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

DEFINITIONS
