diff --git a/daemons/attrd/attrd_attributes.c b/daemons/attrd/attrd_attributes.c index fdc238375e..a18076f7a3 100644 --- a/daemons/attrd/attrd_attributes.c +++ b/daemons/attrd/attrd_attributes.c @@ -1,284 +1,476 @@ /* - * Copyright 2013-2024 the Pacemaker project contributors + * Copyright 2013-2025 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #include #include "pacemaker-attrd.h" static attribute_t * attrd_create_attribute(xmlNode *xml) { int is_private = 0; long long dampen = 0; const char *name = crm_element_value(xml, PCMK__XA_ATTR_NAME); const char *set_type = crm_element_value(xml, PCMK__XA_ATTR_SET_TYPE); const char *dampen_s = crm_element_value(xml, PCMK__XA_ATTR_DAMPENING); attribute_t *a = NULL; if (set_type == NULL) { set_type = PCMK_XE_INSTANCE_ATTRIBUTES; } /* Set type is meaningful only when writing to the CIB. Private * attributes are not written. */ crm_element_value_int(xml, PCMK__XA_ATTR_IS_PRIVATE, &is_private); if (!is_private && !pcmk__str_any_of(set_type, PCMK_XE_INSTANCE_ATTRIBUTES, PCMK_XE_UTILIZATION, NULL)) { crm_warn("Ignoring attribute %s with invalid set type %s", pcmk__s(name, "(unidentified)"), set_type); return NULL; } a = pcmk__assert_alloc(1, sizeof(attribute_t)); a->id = pcmk__str_copy(name); a->set_type = pcmk__str_copy(set_type); a->set_id = crm_element_value_copy(xml, PCMK__XA_ATTR_SET); a->user = crm_element_value_copy(xml, PCMK__XA_ATTR_USER); a->values = pcmk__strikey_table(NULL, attrd_free_attribute_value); if (is_private) { attrd_set_attr_flags(a, attrd_attr_is_private); } if (dampen_s != NULL) { dampen = crm_get_msec(dampen_s); } if (dampen > 0) { a->timeout_ms = (int) QB_MIN(dampen, INT_MAX); a->timer = attrd_add_timer(a->id, a->timeout_ms, a); } else if (dampen < 0) { crm_warn("Ignoring invalid delay %s for attribute %s", dampen_s, a->id); } crm_trace("Created attribute %s with %s write delay and %s CIB user", a->id, ((dampen > 0)? pcmk__readable_interval(a->timeout_ms) : "no"), pcmk__s(a->user, "default")); g_hash_table_replace(attributes, a->id, a); return a; } static int attrd_update_dampening(attribute_t *a, xmlNode *xml, const char *attr) { const char *dvalue = crm_element_value(xml, PCMK__XA_ATTR_DAMPENING); long long dampen = 0; if (dvalue == NULL) { crm_warn("Could not update %s: peer did not specify value for delay", attr); return EINVAL; } dampen = crm_get_msec(dvalue); if (dampen < 0) { crm_warn("Could not update %s: invalid delay value %dms (%s)", attr, dampen, dvalue); return EINVAL; } if (a->timeout_ms != dampen) { mainloop_timer_del(a->timer); a->timeout_ms = (int) QB_MIN(dampen, INT_MAX); if (dampen > 0) { a->timer = attrd_add_timer(attr, a->timeout_ms, a); crm_info("Update attribute %s delay to %dms (%s)", attr, dampen, dvalue); } else { a->timer = NULL; crm_info("Update attribute %s to remove delay", attr); } /* If dampening changed, do an immediate write-out, * otherwise repeated dampening changes would prevent write-outs */ attrd_write_or_elect_attribute(a); } return pcmk_rc_ok; } GHashTable *attributes = NULL; /*! * \internal * \brief Create an XML representation of an attribute for use in peer messages * * \param[in,out] parent Create attribute XML as child element of this * \param[in] a Attribute to represent * \param[in] v Attribute value to represent * \param[in] force_write If true, value should be written even if unchanged * * \return XML representation of attribute */ xmlNode * attrd_add_value_xml(xmlNode *parent, const attribute_t *a, const attribute_value_t *v, bool force_write) { xmlNode *xml = pcmk__xe_create(parent, __func__); crm_xml_add(xml, PCMK__XA_ATTR_NAME, a->id); crm_xml_add(xml, PCMK__XA_ATTR_SET_TYPE, a->set_type); crm_xml_add(xml, PCMK__XA_ATTR_SET, a->set_id); crm_xml_add(xml, PCMK__XA_ATTR_USER, a->user); crm_xml_add(xml, PCMK__XA_ATTR_HOST, v->nodename); /* @COMPAT Prior to 2.1.10 and 3.0.1, the node's cluster ID was added * instead of its XML ID. For Corosync and Pacemaker Remote nodes, those are * the same, but if we ever support node XML IDs that differ from their * cluster IDs, we will have to drop support for rolling upgrades from * versions before those. */ crm_xml_add(xml, PCMK__XA_ATTR_HOST_ID, attrd_get_node_xml_id(v->nodename)); crm_xml_add(xml, PCMK__XA_ATTR_VALUE, v->current); crm_xml_add_int(xml, PCMK__XA_ATTR_DAMPENING, pcmk__timeout_ms2s(a->timeout_ms)); crm_xml_add_int(xml, PCMK__XA_ATTR_IS_PRIVATE, pcmk_is_set(a->flags, attrd_attr_is_private)); crm_xml_add_int(xml, PCMK__XA_ATTR_IS_REMOTE, pcmk_is_set(v->flags, attrd_value_remote)); crm_xml_add_int(xml, PCMK__XA_ATTRD_IS_FORCE_WRITE, force_write); return xml; } void attrd_clear_value_seen(void) { GHashTableIter aIter; GHashTableIter vIter; attribute_t *a; attribute_value_t *v = NULL; g_hash_table_iter_init(&aIter, attributes); while (g_hash_table_iter_next(&aIter, NULL, (gpointer *) & a)) { g_hash_table_iter_init(&vIter, a->values); while (g_hash_table_iter_next(&vIter, NULL, (gpointer *) & v)) { attrd_clear_value_flags(v, attrd_value_from_peer); } } } attribute_t * attrd_populate_attribute(xmlNode *xml, const char *attr) { attribute_t *a = NULL; bool update_both = false; const char *op = crm_element_value(xml, PCMK_XA_TASK); // NULL because PCMK__ATTRD_CMD_SYNC_RESPONSE has no PCMK_XA_TASK update_both = pcmk__str_eq(op, PCMK__ATTRD_CMD_UPDATE_BOTH, pcmk__str_null_matches); // Look up or create attribute entry a = g_hash_table_lookup(attributes, attr); if (a == NULL) { if (update_both || pcmk__str_eq(op, PCMK__ATTRD_CMD_UPDATE, pcmk__str_none)) { a = attrd_create_attribute(xml); if (a == NULL) { return NULL; } } else { crm_warn("Could not update %s: attribute not found", attr); return NULL; } } // Update attribute dampening if (update_both || pcmk__str_eq(op, PCMK__ATTRD_CMD_UPDATE_DELAY, pcmk__str_none)) { int rc = attrd_update_dampening(a, xml, attr); if (rc != pcmk_rc_ok || !update_both) { return NULL; } } return a; } /*! * \internal * \brief Get the XML ID used to write out an attribute set * * \param[in] attr Attribute to get set ID for * \param[in] node_state_id XML ID of node state that attribute value is for * * \return Newly allocated string with XML ID to use for \p attr set */ char * attrd_set_id(const attribute_t *attr, const char *node_state_id) { char *set_id = NULL; pcmk__assert((attr != NULL) && (node_state_id != NULL)); if (pcmk__str_empty(attr->set_id)) { /* @COMPAT This should really take the set type into account. Currently * we use the same XML ID for transient attributes and utilization * attributes. It doesn't cause problems because the status section is * not limited by the schema in any way, but it's still unfortunate. * For backward compatibility reasons, we can't change this. */ set_id = crm_strdup_printf("%s-%s", PCMK_XE_STATUS, node_state_id); } else { /* @COMPAT When the user specifies a set ID for an attribute, it is the * same for every node. That is less than ideal, but again, the schema * doesn't enforce anything for the status section. We couldn't change * it without allowing the set ID to vary per value rather than per * attribute, which would break backward compatibility, pose design * challenges, and potentially cause problems in rolling upgrades. */ set_id = pcmk__str_copy(attr->set_id); } pcmk__xml_sanitize_id(set_id); return set_id; } /*! * \internal * \brief Get the XML ID used to write out an attribute value * * \param[in] attr Attribute to get value XML ID for * \param[in] node_state_id UUID of node that attribute value is for * * \return Newly allocated string with XML ID of \p attr value */ char * attrd_nvpair_id(const attribute_t *attr, const char *node_state_id) { char *nvpair_id = NULL; if (attr->set_id != NULL) { nvpair_id = crm_strdup_printf("%s-%s", attr->set_id, attr->id); } else { nvpair_id = crm_strdup_printf(PCMK_XE_STATUS "-%s-%s", node_state_id, attr->id); } pcmk__xml_sanitize_id(nvpair_id); return nvpair_id; } + +/*! + * \internal + * \brief Check whether an attribute is one that must be written to the CIB + * + * \param[in] a Attribute to check + * + * \return false if we are in standalone mode or \p a is private, otherwise true + */ +bool +attrd_for_cib(const attribute_t *a) +{ + return !stand_alone && (a != NULL) + && !pcmk_is_set(a->flags, attrd_attr_is_private); +} + +/*! + * \internal + * \brief Drop NULL attribute values as indicated by given function + * + * Drop all NULL node attribute values that a given function indicates should + * be, based on the XML ID of an element that was removed from the CIB. + * + * \param[in] cib_id ID of XML element that was removed from CIB + * (a name/value pair, an attribute set, or a node state) + * \param[in] set_type If not NULL, drop only attributes with this set type + * \param[in] func Call this function for every attribute/value + * combination; keep the value if it returns 0, drop + * the value and keep checking the attribute's other + * values if it returns 1, or drop value and stop checking + * the attribute's further values if it returns 2 + */ +static void +drop_removed_values(const char *cib_id, const char *set_type, + int (*func)(const attribute_t *, const attribute_value_t *, + const char *)) +{ + attribute_t *a = NULL; + GHashTableIter attr_iter; + const char *entry_type = pcmk__s(set_type, "status entry"); // for log + + CRM_CHECK((cib_id != NULL) && (func != NULL), return); + + // Check every attribute ... + g_hash_table_iter_init(&attr_iter, attributes); + while (g_hash_table_iter_next(&attr_iter, NULL, (gpointer *) &a)) { + attribute_value_t *v = NULL; + GHashTableIter value_iter; + + if (!attrd_for_cib(a) + || ((set_type != NULL) + && !pcmk__str_eq(a->set_type, set_type, pcmk__str_none))) { + continue; + } + + // Check every value of the attribute ... + g_hash_table_iter_init(&value_iter, a->values); + while (g_hash_table_iter_next(&value_iter, NULL, (gpointer *) &v)) { + int rc = 0; + + if (v->current != NULL) { + continue; + } + + if (attrd_get_node_xml_id(v->nodename) == NULL) { + /* This shouldn't be a significant issue, since we will know the + * XML ID if *any* attribute for the node has ever been written. + */ + crm_trace("Ignoring %s[%s] after CIB erasure of %s %s because " + "its node XML ID is unknown (possibly attribute was " + "never written to CIB)", + a->id, v->nodename, entry_type, cib_id); + continue; + } + + rc = func(a, v, cib_id); + if (rc > 0) { + crm_debug("Dropping %s[%s] after CIB erasure of %s %s", + a->id, v->nodename, entry_type, cib_id); + g_hash_table_iter_remove(&value_iter); + if (rc > 1) { + return; + } + } + } + } +} + +/*! + * \internal + * \brief Check whether an attribute value has a given XML ID + * + * \param[in] a Attribute being checked + * \param[in] v Attribute value being checked + * \param[in] cib_id ID of name/value pair element that was removed from CIB + * + * \return 2 if value matches XML ID, otherwise 0 + */ +static int +nvpair_matches(const attribute_t *a, const attribute_value_t *v, + const char *cib_id) +{ + char *id = attrd_nvpair_id(a, attrd_get_node_xml_id(v->nodename)); + + /* The attribute manager doesn't enforce uniqueness for value XML IDs + * (schema validation could be disabled), but in practice they should be, + * so we can stop looping if we find a match. + */ + int rc = pcmk__str_eq(id, cib_id, pcmk__str_none)? 2 : 0; + + free(id); + return rc; +} + +/*! + * \internal + * \brief Drop attribute value corresponding to given removed CIB entry + * + * \param[in] cib_id ID of name/value pair element that was removed from CIB + */ +void +attrd_drop_removed_value(const char *cib_id) +{ + drop_removed_values(cib_id, NULL, nvpair_matches); +} + +/*! + * \internal + * \brief Check whether an attribute value has a given attribute set ID + * + * \param[in] a Attribute being checked + * \param[in] v Attribute value being checked + * \param[in] cib_id ID of attribute set that was removed from CIB + * + * \return 1 if value matches XML ID, otherwise 0 + */ +static int +set_id_matches(const attribute_t *a, const attribute_value_t *v, + const char *cib_id) +{ + char *id = attrd_set_id(a, attrd_get_node_xml_id(v->nodename)); + int rc = pcmk__str_eq(id, cib_id, pcmk__str_none)? 1: 0; + + free(id); + return rc; +} + +/*! + * \internal + * \brief Drop all removed attribute values for an attribute set + * + * \param[in] set_type XML element name of set that was removed + * \param[in] cib_id ID of attribute set that was removed from CIB + */ +void +attrd_drop_removed_set(const char *set_type, const char *cib_id) +{ + drop_removed_values(cib_id, set_type, set_id_matches); +} + +/*! + * \internal + * \brief Check whether an attribute value has a given node state XML ID + * + * \param[in] a Attribute being checked + * \param[in] v Attribute value being checked + * \param[in] cib_id ID of node state that was removed from CIB + * + * \return 1 if value matches removed ID, otherwise 0 + */ +static int +node_matches(const attribute_t *a, const attribute_value_t *v, + const char *cib_id) +{ + if (pcmk__str_eq(cib_id, attrd_get_node_xml_id(v->nodename), + pcmk__str_none)) { + return 1; + } + return 0; +} + +/*! + * \internal + * \brief Drop all removed attribute values for a node + * + * \param[in] cib_id ID of node state that was removed from CIB + */ +void +attrd_drop_removed_values(const char *cib_id) +{ + drop_removed_values(cib_id, NULL, node_matches); +} diff --git a/daemons/attrd/attrd_cib.c b/daemons/attrd/attrd_cib.c index 4231e4a668..588e779348 100644 --- a/daemons/attrd/attrd_cib.c +++ b/daemons/attrd/attrd_cib.c @@ -1,712 +1,871 @@ /* - * Copyright 2013-2024 the Pacemaker project contributors + * Copyright 2013-2025 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include +#include // for regex.h #include +#include // strndup() +#include // regcomp(), regexec(), regex_t, regmatch_t, regoff_t #include #include #include #include // cib__* #include #include #include #include #include // pcmk__get_node() #include "pacemaker-attrd.h" static int last_cib_op_done = 0; static void write_attribute(attribute_t *a, bool ignore_delay); static void attrd_cib_destroy_cb(gpointer user_data) { cib_t *cib = user_data; cib->cmds->signoff(cib); - if (attrd_shutting_down(false)) { + if (attrd_shutting_down()) { crm_info("Disconnected from the CIB manager"); } else { // @TODO This should trigger a reconnect, not a shutdown crm_crit("Lost connection to the CIB manager, shutting down"); attrd_exit_status = CRM_EX_DISCONNECT; attrd_shutdown(0); } } +/* In a CIB patchset, deletions have the XPath to the deleted element, like: + * + * /cib/status/node_state[@id='X'] + * + * This regular expression checks whether a node state was deleted, or transient + * attributes beneath that, or an attribute set beneath that, or a name/value + * pair beneath that. + */ +#define ID_REGEX "\\[@" PCMK_XA_ID "='([^']+)'\\]" +#define DELETION_REGEX "/" PCMK_XE_CIB "/" PCMK_XE_STATUS \ + "/" PCMK__XE_NODE_STATE ID_REGEX \ + "(/" PCMK__XE_TRANSIENT_ATTRIBUTES ID_REGEX ")?" \ + "(/([^[/]+)" ID_REGEX ")?" "(/" PCMK_XE_NVPAIR ID_REGEX ")?$" + +// Number of parenthesized submatches in DELETION_REGEX plus 1 for entire match +#define DELETION_NMATCH 9 + +/*! + * \internal + * \brief Duplicate a regular expression submatch + * + * \param[in] string String being matched against a regular expression + * \param[in] matches Submatches, as determined by regexec() + * \param[in] submatch Desired index into \p matches + * + * \return Newly allocated string with desired submatch + * \note This asserts on allocation failure, so the result is guaranteed to be + * non-NULL. + */ +static char * +re_submatch(const char *string, regmatch_t matches[], size_t submatch) +{ + const regoff_t start = matches[submatch].rm_so; + char *match = strndup(string + start, matches[submatch].rm_eo - start); + + pcmk__mem_assert(match); + return match; +} + +/*! + * \internal + * \brief Check a patchset change for deletion of node attribute values + * + * \param[in] xml Patchset change element + * \param[in] data Ignored + * + * \return pcmk_rc_ok (to always continue to next patchset change) + */ +static int +drop_values_in_deletion(xmlNode *xml, void *data) +{ + const char *value = NULL; + char *id = NULL; + regmatch_t matches[DELETION_NMATCH]; + + static regex_t re; + static bool re_compiled = false; + + // Skip this change if it does not look like a deletion + value = crm_element_value(xml, PCMK_XA_OPERATION); + if (!pcmk__str_eq(value, "delete", pcmk__str_none)) { + return pcmk_rc_ok; + } + value = crm_element_value(xml, PCMK_XA_PATH); + if (value == NULL) { + crm_warn("Ignoring malformed deletion in " + "CIB change notification: No " PCMK_XA_PATH); + return pcmk_rc_ok; + } + + // Check whether deleted XPath could contain node attribute values + if (!re_compiled) { + CRM_CHECK(regcomp(&re, DELETION_REGEX, REG_EXTENDED) == 0, + return pcmk_rc_ok); + re_compiled = true; + } + if (regexec(&re, value, DELETION_NMATCH, matches, 0) != 0) { + return pcmk_rc_ok; // This is not an attribute deletion + } + + /* matches[0] = entire node state match + * matches[1] = node state ID + * + * Optional: + * matches[2] = transient attributes element with ID + * matches[3] = transient attributes ID + * matches[4] = attribute set element with ID + * matches[5] = attribute set element name + * matches[6] = attribute set ID + * matches[7] = name/value pair element with ID + * matches[8] = name/value pair ID + */ + + // If we get here, we must have matched at least node state and its ID + CRM_CHECK((matches[0].rm_so == 0) && (matches[1].rm_so > 0), + return pcmk_rc_ok); + + /* Check whether all of node's attributes were deleted (if no matches[2], + * entire node state was deleted; if no matches[4], entire + * transient attributes section was deleted) + */ + if ((matches[2].rm_so < 0) || (matches[4].rm_so < 0)) { + id = re_submatch(value, matches, 1); + attrd_drop_removed_values(id); + free(id); + return pcmk_rc_ok; + } + + /* We matched transient attributes, so we must have its ID, as well as an + * attribute set, so we must have its element name and ID + */ + CRM_CHECK((matches[3].rm_so > 0) && (matches[5].rm_so > 0) + && (matches[6].rm_so > 0), return pcmk_rc_ok); + + // Check whether the entire set was deleted + if (matches[7].rm_so < 0) { + char *set_type = re_submatch(value, matches, 5); + + id = re_submatch(value, matches, 6); + attrd_drop_removed_set(set_type, id); + free(id); + free(set_type); + return pcmk_rc_ok; + } + + // We matched a single name/value pair, so we must have its ID + CRM_CHECK(matches[8].rm_so > 0, return pcmk_rc_ok); + + // Drop the one value + id = re_submatch(value, matches, 8); + attrd_drop_removed_value(id); + free(id); + return pcmk_rc_ok; +} + static void attrd_cib_updated_cb(const char *event, xmlNode *msg) { const xmlNode *patchset = NULL; const char *client_name = NULL; bool status_changed = false; if (cib__get_notify_patchset(msg, &patchset) != pcmk_rc_ok) { return; } if (pcmk__cib_element_in_patchset(patchset, PCMK_XE_ALERTS)) { - if (attrd_shutting_down(true)) { + if (attrd_shutting_down()) { crm_debug("Ignoring alerts change in CIB during shutdown"); } else { mainloop_set_trigger(attrd_config_read); } } status_changed = pcmk__cib_element_in_patchset(patchset, PCMK_XE_STATUS); client_name = crm_element_value(msg, PCMK__XA_CIB_CLIENTNAME); if (!cib__client_triggers_refresh(client_name)) { /* This change came from a source that ensured the CIB is consistent * with our attributes table, so we don't need to write anything out. + * If a removed attribute has been erased, we can forget it now. */ + int format = 1; + + if ((crm_element_value_int(patchset, PCMK_XA_FORMAT, &format) != 0) + || (format != 2)) { + crm_warn("Can't handle CIB patch format %d", format); + return; + } + + /* This won't modify patchset, but we need to break const to match the + * function signature. + */ + pcmk__xe_foreach_child((xmlNode *) patchset, PCMK_XE_CHANGE, + drop_values_in_deletion, NULL); return; } if (!attrd_election_won()) { // Don't write attributes if we're not the writer return; } if (status_changed || pcmk__cib_element_in_patchset(patchset, PCMK_XE_NODES)) { - if (attrd_shutting_down(true)) { + if (attrd_shutting_down()) { crm_debug("Ignoring node change in CIB during shutdown"); return; } /* An unsafe client modified the PCMK_XE_NODES or PCMK_XE_STATUS * section. Write transient attributes to ensure they're up-to-date in * the CIB. */ if (client_name == NULL) { client_name = crm_element_value(msg, PCMK__XA_CIB_CLIENTID); } crm_notice("Updating all attributes after %s event triggered by %s", event, pcmk__s(client_name, "unidentified client")); attrd_write_attributes(attrd_write_all); } } int attrd_cib_connect(int max_retry) { static int attempts = 0; int rc = -ENOTCONN; the_cib = cib_new(); if (the_cib == NULL) { return -ENOTCONN; } do { if (attempts > 0) { sleep(attempts); } attempts++; crm_debug("Connection attempt %d to the CIB manager", attempts); rc = the_cib->cmds->signon(the_cib, crm_system_name, cib_command); } while ((rc != pcmk_ok) && (attempts < max_retry)); if (rc != pcmk_ok) { crm_err("Connection to the CIB manager failed: %s " QB_XS " rc=%d", pcmk_strerror(rc), rc); goto cleanup; } crm_debug("Connected to the CIB manager after %d attempts", attempts); rc = the_cib->cmds->set_connection_dnotify(the_cib, attrd_cib_destroy_cb); if (rc != pcmk_ok) { crm_err("Could not set disconnection callback"); goto cleanup; } rc = the_cib->cmds->add_notify_callback(the_cib, PCMK__VALUE_CIB_DIFF_NOTIFY, attrd_cib_updated_cb); if (rc != pcmk_ok) { crm_err("Could not set CIB notification callback"); goto cleanup; } return pcmk_ok; cleanup: cib__clean_up_connection(&the_cib); return -ENOTCONN; } void attrd_cib_disconnect(void) { CRM_CHECK(the_cib != NULL, return); the_cib->cmds->del_notify_callback(the_cib, PCMK__VALUE_CIB_DIFF_NOTIFY, attrd_cib_updated_cb); cib__clean_up_connection(&the_cib); mainloop_destroy_trigger(attrd_config_read); } static void attrd_erase_cb(xmlNode *msg, int call_id, int rc, xmlNode *output, void *user_data) { const char *node = pcmk__s((const char *) user_data, "a node"); if (rc == pcmk_ok) { crm_info("Cleared transient node attributes for %s from CIB", node); } else { crm_err("Unable to clear transient node attributes for %s from CIB: %s", node, pcmk_strerror(rc)); } } #define XPATH_TRANSIENT "//" PCMK__XE_NODE_STATE \ "[@" PCMK_XA_UNAME "='%s']" \ "/" PCMK__XE_TRANSIENT_ATTRIBUTES /*! * \internal * \brief Wipe all transient node attributes for a node from the CIB * * \param[in] node Node to clear attributes for */ void attrd_cib_erase_transient_attrs(const char *node) { int call_id = 0; char *xpath = NULL; CRM_CHECK(node != NULL, return); xpath = crm_strdup_printf(XPATH_TRANSIENT, node); crm_debug("Clearing transient node attributes for %s from CIB using %s", node, xpath); call_id = the_cib->cmds->remove(the_cib, xpath, NULL, cib_xpath); free(xpath); the_cib->cmds->register_callback_full(the_cib, call_id, 120, FALSE, pcmk__str_copy(node), "attrd_erase_cb", attrd_erase_cb, free); } /*! * \internal * \brief Prepare the CIB after cluster is connected */ void attrd_cib_init(void) { /* We have no attribute values in memory, so wipe the CIB to match. This is - * normally done by the DC's controller when this node leaves the cluster, but - * this handles the case where the node restarted so quickly that the + * normally done by the writer when this node leaves the cluster, but this + * handles the case where the node restarted so quickly that the * cluster layer didn't notice. * * \todo If the attribute manager respawns after crashing (see * PCMK_ENV_RESPAWNED), ideally we'd skip this and sync our attributes * from the writer. However, currently we reject any values for us * that the writer has, in attrd_peer_update(). */ attrd_cib_erase_transient_attrs(attrd_cluster->priv->node_name); // Set a trigger for reading the CIB (for the alerts section) attrd_config_read = mainloop_add_trigger(G_PRIORITY_HIGH, attrd_read_options, NULL); // Always read the CIB at start-up mainloop_set_trigger(attrd_config_read); } static gboolean attribute_timer_cb(gpointer data) { attribute_t *a = data; crm_trace("Dampen interval expired for %s", a->id); attrd_write_or_elect_attribute(a); return FALSE; } static void attrd_cib_callback(xmlNode *msg, int call_id, int rc, xmlNode *output, void *user_data) { int level = LOG_ERR; GHashTableIter iter; const char *peer = NULL; attribute_value_t *v = NULL; char *name = user_data; attribute_t *a = g_hash_table_lookup(attributes, name); if(a == NULL) { crm_info("Attribute %s no longer exists", name); return; } a->update = 0; if (rc == pcmk_ok && call_id < 0) { rc = call_id; } switch (rc) { case pcmk_ok: level = LOG_INFO; last_cib_op_done = call_id; if (a->timer && !a->timeout_ms) { // Remove temporary dampening for failed writes mainloop_timer_del(a->timer); a->timer = NULL; } break; case -pcmk_err_diff_failed: /* When an attr changes while the CIB is syncing */ case -ETIME: /* When an attr changes while there is a DC election */ case -ENXIO: /* When an attr changes while the CIB is syncing a * newer config from a node that just came up */ level = LOG_WARNING; break; } do_crm_log(level, "CIB update %d result for %s: %s " QB_XS " rc=%d", call_id, a->id, pcmk_strerror(rc), rc); g_hash_table_iter_init(&iter, a->values); while (g_hash_table_iter_next(&iter, (gpointer *) & peer, (gpointer *) & v)) { if (rc == pcmk_ok) { crm_info("* Wrote %s[%s]=%s", a->id, peer, pcmk__s(v->requested, "(unset)")); pcmk__str_update(&(v->requested), NULL); } else { do_crm_log(level, "* Could not write %s[%s]=%s", a->id, peer, pcmk__s(v->requested, "(unset)")); /* Reattempt write below if we are still the writer */ attrd_set_attr_flags(a, attrd_attr_changed); } } if (pcmk_is_set(a->flags, attrd_attr_changed) && attrd_election_won()) { if (rc == pcmk_ok) { /* We deferred a write of a new update because this update was in * progress. Write out the new value without additional delay. */ crm_debug("Pending update for %s can be written now", a->id); write_attribute(a, false); /* We're re-attempting a write because the original failed; delay * the next attempt so we don't potentially flood the CIB manager * and logs with a zillion attempts per second. * * @TODO We could elect a new writer instead. However, we'd have to * somehow downgrade our vote, and we'd still need something like this * if all peers similarly fail to write this attribute (which may * indicate a corrupted attribute entry rather than a CIB issue). */ } else if (a->timer) { // Attribute has a dampening value, so use that as delay if (!mainloop_timer_running(a->timer)) { crm_trace("Delayed re-attempted write for %s by %s", name, pcmk__readable_interval(a->timeout_ms)); mainloop_timer_start(a->timer); } } else { /* Set a temporary dampening of 2 seconds (timer will continue * to exist until the attribute's dampening gets set or the * write succeeds). */ a->timer = attrd_add_timer(a->id, 2000, a); mainloop_timer_start(a->timer); } } } /*! * \internal * \brief Add a set-attribute update request to the current CIB transaction * * \param[in] attr Attribute to update * \param[in] attr_id ID of attribute to update * \param[in] node_id ID of node for which to update attribute value * \param[in] set_id ID of attribute set * \param[in] value New value for attribute * * \return Standard Pacemaker return code */ static int add_set_attr_update(const attribute_t *attr, const char *attr_id, const char *node_id, const char *set_id, const char *value) { xmlNode *update = pcmk__xe_create(NULL, PCMK__XE_NODE_STATE); xmlNode *child = update; int rc = ENOMEM; crm_xml_add(child, PCMK_XA_ID, node_id); child = pcmk__xe_create(child, PCMK__XE_TRANSIENT_ATTRIBUTES); crm_xml_add(child, PCMK_XA_ID, node_id); child = pcmk__xe_create(child, attr->set_type); crm_xml_add(child, PCMK_XA_ID, set_id); child = pcmk__xe_create(child, PCMK_XE_NVPAIR); crm_xml_add(child, PCMK_XA_ID, attr_id); crm_xml_add(child, PCMK_XA_NAME, attr->id); crm_xml_add(child, PCMK_XA_VALUE, value); rc = the_cib->cmds->modify(the_cib, PCMK_XE_STATUS, update, cib_can_create|cib_transaction); rc = pcmk_legacy2rc(rc); pcmk__xml_free(update); return rc; } /*! * \internal * \brief Add an unset-attribute update request to the current CIB transaction * * \param[in] attr Attribute to update * \param[in] attr_id ID of attribute to update * \param[in] node_id ID of node for which to update attribute value * \param[in] set_id ID of attribute set * * \return Standard Pacemaker return code */ static int add_unset_attr_update(const attribute_t *attr, const char *attr_id, const char *node_id, const char *set_id) { char *xpath = crm_strdup_printf("/" PCMK_XE_CIB "/" PCMK_XE_STATUS "/" PCMK__XE_NODE_STATE "[@" PCMK_XA_ID "='%s']" "/" PCMK__XE_TRANSIENT_ATTRIBUTES "[@" PCMK_XA_ID "='%s']" "/%s[@" PCMK_XA_ID "='%s']" "/" PCMK_XE_NVPAIR "[@" PCMK_XA_ID "='%s' " "and @" PCMK_XA_NAME "='%s']", node_id, node_id, attr->set_type, set_id, attr_id, attr->id); int rc = the_cib->cmds->remove(the_cib, xpath, NULL, cib_xpath|cib_transaction); free(xpath); return pcmk_legacy2rc(rc); } /*! * \internal * \brief Add an attribute update request to the current CIB transaction * * \param[in] attr Attribute to update * \param[in] value New value for attribute * \param[in] node_id ID of node for which to update attribute value * * \return Standard Pacemaker return code */ static int add_attr_update(const attribute_t *attr, const char *value, const char *node_id) { char *set_id = attrd_set_id(attr, node_id); char *nvpair_id = attrd_nvpair_id(attr, node_id); int rc = pcmk_rc_ok; if (value == NULL) { rc = add_unset_attr_update(attr, nvpair_id, node_id, set_id); } else { rc = add_set_attr_update(attr, nvpair_id, node_id, set_id, value); } free(set_id); free(nvpair_id); return rc; } static void send_alert_attributes_value(attribute_t *a, GHashTable *t) { int rc = 0; attribute_value_t *at = NULL; GHashTableIter vIter; g_hash_table_iter_init(&vIter, t); while (g_hash_table_iter_next(&vIter, NULL, (gpointer *) & at)) { const char *node_xml_id = attrd_get_node_xml_id(at->nodename); rc = attrd_send_attribute_alert(at->nodename, node_xml_id, a->id, at->current); crm_trace("Sent alerts for %s[%s]=%s with node XML ID %s " "(%s agents failed)", a->id, at->nodename, at->current, pcmk__s(node_xml_id, "unknown"), ((rc == 0)? "no" : ((rc == -1)? "some" : "all"))); } } static void set_alert_attribute_value(GHashTable *t, attribute_value_t *v) { attribute_value_t *a_v = pcmk__assert_alloc(1, sizeof(attribute_value_t)); a_v->nodename = pcmk__str_copy(v->nodename); a_v->current = pcmk__str_copy(v->current); g_hash_table_replace(t, a_v->nodename, a_v); } mainloop_timer_t * attrd_add_timer(const char *id, int timeout_ms, attribute_t *attr) { return mainloop_timer_add(id, timeout_ms, FALSE, attribute_timer_cb, attr); } /*! * \internal * \brief Write an attribute's values to the CIB if appropriate * * \param[in,out] a Attribute to write * \param[in] ignore_delay If true, write attribute now regardless of any * configured delay */ static void write_attribute(attribute_t *a, bool ignore_delay) { int private_updates = 0, cib_updates = 0; attribute_value_t *v = NULL; GHashTableIter iter; GHashTable *alert_attribute_value = NULL; int rc = pcmk_ok; - bool should_write = true; + bool should_write = attrd_for_cib(a); if (a == NULL) { return; } - // Private attributes (or any in standalone mode) are not written to the CIB - if (stand_alone || pcmk_is_set(a->flags, attrd_attr_is_private)) { - should_write = false; - } - - /* If this attribute will be written to the CIB ... */ if (should_write) { /* Defer the write if now's not a good time */ if (a->update && (a->update < last_cib_op_done)) { crm_info("Write out of '%s' continuing: update %d considered lost", a->id, a->update); a->update = 0; // Don't log this message again } else if (a->update) { crm_info("Write out of '%s' delayed: update %d in progress", a->id, a->update); goto done; } else if (mainloop_timer_running(a->timer)) { if (ignore_delay) { mainloop_timer_stop(a->timer); crm_debug("Overriding '%s' write delay", a->id); } else { crm_info("Delaying write of '%s'", a->id); goto done; } } // Initiate a transaction for all the peer value updates CRM_CHECK(the_cib != NULL, goto done); the_cib->cmds->set_user(the_cib, a->user); rc = the_cib->cmds->init_transaction(the_cib); if (rc != pcmk_ok) { crm_err("Failed to write %s (set %s): Could not initiate " "CIB transaction", a->id, pcmk__s(a->set_id, "unspecified")); goto done; } } /* The changed and force-write flags apply only to the next write, * which this is, so clear them now. Also clear the "node unknown" flag * because we will check whether it is known below and reset if appopriate. */ attrd_clear_attr_flags(a, attrd_attr_changed |attrd_attr_force_write |attrd_attr_node_unknown); /* Make the table for the attribute trap */ alert_attribute_value = pcmk__strikey_table(NULL, attrd_free_attribute_value); /* Iterate over each peer value of this attribute */ g_hash_table_iter_init(&iter, a->values); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &v)) { const char *node_xml_id = NULL; const char *prev_xml_id = NULL; + pcmk__node_status_t *peer = NULL; if (!should_write) { private_updates++; continue; } /* We need the node's CIB XML ID to write out its attributes, so look * for it now. Check the node caches first, even if the ID was * previously known (in case it changed), but use any previous value as * a fallback. */ prev_xml_id = attrd_get_node_xml_id(v->nodename); if (pcmk_is_set(v->flags, attrd_value_remote)) { // A Pacemaker Remote node's XML ID is the same as its name node_xml_id = v->nodename; + } else if (v->current == NULL) { + /* If a value was removed, check the caches for the node XML ID, + * but don't create a new cache entry. We don't want to re-create a + * purged node. + */ + peer = pcmk__search_node_caches(0, v->nodename, prev_xml_id, + pcmk__node_search_any + |pcmk__node_search_cluster_cib); + node_xml_id = pcmk__cluster_get_xml_id(peer); + if (node_xml_id == NULL) { + node_xml_id = prev_xml_id; + } + } else { // This creates a cluster node cache entry if none exists - pcmk__node_status_t *peer = pcmk__get_node(0, v->nodename, - prev_xml_id, - pcmk__node_search_any); - + peer = pcmk__get_node(0, v->nodename, prev_xml_id, + pcmk__node_search_any); node_xml_id = pcmk__cluster_get_xml_id(peer); if (node_xml_id == NULL) { node_xml_id = prev_xml_id; } } // Defer write if this is a cluster node that's never been seen if (node_xml_id == NULL) { attrd_set_attr_flags(a, attrd_attr_node_unknown); crm_notice("Cannot write %s[%s]='%s' to CIB because node's XML ID " "is unknown (will retry if learned)", a->id, v->nodename, v->current); continue; } if (!pcmk__str_eq(prev_xml_id, node_xml_id, pcmk__str_none)) { crm_trace("Setting %s[%s] node XML ID to %s (was %s)", a->id, v->nodename, node_xml_id, pcmk__s(prev_xml_id, "unknown")); attrd_set_node_xml_id(v->nodename, node_xml_id); } // Update this value as part of the CIB transaction we're building rc = add_attr_update(a, v->current, node_xml_id); if (rc != pcmk_rc_ok) { crm_err("Couldn't add %s[%s]='%s' to CIB transaction: %s " QB_XS " node XML ID %s", - a->id, v->nodename, v->current, pcmk_rc_str(rc), - node_xml_id); + a->id, v->nodename, pcmk__s(v->current, "(unset)"), + pcmk_rc_str(rc), node_xml_id); continue; } crm_debug("Added %s[%s]=%s to CIB transaction (node XML ID %s)", a->id, v->nodename, pcmk__s(v->current, "(unset)"), node_xml_id); cib_updates++; /* Preservation of the attribute to transmit alert */ set_alert_attribute_value(alert_attribute_value, v); // Save this value so we can log it when write completes pcmk__str_update(&(v->requested), v->current); } if (private_updates) { crm_info("Processed %d private change%s for %s (set %s)", private_updates, pcmk__plural_s(private_updates), a->id, pcmk__s(a->set_id, "unspecified")); } if (cib_updates > 0) { char *id = pcmk__str_copy(a->id); // Commit transaction a->update = the_cib->cmds->end_transaction(the_cib, true, cib_none); crm_info("Sent CIB request %d with %d change%s for %s (set %s)", a->update, cib_updates, pcmk__plural_s(cib_updates), a->id, pcmk__s(a->set_id, "unspecified")); if (the_cib->cmds->register_callback_full(the_cib, a->update, CIB_OP_TIMEOUT_S, FALSE, id, "attrd_cib_callback", attrd_cib_callback, free)) { // Transmit alert of the attribute + // @TODO Do this in callback only if write was successful send_alert_attributes_value(a, alert_attribute_value); } } done: // Discard transaction (if any) if (the_cib != NULL) { the_cib->cmds->end_transaction(the_cib, false, cib_none); the_cib->cmds->set_user(the_cib, NULL); } if (alert_attribute_value != NULL) { g_hash_table_destroy(alert_attribute_value); } } /*! * \internal * \brief Write out attributes * * \param[in] options Group of enum attrd_write_options */ void attrd_write_attributes(uint32_t options) { GHashTableIter iter; attribute_t *a = NULL; crm_debug("Writing out %s attributes", pcmk_is_set(options, attrd_write_all)? "all" : "changed"); g_hash_table_iter_init(&iter, attributes); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) & a)) { if (!pcmk_is_set(options, attrd_write_all) && pcmk_is_set(a->flags, attrd_attr_node_unknown)) { // Try writing this attribute again, in case peer ID was learned attrd_set_attr_flags(a, attrd_attr_changed); } else if (pcmk_is_set(a->flags, attrd_attr_force_write)) { /* If the force_write flag is set, write the attribute. */ attrd_set_attr_flags(a, attrd_attr_changed); } if (pcmk_is_set(options, attrd_write_all) || pcmk_is_set(a->flags, attrd_attr_changed)) { bool ignore_delay = pcmk_is_set(options, attrd_write_no_delay); if (pcmk_is_set(a->flags, attrd_attr_force_write)) { // Always ignore delay when forced write flag is set ignore_delay = true; } write_attribute(a, ignore_delay); } else { crm_trace("Skipping unchanged attribute %s", a->id); } } } void attrd_write_or_elect_attribute(attribute_t *a) { if (attrd_election_won()) { write_attribute(a, false); } else { attrd_start_election_if_needed(); } } diff --git a/daemons/attrd/attrd_corosync.c b/daemons/attrd/attrd_corosync.c index e681fb24d4..af4bc015de 100644 --- a/daemons/attrd/attrd_corosync.c +++ b/daemons/attrd/attrd_corosync.c @@ -1,633 +1,645 @@ /* - * Copyright 2013-2024 the Pacemaker project contributors + * Copyright 2013-2025 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #include #include #include #include "pacemaker-attrd.h" static xmlNode * attrd_confirmation(int callid) { xmlNode *node = pcmk__xe_create(NULL, __func__); crm_xml_add(node, PCMK__XA_T, PCMK__VALUE_ATTRD); crm_xml_add(node, PCMK__XA_SRC, pcmk__cluster_local_node_name()); crm_xml_add(node, PCMK_XA_TASK, PCMK__ATTRD_CMD_CONFIRM); crm_xml_add_int(node, PCMK__XA_CALL_ID, callid); return node; } static void attrd_peer_message(pcmk__node_status_t *peer, xmlNode *xml) { const char *election_op = crm_element_value(xml, PCMK__XA_CRM_TASK); if (election_op) { attrd_handle_election_op(peer, xml); return; } - if (attrd_shutting_down(false)) { + if (attrd_shutting_down()) { /* If we're shutting down, we want to continue responding to election * ops as long as we're a cluster member (because our vote may be * needed). Ignore all other messages. */ return; } else { pcmk__request_t request = { .ipc_client = NULL, .ipc_id = 0, .ipc_flags = 0, .peer = peer->name, .xml = xml, .call_options = 0, .result = PCMK__UNKNOWN_RESULT, }; request.op = crm_element_value_copy(request.xml, PCMK_XA_TASK); CRM_CHECK(request.op != NULL, return); attrd_handle_request(&request); /* Having finished handling the request, check to see if the originating * peer requested confirmation. If so, send that confirmation back now. */ if (pcmk__xe_attr_is_true(xml, PCMK__XA_CONFIRM) && !pcmk__str_eq(request.op, PCMK__ATTRD_CMD_CONFIRM, pcmk__str_none)) { int callid = 0; xmlNode *reply = NULL; /* Add the confirmation ID for the message we are confirming to the * response so the originating peer knows what they're a confirmation * for. */ crm_element_value_int(xml, PCMK__XA_CALL_ID, &callid); reply = attrd_confirmation(callid); /* And then send the confirmation back to the originating peer. This * ends up right back in this same function (attrd_peer_message) on the * peer where it will have to do something with a PCMK__XA_CONFIRM type * message. */ crm_debug("Sending %s a confirmation", peer->name); attrd_send_message(peer, reply, false); pcmk__xml_free(reply); } pcmk__reset_request(&request); } } static void attrd_cpg_dispatch(cpg_handle_t handle, const struct cpg_name *groupName, uint32_t nodeid, uint32_t pid, void *msg, size_t msg_len) { xmlNode *xml = NULL; const char *from = NULL; char *data = pcmk__cpg_message_data(handle, nodeid, pid, msg, &from); if(data == NULL) { return; } xml = pcmk__xml_parse(data); if (xml == NULL) { crm_err("Bad message received from %s[%" PRIu32 "]: '%.120s'", from, nodeid, data); } else { attrd_peer_message(pcmk__get_node(nodeid, from, NULL, pcmk__node_search_cluster_member), xml); } pcmk__xml_free(xml); free(data); } static void attrd_cpg_destroy(gpointer unused) { - if (attrd_shutting_down(false)) { + if (attrd_shutting_down()) { crm_info("Disconnected from Corosync process group"); } else { crm_crit("Lost connection to Corosync process group, shutting down"); attrd_exit_status = CRM_EX_DISCONNECT; attrd_shutdown(0); } } /*! * \internal * \brief Broadcast an update for a single attribute value * * \param[in] a Attribute to broadcast * \param[in] v Attribute value to broadcast */ void attrd_broadcast_value(const attribute_t *a, const attribute_value_t *v) { xmlNode *op = pcmk__xe_create(NULL, PCMK_XE_OP); crm_xml_add(op, PCMK_XA_TASK, PCMK__ATTRD_CMD_UPDATE); attrd_add_value_xml(op, a, v, false); attrd_send_message(NULL, op, false); pcmk__xml_free(op); } #define state_text(state) pcmk__s((state), "in unknown state") static void attrd_peer_change_cb(enum pcmk__node_update kind, pcmk__node_status_t *peer, const void *data) { bool gone = false; bool is_remote = pcmk_is_set(peer->flags, pcmk__node_status_remote); switch (kind) { case pcmk__node_update_name: crm_debug("%s node %s[%" PRIu32 "] is now %s", (is_remote? "Remote" : "Cluster"), pcmk__s(peer->name, "unknown"), peer->cluster_layer_id, state_text(peer->state)); break; case pcmk__node_update_processes: if (!pcmk_is_set(peer->processes, crm_get_cluster_proc())) { gone = true; } crm_debug("Node %s[%" PRIu32 "] is %s a peer", pcmk__s(peer->name, "unknown"), peer->cluster_layer_id, (gone? "no longer" : "now")); break; case pcmk__node_update_state: crm_debug("%s node %s[%" PRIu32 "] is now %s (was %s)", (is_remote? "Remote" : "Cluster"), pcmk__s(peer->name, "unknown"), peer->cluster_layer_id, state_text(peer->state), state_text(data)); if (pcmk__str_eq(peer->state, PCMK_VALUE_MEMBER, pcmk__str_none)) { /* If we're the writer, send new peers a list of all attributes * (unless it's a remote node, which doesn't run its own attrd) */ if (!is_remote) { if (attrd_election_won()) { attrd_peer_sync(peer); } else { // Anyway send a message so that the peer learns our name attrd_send_protocol(peer); } } } else { // Remove all attribute values associated with lost nodes if (peer->name != NULL) { attrd_peer_remove(peer->name, false, "loss"); } gone = true; } break; } // Remove votes from cluster nodes that leave, in case election in progress if (gone && !is_remote && peer->name != NULL) { attrd_remove_voter(peer); attrd_remove_peer_protocol_ver(peer->name); attrd_do_not_expect_from_peer(peer->name); } } #define readable_value(rv_v) pcmk__s((rv_v)->current, "(unset)") #define readable_peer(p) \ (((p) == NULL)? "all peers" : pcmk__s((p)->name, "unknown peer")) static void update_attr_on_host(attribute_t *a, const pcmk__node_status_t *peer, const xmlNode *xml, const char *attr, const char *value, const char *host, bool filter) { int is_remote = 0; bool changed = false; attribute_value_t *v = NULL; const char *prev_xml_id = NULL; const char *node_xml_id = crm_element_value(xml, PCMK__XA_ATTR_HOST_ID); // Create entry for value if not already existing v = g_hash_table_lookup(a->values, host); if (v == NULL) { v = pcmk__assert_alloc(1, sizeof(attribute_value_t)); v->nodename = pcmk__str_copy(host); g_hash_table_replace(a->values, v->nodename, v); } /* If update doesn't contain the node XML ID, fall back to any previously * known value (for logging) */ prev_xml_id = attrd_get_node_xml_id(v->nodename); if (node_xml_id == NULL) { node_xml_id = prev_xml_id; } // If value is for a Pacemaker Remote node, remember that crm_element_value_int(xml, PCMK__XA_ATTR_IS_REMOTE, &is_remote); if (is_remote) { attrd_set_value_flags(v, attrd_value_remote); pcmk__assert(pcmk__cluster_lookup_remote_node(host) != NULL); } // Check whether the value changed changed = !pcmk__str_eq(v->current, value, pcmk__str_casei); if (changed && filter && pcmk__str_eq(host, attrd_cluster->priv->node_name, pcmk__str_casei)) { /* Broadcast the local value for an attribute that differs from the * value provided in a peer's attribute synchronization response. This * ensures a node's values for itself take precedence and all peers are * kept in sync. */ v = g_hash_table_lookup(a->values, attrd_cluster->priv->node_name); crm_notice("%s[%s]: local value '%s' takes priority over '%s' from %s", attr, host, readable_value(v), value, peer->name); attrd_broadcast_value(a, v); } else if (changed) { crm_notice("Setting %s[%s]%s%s: %s -> %s " QB_XS " from %s with %s write delay and node XML ID %s", attr, host, a->set_type ? " in " : "", pcmk__s(a->set_type, ""), readable_value(v), pcmk__s(value, "(unset)"), peer->name, (a->timeout_ms == 0)? "no" : pcmk__readable_interval(a->timeout_ms), pcmk__s(node_xml_id, "unknown")); pcmk__str_update(&v->current, value); attrd_set_attr_flags(a, attrd_attr_changed); - if (pcmk__str_eq(host, attrd_cluster->priv->node_name, pcmk__str_casei) - && pcmk__str_eq(attr, PCMK__NODE_ATTR_SHUTDOWN, pcmk__str_none)) { - - if (!pcmk__str_eq(value, "0", pcmk__str_null_matches)) { - attrd_set_requesting_shutdown(); - - } else { - attrd_clear_requesting_shutdown(); - } - } - // Write out new value or start dampening timer if (a->timeout_ms && a->timer) { crm_trace("Delaying write of %s %s for dampening", attr, pcmk__readable_interval(a->timeout_ms)); mainloop_timer_start(a->timer); } else { attrd_write_or_elect_attribute(a); } } else { int is_force_write = 0; crm_element_value_int(xml, PCMK__XA_ATTRD_IS_FORCE_WRITE, &is_force_write); if (is_force_write == 1 && a->timeout_ms && a->timer) { /* Save forced writing and set change flag. */ /* The actual attribute is written by Writer after election. */ crm_trace("%s[%s] from %s is unchanged (%s), forcing write", attr, host, peer->name, pcmk__s(value, "unset")); attrd_set_attr_flags(a, attrd_attr_force_write); } else { crm_trace("%s[%s] from %s is unchanged (%s)", attr, host, peer->name, pcmk__s(value, "unset")); } } // This allows us to later detect local values that peer doesn't know about attrd_set_value_flags(v, attrd_value_from_peer); // Remember node's XML ID if we're just learning it if ((node_xml_id != NULL) && !pcmk__str_eq(node_xml_id, prev_xml_id, pcmk__str_none)) { // Remember node's name in case unknown in the membership cache pcmk__node_status_t *known_peer = pcmk__get_node(0, host, node_xml_id, pcmk__node_search_cluster_member); crm_trace("Learned %s[%s] node XML ID is %s (was %s)", a->id, known_peer->name, node_xml_id, pcmk__s(prev_xml_id, "unknown")); attrd_set_node_xml_id(v->nodename, node_xml_id); if (attrd_election_won()) { // In case we couldn't write a value missing the XML ID before attrd_write_attributes(attrd_write_changed); } } } static void attrd_peer_update_one(const pcmk__node_status_t *peer, xmlNode *xml, bool filter) { attribute_t *a = NULL; const char *attr = crm_element_value(xml, PCMK__XA_ATTR_NAME); const char *value = crm_element_value(xml, PCMK__XA_ATTR_VALUE); const char *host = crm_element_value(xml, PCMK__XA_ATTR_HOST); if (attr == NULL) { crm_warn("Could not update attribute: peer did not specify name"); return; } a = attrd_populate_attribute(xml, attr); if (a == NULL) { return; } if (host == NULL) { // If no host was specified, update all hosts GHashTableIter vIter; crm_debug("Setting %s for all hosts to %s", attr, value); pcmk__xe_remove_attr(xml, PCMK__XA_ATTR_HOST_ID); g_hash_table_iter_init(&vIter, a->values); while (g_hash_table_iter_next(&vIter, (gpointer *) & host, NULL)) { update_attr_on_host(a, peer, xml, attr, value, host, filter); } } else { // Update attribute value for the given host update_attr_on_host(a, peer, xml, attr, value, host, filter); } /* If this is a message from some attrd instance broadcasting its protocol * version, check to see if it's a new minimum version. */ if (pcmk__str_eq(attr, CRM_ATTR_PROTOCOL, pcmk__str_none)) { attrd_update_minimum_protocol_ver(peer->name, value); } } static void broadcast_unseen_local_values(void) { GHashTableIter aIter; GHashTableIter vIter; attribute_t *a = NULL; attribute_value_t *v = NULL; xmlNode *sync = NULL; g_hash_table_iter_init(&aIter, attributes); while (g_hash_table_iter_next(&aIter, NULL, (gpointer *) & a)) { g_hash_table_iter_init(&vIter, a->values); while (g_hash_table_iter_next(&vIter, NULL, (gpointer *) & v)) { if (!pcmk_is_set(v->flags, attrd_value_from_peer) && pcmk__str_eq(v->nodename, attrd_cluster->priv->node_name, pcmk__str_casei)) { crm_trace("* %s[%s]='%s' is local-only", a->id, v->nodename, readable_value(v)); if (sync == NULL) { sync = pcmk__xe_create(NULL, __func__); crm_xml_add(sync, PCMK_XA_TASK, PCMK__ATTRD_CMD_SYNC_RESPONSE); } attrd_add_value_xml(sync, a, v, a->timeout_ms && a->timer); } } } if (sync != NULL) { crm_debug("Broadcasting local-only values"); attrd_send_message(NULL, sync, false); pcmk__xml_free(sync); } } int attrd_cluster_connect(void) { int rc = pcmk_rc_ok; attrd_cluster = pcmk_cluster_new(); pcmk_cluster_set_destroy_fn(attrd_cluster, attrd_cpg_destroy); pcmk_cpg_set_deliver_fn(attrd_cluster, attrd_cpg_dispatch); pcmk_cpg_set_confchg_fn(attrd_cluster, pcmk__cpg_confchg_cb); pcmk__cluster_set_status_callback(&attrd_peer_change_cb); rc = pcmk_cluster_connect(attrd_cluster); rc = pcmk_rc2legacy(rc); if (rc != pcmk_ok) { crm_err("Cluster connection failed"); return rc; } return pcmk_ok; } void attrd_peer_clear_failure(pcmk__request_t *request) { xmlNode *xml = request->xml; const char *rsc = crm_element_value(xml, PCMK__XA_ATTR_RESOURCE); const char *host = crm_element_value(xml, PCMK__XA_ATTR_HOST); const char *op = crm_element_value(xml, PCMK__XA_ATTR_CLEAR_OPERATION); const char *interval_spec = crm_element_value(xml, PCMK__XA_ATTR_CLEAR_INTERVAL); guint interval_ms = 0U; char *attr = NULL; GHashTableIter iter; regex_t regex; pcmk__node_status_t *peer = pcmk__get_node(0, request->peer, NULL, pcmk__node_search_cluster_member); pcmk_parse_interval_spec(interval_spec, &interval_ms); if (attrd_failure_regex(®ex, rsc, op, interval_ms) != pcmk_ok) { crm_info("Ignoring invalid request to clear failures for %s", pcmk__s(rsc, "all resources")); return; } crm_xml_add(xml, PCMK_XA_TASK, PCMK__ATTRD_CMD_UPDATE); /* Make sure value is not set, so we delete */ pcmk__xe_remove_attr(xml, PCMK__XA_ATTR_VALUE); g_hash_table_iter_init(&iter, attributes); while (g_hash_table_iter_next(&iter, (gpointer *) &attr, NULL)) { if (regexec(®ex, attr, 0, NULL, 0) == 0) { crm_trace("Matched %s when clearing %s", attr, pcmk__s(rsc, "all resources")); crm_xml_add(xml, PCMK__XA_ATTR_NAME, attr); attrd_peer_update(peer, xml, host, false); } } regfree(®ex); } /*! * \internal * \brief Load attributes from a peer sync response * * \param[in] peer Peer that sent sync response * \param[in] peer_won Whether peer is the attribute writer * \param[in,out] xml Request XML */ void attrd_peer_sync_response(const pcmk__node_status_t *peer, bool peer_won, xmlNode *xml) { crm_info("Processing " PCMK__ATTRD_CMD_SYNC_RESPONSE " from %s", peer->name); if (peer_won) { /* Initialize the "seen" flag for all attributes to cleared, so we can * detect attributes that local node has but the writer doesn't. */ attrd_clear_value_seen(); } // Process each attribute update in the sync response for (xmlNode *child = pcmk__xe_first_child(xml, NULL, NULL, NULL); child != NULL; child = pcmk__xe_next(child, NULL)) { attrd_peer_update(peer, child, crm_element_value(child, PCMK__XA_ATTR_HOST), true); } if (peer_won) { /* If any attributes are still not marked as seen, the writer doesn't * know about them, so send all peers an update with them. */ broadcast_unseen_local_values(); } } /*! * \internal * \brief Remove all attributes and optionally peer cache entries for a node * * \param[in] host Name of node to purge * \param[in] uncache If true, remove node from peer caches * \param[in] source Who requested removal (only used for logging) */ void attrd_peer_remove(const char *host, bool uncache, const char *source) { attribute_t *a = NULL; GHashTableIter aIter; CRM_CHECK(host != NULL, return); crm_notice("Removing all %s attributes for node %s " QB_XS " %s reaping node from cache", host, source, (uncache? "and" : "without")); g_hash_table_iter_init(&aIter, attributes); - while (g_hash_table_iter_next(&aIter, NULL, (gpointer *) & a)) { - if(g_hash_table_remove(a->values, host)) { - crm_debug("Removed %s[%s] for peer %s", a->id, host, source); + while (g_hash_table_iter_next(&aIter, NULL, (gpointer *) &a)) { + /* If the attribute won't be written to the CIB, we can drop the value + * now. Otherwise we need to set it NULL and wait for a notification + * that it was erased, because if there's no writer or the current + * writer fails to write it then leaves, we may become the writer and + * need to do it. + */ + if (attrd_for_cib(a)) { + attribute_value_t *v = g_hash_table_lookup(a->values, host); + + if ((v != NULL) && (v->current != NULL)) { + crm_debug("Removed %s[%s] (by setting NULL) for %s", + a->id, host, source); + pcmk__str_update(&(v->current), NULL); + attrd_set_attr_flags(a, attrd_attr_changed); + } + } else if (g_hash_table_remove(a->values, host)) { + crm_debug("Removed %s[%s] immediately for %s", + a->id, host, source); } } + if (attrd_election_won()) { + attrd_cib_erase_transient_attrs(host); // Wipe from CIB + } else { + attrd_start_election_if_needed(); // Make sure CIB gets updated + } + + // Remove node from caches if requested if (uncache) { pcmk__purge_node_from_cache(host, 0); attrd_forget_node_xml_id(host); } } /*! * \internal * \brief Send all known attributes and values to a peer * * \param[in] peer Peer to send sync to (if NULL, broadcast to all peers) */ void attrd_peer_sync(pcmk__node_status_t *peer) { GHashTableIter aIter; GHashTableIter vIter; attribute_t *a = NULL; attribute_value_t *v = NULL; xmlNode *sync = pcmk__xe_create(NULL, __func__); crm_xml_add(sync, PCMK_XA_TASK, PCMK__ATTRD_CMD_SYNC_RESPONSE); g_hash_table_iter_init(&aIter, attributes); while (g_hash_table_iter_next(&aIter, NULL, (gpointer *) & a)) { g_hash_table_iter_init(&vIter, a->values); while (g_hash_table_iter_next(&vIter, NULL, (gpointer *) & v)) { crm_debug("Syncing %s[%s]='%s' to %s", a->id, v->nodename, readable_value(v), readable_peer(peer)); attrd_add_value_xml(sync, a, v, false); } } crm_debug("Syncing values to %s", readable_peer(peer)); attrd_send_message(peer, sync, false); pcmk__xml_free(sync); } void attrd_peer_update(const pcmk__node_status_t *peer, xmlNode *xml, const char *host, bool filter) { bool handle_sync_point = false; CRM_CHECK((peer != NULL) && (xml != NULL), return); if (xml->children != NULL) { for (xmlNode *child = pcmk__xe_first_child(xml, PCMK_XE_OP, NULL, NULL); child != NULL; child = pcmk__xe_next(child, PCMK_XE_OP)) { pcmk__xe_copy_attrs(child, xml, pcmk__xaf_no_overwrite); attrd_peer_update_one(peer, child, filter); if (attrd_request_has_sync_point(child)) { handle_sync_point = true; } } } else { attrd_peer_update_one(peer, xml, filter); if (attrd_request_has_sync_point(xml)) { handle_sync_point = true; } } /* If the update XML specified that the client wanted to wait for a sync * point, process that now. */ if (handle_sync_point) { crm_trace("Hit local sync point for attribute update"); attrd_ack_waitlist_clients(attrd_sync_point_local, xml); } } diff --git a/daemons/attrd/attrd_elections.c b/daemons/attrd/attrd_elections.c index 281ec12c2f..1a79a93df2 100644 --- a/daemons/attrd/attrd_elections.c +++ b/daemons/attrd/attrd_elections.c @@ -1,176 +1,176 @@ /* - * Copyright 2013-2024 the Pacemaker project contributors + * Copyright 2013-2025 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include "pacemaker-attrd.h" static char *peer_writer = NULL; static void attrd_election_cb(pcmk_cluster_t *cluster) { attrd_declare_winner(); /* Update the peers after an election */ attrd_peer_sync(NULL); /* After winning an election, update the CIB with the values of all * attributes as the winner knows them. */ attrd_write_attributes(attrd_write_all); } void attrd_election_init(void) { election_init(attrd_cluster, attrd_election_cb); } void attrd_start_election_if_needed(void) { if ((peer_writer == NULL) && (election_state(attrd_cluster) != election_in_progress) - && !attrd_shutting_down(false)) { + && !attrd_shutting_down()) { crm_info("Starting an election to determine the writer"); election_vote(attrd_cluster); } } bool attrd_election_won(void) { return (election_state(attrd_cluster) == election_won); } void attrd_handle_election_op(const pcmk__node_status_t *peer, xmlNode *xml) { enum election_result rc = 0; enum election_result previous = election_state(attrd_cluster); crm_xml_add(xml, PCMK__XA_SRC, peer->name); // Don't become writer if we're shutting down - rc = election_count_vote(attrd_cluster, xml, !attrd_shutting_down(false)); + rc = election_count_vote(attrd_cluster, xml, !attrd_shutting_down()); switch(rc) { case election_start: crm_debug("Unsetting writer (was %s) and starting new election", peer_writer? peer_writer : "unset"); free(peer_writer); peer_writer = NULL; election_vote(attrd_cluster); break; case election_lost: /* The election API should really distinguish between "we just lost * to this peer" and "we already lost previously, and we are * discarding this vote for some reason", but it doesn't. * * In the first case, we want to tentatively set the peer writer to * this peer, even though another peer may eventually win (which we * will learn via attrd_check_for_new_writer()), so * attrd_start_election_if_needed() doesn't start a new election. * * Approximate a test for that case as best as possible. */ if ((peer_writer == NULL) || (previous != election_lost)) { pcmk__str_update(&peer_writer, peer->name); crm_debug("Election lost, presuming %s is writer for now", peer_writer); } break; case election_in_progress: election_check(attrd_cluster); break; default: crm_info("Ignoring election op from %s due to error", peer->name); break; } } bool attrd_check_for_new_writer(const pcmk__node_status_t *peer, const xmlNode *xml) { int peer_state = 0; crm_element_value_int(xml, PCMK__XA_ATTR_WRITER, &peer_state); if (peer_state == election_won) { if ((election_state(attrd_cluster) == election_won) && !pcmk__str_eq(peer->name, attrd_cluster->priv->node_name, pcmk__str_casei)) { crm_notice("Detected another attribute writer (%s), starting new " "election", peer->name); election_vote(attrd_cluster); } else if (!pcmk__str_eq(peer->name, peer_writer, pcmk__str_casei)) { crm_notice("Recorded new attribute writer: %s (was %s)", peer->name, pcmk__s(peer_writer, "unset")); pcmk__str_update(&peer_writer, peer->name); } } return (peer_state == election_won); } void attrd_declare_winner(void) { crm_notice("Recorded local node as attribute writer (was %s)", (peer_writer? peer_writer : "unset")); pcmk__str_update(&peer_writer, attrd_cluster->priv->node_name); } void attrd_remove_voter(const pcmk__node_status_t *peer) { election_remove(attrd_cluster, peer->name); if ((peer_writer != NULL) && pcmk__str_eq(peer->name, peer_writer, pcmk__str_casei)) { free(peer_writer); peer_writer = NULL; crm_notice("Lost attribute writer %s", peer->name); /* Clear any election dampening in effect. Otherwise, if the lost writer * had just won, the election could fizzle out with no new writer. */ election_clear_dampening(attrd_cluster); /* If the writer received attribute updates during its shutdown, it will * not have written them to the CIB. Ensure we get a new writer so they * are written out. This means that every node that sees the writer * leave will start a new election, but that's better than losing * attributes. */ attrd_start_election_if_needed(); /* If an election is in progress, we need to call election_check(), in case * this lost peer is the only one that hasn't voted, otherwise the election * would be pending until it's timed out. */ } else if (election_state(attrd_cluster) == election_in_progress) { crm_debug("Checking election status upon loss of voter %s", peer->name); election_check(attrd_cluster); } } void attrd_xml_add_writer(xmlNode *xml) { crm_xml_add_int(xml, PCMK__XA_ATTR_WRITER, election_state(attrd_cluster)); } diff --git a/daemons/attrd/attrd_ipc.c b/daemons/attrd/attrd_ipc.c index 07fe84399a..4e83f7623a 100644 --- a/daemons/attrd/attrd_ipc.c +++ b/daemons/attrd/attrd_ipc.c @@ -1,630 +1,630 @@ /* * Copyright 2004-2025 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include // PRIu32 #include #include #include #include #include #include #include #include #include #include #include "pacemaker-attrd.h" static qb_ipcs_service_t *ipcs = NULL; /*! * \internal * \brief Build the XML reply to a client query * * \param[in] attr Name of requested attribute * \param[in] host Name of requested host (or NULL for all hosts) * * \return New XML reply * \note Caller is responsible for freeing the resulting XML */ static xmlNode *build_query_reply(const char *attr, const char *host) { xmlNode *reply = pcmk__xe_create(NULL, __func__); attribute_t *a; crm_xml_add(reply, PCMK__XA_T, PCMK__VALUE_ATTRD); crm_xml_add(reply, PCMK__XA_SUBT, PCMK__ATTRD_CMD_QUERY); crm_xml_add(reply, PCMK__XA_ATTR_VERSION, ATTRD_PROTOCOL_VERSION); /* If desired attribute exists, add its value(s) to the reply */ a = g_hash_table_lookup(attributes, attr); if (a) { attribute_value_t *v; xmlNode *host_value; crm_xml_add(reply, PCMK__XA_ATTR_NAME, attr); /* Allow caller to use "localhost" to refer to local node */ if (pcmk__str_eq(host, "localhost", pcmk__str_casei)) { host = attrd_cluster->priv->node_name; crm_trace("Mapped localhost to %s", host); } /* If a specific node was requested, add its value */ if (host) { v = g_hash_table_lookup(a->values, host); host_value = pcmk__xe_create(reply, PCMK_XE_NODE); crm_xml_add(host_value, PCMK__XA_ATTR_HOST, host); crm_xml_add(host_value, PCMK__XA_ATTR_VALUE, (v? v->current : NULL)); /* Otherwise, add all nodes' values */ } else { GHashTableIter iter; g_hash_table_iter_init(&iter, a->values); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &v)) { host_value = pcmk__xe_create(reply, PCMK_XE_NODE); crm_xml_add(host_value, PCMK__XA_ATTR_HOST, v->nodename); crm_xml_add(host_value, PCMK__XA_ATTR_VALUE, v->current); } } } return reply; } xmlNode * attrd_client_clear_failure(pcmk__request_t *request) { xmlNode *xml = request->xml; const char *rsc, *op, *interval_spec; if (minimum_protocol_version >= 2) { /* Propagate to all peers (including ourselves). * This ends up at attrd_peer_message(). */ attrd_send_message(NULL, xml, false); pcmk__set_result(&request->result, CRM_EX_OK, PCMK_EXEC_DONE, NULL); return NULL; } rsc = crm_element_value(xml, PCMK__XA_ATTR_RESOURCE); op = crm_element_value(xml, PCMK__XA_ATTR_CLEAR_OPERATION); interval_spec = crm_element_value(xml, PCMK__XA_ATTR_CLEAR_INTERVAL); /* Map this to an update */ crm_xml_add(xml, PCMK_XA_TASK, PCMK__ATTRD_CMD_UPDATE); /* Add regular expression matching desired attributes */ if (rsc) { char *pattern; if (op == NULL) { pattern = crm_strdup_printf(ATTRD_RE_CLEAR_ONE, rsc); } else { guint interval_ms = 0U; pcmk_parse_interval_spec(interval_spec, &interval_ms); pattern = crm_strdup_printf(ATTRD_RE_CLEAR_OP, rsc, op, interval_ms); } crm_xml_add(xml, PCMK__XA_ATTR_REGEX, pattern); free(pattern); } else { crm_xml_add(xml, PCMK__XA_ATTR_REGEX, ATTRD_RE_CLEAR_ALL); } /* Make sure attribute and value are not set, so we delete via regex */ pcmk__xe_remove_attr(xml, PCMK__XA_ATTR_NAME); pcmk__xe_remove_attr(xml, PCMK__XA_ATTR_VALUE); return attrd_client_update(request); } xmlNode * attrd_client_peer_remove(pcmk__request_t *request) { xmlNode *xml = request->xml; // Host and ID are not used in combination, rather host has precedence const char *host = crm_element_value(xml, PCMK__XA_ATTR_HOST); char *host_alloc = NULL; attrd_send_ack(request->ipc_client, request->ipc_id, request->ipc_flags); if (host == NULL) { int nodeid = 0; crm_element_value_int(xml, PCMK__XA_ATTR_HOST_ID, &nodeid); if (nodeid > 0) { pcmk__node_status_t *node = NULL; char *host_alloc = NULL; node = pcmk__search_node_caches(nodeid, NULL, NULL, pcmk__node_search_cluster_member); if ((node != NULL) && (node->name != NULL)) { // Use cached name if available host = node->name; } else { // Otherwise ask cluster layer host_alloc = pcmk__cluster_node_name(nodeid); host = host_alloc; } crm_xml_add(xml, PCMK__XA_ATTR_HOST, host); } } if (host) { crm_info("Client %s is requesting all values for %s be removed", pcmk__client_name(request->ipc_client), host); attrd_send_message(NULL, xml, false); /* ends up at attrd_peer_message() */ free(host_alloc); } else { crm_info("Ignoring request by client %s to remove all peer values without specifying peer", pcmk__client_name(request->ipc_client)); } pcmk__set_result(&request->result, CRM_EX_OK, PCMK_EXEC_DONE, NULL); return NULL; } xmlNode * attrd_client_query(pcmk__request_t *request) { xmlNode *query = request->xml; xmlNode *reply = NULL; const char *attr = NULL; crm_debug("Query arrived from %s", pcmk__client_name(request->ipc_client)); /* Request must specify attribute name to query */ attr = crm_element_value(query, PCMK__XA_ATTR_NAME); if (attr == NULL) { pcmk__format_result(&request->result, CRM_EX_ERROR, PCMK_EXEC_ERROR, "Ignoring malformed query from %s (no attribute name given)", pcmk__client_name(request->ipc_client)); return NULL; } /* Build the XML reply */ reply = build_query_reply(attr, crm_element_value(query, PCMK__XA_ATTR_HOST)); if (reply == NULL) { pcmk__format_result(&request->result, CRM_EX_ERROR, PCMK_EXEC_ERROR, "Could not respond to query from %s: could not create XML reply", pcmk__client_name(request->ipc_client)); return NULL; } else { pcmk__set_result(&request->result, CRM_EX_OK, PCMK_EXEC_DONE, NULL); } request->ipc_client->request_id = 0; return reply; } xmlNode * attrd_client_refresh(pcmk__request_t *request) { crm_info("Updating all attributes"); attrd_send_ack(request->ipc_client, request->ipc_id, request->ipc_flags); attrd_write_attributes(attrd_write_all|attrd_write_no_delay); pcmk__set_result(&request->result, CRM_EX_OK, PCMK_EXEC_DONE, NULL); return NULL; } static void handle_missing_host(xmlNode *xml) { if (crm_element_value(xml, PCMK__XA_ATTR_HOST) == NULL) { crm_trace("Inferring local node %s with XML ID %s", attrd_cluster->priv->node_name, attrd_cluster->priv->node_xml_id); crm_xml_add(xml, PCMK__XA_ATTR_HOST, attrd_cluster->priv->node_name); crm_xml_add(xml, PCMK__XA_ATTR_HOST_ID, attrd_cluster->priv->node_xml_id); } } /* Convert a single IPC message with a regex into one with multiple children, one * for each regex match. */ static int expand_regexes(xmlNode *xml, const char *attr, const char *value, const char *regex) { if (attr == NULL && regex) { bool matched = false; GHashTableIter aIter; regex_t r_patt; crm_debug("Setting %s to %s", regex, value); if (regcomp(&r_patt, regex, REG_EXTENDED|REG_NOSUB)) { return EINVAL; } g_hash_table_iter_init(&aIter, attributes); while (g_hash_table_iter_next(&aIter, (gpointer *) & attr, NULL)) { int status = regexec(&r_patt, attr, 0, NULL, 0); if (status == 0) { xmlNode *child = pcmk__xe_create(xml, PCMK_XE_OP); crm_trace("Matched %s with %s", attr, regex); matched = true; /* Copy all the non-conflicting attributes from the parent over, * but remove the regex and replace it with the name. */ pcmk__xe_copy_attrs(child, xml, pcmk__xaf_no_overwrite); pcmk__xe_remove_attr(child, PCMK__XA_ATTR_REGEX); crm_xml_add(child, PCMK__XA_ATTR_NAME, attr); } } regfree(&r_patt); /* Return a code if we never matched anything. This should not be treated * as an error. It indicates there was a regex, and it was a valid regex, * but simply did not match anything and the caller should not continue * doing any regex-related processing. */ if (!matched) { return pcmk_rc_op_unsatisfied; } } else if (attr == NULL) { return pcmk_rc_bad_nvpair; } return pcmk_rc_ok; } static int handle_regexes(pcmk__request_t *request) { xmlNode *xml = request->xml; int rc = pcmk_rc_ok; const char *attr = crm_element_value(xml, PCMK__XA_ATTR_NAME); const char *value = crm_element_value(xml, PCMK__XA_ATTR_VALUE); const char *regex = crm_element_value(xml, PCMK__XA_ATTR_REGEX); rc = expand_regexes(xml, attr, value, regex); if (rc == EINVAL) { pcmk__format_result(&request->result, CRM_EX_ERROR, PCMK_EXEC_ERROR, "Bad regex '%s' for update from client %s", regex, pcmk__client_name(request->ipc_client)); } else if (rc == pcmk_rc_bad_nvpair) { crm_err("Update request did not specify attribute or regular expression"); pcmk__format_result(&request->result, CRM_EX_ERROR, PCMK_EXEC_ERROR, "Client %s update request did not specify attribute or regular expression", pcmk__client_name(request->ipc_client)); } return rc; } static int handle_value_expansion(const char **value, xmlNode *xml, const char *op, const char *attr) { attribute_t *a = g_hash_table_lookup(attributes, attr); if (a == NULL && pcmk__str_eq(op, PCMK__ATTRD_CMD_UPDATE_DELAY, pcmk__str_none)) { return EINVAL; } if (*value && attrd_value_needs_expansion(*value)) { int int_value; attribute_value_t *v = NULL; if (a) { const char *host = crm_element_value(xml, PCMK__XA_ATTR_HOST); v = g_hash_table_lookup(a->values, host); } int_value = attrd_expand_value(*value, (v? v->current : NULL)); crm_info("Expanded %s=%s to %d", attr, *value, int_value); crm_xml_add_int(xml, PCMK__XA_ATTR_VALUE, int_value); /* Replacing the value frees the previous memory, so re-query it */ *value = crm_element_value(xml, PCMK__XA_ATTR_VALUE); } return pcmk_rc_ok; } static void send_update_msg_to_cluster(pcmk__request_t *request, xmlNode *xml) { if (pcmk__str_eq(attrd_request_sync_point(xml), PCMK__VALUE_CLUSTER, pcmk__str_none)) { /* The client is waiting on the cluster-wide sync point. In this case, * the response ACK is not sent until this attrd broadcasts the update * and receives its own confirmation back from all peers. */ attrd_expect_confirmations(request, attrd_cluster_sync_point_update); attrd_send_message(NULL, xml, true); /* ends up at attrd_peer_message() */ } else { /* The client is either waiting on the local sync point or was not * waiting on any sync point at all. For the local sync point, the * response ACK is sent in attrd_peer_update. For clients not * waiting on any sync point, the response ACK is sent in * handle_update_request immediately before this function was called. */ attrd_send_message(NULL, xml, false); /* ends up at attrd_peer_message() */ } } static int send_child_update(xmlNode *child, void *data) { pcmk__request_t *request = (pcmk__request_t *) data; /* Calling pcmk__set_result is handled by one of these calls to * attrd_client_update, so no need to do it again here. */ request->xml = child; attrd_client_update(request); return pcmk_rc_ok; } xmlNode * attrd_client_update(pcmk__request_t *request) { xmlNode *xml = NULL; const char *attr, *value, *regex; CRM_CHECK((request != NULL) && (request->xml != NULL), return NULL); xml = request->xml; /* If the message has children, that means it is a message from a newer * client that supports sending multiple operations at a time. There are * two ways we can handle that. */ if (xml->children != NULL) { if (ATTRD_SUPPORTS_MULTI_MESSAGE(minimum_protocol_version)) { /* First, if all peers support a certain protocol version, we can * just broadcast the big message and they'll handle it. However, * we also need to apply all the transformations in this function * to the children since they don't happen anywhere else. */ for (xmlNode *child = pcmk__xe_first_child(xml, PCMK_XE_OP, NULL, NULL); child != NULL; child = pcmk__xe_next(child, PCMK_XE_OP)) { attr = crm_element_value(child, PCMK__XA_ATTR_NAME); value = crm_element_value(child, PCMK__XA_ATTR_VALUE); handle_missing_host(child); if (handle_value_expansion(&value, child, request->op, attr) == EINVAL) { pcmk__format_result(&request->result, CRM_EX_NOSUCH, PCMK_EXEC_ERROR, "Attribute %s does not exist", attr); return NULL; } } send_update_msg_to_cluster(request, xml); pcmk__set_result(&request->result, CRM_EX_OK, PCMK_EXEC_DONE, NULL); } else { /* Save the original xml node pointer so it can be restored after iterating * over all the children. */ xmlNode *orig_xml = request->xml; /* Second, if they do not support that protocol version, split it * up into individual messages and call attrd_client_update on * each one. */ pcmk__xe_foreach_child(xml, PCMK_XE_OP, send_child_update, request); request->xml = orig_xml; } return NULL; } attr = crm_element_value(xml, PCMK__XA_ATTR_NAME); value = crm_element_value(xml, PCMK__XA_ATTR_VALUE); regex = crm_element_value(xml, PCMK__XA_ATTR_REGEX); if (handle_regexes(request) != pcmk_rc_ok) { /* Error handling was already dealt with in handle_regexes, so just return. */ return NULL; } else if (regex) { /* Recursively call attrd_client_update on the new message with regexes * expanded. If supported by the attribute daemon, this means that all * matches can also be handled atomically. */ return attrd_client_update(request); } handle_missing_host(xml); if (handle_value_expansion(&value, xml, request->op, attr) == EINVAL) { pcmk__format_result(&request->result, CRM_EX_NOSUCH, PCMK_EXEC_ERROR, "Attribute %s does not exist", attr); return NULL; } crm_debug("Broadcasting %s[%s]=%s%s", attr, crm_element_value(xml, PCMK__XA_ATTR_HOST), value, (attrd_election_won()? " (writer)" : "")); send_update_msg_to_cluster(request, xml); pcmk__set_result(&request->result, CRM_EX_OK, PCMK_EXEC_DONE, NULL); return NULL; } /*! * \internal * \brief Accept a new client IPC connection * * \param[in,out] c New connection * \param[in] uid Client user id * \param[in] gid Client group id * * \return pcmk_ok on success, -errno otherwise */ static int32_t attrd_ipc_accept(qb_ipcs_connection_t *c, uid_t uid, gid_t gid) { crm_trace("New client connection %p", c); - if (attrd_shutting_down(false)) { + if (attrd_shutting_down()) { crm_info("Ignoring new connection from pid %d during shutdown", pcmk__client_pid(c)); return -ECONNREFUSED; } if (pcmk__new_client(c, uid, gid) == NULL) { return -ENOMEM; } return pcmk_ok; } /*! * \internal * \brief Destroy a client IPC connection * * \param[in] c Connection to destroy * * \return FALSE (i.e. do not re-run this callback) */ static int32_t attrd_ipc_closed(qb_ipcs_connection_t *c) { pcmk__client_t *client = pcmk__find_client(c); if (client == NULL) { crm_trace("Ignoring request to clean up unknown connection %p", c); } else { crm_trace("Cleaning up closed client connection %p", c); /* Remove the client from the sync point waitlist if it's present. */ attrd_remove_client_from_waitlist(client); /* And no longer wait for confirmations from any peers. */ attrd_do_not_wait_for_client(client); pcmk__free_client(client); } return FALSE; } /*! * \internal * \brief Destroy a client IPC connection * * \param[in,out] c Connection to destroy * * \note We handle a destroyed connection the same as a closed one, * but we need a separate handler because the return type is different. */ static void attrd_ipc_destroy(qb_ipcs_connection_t *c) { crm_trace("Destroying client connection %p", c); attrd_ipc_closed(c); } static int32_t attrd_ipc_dispatch(qb_ipcs_connection_t * c, void *data, size_t size) { uint32_t id = 0; uint32_t flags = 0; pcmk__client_t *client = pcmk__find_client(c); xmlNode *xml = NULL; // Sanity-check, and parse XML from IPC data CRM_CHECK((c != NULL) && (client != NULL), return 0); if (data == NULL) { crm_debug("No IPC data from PID %d", pcmk__client_pid(c)); return 0; } xml = pcmk__client_data2xml(client, data, &id, &flags); if (xml == NULL) { crm_debug("Unrecognizable IPC data from PID %d", pcmk__client_pid(c)); pcmk__ipc_send_ack(client, id, flags, PCMK__XE_ACK, NULL, CRM_EX_PROTOCOL); return 0; } else { pcmk__request_t request = { .ipc_client = client, .ipc_id = id, .ipc_flags = flags, .peer = NULL, .xml = xml, .call_options = 0, .result = PCMK__UNKNOWN_RESULT, }; pcmk__assert(client->user != NULL); pcmk__update_acl_user(xml, PCMK__XA_ATTR_USER, client->user); request.op = crm_element_value_copy(request.xml, PCMK_XA_TASK); CRM_CHECK(request.op != NULL, return 0); attrd_handle_request(&request); pcmk__reset_request(&request); } pcmk__xml_free(xml); return 0; } static struct qb_ipcs_service_handlers ipc_callbacks = { .connection_accept = attrd_ipc_accept, .connection_created = NULL, .msg_process = attrd_ipc_dispatch, .connection_closed = attrd_ipc_closed, .connection_destroyed = attrd_ipc_destroy }; void attrd_ipc_fini(void) { if (ipcs != NULL) { pcmk__drop_all_clients(ipcs); qb_ipcs_destroy(ipcs); ipcs = NULL; } attrd_unregister_handlers(); pcmk__client_cleanup(); } /*! * \internal * \brief Set up attrd IPC communication */ void attrd_init_ipc(void) { pcmk__serve_attrd_ipc(&ipcs, &ipc_callbacks); } diff --git a/daemons/attrd/attrd_utils.c b/daemons/attrd/attrd_utils.c index f219b8862d..f6c6f0ac52 100644 --- a/daemons/attrd/attrd_utils.c +++ b/daemons/attrd/attrd_utils.c @@ -1,306 +1,277 @@ /* - * Copyright 2004-2024 the Pacemaker project contributors + * Copyright 2004-2025 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #include #include #include #include "pacemaker-attrd.h" cib_t *the_cib = NULL; -static bool requesting_shutdown = false; static bool shutting_down = false; static GMainLoop *mloop = NULL; /* A hash table storing information on the protocol version of each peer attrd. * The key is the peer's uname, and the value is the protocol version number. */ GHashTable *peer_protocol_vers = NULL; -/*! - * \internal - * \brief Set requesting_shutdown state - */ -void -attrd_set_requesting_shutdown(void) -{ - requesting_shutdown = true; -} - -/*! - * \internal - * \brief Clear requesting_shutdown state - */ -void -attrd_clear_requesting_shutdown(void) -{ - requesting_shutdown = false; -} - /*! * \internal * \brief Check whether local attribute manager is shutting down * - * \param[in] if_requested If \c true, also consider presence of - * \c PCMK__NODE_ATTR_SHUTDOWN attribute - * - * \return \c true if local attribute manager has begun shutdown sequence - * or (if \p if_requested is \c true) whether local node has a nonzero - * \c PCMK__NODE_ATTR_SHUTDOWN attribute set, otherwise \c false - * \note Most callers should pass \c false for \p if_requested, because the - * attribute manager needs to continue performing while the controller is - * shutting down, and even needs to be eligible for election in case all - * nodes are shutting down. + * \return \c true if local attribute manager has begun shutdown sequence, + * otherwise \c false */ bool -attrd_shutting_down(bool if_requested) +attrd_shutting_down(void) { - return shutting_down || (if_requested && requesting_shutdown); + return shutting_down; } /*! * \internal * \brief Exit (using mainloop or not, as appropriate) * * \param[in] nsig Ignored */ void attrd_shutdown(int nsig) { // Tell various functions not to do anthing shutting_down = true; // Don't respond to signals while shutting down mainloop_destroy_signal(SIGTERM); mainloop_destroy_signal(SIGCHLD); mainloop_destroy_signal(SIGPIPE); mainloop_destroy_signal(SIGUSR1); mainloop_destroy_signal(SIGUSR2); mainloop_destroy_signal(SIGTRAP); attrd_free_waitlist(); attrd_free_confirmations(); if (peer_protocol_vers != NULL) { g_hash_table_destroy(peer_protocol_vers); peer_protocol_vers = NULL; } if ((mloop == NULL) || !g_main_loop_is_running(mloop)) { /* If there's no main loop active, just exit. This should be possible * only if we get SIGTERM in brief windows at start-up and shutdown. */ crm_exit(CRM_EX_OK); } else { g_main_loop_quit(mloop); g_main_loop_unref(mloop); } } /*! * \internal * \brief Create a main loop for attrd */ void attrd_init_mainloop(void) { mloop = g_main_loop_new(NULL, FALSE); } /*! * \internal * \brief Run attrd main loop */ void attrd_run_mainloop(void) { g_main_loop_run(mloop); } /* strlen("value") */ #define plus_plus_len (5) /*! * \internal * \brief Check whether an attribute value should be expanded * * \param[in] value Attribute value to check * * \return true if value needs expansion, false otherwise */ bool attrd_value_needs_expansion(const char *value) { return ((strlen(value) >= (plus_plus_len + 2)) && (value[plus_plus_len] == '+') && ((value[plus_plus_len + 1] == '+') || (value[plus_plus_len + 1] == '='))); } /*! * \internal * \brief Expand an increment expression into an integer * * \param[in] value Attribute increment expression to expand * \param[in] old_value Previous value of attribute * * \return Expanded value */ int attrd_expand_value(const char *value, const char *old_value) { int increment = 1; int score = 0; if (pcmk_parse_score(old_value, &score, 0) != pcmk_rc_ok) { return 0; // Original value is not a score } // value++ means increment by one, value+=OFFSET means incremement by OFFSET if ((value[plus_plus_len + 1] != '+') && (pcmk_parse_score(value + plus_plus_len + 2, &increment, 0) != pcmk_rc_ok)) { increment = 0; // Invalid increment } if (increment < 0) { return QB_MAX(score + increment, -PCMK_SCORE_INFINITY); } return QB_MIN(score + increment, PCMK_SCORE_INFINITY); } /*! * \internal * \brief Create regular expression matching failure-related attributes * * \param[out] regex Where to store created regular expression * \param[in] rsc Name of resource to clear (or NULL for all) * \param[in] op Operation to clear if rsc is specified (or NULL for all) * \param[in] interval_ms Interval of operation to clear if op is specified * * \return pcmk_ok on success, -EINVAL if arguments are invalid * * \note The caller is responsible for freeing the result with regfree(). */ int attrd_failure_regex(regex_t *regex, const char *rsc, const char *op, guint interval_ms) { char *pattern = NULL; int rc; /* Create a pattern that matches desired attributes */ if (rsc == NULL) { pattern = pcmk__str_copy(ATTRD_RE_CLEAR_ALL); } else if (op == NULL) { pattern = crm_strdup_printf(ATTRD_RE_CLEAR_ONE, rsc); } else { pattern = crm_strdup_printf(ATTRD_RE_CLEAR_OP, rsc, op, interval_ms); } /* Compile pattern into regular expression */ crm_trace("Clearing attributes matching %s", pattern); rc = regcomp(regex, pattern, REG_EXTENDED|REG_NOSUB); free(pattern); return (rc == 0)? pcmk_ok : -EINVAL; } void attrd_free_attribute_value(gpointer data) { attribute_value_t *v = data; free(v->nodename); free(v->current); free(v->requested); free(v); } void attrd_free_attribute(gpointer data) { attribute_t *a = data; if(a) { free(a->id); free(a->set_id); free(a->set_type); free(a->user); mainloop_timer_del(a->timer); g_hash_table_destroy(a->values); free(a); } } /*! * \internal * \brief When a peer node leaves the cluster, stop tracking its protocol version. * * \param[in] host The peer node's uname to be removed */ void attrd_remove_peer_protocol_ver(const char *host) { if (peer_protocol_vers != NULL) { g_hash_table_remove(peer_protocol_vers, host); } } /*! * \internal * \brief When a peer node broadcasts a message with its protocol version, keep * track of that information. * * We keep track of each peer's protocol version so we know which peers to * expect confirmation messages from when handling cluster-wide sync points. * We additionally keep track of the lowest protocol version supported by all * peers so we know when we can send IPC messages containing more than one * request. * * \param[in] host The peer node's uname to be tracked * \param[in] value The peer node's protocol version */ void attrd_update_minimum_protocol_ver(const char *host, const char *value) { int ver; if (peer_protocol_vers == NULL) { peer_protocol_vers = pcmk__strkey_table(free, NULL); } pcmk__scan_min_int(value, &ver, 0); if (ver > 0) { /* Record the peer attrd's protocol version. */ g_hash_table_insert(peer_protocol_vers, pcmk__str_copy(host), GINT_TO_POINTER(ver)); /* If the protocol version is a new minimum, record it as such. */ if (minimum_protocol_version == -1 || ver < minimum_protocol_version) { minimum_protocol_version = ver; crm_trace("Set minimum attrd protocol version to %d", minimum_protocol_version); } } } diff --git a/daemons/attrd/pacemaker-attrd.h b/daemons/attrd/pacemaker-attrd.h index d9423c8915..627dd97976 100644 --- a/daemons/attrd/pacemaker-attrd.h +++ b/daemons/attrd/pacemaker-attrd.h @@ -1,260 +1,262 @@ /* * Copyright 2013-2025 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #ifndef PACEMAKER_ATTRD__H # define PACEMAKER_ATTRD__H #include #include #include #include #include #include #include /* * Legacy attrd (all pre-1.1.11 Pacemaker versions, plus all versions when used * with the no-longer-supported CMAN or corosync-plugin stacks) is unversioned. * * With atomic attrd, each attrd will send ATTRD_PROTOCOL_VERSION with every * peer request and reply. As of Pacemaker 2.0.0, at start-up each attrd will * also set a private attribute for itself with its version, so any attrd can * determine the minimum version supported by all peers. * * Protocol Pacemaker Significant changes * -------- --------- ------------------- * 1 1.1.11 PCMK__ATTRD_CMD_UPDATE (PCMK__XA_ATTR_NAME only), * PCMK__ATTRD_CMD_PEER_REMOVE, PCMK__ATTRD_CMD_REFRESH, * "flush", PCMK__ATTRD_CMD_SYNC_RESPONSE * 1 1.1.13 PCMK__ATTRD_CMD_UPDATE (with PCMK__XA_ATTR_REGEX), * PCMK__ATTRD_CMD_QUERY * 1 1.1.15 PCMK__ATTRD_CMD_UPDATE_BOTH, * PCMK__ATTRD_CMD_UPDATE_DELAY * 2 1.1.17 PCMK__ATTRD_CMD_CLEAR_FAILURE * 3 2.1.1 PCMK__ATTRD_CMD_SYNC_RESPONSE indicates remote nodes * 4 2.1.5 Multiple attributes can be updated in a single IPC * message * 5 2.1.5 Peers can request confirmation of a sent message * 6 2.1.7 PCMK__ATTRD_CMD_PEER_REMOVE supports PCMK__XA_REAP * 7 3.0.0 "flush" support dropped */ #define ATTRD_PROTOCOL_VERSION "7" #define ATTRD_SUPPORTS_MULTI_MESSAGE(x) ((x) >= 4) #define ATTRD_SUPPORTS_CONFIRMATION(x) ((x) >= 5) #define attrd_send_ack(client, id, flags) \ pcmk__ipc_send_ack((client), (id), (flags), PCMK__XE_ACK, \ ATTRD_PROTOCOL_VERSION, CRM_EX_INDETERMINATE) void attrd_init_mainloop(void); void attrd_run_mainloop(void); -void attrd_set_requesting_shutdown(void); -void attrd_clear_requesting_shutdown(void); void attrd_free_waitlist(void); -bool attrd_shutting_down(bool if_requested); +bool attrd_shutting_down(void); void attrd_shutdown(int nsig); void attrd_init_ipc(void); void attrd_ipc_fini(void); int attrd_cib_connect(int max_retry); void attrd_cib_disconnect(void); void attrd_cib_init(void); void attrd_cib_erase_transient_attrs(const char *node); bool attrd_value_needs_expansion(const char *value); int attrd_expand_value(const char *value, const char *old_value); /* regular expression to clear failures of all resources */ #define ATTRD_RE_CLEAR_ALL \ "^(" PCMK__FAIL_COUNT_PREFIX "|" PCMK__LAST_FAILURE_PREFIX ")-" /* regular expression to clear failure of all operations for one resource * (format takes resource name) */ #define ATTRD_RE_CLEAR_ONE ATTRD_RE_CLEAR_ALL "%s#.+_[0-9]+$" /* regular expression to clear failure of one operation for one resource * (format takes resource name, operation name, and interval) */ #define ATTRD_RE_CLEAR_OP ATTRD_RE_CLEAR_ALL "%s#%s_%u$" int attrd_failure_regex(regex_t *regex, const char *rsc, const char *op, guint interval_ms); extern cib_t *the_cib; extern crm_exit_t attrd_exit_status; /* Alerts */ extern lrmd_t *the_lrmd; extern crm_trigger_t *attrd_config_read; void attrd_lrmd_disconnect(void); gboolean attrd_read_options(gpointer user_data); int attrd_send_attribute_alert(const char *node, const char *node_xml_id, const char *attr, const char *value); // Elections void attrd_election_init(void); void attrd_start_election_if_needed(void); bool attrd_election_won(void); void attrd_handle_election_op(const pcmk__node_status_t *peer, xmlNode *xml); bool attrd_check_for_new_writer(const pcmk__node_status_t *peer, const xmlNode *xml); void attrd_declare_winner(void); void attrd_remove_voter(const pcmk__node_status_t *peer); void attrd_xml_add_writer(xmlNode *xml); enum attrd_attr_flags { attrd_attr_none = 0U, // At least one of attribute's values has changed since last write attrd_attr_changed = (1U << 0), // At least one of attribute's values has an unknown node XML ID attrd_attr_node_unknown = (1U << 1), // This attribute should never be written to the CIB attrd_attr_is_private = (1U << 2), // Ignore any configured delay for next write of this attribute attrd_attr_force_write = (1U << 3), }; typedef struct attribute_s { char *id; // Attribute name char *set_type; // PCMK_XE_INSTANCE_ATTRIBUTES or PCMK_XE_UTILIZATION char *set_id; // Set's XML ID to use when writing char *user; // ACL user to use for CIB writes int update; // Call ID of pending write int timeout_ms; // How long to wait for more changes before writing uint32_t flags; // Group of enum attrd_attr_flags GHashTable *values; // Key: node name, value: attribute_value_t mainloop_timer_t *timer; // Timer to use for timeout_ms } attribute_t; #define attrd_set_attr_flags(attr, flags_to_set) do { \ (attr)->flags = pcmk__set_flags_as(__func__, __LINE__, \ LOG_TRACE, "Value for attribute", (attr)->id, \ (attr)->flags, (flags_to_set), #flags_to_set); \ } while (0) #define attrd_clear_attr_flags(attr, flags_to_clear) do { \ (attr)->flags = pcmk__clear_flags_as(__func__, __LINE__, \ LOG_TRACE, "Value for attribute", (attr)->id, \ (attr)->flags, (flags_to_clear), #flags_to_clear); \ } while (0) enum attrd_value_flags { attrd_value_none = 0U, attrd_value_remote = (1U << 0), // Value is for Pacemaker Remote node attrd_value_from_peer = (1U << 1), // Value is from peer sync response }; typedef struct attribute_value_s { char *nodename; // Node that this value is for char *current; // Attribute value char *requested; // Value specified in pending CIB write, if any uint32_t flags; // Group of attrd_value_flags } attribute_value_t; #define attrd_set_value_flags(attr_value, flags_to_set) do { \ (attr_value)->flags = pcmk__set_flags_as(__func__, __LINE__, \ LOG_TRACE, "Value for node", (attr_value)->nodename, \ (attr_value)->flags, (flags_to_set), #flags_to_set); \ } while (0) #define attrd_clear_value_flags(attr_value, flags_to_clear) do { \ (attr_value)->flags = pcmk__clear_flags_as(__func__, __LINE__, \ LOG_TRACE, "Value for node", (attr_value)->nodename, \ (attr_value)->flags, (flags_to_clear), #flags_to_clear); \ } while (0) extern pcmk_cluster_t *attrd_cluster; extern GHashTable *attributes; extern GHashTable *peer_protocol_vers; #define CIB_OP_TIMEOUT_S 120 int attrd_cluster_connect(void); void attrd_broadcast_value(const attribute_t *a, const attribute_value_t *v); void attrd_peer_update(const pcmk__node_status_t *peer, xmlNode *xml, const char *host, bool filter); void attrd_peer_sync(pcmk__node_status_t *peer); void attrd_peer_remove(const char *host, bool uncache, const char *source); void attrd_peer_clear_failure(pcmk__request_t *request); void attrd_peer_sync_response(const pcmk__node_status_t *peer, bool peer_won, xmlNode *xml); void attrd_send_protocol(const pcmk__node_status_t *peer); xmlNode *attrd_client_peer_remove(pcmk__request_t *request); xmlNode *attrd_client_clear_failure(pcmk__request_t *request); xmlNode *attrd_client_update(pcmk__request_t *request); xmlNode *attrd_client_refresh(pcmk__request_t *request); xmlNode *attrd_client_query(pcmk__request_t *request); gboolean attrd_send_message(const pcmk__node_status_t *node, xmlNode *data, bool confirm); xmlNode *attrd_add_value_xml(xmlNode *parent, const attribute_t *a, const attribute_value_t *v, bool force_write); void attrd_clear_value_seen(void); void attrd_free_attribute(gpointer data); void attrd_free_attribute_value(gpointer data); attribute_t *attrd_populate_attribute(xmlNode *xml, const char *attr); char *attrd_set_id(const attribute_t *attr, const char *node_state_id); char *attrd_nvpair_id(const attribute_t *attr, const char *node_state_id); +bool attrd_for_cib(const attribute_t *a); +void attrd_drop_removed_value(const char *cib_id); +void attrd_drop_removed_set(const char *set_type, const char *cib_id); +void attrd_drop_removed_values(const char *cib_id); enum attrd_write_options { attrd_write_changed = 0, attrd_write_all = (1 << 0), attrd_write_no_delay = (1 << 1), }; void attrd_write_attributes(uint32_t options); void attrd_write_or_elect_attribute(attribute_t *a); extern int minimum_protocol_version; void attrd_remove_peer_protocol_ver(const char *host); void attrd_update_minimum_protocol_ver(const char *host, const char *value); mainloop_timer_t *attrd_add_timer(const char *id, int timeout_ms, attribute_t *attr); void attrd_unregister_handlers(void); void attrd_handle_request(pcmk__request_t *request); enum attrd_sync_point { attrd_sync_point_local, attrd_sync_point_cluster, }; typedef int (*attrd_confirmation_action_fn)(xmlNode *); void attrd_add_client_to_waitlist(pcmk__request_t *request); void attrd_ack_waitlist_clients(enum attrd_sync_point sync_point, const xmlNode *xml); int attrd_cluster_sync_point_update(xmlNode *xml); void attrd_do_not_expect_from_peer(const char *host); void attrd_do_not_wait_for_client(pcmk__client_t *client); void attrd_expect_confirmations(pcmk__request_t *request, attrd_confirmation_action_fn fn); void attrd_free_confirmations(void); void attrd_handle_confirmation(int callid, const char *host); void attrd_remove_client_from_waitlist(pcmk__client_t *client); const char *attrd_request_sync_point(xmlNode *xml); bool attrd_request_has_sync_point(xmlNode *xml); extern gboolean stand_alone; // Node utilities (from attrd_nodes.c) const char *attrd_get_node_xml_id(const char *node_name); void attrd_set_node_xml_id(const char *node_name, const char *node_xml_id); void attrd_forget_node_xml_id(const char *node_name); void attrd_cleanup_xml_ids(void); #endif /* PACEMAKER_ATTRD__H */ diff --git a/daemons/controld/controld_attrd.c b/daemons/controld/controld_attrd.c index eff8070818..32fcbe2568 100644 --- a/daemons/controld/controld_attrd.c +++ b/daemons/controld/controld_attrd.c @@ -1,158 +1,165 @@ /* - * Copyright 2006-2024 the Pacemaker project contributors + * Copyright 2006-2025 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include static pcmk_ipc_api_t *attrd_api = NULL; void controld_close_attrd_ipc(void) { if (attrd_api != NULL) { crm_trace("Closing connection to " PCMK__SERVER_ATTRD); pcmk_disconnect_ipc(attrd_api); pcmk_free_ipc_api(attrd_api); attrd_api = NULL; } } static inline const char * node_type(bool is_remote) { return is_remote? "Pacemaker Remote" : "cluster"; } static inline const char * when(void) { return pcmk_is_set(controld_globals.fsa_input_register, R_SHUTDOWN)? " at shutdown" : ""; } static void handle_attr_error(void) { if (AM_I_DC) { /* We are unable to provide accurate information to the * scheduler, so allow another node to take over DC. * @TODO Should we do this unconditionally on any failure? */ crmd_exit(CRM_EX_FATAL); } else if (pcmk_is_set(controld_globals.fsa_input_register, R_SHUTDOWN)) { // Fast-track shutdown since unable to request via attribute register_fsa_input(C_FSA_INTERNAL, I_FAIL, NULL); } } void update_attrd(const char *host, const char *name, const char *value, const char *user_name, gboolean is_remote_node) { int rc = pcmk_rc_ok; if (attrd_api == NULL) { rc = pcmk_new_ipc_api(&attrd_api, pcmk_ipc_attrd); } if (rc == pcmk_rc_ok) { uint32_t attrd_opts = pcmk__node_attr_value; if (is_remote_node) { pcmk__set_node_attr_flags(attrd_opts, pcmk__node_attr_remote); } rc = pcmk__attrd_api_update(attrd_api, host, name, value, NULL, NULL, user_name, attrd_opts); } if (rc != pcmk_rc_ok) { do_crm_log(AM_I_DC? LOG_CRIT : LOG_ERR, "Could not update attribute %s=%s for %s node %s%s: %s " QB_XS " rc=%d", name, value, node_type(is_remote_node), host, when(), pcmk_rc_str(rc), rc); handle_attr_error(); } } void update_attrd_list(GList *attrs, uint32_t opts) { int rc = pcmk_rc_ok; if (attrd_api == NULL) { rc = pcmk_new_ipc_api(&attrd_api, pcmk_ipc_attrd); } if (rc == pcmk_rc_ok) { rc = pcmk__attrd_api_update_list(attrd_api, attrs, NULL, NULL, NULL, opts | pcmk__node_attr_value); } if (rc != pcmk_rc_ok) { do_crm_log(AM_I_DC? LOG_CRIT : LOG_ERR, "Could not update multiple node attributes: %s " QB_XS " rc=%d", pcmk_rc_str(rc), rc); handle_attr_error(); } } +/*! + * \internal + * \brief Ask attribute manager to purge a node and its transient attributes + * + * \param[in] node_name Node to purge + * \param[in] from_cache If true, purge from node caches as well + */ void -update_attrd_remote_node_removed(const char *host, const char *user_name) +controld_purge_node_attrs(const char *node_name, bool from_cache) { int rc = pcmk_rc_ok; if (attrd_api == NULL) { rc = pcmk_new_ipc_api(&attrd_api, pcmk_ipc_attrd); } if (rc == pcmk_rc_ok) { - crm_trace("Asking attribute manager to purge Pacemaker Remote node %s", - host); - rc = pcmk__attrd_api_purge(attrd_api, host, true); + crm_debug("Asking %s to purge transient attributes%s for %s ", + pcmk_ipc_name(attrd_api, true), + (from_cache? " and node cache" : ""), node_name); + rc = pcmk__attrd_api_purge(attrd_api, node_name, from_cache); } if (rc != pcmk_rc_ok) { - crm_err("Could not purge Pacemaker Remote node %s " - "in attribute manager%s: %s " QB_XS " rc=%d", - host, when(), pcmk_rc_str(rc), rc); + crm_err("Could not purge node %s from attribute manager%s: %s " + QB_XS " rc=%d", node_name, when(), pcmk_rc_str(rc), rc); } } void update_attrd_clear_failures(const char *host, const char *rsc, const char *op, const char *interval_spec, gboolean is_remote_node) { int rc = pcmk_rc_ok; if (attrd_api == NULL) { rc = pcmk_new_ipc_api(&attrd_api, pcmk_ipc_attrd); } if (rc == pcmk_rc_ok) { uint32_t attrd_opts = pcmk__node_attr_none; if (is_remote_node) { pcmk__set_node_attr_flags(attrd_opts, pcmk__node_attr_remote); } rc = pcmk__attrd_api_clear_failures(attrd_api, host, rsc, op, interval_spec, NULL, attrd_opts); } if (rc != pcmk_rc_ok) { const char *interval_desc = "all"; if (op != NULL) { interval_desc = pcmk__s(interval_spec, "nonrecurring"); } crm_err("Could not clear failure of %s %s for %s on %s node %s%s: %s " QB_XS " rc=%d", interval_desc, pcmk__s(op, "operations"), pcmk__s(rsc, "all resources"), node_type(is_remote_node), host, when(), pcmk_rc_str(rc), rc); } } diff --git a/daemons/controld/controld_callbacks.c b/daemons/controld/controld_callbacks.c index 48c255ed19..cd4b077dd5 100644 --- a/daemons/controld/controld_callbacks.c +++ b/daemons/controld/controld_callbacks.c @@ -1,396 +1,382 @@ /* * Copyright 2004-2025 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include // PRIu32 #include // bool #include // uint32_t #include // NULL #include #include #include #include #include #include #include /* From join_dc... */ extern gboolean check_join_state(enum crmd_fsa_state cur_state, const char *source); void crmd_ha_msg_filter(xmlNode * msg) { if (AM_I_DC) { const char *sys_from = crm_element_value(msg, PCMK__XA_CRM_SYS_FROM); if (pcmk__str_eq(sys_from, CRM_SYSTEM_DC, pcmk__str_casei)) { const char *from = crm_element_value(msg, PCMK__XA_SRC); if (!controld_is_local_node(from)) { int level = LOG_INFO; const char *op = crm_element_value(msg, PCMK__XA_CRM_TASK); /* make sure the election happens NOW */ if (controld_globals.fsa_state != S_ELECTION) { ha_msg_input_t new_input; level = LOG_WARNING; new_input.msg = msg; register_fsa_error_adv(C_FSA_INTERNAL, I_ELECTION, NULL, &new_input, __func__); } do_crm_log(level, "Another DC detected: %s (op=%s)", from, op); goto done; } } } else { const char *sys_to = crm_element_value(msg, PCMK__XA_CRM_SYS_TO); if (pcmk__str_eq(sys_to, CRM_SYSTEM_DC, pcmk__str_casei)) { return; } } /* crm_log_xml_trace(msg, "HA[inbound]"); */ route_message(C_HA_MESSAGE, msg); done: controld_trigger_fsa(); } /*! * \internal * \brief Check whether a node is online * * \param[in] node Node to check * * \retval -1 if completely dead * \retval 0 if partially alive * \retval 1 if completely alive */ static int node_alive(const pcmk__node_status_t *node) { if (pcmk_is_set(node->flags, pcmk__node_status_remote)) { // Pacemaker Remote nodes can't be partially alive if (pcmk__str_eq(node->state, PCMK_VALUE_MEMBER, pcmk__str_none)) { return 1; } return -1; } else if (pcmk__cluster_is_node_active(node)) { // Completely up cluster node: both cluster member and peer return 1; } else if (!pcmk_is_set(node->processes, crm_get_cluster_proc()) && !pcmk__str_eq(node->state, PCMK_VALUE_MEMBER, pcmk__str_none)) { // Completely down cluster node: neither cluster member nor peer return -1; } // Partially up cluster node: only cluster member or only peer return 0; } #define state_text(state) ((state)? (const char *)(state) : "in unknown state") // @TODO This is insanely long, and some parts should be functionized void peer_update_callback(enum pcmk__node_update type, pcmk__node_status_t *node, const void *data) { uint32_t old = 0; bool appeared = FALSE; bool is_remote = pcmk_is_set(node->flags, pcmk__node_status_remote); controld_node_pending_timer(node); /* The controller waits to receive some information from the membership * layer before declaring itself operational. If this is being called for a * cluster node, indicate that we have it. */ if (!is_remote) { controld_set_fsa_input_flags(R_PEER_DATA); } if ((type == pcmk__node_update_processes) && pcmk_is_set(node->processes, crm_get_cluster_proc()) && !AM_I_DC && !is_remote) { /* relay_message() on the recipient ignores these messages, but * libcrmcluster will have cached the node name by then */ xmlNode *query = pcmk__new_request(pcmk_ipc_controld, CRM_SYSTEM_CRMD, NULL, CRM_SYSTEM_CRMD, CRM_OP_HELLO, NULL); crm_debug("Sending hello to node %" PRIu32 " so that it learns our " "node name", node->cluster_layer_id); pcmk__cluster_send_message(node, pcmk_ipc_controld, query); pcmk__xml_free(query); } if (node->name == NULL) { return; } switch (type) { case pcmk__node_update_name: /* If we've never seen the node, then it also won't be in the status section */ crm_info("%s node %s is now %s", (is_remote? "Remote" : "Cluster"), node->name, state_text(node->state)); return; case pcmk__node_update_state: /* This callback should not be called unless the state actually * changed, but here's a failsafe just in case. */ CRM_CHECK(!pcmk__str_eq(data, node->state, pcmk__str_casei), return); crm_info("%s node %s is now %s (was %s)", (is_remote? "Remote" : "Cluster"), node->name, state_text(node->state), state_text(data)); if (pcmk__str_eq(PCMK_VALUE_MEMBER, node->state, pcmk__str_none)) { appeared = TRUE; if (!is_remote) { remove_stonith_cleanup(node->name); } } else { controld_remove_failed_sync_node(node->name); controld_remove_voter(node->name); } crmd_alert_node_event(node); break; case pcmk__node_update_processes: CRM_CHECK(data != NULL, return); old = *(const uint32_t *)data; appeared = pcmk_is_set(node->processes, crm_get_cluster_proc()); { const char *dc_s = controld_globals.dc_name; if ((dc_s == NULL) && AM_I_DC) { dc_s = PCMK_VALUE_TRUE; } crm_info("Node %s is %s a peer " QB_XS " DC=%s old=%#07x new=%#07x", node->name, (appeared? "now" : "no longer"), pcmk__s(dc_s, ""), old, node->processes); } if (!pcmk_is_set((node->processes ^ old), crm_get_cluster_proc())) { /* Peer status did not change. This should not be possible, * since we don't track process flags other than peer status. */ crm_trace("Process flag %#7x did not change from %#7x to %#7x", crm_get_cluster_proc(), old, node->processes); return; } if (!appeared) { node->peer_lost = time(NULL); controld_remove_failed_sync_node(node->name); controld_remove_voter(node->name); } if (!pcmk_is_set(controld_globals.fsa_input_register, R_CIB_CONNECTED)) { crm_trace("Ignoring peer status change because not connected to CIB"); return; } else if (controld_globals.fsa_state == S_STOPPING) { crm_trace("Ignoring peer status change because stopping"); return; } if (!appeared && controld_is_local_node(node->name)) { /* Did we get evicted? */ crm_notice("Our peer connection failed"); register_fsa_input(C_CRMD_STATUS_CALLBACK, I_ERROR, NULL); } else if (pcmk__str_eq(node->name, controld_globals.dc_name, pcmk__str_casei) && !pcmk__cluster_is_node_active(node)) { - /* The DC has left, so delete its transient attributes and - * trigger a new election. - * - * A DC sends its shutdown request to all peers, who update the - * DC's expected state to down. This avoids fencing upon - * deletion of its transient attributes. - */ + // The DC has left, so trigger a new election crm_notice("Our peer on the DC (%s) is dead", controld_globals.dc_name); - register_fsa_input(C_CRMD_STATUS_CALLBACK, I_ELECTION, NULL); - controld_delete_node_state(node->name, controld_section_attrs, - cib_none); - } else if (AM_I_DC || pcmk_is_set(controld_globals.flags, controld_dc_left) || (controld_globals.dc_name == NULL)) { /* This only needs to be done once, so normally the DC should do * it. However if there is no DC, every node must do it, since * there is no other way to ensure some one node does it. */ if (appeared) { te_trigger_stonith_history_sync(FALSE); - } else { - controld_delete_node_state(node->name, - controld_section_attrs, - cib_none); } } break; } if (AM_I_DC) { xmlNode *update = NULL; uint32_t flags = controld_node_update_peer; int alive = node_alive(node); pcmk__graph_action_t *down = match_down_event(node->xml_id); crm_trace("Alive=%d, appeared=%d, down=%d", alive, appeared, (down? down->id : -1)); if (appeared && (alive > 0) && !is_remote) { register_fsa_input_before(C_FSA_INTERNAL, I_NODE_JOIN, NULL); } if (down) { const char *task = crm_element_value(down->xml, PCMK_XA_OPERATION); if (pcmk__str_eq(task, PCMK_ACTION_STONITH, pcmk__str_casei)) { const bool confirmed = pcmk_is_set(down->flags, pcmk__graph_action_confirmed); /* tengine_stonith_callback() confirms fence actions */ crm_trace("Updating CIB %s fencer reported fencing of %s complete", (confirmed? "after" : "before"), node->name); } else if (!appeared && pcmk__str_eq(task, PCMK_ACTION_DO_SHUTDOWN, pcmk__str_casei)) { // Shutdown actions are immediately confirmed (i.e. no_wait) if (!is_remote) { flags |= controld_node_update_join |controld_node_update_expected; crmd_peer_down(node, FALSE); check_join_state(controld_globals.fsa_state, __func__); } if (alive >= 0) { crm_info("%s of peer %s is in progress " QB_XS " action=%d", task, node->name, down->id); } else { crm_notice("%s of peer %s is complete " QB_XS " action=%d", task, node->name, down->id); pcmk__update_graph(controld_globals.transition_graph, down); trigger_graph(); } } else { const char *liveness = "alive"; if (alive == 0) { liveness = "partially alive"; } else if (alive < 0) { liveness = "dead"; } crm_trace("Node %s is %s, was expected to %s (op %d)", node->name, liveness, task, down->id); } } else if (appeared == FALSE) { if ((controld_globals.transition_graph == NULL) || (controld_globals.transition_graph->id <= 0)) { crm_info("Stonith/shutdown of node %s is unknown to the " "current DC", node->name); } else { crm_warn("Stonith/shutdown of node %s was not expected", node->name); } if (!is_remote) { crm_update_peer_join(__func__, node, controld_join_none); check_join_state(controld_globals.fsa_state, __func__); } abort_transition(PCMK_SCORE_INFINITY, pcmk__graph_restart, "Node failure", NULL); fail_incompletable_actions(controld_globals.transition_graph, node->xml_id); } else { crm_trace("Node %s came up, was not expected to be down", node->name); } if (is_remote) { /* A pacemaker_remote node won't have its cluster status updated * in the CIB by membership-layer callbacks, so do it here. */ flags |= controld_node_update_cluster; /* Trigger resource placement on newly integrated nodes */ if (appeared) { abort_transition(PCMK_SCORE_INFINITY, pcmk__graph_restart, "Pacemaker Remote node integrated", NULL); } } if (!appeared && (type == pcmk__node_update_processes) && (node->when_member > 1)) { /* The node left CPG but is still a cluster member. Set its * membership time to 1 to record it in the cluster state as a * boolean, so we don't fence it due to * PCMK_OPT_NODE_PENDING_TIMEOUT. */ node->when_member = 1; flags |= controld_node_update_cluster; controld_node_pending_timer(node); } /* Update the CIB node state */ update = create_node_state_update(node, flags, NULL, __func__); if (update == NULL) { crm_debug("Node state update not yet possible for %s", node->name); } else { fsa_cib_anon_update(PCMK_XE_STATUS, update); } pcmk__xml_free(update); } controld_trigger_fsa(); } gboolean crm_fsa_trigger(gpointer user_data) { crm_trace("Invoked (queue len: %d)", g_list_length(controld_globals.fsa_message_queue)); s_crmd_fsa(C_FSA_INTERNAL); crm_trace("Exited (queue len: %d)", g_list_length(controld_globals.fsa_message_queue)); return TRUE; } diff --git a/daemons/controld/controld_cib.c b/daemons/controld/controld_cib.c index e2a0d50179..b5e6a29146 100644 --- a/daemons/controld/controld_cib.c +++ b/daemons/controld/controld_cib.c @@ -1,1061 +1,1028 @@ /* * Copyright 2004-2025 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include /* sleep */ #include #include #include #include #include // Call ID of the most recent in-progress CIB resource update (or 0 if none) static int pending_rsc_update = 0; /*! * \internal * \brief Respond to a dropped CIB connection * * \param[in] user_data CIB connection that dropped */ static void handle_cib_disconnect(gpointer user_data) { CRM_LOG_ASSERT(user_data == controld_globals.cib_conn); controld_trigger_fsa(); controld_globals.cib_conn->state = cib_disconnected; if (pcmk_is_set(controld_globals.fsa_input_register, R_CIB_CONNECTED)) { // @TODO This should trigger a reconnect, not a shutdown crm_crit("Lost connection to the CIB manager, shutting down"); register_fsa_input(C_FSA_INTERNAL, I_ERROR, NULL); controld_clear_fsa_input_flags(R_CIB_CONNECTED); } else { // Expected crm_info("Disconnected from the CIB manager"); } } static void do_cib_updated(const char *event, xmlNode * msg) { const xmlNode *patchset = NULL; const char *client_name = NULL; crm_debug("Received CIB diff notification: DC=%s", pcmk__btoa(AM_I_DC)); if (cib__get_notify_patchset(msg, &patchset) != pcmk_rc_ok) { return; } if (pcmk__cib_element_in_patchset(patchset, PCMK_XE_ALERTS) || pcmk__cib_element_in_patchset(patchset, PCMK_XE_CRM_CONFIG)) { controld_trigger_config(); } if (!AM_I_DC) { // We're not in control of the join sequence return; } client_name = crm_element_value(msg, PCMK__XA_CIB_CLIENTNAME); if (!cib__client_triggers_refresh(client_name)) { // The CIB is still accurate return; } if (pcmk__cib_element_in_patchset(patchset, PCMK_XE_NODES) || pcmk__cib_element_in_patchset(patchset, PCMK_XE_STATUS)) { /* An unsafe client modified the PCMK_XE_NODES or PCMK_XE_STATUS * section. Ensure the node list is up-to-date, and start the join * process again so we get everyone's current resource history. */ if (client_name == NULL) { client_name = crm_element_value(msg, PCMK__XA_CIB_CLIENTID); } crm_notice("Populating nodes and starting an election after %s event " "triggered by %s", event, pcmk__s(client_name, "(unidentified client)")); populate_cib_nodes(controld_node_update_quick|controld_node_update_all, __func__); register_fsa_input(C_FSA_INTERNAL, I_ELECTION, NULL); } } void controld_disconnect_cib_manager(void) { cib_t *cib_conn = controld_globals.cib_conn; pcmk__assert(cib_conn != NULL); crm_debug("Disconnecting from the CIB manager"); controld_clear_fsa_input_flags(R_CIB_CONNECTED); cib_conn->cmds->del_notify_callback(cib_conn, PCMK__VALUE_CIB_DIFF_NOTIFY, do_cib_updated); cib_free_callbacks(cib_conn); if (cib_conn->state != cib_disconnected) { cib_conn->cmds->set_secondary(cib_conn, cib_discard_reply); cib_conn->cmds->signoff(cib_conn); } } /* A_CIB_STOP, A_CIB_START, O_CIB_RESTART */ void do_cib_control(long long action, enum crmd_fsa_cause cause, enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data) { static int cib_retries = 0; cib_t *cib_conn = controld_globals.cib_conn; void (*dnotify_fn) (gpointer user_data) = handle_cib_disconnect; void (*update_cb) (const char *event, xmlNodePtr msg) = do_cib_updated; int rc = pcmk_ok; pcmk__assert(cib_conn != NULL); if (pcmk_is_set(action, A_CIB_STOP)) { if ((cib_conn->state != cib_disconnected) && (pending_rsc_update != 0)) { crm_info("Waiting for resource update %d to complete", pending_rsc_update); crmd_fsa_stall(FALSE); return; } controld_disconnect_cib_manager(); } if (!pcmk_is_set(action, A_CIB_START)) { return; } if (cur_state == S_STOPPING) { crm_err("Ignoring request to connect to the CIB manager after " "shutdown"); return; } rc = cib_conn->cmds->signon(cib_conn, crm_system_name, cib_command_nonblocking); if (rc != pcmk_ok) { // A short wait that usually avoids stalling the FSA sleep(1); rc = cib_conn->cmds->signon(cib_conn, crm_system_name, cib_command_nonblocking); } if (rc != pcmk_ok) { crm_info("Could not connect to the CIB manager: %s", pcmk_strerror(rc)); } else if (cib_conn->cmds->set_connection_dnotify(cib_conn, dnotify_fn) != pcmk_ok) { crm_err("Could not set dnotify callback"); } else if (cib_conn->cmds->add_notify_callback(cib_conn, PCMK__VALUE_CIB_DIFF_NOTIFY, update_cb) != pcmk_ok) { crm_err("Could not set CIB notification callback (update)"); } else { controld_set_fsa_input_flags(R_CIB_CONNECTED); cib_retries = 0; } if (!pcmk_is_set(controld_globals.fsa_input_register, R_CIB_CONNECTED)) { cib_retries++; if (cib_retries < 30) { crm_warn("Couldn't complete CIB registration %d times... " "pause and retry", cib_retries); controld_start_wait_timer(); crmd_fsa_stall(FALSE); } else { crm_err("Could not complete CIB registration %d times... " "hard error", cib_retries); register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); } } } #define MIN_CIB_OP_TIMEOUT (30) /*! * \internal * \brief Get the timeout (in seconds) that should be used with CIB operations * * \return The maximum of 30 seconds, the value of the PCMK_cib_timeout * environment variable, or 10 seconds times one more than the number of * nodes in the cluster. */ unsigned int cib_op_timeout(void) { unsigned int calculated_timeout = 10U * (pcmk__cluster_num_active_nodes() + pcmk__cluster_num_remote_nodes() + 1U); calculated_timeout = QB_MAX(calculated_timeout, MIN_CIB_OP_TIMEOUT); crm_trace("Calculated timeout: %s", pcmk__readable_interval(calculated_timeout * 1000)); if (controld_globals.cib_conn) { controld_globals.cib_conn->call_timeout = calculated_timeout; } return calculated_timeout; } /*! * \internal * \brief Get CIB call options to use local scope if primary is unavailable * * \return CIB call options */ int crmd_cib_smart_opt(void) { int call_opt = cib_none; if ((controld_globals.fsa_state == S_ELECTION) || (controld_globals.fsa_state == S_PENDING)) { crm_info("Sending update to local CIB in state: %s", fsa_state2string(controld_globals.fsa_state)); cib__set_call_options(call_opt, "update", cib_none); } return call_opt; } static void cib_delete_callback(xmlNode *msg, int call_id, int rc, xmlNode *output, void *user_data) { char *desc = user_data; if (rc == 0) { crm_debug("Deletion of %s (via CIB call %d) succeeded", desc, call_id); } else { crm_warn("Deletion of %s (via CIB call %d) failed: %s " QB_XS " rc=%d", desc, call_id, pcmk_strerror(rc), rc); } } // Searches for various portions of PCMK__XE_NODE_STATE to delete // Match a particular node's PCMK__XE_NODE_STATE (takes node name 1x) #define XPATH_NODE_STATE "//" PCMK__XE_NODE_STATE "[@" PCMK_XA_UNAME "='%s']" // Node's lrm section (name 1x) #define XPATH_NODE_LRM XPATH_NODE_STATE "/" PCMK__XE_LRM /* Node's PCMK__XE_LRM_RSC_OP entries and PCMK__XE_LRM_RESOURCE entries without * unexpired lock * (name 2x, (seconds_since_epoch - PCMK_OPT_SHUTDOWN_LOCK_LIMIT) 1x) */ #define XPATH_NODE_LRM_UNLOCKED XPATH_NODE_STATE "//" PCMK__XE_LRM_RSC_OP \ "|" XPATH_NODE_STATE \ "//" PCMK__XE_LRM_RESOURCE \ "[not(@" PCMK_OPT_SHUTDOWN_LOCK ") " \ "or " PCMK_OPT_SHUTDOWN_LOCK "<%lld]" -// Node's PCMK__XE_TRANSIENT_ATTRIBUTES section (name 1x) -#define XPATH_NODE_ATTRS XPATH_NODE_STATE "/" PCMK__XE_TRANSIENT_ATTRIBUTES - -// Everything under PCMK__XE_NODE_STATE (name 1x) -#define XPATH_NODE_ALL XPATH_NODE_STATE "/*" - -/* Unlocked history + transient attributes - * (name 2x, (seconds_since_epoch - PCMK_OPT_SHUTDOWN_LOCK_LIMIT) 1x, name 1x) - */ -#define XPATH_NODE_ALL_UNLOCKED XPATH_NODE_LRM_UNLOCKED "|" XPATH_NODE_ATTRS - /*! * \internal - * \brief Get the XPath and description of a node state section to be deleted + * \brief Get the XPath and description of resource history to be deleted * - * \param[in] uname Desired node - * \param[in] section Subsection of \c PCMK__XE_NODE_STATE to be deleted - * \param[out] xpath Where to store XPath of \p section - * \param[out] desc If not \c NULL, where to store description of \p section + * \param[in] uname Name of node to delete resource history for + * \param[in] unlocked_only If true, delete history of only unlocked resources + * \param[out] xpath Where to store XPath for history deletion + * \param[out] desc If not NULL, where to store loggable description */ void -controld_node_state_deletion_strings(const char *uname, - enum controld_section_e section, - char **xpath, char **desc) +controld_node_history_deletion_strings(const char *uname, bool unlocked_only, + char **xpath, char **desc) { const char *desc_pre = NULL; // Shutdown locks that started before this time are expired long long expire = (long long) time(NULL) - controld_globals.shutdown_lock_limit; - switch (section) { - case controld_section_lrm: - *xpath = crm_strdup_printf(XPATH_NODE_LRM, uname); - desc_pre = "resource history"; - break; - case controld_section_lrm_unlocked: - *xpath = crm_strdup_printf(XPATH_NODE_LRM_UNLOCKED, - uname, uname, expire); - desc_pre = "resource history (other than shutdown locks)"; - break; - case controld_section_attrs: - *xpath = crm_strdup_printf(XPATH_NODE_ATTRS, uname); - desc_pre = "transient attributes"; - break; - case controld_section_all: - *xpath = crm_strdup_printf(XPATH_NODE_ALL, uname); - desc_pre = "all state"; - break; - case controld_section_all_unlocked: - *xpath = crm_strdup_printf(XPATH_NODE_ALL_UNLOCKED, - uname, uname, expire, uname); - desc_pre = "all state (other than shutdown locks)"; - break; - default: - // We called this function incorrectly - pcmk__assert(false); - break; + if (unlocked_only) { + *xpath = crm_strdup_printf(XPATH_NODE_LRM_UNLOCKED, + uname, uname, expire); + desc_pre = "resource history (other than shutdown locks)"; + } else { + *xpath = crm_strdup_printf(XPATH_NODE_LRM, uname); + desc_pre = "resource history"; } if (desc != NULL) { *desc = crm_strdup_printf("%s for node %s", desc_pre, uname); } } + /*! * \internal - * \brief Delete subsection of a node's CIB \c PCMK__XE_NODE_STATE + * \brief Delete a node's resource history from the CIB * - * \param[in] uname Desired node - * \param[in] section Subsection of \c PCMK__XE_NODE_STATE to delete - * \param[in] options CIB call options to use + * \param[in] uname Desired node + * \param[in] unlocked_only If true, delete history of only unlocked resources + * \param[in] options CIB call options to use */ void -controld_delete_node_state(const char *uname, enum controld_section_e section, - int options) +controld_delete_node_history(const char *uname, bool unlocked_only, int options) { cib_t *cib = controld_globals.cib_conn; char *xpath = NULL; char *desc = NULL; int cib_rc = pcmk_ok; pcmk__assert((uname != NULL) && (cib != NULL)); - controld_node_state_deletion_strings(uname, section, &xpath, &desc); - + controld_node_history_deletion_strings(uname, unlocked_only, &xpath, &desc); cib__set_call_options(options, "node state deletion", cib_xpath|cib_multiple); cib_rc = cib->cmds->remove(cib, xpath, NULL, options); fsa_register_cib_callback(cib_rc, desc, cib_delete_callback); crm_info("Deleting %s (via CIB call %d) " QB_XS " xpath=%s", desc, cib_rc, xpath); // CIB library handles freeing desc free(xpath); } // Takes node name and resource ID #define XPATH_RESOURCE_HISTORY "//" PCMK__XE_NODE_STATE \ "[@" PCMK_XA_UNAME "='%s']/" \ PCMK__XE_LRM "/" PCMK__XE_LRM_RESOURCES \ "/" PCMK__XE_LRM_RESOURCE \ "[@" PCMK_XA_ID "='%s']" // @TODO could add "and @PCMK_OPT_SHUTDOWN_LOCK" to limit to locks /*! * \internal * \brief Clear resource history from CIB for a given resource and node * * \param[in] rsc_id ID of resource to be cleared * \param[in] node Node whose resource history should be cleared * \param[in] user_name ACL user name to use * \param[in] call_options CIB call options * * \return Standard Pacemaker return code */ int controld_delete_resource_history(const char *rsc_id, const char *node, const char *user_name, int call_options) { char *desc = NULL; char *xpath = NULL; int rc = pcmk_rc_ok; cib_t *cib = controld_globals.cib_conn; CRM_CHECK((rsc_id != NULL) && (node != NULL), return EINVAL); desc = crm_strdup_printf("resource history for %s on %s", rsc_id, node); if (cib == NULL) { crm_err("Unable to clear %s: no CIB connection", desc); free(desc); return ENOTCONN; } // Ask CIB to delete the entry xpath = crm_strdup_printf(XPATH_RESOURCE_HISTORY, node, rsc_id); cib->cmds->set_user(cib, user_name); rc = cib->cmds->remove(cib, xpath, NULL, call_options|cib_xpath); cib->cmds->set_user(cib, NULL); if (rc < 0) { rc = pcmk_legacy2rc(rc); crm_err("Could not delete resource status of %s on %s%s%s: %s " QB_XS " rc=%d", rsc_id, node, (user_name? " for user " : ""), (user_name? user_name : ""), pcmk_rc_str(rc), rc); free(desc); free(xpath); return rc; } if (pcmk_is_set(call_options, cib_sync_call)) { if (pcmk_is_set(call_options, cib_dryrun)) { crm_debug("Deletion of %s would succeed", desc); } else { crm_debug("Deletion of %s succeeded", desc); } free(desc); } else { crm_info("Clearing %s (via CIB call %d) " QB_XS " xpath=%s", desc, rc, xpath); fsa_register_cib_callback(rc, desc, cib_delete_callback); // CIB library handles freeing desc } free(xpath); return pcmk_rc_ok; } /*! * \internal * \brief Build XML and string of parameters meeting some criteria, for digest * * \param[in] op Executor event with parameter table to use * \param[in] metadata Parsed meta-data for executed resource agent * \param[in] param_type Flag used for selection criteria * \param[out] result Will be set to newly created XML with selected * parameters as attributes * * \return Newly allocated space-separated string of parameter names * \note Selection criteria varies by param_type: for the restart digest, we * want parameters that are *not* marked reloadable (OCF 1.1) or that * *are* marked unique (pre-1.1), for both string and XML results; for the * secure digest, we want parameters that *are* marked private for the * string, but parameters that are *not* marked private for the XML. * \note It is the caller's responsibility to free the string return value with * \p g_string_free() and the XML result with \p pcmk__xml_free(). */ static GString * build_parameter_list(const lrmd_event_data_t *op, const struct ra_metadata_s *metadata, enum ra_param_flags_e param_type, xmlNode **result) { GString *list = NULL; *result = pcmk__xe_create(NULL, PCMK_XE_PARAMETERS); /* Consider all parameters only except private ones to be consistent with * what scheduler does with calculate_secure_digest(). */ if (param_type == ra_param_private && compare_version(controld_globals.dc_version, "3.16.0") >= 0) { g_hash_table_foreach(op->params, hash2field, *result); pcmk__filter_op_for_digest(*result); } for (GList *iter = metadata->ra_params; iter != NULL; iter = iter->next) { struct ra_param_s *param = (struct ra_param_s *) iter->data; bool accept_for_list = false; bool accept_for_xml = false; switch (param_type) { case ra_param_reloadable: accept_for_list = !pcmk_is_set(param->rap_flags, param_type); accept_for_xml = accept_for_list; break; case ra_param_unique: accept_for_list = pcmk_is_set(param->rap_flags, param_type); accept_for_xml = accept_for_list; break; case ra_param_private: accept_for_list = pcmk_is_set(param->rap_flags, param_type); accept_for_xml = !accept_for_list; break; } if (accept_for_list) { crm_trace("Attr %s is %s", param->rap_name, ra_param_flag2text(param_type)); if (list == NULL) { // We will later search for " WORD ", so start list with a space pcmk__add_word(&list, 256, " "); } pcmk__add_word(&list, 0, param->rap_name); } else { crm_trace("Rejecting %s for %s", param->rap_name, ra_param_flag2text(param_type)); } if (accept_for_xml) { const char *v = g_hash_table_lookup(op->params, param->rap_name); if (v != NULL) { crm_trace("Adding attr %s=%s to the xml result", param->rap_name, v); crm_xml_add(*result, param->rap_name, v); } } else { crm_trace("Removing attr %s from the xml result", param->rap_name); pcmk__xe_remove_attr(*result, param->rap_name); } } if (list != NULL) { // We will later search for " WORD ", so end list with a space pcmk__add_word(&list, 0, " "); } return list; } static void append_restart_list(lrmd_event_data_t *op, struct ra_metadata_s *metadata, xmlNode *update, const char *version) { GString *list = NULL; char *digest = NULL; xmlNode *restart = NULL; CRM_LOG_ASSERT(op->params != NULL); if (op->interval_ms > 0) { /* monitors are not reloadable */ return; } if (pcmk_is_set(metadata->ra_flags, ra_supports_reload_agent)) { /* Add parameters not marked reloadable to the PCMK__XA_OP_FORCE_RESTART * list */ list = build_parameter_list(op, metadata, ra_param_reloadable, &restart); } else if (pcmk_is_set(metadata->ra_flags, ra_supports_legacy_reload)) { /* @COMPAT pre-OCF-1.1 resource agents * * Before OCF 1.1, Pacemaker abused "unique=0" to indicate * reloadability. Add any parameters with unique="1" to the * PCMK__XA_OP_FORCE_RESTART list. */ list = build_parameter_list(op, metadata, ra_param_unique, &restart); } else { // Resource does not support agent reloads return; } digest = pcmk__digest_operation(restart); /* Add PCMK__XA_OP_FORCE_RESTART and PCMK__XA_OP_RESTART_DIGEST to indicate * the resource supports reload, no matter if it actually supports any * reloadable parameters */ crm_xml_add(update, PCMK__XA_OP_FORCE_RESTART, (list == NULL)? "" : (const char *) list->str); crm_xml_add(update, PCMK__XA_OP_RESTART_DIGEST, digest); if ((list != NULL) && (list->len > 0)) { crm_trace("%s: %s, %s", op->rsc_id, digest, (const char *) list->str); } else { crm_trace("%s: %s", op->rsc_id, digest); } if (list != NULL) { g_string_free(list, TRUE); } pcmk__xml_free(restart); free(digest); } static void append_secure_list(lrmd_event_data_t *op, struct ra_metadata_s *metadata, xmlNode *update, const char *version) { GString *list = NULL; char *digest = NULL; xmlNode *secure = NULL; CRM_LOG_ASSERT(op->params != NULL); /* To keep PCMK__XA_OP_SECURE_PARAMS short, we want it to contain the secure * parameters but PCMK__XA_OP_SECURE_DIGEST to be based on the insecure ones */ list = build_parameter_list(op, metadata, ra_param_private, &secure); if (list != NULL) { digest = pcmk__digest_operation(secure); crm_xml_add(update, PCMK__XA_OP_SECURE_PARAMS, (const char *) list->str); crm_xml_add(update, PCMK__XA_OP_SECURE_DIGEST, digest); crm_trace("%s: %s, %s", op->rsc_id, digest, (const char *) list->str); g_string_free(list, TRUE); } else { crm_trace("%s: no secure parameters", op->rsc_id); } pcmk__xml_free(secure); free(digest); } /*! * \internal * \brief Create XML for a resource history entry * * \param[in] func Function name of caller * \param[in,out] parent XML to add entry to * \param[in] rsc Affected resource * \param[in,out] op Action to add an entry for (or NULL to do nothing) * \param[in] node_name Node where action occurred */ void controld_add_resource_history_xml_as(const char *func, xmlNode *parent, const lrmd_rsc_info_t *rsc, lrmd_event_data_t *op, const char *node_name) { int target_rc = 0; xmlNode *xml_op = NULL; struct ra_metadata_s *metadata = NULL; const char *caller_version = NULL; lrm_state_t *lrm_state = NULL; if (op == NULL) { return; } target_rc = rsc_op_expected_rc(op); caller_version = g_hash_table_lookup(op->params, PCMK_XA_CRM_FEATURE_SET); CRM_CHECK(caller_version != NULL, caller_version = CRM_FEATURE_SET); xml_op = pcmk__create_history_xml(parent, op, caller_version, target_rc, controld_globals.cluster->priv->node_name, func); if (xml_op == NULL) { return; } if ((rsc == NULL) || (op->params == NULL) || !crm_op_needs_metadata(rsc->standard, op->op_type)) { crm_trace("No digests needed for %s action on %s (params=%p rsc=%p)", op->op_type, op->rsc_id, op->params, rsc); return; } lrm_state = controld_get_executor_state(node_name, false); if (lrm_state == NULL) { crm_warn("Cannot calculate digests for operation " PCMK__OP_FMT " because we have no connection to executor for %s", op->rsc_id, op->op_type, op->interval_ms, node_name); return; } /* Ideally the metadata is cached, and the agent is just a fallback. * * @TODO Go through all callers and ensure they get metadata asynchronously * first. */ metadata = controld_get_rsc_metadata(lrm_state, rsc, controld_metadata_from_agent |controld_metadata_from_cache); if (metadata == NULL) { return; } crm_trace("Including additional digests for %s:%s:%s", rsc->standard, rsc->provider, rsc->type); append_restart_list(op, metadata, xml_op, caller_version); append_secure_list(op, metadata, xml_op, caller_version); return; } /*! * \internal * \brief Record an action as pending in the CIB, if appropriate * * \param[in] node_name Node where the action is pending * \param[in] rsc Resource that action is for * \param[in,out] op Pending action * * \return true if action was recorded in CIB, otherwise false */ bool controld_record_pending_op(const char *node_name, const lrmd_rsc_info_t *rsc, lrmd_event_data_t *op) { const char *record_pending = NULL; CRM_CHECK((node_name != NULL) && (rsc != NULL) && (op != NULL), return false); // Never record certain operation types as pending if ((op->op_type == NULL) || (op->params == NULL) || !controld_action_is_recordable(op->op_type)) { return false; } // Check action's PCMK_META_RECORD_PENDING meta-attribute (defaults to true) record_pending = crm_meta_value(op->params, PCMK_META_RECORD_PENDING); if ((record_pending != NULL) && !crm_is_true(record_pending)) { pcmk__warn_once(pcmk__wo_record_pending, "The " PCMK_META_RECORD_PENDING " option (for example, " "for the %s resource's %s operation) is deprecated and " "will be removed in a future release", rsc->id, op->op_type); return false; } op->call_id = -1; op->t_run = time(NULL); op->t_rcchange = op->t_run; lrmd__set_result(op, PCMK_OCF_UNKNOWN, PCMK_EXEC_PENDING, NULL); crm_debug("Recording pending %s-interval %s for %s on %s in the CIB", pcmk__readable_interval(op->interval_ms), op->op_type, op->rsc_id, node_name); controld_update_resource_history(node_name, rsc, op, 0); return true; } static void cib_rsc_callback(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data) { switch (rc) { case pcmk_ok: case -pcmk_err_diff_failed: case -pcmk_err_diff_resync: crm_trace("Resource history update completed (call=%d rc=%d)", call_id, rc); break; default: if (call_id > 0) { crm_warn("Resource history update %d failed: %s " QB_XS " rc=%d", call_id, pcmk_strerror(rc), rc); } else { crm_warn("Resource history update failed: %s " QB_XS " rc=%d", pcmk_strerror(rc), rc); } } if (call_id == pending_rsc_update) { pending_rsc_update = 0; controld_trigger_fsa(); } } /* Only successful stops, and probes that found the resource inactive, get locks * recorded in the history. This ensures the resource stays locked to the node * until it is active there again after the node comes back up. */ static bool should_preserve_lock(lrmd_event_data_t *op) { if (!pcmk_is_set(controld_globals.flags, controld_shutdown_lock_enabled)) { return false; } if (!strcmp(op->op_type, PCMK_ACTION_STOP) && (op->rc == PCMK_OCF_OK)) { return true; } if (!strcmp(op->op_type, PCMK_ACTION_MONITOR) && (op->rc == PCMK_OCF_NOT_RUNNING)) { return true; } return false; } /*! * \internal * \brief Request a CIB update * * \param[in] section Section of CIB to update * \param[in] data New XML of CIB section to update * \param[in] options CIB call options * \param[in] callback If not \c NULL, set this as the operation callback * * \return Standard Pacemaker return code * * \note If \p callback is \p cib_rsc_callback(), the CIB update's call ID is * stored in \p pending_rsc_update on success. */ int controld_update_cib(const char *section, xmlNode *data, int options, void (*callback)(xmlNode *, int, int, xmlNode *, void *)) { cib_t *cib = controld_globals.cib_conn; int cib_rc = -ENOTCONN; pcmk__assert(data != NULL); if (cib != NULL) { cib_rc = cib->cmds->modify(cib, section, data, options); if (cib_rc >= 0) { crm_debug("Submitted CIB update %d for %s section", cib_rc, section); } } if (callback == NULL) { if (cib_rc < 0) { crm_err("Failed to update CIB %s section: %s", section, pcmk_rc_str(pcmk_legacy2rc(cib_rc))); } } else { if ((cib_rc >= 0) && (callback == cib_rsc_callback)) { /* Checking for a particular callback is a little hacky, but it * didn't seem worth adding an output argument for cib_rc for just * one use case. */ pending_rsc_update = cib_rc; } fsa_register_cib_callback(cib_rc, NULL, callback); } return (cib_rc >= 0)? pcmk_rc_ok : pcmk_legacy2rc(cib_rc); } /*! * \internal * \brief Update resource history entry in CIB * * \param[in] node_name Node where action occurred * \param[in] rsc Resource that action is for * \param[in,out] op Action to record * \param[in] lock_time If nonzero, when resource was locked to node * * \note On success, the CIB update's call ID will be stored in * pending_rsc_update. */ void controld_update_resource_history(const char *node_name, const lrmd_rsc_info_t *rsc, lrmd_event_data_t *op, time_t lock_time) { xmlNode *update = NULL; xmlNode *xml = NULL; int call_opt = crmd_cib_smart_opt(); const char *node_id = NULL; const char *container = NULL; CRM_CHECK((node_name != NULL) && (op != NULL), return); if (rsc == NULL) { crm_warn("Resource %s no longer exists in the executor", op->rsc_id); controld_ack_event_directly(NULL, NULL, rsc, op, op->rsc_id); return; } // update = pcmk__xe_create(NULL, PCMK_XE_STATUS); // xml = pcmk__xe_create(update, PCMK__XE_NODE_STATE); if (controld_is_local_node(node_name)) { node_id = controld_globals.our_uuid; } else { node_id = node_name; pcmk__xe_set_bool_attr(xml, PCMK_XA_REMOTE_NODE, true); } crm_xml_add(xml, PCMK_XA_ID, node_id); crm_xml_add(xml, PCMK_XA_UNAME, node_name); crm_xml_add(xml, PCMK_XA_CRM_DEBUG_ORIGIN, __func__); // xml = pcmk__xe_create(xml, PCMK__XE_LRM); crm_xml_add(xml, PCMK_XA_ID, node_id); // xml = pcmk__xe_create(xml, PCMK__XE_LRM_RESOURCES); // xml = pcmk__xe_create(xml, PCMK__XE_LRM_RESOURCE); crm_xml_add(xml, PCMK_XA_ID, op->rsc_id); crm_xml_add(xml, PCMK_XA_CLASS, rsc->standard); crm_xml_add(xml, PCMK_XA_PROVIDER, rsc->provider); crm_xml_add(xml, PCMK_XA_TYPE, rsc->type); if (lock_time != 0) { /* Actions on a locked resource should either preserve the lock by * recording it with the action result, or clear it. */ if (!should_preserve_lock(op)) { lock_time = 0; } crm_xml_add_ll(xml, PCMK_OPT_SHUTDOWN_LOCK, (long long) lock_time); } if (op->params != NULL) { container = g_hash_table_lookup(op->params, CRM_META "_" PCMK__META_CONTAINER); if (container != NULL) { crm_trace("Resource %s is a part of container resource %s", op->rsc_id, container); crm_xml_add(xml, PCMK__META_CONTAINER, container); } } // (possibly more than one) controld_add_resource_history_xml(xml, rsc, op, node_name); /* Update CIB asynchronously. Even if it fails, the resource state should be * discovered during the next election. Worst case, the node is wrongly * fenced for running a resource it isn't. */ crm_log_xml_trace(update, __func__); controld_update_cib(PCMK_XE_STATUS, update, call_opt, cib_rsc_callback); pcmk__xml_free(update); } /*! * \internal * \brief Erase an LRM history entry from the CIB, given the operation data * * \param[in] op Operation whose history should be deleted */ void controld_delete_action_history(const lrmd_event_data_t *op) { xmlNode *xml_top = NULL; CRM_CHECK(op != NULL, return); xml_top = pcmk__xe_create(NULL, PCMK__XE_LRM_RSC_OP); crm_xml_add_int(xml_top, PCMK__XA_CALL_ID, op->call_id); crm_xml_add(xml_top, PCMK__XA_TRANSITION_KEY, op->user_data); if (op->interval_ms > 0) { char *op_id = pcmk__op_key(op->rsc_id, op->op_type, op->interval_ms); /* Avoid deleting last_failure too (if it was a result of this recurring op failing) */ crm_xml_add(xml_top, PCMK_XA_ID, op_id); free(op_id); } crm_debug("Erasing resource operation history for " PCMK__OP_FMT " (call=%d)", op->rsc_id, op->op_type, op->interval_ms, op->call_id); controld_globals.cib_conn->cmds->remove(controld_globals.cib_conn, PCMK_XE_STATUS, xml_top, cib_none); crm_log_xml_trace(xml_top, "op:cancel"); pcmk__xml_free(xml_top); } /* Define xpath to find LRM resource history entry by node and resource */ #define XPATH_HISTORY \ "/" PCMK_XE_CIB "/" PCMK_XE_STATUS \ "/" PCMK__XE_NODE_STATE "[@" PCMK_XA_UNAME "='%s']" \ "/" PCMK__XE_LRM "/" PCMK__XE_LRM_RESOURCES \ "/" PCMK__XE_LRM_RESOURCE "[@" PCMK_XA_ID "='%s']" \ "/" PCMK__XE_LRM_RSC_OP /* ... and also by operation key */ #define XPATH_HISTORY_ID XPATH_HISTORY "[@" PCMK_XA_ID "='%s']" /* ... and also by operation key and operation call ID */ #define XPATH_HISTORY_CALL XPATH_HISTORY \ "[@" PCMK_XA_ID "='%s' and @" PCMK__XA_CALL_ID "='%d']" /* ... and also by operation key and original operation key */ #define XPATH_HISTORY_ORIG XPATH_HISTORY \ "[@" PCMK_XA_ID "='%s' and @" PCMK__XA_OPERATION_KEY "='%s']" /*! * \internal * \brief Delete a last_failure resource history entry from the CIB * * \param[in] rsc_id Name of resource to clear history for * \param[in] node Name of node to clear history for * \param[in] action If specified, delete only if this was failed action * \param[in] interval_ms If \p action is specified, it has this interval */ void controld_cib_delete_last_failure(const char *rsc_id, const char *node, const char *action, guint interval_ms) { char *xpath = NULL; char *last_failure_key = NULL; CRM_CHECK((rsc_id != NULL) && (node != NULL), return); // Generate XPath to match desired entry last_failure_key = pcmk__op_key(rsc_id, "last_failure", 0); if (action == NULL) { xpath = crm_strdup_printf(XPATH_HISTORY_ID, node, rsc_id, last_failure_key); } else { char *action_key = pcmk__op_key(rsc_id, action, interval_ms); xpath = crm_strdup_printf(XPATH_HISTORY_ORIG, node, rsc_id, last_failure_key, action_key); free(action_key); } free(last_failure_key); controld_globals.cib_conn->cmds->remove(controld_globals.cib_conn, xpath, NULL, cib_xpath); free(xpath); } /*! * \internal * \brief Delete resource history entry from the CIB, given operation key * * \param[in] rsc_id Name of resource to clear history for * \param[in] node Name of node to clear history for * \param[in] key Operation key of operation to clear history for * \param[in] call_id If specified, delete entry only if it has this call ID */ void controld_delete_action_history_by_key(const char *rsc_id, const char *node, const char *key, int call_id) { char *xpath = NULL; CRM_CHECK((rsc_id != NULL) && (node != NULL) && (key != NULL), return); if (call_id > 0) { xpath = crm_strdup_printf(XPATH_HISTORY_CALL, node, rsc_id, key, call_id); } else { xpath = crm_strdup_printf(XPATH_HISTORY_ID, node, rsc_id, key); } controld_globals.cib_conn->cmds->remove(controld_globals.cib_conn, xpath, NULL, cib_xpath); free(xpath); } diff --git a/daemons/controld/controld_cib.h b/daemons/controld/controld_cib.h index b8622d5228..116db64924 100644 --- a/daemons/controld/controld_cib.h +++ b/daemons/controld/controld_cib.h @@ -1,121 +1,112 @@ /* - * Copyright 2004-2024 the Pacemaker project contributors + * Copyright 2004-2025 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #ifndef PCMK__CONTROLD_CIB__H #define PCMK__CONTROLD_CIB__H #include #include #include #include #include // cib__* #include "controld_globals.h" // controld_globals.cib_conn static inline void fsa_cib_anon_update(const char *section, xmlNode *data) { if (controld_globals.cib_conn == NULL) { crm_err("No CIB connection available"); } else { controld_globals.cib_conn->cmds->modify(controld_globals.cib_conn, section, data, cib_can_create); } } static inline void fsa_cib_anon_update_discard_reply(const char *section, xmlNode *data) { if (controld_globals.cib_conn == NULL) { crm_err("No CIB connection available"); } else { controld_globals.cib_conn->cmds->modify(controld_globals.cib_conn, section, data, cib_can_create |cib_discard_reply); } } int controld_update_cib(const char *section, xmlNode *data, int options, void (*callback)(xmlNode *, int, int, xmlNode *, void *)); unsigned int cib_op_timeout(void); -// Subsections of PCMK__XE_NODE_STATE -enum controld_section_e { - controld_section_lrm, - controld_section_lrm_unlocked, - controld_section_attrs, - controld_section_all, - controld_section_all_unlocked -}; - -void controld_node_state_deletion_strings(const char *uname, - enum controld_section_e section, - char **xpath, char **desc); -void controld_delete_node_state(const char *uname, - enum controld_section_e section, int options); +void controld_node_history_deletion_strings(const char *uname, + bool unlocked_only, + char **xpath, char **desc); +void controld_delete_node_history(const char *uname, bool unlocked_only, + int options); int controld_delete_resource_history(const char *rsc_id, const char *node, const char *user_name, int call_options); /* Convenience macro for registering a CIB callback * (assumes that data can be freed with free()) */ # define fsa_register_cib_callback(id, data, fn) do { \ cib_t *cib_conn = controld_globals.cib_conn; \ \ pcmk__assert(cib_conn != NULL); \ cib_conn->cmds->register_callback_full(cib_conn, id, cib_op_timeout(), \ FALSE, data, #fn, fn, free); \ } while(0) void controld_add_resource_history_xml_as(const char *func, xmlNode *parent, const lrmd_rsc_info_t *rsc, lrmd_event_data_t *op, const char *node_name); #define controld_add_resource_history_xml(parent, rsc, op, node_name) \ controld_add_resource_history_xml_as(__func__, (parent), (rsc), \ (op), (node_name)) bool controld_record_pending_op(const char *node_name, const lrmd_rsc_info_t *rsc, lrmd_event_data_t *op); void controld_update_resource_history(const char *node_name, const lrmd_rsc_info_t *rsc, lrmd_event_data_t *op, time_t lock_time); void controld_delete_action_history(const lrmd_event_data_t *op); void controld_cib_delete_last_failure(const char *rsc_id, const char *node, const char *action, guint interval_ms); void controld_delete_action_history_by_key(const char *rsc_id, const char *node, const char *key, int call_id); void controld_disconnect_cib_manager(void); int crmd_cib_smart_opt(void); /*! * \internal * \brief Check whether an action type should be recorded in the CIB * * \param[in] action Action type * * \return true if action should be recorded, false otherwise */ static inline bool controld_action_is_recordable(const char *action) { return !pcmk__str_any_of(action, PCMK_ACTION_CANCEL, PCMK_ACTION_DELETE, PCMK_ACTION_NOTIFY, PCMK_ACTION_META_DATA, NULL); } #endif // PCMK__CONTROLD_CIB__H diff --git a/daemons/controld/controld_execd.c b/daemons/controld/controld_execd.c index 2ec689394d..977acf01ab 100644 --- a/daemons/controld/controld_execd.c +++ b/daemons/controld/controld_execd.c @@ -1,2406 +1,2405 @@ /* * Copyright 2004-2025 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include // lrmd_event_data_t, lrmd_rsc_info_t, etc. #include #include #include #include #include #define START_DELAY_THRESHOLD 5 * 60 * 1000 #define MAX_LRM_REG_FAILS 30 struct delete_event_s { int rc; const char *rsc; lrm_state_t *lrm_state; }; static gboolean is_rsc_active(lrm_state_t * lrm_state, const char *rsc_id); static gboolean build_active_RAs(lrm_state_t * lrm_state, xmlNode * rsc_list); static gboolean stop_recurring_actions(gpointer key, gpointer value, gpointer user_data); static lrmd_event_data_t *construct_op(const lrm_state_t *lrm_state, const xmlNode *rsc_op, const char *rsc_id, const char *operation); static void do_lrm_rsc_op(lrm_state_t *lrm_state, lrmd_rsc_info_t *rsc, xmlNode *msg, struct ra_metadata_s *md); static gboolean lrm_state_verify_stopped(lrm_state_t * lrm_state, enum crmd_fsa_state cur_state, int log_level); static void lrm_connection_destroy(void) { if (pcmk_is_set(controld_globals.fsa_input_register, R_LRM_CONNECTED)) { crm_crit("Lost connection to local executor"); register_fsa_input(C_FSA_INTERNAL, I_ERROR, NULL); controld_clear_fsa_input_flags(R_LRM_CONNECTED); } } static char * make_stop_id(const char *rsc, int call_id) { return crm_strdup_printf("%s:%d", rsc, call_id); } static void copy_instance_keys(gpointer key, gpointer value, gpointer user_data) { if (strstr(key, CRM_META "_") == NULL) { pcmk__insert_dup(user_data, (const char *) key, (const char *) value); } } static void copy_meta_keys(gpointer key, gpointer value, gpointer user_data) { if (strstr(key, CRM_META "_") != NULL) { pcmk__insert_dup(user_data, (const char *) key, (const char *) value); } } /*! * \internal * \brief Remove a recurring operation from a resource's history * * \param[in,out] history Resource history to modify * \param[in] op Operation to remove * * \return TRUE if the operation was found and removed, FALSE otherwise */ static gboolean history_remove_recurring_op(rsc_history_t *history, const lrmd_event_data_t *op) { GList *iter; for (iter = history->recurring_op_list; iter != NULL; iter = iter->next) { lrmd_event_data_t *existing = iter->data; if ((op->interval_ms == existing->interval_ms) && pcmk__str_eq(op->rsc_id, existing->rsc_id, pcmk__str_none) && pcmk__str_eq(op->op_type, existing->op_type, pcmk__str_casei)) { history->recurring_op_list = g_list_delete_link(history->recurring_op_list, iter); lrmd_free_event(existing); return TRUE; } } return FALSE; } /*! * \internal * \brief Free all recurring operations in resource history * * \param[in,out] history Resource history to modify */ static void history_free_recurring_ops(rsc_history_t *history) { GList *iter; for (iter = history->recurring_op_list; iter != NULL; iter = iter->next) { lrmd_free_event(iter->data); } g_list_free(history->recurring_op_list); history->recurring_op_list = NULL; } /*! * \internal * \brief Free resource history * * \param[in,out] history Resource history to free */ void history_free(gpointer data) { rsc_history_t *history = (rsc_history_t*)data; if (history->stop_params) { g_hash_table_destroy(history->stop_params); } /* Don't need to free history->rsc.id because it's set to history->id */ free(history->rsc.type); free(history->rsc.standard); free(history->rsc.provider); lrmd_free_event(history->failed); lrmd_free_event(history->last); free(history->id); history_free_recurring_ops(history); free(history); } static void update_history_cache(lrm_state_t * lrm_state, lrmd_rsc_info_t * rsc, lrmd_event_data_t * op) { int target_rc = 0; rsc_history_t *entry = NULL; if (op->rsc_deleted) { crm_debug("Purged history for '%s' after %s", op->rsc_id, op->op_type); controld_delete_resource_history(op->rsc_id, lrm_state->node_name, NULL, crmd_cib_smart_opt()); return; } if (pcmk__str_eq(op->op_type, PCMK_ACTION_NOTIFY, pcmk__str_casei)) { return; } crm_debug("Updating history for '%s' with %s op", op->rsc_id, op->op_type); entry = g_hash_table_lookup(lrm_state->resource_history, op->rsc_id); if (entry == NULL && rsc) { entry = pcmk__assert_alloc(1, sizeof(rsc_history_t)); entry->id = pcmk__str_copy(op->rsc_id); g_hash_table_insert(lrm_state->resource_history, entry->id, entry); entry->rsc.id = entry->id; entry->rsc.type = pcmk__str_copy(rsc->type); entry->rsc.standard = pcmk__str_copy(rsc->standard); entry->rsc.provider = pcmk__str_copy(rsc->provider); } else if (entry == NULL) { crm_info("Resource %s no longer exists, not updating cache", op->rsc_id); return; } entry->last_callid = op->call_id; target_rc = rsc_op_expected_rc(op); if (op->op_status == PCMK_EXEC_CANCELLED) { if (op->interval_ms > 0) { crm_trace("Removing cancelled recurring op: " PCMK__OP_FMT, op->rsc_id, op->op_type, op->interval_ms); history_remove_recurring_op(entry, op); return; } else { crm_trace("Skipping " PCMK__OP_FMT " rc=%d, status=%d", op->rsc_id, op->op_type, op->interval_ms, op->rc, op->op_status); } } else if (did_rsc_op_fail(op, target_rc)) { /* Store failed monitors here, otherwise the block below will cause them * to be forgotten when a stop happens. */ if (entry->failed) { lrmd_free_event(entry->failed); } entry->failed = lrmd_copy_event(op); } else if (op->interval_ms == 0) { if (entry->last) { lrmd_free_event(entry->last); } entry->last = lrmd_copy_event(op); if (op->params && pcmk__strcase_any_of(op->op_type, PCMK_ACTION_START, PCMK_ACTION_RELOAD, PCMK_ACTION_RELOAD_AGENT, PCMK_ACTION_MONITOR, NULL)) { if (entry->stop_params) { g_hash_table_destroy(entry->stop_params); } entry->stop_params = pcmk__strkey_table(free, free); g_hash_table_foreach(op->params, copy_instance_keys, entry->stop_params); } } if (op->interval_ms > 0) { /* Ensure there are no duplicates */ history_remove_recurring_op(entry, op); crm_trace("Adding recurring op: " PCMK__OP_FMT, op->rsc_id, op->op_type, op->interval_ms); entry->recurring_op_list = g_list_prepend(entry->recurring_op_list, lrmd_copy_event(op)); } else if ((entry->recurring_op_list != NULL) && !pcmk__str_eq(op->op_type, PCMK_ACTION_MONITOR, pcmk__str_casei)) { crm_trace("Dropping %d recurring ops because of: " PCMK__OP_FMT, g_list_length(entry->recurring_op_list), op->rsc_id, op->op_type, op->interval_ms); history_free_recurring_ops(entry); } } /*! * \internal * \brief Send a direct OK ack for a resource task * * \param[in] lrm_state LRM connection * \param[in] input Input message being ack'ed * \param[in] rsc_id ID of affected resource * \param[in] rsc Affected resource (if available) * \param[in] task Operation task being ack'ed * \param[in] ack_host Name of host to send ack to * \param[in] ack_sys IPC system name to ack */ static void send_task_ok_ack(const lrm_state_t *lrm_state, const ha_msg_input_t *input, const char *rsc_id, const lrmd_rsc_info_t *rsc, const char *task, const char *ack_host, const char *ack_sys) { lrmd_event_data_t *op = construct_op(lrm_state, input->xml, rsc_id, task); lrmd__set_result(op, PCMK_OCF_OK, PCMK_EXEC_DONE, NULL); controld_ack_event_directly(ack_host, ack_sys, rsc, op, rsc_id); lrmd_free_event(op); } static inline const char * op_node_name(lrmd_event_data_t *op) { return pcmk__s(op->remote_nodename, controld_globals.cluster->priv->node_name); } void lrm_op_callback(lrmd_event_data_t * op) { CRM_CHECK(op != NULL, return); switch (op->type) { case lrmd_event_disconnect: if (op->remote_nodename == NULL) { /* If this is the local executor IPC connection, set the right * bits in the controller when the connection goes down. */ lrm_connection_destroy(); } break; case lrmd_event_exec_complete: { lrm_state_t *lrm_state = controld_get_executor_state(op_node_name(op), false); pcmk__assert(lrm_state != NULL); process_lrm_event(lrm_state, op, NULL, NULL); } break; default: break; } } static void try_local_executor_connect(long long action, fsa_data_t *msg_data, lrm_state_t *lrm_state) { int rc = pcmk_rc_ok; crm_debug("Connecting to the local executor"); // If we can connect, great rc = controld_connect_local_executor(lrm_state); if (rc == pcmk_rc_ok) { controld_set_fsa_input_flags(R_LRM_CONNECTED); crm_info("Connection to the local executor established"); return; } // Otherwise, if we can try again, set a timer to do so if (lrm_state->num_lrm_register_fails < MAX_LRM_REG_FAILS) { crm_warn("Failed to connect to the local executor %d time%s " "(%d max): %s", lrm_state->num_lrm_register_fails, pcmk__plural_s(lrm_state->num_lrm_register_fails), MAX_LRM_REG_FAILS, pcmk_rc_str(rc)); controld_start_wait_timer(); crmd_fsa_stall(FALSE); return; } // Otherwise give up crm_err("Failed to connect to the executor the max allowed " "%d time%s: %s", lrm_state->num_lrm_register_fails, pcmk__plural_s(lrm_state->num_lrm_register_fails), pcmk_rc_str(rc)); register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); } /* A_LRM_CONNECT */ void do_lrm_control(long long action, enum crmd_fsa_cause cause, enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data) { /* This only pertains to local executor connections. Remote connections are * handled as resources within the scheduler. Connecting and disconnecting * from remote executor instances is handled differently. */ lrm_state_t *lrm_state = NULL; if (controld_globals.cluster->priv->node_name == NULL) { return; // Shouldn't be possible } lrm_state = controld_get_executor_state(NULL, true); if (lrm_state == NULL) { register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); return; } if (action & A_LRM_DISCONNECT) { if (lrm_state_verify_stopped(lrm_state, cur_state, LOG_INFO) == FALSE) { if (action == A_LRM_DISCONNECT) { crmd_fsa_stall(FALSE); return; } } controld_clear_fsa_input_flags(R_LRM_CONNECTED); lrm_state_disconnect(lrm_state); lrm_state_reset_tables(lrm_state, FALSE); } if (action & A_LRM_CONNECT) { try_local_executor_connect(action, msg_data, lrm_state); } if (action & ~(A_LRM_CONNECT | A_LRM_DISCONNECT)) { crm_err("Unexpected action %s in %s", fsa_action2string(action), __func__); } } static gboolean lrm_state_verify_stopped(lrm_state_t * lrm_state, enum crmd_fsa_state cur_state, int log_level) { int counter = 0; gboolean rc = TRUE; const char *when = "lrm disconnect"; GHashTableIter gIter; const char *key = NULL; rsc_history_t *entry = NULL; active_op_t *pending = NULL; crm_debug("Checking for active resources before exit"); if (cur_state == S_TERMINATE) { log_level = LOG_ERR; when = "shutdown"; } else if (pcmk_is_set(controld_globals.fsa_input_register, R_SHUTDOWN)) { when = "shutdown... waiting"; } if ((lrm_state->active_ops != NULL) && lrm_state_is_connected(lrm_state)) { guint removed = g_hash_table_foreach_remove(lrm_state->active_ops, stop_recurring_actions, lrm_state); guint nremaining = g_hash_table_size(lrm_state->active_ops); if (removed || nremaining) { crm_notice("Stopped %u recurring operation%s at %s (%u remaining)", removed, pcmk__plural_s(removed), when, nremaining); } } if (lrm_state->active_ops != NULL) { g_hash_table_iter_init(&gIter, lrm_state->active_ops); while (g_hash_table_iter_next(&gIter, NULL, (void **)&pending)) { /* Ignore recurring actions in the shutdown calculations */ if (pending->interval_ms == 0) { counter++; } } } if (counter > 0) { do_crm_log(log_level, "%d pending executor operation%s at %s", counter, pcmk__plural_s(counter), when); if ((cur_state == S_TERMINATE) || !pcmk_is_set(controld_globals.fsa_input_register, R_SENT_RSC_STOP)) { g_hash_table_iter_init(&gIter, lrm_state->active_ops); while (g_hash_table_iter_next(&gIter, (gpointer*)&key, (gpointer*)&pending)) { do_crm_log(log_level, "Pending action: %s (%s)", key, pending->op_key); } } else { rc = FALSE; } return rc; } if (lrm_state->resource_history == NULL) { return rc; } if (pcmk_is_set(controld_globals.fsa_input_register, R_SHUTDOWN)) { /* At this point we're not waiting, we're just shutting down */ when = "shutdown"; } counter = 0; g_hash_table_iter_init(&gIter, lrm_state->resource_history); while (g_hash_table_iter_next(&gIter, NULL, (gpointer*)&entry)) { if (is_rsc_active(lrm_state, entry->id) == FALSE) { continue; } counter++; if (log_level == LOG_ERR) { crm_info("Found %s active at %s", entry->id, when); } else { crm_trace("Found %s active at %s", entry->id, when); } if (lrm_state->active_ops != NULL) { GHashTableIter hIter; g_hash_table_iter_init(&hIter, lrm_state->active_ops); while (g_hash_table_iter_next(&hIter, (gpointer*)&key, (gpointer*)&pending)) { if (pcmk__str_eq(entry->id, pending->rsc_id, pcmk__str_none)) { crm_notice("%sction %s (%s) incomplete at %s", pending->interval_ms == 0 ? "A" : "Recurring a", key, pending->op_key, when); } } } } if (counter) { crm_err("%d resource%s active at %s", counter, (counter == 1)? " was" : "s were", when); } return rc; } static gboolean is_rsc_active(lrm_state_t * lrm_state, const char *rsc_id) { rsc_history_t *entry = NULL; entry = g_hash_table_lookup(lrm_state->resource_history, rsc_id); if (entry == NULL || entry->last == NULL) { return FALSE; } crm_trace("Processing %s: %s.%d=%d", rsc_id, entry->last->op_type, entry->last->interval_ms, entry->last->rc); if ((entry->last->rc == PCMK_OCF_OK) && pcmk__str_eq(entry->last->op_type, PCMK_ACTION_STOP, pcmk__str_casei)) { return FALSE; } else if (entry->last->rc == PCMK_OCF_OK && pcmk__str_eq(entry->last->op_type, PCMK_ACTION_MIGRATE_TO, pcmk__str_casei)) { // A stricter check is too complex ... leave that to the scheduler return FALSE; } else if (entry->last->rc == PCMK_OCF_NOT_RUNNING) { return FALSE; } else if ((entry->last->interval_ms == 0) && (entry->last->rc == PCMK_OCF_NOT_CONFIGURED)) { /* Badly configured resources can't be reliably stopped */ return FALSE; } return TRUE; } static gboolean build_active_RAs(lrm_state_t * lrm_state, xmlNode * rsc_list) { GHashTableIter iter; rsc_history_t *entry = NULL; g_hash_table_iter_init(&iter, lrm_state->resource_history); while (g_hash_table_iter_next(&iter, NULL, (void **)&entry)) { GList *gIter = NULL; xmlNode *xml_rsc = pcmk__xe_create(rsc_list, PCMK__XE_LRM_RESOURCE); crm_xml_add(xml_rsc, PCMK_XA_ID, entry->id); crm_xml_add(xml_rsc, PCMK_XA_TYPE, entry->rsc.type); crm_xml_add(xml_rsc, PCMK_XA_CLASS, entry->rsc.standard); crm_xml_add(xml_rsc, PCMK_XA_PROVIDER, entry->rsc.provider); if (entry->last && entry->last->params) { static const char *name = CRM_META "_" PCMK__META_CONTAINER; const char *container = g_hash_table_lookup(entry->last->params, name); if (container) { crm_trace("Resource %s is a part of container resource %s", entry->id, container); crm_xml_add(xml_rsc, PCMK__META_CONTAINER, container); } } controld_add_resource_history_xml(xml_rsc, &(entry->rsc), entry->failed, lrm_state->node_name); controld_add_resource_history_xml(xml_rsc, &(entry->rsc), entry->last, lrm_state->node_name); for (gIter = entry->recurring_op_list; gIter != NULL; gIter = gIter->next) { controld_add_resource_history_xml(xml_rsc, &(entry->rsc), gIter->data, lrm_state->node_name); } } return FALSE; } xmlNode * controld_query_executor_state(void) { // @TODO Ensure all callers handle NULL returns xmlNode *xml_state = NULL; xmlNode *xml_data = NULL; xmlNode *rsc_list = NULL; pcmk__node_status_t *peer = NULL; lrm_state_t *lrm_state = controld_get_executor_state(NULL, false); if (!lrm_state) { crm_err("Could not get executor state for local node"); return NULL; } peer = pcmk__get_node(0, lrm_state->node_name, NULL, pcmk__node_search_any); CRM_CHECK(peer != NULL, return NULL); xml_state = create_node_state_update(peer, controld_node_update_cluster |controld_node_update_peer, NULL, __func__); if (xml_state == NULL) { return NULL; } xml_data = pcmk__xe_create(xml_state, PCMK__XE_LRM); crm_xml_add(xml_data, PCMK_XA_ID, peer->xml_id); rsc_list = pcmk__xe_create(xml_data, PCMK__XE_LRM_RESOURCES); // Build a list of active (not necessarily running) resources build_active_RAs(lrm_state, rsc_list); crm_log_xml_trace(xml_state, "Current executor state"); return xml_state; } /*! * \internal * \brief Map standard Pacemaker return code to operation status and OCF code * * \param[out] event Executor event whose status and return code should be set * \param[in] rc Standard Pacemaker return code */ void controld_rc2event(lrmd_event_data_t *event, int rc) { /* This is called for cleanup requests from controller peers/clients, not * for resource actions, so no exit reason is needed. */ switch (rc) { case pcmk_rc_ok: lrmd__set_result(event, PCMK_OCF_OK, PCMK_EXEC_DONE, NULL); break; case EACCES: lrmd__set_result(event, PCMK_OCF_INSUFFICIENT_PRIV, PCMK_EXEC_ERROR, NULL); break; default: lrmd__set_result(event, PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_ERROR, NULL); break; } } /*! * \internal * \brief Trigger a new transition after CIB status was deleted * * If a CIB status delete was not expected (as part of the transition graph), * trigger a new transition by updating the (arbitrary) "last-lrm-refresh" * cluster property. * * \param[in] from_sys IPC name that requested the delete * \param[in] rsc_id Resource whose status was deleted (for logging only) */ void controld_trigger_delete_refresh(const char *from_sys, const char *rsc_id) { if (!pcmk__str_eq(from_sys, CRM_SYSTEM_TENGINE, pcmk__str_casei)) { char *now_s = crm_strdup_printf("%lld", (long long) time(NULL)); crm_debug("Triggering a refresh after %s cleaned %s", from_sys, rsc_id); cib__update_node_attr(controld_globals.logger_out, controld_globals.cib_conn, cib_none, PCMK_XE_CRM_CONFIG, NULL, NULL, NULL, NULL, "last-lrm-refresh", now_s, NULL, NULL); free(now_s); } } static void notify_deleted(lrm_state_t * lrm_state, ha_msg_input_t * input, const char *rsc_id, int rc) { lrmd_event_data_t *op = NULL; const char *from_sys = crm_element_value(input->msg, PCMK__XA_CRM_SYS_FROM); const char *from_host = crm_element_value(input->msg, PCMK__XA_SRC); crm_info("Notifying %s on %s that %s was%s deleted", from_sys, (from_host? from_host : "localhost"), rsc_id, ((rc == pcmk_ok)? "" : " not")); op = construct_op(lrm_state, input->xml, rsc_id, PCMK_ACTION_DELETE); controld_rc2event(op, pcmk_legacy2rc(rc)); controld_ack_event_directly(from_host, from_sys, NULL, op, rsc_id); lrmd_free_event(op); controld_trigger_delete_refresh(from_sys, rsc_id); } static gboolean lrm_remove_deleted_rsc(gpointer key, gpointer value, gpointer user_data) { struct delete_event_s *event = user_data; struct pending_deletion_op_s *op = value; if (pcmk__str_eq(event->rsc, op->rsc, pcmk__str_none)) { notify_deleted(event->lrm_state, op->input, event->rsc, event->rc); return TRUE; } return FALSE; } static gboolean lrm_remove_deleted_op(gpointer key, gpointer value, gpointer user_data) { const char *rsc = user_data; active_op_t *pending = value; if (pcmk__str_eq(rsc, pending->rsc_id, pcmk__str_none)) { crm_info("Removing op %s:%d for deleted resource %s", pending->op_key, pending->call_id, rsc); return TRUE; } return FALSE; } static void delete_rsc_entry(lrm_state_t *lrm_state, ha_msg_input_t *input, const char *rsc_id, GHashTableIter *rsc_iter, int rc, const char *user_name, bool from_cib) { struct delete_event_s event; CRM_CHECK(rsc_id != NULL, return); if (rc == pcmk_ok) { char *rsc_id_copy = pcmk__str_copy(rsc_id); if (rsc_iter) { g_hash_table_iter_remove(rsc_iter); } else { g_hash_table_remove(lrm_state->resource_history, rsc_id_copy); } if (from_cib) { controld_delete_resource_history(rsc_id_copy, lrm_state->node_name, user_name, crmd_cib_smart_opt()); } g_hash_table_foreach_remove(lrm_state->active_ops, lrm_remove_deleted_op, rsc_id_copy); free(rsc_id_copy); } if (input) { notify_deleted(lrm_state, input, rsc_id, rc); } event.rc = rc; event.rsc = rsc_id; event.lrm_state = lrm_state; g_hash_table_foreach_remove(lrm_state->deletion_ops, lrm_remove_deleted_rsc, &event); } static inline gboolean last_failed_matches_op(rsc_history_t *entry, const char *op, guint interval_ms) { if (entry == NULL) { return FALSE; } if (op == NULL) { return TRUE; } return (pcmk__str_eq(op, entry->failed->op_type, pcmk__str_casei) && (interval_ms == entry->failed->interval_ms)); } /*! * \internal * \brief Clear a resource's last failure * * Erase a resource's last failure on a particular node from both the * LRM resource history in the CIB, and the resource history remembered * for the LRM state. * * \param[in] rsc_id Resource name * \param[in] node_name Node name * \param[in] operation If specified, only clear if matching this operation * \param[in] interval_ms If operation is specified, it has this interval */ void lrm_clear_last_failure(const char *rsc_id, const char *node_name, const char *operation, guint interval_ms) { lrm_state_t *lrm_state = controld_get_executor_state(node_name, false); if (lrm_state == NULL) { return; } if (lrm_state->resource_history != NULL) { rsc_history_t *entry = g_hash_table_lookup(lrm_state->resource_history, rsc_id); if (last_failed_matches_op(entry, operation, interval_ms)) { lrmd_free_event(entry->failed); entry->failed = NULL; } } } /* Returns: gboolean - cancellation is in progress */ static gboolean cancel_op(lrm_state_t * lrm_state, const char *rsc_id, const char *key, int op, gboolean remove) { int rc = pcmk_ok; char *local_key = NULL; active_op_t *pending = NULL; CRM_CHECK(op != 0, return FALSE); CRM_CHECK(rsc_id != NULL, return FALSE); if (key == NULL) { local_key = make_stop_id(rsc_id, op); key = local_key; } pending = g_hash_table_lookup(lrm_state->active_ops, key); if (pending) { if (remove && !pcmk_is_set(pending->flags, active_op_remove)) { controld_set_active_op_flags(pending, active_op_remove); crm_debug("Scheduling %s for removal", key); } if (pcmk_is_set(pending->flags, active_op_cancelled)) { crm_debug("Operation %s already cancelled", key); free(local_key); return FALSE; } controld_set_active_op_flags(pending, active_op_cancelled); } else { crm_info("No pending op found for %s", key); free(local_key); return FALSE; } crm_debug("Cancelling op %d for %s (%s)", op, rsc_id, key); rc = lrm_state_cancel(lrm_state, pending->rsc_id, pending->op_type, pending->interval_ms); if (rc == pcmk_ok) { crm_debug("Op %d for %s (%s): cancelled", op, rsc_id, key); free(local_key); return TRUE; } crm_debug("Op %d for %s (%s): Nothing to cancel", op, rsc_id, key); /* The caller needs to make sure the entry is * removed from the active operations list * * Usually by returning TRUE inside the worker function * supplied to g_hash_table_foreach_remove() * * Not removing the entry from active operations will block * the node from shutting down */ free(local_key); return FALSE; } struct cancel_data { gboolean done; gboolean remove; const char *key; lrmd_rsc_info_t *rsc; lrm_state_t *lrm_state; }; static gboolean cancel_action_by_key(gpointer key, gpointer value, gpointer user_data) { gboolean remove = FALSE; struct cancel_data *data = user_data; active_op_t *op = value; if (pcmk__str_eq(op->op_key, data->key, pcmk__str_none)) { data->done = TRUE; remove = !cancel_op(data->lrm_state, data->rsc->id, key, op->call_id, data->remove); } return remove; } static gboolean cancel_op_key(lrm_state_t * lrm_state, lrmd_rsc_info_t * rsc, const char *key, gboolean remove) { guint removed = 0; struct cancel_data data; CRM_CHECK(rsc != NULL, return FALSE); CRM_CHECK(key != NULL, return FALSE); data.key = key; data.rsc = rsc; data.done = FALSE; data.remove = remove; data.lrm_state = lrm_state; removed = g_hash_table_foreach_remove(lrm_state->active_ops, cancel_action_by_key, &data); crm_trace("Removed %u op cache entries, new size: %u", removed, g_hash_table_size(lrm_state->active_ops)); return data.done; } /*! * \internal * \brief Retrieve resource information from LRM * * \param[in,out] lrm_state Executor connection state to use * \param[in] rsc_xml XML containing resource configuration * \param[in] do_create If true, register resource if not already * \param[out] rsc_info Where to store information obtained from executor * * \retval pcmk_ok Success (and rsc_info holds newly allocated result) * \retval -EINVAL Required information is missing from arguments * \retval -ENOTCONN No active connection to LRM * \retval -ENODEV Resource not found * \retval -errno Error communicating with executor when registering resource * * \note Caller is responsible for freeing result on success. */ static int get_lrm_resource(lrm_state_t *lrm_state, const xmlNode *rsc_xml, gboolean do_create, lrmd_rsc_info_t **rsc_info) { const char *id = pcmk__xe_id(rsc_xml); CRM_CHECK(lrm_state && rsc_xml && rsc_info, return -EINVAL); CRM_CHECK(id, return -EINVAL); if (lrm_state_is_connected(lrm_state) == FALSE) { return -ENOTCONN; } crm_trace("Retrieving resource information for %s from the executor", id); *rsc_info = lrm_state_get_rsc_info(lrm_state, id, 0); // If resource isn't known by ID, try clone name, if provided if (!*rsc_info) { const char *long_id = crm_element_value(rsc_xml, PCMK__XA_LONG_ID); if (long_id) { *rsc_info = lrm_state_get_rsc_info(lrm_state, long_id, 0); } } if ((*rsc_info == NULL) && do_create) { const char *class = crm_element_value(rsc_xml, PCMK_XA_CLASS); const char *provider = crm_element_value(rsc_xml, PCMK_XA_PROVIDER); const char *type = crm_element_value(rsc_xml, PCMK_XA_TYPE); int rc; crm_trace("Registering resource %s with the executor", id); rc = lrm_state_register_rsc(lrm_state, id, class, provider, type, lrmd_opt_drop_recurring); if (rc != pcmk_ok) { fsa_data_t *msg_data = NULL; crm_err("Could not register resource %s with the executor on %s: %s " QB_XS " rc=%d", id, lrm_state->node_name, pcmk_strerror(rc), rc); /* Register this as an internal error if this involves the local * executor. Otherwise, we're likely dealing with an unresponsive * remote node, which is not an FSA failure. */ if (lrm_state_is_local(lrm_state) == TRUE) { register_fsa_error(C_FSA_INTERNAL, I_FAIL, NULL); } return rc; } *rsc_info = lrm_state_get_rsc_info(lrm_state, id, 0); } return *rsc_info? pcmk_ok : -ENODEV; } static void delete_resource(lrm_state_t *lrm_state, const char *id, lrmd_rsc_info_t *rsc, GHashTableIter *iter, const char *sys, const char *user, ha_msg_input_t *request, bool unregister, bool from_cib) { int rc = pcmk_ok; crm_info("Removing resource %s from executor for %s%s%s", id, sys, (user? " as " : ""), (user? user : "")); if (rsc && unregister) { rc = lrm_state_unregister_rsc(lrm_state, id, 0); } if (rc == pcmk_ok) { crm_trace("Resource %s deleted from executor", id); } else if (rc == -EINPROGRESS) { crm_info("Deletion of resource '%s' from executor is pending", id); if (request) { struct pending_deletion_op_s *op = NULL; char *ref = crm_element_value_copy(request->msg, PCMK_XA_REFERENCE); op = pcmk__assert_alloc(1, sizeof(struct pending_deletion_op_s)); op->rsc = pcmk__str_copy(rsc->id); op->input = copy_ha_msg_input(request); g_hash_table_insert(lrm_state->deletion_ops, ref, op); } return; } else { crm_warn("Could not delete '%s' from executor for %s%s%s: %s " QB_XS " rc=%d", id, sys, (user? " as " : ""), (user? user : ""), pcmk_strerror(rc), rc); } delete_rsc_entry(lrm_state, request, id, iter, rc, user, from_cib); } static int get_fake_call_id(lrm_state_t *lrm_state, const char *rsc_id) { int call_id = 999999999; rsc_history_t *entry = NULL; if(lrm_state) { entry = g_hash_table_lookup(lrm_state->resource_history, rsc_id); } /* Make sure the call id is greater than the last successful operation, * otherwise the failure will not result in a possible recovery of the resource * as it could appear the failure occurred before the successful start */ if (entry) { call_id = entry->last_callid + 1; } if (call_id < 0) { call_id = 1; } return call_id; } static void fake_op_status(lrm_state_t *lrm_state, lrmd_event_data_t *op, int op_status, enum ocf_exitcode op_exitcode, const char *exit_reason) { op->call_id = get_fake_call_id(lrm_state, op->rsc_id); op->t_run = time(NULL); op->t_rcchange = op->t_run; lrmd__set_result(op, op_exitcode, op_status, exit_reason); } static void force_reprobe(lrm_state_t *lrm_state, const char *from_sys, const char *from_host, const char *user_name, gboolean is_remote_node, bool reprobe_all_nodes) { GHashTableIter gIter; rsc_history_t *entry = NULL; crm_info("Clearing resource history on node %s", lrm_state->node_name); g_hash_table_iter_init(&gIter, lrm_state->resource_history); while (g_hash_table_iter_next(&gIter, NULL, (void **)&entry)) { /* only unregister the resource during a reprobe if it is not a remote connection * resource. otherwise unregistering the connection will terminate remote-node * membership */ bool unregister = true; if (is_remote_lrmd_ra(NULL, NULL, entry->id)) { unregister = false; if (reprobe_all_nodes) { lrm_state_t *remote_lrm_state = controld_get_executor_state(entry->id, false); if (remote_lrm_state != NULL) { /* If reprobing all nodes, be sure to reprobe the remote * node before clearing its connection resource */ force_reprobe(remote_lrm_state, from_sys, from_host, user_name, TRUE, reprobe_all_nodes); } } } /* Don't delete from the CIB, since we'll delete the whole node's LRM * state from the CIB soon */ delete_resource(lrm_state, entry->id, &entry->rsc, &gIter, from_sys, user_name, NULL, unregister, false); } /* Now delete the copy in the CIB */ - controld_delete_node_state(lrm_state->node_name, controld_section_lrm, - cib_none); + controld_delete_node_history(lrm_state->node_name, false, cib_none); } /*! * \internal * \brief Fail a requested action without actually executing it * * For an action that can't be executed, process it similarly to an actual * execution result, with specified error status (except for notify actions, * which will always be treated as successful). * * \param[in,out] lrm_state Executor connection that action is for * \param[in] action Action XML from request * \param[in] rc Desired return code to use * \param[in] op_status Desired operation status to use * \param[in] exit_reason Human-friendly detail, if error */ static void synthesize_lrmd_failure(lrm_state_t *lrm_state, const xmlNode *action, int op_status, enum ocf_exitcode rc, const char *exit_reason) { lrmd_event_data_t *op = NULL; const char *operation = crm_element_value(action, PCMK_XA_OPERATION); const char *target_node = crm_element_value(action, PCMK__META_ON_NODE); xmlNode *xml_rsc = pcmk__xe_first_child(action, PCMK_XE_PRIMITIVE, NULL, NULL); if ((xml_rsc == NULL) || (pcmk__xe_id(xml_rsc) == NULL)) { /* @TODO Should we do something else, like direct ack? */ crm_info("Can't fake %s failure (%d) on %s without resource configuration", crm_element_value(action, PCMK__XA_OPERATION_KEY), rc, target_node); return; } else if(operation == NULL) { /* This probably came from crm_resource -C, nothing to do */ crm_info("Can't fake %s failure (%d) on %s without operation", pcmk__xe_id(xml_rsc), rc, target_node); return; } op = construct_op(lrm_state, action, pcmk__xe_id(xml_rsc), operation); if (pcmk__str_eq(operation, PCMK_ACTION_NOTIFY, pcmk__str_casei)) { // Notifications can't fail fake_op_status(lrm_state, op, PCMK_EXEC_DONE, PCMK_OCF_OK, NULL); } else { fake_op_status(lrm_state, op, op_status, rc, exit_reason); } crm_info("Faking " PCMK__OP_FMT " result (%d) on %s", op->rsc_id, op->op_type, op->interval_ms, op->rc, target_node); // Process the result as if it came from the LRM process_lrm_event(lrm_state, op, NULL, action); lrmd_free_event(op); } /*! * \internal * \brief Get target of an LRM operation (replacing \p NULL with local node * name) * * \param[in] xml LRM operation data XML * * \return LRM operation target node name (local node or Pacemaker Remote node) */ static const char * lrm_op_target(const xmlNode *xml) { const char *target = NULL; if (xml) { target = crm_element_value(xml, PCMK__META_ON_NODE); } if (target == NULL) { target = controld_globals.cluster->priv->node_name; } return target; } static void fail_lrm_resource(xmlNode *xml, lrm_state_t *lrm_state, const char *user_name, const char *from_host, const char *from_sys) { lrmd_event_data_t *op = NULL; lrmd_rsc_info_t *rsc = NULL; xmlNode *xml_rsc = pcmk__xe_first_child(xml, PCMK_XE_PRIMITIVE, NULL, NULL); CRM_CHECK(xml_rsc != NULL, return); /* The executor simply executes operations and reports the results, without * any concept of success or failure, so to fail a resource, we must fake * what a failure looks like. * * To do this, we create a fake executor operation event for the resource, * and pass that event to the executor client callback so it will be * processed as if it came from the executor. */ op = construct_op(lrm_state, xml, pcmk__xe_id(xml_rsc), "asyncmon"); free((char*) op->user_data); op->user_data = NULL; op->interval_ms = 0; if (user_name && !pcmk__is_privileged(user_name)) { crm_err("%s does not have permission to fail %s", user_name, pcmk__xe_id(xml_rsc)); fake_op_status(lrm_state, op, PCMK_EXEC_ERROR, PCMK_OCF_INSUFFICIENT_PRIV, "Unprivileged user cannot fail resources"); controld_ack_event_directly(from_host, from_sys, NULL, op, pcmk__xe_id(xml_rsc)); lrmd_free_event(op); return; } if (get_lrm_resource(lrm_state, xml_rsc, TRUE, &rsc) == pcmk_ok) { crm_info("Failing resource %s...", rsc->id); fake_op_status(lrm_state, op, PCMK_EXEC_DONE, PCMK_OCF_UNKNOWN_ERROR, "Simulated failure"); process_lrm_event(lrm_state, op, NULL, xml); op->rc = PCMK_OCF_OK; // The request to fail the resource succeeded lrmd_free_rsc_info(rsc); } else { crm_info("Cannot find/create resource in order to fail it..."); crm_log_xml_warn(xml, "bad input"); fake_op_status(lrm_state, op, PCMK_EXEC_ERROR, PCMK_OCF_UNKNOWN_ERROR, "Cannot fail unknown resource"); } controld_ack_event_directly(from_host, from_sys, NULL, op, pcmk__xe_id(xml_rsc)); lrmd_free_event(op); } static void handle_reprobe_op(lrm_state_t *lrm_state, xmlNode *msg, const char *from_sys, const char *from_host, const char *user_name, gboolean is_remote_node, bool reprobe_all_nodes) { crm_notice("Forcing the status of all resources to be redetected"); force_reprobe(lrm_state, from_sys, from_host, user_name, is_remote_node, reprobe_all_nodes); if (!pcmk__strcase_any_of(from_sys, CRM_SYSTEM_PENGINE, CRM_SYSTEM_TENGINE, NULL)) { xmlNode *reply = pcmk__new_reply(msg, NULL); crm_debug("ACK'ing re-probe from %s (%s)", from_sys, from_host); if (relay_message(reply, TRUE) == FALSE) { crm_log_xml_err(reply, "Unable to route reply"); } pcmk__xml_free(reply); } } static bool do_lrm_cancel(ha_msg_input_t *input, lrm_state_t *lrm_state, lrmd_rsc_info_t *rsc, const char *from_host, const char *from_sys) { char *op_key = NULL; char *meta_key = NULL; int call = 0; const char *call_id = NULL; const char *op_task = NULL; guint interval_ms = 0; gboolean in_progress = FALSE; xmlNode *params = pcmk__xe_first_child(input->xml, PCMK__XE_ATTRIBUTES, NULL, NULL); CRM_CHECK(params != NULL, return FALSE); meta_key = crm_meta_name(PCMK_XA_OPERATION); op_task = crm_element_value(params, meta_key); free(meta_key); CRM_CHECK(op_task != NULL, return FALSE); meta_key = crm_meta_name(PCMK_META_INTERVAL); if (crm_element_value_ms(params, meta_key, &interval_ms) != pcmk_ok) { free(meta_key); return FALSE; } free(meta_key); op_key = pcmk__op_key(rsc->id, op_task, interval_ms); meta_key = crm_meta_name(PCMK__XA_CALL_ID); call_id = crm_element_value(params, meta_key); free(meta_key); crm_debug("Scheduler requested op %s (call=%s) be cancelled", op_key, (call_id? call_id : "NA")); pcmk__scan_min_int(call_id, &call, 0); if (call == 0) { // Normal case when the scheduler cancels a recurring op in_progress = cancel_op_key(lrm_state, rsc, op_key, TRUE); } else { // Normal case when the scheduler cancels an orphan op in_progress = cancel_op(lrm_state, rsc->id, NULL, call, TRUE); } // Acknowledge cancellation operation if for a remote connection resource if (!in_progress || is_remote_lrmd_ra(NULL, NULL, rsc->id)) { char *op_id = make_stop_id(rsc->id, call); if (is_remote_lrmd_ra(NULL, NULL, rsc->id) == FALSE) { crm_info("Nothing known about operation %d for %s", call, op_key); } controld_delete_action_history_by_key(rsc->id, lrm_state->node_name, op_key, call); send_task_ok_ack(lrm_state, input, rsc->id, rsc, op_task, from_host, from_sys); /* needed at least for cancellation of a remote operation */ if (lrm_state->active_ops != NULL) { g_hash_table_remove(lrm_state->active_ops, op_id); } free(op_id); } free(op_key); return TRUE; } static void do_lrm_delete(ha_msg_input_t *input, lrm_state_t *lrm_state, lrmd_rsc_info_t *rsc, const char *from_sys, const char *from_host, bool crm_rsc_delete, const char *user_name) { bool unregister = true; int cib_rc = controld_delete_resource_history(rsc->id, lrm_state->node_name, user_name, cib_dryrun|cib_sync_call); if (cib_rc != pcmk_rc_ok) { lrmd_event_data_t *op = NULL; op = construct_op(lrm_state, input->xml, rsc->id, PCMK_ACTION_DELETE); /* These are resource clean-ups, not actions, so no exit reason is * needed. */ lrmd__set_result(op, pcmk_rc2ocf(cib_rc), PCMK_EXEC_ERROR, NULL); controld_ack_event_directly(from_host, from_sys, NULL, op, rsc->id); lrmd_free_event(op); return; } if (crm_rsc_delete && is_remote_lrmd_ra(NULL, NULL, rsc->id)) { unregister = false; } delete_resource(lrm_state, rsc->id, rsc, NULL, from_sys, user_name, input, unregister, true); } // User data for asynchronous metadata execution struct metadata_cb_data { lrmd_rsc_info_t *rsc; // Copy of resource information xmlNode *input_xml; // Copy of FSA input XML }; static struct metadata_cb_data * new_metadata_cb_data(lrmd_rsc_info_t *rsc, xmlNode *input_xml) { struct metadata_cb_data *data = NULL; data = pcmk__assert_alloc(1, sizeof(struct metadata_cb_data)); data->input_xml = pcmk__xml_copy(NULL, input_xml); data->rsc = lrmd_copy_rsc_info(rsc); return data; } static void free_metadata_cb_data(struct metadata_cb_data *data) { lrmd_free_rsc_info(data->rsc); pcmk__xml_free(data->input_xml); free(data); } /*! * \internal * \brief Execute an action after metadata has been retrieved * * \param[in] pid Ignored * \param[in] result Result of metadata action * \param[in] user_data Metadata callback data */ static void metadata_complete(int pid, const pcmk__action_result_t *result, void *user_data) { struct metadata_cb_data *data = (struct metadata_cb_data *) user_data; struct ra_metadata_s *md = NULL; lrm_state_t *lrm_state = controld_get_executor_state(lrm_op_target(data->input_xml), false); if ((lrm_state != NULL) && pcmk__result_ok(result)) { md = controld_cache_metadata(lrm_state->metadata_cache, data->rsc, result->action_stdout); } if (!pcmk_is_set(controld_globals.fsa_input_register, R_HA_DISCONNECTED)) { do_lrm_rsc_op(lrm_state, data->rsc, data->input_xml, md); } free_metadata_cb_data(data); } /* A_LRM_INVOKE */ void do_lrm_invoke(long long action, enum crmd_fsa_cause cause, enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data) { lrm_state_t *lrm_state = NULL; const char *crm_op = NULL; const char *from_sys = NULL; const char *from_host = NULL; const char *operation = NULL; ha_msg_input_t *input = fsa_typed_data(fsa_dt_ha_msg); const char *user_name = NULL; const char *target_node = lrm_op_target(input->xml); gboolean is_remote_node = FALSE; bool crm_rsc_delete = FALSE; // Message routed to the local node is targeting a specific, non-local node is_remote_node = !controld_is_local_node(target_node); lrm_state = controld_get_executor_state(target_node, false); if ((lrm_state == NULL) && is_remote_node) { crm_err("Failing action because local node has never had connection to remote node %s", target_node); synthesize_lrmd_failure(NULL, input->xml, PCMK_EXEC_NOT_CONNECTED, PCMK_OCF_UNKNOWN_ERROR, "Local node has no connection to remote"); return; } pcmk__assert(lrm_state != NULL); user_name = pcmk__update_acl_user(input->msg, PCMK__XA_CRM_USER, NULL); crm_op = crm_element_value(input->msg, PCMK__XA_CRM_TASK); from_sys = crm_element_value(input->msg, PCMK__XA_CRM_SYS_FROM); if (!pcmk__str_eq(from_sys, CRM_SYSTEM_TENGINE, pcmk__str_none)) { from_host = crm_element_value(input->msg, PCMK__XA_SRC); } if (pcmk__str_eq(crm_op, PCMK_ACTION_LRM_DELETE, pcmk__str_none)) { if (!pcmk__str_eq(from_sys, CRM_SYSTEM_TENGINE, pcmk__str_none)) { crm_rsc_delete = TRUE; // from crm_resource } operation = PCMK_ACTION_DELETE; } else if (input->xml != NULL) { operation = crm_element_value(input->xml, PCMK_XA_OPERATION); } CRM_CHECK(!pcmk__str_empty(crm_op) || !pcmk__str_empty(operation), return); crm_trace("'%s' execution request from %s as %s user", pcmk__s(crm_op, operation), pcmk__s(from_sys, "unknown subsystem"), pcmk__s(user_name, "current")); if (pcmk__str_eq(crm_op, CRM_OP_LRM_FAIL, pcmk__str_none)) { fail_lrm_resource(input->xml, lrm_state, user_name, from_host, from_sys); } else if (pcmk__str_eq(crm_op, CRM_OP_REPROBE, pcmk__str_none) || pcmk__str_eq(operation, CRM_OP_REPROBE, pcmk__str_none)) { const char *raw_target = NULL; if (input->xml != NULL) { // For CRM_OP_REPROBE, a NULL target means we're targeting all nodes raw_target = crm_element_value(input->xml, PCMK__META_ON_NODE); } handle_reprobe_op(lrm_state, input->msg, from_sys, from_host, user_name, is_remote_node, (raw_target == NULL)); } else if (operation != NULL) { lrmd_rsc_info_t *rsc = NULL; xmlNode *xml_rsc = pcmk__xe_first_child(input->xml, PCMK_XE_PRIMITIVE, NULL, NULL); gboolean create_rsc = !pcmk__str_eq(operation, PCMK_ACTION_DELETE, pcmk__str_none); int rc; // We can't return anything meaningful without a resource ID CRM_CHECK((xml_rsc != NULL) && (pcmk__xe_id(xml_rsc) != NULL), return); rc = get_lrm_resource(lrm_state, xml_rsc, create_rsc, &rsc); if (rc == -ENOTCONN) { synthesize_lrmd_failure(lrm_state, input->xml, PCMK_EXEC_NOT_CONNECTED, PCMK_OCF_UNKNOWN_ERROR, "Not connected to remote executor"); return; } else if ((rc < 0) && !create_rsc) { /* Delete of malformed or nonexistent resource * (deleting something that does not exist is a success) */ crm_debug("Not registering resource '%s' for a %s event " QB_XS " get-rc=%d (%s) transition-key=%s", pcmk__xe_id(xml_rsc), operation, rc, pcmk_strerror(rc), pcmk__xe_id(input->xml)); delete_rsc_entry(lrm_state, input, pcmk__xe_id(xml_rsc), NULL, pcmk_ok, user_name, true); return; } else if (rc == -EINVAL) { // Resource operation on malformed resource crm_err("Invalid resource definition for %s", pcmk__xe_id(xml_rsc)); crm_log_xml_warn(input->msg, "invalid resource"); synthesize_lrmd_failure(lrm_state, input->xml, PCMK_EXEC_ERROR, PCMK_OCF_NOT_CONFIGURED, // fatal error "Invalid resource definition"); return; } else if (rc < 0) { // Error communicating with the executor crm_err("Could not register resource '%s' with executor: %s " QB_XS " rc=%d", pcmk__xe_id(xml_rsc), pcmk_strerror(rc), rc); crm_log_xml_warn(input->msg, "failed registration"); synthesize_lrmd_failure(lrm_state, input->xml, PCMK_EXEC_ERROR, PCMK_OCF_INVALID_PARAM, // hard error "Could not register resource with executor"); return; } if (pcmk__str_eq(operation, PCMK_ACTION_CANCEL, pcmk__str_none)) { if (!do_lrm_cancel(input, lrm_state, rsc, from_host, from_sys)) { crm_log_xml_warn(input->xml, "Bad command"); } } else if (pcmk__str_eq(operation, PCMK_ACTION_DELETE, pcmk__str_none)) { do_lrm_delete(input, lrm_state, rsc, from_sys, from_host, crm_rsc_delete, user_name); } else { struct ra_metadata_s *md = NULL; /* Getting metadata from cache is OK except for start actions -- * always refresh from the agent for those, in case the resource * agent was updated. * * @TODO Only refresh metadata for starts if the agent actually * changed (using something like inotify, or a hash or modification * time of the agent executable). */ if (strcmp(operation, PCMK_ACTION_START) != 0) { md = controld_get_rsc_metadata(lrm_state, rsc, controld_metadata_from_cache); } if ((md == NULL) && crm_op_needs_metadata(rsc->standard, operation)) { /* Most likely, we'll need the agent metadata to record the * pending operation and the operation result. Get it now rather * than wait until then, so the metadata action doesn't eat into * the real action's timeout. * * @TODO Metadata is retrieved via direct execution of the * agent, which has a couple of related issues: the executor * should execute agents, not the controller; and metadata for * Pacemaker Remote nodes should be collected on those nodes, * not locally. */ struct metadata_cb_data *data = NULL; data = new_metadata_cb_data(rsc, input->xml); crm_info("Retrieving metadata for %s (%s%s%s:%s) asynchronously", rsc->id, rsc->standard, ((rsc->provider == NULL)? "" : ":"), ((rsc->provider == NULL)? "" : rsc->provider), rsc->type); (void) lrmd__metadata_async(rsc, metadata_complete, (void *) data); } else { do_lrm_rsc_op(lrm_state, rsc, input->xml, md); } } lrmd_free_rsc_info(rsc); } else { crm_err("Invalid execution request: unknown command '%s' (bug?)", crm_op); register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); } } static lrmd_event_data_t * construct_op(const lrm_state_t *lrm_state, const xmlNode *rsc_op, const char *rsc_id, const char *operation) { lrmd_event_data_t *op = NULL; const char *op_delay = NULL; const char *op_timeout = NULL; GHashTable *params = NULL; xmlNode *primitive = NULL; const char *class = NULL; const char *transition = NULL; pcmk__assert((rsc_id != NULL) && (operation != NULL)); op = lrmd_new_event(rsc_id, operation, 0); op->type = lrmd_event_exec_complete; op->timeout = 0; op->start_delay = 0; lrmd__set_result(op, PCMK_OCF_UNKNOWN, PCMK_EXEC_PENDING, NULL); if (rsc_op == NULL) { CRM_LOG_ASSERT(pcmk__str_eq(operation, PCMK_ACTION_STOP, pcmk__str_casei)); op->user_data = NULL; /* the stop_all_resources() case * by definition there is no DC (or they'd be shutting * us down). * So we should put our version here. */ op->params = pcmk__strkey_table(free, free); pcmk__insert_dup(op->params, PCMK_XA_CRM_FEATURE_SET, CRM_FEATURE_SET); crm_trace("Constructed %s op for %s", operation, rsc_id); return op; } params = xml2list(rsc_op); g_hash_table_remove(params, CRM_META "_" PCMK__META_OP_TARGET_RC); op_delay = crm_meta_value(params, PCMK_META_START_DELAY); pcmk__scan_min_int(op_delay, &op->start_delay, 0); op_timeout = crm_meta_value(params, PCMK_META_TIMEOUT); pcmk__scan_min_int(op_timeout, &op->timeout, 0); if (pcmk__guint_from_hash(params, CRM_META "_" PCMK_META_INTERVAL, 0, &(op->interval_ms)) != pcmk_rc_ok) { op->interval_ms = 0; } /* Use pcmk_monitor_timeout instead of meta timeout for stonith recurring monitor, if set */ primitive = pcmk__xe_first_child(rsc_op, PCMK_XE_PRIMITIVE, NULL, NULL); class = crm_element_value(primitive, PCMK_XA_CLASS); if (pcmk_is_set(pcmk_get_ra_caps(class), pcmk_ra_cap_fence_params) && pcmk__str_eq(operation, PCMK_ACTION_MONITOR, pcmk__str_casei) && (op->interval_ms > 0)) { op_timeout = g_hash_table_lookup(params, "pcmk_monitor_timeout"); if (op_timeout != NULL) { long long timeout_ms = crm_get_msec(op_timeout); op->timeout = (int) QB_MIN(timeout_ms, INT_MAX); } } if (!pcmk__str_eq(operation, PCMK_ACTION_STOP, pcmk__str_casei)) { op->params = params; } else { rsc_history_t *entry = NULL; if (lrm_state) { entry = g_hash_table_lookup(lrm_state->resource_history, rsc_id); } /* If we do not have stop parameters cached, use * whatever we are given */ if (!entry || !entry->stop_params) { op->params = params; } else { /* Copy the cached parameter list so that we stop the resource * with the old attributes, not the new ones */ op->params = pcmk__strkey_table(free, free); g_hash_table_foreach(params, copy_meta_keys, op->params); g_hash_table_foreach(entry->stop_params, copy_instance_keys, op->params); g_hash_table_destroy(params); params = NULL; } } /* sanity */ if (op->timeout <= 0) { op->timeout = op->interval_ms; } if (op->start_delay < 0) { op->start_delay = 0; } transition = crm_element_value(rsc_op, PCMK__XA_TRANSITION_KEY); CRM_CHECK(transition != NULL, return op); op->user_data = pcmk__str_copy(transition); if (op->interval_ms != 0) { if (pcmk__strcase_any_of(operation, PCMK_ACTION_START, PCMK_ACTION_STOP, NULL)) { crm_err("Start and Stop actions cannot have an interval: %u", op->interval_ms); op->interval_ms = 0; } } crm_trace("Constructed %s op for %s: interval=%u", operation, rsc_id, op->interval_ms); return op; } /*! * \internal * \brief Send a (synthesized) event result * * Reply with a synthesized event result directly, as opposed to going through * the executor. * * \param[in] to_host Host to send result to * \param[in] to_sys IPC name to send result (NULL for transition engine) * \param[in] rsc Type information about resource the result is for * \param[in,out] op Event with result to send * \param[in] rsc_id ID of resource the result is for */ void controld_ack_event_directly(const char *to_host, const char *to_sys, const lrmd_rsc_info_t *rsc, lrmd_event_data_t *op, const char *rsc_id) { xmlNode *reply = NULL; xmlNode *update, *iter; pcmk__node_status_t *peer = NULL; CRM_CHECK(op != NULL, return); if (op->rsc_id == NULL) { // op->rsc_id is a (const char *) but lrmd_free_event() frees it pcmk__assert(rsc_id != NULL); op->rsc_id = pcmk__str_copy(rsc_id); } if (to_sys == NULL) { to_sys = CRM_SYSTEM_TENGINE; } peer = controld_get_local_node_status(); update = create_node_state_update(peer, controld_node_update_none, NULL, __func__); iter = pcmk__xe_create(update, PCMK__XE_LRM); crm_xml_add(iter, PCMK_XA_ID, controld_globals.our_uuid); iter = pcmk__xe_create(iter, PCMK__XE_LRM_RESOURCES); iter = pcmk__xe_create(iter, PCMK__XE_LRM_RESOURCE); crm_xml_add(iter, PCMK_XA_ID, op->rsc_id); controld_add_resource_history_xml(iter, rsc, op, controld_globals.cluster->priv->node_name); /* We don't have the original message ID, so use "direct-ack" (we just need * something non-NULL for this to create a reply) * * @TODO It would be better to use the server, message ID, and task from the * original request when callers have it available */ reply = pcmk__new_message(pcmk_ipc_controld, "direct-ack", CRM_SYSTEM_LRMD, to_host, to_sys, CRM_OP_INVOKE_LRM, update); crm_log_xml_trace(update, "[direct ACK]"); crm_debug("ACK'ing resource op " PCMK__OP_FMT " from %s: %s", op->rsc_id, op->op_type, op->interval_ms, op->user_data, crm_element_value(reply, PCMK_XA_REFERENCE)); if (relay_message(reply, TRUE) == FALSE) { crm_log_xml_err(reply, "Unable to route reply"); } pcmk__xml_free(update); pcmk__xml_free(reply); } gboolean verify_stopped(enum crmd_fsa_state cur_state, int log_level) { gboolean res = TRUE; GList *lrm_state_list = lrm_state_get_list(); GList *state_entry; for (state_entry = lrm_state_list; state_entry != NULL; state_entry = state_entry->next) { lrm_state_t *lrm_state = state_entry->data; if (!lrm_state_verify_stopped(lrm_state, cur_state, log_level)) { /* keep iterating through all even when false is returned */ res = FALSE; } } controld_set_fsa_input_flags(R_SENT_RSC_STOP); g_list_free(lrm_state_list); lrm_state_list = NULL; return res; } struct stop_recurring_action_s { lrmd_rsc_info_t *rsc; lrm_state_t *lrm_state; }; static gboolean stop_recurring_action_by_rsc(gpointer key, gpointer value, gpointer user_data) { gboolean remove = FALSE; struct stop_recurring_action_s *event = user_data; active_op_t *op = value; if ((op->interval_ms != 0) && pcmk__str_eq(op->rsc_id, event->rsc->id, pcmk__str_none)) { crm_debug("Cancelling op %d for %s (%s)", op->call_id, op->rsc_id, (char*)key); remove = !cancel_op(event->lrm_state, event->rsc->id, key, op->call_id, FALSE); } return remove; } static gboolean stop_recurring_actions(gpointer key, gpointer value, gpointer user_data) { gboolean remove = FALSE; lrm_state_t *lrm_state = user_data; active_op_t *op = value; if (op->interval_ms != 0) { crm_info("Cancelling op %d for %s (%s)", op->call_id, op->rsc_id, (const char *) key); remove = !cancel_op(lrm_state, op->rsc_id, key, op->call_id, FALSE); } return remove; } /*! * \internal * \brief Check whether recurring actions should be cancelled before an action * * \param[in] rsc_id Resource that action is for * \param[in] action Action being performed * \param[in] interval_ms Operation interval of \p action (in milliseconds) * * \return true if recurring actions should be cancelled, otherwise false */ static bool should_cancel_recurring(const char *rsc_id, const char *action, guint interval_ms) { if (is_remote_lrmd_ra(NULL, NULL, rsc_id) && (interval_ms == 0) && (strcmp(action, PCMK_ACTION_MIGRATE_TO) == 0)) { /* Don't stop monitoring a migrating Pacemaker Remote connection * resource until the entire migration has completed. We must detect if * the connection is unexpectedly severed, even during a migration. */ return false; } // Cancel recurring actions before changing resource state return (interval_ms == 0) && !pcmk__str_any_of(action, PCMK_ACTION_MONITOR, PCMK_ACTION_NOTIFY, NULL); } /*! * \internal * \brief Check whether an action should not be performed at this time * * \param[in] operation Action to be performed * * \return Readable description of why action should not be performed, * or NULL if it should be performed */ static const char * should_nack_action(const char *action) { if (pcmk_is_set(controld_globals.fsa_input_register, R_SHUTDOWN) && pcmk__str_eq(action, PCMK_ACTION_START, pcmk__str_none)) { register_fsa_input(C_SHUTDOWN, I_SHUTDOWN, NULL); return "Not attempting start due to shutdown in progress"; } switch (controld_globals.fsa_state) { case S_NOT_DC: case S_POLICY_ENGINE: // Recalculating case S_TRANSITION_ENGINE: break; default: if (!pcmk__str_eq(action, PCMK_ACTION_STOP, pcmk__str_none)) { return "Controller cannot attempt actions at this time"; } break; } return NULL; } static void do_lrm_rsc_op(lrm_state_t *lrm_state, lrmd_rsc_info_t *rsc, xmlNode *msg, struct ra_metadata_s *md) { int rc; int call_id = 0; char *op_id = NULL; lrmd_event_data_t *op = NULL; fsa_data_t *msg_data = NULL; const char *transition = NULL; const char *operation = NULL; const char *nack_reason = NULL; CRM_CHECK((rsc != NULL) && (msg != NULL), return); operation = crm_element_value(msg, PCMK_XA_OPERATION); CRM_CHECK(!pcmk__str_empty(operation), return); transition = crm_element_value(msg, PCMK__XA_TRANSITION_KEY); if (pcmk__str_empty(transition)) { crm_log_xml_err(msg, "Missing transition number"); } if (lrm_state == NULL) { // This shouldn't be possible, but provide a failsafe just in case crm_err("Cannot execute %s of %s: No executor connection " QB_XS " transition_key=%s", operation, rsc->id, pcmk__s(transition, "")); synthesize_lrmd_failure(NULL, msg, PCMK_EXEC_INVALID, PCMK_OCF_UNKNOWN_ERROR, "No executor connection"); return; } if (pcmk__str_any_of(operation, PCMK_ACTION_RELOAD, PCMK_ACTION_RELOAD_AGENT, NULL)) { /* Pre-2.1.0 DCs will schedule reload actions only, and 2.1.0+ DCs * will schedule reload-agent actions only. In either case, we need * to map that to whatever the resource agent actually supports. * Default to the OCF 1.1 name. */ if ((md != NULL) && pcmk_is_set(md->ra_flags, ra_supports_legacy_reload)) { operation = PCMK_ACTION_RELOAD; } else { operation = PCMK_ACTION_RELOAD_AGENT; } } op = construct_op(lrm_state, msg, rsc->id, operation); CRM_CHECK(op != NULL, return); if (should_cancel_recurring(rsc->id, operation, op->interval_ms)) { guint removed = 0; struct stop_recurring_action_s data; data.rsc = rsc; data.lrm_state = lrm_state; removed = g_hash_table_foreach_remove(lrm_state->active_ops, stop_recurring_action_by_rsc, &data); if (removed) { crm_debug("Stopped %u recurring operation%s in preparation for " PCMK__OP_FMT, removed, pcmk__plural_s(removed), rsc->id, operation, op->interval_ms); } } nack_reason = should_nack_action(operation); if (nack_reason != NULL) { crm_notice("Not requesting local execution of %s operation for %s on %s" " in state %s: %s", pcmk__readable_action(op->op_type, op->interval_ms), rsc->id, lrm_state->node_name, fsa_state2string(controld_globals.fsa_state), nack_reason); lrmd__set_result(op, PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_INVALID, nack_reason); controld_ack_event_directly(NULL, NULL, rsc, op, rsc->id); lrmd_free_event(op); free(op_id); return; } crm_notice("Requesting local execution of %s operation for %s on %s " QB_XS " transition %s", pcmk__readable_action(op->op_type, op->interval_ms), rsc->id, lrm_state->node_name, pcmk__s(transition, "")); controld_record_pending_op(lrm_state->node_name, rsc, op); op_id = pcmk__op_key(rsc->id, op->op_type, op->interval_ms); if (op->interval_ms > 0) { /* cancel it so we can then restart it without conflict */ cancel_op_key(lrm_state, rsc, op_id, FALSE); } rc = controld_execute_resource_agent(lrm_state, rsc->id, op->op_type, op->user_data, op->interval_ms, op->timeout, op->start_delay, op->params, &call_id); if (rc == pcmk_rc_ok) { /* record all operations so we can wait * for them to complete during shutdown */ char *call_id_s = make_stop_id(rsc->id, call_id); active_op_t *pending = NULL; pending = pcmk__assert_alloc(1, sizeof(active_op_t)); crm_trace("Recording pending op: %d - %s %s", call_id, op_id, call_id_s); pending->call_id = call_id; pending->interval_ms = op->interval_ms; pending->op_type = pcmk__str_copy(operation); pending->op_key = pcmk__str_copy(op_id); pending->rsc_id = pcmk__str_copy(rsc->id); pending->start_time = time(NULL); pending->user_data = pcmk__str_copy(op->user_data); if (crm_element_value_epoch(msg, PCMK_OPT_SHUTDOWN_LOCK, &(pending->lock_time)) != pcmk_ok) { pending->lock_time = 0; } g_hash_table_replace(lrm_state->active_ops, call_id_s, pending); if ((op->interval_ms > 0) && (op->start_delay > START_DELAY_THRESHOLD)) { int target_rc = PCMK_OCF_OK; crm_info("Faking confirmation of %s: execution postponed for over 5 minutes", op_id); decode_transition_key(op->user_data, NULL, NULL, NULL, &target_rc); lrmd__set_result(op, target_rc, PCMK_EXEC_DONE, NULL); controld_ack_event_directly(NULL, NULL, rsc, op, rsc->id); } pending->params = op->params; op->params = NULL; } else if (lrm_state_is_local(lrm_state)) { crm_err("Could not initiate %s action for resource %s locally: %s " QB_XS " rc=%d", operation, rsc->id, pcmk_rc_str(rc), rc); fake_op_status(lrm_state, op, PCMK_EXEC_NOT_CONNECTED, PCMK_OCF_UNKNOWN_ERROR, pcmk_rc_str(rc)); process_lrm_event(lrm_state, op, NULL, NULL); register_fsa_error(C_FSA_INTERNAL, I_FAIL, NULL); } else { crm_err("Could not initiate %s action for resource %s remotely on %s: " "%s " QB_XS " rc=%d", operation, rsc->id, lrm_state->node_name, pcmk_rc_str(rc), rc); fake_op_status(lrm_state, op, PCMK_EXEC_NOT_CONNECTED, PCMK_OCF_UNKNOWN_ERROR, pcmk_rc_str(rc)); process_lrm_event(lrm_state, op, NULL, NULL); } free(op_id); lrmd_free_event(op); } static char * unescape_newlines(const char *string) { char *pch = NULL; char *ret = NULL; static const char *escaped_newline = "\\n"; if (!string) { return NULL; } ret = pcmk__str_copy(string); pch = strstr(ret, escaped_newline); while (pch != NULL) { /* Replace newline escape pattern with actual newline (and a space so we * don't have to shuffle the rest of the buffer) */ pch[0] = '\n'; pch[1] = ' '; pch = strstr(pch, escaped_newline); } return ret; } static bool did_lrm_rsc_op_fail(lrm_state_t *lrm_state, const char * rsc_id, const char * op_type, guint interval_ms) { rsc_history_t *entry = NULL; CRM_CHECK(lrm_state != NULL, return FALSE); CRM_CHECK(rsc_id != NULL, return FALSE); CRM_CHECK(op_type != NULL, return FALSE); entry = g_hash_table_lookup(lrm_state->resource_history, rsc_id); if (entry == NULL || entry->failed == NULL) { return FALSE; } if (pcmk__str_eq(entry->failed->rsc_id, rsc_id, pcmk__str_none) && pcmk__str_eq(entry->failed->op_type, op_type, pcmk__str_casei) && entry->failed->interval_ms == interval_ms) { return TRUE; } return FALSE; } /*! * \internal * \brief Log the result of an executor action (actual or synthesized) * * \param[in] op Executor action to log result for * \param[in] op_key Operation key for action * \param[in] node_name Name of node action was performed on, if known * \param[in] confirmed Whether to log that graph action was confirmed */ static void log_executor_event(const lrmd_event_data_t *op, const char *op_key, const char *node_name, gboolean confirmed) { int log_level = LOG_ERR; GString *str = g_string_sized_new(100); // reasonable starting size pcmk__g_strcat(str, "Result of ", pcmk__readable_action(op->op_type, op->interval_ms), " operation for ", op->rsc_id, NULL); if (node_name != NULL) { pcmk__g_strcat(str, " on ", node_name, NULL); } switch (op->op_status) { case PCMK_EXEC_DONE: log_level = LOG_NOTICE; pcmk__g_strcat(str, ": ", crm_exit_str((crm_exit_t) op->rc), NULL); break; case PCMK_EXEC_TIMEOUT: pcmk__g_strcat(str, ": ", pcmk_exec_status_str(op->op_status), " after ", pcmk__readable_interval(op->timeout), NULL); break; case PCMK_EXEC_CANCELLED: log_level = LOG_INFO; pcmk__g_strcat(str, ": ", pcmk_exec_status_str(op->op_status), NULL); break; default: pcmk__g_strcat(str, ": ", pcmk_exec_status_str(op->op_status), NULL); break; } if ((op->exit_reason != NULL) && ((op->op_status != PCMK_EXEC_DONE) || (op->rc != PCMK_OCF_OK))) { pcmk__g_strcat(str, " (", op->exit_reason, ")", NULL); } g_string_append(str, " " QB_XS); g_string_append_printf(str, " graph action %sconfirmed; call=%d key=%s", (confirmed? "" : "un"), op->call_id, op_key); if (op->op_status == PCMK_EXEC_DONE) { g_string_append_printf(str, " rc=%d", op->rc); } do_crm_log(log_level, "%s", str->str); g_string_free(str, TRUE); /* The services library has already logged the output at info or debug * level, so just raise to notice if it looks like a failure. */ if ((op->output != NULL) && (op->rc != PCMK_OCF_OK)) { char *prefix = crm_strdup_printf(PCMK__OP_FMT "@%s output", op->rsc_id, op->op_type, op->interval_ms, node_name); crm_log_output(LOG_NOTICE, prefix, op->output); free(prefix); } } void process_lrm_event(lrm_state_t *lrm_state, lrmd_event_data_t *op, active_op_t *pending, const xmlNode *action_xml) { char *op_id = NULL; char *op_key = NULL; gboolean remove = FALSE; gboolean removed = FALSE; bool need_direct_ack = FALSE; lrmd_rsc_info_t *rsc = NULL; const char *node_name = NULL; CRM_CHECK(op != NULL, return); CRM_CHECK(op->rsc_id != NULL, return); // Remap new status codes for older DCs if (compare_version(controld_globals.dc_version, "3.2.0") < 0) { switch (op->op_status) { case PCMK_EXEC_NOT_CONNECTED: lrmd__set_result(op, PCMK_OCF_CONNECTION_DIED, PCMK_EXEC_ERROR, op->exit_reason); break; case PCMK_EXEC_INVALID: lrmd__set_result(op, CRM_DIRECT_NACK_RC, PCMK_EXEC_ERROR, op->exit_reason); break; default: break; } } op_id = make_stop_id(op->rsc_id, op->call_id); op_key = pcmk__op_key(op->rsc_id, op->op_type, op->interval_ms); // Get resource info if available (from executor state or action XML) if (lrm_state) { rsc = lrm_state_get_rsc_info(lrm_state, op->rsc_id, 0); } if ((rsc == NULL) && action_xml) { xmlNode *xml = pcmk__xe_first_child(action_xml, PCMK_XE_PRIMITIVE, NULL, NULL); const char *standard = crm_element_value(xml, PCMK_XA_CLASS); const char *provider = crm_element_value(xml, PCMK_XA_PROVIDER); const char *type = crm_element_value(xml, PCMK_XA_TYPE); if (standard && type) { crm_info("%s agent information not cached, using %s%s%s:%s from action XML", op->rsc_id, standard, (provider? ":" : ""), (provider? provider : ""), type); rsc = lrmd_new_rsc_info(op->rsc_id, standard, provider, type); } else { crm_err("Can't process %s result because %s agent information not cached or in XML", op_key, op->rsc_id); } } // Get node name if available (from executor state or action XML) if (lrm_state) { node_name = lrm_state->node_name; } else if (action_xml) { node_name = crm_element_value(action_xml, PCMK__META_ON_NODE); } if(pending == NULL) { remove = TRUE; if (lrm_state) { pending = g_hash_table_lookup(lrm_state->active_ops, op_id); } } if (op->op_status == PCMK_EXEC_ERROR) { switch(op->rc) { case PCMK_OCF_NOT_RUNNING: case PCMK_OCF_RUNNING_PROMOTED: case PCMK_OCF_DEGRADED: case PCMK_OCF_DEGRADED_PROMOTED: // Leave it to the TE/scheduler to decide if this is an error op->op_status = PCMK_EXEC_DONE; break; default: /* Nothing to do */ break; } } if (op->op_status != PCMK_EXEC_CANCELLED) { /* We might not record the result, so directly acknowledge it to the * originator instead, so it doesn't time out waiting for the result * (especially important if part of a transition). */ need_direct_ack = TRUE; if (controld_action_is_recordable(op->op_type)) { if (node_name && rsc) { // We should record the result, and happily, we can time_t lock_time = (pending == NULL)? 0 : pending->lock_time; controld_update_resource_history(node_name, rsc, op, lock_time); need_direct_ack = FALSE; } else if (op->rsc_deleted) { /* We shouldn't record the result (likely the resource was * refreshed, cleaned, or removed while this operation was * in flight). */ crm_notice("Not recording %s result in CIB because " "resource information was removed since it was initiated", op_key); } else { /* This shouldn't be possible; the executor didn't consider the * resource deleted, but we couldn't find resource or node * information. */ crm_err("Unable to record %s result in CIB: %s", op_key, (node_name? "No resource information" : "No node name")); } } } else if (op->interval_ms == 0) { /* A non-recurring operation was cancelled. Most likely, the * never-initiated action was removed from the executor's pending * operations list upon resource removal. */ need_direct_ack = TRUE; } else if (pending == NULL) { /* This recurring operation was cancelled, but was not pending. No * transition actions are waiting on it, nothing needs to be done. */ } else if (op->user_data == NULL) { /* This recurring operation was cancelled and pending, but we don't * have a transition key. This should never happen. */ crm_err("Recurring operation %s was cancelled without transition information", op_key); } else if (pcmk_is_set(pending->flags, active_op_remove)) { /* This recurring operation was cancelled (by us) and pending, and we * have been waiting for it to finish. */ if (lrm_state) { controld_delete_action_history(op); } /* Directly acknowledge failed recurring actions here. The above call to * controld_delete_action_history() will not erase any corresponding * last_failure entry, which means that the DC won't confirm the * cancellation via process_op_deletion(), and the transition would * otherwise wait for the action timer to pop. */ if (did_lrm_rsc_op_fail(lrm_state, pending->rsc_id, pending->op_type, pending->interval_ms)) { need_direct_ack = TRUE; } } else if (op->rsc_deleted) { /* This recurring operation was cancelled (but not by us, and the * executor does not have resource information, likely due to resource * cleanup, refresh, or removal) and pending. */ crm_debug("Recurring op %s was cancelled due to resource deletion", op_key); need_direct_ack = TRUE; } else { /* This recurring operation was cancelled (but not by us, likely by the * executor before stopping the resource) and pending. We don't need to * do anything special. */ } if (need_direct_ack) { controld_ack_event_directly(NULL, NULL, NULL, op, op->rsc_id); } if(remove == FALSE) { /* The caller will do this afterwards, but keep the logging consistent */ removed = TRUE; } else if (lrm_state && ((op->interval_ms == 0) || (op->op_status == PCMK_EXEC_CANCELLED))) { gboolean found = g_hash_table_remove(lrm_state->active_ops, op_id); if (op->interval_ms != 0) { removed = TRUE; } else if (found) { removed = TRUE; crm_trace("Op %s (call=%d, stop-id=%s, remaining=%u): Confirmed", op_key, op->call_id, op_id, g_hash_table_size(lrm_state->active_ops)); } } log_executor_event(op, op_key, node_name, removed); if (lrm_state) { if (!pcmk__str_eq(op->op_type, PCMK_ACTION_META_DATA, pcmk__str_casei)) { crmd_alert_resource_op(lrm_state->node_name, op); } else if (rsc && (op->rc == PCMK_OCF_OK)) { char *metadata = unescape_newlines(op->output); controld_cache_metadata(lrm_state->metadata_cache, rsc, metadata); free(metadata); } } if (op->rsc_deleted) { crm_info("Deletion of resource '%s' complete after %s", op->rsc_id, op_key); if (lrm_state) { delete_rsc_entry(lrm_state, NULL, op->rsc_id, NULL, pcmk_ok, NULL, true); } } /* If a shutdown was escalated while operations were pending, * then the FSA will be stalled right now... allow it to continue */ controld_trigger_fsa(); if (lrm_state && rsc) { update_history_cache(lrm_state, rsc, op); } lrmd_free_rsc_info(rsc); free(op_key); free(op_id); } diff --git a/daemons/controld/controld_fencing.c b/daemons/controld/controld_fencing.c index 51367ca1a7..5b689b8ec4 100644 --- a/daemons/controld/controld_fencing.c +++ b/daemons/controld/controld_fencing.c @@ -1,1113 +1,1117 @@ /* * Copyright 2004-2025 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include static void tengine_stonith_history_synced(stonith_t *st, stonith_event_t *st_event); /* * stonith failure counting * * We don't want to get stuck in a permanent fencing loop. Keep track of the * number of fencing failures for each target node, and the most we'll restart a * transition for. */ struct st_fail_rec { int count; }; #define DEFAULT_STONITH_MAX_ATTEMPTS 10 static bool fence_reaction_panic = false; static unsigned long int stonith_max_attempts = DEFAULT_STONITH_MAX_ATTEMPTS; static GHashTable *stonith_failures = NULL; /*! * \internal * \brief Update max fencing attempts before giving up * * \param[in] value New max fencing attempts */ static void update_stonith_max_attempts(const char *value) { int score = 0; int rc = pcmk_parse_score(value, &score, DEFAULT_STONITH_MAX_ATTEMPTS); // The option validator ensures invalid values shouldn't be possible CRM_CHECK((rc == pcmk_rc_ok) && (score > 0), return); if (stonith_max_attempts != score) { crm_debug("Maximum fencing attempts per transition is now %d (was %lu)", score, stonith_max_attempts); } stonith_max_attempts = score; } /*! * \internal * \brief Configure reaction to notification of local node being fenced * * \param[in] reaction_s Reaction type */ static void set_fence_reaction(const char *reaction_s) { if (pcmk__str_eq(reaction_s, "panic", pcmk__str_casei)) { fence_reaction_panic = true; } else { if (!pcmk__str_eq(reaction_s, PCMK_VALUE_STOP, pcmk__str_casei)) { crm_warn("Invalid value '%s' for %s, using 'stop'", reaction_s, PCMK_OPT_FENCE_REACTION); } fence_reaction_panic = false; } } /*! * \internal * \brief Configure fencing options based on the CIB * * \param[in,out] options Name/value pairs for configured options */ void controld_configure_fencing(GHashTable *options) { const char *value = NULL; value = g_hash_table_lookup(options, PCMK_OPT_FENCE_REACTION); set_fence_reaction(value); value = g_hash_table_lookup(options, PCMK_OPT_STONITH_MAX_ATTEMPTS); update_stonith_max_attempts(value); } static gboolean too_many_st_failures(const char *target) { GHashTableIter iter; const char *key = NULL; struct st_fail_rec *value = NULL; if (stonith_failures == NULL) { return FALSE; } if (target == NULL) { g_hash_table_iter_init(&iter, stonith_failures); while (g_hash_table_iter_next(&iter, (gpointer *) &key, (gpointer *) &value)) { if (value->count >= stonith_max_attempts) { target = (const char*)key; goto too_many; } } } else { value = g_hash_table_lookup(stonith_failures, target); if ((value != NULL) && (value->count >= stonith_max_attempts)) { goto too_many; } } return FALSE; too_many: crm_warn("Too many failures (%d) to fence %s, giving up", value->count, target); return TRUE; } /*! * \internal * \brief Reset a stonith fail count * * \param[in] target Name of node to reset, or NULL for all */ void st_fail_count_reset(const char *target) { if (stonith_failures == NULL) { return; } if (target) { struct st_fail_rec *rec = NULL; rec = g_hash_table_lookup(stonith_failures, target); if (rec) { rec->count = 0; } } else { GHashTableIter iter; const char *key = NULL; struct st_fail_rec *rec = NULL; g_hash_table_iter_init(&iter, stonith_failures); while (g_hash_table_iter_next(&iter, (gpointer *) &key, (gpointer *) &rec)) { rec->count = 0; } } } static void st_fail_count_increment(const char *target) { struct st_fail_rec *rec = NULL; if (stonith_failures == NULL) { stonith_failures = pcmk__strkey_table(free, free); } rec = g_hash_table_lookup(stonith_failures, target); if (rec) { rec->count++; } else { rec = malloc(sizeof(struct st_fail_rec)); if(rec == NULL) { return; } rec->count = 1; g_hash_table_insert(stonith_failures, pcmk__str_copy(target), rec); } } /* end stonith fail count functions */ static void cib_fencing_updated(xmlNode *msg, int call_id, int rc, xmlNode *output, void *user_data) { if (rc < pcmk_ok) { crm_err("Fencing update %d for %s: failed - %s (%d)", call_id, (char *)user_data, pcmk_strerror(rc), rc); crm_log_xml_warn(msg, "Failed update"); abort_transition(PCMK_SCORE_INFINITY, pcmk__graph_shutdown, "CIB update failed", NULL); } else { crm_info("Fencing update %d for %s: complete", call_id, (char *)user_data); } } /*! * \internal * \brief Update a fencing target's node state * * \param[in] target Node that was successfully fenced * \param[in] target_xml_id CIB XML ID of target */ static void update_node_state_after_fencing(const char *target, const char *target_xml_id) { int rc = pcmk_ok; pcmk__node_status_t *peer = NULL; xmlNode *node_state = NULL; /* We (usually) rely on the membership layer to do * controld_node_update_cluster, and the peer status callback to do * controld_node_update_peer, because the node might have already rejoined * before we get the stonith result here. */ uint32_t flags = controld_node_update_join|controld_node_update_expected; CRM_CHECK((target != NULL) && (target_xml_id != NULL), return); // Ensure target is cached peer = pcmk__get_node(0, target, target_xml_id, pcmk__node_search_any); CRM_CHECK(peer != NULL, return); if (peer->state == NULL) { /* Usually, we rely on the membership layer to update the cluster state * in the CIB. However, if the node has never been seen, do it here, so * the node is not considered unclean. */ flags |= controld_node_update_cluster; } if (peer->xml_id == NULL) { crm_info("Recording XML ID '%s' for node '%s'", target_xml_id, target); peer->xml_id = pcmk__str_copy(target_xml_id); } crmd_peer_down(peer, TRUE); node_state = create_node_state_update(peer, flags, NULL, __func__); crm_xml_add(node_state, PCMK_XA_ID, target_xml_id); if (pcmk_is_set(peer->flags, pcmk__node_status_remote)) { char *now_s = pcmk__ttoa(time(NULL)); crm_xml_add(node_state, PCMK__XA_NODE_FENCED, now_s); free(now_s); } rc = controld_globals.cib_conn->cmds->modify(controld_globals.cib_conn, PCMK_XE_STATUS, node_state, cib_can_create); pcmk__xml_free(node_state); crm_debug("Updating node state for %s after fencing (call %d)", target, rc); fsa_register_cib_callback(rc, pcmk__str_copy(target), cib_fencing_updated); - controld_delete_node_state(peer->name, controld_section_all, cib_none); + // Delete node's resource history from CIB + controld_delete_node_history(peer->name, false, cib_none); + + // Ask attribute manager to delete node's transient attributes + controld_purge_node_attrs(peer->name, false); } /*! * \internal * \brief Abort transition due to stonith failure * * \param[in] abort_action Whether to restart or stop transition * \param[in] target Don't restart if this (NULL for any) has too many failures * \param[in] reason Log this stonith action XML as abort reason (or NULL) */ static void abort_for_stonith_failure(enum pcmk__graph_next abort_action, const char *target, const xmlNode *reason) { /* If stonith repeatedly fails, we eventually give up on starting a new * transition for that reason. */ if ((abort_action != pcmk__graph_wait) && too_many_st_failures(target)) { abort_action = pcmk__graph_wait; } abort_transition(PCMK_SCORE_INFINITY, abort_action, "Stonith failed", reason); } /* * stonith cleanup list * * If the DC is shot, proper notifications might not go out. * The stonith cleanup list allows the cluster to (re-)send * notifications once a new DC is elected. */ static GList *stonith_cleanup_list = NULL; /*! * \internal * \brief Add a node to the stonith cleanup list * * \param[in] target Name of node to add */ void add_stonith_cleanup(const char *target) { stonith_cleanup_list = g_list_append(stonith_cleanup_list, pcmk__str_copy(target)); } /*! * \internal * \brief Remove a node from the stonith cleanup list * * \param[in] Name of node to remove */ void remove_stonith_cleanup(const char *target) { GList *iter = stonith_cleanup_list; while (iter != NULL) { GList *tmp = iter; char *iter_name = tmp->data; iter = iter->next; if (pcmk__str_eq(target, iter_name, pcmk__str_casei)) { crm_trace("Removing %s from the cleanup list", iter_name); stonith_cleanup_list = g_list_delete_link(stonith_cleanup_list, tmp); free(iter_name); } } } /*! * \internal * \brief Purge all entries from the stonith cleanup list */ void purge_stonith_cleanup(void) { if (stonith_cleanup_list) { GList *iter = NULL; for (iter = stonith_cleanup_list; iter != NULL; iter = iter->next) { char *target = iter->data; crm_info("Purging %s from stonith cleanup list", target); free(target); } g_list_free(stonith_cleanup_list); stonith_cleanup_list = NULL; } } /*! * \internal * \brief Send stonith updates for all entries in cleanup list, then purge it */ void execute_stonith_cleanup(void) { GList *iter; for (iter = stonith_cleanup_list; iter != NULL; iter = iter->next) { char *target = iter->data; pcmk__node_status_t *target_node = pcmk__get_node(0, target, NULL, pcmk__node_search_cluster_member); const char *uuid = pcmk__cluster_get_xml_id(target_node); crm_notice("Marking %s, target of a previous stonith action, as clean", target); update_node_state_after_fencing(target, uuid); free(target); } g_list_free(stonith_cleanup_list); stonith_cleanup_list = NULL; } /* end stonith cleanup list functions */ /* stonith API client * * Functions that need to interact directly with the fencer via its API */ static stonith_t *stonith_api = NULL; static mainloop_timer_t *controld_fencer_connect_timer = NULL; static char *te_client_id = NULL; static gboolean fail_incompletable_stonith(pcmk__graph_t *graph) { GList *lpc = NULL; const char *task = NULL; xmlNode *last_action = NULL; if (graph == NULL) { return FALSE; } for (lpc = graph->synapses; lpc != NULL; lpc = lpc->next) { GList *lpc2 = NULL; pcmk__graph_synapse_t *synapse = (pcmk__graph_synapse_t *) lpc->data; if (pcmk_is_set(synapse->flags, pcmk__synapse_confirmed)) { continue; } for (lpc2 = synapse->actions; lpc2 != NULL; lpc2 = lpc2->next) { pcmk__graph_action_t *action = (pcmk__graph_action_t *) lpc2->data; if ((action->type != pcmk__cluster_graph_action) || pcmk_is_set(action->flags, pcmk__graph_action_confirmed)) { continue; } task = crm_element_value(action->xml, PCMK_XA_OPERATION); if (pcmk__str_eq(task, PCMK_ACTION_STONITH, pcmk__str_casei)) { pcmk__set_graph_action_flags(action, pcmk__graph_action_failed); last_action = action->xml; pcmk__update_graph(graph, action); crm_notice("Failing action %d (%s): fencer terminated", action->id, pcmk__xe_id(action->xml)); } } } if (last_action != NULL) { crm_warn("Fencer failure resulted in unrunnable actions"); abort_for_stonith_failure(pcmk__graph_restart, NULL, last_action); return TRUE; } return FALSE; } static void tengine_stonith_connection_destroy(stonith_t *st, stonith_event_t *e) { te_cleanup_stonith_history_sync(st, FALSE); if (pcmk_is_set(controld_globals.fsa_input_register, R_ST_REQUIRED)) { crm_err("Lost fencer connection (will attempt to reconnect)"); if (!mainloop_timer_running(controld_fencer_connect_timer)) { mainloop_timer_start(controld_fencer_connect_timer); } } else { crm_info("Disconnected from fencer"); } if (stonith_api) { /* the client API won't properly reconnect notifications * if they are still in the table - so remove them */ if (stonith_api->state != stonith_disconnected) { stonith_api->cmds->disconnect(st); } stonith_api->cmds->remove_notification(stonith_api, NULL); } if (AM_I_DC) { fail_incompletable_stonith(controld_globals.transition_graph); trigger_graph(); } } /*! * \internal * \brief Handle an event notification from the fencing API * * \param[in] st Fencing API connection (ignored) * \param[in] event Fencing API event notification */ static void handle_fence_notification(stonith_t *st, stonith_event_t *event) { bool succeeded = true; const char *executioner = "the cluster"; const char *client = "a client"; const char *reason = NULL; int exec_status; if (te_client_id == NULL) { te_client_id = crm_strdup_printf("%s.%lu", crm_system_name, (unsigned long) getpid()); } if (event == NULL) { crm_err("Notify data not found"); return; } if (event->executioner != NULL) { executioner = event->executioner; } if (event->client_origin != NULL) { client = event->client_origin; } exec_status = stonith__event_execution_status(event); if ((stonith__event_exit_status(event) != CRM_EX_OK) || (exec_status != PCMK_EXEC_DONE)) { succeeded = false; if (exec_status == PCMK_EXEC_DONE) { exec_status = PCMK_EXEC_ERROR; } } reason = stonith__event_exit_reason(event); crmd_alert_fencing_op(event); if (pcmk__str_eq(PCMK_ACTION_ON, event->action, pcmk__str_none)) { // Unfencing doesn't need special handling, just a log message if (succeeded) { crm_notice("%s was unfenced by %s at the request of %s@%s", event->target, executioner, client, event->origin); } else { crm_err("Unfencing of %s by %s failed (%s%s%s) with exit status %d", event->target, executioner, pcmk_exec_status_str(exec_status), ((reason == NULL)? "" : ": "), ((reason == NULL)? "" : reason), stonith__event_exit_status(event)); } return; } if (succeeded && controld_is_local_node(event->target)) { /* We were notified of our own fencing. Most likely, either fencing was * misconfigured, or fabric fencing that doesn't cut cluster * communication is in use. * * Either way, shutting down the local host is a good idea, to require * administrator intervention. Also, other nodes would otherwise likely * set our status to lost because of the fencing callback and discard * our subsequent election votes as "not part of our cluster". */ crm_crit("We were allegedly just fenced by %s for %s!", executioner, event->origin); // Dumps blackbox if enabled if (fence_reaction_panic) { pcmk__panic("Notified of own fencing"); } else { crm_exit(CRM_EX_FATAL); } return; // Should never get here } /* Update the count of fencing failures for this target, in case we become * DC later. The current DC has already updated its fail count in * tengine_stonith_callback(). */ if (!AM_I_DC) { if (succeeded) { st_fail_count_reset(event->target); } else { st_fail_count_increment(event->target); } } crm_notice("Peer %s was%s terminated (%s) by %s on behalf of %s@%s: " "%s%s%s%s " QB_XS " event=%s", event->target, (succeeded? "" : " not"), event->action, executioner, client, event->origin, (succeeded? "OK" : pcmk_exec_status_str(exec_status)), ((reason == NULL)? "" : " ("), ((reason == NULL)? "" : reason), ((reason == NULL)? "" : ")"), event->id); if (succeeded) { const uint32_t flags = pcmk__node_search_any |pcmk__node_search_cluster_cib; pcmk__node_status_t *peer = pcmk__search_node_caches(0, event->target, NULL, flags); const char *uuid = NULL; if (peer == NULL) { return; } uuid = pcmk__cluster_get_xml_id(peer); if (AM_I_DC) { /* The DC always sends updates */ update_node_state_after_fencing(event->target, uuid); /* @TODO Ideally, at this point, we'd check whether the fenced node * hosted any guest nodes, and call remote_node_down() for them. * Unfortunately, the controller doesn't have a simple, reliable way * to map hosts to guests. It might be possible to track this in the * peer cache via refresh_remote_nodes(). For now, we rely on the * scheduler creating fence pseudo-events for the guests. */ if (!pcmk__str_eq(client, te_client_id, pcmk__str_casei)) { /* Abort the current transition if it wasn't the cluster that * initiated fencing. */ crm_info("External fencing operation from %s fenced %s", client, event->target); abort_transition(PCMK_SCORE_INFINITY, pcmk__graph_restart, "External Fencing Operation", NULL); } } else if (pcmk__str_eq(controld_globals.dc_name, event->target, pcmk__str_null_matches|pcmk__str_casei) && !pcmk_is_set(peer->flags, pcmk__node_status_remote)) { // Assume the target was our DC if we don't currently have one if (controld_globals.dc_name != NULL) { crm_notice("Fencing target %s was our DC", event->target); } else { crm_notice("Fencing target %s may have been our DC", event->target); } /* Given the CIB resyncing that occurs around elections, * have one node update the CIB now and, if the new DC is different, * have them do so too after the election */ if (controld_is_local_node(event->executioner)) { update_node_state_after_fencing(event->target, uuid); } add_stonith_cleanup(event->target); } /* If the target is a remote node, and we host its connection, * immediately fail all monitors so it can be recovered quickly. * The connection won't necessarily drop when a remote node is fenced, * so the failure might not otherwise be detected until the next poke. */ if (pcmk_is_set(peer->flags, pcmk__node_status_remote)) { remote_ra_fail(event->target); } crmd_peer_down(peer, TRUE); } } /*! * \brief Connect to fencer * * \param[in] user_data If NULL, retry failures now, otherwise retry in mainloop timer * * \return G_SOURCE_REMOVE on success, G_SOURCE_CONTINUE to retry * \note If user_data is NULL, this will wait 2s between attempts, for up to * 30 attempts, meaning the controller could be blocked as long as 58s. */ gboolean controld_timer_fencer_connect(gpointer user_data) { int rc = pcmk_ok; if (stonith_api == NULL) { stonith_api = stonith__api_new(); if (stonith_api == NULL) { crm_err("Could not connect to fencer: API memory allocation failed"); return G_SOURCE_REMOVE; } } if (stonith_api->state != stonith_disconnected) { crm_trace("Already connected to fencer, no need to retry"); return G_SOURCE_REMOVE; } if (user_data == NULL) { // Blocking (retry failures now until successful) rc = stonith__api_connect_retry(stonith_api, crm_system_name, 30); if (rc != pcmk_rc_ok) { crm_err("Could not connect to fencer in 30 attempts: %s " QB_XS " rc=%d", pcmk_rc_str(rc), rc); } } else { // Non-blocking (retry failures later in main loop) rc = stonith_api->cmds->connect(stonith_api, crm_system_name, NULL); if (controld_fencer_connect_timer == NULL) { controld_fencer_connect_timer = mainloop_timer_add("controld_fencer_connect", 1000, TRUE, controld_timer_fencer_connect, GINT_TO_POINTER(TRUE)); } if (rc != pcmk_ok) { if (pcmk_is_set(controld_globals.fsa_input_register, R_ST_REQUIRED)) { crm_notice("Fencer connection failed (will retry): %s " QB_XS " rc=%d", pcmk_strerror(rc), rc); if (!mainloop_timer_running(controld_fencer_connect_timer)) { mainloop_timer_start(controld_fencer_connect_timer); } return G_SOURCE_CONTINUE; } else { crm_info("Fencer connection failed (ignoring because no longer required): %s " QB_XS " rc=%d", pcmk_strerror(rc), rc); } return G_SOURCE_REMOVE; } } if (rc == pcmk_ok) { stonith_api_operations_t *cmds = stonith_api->cmds; cmds->register_notification(stonith_api, PCMK__VALUE_ST_NOTIFY_DISCONNECT, tengine_stonith_connection_destroy); cmds->register_notification(stonith_api, PCMK__VALUE_ST_NOTIFY_FENCE, handle_fence_notification); cmds->register_notification(stonith_api, PCMK__VALUE_ST_NOTIFY_HISTORY_SYNCED, tengine_stonith_history_synced); te_trigger_stonith_history_sync(TRUE); crm_notice("Fencer successfully connected"); } return G_SOURCE_REMOVE; } void controld_disconnect_fencer(bool destroy) { if (stonith_api) { // Prevent fencer connection from coming up again controld_clear_fsa_input_flags(R_ST_REQUIRED); if (stonith_api->state != stonith_disconnected) { stonith_api->cmds->disconnect(stonith_api); } stonith_api->cmds->remove_notification(stonith_api, NULL); } if (destroy) { if (stonith_api) { stonith_api->cmds->free(stonith_api); stonith_api = NULL; } if (controld_fencer_connect_timer) { mainloop_timer_del(controld_fencer_connect_timer); controld_fencer_connect_timer = NULL; } if (te_client_id) { free(te_client_id); te_client_id = NULL; } } } static gboolean do_stonith_history_sync(gpointer user_data) { if (stonith_api && (stonith_api->state != stonith_disconnected)) { stonith_history_t *history = NULL; te_cleanup_stonith_history_sync(stonith_api, FALSE); stonith_api->cmds->history(stonith_api, st_opt_sync_call | st_opt_broadcast, NULL, &history, 5); stonith__history_free(history); return TRUE; } else { crm_info("Skip triggering stonith history-sync as stonith is disconnected"); return FALSE; } } static void tengine_stonith_callback(stonith_t *stonith, stonith_callback_data_t *data) { char *uuid = NULL; int stonith_id = -1; int transition_id = -1; pcmk__graph_action_t *action = NULL; const char *target = NULL; if ((data == NULL) || (data->userdata == NULL)) { crm_err("Ignoring fence operation %d result: " "No transition key given (bug?)", ((data == NULL)? -1 : data->call_id)); return; } if (!AM_I_DC) { const char *reason = stonith__exit_reason(data); if (reason == NULL) { reason = pcmk_exec_status_str(stonith__execution_status(data)); } crm_notice("Result of fence operation %d: %d (%s) " QB_XS " key=%s", data->call_id, stonith__exit_status(data), reason, (const char *) data->userdata); return; } CRM_CHECK(decode_transition_key(data->userdata, &uuid, &transition_id, &stonith_id, NULL), goto bail); if (controld_globals.transition_graph->complete || (stonith_id < 0) || !pcmk__str_eq(uuid, controld_globals.te_uuid, pcmk__str_none) || (controld_globals.transition_graph->id != transition_id)) { crm_info("Ignoring fence operation %d result: " "Not from current transition " QB_XS " complete=%s action=%d uuid=%s (vs %s) transition=%d (vs %d)", data->call_id, pcmk__btoa(controld_globals.transition_graph->complete), stonith_id, uuid, controld_globals.te_uuid, transition_id, controld_globals.transition_graph->id); goto bail; } action = controld_get_action(stonith_id); if (action == NULL) { crm_err("Ignoring fence operation %d result: " "Action %d not found in transition graph (bug?) " QB_XS " uuid=%s transition=%d", data->call_id, stonith_id, uuid, transition_id); goto bail; } target = crm_element_value(action->xml, PCMK__META_ON_NODE); if (target == NULL) { crm_err("Ignoring fence operation %d result: No target given (bug?)", data->call_id); goto bail; } stop_te_timer(action); if (stonith__exit_status(data) == CRM_EX_OK) { const char *uuid = crm_element_value(action->xml, PCMK__META_ON_NODE_UUID); const char *op = crm_meta_value(action->params, PCMK__META_STONITH_ACTION); crm_info("Fence operation %d for %s succeeded", data->call_id, target); if (!(pcmk_is_set(action->flags, pcmk__graph_action_confirmed))) { te_action_confirmed(action, NULL); if (pcmk__str_eq(PCMK_ACTION_ON, op, pcmk__str_casei)) { const char *value = NULL; char *now = pcmk__ttoa(time(NULL)); gboolean is_remote_node = FALSE; /* This check is not 100% reliable, since this node is not * guaranteed to have the remote node cached. However, it * doesn't have to be reliable, since the attribute manager can * learn a node's "remoteness" by other means sooner or later. * This allows it to learn more quickly if this node does have * the information. */ if (g_hash_table_lookup(pcmk__remote_peer_cache, uuid) != NULL) { is_remote_node = TRUE; } update_attrd(target, CRM_ATTR_UNFENCED, now, NULL, is_remote_node); free(now); value = crm_meta_value(action->params, PCMK__META_DIGESTS_ALL); update_attrd(target, CRM_ATTR_DIGESTS_ALL, value, NULL, is_remote_node); value = crm_meta_value(action->params, PCMK__META_DIGESTS_SECURE); update_attrd(target, CRM_ATTR_DIGESTS_SECURE, value, NULL, is_remote_node); } else if (!(pcmk_is_set(action->flags, pcmk__graph_action_sent_update))) { update_node_state_after_fencing(target, uuid); pcmk__set_graph_action_flags(action, pcmk__graph_action_sent_update); } } st_fail_count_reset(target); } else { enum pcmk__graph_next abort_action = pcmk__graph_restart; int status = stonith__execution_status(data); const char *reason = stonith__exit_reason(data); if (reason == NULL) { if (status == PCMK_EXEC_DONE) { reason = "Agent returned error"; } else { reason = pcmk_exec_status_str(status); } } pcmk__set_graph_action_flags(action, pcmk__graph_action_failed); /* If no fence devices were available, there's no use in immediately * checking again, so don't start a new transition in that case. */ if (status == PCMK_EXEC_NO_FENCE_DEVICE) { crm_warn("Fence operation %d for %s failed: %s " "(aborting transition and giving up for now)", data->call_id, target, reason); abort_action = pcmk__graph_wait; } else { crm_notice("Fence operation %d for %s failed: %s " "(aborting transition)", data->call_id, target, reason); } /* Increment the fail count now, so abort_for_stonith_failure() can * check it. Non-DC nodes will increment it in * handle_fence_notification(). */ st_fail_count_increment(target); abort_for_stonith_failure(abort_action, target, NULL); } pcmk__update_graph(controld_globals.transition_graph, action); trigger_graph(); bail: free(data->userdata); free(uuid); return; } static int fence_with_delay(const char *target, const char *type, int delay) { uint32_t options = st_opt_none; // Group of enum stonith_call_options int timeout_sec = pcmk__timeout_ms2s(controld_globals.transition_graph->stonith_timeout); if (crmd_join_phase_count(controld_join_confirmed) == 1) { stonith__set_call_options(options, target, st_opt_allow_self_fencing); } return stonith_api->cmds->fence_with_delay(stonith_api, options, target, type, timeout_sec, 0, delay); } /*! * \internal * \brief Execute a fencing action from a transition graph * * \param[in] graph Transition graph being executed (ignored) * \param[in] action Fencing action to execute * * \return Standard Pacemaker return code */ int controld_execute_fence_action(pcmk__graph_t *graph, pcmk__graph_action_t *action) { int rc = 0; const char *id = pcmk__xe_id(action->xml); const char *uuid = crm_element_value(action->xml, PCMK__META_ON_NODE_UUID); const char *target = crm_element_value(action->xml, PCMK__META_ON_NODE); const char *type = crm_meta_value(action->params, PCMK__META_STONITH_ACTION); char *transition_key = NULL; const char *priority_delay = NULL; int delay_i = 0; gboolean invalid_action = FALSE; int stonith_timeout = pcmk__timeout_ms2s(controld_globals.transition_graph->stonith_timeout); CRM_CHECK(id != NULL, invalid_action = TRUE); CRM_CHECK(uuid != NULL, invalid_action = TRUE); CRM_CHECK(type != NULL, invalid_action = TRUE); CRM_CHECK(target != NULL, invalid_action = TRUE); if (invalid_action) { crm_log_xml_warn(action->xml, "BadAction"); return EPROTO; } priority_delay = crm_meta_value(action->params, PCMK_OPT_PRIORITY_FENCING_DELAY); crm_notice("Requesting fencing (%s) targeting node %s " QB_XS " action=%s timeout=%i%s%s", type, target, id, stonith_timeout, priority_delay ? " priority_delay=" : "", priority_delay ? priority_delay : ""); /* Passing NULL means block until we can connect... */ controld_timer_fencer_connect(NULL); pcmk__scan_min_int(priority_delay, &delay_i, 0); rc = fence_with_delay(target, type, delay_i); transition_key = pcmk__transition_key(controld_globals.transition_graph->id, action->id, 0, controld_globals.te_uuid), stonith_api->cmds->register_callback(stonith_api, rc, (stonith_timeout + (delay_i > 0 ? delay_i : 0)), st_opt_timeout_updates, transition_key, "tengine_stonith_callback", tengine_stonith_callback); return pcmk_rc_ok; } bool controld_verify_stonith_watchdog_timeout(const char *value) { long long st_timeout = (value != NULL)? crm_get_msec(value) : 0; const char *our_nodename = controld_globals.cluster->priv->node_name; if (st_timeout == 0 || (stonith_api && (stonith_api->state != stonith_disconnected) && stonith__watchdog_fencing_enabled_for_node_api(stonith_api, our_nodename))) { return pcmk__valid_stonith_watchdog_timeout(value); } return true; } /* end stonith API client functions */ /* * stonith history synchronization * * Each node's fencer keeps track of a cluster-wide fencing history. When a node * joins or leaves, we need to synchronize the history across all nodes. */ static crm_trigger_t *stonith_history_sync_trigger = NULL; static mainloop_timer_t *stonith_history_sync_timer_short = NULL; static mainloop_timer_t *stonith_history_sync_timer_long = NULL; void te_cleanup_stonith_history_sync(stonith_t *st, bool free_timers) { if (free_timers) { mainloop_timer_del(stonith_history_sync_timer_short); stonith_history_sync_timer_short = NULL; mainloop_timer_del(stonith_history_sync_timer_long); stonith_history_sync_timer_long = NULL; } else { mainloop_timer_stop(stonith_history_sync_timer_short); mainloop_timer_stop(stonith_history_sync_timer_long); } if (st) { st->cmds->remove_notification(st, PCMK__VALUE_ST_NOTIFY_HISTORY_SYNCED); } } static void tengine_stonith_history_synced(stonith_t *st, stonith_event_t *st_event) { te_cleanup_stonith_history_sync(st, FALSE); crm_debug("Fence-history synced - cancel all timers"); } static gboolean stonith_history_sync_set_trigger(gpointer user_data) { mainloop_set_trigger(stonith_history_sync_trigger); return FALSE; } void te_trigger_stonith_history_sync(bool long_timeout) { /* trigger a sync in 5s to give more nodes the * chance to show up so that we don't create * unnecessary stonith-history-sync traffic * * the long timeout of 30s is there as a fallback * so that after a successful connection to fenced * we will wait for 30s for the DC to trigger a * history-sync * if this doesn't happen we trigger a sync locally * (e.g. fenced segfaults and is restarted by pacemakerd) */ /* as we are finally checking the stonith-connection * in do_stonith_history_sync we should be fine * leaving stonith_history_sync_time & stonith_history_sync_trigger * around */ if (stonith_history_sync_trigger == NULL) { stonith_history_sync_trigger = mainloop_add_trigger(G_PRIORITY_LOW, do_stonith_history_sync, NULL); } if (long_timeout) { if(stonith_history_sync_timer_long == NULL) { stonith_history_sync_timer_long = mainloop_timer_add("history_sync_long", 30000, FALSE, stonith_history_sync_set_trigger, NULL); } crm_info("Fence history will be synchronized cluster-wide within 30 seconds"); mainloop_timer_start(stonith_history_sync_timer_long); } else { if(stonith_history_sync_timer_short == NULL) { stonith_history_sync_timer_short = mainloop_timer_add("history_sync_short", 5000, FALSE, stonith_history_sync_set_trigger, NULL); } crm_info("Fence history will be synchronized cluster-wide within 5 seconds"); mainloop_timer_start(stonith_history_sync_timer_short); } } /* end stonith history synchronization functions */ diff --git a/daemons/controld/controld_join_dc.c b/daemons/controld/controld_join_dc.c index a91fbfa3a7..8c494b97e0 100644 --- a/daemons/controld/controld_join_dc.c +++ b/daemons/controld/controld_join_dc.c @@ -1,1096 +1,1095 @@ /* - * Copyright 2004-2024 the Pacemaker project contributors + * Copyright 2004-2025 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include // PRIu32 #include // bool, true, false #include // NULL #include // free(), etc. #include // gboolean, etc. #include // xmlNode #include #include #include #include static char *max_generation_from = NULL; static xmlNodePtr max_generation_xml = NULL; /*! * \internal * \brief Nodes from which a CIB sync has failed since the peer joined * * This table is of the form (node_name -> join_id). \p node_name is * the name of a client node from which a CIB \p sync_from() call has failed in * \p do_dc_join_finalize() since the client joined the cluster as a peer. * \p join_id is the ID of the join round in which the \p sync_from() failed, * and is intended for use in nack log messages. */ static GHashTable *failed_sync_nodes = NULL; void finalize_join_for(gpointer key, gpointer value, gpointer user_data); void finalize_sync_callback(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data); gboolean check_join_state(enum crmd_fsa_state cur_state, const char *source); /* Numeric counter used to identify join rounds (an unsigned int would be * appropriate, except we get and set it in XML as int) */ static int current_join_id = 0; /*! * \internal * \brief Get log-friendly string equivalent of a controller group join phase * * \param[in] phase Join phase * * \return Log-friendly string equivalent of \p phase */ static const char * join_phase_text(enum controld_join_phase phase) { switch (phase) { case controld_join_nack: return "nack"; case controld_join_none: return "none"; case controld_join_welcomed: return "welcomed"; case controld_join_integrated: return "integrated"; case controld_join_finalized: return "finalized"; case controld_join_confirmed: return "confirmed"; default: return "invalid"; } } /*! * \internal * \brief Destroy the hash table containing failed sync nodes */ void controld_destroy_failed_sync_table(void) { if (failed_sync_nodes != NULL) { g_hash_table_destroy(failed_sync_nodes); failed_sync_nodes = NULL; } } /*! * \internal * \brief Remove a node from the failed sync nodes table if present * * \param[in] node_name Node name to remove */ void controld_remove_failed_sync_node(const char *node_name) { if (failed_sync_nodes != NULL) { g_hash_table_remove(failed_sync_nodes, (gchar *) node_name); } } /*! * \internal * \brief Add to a hash table a node whose CIB failed to sync * * \param[in] node_name Name of node whose CIB failed to sync * \param[in] join_id Join round when the failure occurred */ static void record_failed_sync_node(const char *node_name, gint join_id) { if (failed_sync_nodes == NULL) { failed_sync_nodes = pcmk__strikey_table(g_free, NULL); } /* If the node is already in the table then we failed to nack it during the * filter offer step */ CRM_LOG_ASSERT(g_hash_table_insert(failed_sync_nodes, g_strdup(node_name), GINT_TO_POINTER(join_id))); } /*! * \internal * \brief Look up a node name in the failed sync table * * \param[in] node_name Name of node to look up * \param[out] join_id Where to store the join ID of when the sync failed * * \return Standard Pacemaker return code. Specifically, \p pcmk_rc_ok if the * node name was found, or \p pcmk_rc_node_unknown otherwise. * \note \p *join_id is set to -1 if the node is not found. */ static int lookup_failed_sync_node(const char *node_name, gint *join_id) { *join_id = -1; if (failed_sync_nodes != NULL) { gpointer result = g_hash_table_lookup(failed_sync_nodes, (gchar *) node_name); if (result != NULL) { *join_id = GPOINTER_TO_INT(result); return pcmk_rc_ok; } } return pcmk_rc_node_unknown; } void crm_update_peer_join(const char *source, pcmk__node_status_t *node, enum controld_join_phase phase) { enum controld_join_phase last = controld_get_join_phase(node); CRM_CHECK(node != NULL, return); /* Remote nodes do not participate in joins */ if (pcmk_is_set(node->flags, pcmk__node_status_remote)) { return; } if (phase == last) { crm_trace("Node %s join-%d phase is still %s " QB_XS " nodeid=%" PRIu32 " source=%s", node->name, current_join_id, join_phase_text(last), node->cluster_layer_id, source); return; } if ((phase <= controld_join_none) || (phase == (last + 1))) { struct controld_node_status_data *data = NULL; if (node->user_data == NULL) { node->user_data = pcmk__assert_alloc(1, sizeof(struct controld_node_status_data)); } data = node->user_data; data->join_phase = phase; crm_trace("Node %s join-%d phase is now %s (was %s) " QB_XS " nodeid=%" PRIu32 " source=%s", node->name, current_join_id, join_phase_text(phase), join_phase_text(last), node->cluster_layer_id, source); return; } crm_warn("Rejecting join-%d phase update for node %s because can't go from " "%s to %s " QB_XS " nodeid=%" PRIu32 " source=%s", current_join_id, node->name, join_phase_text(last), join_phase_text(phase), node->cluster_layer_id, source); } static void start_join_round(void) { GHashTableIter iter; pcmk__node_status_t *peer = NULL; crm_debug("Starting new join round join-%d", current_join_id); g_hash_table_iter_init(&iter, pcmk__peer_cache); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &peer)) { crm_update_peer_join(__func__, peer, controld_join_none); } if (max_generation_from != NULL) { free(max_generation_from); max_generation_from = NULL; } if (max_generation_xml != NULL) { pcmk__xml_free(max_generation_xml); max_generation_xml = NULL; } controld_clear_fsa_input_flags(R_HAVE_CIB); } /*! * \internal * \brief Create a join message from the DC * * \param[in] join_op Join operation name * \param[in] host_to Recipient of message */ static xmlNode * create_dc_message(const char *join_op, const char *host_to) { xmlNode *msg = pcmk__new_request(pcmk_ipc_controld, CRM_SYSTEM_DC, host_to, CRM_SYSTEM_CRMD, join_op, NULL); /* Identify which election this is a part of */ crm_xml_add_int(msg, PCMK__XA_JOIN_ID, current_join_id); /* Add a field specifying whether the DC is shutting down. This keeps the * joining node from fencing the old DC if it becomes the new DC. */ pcmk__xe_set_bool_attr(msg, PCMK__XA_DC_LEAVING, pcmk_is_set(controld_globals.fsa_input_register, R_SHUTDOWN)); return msg; } static void join_make_offer(gpointer key, gpointer value, gpointer user_data) { /* @TODO We don't use user_data except to distinguish one particular call * from others. Make this clearer. */ xmlNode *offer = NULL; pcmk__node_status_t *member = (pcmk__node_status_t *) value; pcmk__assert(member != NULL); if (!pcmk__cluster_is_node_active(member)) { crm_info("Not making join-%d offer to inactive node %s", current_join_id, pcmk__s(member->name, "with unknown name")); if ((member->expected == NULL) && pcmk__str_eq(member->state, PCMK__VALUE_LOST, pcmk__str_none)) { /* You would think this unsafe, but in fact this plus an * active resource is what causes it to be fenced. * * Yes, this does mean that any node that dies at the same * time as the old DC and is not running resource (still) * won't be fenced. * * I'm not happy about this either. */ pcmk__update_peer_expected(__func__, member, CRMD_JOINSTATE_DOWN); } return; } if (member->name == NULL) { crm_info("Not making join-%d offer to node uuid %s with unknown name", current_join_id, member->xml_id); return; } if (controld_globals.membership_id != controld_globals.peer_seq) { controld_globals.membership_id = controld_globals.peer_seq; crm_info("Making join-%d offers based on membership event %llu", current_join_id, controld_globals.peer_seq); } if (user_data != NULL) { enum controld_join_phase phase = controld_get_join_phase(member); if (phase > controld_join_none) { crm_info("Not making join-%d offer to already known node %s (%s)", current_join_id, member->name, join_phase_text(phase)); return; } } crm_update_peer_join(__func__, (pcmk__node_status_t*) member, controld_join_none); offer = create_dc_message(CRM_OP_JOIN_OFFER, member->name); // Advertise our feature set so the joining node can bail if not compatible crm_xml_add(offer, PCMK_XA_CRM_FEATURE_SET, CRM_FEATURE_SET); crm_info("Sending join-%d offer to %s", current_join_id, member->name); pcmk__cluster_send_message(member, pcmk_ipc_controld, offer); pcmk__xml_free(offer); crm_update_peer_join(__func__, member, controld_join_welcomed); } /* A_DC_JOIN_OFFER_ALL */ void do_dc_join_offer_all(long long action, enum crmd_fsa_cause cause, enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data) { int count; /* Reset everyone's status back to down or in_ccm in the CIB. * Any nodes that are active in the CIB but not in the cluster membership * will be seen as offline by the scheduler anyway. */ current_join_id++; start_join_round(); update_dc(NULL); if (cause == C_HA_MESSAGE && current_input == I_NODE_JOIN) { crm_info("A new node joined the cluster"); } g_hash_table_foreach(pcmk__peer_cache, join_make_offer, NULL); count = crmd_join_phase_count(controld_join_welcomed); crm_info("Waiting on join-%d requests from %d outstanding node%s", current_join_id, count, pcmk__plural_s(count)); // Don't waste time by invoking the scheduler yet } /* A_DC_JOIN_OFFER_ONE */ void do_dc_join_offer_one(long long action, enum crmd_fsa_cause cause, enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data) { pcmk__node_status_t *member = NULL; ha_msg_input_t *welcome = NULL; int count; const char *join_to = NULL; if (msg_data->data == NULL) { crm_info("Making join-%d offers to any unconfirmed nodes " "because an unknown node joined", current_join_id); g_hash_table_foreach(pcmk__peer_cache, join_make_offer, &member); check_join_state(cur_state, __func__); return; } welcome = fsa_typed_data(fsa_dt_ha_msg); if (welcome == NULL) { // fsa_typed_data() already logged an error return; } join_to = crm_element_value(welcome->msg, PCMK__XA_SRC); if (join_to == NULL) { crm_err("Can't make join-%d offer to unknown node", current_join_id); return; } member = pcmk__get_node(0, join_to, NULL, pcmk__node_search_cluster_member); /* It is possible that a node will have been sick or starting up when the * original offer was made. However, it will either re-announce itself in * due course, or we can re-store the original offer on the client. */ crm_update_peer_join(__func__, member, controld_join_none); join_make_offer(NULL, member, NULL); /* If the offer isn't to the local node, make an offer to the local node as * well, to ensure the correct value for max_generation_from. */ if (!controld_is_local_node(join_to)) { member = controld_get_local_node_status(); join_make_offer(NULL, member, NULL); } /* This was a genuine join request; cancel any existing transition and * invoke the scheduler. */ abort_transition(PCMK_SCORE_INFINITY, pcmk__graph_restart, "Node join", NULL); count = crmd_join_phase_count(controld_join_welcomed); crm_info("Waiting on join-%d requests from %d outstanding node%s", current_join_id, count, pcmk__plural_s(count)); // Don't waste time by invoking the scheduler yet } static int compare_int_fields(xmlNode * left, xmlNode * right, const char *field) { const char *elem_l = crm_element_value(left, field); const char *elem_r = crm_element_value(right, field); long long int_elem_l; long long int_elem_r; int rc = pcmk_rc_ok; rc = pcmk__scan_ll(elem_l, &int_elem_l, -1LL); if (rc != pcmk_rc_ok) { // Shouldn't be possible crm_warn("Comparing current CIB %s as -1 " "because '%s' is not an integer", field, elem_l); } rc = pcmk__scan_ll(elem_r, &int_elem_r, -1LL); if (rc != pcmk_rc_ok) { // Shouldn't be possible crm_warn("Comparing joining node's CIB %s as -1 " "because '%s' is not an integer", field, elem_r); } if (int_elem_l < int_elem_r) { return -1; } else if (int_elem_l > int_elem_r) { return 1; } return 0; } /* A_DC_JOIN_PROCESS_REQ */ void do_dc_join_filter_offer(long long action, enum crmd_fsa_cause cause, enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data) { xmlNode *generation = NULL; int cmp = 0; int join_id = -1; int count = 0; gint value = 0; gboolean ack_nack_bool = TRUE; ha_msg_input_t *join_ack = fsa_typed_data(fsa_dt_ha_msg); const char *join_from = crm_element_value(join_ack->msg, PCMK__XA_SRC); const char *ref = crm_element_value(join_ack->msg, PCMK_XA_REFERENCE); const char *join_version = crm_element_value(join_ack->msg, PCMK_XA_CRM_FEATURE_SET); pcmk__node_status_t *join_node = NULL; if (join_from == NULL) { crm_err("Ignoring invalid join request without node name"); return; } join_node = pcmk__get_node(0, join_from, NULL, pcmk__node_search_cluster_member); crm_element_value_int(join_ack->msg, PCMK__XA_JOIN_ID, &join_id); if (join_id != current_join_id) { crm_debug("Ignoring join-%d request from %s because we are on join-%d", join_id, join_from, current_join_id); check_join_state(cur_state, __func__); return; } generation = join_ack->xml; if (max_generation_xml != NULL && generation != NULL) { int lpc = 0; const char *attributes[] = { PCMK_XA_ADMIN_EPOCH, PCMK_XA_EPOCH, PCMK_XA_NUM_UPDATES, }; /* It's not obvious that join_ack->xml is the PCMK__XE_GENERATION_TUPLE * element from the join client. The "if" guard is for clarity. */ if (pcmk__xe_is(generation, PCMK__XE_GENERATION_TUPLE)) { for (lpc = 0; cmp == 0 && lpc < PCMK__NELEM(attributes); lpc++) { cmp = compare_int_fields(max_generation_xml, generation, attributes[lpc]); } } else { // Should always be PCMK__XE_GENERATION_TUPLE CRM_LOG_ASSERT(false); } } if (ref == NULL) { ref = "none"; // for logging only } if (lookup_failed_sync_node(join_from, &value) == pcmk_rc_ok) { crm_err("Rejecting join-%d request from node %s because we failed to " "sync its CIB in join-%d " QB_XS " ref=%s", join_id, join_from, value, ref); ack_nack_bool = FALSE; } else if (!pcmk__cluster_is_node_active(join_node)) { if (match_down_event(join_from) != NULL) { /* The join request was received after the node was fenced or * otherwise shutdown in a way that we're aware of. No need to log * an error in this rare occurrence; we know the client was recently * shut down, and receiving a lingering in-flight request is not * cause for alarm. */ crm_debug("Rejecting join-%d request from inactive node %s " QB_XS " ref=%s", join_id, join_from, ref); } else { crm_err("Rejecting join-%d request from inactive node %s " QB_XS " ref=%s", join_id, join_from, ref); } ack_nack_bool = FALSE; } else if (generation == NULL) { crm_err("Rejecting invalid join-%d request from node %s " "missing CIB generation " QB_XS " ref=%s", join_id, join_from, ref); ack_nack_bool = FALSE; } else if ((join_version == NULL) || !feature_set_compatible(CRM_FEATURE_SET, join_version)) { crm_err("Rejecting join-%d request from node %s because feature set %s" " is incompatible with ours (%s) " QB_XS " ref=%s", join_id, join_from, (join_version? join_version : "pre-3.1.0"), CRM_FEATURE_SET, ref); ack_nack_bool = FALSE; } else if (max_generation_xml == NULL) { const char *validation = crm_element_value(generation, PCMK_XA_VALIDATE_WITH); if (pcmk__get_schema(validation) == NULL) { crm_err("Rejecting join-%d request from %s (with first CIB " "generation) due to %s schema version %s " QB_XS " ref=%s", join_id, join_from, ((validation == NULL)? "missing" : "unknown"), pcmk__s(validation, ""), ref); ack_nack_bool = FALSE; } else { crm_debug("Accepting join-%d request from %s (with first CIB " "generation) " QB_XS " ref=%s", join_id, join_from, ref); max_generation_xml = pcmk__xml_copy(NULL, generation); pcmk__str_update(&max_generation_from, join_from); } } else if ((cmp < 0) || ((cmp == 0) && controld_is_local_node(join_from))) { const char *validation = crm_element_value(generation, PCMK_XA_VALIDATE_WITH); if (pcmk__get_schema(validation) == NULL) { crm_err("Rejecting join-%d request from %s (with better CIB " "generation than current best from %s) due to %s " "schema version %s " QB_XS " ref=%s", join_id, join_from, max_generation_from, ((validation == NULL)? "missing" : "unknown"), pcmk__s(validation, ""), ref); ack_nack_bool = FALSE; } else { crm_debug("Accepting join-%d request from %s (with better CIB " "generation than current best from %s) " QB_XS " ref=%s", join_id, join_from, max_generation_from, ref); crm_log_xml_debug(max_generation_xml, "Old max generation"); crm_log_xml_debug(generation, "New max generation"); pcmk__xml_free(max_generation_xml); max_generation_xml = pcmk__xml_copy(NULL, join_ack->xml); pcmk__str_update(&max_generation_from, join_from); } } else { crm_debug("Accepting join-%d request from %s " QB_XS " ref=%s", join_id, join_from, ref); } if (!ack_nack_bool) { crm_update_peer_join(__func__, join_node, controld_join_nack); pcmk__update_peer_expected(__func__, join_node, CRMD_JOINSTATE_NACK); } else { crm_update_peer_join(__func__, join_node, controld_join_integrated); pcmk__update_peer_expected(__func__, join_node, CRMD_JOINSTATE_MEMBER); } count = crmd_join_phase_count(controld_join_integrated); crm_debug("%d node%s currently integrated in join-%d", count, pcmk__plural_s(count), join_id); if (check_join_state(cur_state, __func__) == FALSE) { // Don't waste time by invoking the scheduler yet count = crmd_join_phase_count(controld_join_welcomed); crm_debug("Waiting on join-%d requests from %d outstanding node%s", join_id, count, pcmk__plural_s(count)); } } /* A_DC_JOIN_FINALIZE */ void do_dc_join_finalize(long long action, enum crmd_fsa_cause cause, enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data) { char *sync_from = NULL; int rc = pcmk_ok; int count_welcomed = crmd_join_phase_count(controld_join_welcomed); int count_finalizable = crmd_join_phase_count(controld_join_integrated) + crmd_join_phase_count(controld_join_nack); /* This we can do straight away and avoid clients timing us out * while we compute the latest CIB */ if (count_welcomed != 0) { crm_debug("Waiting on join-%d requests from %d outstanding node%s " "before finalizing join", current_join_id, count_welcomed, pcmk__plural_s(count_welcomed)); crmd_join_phase_log(LOG_DEBUG); /* crmd_fsa_stall(FALSE); Needed? */ return; } else if (count_finalizable == 0) { crm_debug("Finalization not needed for join-%d at the current time", current_join_id); crmd_join_phase_log(LOG_DEBUG); check_join_state(controld_globals.fsa_state, __func__); return; } controld_clear_fsa_input_flags(R_HAVE_CIB); if ((max_generation_from == NULL) || controld_is_local_node(max_generation_from)) { controld_set_fsa_input_flags(R_HAVE_CIB); } if (!controld_globals.transition_graph->complete) { crm_warn("Delaying join-%d finalization while transition in progress", current_join_id); crmd_join_phase_log(LOG_DEBUG); crmd_fsa_stall(FALSE); return; } if (pcmk_is_set(controld_globals.fsa_input_register, R_HAVE_CIB)) { // Send our CIB out to everyone sync_from = pcmk__str_copy(controld_globals.cluster->priv->node_name); } else { // Ask for the agreed best CIB sync_from = pcmk__str_copy(max_generation_from); } crm_notice("Finalizing join-%d for %d node%s (sync'ing CIB %s.%s.%s " "with schema %s and feature set %s from %s)", current_join_id, count_finalizable, pcmk__plural_s(count_finalizable), crm_element_value(max_generation_xml, PCMK_XA_ADMIN_EPOCH), crm_element_value(max_generation_xml, PCMK_XA_EPOCH), crm_element_value(max_generation_xml, PCMK_XA_NUM_UPDATES), crm_element_value(max_generation_xml, PCMK_XA_VALIDATE_WITH), crm_element_value(max_generation_xml, PCMK_XA_CRM_FEATURE_SET), sync_from); crmd_join_phase_log(LOG_DEBUG); rc = controld_globals.cib_conn->cmds->sync_from(controld_globals.cib_conn, sync_from, NULL, cib_none); fsa_register_cib_callback(rc, sync_from, finalize_sync_callback); } void free_max_generation(void) { free(max_generation_from); max_generation_from = NULL; pcmk__xml_free(max_generation_xml); max_generation_xml = NULL; } void finalize_sync_callback(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data) { CRM_LOG_ASSERT(-EPERM != rc); if (rc != pcmk_ok) { const char *sync_from = (const char *) user_data; do_crm_log(((rc == -pcmk_err_old_data)? LOG_WARNING : LOG_ERR), "Could not sync CIB from %s in join-%d: %s", sync_from, current_join_id, pcmk_strerror(rc)); if (rc != -pcmk_err_old_data) { record_failed_sync_node(sync_from, current_join_id); } /* restart the whole join process */ register_fsa_error_adv(C_FSA_INTERNAL, I_ELECTION_DC, NULL, NULL, __func__); } else if (!AM_I_DC) { crm_debug("Sync'ed CIB for join-%d but no longer DC", current_join_id); } else if (controld_globals.fsa_state != S_FINALIZE_JOIN) { crm_debug("Sync'ed CIB for join-%d but no longer in S_FINALIZE_JOIN " "(%s)", current_join_id, fsa_state2string(controld_globals.fsa_state)); } else { controld_set_fsa_input_flags(R_HAVE_CIB); /* make sure dc_uuid is re-set to us */ if (!check_join_state(controld_globals.fsa_state, __func__)) { int count_finalizable = 0; count_finalizable = crmd_join_phase_count(controld_join_integrated) + crmd_join_phase_count(controld_join_nack); crm_debug("Notifying %d node%s of join-%d results", count_finalizable, pcmk__plural_s(count_finalizable), current_join_id); g_hash_table_foreach(pcmk__peer_cache, finalize_join_for, NULL); } } } static void join_node_state_commit_callback(xmlNode *msg, int call_id, int rc, xmlNode *output, void *user_data) { const char *node = user_data; if (rc != pcmk_ok) { fsa_data_t *msg_data = NULL; // for register_fsa_error() macro crm_crit("join-%d node history update (via CIB call %d) for node %s " "failed: %s", current_join_id, call_id, node, pcmk_strerror(rc)); crm_log_xml_debug(msg, "failed"); register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); } crm_debug("join-%d node history update (via CIB call %d) for node %s " "complete", current_join_id, call_id, node); check_join_state(controld_globals.fsa_state, __func__); } /* A_DC_JOIN_PROCESS_ACK */ void do_dc_join_ack(long long action, enum crmd_fsa_cause cause, enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data) { int join_id = -1; ha_msg_input_t *join_ack = fsa_typed_data(fsa_dt_ha_msg); const char *op = crm_element_value(join_ack->msg, PCMK__XA_CRM_TASK); char *join_from = crm_element_value_copy(join_ack->msg, PCMK__XA_SRC); pcmk__node_status_t *peer = NULL; enum controld_join_phase phase = controld_join_none; - enum controld_section_e section = controld_section_lrm; + const bool unlocked_only = pcmk_is_set(controld_globals.flags, + controld_shutdown_lock_enabled); char *xpath = NULL; xmlNode *state = join_ack->xml; xmlNode *execd_state = NULL; cib_t *cib = controld_globals.cib_conn; int rc = pcmk_ok; // Sanity checks if (join_from == NULL) { crm_warn("Ignoring message received without node identification"); goto done; } if (op == NULL) { crm_warn("Ignoring message received from %s without task", join_from); goto done; } if (strcmp(op, CRM_OP_JOIN_CONFIRM)) { crm_debug("Ignoring '%s' message from %s while waiting for '%s'", op, join_from, CRM_OP_JOIN_CONFIRM); goto done; } if (crm_element_value_int(join_ack->msg, PCMK__XA_JOIN_ID, &join_id) != 0) { crm_warn("Ignoring join confirmation from %s without valid join ID", join_from); goto done; } peer = pcmk__get_node(0, join_from, NULL, pcmk__node_search_cluster_member); phase = controld_get_join_phase(peer); if (phase != controld_join_finalized) { crm_info("Ignoring out-of-sequence join-%d confirmation from %s " "(currently %s not %s)", join_id, join_from, join_phase_text(phase), join_phase_text(controld_join_finalized)); goto done; } if (join_id != current_join_id) { crm_err("Rejecting join-%d confirmation from %s " "because currently on join-%d", join_id, join_from, current_join_id); crm_update_peer_join(__func__, peer, controld_join_nack); goto done; } crm_update_peer_join(__func__, peer, controld_join_confirmed); /* Update CIB with node's current executor state. A new transition will be * triggered later, when the CIB manager notifies us of the change. * * The delete and modify requests are part of an atomic transaction. */ rc = cib->cmds->init_transaction(cib); if (rc != pcmk_ok) { goto done; } // Delete relevant parts of node's current executor state from CIB - if (pcmk_is_set(controld_globals.flags, controld_shutdown_lock_enabled)) { - section = controld_section_lrm_unlocked; - } - controld_node_state_deletion_strings(join_from, section, &xpath, NULL); + controld_node_history_deletion_strings(join_from, unlocked_only, &xpath, + NULL); rc = cib->cmds->remove(cib, xpath, NULL, cib_xpath|cib_multiple|cib_transaction); if (rc != pcmk_ok) { goto done; } // Update CIB with node's latest known executor state if (controld_is_local_node(join_from)) { // Use the latest possible state if processing our own join ack execd_state = controld_query_executor_state(); if (execd_state != NULL) { crm_debug("Updating local node history for join-%d from query " "result", current_join_id); state = execd_state; } else { crm_warn("Updating local node history from join-%d confirmation " "because query failed", current_join_id); } } else { crm_debug("Updating node history for %s from join-%d confirmation", join_from, current_join_id); } rc = cib->cmds->modify(cib, PCMK_XE_STATUS, state, cib_can_create|cib_transaction); pcmk__xml_free(execd_state); if (rc != pcmk_ok) { goto done; } // Commit the transaction rc = cib->cmds->end_transaction(cib, true, cib_none); fsa_register_cib_callback(rc, join_from, join_node_state_commit_callback); if (rc > 0) { // join_from will be freed after callback join_from = NULL; rc = pcmk_ok; } done: if (rc != pcmk_ok) { crm_crit("join-%d node history update for node %s failed: %s", current_join_id, join_from, pcmk_strerror(rc)); register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); } free(join_from); free(xpath); } void finalize_join_for(gpointer key, gpointer value, gpointer user_data) { xmlNode *acknak = NULL; xmlNode *tmp1 = NULL; pcmk__node_status_t *join_node = value; const char *join_to = join_node->name; enum controld_join_phase phase = controld_get_join_phase(join_node); bool integrated = false; switch (phase) { case controld_join_integrated: integrated = true; break; case controld_join_nack: break; default: crm_trace("Not updating non-integrated and non-nacked node %s (%s) " "for join-%d", join_to, join_phase_text(phase), current_join_id); return; } /* Update the element with the node's name and UUID, in case they * weren't known before */ crm_trace("Updating node name and UUID in CIB for %s", join_to); tmp1 = pcmk__xe_create(NULL, PCMK_XE_NODE); crm_xml_add(tmp1, PCMK_XA_ID, pcmk__cluster_get_xml_id(join_node)); crm_xml_add(tmp1, PCMK_XA_UNAME, join_to); fsa_cib_anon_update(PCMK_XE_NODES, tmp1); pcmk__xml_free(tmp1); join_node = pcmk__get_node(0, join_to, NULL, pcmk__node_search_cluster_member); if (!pcmk__cluster_is_node_active(join_node)) { /* * NACK'ing nodes that the membership layer doesn't know about yet * simply creates more churn * * Better to leave them waiting and let the join restart when * the new membership event comes in * * All other NACKs (due to versions etc) should still be processed */ pcmk__update_peer_expected(__func__, join_node, CRMD_JOINSTATE_PENDING); return; } // Acknowledge or nack node's join request crm_debug("%sing join-%d request from %s", integrated? "Acknowledg" : "Nack", current_join_id, join_to); acknak = create_dc_message(CRM_OP_JOIN_ACKNAK, join_to); pcmk__xe_set_bool_attr(acknak, CRM_OP_JOIN_ACKNAK, integrated); if (integrated) { // No change needed for a nacked node crm_update_peer_join(__func__, join_node, controld_join_finalized); pcmk__update_peer_expected(__func__, join_node, CRMD_JOINSTATE_MEMBER); /* Iterate through the remote peer cache and add information on which * node hosts each to the ACK message. This keeps new controllers in * sync with what has already happened. */ if (pcmk__cluster_num_remote_nodes() > 0) { GHashTableIter iter; pcmk__node_status_t *node = NULL; xmlNode *remotes = pcmk__xe_create(acknak, PCMK_XE_NODES); g_hash_table_iter_init(&iter, pcmk__remote_peer_cache); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &node)) { xmlNode *remote = NULL; if (!node->conn_host) { continue; } remote = pcmk__xe_create(remotes, PCMK_XE_NODE); pcmk__xe_set_props(remote, PCMK_XA_ID, node->name, PCMK__XA_NODE_STATE, node->state, PCMK__XA_CONNECTION_HOST, node->conn_host, NULL); } } } pcmk__cluster_send_message(join_node, pcmk_ipc_controld, acknak); pcmk__xml_free(acknak); return; } gboolean check_join_state(enum crmd_fsa_state cur_state, const char *source) { static unsigned long long highest_seq = 0; if (controld_globals.membership_id != controld_globals.peer_seq) { crm_debug("join-%d: Membership changed from %llu to %llu " QB_XS " highest=%llu state=%s for=%s", current_join_id, controld_globals.membership_id, controld_globals.peer_seq, highest_seq, fsa_state2string(cur_state), source); if (highest_seq < controld_globals.peer_seq) { /* Don't spam the FSA with duplicates */ highest_seq = controld_globals.peer_seq; register_fsa_input_before(C_FSA_INTERNAL, I_NODE_JOIN, NULL); } } else if (cur_state == S_INTEGRATION) { if (crmd_join_phase_count(controld_join_welcomed) == 0) { int count = crmd_join_phase_count(controld_join_integrated); crm_debug("join-%d: Integration of %d peer%s complete " QB_XS " state=%s for=%s", current_join_id, count, pcmk__plural_s(count), fsa_state2string(cur_state), source); register_fsa_input_before(C_FSA_INTERNAL, I_INTEGRATED, NULL); return TRUE; } } else if (cur_state == S_FINALIZE_JOIN) { if (!pcmk_is_set(controld_globals.fsa_input_register, R_HAVE_CIB)) { crm_debug("join-%d: Delaying finalization until we have CIB " QB_XS " state=%s for=%s", current_join_id, fsa_state2string(cur_state), source); return TRUE; } else if (crmd_join_phase_count(controld_join_welcomed) != 0) { int count = crmd_join_phase_count(controld_join_welcomed); crm_debug("join-%d: Still waiting on %d welcomed node%s " QB_XS " state=%s for=%s", current_join_id, count, pcmk__plural_s(count), fsa_state2string(cur_state), source); crmd_join_phase_log(LOG_DEBUG); } else if (crmd_join_phase_count(controld_join_integrated) != 0) { int count = crmd_join_phase_count(controld_join_integrated); crm_debug("join-%d: Still waiting on %d integrated node%s " QB_XS " state=%s for=%s", current_join_id, count, pcmk__plural_s(count), fsa_state2string(cur_state), source); crmd_join_phase_log(LOG_DEBUG); } else if (crmd_join_phase_count(controld_join_finalized) != 0) { int count = crmd_join_phase_count(controld_join_finalized); crm_debug("join-%d: Still waiting on %d finalized node%s " QB_XS " state=%s for=%s", current_join_id, count, pcmk__plural_s(count), fsa_state2string(cur_state), source); crmd_join_phase_log(LOG_DEBUG); } else { crm_debug("join-%d: Complete " QB_XS " state=%s for=%s", current_join_id, fsa_state2string(cur_state), source); register_fsa_input_later(C_FSA_INTERNAL, I_FINALIZED, NULL); return TRUE; } } return FALSE; } void do_dc_join_final(long long action, enum crmd_fsa_cause cause, enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data) { crm_debug("Ensuring DC, quorum and node attributes are up-to-date"); crm_update_quorum(pcmk__cluster_has_quorum(), TRUE); } int crmd_join_phase_count(enum controld_join_phase phase) { int count = 0; pcmk__node_status_t *peer; GHashTableIter iter; g_hash_table_iter_init(&iter, pcmk__peer_cache); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &peer)) { if (controld_get_join_phase(peer) == phase) { count++; } } return count; } void crmd_join_phase_log(int level) { pcmk__node_status_t *peer; GHashTableIter iter; g_hash_table_iter_init(&iter, pcmk__peer_cache); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &peer)) { do_crm_log(level, "join-%d: %s=%s", current_join_id, peer->name, join_phase_text(controld_get_join_phase(peer))); } } diff --git a/daemons/controld/controld_remote_ra.c b/daemons/controld/controld_remote_ra.c index 1cc4ae008b..1c52477689 100644 --- a/daemons/controld/controld_remote_ra.c +++ b/daemons/controld/controld_remote_ra.c @@ -1,1493 +1,1472 @@ /* * Copyright 2013-2025 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include // xmlXPathObject, etc. #include #define REMOTE_LRMD_RA "remote" /* The max start timeout before cmd retry */ #define MAX_START_TIMEOUT_MS 10000 #define cmd_set_flags(cmd, flags_to_set) do { \ (cmd)->status = pcmk__set_flags_as(__func__, __LINE__, LOG_TRACE, \ "Remote command", (cmd)->rsc_id, (cmd)->status, \ (flags_to_set), #flags_to_set); \ } while (0) #define cmd_clear_flags(cmd, flags_to_clear) do { \ (cmd)->status = pcmk__clear_flags_as(__func__, __LINE__, LOG_TRACE, \ "Remote command", (cmd)->rsc_id, (cmd)->status, \ (flags_to_clear), #flags_to_clear); \ } while (0) enum remote_cmd_status { cmd_reported_success = (1 << 0), cmd_cancel = (1 << 1), }; typedef struct remote_ra_cmd_s { /*! the local node the cmd is issued from */ char *owner; /*! the remote node the cmd is executed on */ char *rsc_id; /*! the action to execute */ char *action; /*! some string the client wants us to give it back */ char *userdata; /*! start delay in ms */ int start_delay; /*! timer id used for start delay. */ int delay_id; /*! timeout in ms for cmd */ int timeout; /*! recurring interval in ms */ guint interval_ms; /*! interval timer id */ int interval_id; int monitor_timeout_id; int takeover_timeout_id; /*! action parameters */ lrmd_key_value_t *params; pcmk__action_result_t result; int call_id; time_t start_time; uint32_t status; } remote_ra_cmd_t; #define lrm_remote_set_flags(lrm_state, flags_to_set) do { \ lrm_state_t *lrm = (lrm_state); \ remote_ra_data_t *ra = lrm->remote_ra_data; \ ra->status = pcmk__set_flags_as(__func__, __LINE__, LOG_TRACE, "Remote", \ lrm->node_name, ra->status, \ (flags_to_set), #flags_to_set); \ } while (0) #define lrm_remote_clear_flags(lrm_state, flags_to_clear) do { \ lrm_state_t *lrm = (lrm_state); \ remote_ra_data_t *ra = lrm->remote_ra_data; \ ra->status = pcmk__clear_flags_as(__func__, __LINE__, LOG_TRACE, "Remote", \ lrm->node_name, ra->status, \ (flags_to_clear), #flags_to_clear); \ } while (0) enum remote_status { expect_takeover = (1 << 0), takeover_complete = (1 << 1), remote_active = (1 << 2), /* Maintenance mode is difficult to determine from the controller's context, * so we have it signalled back with the transition from the scheduler. */ remote_in_maint = (1 << 3), /* Similar for whether we are controlling a guest node or remote node. * Fortunately there is a meta-attribute in the transition already and * as the situation doesn't change over time we can use the * resource start for noting down the information for later use when * the attributes aren't at hand. */ controlling_guest = (1 << 4), }; typedef struct remote_ra_data_s { crm_trigger_t *work; remote_ra_cmd_t *cur_cmd; GList *cmds; GList *recurring_cmds; uint32_t status; } remote_ra_data_t; static int handle_remote_ra_start(lrm_state_t * lrm_state, remote_ra_cmd_t * cmd, int timeout_ms); static void handle_remote_ra_stop(lrm_state_t * lrm_state, remote_ra_cmd_t * cmd); static GList *fail_all_monitor_cmds(GList * list); static void free_cmd(gpointer user_data) { remote_ra_cmd_t *cmd = user_data; if (!cmd) { return; } if (cmd->delay_id) { g_source_remove(cmd->delay_id); } if (cmd->interval_id) { g_source_remove(cmd->interval_id); } if (cmd->monitor_timeout_id) { g_source_remove(cmd->monitor_timeout_id); } if (cmd->takeover_timeout_id) { g_source_remove(cmd->takeover_timeout_id); } free(cmd->owner); free(cmd->rsc_id); free(cmd->action); free(cmd->userdata); pcmk__reset_result(&(cmd->result)); lrmd_key_value_freeall(cmd->params); free(cmd); } static int generate_callid(void) { static int remote_ra_callid = 0; remote_ra_callid++; if (remote_ra_callid <= 0) { remote_ra_callid = 1; } return remote_ra_callid; } static gboolean recurring_helper(gpointer data) { remote_ra_cmd_t *cmd = data; lrm_state_t *connection_rsc = NULL; cmd->interval_id = 0; connection_rsc = controld_get_executor_state(cmd->rsc_id, false); if (connection_rsc && connection_rsc->remote_ra_data) { remote_ra_data_t *ra_data = connection_rsc->remote_ra_data; ra_data->recurring_cmds = g_list_remove(ra_data->recurring_cmds, cmd); ra_data->cmds = g_list_append(ra_data->cmds, cmd); mainloop_set_trigger(ra_data->work); } return FALSE; } static gboolean start_delay_helper(gpointer data) { remote_ra_cmd_t *cmd = data; lrm_state_t *connection_rsc = NULL; cmd->delay_id = 0; connection_rsc = controld_get_executor_state(cmd->rsc_id, false); if (connection_rsc && connection_rsc->remote_ra_data) { remote_ra_data_t *ra_data = connection_rsc->remote_ra_data; mainloop_set_trigger(ra_data->work); } return FALSE; } static bool should_purge_attributes(pcmk__node_status_t *node) { pcmk__node_status_t *conn_node = NULL; lrm_state_t *connection_rsc = NULL; if ((node->conn_host == NULL) || (node->name == NULL)) { return true; } /* Get the node that was hosting the remote connection resource from the * peer cache. That's the one we really care about here. */ conn_node = pcmk__get_node(0, node->conn_host, NULL, pcmk__node_search_cluster_member); if (conn_node == NULL) { return true; } /* Check the uptime of connection_rsc. If it hasn't been running long * enough, set purge=true. "Long enough" means it started running earlier * than the timestamp when we noticed it went away in the first place. */ connection_rsc = controld_get_executor_state(node->name, false); if (connection_rsc != NULL) { lrmd_t *lrm = connection_rsc->conn; time_t uptime = lrmd__uptime(lrm); time_t now = time(NULL); /* Add 20s of fuzziness to give corosync a while to notice the remote * host is gone. On various error conditions (failure to get uptime, * peer_lost isn't set) we default to purging. */ if (uptime > 0 && conn_node->peer_lost > 0 && uptime + 20 >= now - conn_node->peer_lost) { return false; } } return true; } -static enum controld_section_e -section_to_delete(bool purge) -{ - if (pcmk_is_set(controld_globals.flags, controld_shutdown_lock_enabled)) { - if (purge) { - return controld_section_all_unlocked; - } else { - return controld_section_lrm_unlocked; - } - } else { - if (purge) { - return controld_section_all; - } else { - return controld_section_lrm; - } - } -} - static void purge_remote_node_attrs(int call_opt, pcmk__node_status_t *node) { - bool purge = should_purge_attributes(node); - enum controld_section_e section = section_to_delete(purge); + const bool unlocked_only = pcmk_is_set(controld_globals.flags, + controld_shutdown_lock_enabled); - /* Purge node from attrd's memory */ - if (purge) { - update_attrd_remote_node_removed(node->name, NULL); + // Purge node's transient attributes (from attribute manager and CIB) + if (should_purge_attributes(node)) { + controld_purge_node_attrs(node->name, true); } - controld_delete_node_state(node->name, section, call_opt); + controld_delete_node_history(node->name, unlocked_only, call_opt); } /*! * \internal * \brief Handle cluster communication related to pacemaker_remote node joining * * \param[in] node_name Name of newly integrated pacemaker_remote node */ static void remote_node_up(const char *node_name) { int call_opt; xmlNode *update, *state; pcmk__node_status_t *node = NULL; lrm_state_t *connection_rsc = NULL; CRM_CHECK(node_name != NULL, return); crm_info("Announcing Pacemaker Remote node %s", node_name); call_opt = crmd_cib_smart_opt(); /* Delete node's CRM_OP_PROBED attribute. Deleting any attribute ensures * that the attribute manager learns the node is remote. Deletion of this * specfic attribute is a holdover from when it had special meaning. * * @COMPAT Find another way to tell attrd that the node is remote, without * risking deletion or overwrite of an arbitrary attribute. Then work on * deprecating CRM_OP_PROBED. */ update_attrd(node_name, CRM_OP_PROBED, NULL, NULL, TRUE); /* Ensure node is in the remote peer cache with member status */ node = pcmk__cluster_lookup_remote_node(node_name); CRM_CHECK((node != NULL) && (node->name != NULL), return); purge_remote_node_attrs(call_opt, node); pcmk__update_peer_state(__func__, node, PCMK_VALUE_MEMBER, 0); /* Apply any start state that we were given from the environment on the * remote node. */ connection_rsc = controld_get_executor_state(node->name, false); if (connection_rsc != NULL) { lrmd_t *lrm = connection_rsc->conn; const char *start_state = lrmd__node_start_state(lrm); if (start_state) { set_join_state(start_state, node->name, node->xml_id, true); } } /* pacemaker_remote nodes don't participate in the membership layer, * so cluster nodes don't automatically get notified when they come and go. * We send a cluster message to the DC, and update the CIB node state entry, * so the DC will get it sooner (via message) or later (via CIB refresh), * and any other interested parties can query the CIB. */ broadcast_remote_state_message(node_name, true); update = pcmk__xe_create(NULL, PCMK_XE_STATUS); state = create_node_state_update(node, controld_node_update_cluster, update, __func__); /* Clear the PCMK__XA_NODE_FENCED flag in the node state. If the node ever * needs to be fenced, this flag will allow various actions to determine * whether the fencing has happened yet. */ crm_xml_add(state, PCMK__XA_NODE_FENCED, "0"); /* TODO: If the remote connection drops, and this (async) CIB update either * failed or has not yet completed, later actions could mistakenly think the * node has already been fenced (if the PCMK__XA_NODE_FENCED attribute was * previously set, because it won't have been cleared). This could prevent * actual fencing or allow recurring monitor failures to be cleared too * soon. Ideally, we wouldn't rely on the CIB for the fenced status. */ controld_update_cib(PCMK_XE_STATUS, update, call_opt, NULL); pcmk__xml_free(update); } enum down_opts { DOWN_KEEP_LRM, DOWN_ERASE_LRM }; /*! * \internal * \brief Handle cluster communication related to pacemaker_remote node leaving * * \param[in] node_name Name of lost node * \param[in] opts Whether to keep or erase LRM history */ static void remote_node_down(const char *node_name, const enum down_opts opts) { xmlNode *update; int call_opt = crmd_cib_smart_opt(); pcmk__node_status_t *node = NULL; - /* Purge node from attrd's memory */ - update_attrd_remote_node_removed(node_name, NULL); + // Purge node's transient attributes (from attribute manager and CIB) + controld_purge_node_attrs(node_name, true); - /* Normally, only node attributes should be erased, and the resource history - * should be kept until the node comes back up. However, after a successful - * fence, we want to clear the history as well, so we don't think resources - * are still running on the node. + /* Normally, the resource history should be kept until the node comes back + * up. However, after a successful fence, clear the history so we don't + * think resources are still running on the node. */ if (opts == DOWN_ERASE_LRM) { - controld_delete_node_state(node_name, controld_section_all, call_opt); - } else { - controld_delete_node_state(node_name, controld_section_attrs, call_opt); + controld_delete_node_history(node_name, false, call_opt); } /* Ensure node is in the remote peer cache with lost state */ node = pcmk__cluster_lookup_remote_node(node_name); CRM_CHECK(node != NULL, return); pcmk__update_peer_state(__func__, node, PCMK__VALUE_LOST, 0); /* Notify DC */ broadcast_remote_state_message(node_name, false); /* Update CIB node state */ update = pcmk__xe_create(NULL, PCMK_XE_STATUS); create_node_state_update(node, controld_node_update_cluster, update, __func__); controld_update_cib(PCMK_XE_STATUS, update, call_opt, NULL); pcmk__xml_free(update); } /*! * \internal * \brief Handle effects of a remote RA command on node state * * \param[in] cmd Completed remote RA command */ static void check_remote_node_state(const remote_ra_cmd_t *cmd) { /* Only successful actions can change node state */ if (!pcmk__result_ok(&(cmd->result))) { return; } if (pcmk__str_eq(cmd->action, PCMK_ACTION_START, pcmk__str_casei)) { remote_node_up(cmd->rsc_id); } else if (pcmk__str_eq(cmd->action, PCMK_ACTION_MIGRATE_FROM, pcmk__str_casei)) { /* After a successful migration, we don't need to do remote_node_up() * because the DC already knows the node is up, and we don't want to * clear LRM history etc. We do need to add the remote node to this * host's remote peer cache, because (unless it happens to be DC) * it hasn't been tracking the remote node, and other code relies on * the cache to distinguish remote nodes from unseen cluster nodes. */ pcmk__node_status_t *node = pcmk__cluster_lookup_remote_node(cmd->rsc_id); CRM_CHECK(node != NULL, return); pcmk__update_peer_state(__func__, node, PCMK_VALUE_MEMBER, 0); } else if (pcmk__str_eq(cmd->action, PCMK_ACTION_STOP, pcmk__str_casei)) { lrm_state_t *lrm_state = controld_get_executor_state(cmd->rsc_id, false); remote_ra_data_t *ra_data = lrm_state? lrm_state->remote_ra_data : NULL; if (ra_data) { if (!pcmk_is_set(ra_data->status, takeover_complete)) { /* Stop means down if we didn't successfully migrate elsewhere */ remote_node_down(cmd->rsc_id, DOWN_KEEP_LRM); } else if (AM_I_DC == FALSE) { /* Only the connection host and DC track node state, * so if the connection migrated elsewhere and we aren't DC, * un-cache the node, so we don't have stale info */ pcmk__cluster_forget_remote_node(cmd->rsc_id); } } } /* We don't do anything for successful monitors, which is correct for * routine recurring monitors, and for monitors on nodes where the * connection isn't supposed to be (the cluster will stop the connection in * that case). However, if the initial probe finds the connection already * active on the node where we want it, we probably should do * remote_node_up(). Unfortunately, we can't distinguish that case here. * Given that connections have to be initiated by the cluster, the chance of * that should be close to zero. */ } static void report_remote_ra_result(remote_ra_cmd_t * cmd) { lrmd_event_data_t op = { 0, }; check_remote_node_state(cmd); op.type = lrmd_event_exec_complete; op.rsc_id = cmd->rsc_id; op.op_type = cmd->action; op.user_data = cmd->userdata; op.timeout = cmd->timeout; op.interval_ms = cmd->interval_ms; op.t_run = cmd->start_time; op.t_rcchange = cmd->start_time; lrmd__set_result(&op, cmd->result.exit_status, cmd->result.execution_status, cmd->result.exit_reason); if (pcmk_is_set(cmd->status, cmd_reported_success) && !pcmk__result_ok(&(cmd->result))) { op.t_rcchange = time(NULL); /* This edge case will likely never ever occur, but if it does the * result is that a failure will not be processed correctly. This is only * remotely possible because we are able to detect a connection resource's tcp * connection has failed at any moment after start has completed. The actual * recurring operation is just a connectivity ping. * * basically, we are not guaranteed that the first successful monitor op and * a subsequent failed monitor op will not occur in the same timestamp. We have to * make it look like the operations occurred at separate times though. */ if (op.t_rcchange == op.t_run) { op.t_rcchange++; } } if (cmd->params) { lrmd_key_value_t *tmp; op.params = pcmk__strkey_table(free, free); for (tmp = cmd->params; tmp; tmp = tmp->next) { pcmk__insert_dup(op.params, tmp->key, tmp->value); } } op.call_id = cmd->call_id; op.remote_nodename = cmd->owner; lrm_op_callback(&op); if (op.params) { g_hash_table_destroy(op.params); } lrmd__reset_result(&op); } /*! * \internal * \brief Return a remote command's remaining timeout in seconds * * \param[in] cmd Remote command to check * * \return Command's remaining timeout in seconds */ static int remaining_timeout_sec(const remote_ra_cmd_t *cmd) { return pcmk__timeout_ms2s(cmd->timeout) - (time(NULL) - cmd->start_time); } static gboolean retry_start_cmd_cb(gpointer data) { lrm_state_t *lrm_state = data; remote_ra_data_t *ra_data = lrm_state->remote_ra_data; remote_ra_cmd_t *cmd = NULL; int rc = ETIME; int remaining = 0; if (!ra_data || !ra_data->cur_cmd) { return FALSE; } cmd = ra_data->cur_cmd; if (!pcmk__is_up_action(cmd->action)) { return FALSE; } remaining = remaining_timeout_sec(cmd); if (remaining > 0) { rc = handle_remote_ra_start(lrm_state, cmd, remaining * 1000); } else { pcmk__set_result(&(cmd->result), PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_TIMEOUT, "Not enough time remains to retry remote connection"); } if (rc != pcmk_rc_ok) { report_remote_ra_result(cmd); if (ra_data->cmds) { mainloop_set_trigger(ra_data->work); } ra_data->cur_cmd = NULL; free_cmd(cmd); } else { /* wait for connection event */ } return FALSE; } static gboolean connection_takeover_timeout_cb(gpointer data) { lrm_state_t *lrm_state = NULL; remote_ra_cmd_t *cmd = data; crm_info("takeover event timed out for node %s", cmd->rsc_id); cmd->takeover_timeout_id = 0; lrm_state = controld_get_executor_state(cmd->rsc_id, false); handle_remote_ra_stop(lrm_state, cmd); free_cmd(cmd); return FALSE; } static gboolean monitor_timeout_cb(gpointer data) { lrm_state_t *lrm_state = NULL; remote_ra_cmd_t *cmd = data; lrm_state = controld_get_executor_state(cmd->rsc_id, false); crm_info("Timed out waiting for remote poke response from %s%s", cmd->rsc_id, (lrm_state? "" : " (no LRM state)")); cmd->monitor_timeout_id = 0; pcmk__set_result(&(cmd->result), PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_TIMEOUT, "Remote executor did not respond"); if (lrm_state && lrm_state->remote_ra_data) { remote_ra_data_t *ra_data = lrm_state->remote_ra_data; if (ra_data->cur_cmd == cmd) { ra_data->cur_cmd = NULL; } if (ra_data->cmds) { mainloop_set_trigger(ra_data->work); } } report_remote_ra_result(cmd); free_cmd(cmd); if(lrm_state) { // @TODO Should we move this before reporting the result above? lrm_state_disconnect(lrm_state); } return FALSE; } static void synthesize_lrmd_success(lrm_state_t *lrm_state, const char *rsc_id, const char *op_type) { lrmd_event_data_t op = { 0, }; if (lrm_state == NULL) { /* if lrm_state not given assume local */ lrm_state = controld_get_executor_state(NULL, false); } pcmk__assert(lrm_state != NULL); op.type = lrmd_event_exec_complete; op.rsc_id = rsc_id; op.op_type = op_type; op.t_run = time(NULL); op.t_rcchange = op.t_run; op.call_id = generate_callid(); lrmd__set_result(&op, PCMK_OCF_OK, PCMK_EXEC_DONE, NULL); process_lrm_event(lrm_state, &op, NULL, NULL); } void remote_lrm_op_callback(lrmd_event_data_t * op) { gboolean cmd_handled = FALSE; lrm_state_t *lrm_state = NULL; remote_ra_data_t *ra_data = NULL; remote_ra_cmd_t *cmd = NULL; CRM_CHECK((op != NULL) && (op->remote_nodename != NULL), return); crm_debug("Processing '%s%s%s' event on remote connection to %s: %s " "(%d) status=%s (%d)", (op->op_type? op->op_type : ""), (op->op_type? " " : ""), lrmd_event_type2str(op->type), op->remote_nodename, crm_exit_str((crm_exit_t) op->rc), op->rc, pcmk_exec_status_str(op->op_status), op->op_status); lrm_state = controld_get_executor_state(op->remote_nodename, false); if (!lrm_state || !lrm_state->remote_ra_data) { crm_debug("No state information found for remote connection event"); return; } ra_data = lrm_state->remote_ra_data; if (op->type == lrmd_event_new_client) { // Another client has connected to the remote daemon if (pcmk_is_set(ra_data->status, expect_takeover)) { // Great, we knew this was coming lrm_remote_clear_flags(lrm_state, expect_takeover); lrm_remote_set_flags(lrm_state, takeover_complete); } else { crm_err("Disconnecting from Pacemaker Remote node %s due to " "unexpected client takeover", op->remote_nodename); /* In this case, lrmd_tls_connection_destroy() will be called under the control of mainloop. */ /* Do not free lrm_state->conn yet. */ /* It'll be freed in the following stop action. */ lrm_state_disconnect_only(lrm_state); } return; } /* filter all EXEC events up */ if (op->type == lrmd_event_exec_complete) { if (pcmk_is_set(ra_data->status, takeover_complete)) { crm_debug("ignoring event, this connection is taken over by another node"); } else { lrm_op_callback(op); } return; } if ((op->type == lrmd_event_disconnect) && (ra_data->cur_cmd == NULL)) { if (!pcmk_is_set(ra_data->status, remote_active)) { crm_debug("Disconnection from Pacemaker Remote node %s complete", lrm_state->node_name); } else if (!remote_ra_is_in_maintenance(lrm_state)) { crm_err("Lost connection to Pacemaker Remote node %s", lrm_state->node_name); ra_data->recurring_cmds = fail_all_monitor_cmds(ra_data->recurring_cmds); ra_data->cmds = fail_all_monitor_cmds(ra_data->cmds); } else { crm_notice("Unmanaged Pacemaker Remote node %s disconnected", lrm_state->node_name); /* Do roughly what a 'stop' on the remote-resource would do */ handle_remote_ra_stop(lrm_state, NULL); remote_node_down(lrm_state->node_name, DOWN_KEEP_LRM); /* now fake the reply of a successful 'stop' */ synthesize_lrmd_success(NULL, lrm_state->node_name, PCMK_ACTION_STOP); } return; } if (!ra_data->cur_cmd) { crm_debug("no event to match"); return; } cmd = ra_data->cur_cmd; /* Start actions and migrate from actions complete after connection * comes back to us. */ if ((op->type == lrmd_event_connect) && pcmk__is_up_action(cmd->action)) { if (op->connection_rc < 0) { int remaining = remaining_timeout_sec(cmd); if ((op->connection_rc == -ENOKEY) || (op->connection_rc == -EKEYREJECTED)) { // Hard error, don't retry pcmk__set_result(&(cmd->result), PCMK_OCF_INVALID_PARAM, PCMK_EXEC_ERROR, pcmk_strerror(op->connection_rc)); } else if (remaining > 3) { crm_trace("Rescheduling start (%ds remains before timeout)", remaining); pcmk__create_timer(1000, retry_start_cmd_cb, lrm_state); return; } else { crm_trace("Not enough time before timeout (%ds) " "to reschedule start", remaining); pcmk__format_result(&(cmd->result), PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_TIMEOUT, "%s without enough time to retry", pcmk_strerror(op->connection_rc)); } } else { lrm_state_reset_tables(lrm_state, TRUE); pcmk__set_result(&(cmd->result), PCMK_OCF_OK, PCMK_EXEC_DONE, NULL); lrm_remote_set_flags(lrm_state, remote_active); } crm_debug("Remote connection event matched %s action", cmd->action); report_remote_ra_result(cmd); cmd_handled = TRUE; } else if ((op->type == lrmd_event_poke) && pcmk__str_eq(cmd->action, PCMK_ACTION_MONITOR, pcmk__str_casei)) { if (cmd->monitor_timeout_id) { g_source_remove(cmd->monitor_timeout_id); cmd->monitor_timeout_id = 0; } /* Only report success the first time, after that only worry about failures. * For this function, if we get the poke pack, it is always a success. Pokes * only fail if the send fails, or the response times out. */ if (!pcmk_is_set(cmd->status, cmd_reported_success)) { pcmk__set_result(&(cmd->result), PCMK_OCF_OK, PCMK_EXEC_DONE, NULL); report_remote_ra_result(cmd); cmd_set_flags(cmd, cmd_reported_success); } crm_debug("Remote poke event matched %s action", cmd->action); /* success, keep rescheduling if interval is present. */ if (cmd->interval_ms && !pcmk_is_set(cmd->status, cmd_cancel)) { ra_data->recurring_cmds = g_list_append(ra_data->recurring_cmds, cmd); cmd->interval_id = pcmk__create_timer(cmd->interval_ms, recurring_helper, cmd); cmd = NULL; /* prevent free */ } cmd_handled = TRUE; } else if ((op->type == lrmd_event_disconnect) && pcmk__str_eq(cmd->action, PCMK_ACTION_MONITOR, pcmk__str_casei)) { if (pcmk_is_set(ra_data->status, remote_active) && !pcmk_is_set(cmd->status, cmd_cancel)) { pcmk__set_result(&(cmd->result), PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_ERROR, "Remote connection unexpectedly dropped " "during monitor"); report_remote_ra_result(cmd); crm_err("Remote connection to %s unexpectedly dropped during monitor", lrm_state->node_name); } cmd_handled = TRUE; } else { crm_debug("Event did not match %s action", ra_data->cur_cmd->action); } if (cmd_handled) { ra_data->cur_cmd = NULL; if (ra_data->cmds) { mainloop_set_trigger(ra_data->work); } free_cmd(cmd); } } static void handle_remote_ra_stop(lrm_state_t * lrm_state, remote_ra_cmd_t * cmd) { remote_ra_data_t *ra_data = NULL; pcmk__assert(lrm_state != NULL); ra_data = lrm_state->remote_ra_data; if (!pcmk_is_set(ra_data->status, takeover_complete)) { /* delete pending ops when ever the remote connection is intentionally stopped */ g_hash_table_remove_all(lrm_state->active_ops); } else { /* we no longer hold the history if this connection has been migrated, * however, we keep metadata cache for future use */ lrm_state_reset_tables(lrm_state, FALSE); } lrm_remote_clear_flags(lrm_state, remote_active); lrm_state_disconnect(lrm_state); if (ra_data->cmds) { g_list_free_full(ra_data->cmds, free_cmd); } if (ra_data->recurring_cmds) { g_list_free_full(ra_data->recurring_cmds, free_cmd); } ra_data->cmds = NULL; ra_data->recurring_cmds = NULL; ra_data->cur_cmd = NULL; if (cmd) { pcmk__set_result(&(cmd->result), PCMK_OCF_OK, PCMK_EXEC_DONE, NULL); report_remote_ra_result(cmd); } } // \return Standard Pacemaker return code static int handle_remote_ra_start(lrm_state_t * lrm_state, remote_ra_cmd_t * cmd, int timeout_ms) { const char *server = NULL; lrmd_key_value_t *tmp = NULL; int port = 0; int timeout_used = timeout_ms > MAX_START_TIMEOUT_MS ? MAX_START_TIMEOUT_MS : timeout_ms; int rc = pcmk_rc_ok; for (tmp = cmd->params; tmp; tmp = tmp->next) { if (pcmk__strcase_any_of(tmp->key, PCMK_REMOTE_RA_ADDR, PCMK_REMOTE_RA_SERVER, NULL)) { server = tmp->value; } else if (pcmk__str_eq(tmp->key, PCMK_REMOTE_RA_PORT, pcmk__str_none)) { port = atoi(tmp->value); } else if (pcmk__str_eq(tmp->key, CRM_META "_" PCMK__META_CONTAINER, pcmk__str_none)) { lrm_remote_set_flags(lrm_state, controlling_guest); } } rc = controld_connect_remote_executor(lrm_state, server, port, timeout_used); if (rc != pcmk_rc_ok) { pcmk__format_result(&(cmd->result), PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_ERROR, "Could not connect to Pacemaker Remote node %s: %s", lrm_state->node_name, pcmk_rc_str(rc)); } return rc; } static gboolean handle_remote_ra_exec(gpointer user_data) { int rc = 0; lrm_state_t *lrm_state = user_data; remote_ra_data_t *ra_data = lrm_state->remote_ra_data; remote_ra_cmd_t *cmd; GList *first = NULL; if (ra_data->cur_cmd) { /* still waiting on previous cmd */ return TRUE; } while (ra_data->cmds) { first = ra_data->cmds; cmd = first->data; if (cmd->delay_id) { /* still waiting for start delay timer to trip */ return TRUE; } ra_data->cmds = g_list_remove_link(ra_data->cmds, first); g_list_free_1(first); if (pcmk__str_any_of(cmd->action, PCMK_ACTION_START, PCMK_ACTION_MIGRATE_FROM, NULL)) { lrm_remote_clear_flags(lrm_state, expect_takeover | takeover_complete); if (handle_remote_ra_start(lrm_state, cmd, cmd->timeout) == pcmk_rc_ok) { /* take care of this later when we get async connection result */ crm_debug("Initiated async remote connection, %s action will complete after connect event", cmd->action); ra_data->cur_cmd = cmd; return TRUE; } report_remote_ra_result(cmd); } else if (!strcmp(cmd->action, PCMK_ACTION_MONITOR)) { if (lrm_state_is_connected(lrm_state) == TRUE) { rc = lrm_state_poke_connection(lrm_state); if (rc < 0) { pcmk__set_result(&(cmd->result), PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_ERROR, pcmk_strerror(rc)); } } else { rc = -1; pcmk__set_result(&(cmd->result), PCMK_OCF_NOT_RUNNING, PCMK_EXEC_DONE, "Remote connection inactive"); } if (rc == 0) { crm_debug("Poked Pacemaker Remote at node %s, waiting for async response", cmd->rsc_id); ra_data->cur_cmd = cmd; cmd->monitor_timeout_id = pcmk__create_timer(cmd->timeout, monitor_timeout_cb, cmd); return TRUE; } report_remote_ra_result(cmd); } else if (!strcmp(cmd->action, PCMK_ACTION_STOP)) { if (pcmk_is_set(ra_data->status, expect_takeover)) { /* Briefly wait on stop for an expected takeover to occur. If * the takeover does not occur during the wait, that's fine; it * just means that the remote node's resource history will be * cleared, which will require probing all resources on the * remote node. If the takeover does occur successfully, then we * can leave the status section intact. */ cmd->takeover_timeout_id = pcmk__create_timer((cmd->timeout/2), connection_takeover_timeout_cb, cmd); ra_data->cur_cmd = cmd; return TRUE; } handle_remote_ra_stop(lrm_state, cmd); } else if (strcmp(cmd->action, PCMK_ACTION_MIGRATE_TO) == 0) { lrm_remote_clear_flags(lrm_state, takeover_complete); lrm_remote_set_flags(lrm_state, expect_takeover); pcmk__set_result(&(cmd->result), PCMK_OCF_OK, PCMK_EXEC_DONE, NULL); report_remote_ra_result(cmd); } else if (pcmk__str_any_of(cmd->action, PCMK_ACTION_RELOAD, PCMK_ACTION_RELOAD_AGENT, NULL)) { /* Currently the only reloadable parameter is * PCMK_REMOTE_RA_RECONNECT_INTERVAL, which is only used by the * scheduler via the CIB, so reloads are a no-op. * * @COMPAT DC <2.1.0: We only need to check for "reload" in case * we're in a rolling upgrade with a DC scheduling "reload" instead * of "reload-agent". An OCF 1.1 "reload" would be a no-op anyway, * so this would work for that purpose as well. */ pcmk__set_result(&(cmd->result), PCMK_OCF_OK, PCMK_EXEC_DONE, NULL); report_remote_ra_result(cmd); } free_cmd(cmd); } return TRUE; } static void remote_ra_data_init(lrm_state_t * lrm_state) { remote_ra_data_t *ra_data = NULL; if (lrm_state->remote_ra_data) { return; } ra_data = pcmk__assert_alloc(1, sizeof(remote_ra_data_t)); ra_data->work = mainloop_add_trigger(G_PRIORITY_HIGH, handle_remote_ra_exec, lrm_state); lrm_state->remote_ra_data = ra_data; } void remote_ra_cleanup(lrm_state_t * lrm_state) { remote_ra_data_t *ra_data = lrm_state->remote_ra_data; if (!ra_data) { return; } if (ra_data->cmds) { g_list_free_full(ra_data->cmds, free_cmd); } if (ra_data->recurring_cmds) { g_list_free_full(ra_data->recurring_cmds, free_cmd); } mainloop_destroy_trigger(ra_data->work); free(ra_data); lrm_state->remote_ra_data = NULL; } gboolean is_remote_lrmd_ra(const char *agent, const char *provider, const char *id) { if (agent && provider && !strcmp(agent, REMOTE_LRMD_RA) && !strcmp(provider, "pacemaker")) { return TRUE; } return (id != NULL) && (controld_get_executor_state(id, false) != NULL) && !controld_is_local_node(id); } lrmd_rsc_info_t * remote_ra_get_rsc_info(lrm_state_t * lrm_state, const char *rsc_id) { lrmd_rsc_info_t *info = NULL; CRM_CHECK(rsc_id != NULL, return NULL); if (controld_get_executor_state(rsc_id, false) != NULL) { info = pcmk__assert_alloc(1, sizeof(lrmd_rsc_info_t)); info->id = pcmk__str_copy(rsc_id); info->type = pcmk__str_copy(REMOTE_LRMD_RA); info->standard = pcmk__str_copy(PCMK_RESOURCE_CLASS_OCF); info->provider = pcmk__str_copy("pacemaker"); } return info; } static gboolean is_remote_ra_supported_action(const char *action) { return pcmk__str_any_of(action, PCMK_ACTION_START, PCMK_ACTION_STOP, PCMK_ACTION_MONITOR, PCMK_ACTION_MIGRATE_TO, PCMK_ACTION_MIGRATE_FROM, PCMK_ACTION_RELOAD_AGENT, PCMK_ACTION_RELOAD, NULL); } static GList * fail_all_monitor_cmds(GList * list) { GList *rm_list = NULL; remote_ra_cmd_t *cmd = NULL; GList *gIter = NULL; for (gIter = list; gIter != NULL; gIter = gIter->next) { cmd = gIter->data; if ((cmd->interval_ms > 0) && pcmk__str_eq(cmd->action, PCMK_ACTION_MONITOR, pcmk__str_casei)) { rm_list = g_list_append(rm_list, cmd); } } for (gIter = rm_list; gIter != NULL; gIter = gIter->next) { cmd = gIter->data; pcmk__set_result(&(cmd->result), PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_ERROR, "Lost connection to remote executor"); crm_trace("Pre-emptively failing %s %s (interval=%u, %s)", cmd->action, cmd->rsc_id, cmd->interval_ms, cmd->userdata); report_remote_ra_result(cmd); list = g_list_remove(list, cmd); free_cmd(cmd); } /* frees only the list data, not the cmds */ g_list_free(rm_list); return list; } static GList * remove_cmd(GList * list, const char *action, guint interval_ms) { remote_ra_cmd_t *cmd = NULL; GList *gIter = NULL; for (gIter = list; gIter != NULL; gIter = gIter->next) { cmd = gIter->data; if ((cmd->interval_ms == interval_ms) && pcmk__str_eq(cmd->action, action, pcmk__str_casei)) { break; } cmd = NULL; } if (cmd) { list = g_list_remove(list, cmd); free_cmd(cmd); } return list; } int remote_ra_cancel(lrm_state_t *lrm_state, const char *rsc_id, const char *action, guint interval_ms) { lrm_state_t *connection_rsc = NULL; remote_ra_data_t *ra_data = NULL; CRM_CHECK(rsc_id != NULL, return -EINVAL); connection_rsc = controld_get_executor_state(rsc_id, false); if (!connection_rsc || !connection_rsc->remote_ra_data) { return -EINVAL; } ra_data = connection_rsc->remote_ra_data; ra_data->cmds = remove_cmd(ra_data->cmds, action, interval_ms); ra_data->recurring_cmds = remove_cmd(ra_data->recurring_cmds, action, interval_ms); if (ra_data->cur_cmd && (ra_data->cur_cmd->interval_ms == interval_ms) && (pcmk__str_eq(ra_data->cur_cmd->action, action, pcmk__str_casei))) { cmd_set_flags(ra_data->cur_cmd, cmd_cancel); } return 0; } static remote_ra_cmd_t * handle_dup_monitor(remote_ra_data_t *ra_data, guint interval_ms, const char *userdata) { GList *gIter = NULL; remote_ra_cmd_t *cmd = NULL; /* there are 3 places a potential duplicate monitor operation * could exist. * 1. recurring_cmds list. where the op is waiting for its next interval * 2. cmds list, where the op is queued to get executed immediately * 3. cur_cmd, which means the monitor op is in flight right now. */ if (interval_ms == 0) { return NULL; } if (ra_data->cur_cmd && !pcmk_is_set(ra_data->cur_cmd->status, cmd_cancel) && (ra_data->cur_cmd->interval_ms == interval_ms) && pcmk__str_eq(ra_data->cur_cmd->action, PCMK_ACTION_MONITOR, pcmk__str_casei)) { cmd = ra_data->cur_cmd; goto handle_dup; } for (gIter = ra_data->recurring_cmds; gIter != NULL; gIter = gIter->next) { cmd = gIter->data; if ((cmd->interval_ms == interval_ms) && pcmk__str_eq(cmd->action, PCMK_ACTION_MONITOR, pcmk__str_casei)) { goto handle_dup; } } for (gIter = ra_data->cmds; gIter != NULL; gIter = gIter->next) { cmd = gIter->data; if ((cmd->interval_ms == interval_ms) && pcmk__str_eq(cmd->action, PCMK_ACTION_MONITOR, pcmk__str_casei)) { goto handle_dup; } } return NULL; handle_dup: crm_trace("merging duplicate monitor cmd " PCMK__OP_FMT, cmd->rsc_id, PCMK_ACTION_MONITOR, interval_ms); /* update the userdata */ if (userdata) { free(cmd->userdata); cmd->userdata = pcmk__str_copy(userdata); } /* if we've already reported success, generate a new call id */ if (pcmk_is_set(cmd->status, cmd_reported_success)) { cmd->start_time = time(NULL); cmd->call_id = generate_callid(); cmd_clear_flags(cmd, cmd_reported_success); } /* if we have an interval_id set, that means we are in the process of * waiting for this cmd's next interval. instead of waiting, cancel * the timer and execute the action immediately */ if (cmd->interval_id) { g_source_remove(cmd->interval_id); cmd->interval_id = 0; recurring_helper(cmd); } return cmd; } /*! * \internal * \brief Execute an action using the (internal) ocf:pacemaker:remote agent * * \param[in] lrm_state Executor state object for remote connection * \param[in] rsc_id Connection resource ID * \param[in] action Action to execute * \param[in] userdata String to copy and pass to execution callback * \param[in] interval_ms Action interval (in milliseconds) * \param[in] timeout_ms Action timeout (in milliseconds) * \param[in] start_delay_ms Delay (in milliseconds) before executing action * \param[in,out] params Connection resource parameters * \param[out] call_id Where to store call ID on success * * \return Standard Pacemaker return code * \note This takes ownership of \p params, which should not be used or freed * after calling this function. */ int controld_execute_remote_agent(const lrm_state_t *lrm_state, const char *rsc_id, const char *action, const char *userdata, guint interval_ms, int timeout_ms, int start_delay_ms, lrmd_key_value_t *params, int *call_id) { lrm_state_t *connection_rsc = NULL; remote_ra_cmd_t *cmd = NULL; remote_ra_data_t *ra_data = NULL; *call_id = 0; CRM_CHECK((lrm_state != NULL) && (rsc_id != NULL) && (action != NULL) && (userdata != NULL) && (call_id != NULL), lrmd_key_value_freeall(params); return EINVAL); if (!is_remote_ra_supported_action(action)) { lrmd_key_value_freeall(params); return EOPNOTSUPP; } connection_rsc = controld_get_executor_state(rsc_id, false); if (connection_rsc == NULL) { lrmd_key_value_freeall(params); return ENOTCONN; } remote_ra_data_init(connection_rsc); ra_data = connection_rsc->remote_ra_data; cmd = handle_dup_monitor(ra_data, interval_ms, userdata); if (cmd) { *call_id = cmd->call_id; lrmd_key_value_freeall(params); return pcmk_rc_ok; } cmd = pcmk__assert_alloc(1, sizeof(remote_ra_cmd_t)); cmd->owner = pcmk__str_copy(lrm_state->node_name); cmd->rsc_id = pcmk__str_copy(rsc_id); cmd->action = pcmk__str_copy(action); cmd->userdata = pcmk__str_copy(userdata); cmd->interval_ms = interval_ms; cmd->timeout = timeout_ms; cmd->start_delay = start_delay_ms; cmd->params = params; cmd->start_time = time(NULL); cmd->call_id = generate_callid(); if (cmd->start_delay) { cmd->delay_id = pcmk__create_timer(cmd->start_delay, start_delay_helper, cmd); } ra_data->cmds = g_list_append(ra_data->cmds, cmd); mainloop_set_trigger(ra_data->work); *call_id = cmd->call_id; return pcmk_rc_ok; } /*! * \internal * \brief Immediately fail all monitors of a remote node, if proxied here * * \param[in] node_name Name of pacemaker_remote node */ void remote_ra_fail(const char *node_name) { lrm_state_t *lrm_state = NULL; CRM_CHECK(node_name != NULL, return); lrm_state = controld_get_executor_state(node_name, false); if (lrm_state && lrm_state_is_connected(lrm_state)) { remote_ra_data_t *ra_data = lrm_state->remote_ra_data; crm_info("Failing monitors on Pacemaker Remote node %s", node_name); ra_data->recurring_cmds = fail_all_monitor_cmds(ra_data->recurring_cmds); ra_data->cmds = fail_all_monitor_cmds(ra_data->cmds); } } /* A guest node fencing implied by host fencing looks like: * * * * * * * */ #define XPATH_PSEUDO_FENCE "/" PCMK__XE_PSEUDO_EVENT \ "[@" PCMK_XA_OPERATION "='stonith']/" PCMK__XE_DOWNED "/" PCMK_XE_NODE /*! * \internal * \brief Check a pseudo-action for Pacemaker Remote node side effects * * \param[in,out] xml XML of pseudo-action to check */ void remote_ra_process_pseudo(xmlNode *xml) { xmlXPathObject *search = pcmk__xpath_search(xml->doc, XPATH_PSEUDO_FENCE); if (pcmk__xpath_num_results(search) == 1) { xmlNode *result = pcmk__xpath_result(search, 0); /* Normally, we handle the necessary side effects of a guest node stop * action when reporting the remote agent's result. However, if the stop * is implied due to fencing, it will be a fencing pseudo-event, and * there won't be a result to report. Handle that case here. * * This will result in a duplicate call to remote_node_down() if the * guest stop was real instead of implied, but that shouldn't hurt. * * There is still one corner case that isn't handled: if a guest node * isn't running any resources when its host is fenced, it will appear * to be cleanly stopped, so there will be no pseudo-fence, and our * peer cache state will be incorrect unless and until the guest is * recovered. */ if (result) { const char *remote = pcmk__xe_id(result); if (remote) { remote_node_down(remote, DOWN_ERASE_LRM); } } } xmlXPathFreeObject(search); } static void remote_ra_maintenance(lrm_state_t * lrm_state, gboolean maintenance) { xmlNode *update, *state; int call_opt; pcmk__node_status_t *node = NULL; call_opt = crmd_cib_smart_opt(); node = pcmk__cluster_lookup_remote_node(lrm_state->node_name); CRM_CHECK(node != NULL, return); update = pcmk__xe_create(NULL, PCMK_XE_STATUS); state = create_node_state_update(node, controld_node_update_none, update, __func__); crm_xml_add(state, PCMK__XA_NODE_IN_MAINTENANCE, (maintenance? "1" : "0")); if (controld_update_cib(PCMK_XE_STATUS, update, call_opt, NULL) == pcmk_rc_ok) { /* TODO: still not 100% sure that async update will succeed ... */ if (maintenance) { lrm_remote_set_flags(lrm_state, remote_in_maint); } else { lrm_remote_clear_flags(lrm_state, remote_in_maint); } } pcmk__xml_free(update); } #define XPATH_PSEUDO_MAINTENANCE "//" PCMK__XE_PSEUDO_EVENT \ "[@" PCMK_XA_OPERATION "='" PCMK_ACTION_MAINTENANCE_NODES "']/" \ PCMK__XE_MAINTENANCE /*! * \internal * \brief Check a pseudo-action holding updates for maintenance state * * \param[in,out] xml XML of pseudo-action to check */ void remote_ra_process_maintenance_nodes(xmlNode *xml) { xmlXPathObject *search = pcmk__xpath_search(xml->doc, XPATH_PSEUDO_MAINTENANCE); if (pcmk__xpath_num_results(search) == 1) { xmlNode *node; int cnt = 0, cnt_remote = 0; for (node = pcmk__xe_first_child(pcmk__xpath_result(search, 0), PCMK_XE_NODE, NULL, NULL); node != NULL; node = pcmk__xe_next(node, PCMK_XE_NODE)) { lrm_state_t *lrm_state = NULL; const char *id = pcmk__xe_id(node); cnt++; if (id == NULL) { continue; // Shouldn't be possible } lrm_state = controld_get_executor_state(id, false); if (lrm_state && lrm_state->remote_ra_data && pcmk_is_set(((remote_ra_data_t *) lrm_state->remote_ra_data)->status, remote_active)) { const char *in_maint_s = NULL; int in_maint; cnt_remote++; in_maint_s = crm_element_value(node, PCMK__XA_NODE_IN_MAINTENANCE); pcmk__scan_min_int(in_maint_s, &in_maint, 0); remote_ra_maintenance(lrm_state, in_maint); } } crm_trace("Action holds %d nodes (%d remotes found) adjusting " PCMK_OPT_MAINTENANCE_MODE, cnt, cnt_remote); } xmlXPathFreeObject(search); } gboolean remote_ra_is_in_maintenance(lrm_state_t * lrm_state) { remote_ra_data_t *ra_data = lrm_state->remote_ra_data; return pcmk_is_set(ra_data->status, remote_in_maint); } gboolean remote_ra_controlling_guest(lrm_state_t * lrm_state) { remote_ra_data_t *ra_data = lrm_state->remote_ra_data; return pcmk_is_set(ra_data->status, controlling_guest); } diff --git a/daemons/controld/controld_utils.h b/daemons/controld/controld_utils.h index e6338882e7..262e0d1f39 100644 --- a/daemons/controld/controld_utils.h +++ b/daemons/controld/controld_utils.h @@ -1,87 +1,87 @@ /* * Copyright 2004-2025 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #ifndef CRMD_UTILS__H #define CRMD_UTILS__H #include // UINT32_C(), uint32_t #include // gboolean #include // xmlNode #include #include // enum controld_join_phase #include // pcmk__node_status_t #include # define FAKE_TE_ID "xxxxxxxx-xxxx-xxxx-xxxx-xxxxxxxxxxxx" //! Flags determining how the controller updates node information in the CIB enum controld_node_update_flags { //! This flag has no effect controld_node_update_none = UINT32_C(0), //! Skip loading the node list from the cluster layer controld_node_update_quick = (UINT32_C(1) << 0), //! Update \c PCMK__XA_IN_CCM with the time the node became a cluster member controld_node_update_cluster = (UINT32_C(1) << 1), //! Update \c PCMK_XA_CRMD with the time the node joined the CPG controld_node_update_peer = (UINT32_C(1) << 2), //! Update \c PCMK__XA_JOIN with the node's join state controld_node_update_join = (UINT32_C(1) << 3), //! Update \c PCMK_XA_EXPECTED with the node's expected join state controld_node_update_expected = (UINT32_C(1) << 4), //! Convenience alias to update all of the attributes mentioned above controld_node_update_all = controld_node_update_cluster |controld_node_update_peer |controld_node_update_join |controld_node_update_expected, }; crm_exit_t crmd_exit(crm_exit_t exit_code); _Noreturn void crmd_fast_exit(crm_exit_t exit_code); void controld_shutdown_schedulerd_ipc(void); void controld_stop_sched_timer(void); void controld_free_sched_timer(void); void controld_expect_sched_reply(char *ref); void fsa_dump_actions(uint64_t action, const char *text); void fsa_dump_inputs(int log_level, const char *text, long long input_register); gboolean update_dc(xmlNode * msg); void crm_update_peer_join(const char *source, pcmk__node_status_t *node, enum controld_join_phase phase); xmlNode *create_node_state_update(pcmk__node_status_t *node, uint32_t flags, xmlNode *parent, const char *source); void populate_cib_nodes(uint32_t flags, const char *source); void crm_update_quorum(gboolean quorum, gboolean force_update); void controld_close_attrd_ipc(void); void update_attrd(const char *host, const char *name, const char *value, const char *user_name, gboolean is_remote_node); void update_attrd_list(GList *attrs, uint32_t opts); -void update_attrd_remote_node_removed(const char *host, const char *user_name); +void controld_purge_node_attrs(const char *node_name, bool from_cache); void update_attrd_clear_failures(const char *host, const char *rsc, const char *op, const char *interval_spec, gboolean is_remote_node); int crmd_join_phase_count(enum controld_join_phase phase); void crmd_join_phase_log(int level); void crmd_peer_down(pcmk__node_status_t *peer, bool full); bool feature_set_compatible(const char *dc_version, const char *join_version); bool controld_is_local_node(const char *name); pcmk__node_status_t *controld_get_local_node_status(void); const char *get_node_id(xmlNode *lrm_rsc_op); #endif