
 831     pcmk__assert(data != NULL);
 832 
 833     if (cib != NULL) {
 834         cib_rc = cib->cmds->modify(cib, section, data, options);
 835         if (cib_rc >= 0) {
 836             crm_debug("Submitted CIB update %d for %s section",
 837                       cib_rc, section);
 838         }
 839     }
 840 
 841     if (callback == NULL) {
 842         if (cib_rc < 0) {
 843             crm_err("Failed to update CIB %s section: %s",
 844                     section, pcmk_rc_str(pcmk_legacy2rc(cib_rc)));
 845         }
 846 
 847     } else {
 848         if ((cib_rc >= 0) && (callback == cib_rsc_callback)) {
 849             /* Checking for a particular callback is a little hacky, but it
 850              * didn't seem worth adding an output argument for cib_rc for just
 851              * one use case.
 852              */
 853             pending_rsc_update = cib_rc;
 854         }
 855         fsa_register_cib_callback(cib_rc, NULL, callback);
 856     }
 857 
 858     return (cib_rc >= 0)? pcmk_rc_ok : pcmk_legacy2rc(cib_rc);
 859 }
 860 
 861 /*!
 862  * \internal
 863  * \brief Update resource history entry in CIB
 864  *
 865  * \param[in]     node_name  Node where action occurred
 866  * \param[in]     rsc        Resource that action is for
 867  * \param[in,out] op         Action to record
 868  * \param[in]     lock_time  If nonzero, when resource was locked to node
 869  *
 870  * \note On success, the CIB update's call ID will be stored in
 871  *       pending_rsc_update.
 872  */
 873 void
 874 controld_update_resource_history(const char *node_name,
     /*  */
 875                                  const lrmd_rsc_info_t *rsc,
 876                                  lrmd_event_data_t *op, time_t lock_time)
 877 {
 878     xmlNode *update = NULL;
 879     xmlNode *xml = NULL;
 880     int call_opt = crmd_cib_smart_opt();
 881     const char *node_id = NULL;
 882     const char *container = NULL;
 883 
 884     CRM_CHECK((node_name != NULL) && (op != NULL), return);
 885 
 886     if (rsc == NULL) {
 887         crm_warn("Resource %s no longer exists in the executor", op->rsc_id);
 888         controld_ack_event_directly(NULL, NULL, rsc, op, op->rsc_id);
 889         return;
 890     }
 891 
 892     // <status>
 893     update = pcmk__xe_create(NULL, PCMK_XE_STATUS);
 894 
 895     //   <node_state ...>
 896     xml = pcmk__xe_create(update, PCMK__XE_NODE_STATE);
 897     if (pcmk__str_eq(node_name, controld_globals.our_nodename,
 898                      pcmk__str_casei)) {
 899         node_id = controld_globals.our_uuid;
 900     } else {
 901         node_id = node_name;
 902         pcmk__xe_set_bool_attr(xml, PCMK_XA_REMOTE_NODE, true);
 903     }
 904     crm_xml_add(xml, PCMK_XA_ID, node_id);
 905     crm_xml_add(xml, PCMK_XA_UNAME, node_name);
 906     crm_xml_add(xml, PCMK_XA_CRM_DEBUG_ORIGIN, __func__);
 907 
 908     //     <lrm ...>
 909     xml = pcmk__xe_create(xml, PCMK__XE_LRM);
 910     crm_xml_add(xml, PCMK_XA_ID, node_id);
 911 
 912     //       <lrm_resources>
 913     xml = pcmk__xe_create(xml, PCMK__XE_LRM_RESOURCES);
 914 
 915     //         <lrm_resource ...>
 916     xml = pcmk__xe_create(xml, PCMK__XE_LRM_RESOURCE);
 917     crm_xml_add(xml, PCMK_XA_ID, op->rsc_id);
 918     crm_xml_add(xml, PCMK_XA_CLASS, rsc->standard);
 919     crm_xml_add(xml, PCMK_XA_PROVIDER, rsc->provider);
 920     crm_xml_add(xml, PCMK_XA_TYPE, rsc->type);
 921     if (lock_time != 0) {
 922         /* Actions on a locked resource should either preserve the lock by
 923          * recording it with the action result, or clear it.
 924          */
 925         if (!should_preserve_lock(op)) {
 926             lock_time = 0;
 927         }
 928         crm_xml_add_ll(xml, PCMK_OPT_SHUTDOWN_LOCK, (long long) lock_time);
 929     }
 930     if (op->params != NULL) {
 931         container = g_hash_table_lookup(op->params,
 932                                         CRM_META "_" PCMK__META_CONTAINER);
 933         if (container != NULL) {
 934             crm_trace("Resource %s is a part of container resource %s",
 935                       op->rsc_id, container);
 936             crm_xml_add(xml, PCMK__META_CONTAINER, container);
 937         }
 938     }
 939 
 940     //           <lrm_resource_op ...> (possibly more than one)
 941     controld_add_resource_history_xml(xml, rsc, op, node_name);
 942 
 943     /* Update CIB asynchronously. Even if it fails, the resource state should be
 944      * discovered during the next election. Worst case, the node is wrongly
 945      * fenced for running a resource it isn't.
 946      */
 947     crm_log_xml_trace(update, __func__);
 948     controld_update_cib(PCMK_XE_STATUS, update, call_opt, cib_rsc_callback);
 949     free_xml(update);
 950 }
 951 
 952 /*!
 953  * \internal
 954  * \brief Erase an LRM history entry from the CIB, given the operation data
 955  *
 956  * \param[in] op         Operation whose history should be deleted
 957  */
 958 void
 959 controld_delete_action_history(const lrmd_event_data_t *op)
     /*  */
 960 {
 961     xmlNode *xml_top = NULL;
 962 
 963     CRM_CHECK(op != NULL, return);
 964 
 965     xml_top = pcmk__xe_create(NULL, PCMK__XE_LRM_RSC_OP);
 966     crm_xml_add_int(xml_top, PCMK__XA_CALL_ID, op->call_id);
 967     crm_xml_add(xml_top, PCMK__XA_TRANSITION_KEY, op->user_data);
 968 
 969     if (op->interval_ms > 0) {
 970         char *op_id = pcmk__op_key(op->rsc_id, op->op_type, op->interval_ms);
 971 
 972         /* Avoid deleting last_failure too (if it was a result of this recurring op failing) */
 973         crm_xml_add(xml_top, PCMK_XA_ID, op_id);
 974         free(op_id);
 975     }
 976 
 977     crm_debug("Erasing resource operation history for " PCMK__OP_FMT " (call=%d)",
 978               op->rsc_id, op->op_type, op->interval_ms, op->call_id);
 979 
 980     controld_globals.cib_conn->cmds->remove(controld_globals.cib_conn,
 981                                             PCMK_XE_STATUS, xml_top, cib_none);
 982     crm_log_xml_trace(xml_top, "op:cancel");
 983     free_xml(xml_top);
 984 }
 985 
 986 /* Define xpath to find LRM resource history entry by node and resource */
 987 #define XPATH_HISTORY                                   \
 988     "/" PCMK_XE_CIB "/" PCMK_XE_STATUS                  \
 989     "/" PCMK__XE_NODE_STATE "[@" PCMK_XA_UNAME "='%s']" \
 990     "/" PCMK__XE_LRM "/" PCMK__XE_LRM_RESOURCES         \
 991     "/" PCMK__XE_LRM_RESOURCE "[@" PCMK_XA_ID "='%s']"  \
 992     "/" PCMK__XE_LRM_RSC_OP
 993 
 994 /* ... and also by operation key */
 995 #define XPATH_HISTORY_ID XPATH_HISTORY "[@" PCMK_XA_ID "='%s']"
 996 
 997 /* ... and also by operation key and operation call ID */
 998 #define XPATH_HISTORY_CALL XPATH_HISTORY \
 999     "[@" PCMK_XA_ID "='%s' and @" PCMK__XA_CALL_ID "='%d']"
1000 
1001 /* ... and also by operation key and original operation key */
1002 #define XPATH_HISTORY_ORIG XPATH_HISTORY \
1003     "[@" PCMK_XA_ID "='%s' and @" PCMK__XA_OPERATION_KEY "='%s']"
1004 
1005 /*!
1006  * \internal
1007  * \brief Delete a last_failure resource history entry from the CIB
1008  *
1009  * \param[in] rsc_id       Name of resource to clear history for
1010  * \param[in] node         Name of node to clear history for
1011  * \param[in] action       If specified, delete only if this was failed action
1012  * \param[in] interval_ms  If \p action is specified, it has this interval
1013  */
1014 void
1015 controld_cib_delete_last_failure(const char *rsc_id, const char *node,
     /*  */
1016                                  const char *action, guint interval_ms)
1017 {
1018     char *xpath = NULL;
1019     char *last_failure_key = NULL;
1020     CRM_CHECK((rsc_id != NULL) && (node != NULL), return);
1021 
1022     // Generate XPath to match desired entry
1023     last_failure_key = pcmk__op_key(rsc_id, "last_failure", 0);
1024     if (action == NULL) {
1025         xpath = crm_strdup_printf(XPATH_HISTORY_ID, node, rsc_id,
1026                                   last_failure_key);
1027     } else {
1028         char *action_key = pcmk__op_key(rsc_id, action, interval_ms);
1029 
1030         xpath = crm_strdup_printf(XPATH_HISTORY_ORIG, node, rsc_id,
1031                                   last_failure_key, action_key);
1032         free(action_key);
1033     }
1034     free(last_failure_key);
1035 
1036     controld_globals.cib_conn->cmds->remove(controld_globals.cib_conn, xpath,
1037                                             NULL, cib_xpath);
1038     free(xpath);
1039 }
1040 
1041 /*!
1042  * \internal
1043  * \brief Delete resource history entry from the CIB, given operation key
1044  *
1045  * \param[in] rsc_id     Name of resource to clear history for
1046  * \param[in] node       Name of node to clear history for
1047  * \param[in] key        Operation key of operation to clear history for
1048  * \param[in] call_id    If specified, delete entry only if it has this call ID
1049  */
1050 void
1051 controld_delete_action_history_by_key(const char *rsc_id, const char *node,
     /*  */
1052                                       const char *key, int call_id)
1053 {
1054     char *xpath = NULL;
1055 
1056     CRM_CHECK((rsc_id != NULL) && (node != NULL) && (key != NULL), return);
1057 
1058     if (call_id > 0) {
1059         xpath = crm_strdup_printf(XPATH_HISTORY_CALL, node, rsc_id, key,
1060                                   call_id);
1061     } else {
1062         xpath = crm_strdup_printf(XPATH_HISTORY_ID, node, rsc_id, key);
1063     }
1064     controld_globals.cib_conn->cmds->remove(controld_globals.cib_conn, xpath,
1065                                             NULL, cib_xpath);
1066     free(xpath);
1067 }
/* */
root/daemons/controld/controld_cib.c

DEFINITIONS
