Page MenuHomeClusterLabs Projects
Files F4639681

No OneTemporary
Actions

View Options

This file is larger than 256 KB, so syntax highlighting was skipped.
diff --git a/daemons/attrd/attrd_cib.c b/daemons/attrd/attrd_cib.c
index 80afec4427..b4d8a2ad38 100644
--- a/daemons/attrd/attrd_cib.c
+++ b/daemons/attrd/attrd_cib.c
@@ -1,685 +1,685 @@
/*
* Copyright 2013-2024 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 <crm_internal.h>
#include <errno.h>
#include <inttypes.h> // PRIu32
#include <stdbool.h>
#include <stdlib.h>
#include <glib.h>
#include <crm/cib/internal.h> // cib__*
#include <crm/common/logging.h>
#include <crm/common/results.h>
#include <crm/common/strings_internal.h>
#include <crm/common/xml.h>
#include <crm/cluster/internal.h> // 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)) {
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);
}
}
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 (attrd_shutting_down(true)) {
crm_debug("Ignoring CIB change during shutdown");
return;
}
if (cib__get_notify_patchset(msg, &patchset) != pcmk_rc_ok) {
return;
}
if (cib__element_in_patchset(patchset, PCMK_XE_ALERTS)) {
mainloop_set_trigger(attrd_config_read);
}
status_changed = 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.
*/
return;
}
if (!attrd_election_won()) {
// Don't write attributes if we're not the writer
return;
}
if (status_changed || cib__element_in_patchset(patchset, PCMK_XE_NODES)) {
/* 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, PCMK__VALUE_ATTRD, 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
* cluster layer didn't notice.
*
* \todo If pacemaker-attrd 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->uname);
// 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)) {
rc = attrd_send_attribute_alert(at->nodename, at->nodeid,
a->id, at->current);
crm_trace("Sent alerts for %s[%s]=%s: nodeid=%d rc=%d",
a->id, at->nodename, at->current, at->nodeid, rc);
}
}
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->nodeid = v->nodeid;
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;
if (a == NULL) {
return;
}
/* If this attribute will be written to the CIB ... */
if (!stand_alone && !pcmk_is_set(a->flags, attrd_attr_is_private)) {
/* 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;
}
}
/* Attribute will be written shortly, so clear changed flag and force
* write flag, and initialize UUID missing flag to false.
*/
attrd_clear_attr_flags(a, attrd_attr_changed|attrd_attr_uuid_missing|attrd_attr_force_write);
/* 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 *uuid = NULL;
if (pcmk_is_set(v->flags, attrd_value_remote)) {
/* If this is a Pacemaker Remote node, the node's UUID is the same
* as its name, which we already have.
*/
uuid = v->nodename;
} else {
// This will create a cluster node cache entry if none exists
pcmk__node_status_t *peer = pcmk__get_node(v->nodeid, v->nodename,
NULL,
pcmk__node_search_any);
- uuid = peer->uuid;
+ uuid = peer->xml_id;
// Remember peer's node ID if we're just now learning it
if ((peer->cluster_layer_id != 0) && (v->nodeid == 0)) {
crm_trace("Learned ID %" PRIu32 " for node %s",
peer->cluster_layer_id, v->nodename);
v->nodeid = peer->cluster_layer_id;
}
}
/* If this is a private attribute, no update needs to be sent */
if (stand_alone || pcmk_is_set(a->flags, attrd_attr_is_private)) {
private_updates++;
continue;
}
// Defer write if this is a cluster node that's never been seen
if (uuid == NULL) {
attrd_set_attr_flags(a, attrd_attr_uuid_missing);
crm_notice("Cannot update %s[%s]='%s' now because node's UUID is "
"unknown (will retry if learned)",
a->id, v->nodename, v->current);
continue;
}
// Update this value as part of the CIB transaction we're building
rc = add_attr_update(a, v->current, uuid);
if (rc != pcmk_rc_ok) {
crm_err("Failed to update %s[%s]='%s': %s "
QB_XS " node uuid=%s id=%" PRIu32,
a->id, v->nodename, v->current, pcmk_rc_str(rc),
uuid, v->nodeid);
continue;
}
crm_debug("Writing %s[%s]=%s (node-state-id=%s node-id=%" PRIu32 ")",
a->id, v->nodename, pcmk__s(v->current, "(unset)"),
uuid, v->nodeid);
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
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_uuid_missing)) {
// 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 a4885762e3..a837476063 100644
--- a/daemons/attrd/attrd_corosync.c
+++ b/daemons/attrd/attrd_corosync.c
@@ -1,612 +1,613 @@
/*
* Copyright 2013-2024 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 <crm_internal.h>
#include <errno.h>
#include <stdbool.h>
#include <stdint.h>
#include <stdlib.h>
#include <crm/cluster.h>
#include <crm/cluster/internal.h>
#include <crm/common/logging.h>
#include <crm/common/results.h>
#include <crm/common/strings_internal.h>
#include <crm/common/xml.h>
#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 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)
{
uint32_t kind = 0;
xmlNode *xml = NULL;
const char *from = NULL;
char *data = pcmk__cpg_message_data(handle, nodeid, pid, msg, &kind, &from);
if(data == NULL) {
return;
}
if (kind == crm_class_cluster) {
xml = pcmk__xml_parse(data);
}
if (xml == NULL) {
crm_err("Bad message of class %d received from %s[%u]: '%.120s'", kind, 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)) {
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 crm_status_type kind, pcmk__node_status_t *peer,
const void *data)
{
bool gone = false;
bool is_remote = pcmk_is_set(peer->flags, crm_remote_node);
switch (kind) {
case crm_status_uname:
crm_debug("%s node %s is now %s",
(is_remote? "Remote" : "Cluster"),
peer->name, state_text(peer->state));
break;
case crm_status_processes:
if (!pcmk_is_set(peer->processes, crm_get_cluster_proc())) {
gone = true;
}
crm_debug("Node %s is %s a peer",
peer->name, (gone? "no longer" : "now"));
break;
case crm_status_nstate:
crm_debug("%s node %s is now %s (was %s)",
(is_remote? "Remote" : "Cluster"),
peer->name, state_text(peer->state), state_text(data));
if (pcmk__str_eq(peer->state, CRM_NODE_MEMBER, pcmk__str_casei)) {
/* 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 (attrd_election_won()
&& !pcmk_is_set(peer->flags, crm_remote_node)) {
attrd_peer_sync(peer);
}
} else {
// Remove all attribute values associated with lost nodes
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) {
attrd_remove_voter(peer);
attrd_remove_peer_protocol_ver(peer->name);
attrd_do_not_expect_from_peer(peer->name);
}
}
static void
record_peer_nodeid(attribute_value_t *v, const char *host)
{
pcmk__node_status_t *known_peer =
pcmk__get_node(v->nodeid, host, NULL, pcmk__node_search_cluster_member);
- crm_trace("Learned %s has node id %s", known_peer->name, known_peer->uuid);
+ crm_trace("Learned %s has node id %s",
+ known_peer->name, known_peer->xml_id);
if (attrd_election_won()) {
attrd_write_attributes(attrd_write_changed);
}
}
#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;
// 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 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);
CRM_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->uname,
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->uname);
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",
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__str_update(&v->current, value);
attrd_set_attr_flags(a, attrd_attr_changed);
if (pcmk__str_eq(host, attrd_cluster->uname, 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);
/* If this is a cluster node whose node ID we are learning, remember it */
if ((v->nodeid == 0) && !pcmk_is_set(v->flags, attrd_value_remote)
&& (crm_element_value_int(xml, PCMK__XA_ATTR_HOST_ID,
(int*)&v->nodeid) == 0) && (v->nodeid > 0)) {
record_peer_nodeid(v, host);
}
}
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->uname,
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(&regex, 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(&regex, 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(&regex);
}
/*!
* \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)) {
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);
}
}
if (uncache) {
pcmk__purge_node_from_cache(host, 0);
}
}
/*!
* \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_same(child)) {
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/controld/controld_callbacks.c b/daemons/controld/controld_callbacks.c
index 04b22161bb..aa6d61f576 100644
--- a/daemons/controld/controld_callbacks.c
+++ b/daemons/controld/controld_callbacks.c
@@ -1,392 +1,392 @@
/*
* Copyright 2004-2024 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 <crm_internal.h>
#include <inttypes.h> // PRIu32
#include <stdbool.h> // bool
#include <stdio.h> // NULL
#include <sys/param.h>
#include <string.h>
#include <crm/crm.h>
#include <crm/common/xml.h>
#include <crm/cluster.h>
#include <crm/cib.h>
#include <pacemaker-controld.h>
/* 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 (!pcmk__str_eq(from, controld_globals.our_nodename,
pcmk__str_casei)) {
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, crm_remote_node)) {
// Pacemaker Remote nodes can't be partially alive
return pcmk__str_eq(node->state, CRM_NODE_MEMBER, pcmk__str_casei) ? 1: -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, CRM_NODE_MEMBER, pcmk__str_casei)) {
// 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")
void
peer_update_callback(enum crm_status_type type, pcmk__node_status_t *node,
const void *data)
{
uint32_t old = 0;
bool appeared = FALSE;
bool is_remote = pcmk_is_set(node->flags, crm_remote_node);
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 == crm_status_processes
&& pcmk_is_set(node->processes, crm_get_cluster_proc())
&& !AM_I_DC
&& !is_remote) {
/*
* This is a hack until we can send to a nodeid and/or we fix node name lookups
* These messages are ignored in crmd_ha_msg_filter()
*/
xmlNode *query = create_request(CRM_OP_HELLO, NULL, NULL, CRM_SYSTEM_CRMD, CRM_SYSTEM_CRMD, NULL);
crm_debug("Sending hello to node %" PRIu32 " so that it learns our "
"node name",
node->cluster_layer_id);
pcmk__cluster_send_message(node, crm_msg_crmd, query);
pcmk__xml_free(query);
}
if (node->name == NULL) {
return;
}
switch (type) {
case crm_status_uname:
/* 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 crm_status_nstate:
/* 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(CRM_NODE_MEMBER, node->state, pcmk__str_casei)) {
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 crm_status_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, "<none>"), 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
&& pcmk__str_eq(node->name, controld_globals.our_nodename,
pcmk__str_casei)) {
/* 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.
*/
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;
int flags = node_update_peer;
int alive = node_alive(node);
- pcmk__graph_action_t *down = match_down_event(node->uuid);
+ 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 |= node_update_join | 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 == -1)) {
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, crm_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->uuid);
+ 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 |= 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 == crm_status_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 |= 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_execd.c b/daemons/controld/controld_execd.c
index 5ef83d75f6..a38e356ff2 100644
--- a/daemons/controld/controld_execd.c
+++ b/daemons/controld/controld_execd.c
@@ -1,2420 +1,2420 @@
/*
* Copyright 2004-2024 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 <crm_internal.h>
#include <regex.h>
#include <sys/param.h>
#include <sys/types.h>
#include <sys/wait.h>
#include <crm/crm.h>
#include <crm/lrmd.h> // lrmd_event_data_t, lrmd_rsc_info_t, etc.
#include <crm/services.h>
#include <crm/common/xml.h>
#include <crm/pengine/rules.h>
#include <crm/lrmd_internal.h>
#include <pacemaker-internal.h>
#include <pacemaker-controld.h>
#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.our_nodename);
}
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 = lrm_state_find(op_node_name(op));
CRM_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.our_nodename == NULL) {
return; /* Nothing to do */
}
lrm_state = lrm_state_find_or_create(controld_globals.our_nodename);
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)
{
xmlNode *xml_state = NULL;
xmlNode *xml_data = NULL;
xmlNode *rsc_list = NULL;
pcmk__node_status_t *peer = NULL;
lrm_state_t *lrm_state = lrm_state_find(controld_globals.our_nodename);
if (!lrm_state) {
crm_err("Could not find executor state for node %s",
controld_globals.our_nodename);
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,
node_update_cluster|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->uuid);
+ 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 always 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 = lrm_state_find(node_name);
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 = lrm_state_find(entry->id);
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);
}
/*!
* \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.our_nodename;
}
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, 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 = create_request(CRM_OP_INVOKE_LRM, NULL, from_host,
from_sys, CRM_SYSTEM_LRMD,
controld_globals.our_uuid);
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 = lrm_state_find(lrm_op_target(data->input_xml));
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 = !pcmk__str_eq(target_node, controld_globals.our_nodename,
pcmk__str_casei);
lrm_state = lrm_state_find(target_node);
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;
}
CRM_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, 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_notice("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;
CRM_ASSERT(rsc_id && operation);
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
CRM_ASSERT(rsc_id != NULL);
op->rsc_id = pcmk__str_copy(rsc_id);
}
if (to_sys == NULL) {
to_sys = CRM_SYSTEM_TENGINE;
}
peer = pcmk__get_node(0, controld_globals.our_nodename, NULL,
pcmk__node_search_cluster_member);
update = create_node_state_update(peer, 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.our_nodename);
reply = create_request(CRM_OP_INVOKE_LRM, update, to_host, to_sys, CRM_SYSTEM_LRMD, NULL);
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);
}
}
/* now do the op */
crm_notice("Requesting local execution of %s operation for %s on %s "
QB_XS " transition_key=%s op_key=" PCMK__OP_FMT,
pcmk__readable_action(op->op_type, op->interval_ms), rsc->id,
lrm_state->node_name, pcmk__s(transition, ""), rsc->id,
operation, op->interval_ms);
nack_reason = should_nack_action(operation);
if (nack_reason != NULL) {
crm_notice("Discarding attempt to perform action %s on %s in state %s "
"(shutdown=%s)", operation, rsc->id,
fsa_state2string(controld_globals.fsa_state),
pcmk__flag_text(controld_globals.fsa_input_register,
R_SHUTDOWN));
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;
}
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);
}
void
do_lrm_event(long long action,
enum crmd_fsa_cause cause,
enum crmd_fsa_state cur_state, enum crmd_fsa_input cur_input, fsa_data_t * msg_data)
{
CRM_CHECK(FALSE, return);
}
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, ": ", services_ocf_exitcode_str(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;
/* order of __attribute__ and Fall through comment is IMPORTANT!
* do not change it without proper testing with both clang and gcc
* in multiple versions.
* the clang check allows to build with all versions of clang.
* the has_c_attribute check is to workaround a bug in clang version
* in rhel7. has_attribute would happily return "YES SIR WE GOT IT"
* and fail the build the next line.
*/
#ifdef __clang__
#ifdef __has_c_attribute
#if __has_attribute(fallthrough)
__attribute__((fallthrough));
#endif
#endif
#endif
// Fall through
default:
pcmk__g_strcat(str, ": ", pcmk_exec_status_str(op->op_status),
NULL);
}
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 6991bf1247..f79944f214 100644
--- a/daemons/controld/controld_fencing.c
+++ b/daemons/controld/controld_fencing.c
@@ -1,1119 +1,1119 @@
/*
* Copyright 2004-2024 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 <crm_internal.h>
#include <crm/crm.h>
#include <crm/common/xml.h>
#include <crm/stonith-ng.h>
#include <crm/fencing/internal.h>
#include <pacemaker-controld.h>
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;
};
static bool fence_reaction_panic = false;
static unsigned long int stonith_max_attempts = 10;
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)
{
stonith_max_attempts = char2score(value);
if (stonith_max_attempts < 1UL) {
stonith_max_attempts = 10UL;
}
}
/*!
* \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);
}
}
static void
send_stonith_update(pcmk__graph_action_t *action, const char *target,
const char *uuid)
{
int rc = pcmk_ok;
pcmk__node_status_t *peer = NULL;
/* We (usually) rely on the membership layer to do node_update_cluster,
* and the peer status callback to do node_update_peer, because the node
* might have already rejoined before we get the stonith result here.
*/
int flags = node_update_join | node_update_expected;
/* zero out the node-status & remove all LRM status info */
xmlNode *node_state = NULL;
CRM_CHECK(target != NULL, return);
CRM_CHECK(uuid != NULL, return);
/* Make sure the membership and join caches are accurate.
* Try getting any existing node cache entry also by node uuid in case it
* doesn't have an uname yet.
*/
peer = pcmk__get_node(0, target, uuid, 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 |= node_update_cluster;
}
- if (peer->uuid == NULL) {
- crm_info("Recording uuid '%s' for node '%s'", uuid, target);
- peer->uuid = pcmk__str_copy(uuid);
+ if (peer->xml_id == NULL) {
+ crm_info("Recording XML ID '%s' for node '%s'", uuid, target);
+ peer->xml_id = pcmk__str_copy(uuid);
}
crmd_peer_down(peer, TRUE);
/* Generate a node state update for the CIB */
node_state = create_node_state_update(peer, flags, NULL, __func__);
/* we have to mark whether or not remote nodes have already been fenced */
if (peer->flags & crm_remote_node) {
char *now_s = pcmk__ttoa(time(NULL));
crm_xml_add(node_state, PCMK__XA_NODE_FENCED, now_s);
free(now_s);
}
/* Force our known ID */
crm_xml_add(node_state, PCMK_XA_ID, uuid);
rc = controld_globals.cib_conn->cmds->modify(controld_globals.cib_conn,
PCMK_XE_STATUS, node_state,
cib_can_create);
/* Delay processing the trigger until the update completes */
crm_debug("Sending fencing update %d for %s", rc, target);
fsa_register_cib_callback(rc, pcmk__str_copy(target), cib_fencing_updated);
// Make sure it sticks
/* controld_globals.cib_conn->cmds->bump_epoch(controld_globals.cib_conn,
* cib_none);
*/
controld_delete_node_state(peer->name, controld_section_all, cib_none);
pcmk__xml_free(node_state);
return;
}
/*!
* \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_node_uuid(target_node);
crm_notice("Marking %s, target of a previous stonith action, as clean", target);
send_stonith_update(NULL, 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
&& pcmk__str_eq(event->target, controld_globals.our_nodename,
pcmk__str_casei)) {
/* 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(__func__);
} 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,
flags);
const char *uuid = NULL;
if (peer == NULL) {
return;
}
uuid = pcmk__cluster_node_uuid(peer);
if (AM_I_DC) {
/* The DC always sends updates */
send_stonith_update(NULL, 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, crm_remote_node)) {
// 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 (pcmk__str_eq(event->executioner, controld_globals.our_nodename,
pcmk__str_casei)) {
send_stonith_update(NULL, 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, crm_remote_node)) {
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_ok) {
crm_err("Could not connect to fencer in 30 attempts: %s "
QB_XS " rc=%d", pcmk_strerror(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(crm_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))) {
send_stonith_update(action, 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 = (int) (controld_globals.transition_graph->stonith_timeout
/ 1000);
if (crmd_join_phase_count(crm_join_confirmed) == 1) {
stonith__set_call_options(options, target, st_opt_allow_suicide);
}
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 = (int) (controld_globals.transition_graph->stonith_timeout
/ 1000);
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.our_nodename;
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 2227e01360..3d947326bb 100644
--- a/daemons/controld/controld_join_dc.c
+++ b/daemons/controld/controld_join_dc.c
@@ -1,1057 +1,1057 @@
/*
* Copyright 2004-2024 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 <crm_internal.h>
#include <inttypes.h> // PRIu32
#include <stdbool.h> // bool, true, false
#include <stdio.h> // NULL
#include <stdlib.h> // free(), etc.
#include <glib.h> // gboolean, etc.
#include <libxml/tree.h> // xmlNode
#include <crm/crm.h>
#include <crm/common/xml.h>
#include <crm/cluster.h>
#include <pacemaker-controld.h>
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 (<tt>node_name -> join_id</tt>). \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 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 crm_join_phase phase)
{
enum crm_join_phase last = 0;
CRM_CHECK(node != NULL, return);
/* Remote nodes do not participate in joins */
if (pcmk_is_set(node->flags, crm_remote_node)) {
return;
}
last = node->join;
if(phase == last) {
crm_trace("Node %s join-%d phase is still %s "
QB_XS " nodeid=%" PRIu32 " source=%s",
node->name, current_join_id, controld_join_phase_text(last),
node->cluster_layer_id, source);
} else if ((phase <= crm_join_none) || (phase == (last + 1))) {
node->join = phase;
crm_trace("Node %s join-%d phase is now %s (was %s) "
QB_XS " nodeid=%" PRIu32 " source=%s",
node->name, current_join_id, controld_join_phase_text(phase),
controld_join_phase_text(last), node->cluster_layer_id,
source);
} else {
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, controld_join_phase_text(last),
controld_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, crm_peer_cache);
while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &peer)) {
crm_update_peer_join(__func__, peer, crm_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 = create_request(join_op, NULL, host_to, CRM_SYSTEM_CRMD,
CRM_SYSTEM_DC, 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)
{
xmlNode *offer = NULL;
pcmk__node_status_t *member = (pcmk__node_status_t *) value;
CRM_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, CRM_NODE_LOST, pcmk__str_casei)) {
/* 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->uuid);
+ current_join_id, member->xml_id);
return;
}
if (controld_globals.membership_id != crm_peer_seq) {
controld_globals.membership_id = crm_peer_seq;
crm_info("Making join-%d offers based on membership event %llu",
current_join_id, crm_peer_seq);
}
if(user_data && member->join > crm_join_none) {
crm_info("Not making join-%d offer to already known node %s (%s)",
current_join_id, member->name,
controld_join_phase_text(member->join));
return;
}
crm_update_peer_join(__func__, (pcmk__node_status_t*) member,
crm_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, crm_msg_crmd, offer);
pcmk__xml_free(offer);
crm_update_peer_join(__func__, member, crm_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(crm_peer_cache, join_make_offer, NULL);
count = crmd_join_phase_count(crm_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(crm_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, crm_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 (strcasecmp(join_to, controld_globals.our_nodename) != 0) {
member = pcmk__get_node(0, controld_globals.our_nodename, NULL,
pcmk__node_search_cluster_member);
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(crm_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;
pcmk__scan_ll(elem_l, &int_elem_l, -1LL);
pcmk__scan_ll(elem_r, &int_elem_r, -1LL);
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 unknown schema version %s "
QB_XS " ref=%s",
join_id, join_from, pcmk__s(validation, "(missing)"), 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)
&& pcmk__str_eq(join_from, controld_globals.our_nodename,
pcmk__str_casei))) {
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 unknown "
"schema version %s " QB_XS " ref=%s",
join_id, join_from, max_generation_from,
pcmk__s(validation, "(missing)"), 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) {
if (compare_version(join_version, "3.17.0") < 0) {
/* Clients with CRM_FEATURE_SET < 3.17.0 may respawn infinitely
* after a nack message, don't send one
*/
crm_update_peer_join(__func__, join_node, crm_join_nack_quiet);
} else {
crm_update_peer_join(__func__, join_node, crm_join_nack);
}
pcmk__update_peer_expected(__func__, join_node, CRMD_JOINSTATE_NACK);
} else {
crm_update_peer_join(__func__, join_node, crm_join_integrated);
pcmk__update_peer_expected(__func__, join_node, CRMD_JOINSTATE_MEMBER);
}
count = crmd_join_phase_count(crm_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(crm_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(crm_join_welcomed);
int count_finalizable = crmd_join_phase_count(crm_join_integrated)
+ crmd_join_phase_count(crm_join_nack)
+ crmd_join_phase_count(crm_join_nack_quiet);
/* 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 (pcmk__str_eq(max_generation_from, controld_globals.our_nodename,
pcmk__str_null_matches|pcmk__str_casei)) {
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.our_nodename);
crm_debug("Finalizing join-%d for %d node%s (sync'ing from local CIB)",
current_join_id, count_finalizable,
pcmk__plural_s(count_finalizable));
crm_log_xml_debug(max_generation_xml, "Requested CIB version");
} 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 from %s)",
current_join_id, count_finalizable,
pcmk__plural_s(count_finalizable), sync_from);
crm_log_xml_notice(max_generation_xml, "Requested CIB version");
}
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(crm_join_integrated)
+ crmd_join_phase_count(crm_join_nack)
+ crmd_join_phase_count(crm_join_nack_quiet);
crm_debug("Notifying %d node%s of join-%d results",
count_finalizable, pcmk__plural_s(count_finalizable),
current_join_id);
g_hash_table_foreach(crm_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_section_e section = controld_section_lrm;
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);
if (peer->join != crm_join_finalized) {
crm_info("Ignoring out-of-sequence join-%d confirmation from %s "
"(currently %s not %s)",
join_id, join_from, controld_join_phase_text(peer->join),
controld_join_phase_text(crm_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, crm_join_nack);
goto done;
}
crm_update_peer_join(__func__, peer, crm_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);
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 (pcmk__str_eq(join_from, controld_globals.our_nodename,
pcmk__str_casei)) {
// 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;
bool integrated = false;
switch (join_node->join) {
case crm_join_integrated:
integrated = true;
break;
case crm_join_nack:
case crm_join_nack_quiet:
break;
default:
crm_trace("Not updating non-integrated and non-nacked node %s (%s) "
"for join-%d", join_to,
controld_join_phase_text(join_node->join),
current_join_id);
return;
}
/* Update the <node> 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_node_uuid(join_node));
crm_xml_add(tmp1, PCMK_XA_UNAME, join_to);
fsa_cib_anon_update(PCMK_XE_NODES, tmp1);
pcmk__xml_free(tmp1);
if (join_node->join == crm_join_nack_quiet) {
crm_trace("Not sending nack message to node %s with feature set older "
"than 3.17.0", join_to);
return;
}
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, crm_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, crm_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, crm_msg_crmd, 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 != crm_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, crm_peer_seq,
highest_seq, fsa_state2string(cur_state), source);
if(highest_seq < crm_peer_seq) {
/* Don't spam the FSA with duplicates */
highest_seq = crm_peer_seq;
register_fsa_input_before(C_FSA_INTERNAL, I_NODE_JOIN, NULL);
}
} else if (cur_state == S_INTEGRATION) {
if (crmd_join_phase_count(crm_join_welcomed) == 0) {
int count = crmd_join_phase_count(crm_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(crm_join_welcomed) != 0) {
int count = crmd_join_phase_count(crm_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(crm_join_integrated) != 0) {
int count = crmd_join_phase_count(crm_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(crm_join_finalized) != 0) {
int count = crmd_join_phase_count(crm_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 crm_join_phase phase)
{
int count = 0;
pcmk__node_status_t *peer;
GHashTableIter iter;
g_hash_table_iter_init(&iter, crm_peer_cache);
while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &peer)) {
if(peer->join == phase) {
count++;
}
}
return count;
}
void crmd_join_phase_log(int level)
{
pcmk__node_status_t *peer;
GHashTableIter iter;
g_hash_table_iter_init(&iter, crm_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,
controld_join_phase_text(peer->join));
}
}
diff --git a/daemons/controld/controld_membership.c b/daemons/controld/controld_membership.c
index f190589ff1..efff40a881 100644
--- a/daemons/controld/controld_membership.c
+++ b/daemons/controld/controld_membership.c
@@ -1,498 +1,498 @@
/*
* Copyright 2004-2024 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.
*/
/* put these first so that uuid_t is defined without conflicts */
#include <crm_internal.h>
#include <string.h>
#include <crm/crm.h>
#include <crm/common/xml.h>
#include <crm/common/xml_internal.h>
#include <crm/cluster/internal.h>
#include <pacemaker-controld.h>
void post_cache_update(int instance);
extern gboolean check_join_state(enum crmd_fsa_state cur_state, const char *source);
/*!
* \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
*/
const char *
controld_join_phase_text(enum crm_join_phase phase)
{
switch (phase) {
case crm_join_nack_quiet:
return "nack_quiet";
case crm_join_nack:
return "nack";
case crm_join_none:
return "none";
case crm_join_welcomed:
return "welcomed";
case crm_join_integrated:
return "integrated";
case crm_join_finalized:
return "finalized";
case crm_join_confirmed:
return "confirmed";
default:
return "invalid";
}
}
static void
reap_dead_nodes(gpointer key, gpointer value, gpointer user_data)
{
pcmk__node_status_t *node = value;
if (pcmk__cluster_is_node_active(node)) {
return;
}
crm_update_peer_join(__func__, node, crm_join_none);
if ((node != NULL) && (node->name != NULL)) {
if (pcmk__str_eq(controld_globals.our_nodename, node->name,
pcmk__str_casei)) {
crm_err("We're not part of the cluster anymore");
register_fsa_input(C_FSA_INTERNAL, I_ERROR, NULL);
} else if (!AM_I_DC
&& pcmk__str_eq(node->name, controld_globals.dc_name,
pcmk__str_casei)) {
crm_warn("Our DC node (%s) left the cluster", node->name);
register_fsa_input(C_FSA_INTERNAL, I_ELECTION, NULL);
}
}
if ((controld_globals.fsa_state == S_INTEGRATION)
|| (controld_globals.fsa_state == S_FINALIZE_JOIN)) {
check_join_state(controld_globals.fsa_state, __func__);
}
- if ((node != NULL) && (node->uuid != NULL)) {
+ if ((node != NULL) && (node->xml_id != NULL)) {
fail_incompletable_actions(controld_globals.transition_graph,
- node->uuid);
+ node->xml_id);
}
}
void
post_cache_update(int instance)
{
xmlNode *no_op = NULL;
crm_peer_seq = instance;
crm_debug("Updated cache after membership event %d.", instance);
g_hash_table_foreach(crm_peer_cache, reap_dead_nodes, NULL);
controld_set_fsa_input_flags(R_MEMBERSHIP);
if (AM_I_DC) {
populate_cib_nodes(node_update_quick | node_update_cluster | node_update_peer |
node_update_expected, __func__);
}
/*
* If we lost nodes, we should re-check the election status
* Safe to call outside of an election
*/
controld_set_fsa_action_flags(A_ELECTION_CHECK);
controld_trigger_fsa();
/* Membership changed, remind everyone we're here.
* This will aid detection of duplicate DCs
*/
no_op = create_request(CRM_OP_NOOP, NULL, NULL, CRM_SYSTEM_CRMD,
AM_I_DC ? CRM_SYSTEM_DC : CRM_SYSTEM_CRMD, NULL);
pcmk__cluster_send_message(NULL, crm_msg_crmd, no_op);
pcmk__xml_free(no_op);
}
static void
crmd_node_update_complete(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data)
{
fsa_data_t *msg_data = NULL;
if (rc == pcmk_ok) {
crm_trace("Node update %d complete", call_id);
} else if(call_id < pcmk_ok) {
crm_err("Node update failed: %s (%d)", pcmk_strerror(call_id), call_id);
crm_log_xml_debug(msg, "failed");
register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL);
} else {
crm_err("Node update %d failed: %s (%d)", call_id, pcmk_strerror(rc), rc);
crm_log_xml_debug(msg, "failed");
register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL);
}
}
/*!
* \internal
* \brief Create an XML node state tag with updates
*
* \param[in,out] node Node whose state will be used for update
* \param[in] flags Bitmask of node_update_flags indicating what to update
* \param[in,out] parent XML node to contain update (or NULL)
* \param[in] source Who requested the update (only used for logging)
*
* \return Pointer to created node state tag
*/
xmlNode *
create_node_state_update(pcmk__node_status_t *node, int flags,
xmlNode *parent, const char *source)
{
const char *value = NULL;
xmlNode *node_state;
if (!node->state) {
crm_info("Node update for %s cancelled: no state, not seen yet",
node->name);
return NULL;
}
node_state = pcmk__xe_create(parent, PCMK__XE_NODE_STATE);
if (pcmk_is_set(node->flags, crm_remote_node)) {
pcmk__xe_set_bool_attr(node_state, PCMK_XA_REMOTE_NODE, true);
}
if (crm_xml_add(node_state, PCMK_XA_ID,
pcmk__cluster_node_uuid(node)) == NULL) {
crm_info("Node update for %s cancelled: no ID", node->name);
pcmk__xml_free(node_state);
return NULL;
}
crm_xml_add(node_state, PCMK_XA_UNAME, node->name);
if ((flags & node_update_cluster) && node->state) {
if (compare_version(controld_globals.dc_version, "3.18.0") >= 0) {
// A value 0 means the node is not a cluster member.
crm_xml_add_ll(node_state, PCMK__XA_IN_CCM, node->when_member);
} else {
pcmk__xe_set_bool_attr(node_state, PCMK__XA_IN_CCM,
pcmk__str_eq(node->state, CRM_NODE_MEMBER,
pcmk__str_casei));
}
}
if (!pcmk_is_set(node->flags, crm_remote_node)) {
if (flags & node_update_peer) {
if (compare_version(controld_globals.dc_version, "3.18.0") >= 0) {
// A value 0 means the peer is offline in CPG.
crm_xml_add_ll(node_state, PCMK_XA_CRMD, node->when_online);
} else {
// @COMPAT DCs < 2.1.7 use online/offline rather than timestamp
value = PCMK_VALUE_OFFLINE;
if (pcmk_is_set(node->processes, crm_get_cluster_proc())) {
value = PCMK_VALUE_ONLINE;
}
crm_xml_add(node_state, PCMK_XA_CRMD, value);
}
}
if (flags & node_update_join) {
if (node->join <= crm_join_none) {
value = CRMD_JOINSTATE_DOWN;
} else {
value = CRMD_JOINSTATE_MEMBER;
}
crm_xml_add(node_state, PCMK__XA_JOIN, value);
}
if (flags & node_update_expected) {
crm_xml_add(node_state, PCMK_XA_EXPECTED, node->expected);
}
}
crm_xml_add(node_state, PCMK_XA_CRM_DEBUG_ORIGIN, source);
return node_state;
}
static void
remove_conflicting_node_callback(xmlNode * msg, int call_id, int rc,
xmlNode * output, void *user_data)
{
char *node_uuid = user_data;
do_crm_log_unlikely(rc == 0 ? LOG_DEBUG : LOG_NOTICE,
"Deletion of the unknown conflicting node \"%s\": %s (rc=%d)",
node_uuid, pcmk_strerror(rc), rc);
}
static void
search_conflicting_node_callback(xmlNode * msg, int call_id, int rc,
xmlNode * output, void *user_data)
{
char *new_node_uuid = user_data;
xmlNode *node_xml = NULL;
if (rc != pcmk_ok) {
if (rc != -ENXIO) {
crm_notice("Searching conflicting nodes for %s failed: %s (%d)",
new_node_uuid, pcmk_strerror(rc), rc);
}
return;
} else if (output == NULL) {
return;
}
if (pcmk__xe_is(output, PCMK_XE_NODE)) {
node_xml = output;
} else {
node_xml = pcmk__xe_first_child(output, PCMK_XE_NODE, NULL, NULL);
}
for (; node_xml != NULL; node_xml = pcmk__xe_next_same(node_xml)) {
const char *node_uuid = NULL;
const char *node_uname = NULL;
GHashTableIter iter;
pcmk__node_status_t *node = NULL;
gboolean known = FALSE;
node_uuid = crm_element_value(node_xml, PCMK_XA_ID);
node_uname = crm_element_value(node_xml, PCMK_XA_UNAME);
if (node_uuid == NULL || node_uname == NULL) {
continue;
}
g_hash_table_iter_init(&iter, crm_peer_cache);
while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &node)) {
if ((node != NULL)
- && pcmk__str_eq(node->uuid, node_uuid, pcmk__str_casei)
+ && pcmk__str_eq(node->xml_id, node_uuid, pcmk__str_casei)
&& pcmk__str_eq(node->name, node_uname, pcmk__str_casei)) {
known = TRUE;
break;
}
}
if (known == FALSE) {
cib_t *cib_conn = controld_globals.cib_conn;
int delete_call_id = 0;
xmlNode *node_state_xml = NULL;
crm_notice("Deleting unknown node %s/%s which has conflicting uname with %s",
node_uuid, node_uname, new_node_uuid);
delete_call_id = cib_conn->cmds->remove(cib_conn, PCMK_XE_NODES,
node_xml, cib_none);
fsa_register_cib_callback(delete_call_id, pcmk__str_copy(node_uuid),
remove_conflicting_node_callback);
node_state_xml = pcmk__xe_create(NULL, PCMK__XE_NODE_STATE);
crm_xml_add(node_state_xml, PCMK_XA_ID, node_uuid);
crm_xml_add(node_state_xml, PCMK_XA_UNAME, node_uname);
delete_call_id = cib_conn->cmds->remove(cib_conn, PCMK_XE_STATUS,
node_state_xml, cib_none);
fsa_register_cib_callback(delete_call_id, pcmk__str_copy(node_uuid),
remove_conflicting_node_callback);
pcmk__xml_free(node_state_xml);
}
}
}
static void
node_list_update_callback(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data)
{
fsa_data_t *msg_data = NULL;
if(call_id < pcmk_ok) {
crm_err("Node list update failed: %s (%d)", pcmk_strerror(call_id), call_id);
crm_log_xml_debug(msg, "update:failed");
register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL);
} else if(rc < pcmk_ok) {
crm_err("Node update %d failed: %s (%d)", call_id, pcmk_strerror(rc), rc);
crm_log_xml_debug(msg, "update:failed");
register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL);
}
}
void
populate_cib_nodes(enum node_update_flags flags, const char *source)
{
cib_t *cib_conn = controld_globals.cib_conn;
int call_id = 0;
gboolean from_hashtable = TRUE;
xmlNode *node_list = pcmk__xe_create(NULL, PCMK_XE_NODES);
#if SUPPORT_COROSYNC
if (!pcmk_is_set(flags, node_update_quick)
&& (pcmk_get_cluster_layer() == pcmk_cluster_layer_corosync)) {
from_hashtable = pcmk__corosync_add_nodes(node_list);
}
#endif
if (from_hashtable) {
GHashTableIter iter;
pcmk__node_status_t *node = NULL;
GString *xpath = NULL;
g_hash_table_iter_init(&iter, crm_peer_cache);
while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &node)) {
xmlNode *new_node = NULL;
- if ((node->uuid != NULL) && (node->name != NULL)) {
+ if ((node->xml_id != NULL) && (node->name != NULL)) {
crm_trace("Creating node entry for %s/%s",
- node->name, node->uuid);
+ node->name, node->xml_id);
if (xpath == NULL) {
xpath = g_string_sized_new(512);
} else {
g_string_truncate(xpath, 0);
}
/* We need both to be valid */
new_node = pcmk__xe_create(node_list, PCMK_XE_NODE);
- crm_xml_add(new_node, PCMK_XA_ID, node->uuid);
+ crm_xml_add(new_node, PCMK_XA_ID, node->xml_id);
crm_xml_add(new_node, PCMK_XA_UNAME, node->name);
/* Search and remove unknown nodes with the conflicting uname from CIB */
pcmk__g_strcat(xpath,
"/" PCMK_XE_CIB "/" PCMK_XE_CONFIGURATION
"/" PCMK_XE_NODES "/" PCMK_XE_NODE
"[@" PCMK_XA_UNAME "='", node->name, "']"
- "[@" PCMK_XA_ID "!='", node->uuid, "']", NULL);
+ "[@" PCMK_XA_ID "!='", node->xml_id, "']", NULL);
call_id = cib_conn->cmds->query(cib_conn,
(const char *) xpath->str, NULL,
cib_xpath);
- fsa_register_cib_callback(call_id, pcmk__str_copy(node->uuid),
+ fsa_register_cib_callback(call_id, pcmk__str_copy(node->xml_id),
search_conflicting_node_callback);
}
}
if (xpath != NULL) {
g_string_free(xpath, TRUE);
}
}
crm_trace("Populating <nodes> section from %s", from_hashtable ? "hashtable" : "cluster");
if ((controld_update_cib(PCMK_XE_NODES, node_list, cib_none,
node_list_update_callback) == pcmk_rc_ok)
&& (crm_peer_cache != NULL) && AM_I_DC) {
/*
* There is no need to update the local CIB with our values if
* we've not seen valid membership data
*/
GHashTableIter iter;
pcmk__node_status_t *node = NULL;
pcmk__xml_free(node_list);
node_list = pcmk__xe_create(NULL, PCMK_XE_STATUS);
g_hash_table_iter_init(&iter, crm_peer_cache);
while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &node)) {
create_node_state_update(node, flags, node_list, source);
}
if (crm_remote_peer_cache) {
g_hash_table_iter_init(&iter, crm_remote_peer_cache);
while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &node)) {
create_node_state_update(node, flags, node_list, source);
}
}
controld_update_cib(PCMK_XE_STATUS, node_list, cib_none,
crmd_node_update_complete);
}
pcmk__xml_free(node_list);
}
static void
cib_quorum_update_complete(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data)
{
fsa_data_t *msg_data = NULL;
if (rc == pcmk_ok) {
crm_trace("Quorum update %d complete", call_id);
} else {
crm_err("Quorum update %d failed: %s (%d)", call_id, pcmk_strerror(rc), rc);
crm_log_xml_debug(msg, "failed");
register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL);
}
}
void
crm_update_quorum(gboolean quorum, gboolean force_update)
{
bool has_quorum = pcmk_is_set(controld_globals.flags, controld_has_quorum);
if (quorum) {
controld_set_global_flags(controld_ever_had_quorum);
} else if (pcmk_all_flags_set(controld_globals.flags,
controld_ever_had_quorum
|controld_no_quorum_suicide)) {
pcmk__panic(__func__);
}
if (AM_I_DC
&& ((has_quorum && !quorum) || (!has_quorum && quorum)
|| force_update)) {
xmlNode *update = NULL;
update = pcmk__xe_create(NULL, PCMK_XE_CIB);
crm_xml_add_int(update, PCMK_XA_HAVE_QUORUM, quorum);
crm_xml_add(update, PCMK_XA_DC_UUID, controld_globals.our_uuid);
crm_debug("Updating quorum status to %s", pcmk__btoa(quorum));
controld_update_cib(PCMK_XE_CIB, update, cib_none,
cib_quorum_update_complete);
pcmk__xml_free(update);
/* Quorum changes usually cause a new transition via other activity:
* quorum gained via a node joining will abort via the node join,
* and quorum lost via a node leaving will usually abort via resource
* activity and/or fencing.
*
* However, it is possible that nothing else causes a transition (e.g.
* someone forces quorum via corosync-cmaptcl, or quorum is lost due to
* a node in standby shutting down cleanly), so here ensure a new
* transition is triggered.
*/
if (quorum) {
/* If quorum was gained, abort after a short delay, in case multiple
* nodes are joining around the same time, so the one that brings us
* to quorum doesn't cause all the remaining ones to be fenced.
*/
abort_after_delay(PCMK_SCORE_INFINITY, pcmk__graph_restart,
"Quorum gained", 5000);
} else {
abort_transition(PCMK_SCORE_INFINITY, pcmk__graph_restart,
"Quorum lost", NULL);
}
}
if (quorum) {
controld_set_global_flags(controld_has_quorum);
} else {
controld_clear_global_flags(controld_has_quorum);
}
}
diff --git a/daemons/controld/controld_messages.c b/daemons/controld/controld_messages.c
index 159efbdbd6..498d2bc427 100644
--- a/daemons/controld/controld_messages.c
+++ b/daemons/controld/controld_messages.c
@@ -1,1359 +1,1359 @@
/*
* Copyright 2004-2024 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 <crm_internal.h>
#include <sys/param.h>
#include <string.h>
#include <time.h>
#include <crm/crm.h>
#include <crm/common/xml.h>
#include <crm/cluster/internal.h>
#include <crm/cib.h>
#include <crm/common/ipc_internal.h>
#include <pacemaker-controld.h>
static enum crmd_fsa_input handle_message(xmlNode *msg,
enum crmd_fsa_cause cause);
static void handle_response(xmlNode *stored_msg);
static enum crmd_fsa_input handle_request(xmlNode *stored_msg,
enum crmd_fsa_cause cause);
static enum crmd_fsa_input handle_shutdown_request(xmlNode *stored_msg);
static void send_msg_via_ipc(xmlNode * msg, const char *sys);
/* debug only, can wrap all it likes */
static int last_data_id = 0;
void
register_fsa_error_adv(enum crmd_fsa_cause cause, enum crmd_fsa_input input,
fsa_data_t * cur_data, void *new_data, const char *raised_from)
{
/* save the current actions if any */
if (controld_globals.fsa_actions != A_NOTHING) {
register_fsa_input_adv(cur_data ? cur_data->fsa_cause : C_FSA_INTERNAL,
I_NULL, cur_data ? cur_data->data : NULL,
controld_globals.fsa_actions, TRUE, __func__);
}
/* reset the action list */
crm_info("Resetting the current action list");
fsa_dump_actions(controld_globals.fsa_actions, "Drop");
controld_globals.fsa_actions = A_NOTHING;
/* register the error */
register_fsa_input_adv(cause, input, new_data, A_NOTHING, TRUE, raised_from);
}
void
register_fsa_input_adv(enum crmd_fsa_cause cause, enum crmd_fsa_input input,
void *data, uint64_t with_actions,
gboolean prepend, const char *raised_from)
{
unsigned old_len = g_list_length(controld_globals.fsa_message_queue);
fsa_data_t *fsa_data = NULL;
if (raised_from == NULL) {
raised_from = "<unknown>";
}
if (input == I_NULL && with_actions == A_NOTHING /* && data == NULL */ ) {
/* no point doing anything */
crm_err("Cannot add entry to queue: no input and no action");
return;
}
if (input == I_WAIT_FOR_EVENT) {
controld_set_global_flags(controld_fsa_is_stalled);
crm_debug("Stalling the FSA pending further input: source=%s cause=%s data=%p queue=%d",
raised_from, fsa_cause2string(cause), data, old_len);
if (old_len > 0) {
fsa_dump_queue(LOG_TRACE);
prepend = FALSE;
}
if (data == NULL) {
controld_set_fsa_action_flags(with_actions);
fsa_dump_actions(with_actions, "Restored");
return;
}
/* Store everything in the new event and reset
* controld_globals.fsa_actions
*/
with_actions |= controld_globals.fsa_actions;
controld_globals.fsa_actions = A_NOTHING;
}
last_data_id++;
crm_trace("%s %s FSA input %d (%s) due to %s, %s data",
raised_from, (prepend? "prepended" : "appended"), last_data_id,
fsa_input2string(input), fsa_cause2string(cause),
(data? "with" : "without"));
fsa_data = pcmk__assert_alloc(1, sizeof(fsa_data_t));
fsa_data->id = last_data_id;
fsa_data->fsa_input = input;
fsa_data->fsa_cause = cause;
fsa_data->origin = raised_from;
fsa_data->data = NULL;
fsa_data->data_type = fsa_dt_none;
fsa_data->actions = with_actions;
if (with_actions != A_NOTHING) {
crm_trace("Adding actions %.16llx to input",
(unsigned long long) with_actions);
}
if (data != NULL) {
switch (cause) {
case C_FSA_INTERNAL:
case C_CRMD_STATUS_CALLBACK:
case C_IPC_MESSAGE:
case C_HA_MESSAGE:
CRM_CHECK(((ha_msg_input_t *) data)->msg != NULL,
crm_err("Bogus data from %s", raised_from));
crm_trace("Copying %s data from %s as cluster message data",
fsa_cause2string(cause), raised_from);
fsa_data->data = copy_ha_msg_input(data);
fsa_data->data_type = fsa_dt_ha_msg;
break;
case C_LRM_OP_CALLBACK:
crm_trace("Copying %s data from %s as lrmd_event_data_t",
fsa_cause2string(cause), raised_from);
fsa_data->data = lrmd_copy_event((lrmd_event_data_t *) data);
fsa_data->data_type = fsa_dt_lrm;
break;
case C_TIMER_POPPED:
case C_SHUTDOWN:
case C_UNKNOWN:
case C_STARTUP:
crm_crit("Copying %s data (from %s) is not yet implemented",
fsa_cause2string(cause), raised_from);
crmd_exit(CRM_EX_SOFTWARE);
break;
}
}
/* make sure to free it properly later */
if (prepend) {
controld_globals.fsa_message_queue
= g_list_prepend(controld_globals.fsa_message_queue, fsa_data);
} else {
controld_globals.fsa_message_queue
= g_list_append(controld_globals.fsa_message_queue, fsa_data);
}
crm_trace("FSA message queue length is %d",
g_list_length(controld_globals.fsa_message_queue));
/* fsa_dump_queue(LOG_TRACE); */
if (old_len == g_list_length(controld_globals.fsa_message_queue)) {
crm_err("Couldn't add message to the queue");
}
if (input != I_WAIT_FOR_EVENT) {
controld_trigger_fsa();
}
}
void
fsa_dump_queue(int log_level)
{
int offset = 0;
for (GList *iter = controld_globals.fsa_message_queue; iter != NULL;
iter = iter->next) {
fsa_data_t *data = (fsa_data_t *) iter->data;
do_crm_log_unlikely(log_level,
"queue[%d.%d]: input %s raised by %s(%p.%d)\t(cause=%s)",
offset++, data->id, fsa_input2string(data->fsa_input),
data->origin, data->data, data->data_type,
fsa_cause2string(data->fsa_cause));
}
}
ha_msg_input_t *
copy_ha_msg_input(ha_msg_input_t * orig)
{
xmlNode *wrapper = NULL;
ha_msg_input_t *copy = pcmk__assert_alloc(1, sizeof(ha_msg_input_t));
copy->msg = (orig != NULL)? pcmk__xml_copy(NULL, orig->msg) : NULL;
wrapper = pcmk__xe_first_child(copy->msg, PCMK__XE_CRM_XML, NULL, NULL);
copy->xml = pcmk__xe_first_child(wrapper, NULL, NULL, NULL);
return copy;
}
void
delete_fsa_input(fsa_data_t * fsa_data)
{
lrmd_event_data_t *op = NULL;
xmlNode *foo = NULL;
if (fsa_data == NULL) {
return;
}
crm_trace("About to free %s data", fsa_cause2string(fsa_data->fsa_cause));
if (fsa_data->data != NULL) {
switch (fsa_data->data_type) {
case fsa_dt_ha_msg:
delete_ha_msg_input(fsa_data->data);
break;
case fsa_dt_xml:
foo = fsa_data->data;
pcmk__xml_free(foo);
break;
case fsa_dt_lrm:
op = (lrmd_event_data_t *) fsa_data->data;
lrmd_free_event(op);
break;
case fsa_dt_none:
if (fsa_data->data != NULL) {
crm_err("Don't know how to free %s data from %s",
fsa_cause2string(fsa_data->fsa_cause), fsa_data->origin);
crmd_exit(CRM_EX_SOFTWARE);
}
break;
}
crm_trace("%s data freed", fsa_cause2string(fsa_data->fsa_cause));
}
free(fsa_data);
}
/* returns the next message */
fsa_data_t *
get_message(void)
{
fsa_data_t *message
= (fsa_data_t *) controld_globals.fsa_message_queue->data;
controld_globals.fsa_message_queue
= g_list_remove(controld_globals.fsa_message_queue, message);
crm_trace("Processing input %d", message->id);
return message;
}
void *
fsa_typed_data_adv(fsa_data_t * fsa_data, enum fsa_data_type a_type, const char *caller)
{
void *ret_val = NULL;
if (fsa_data == NULL) {
crm_err("%s: No FSA data available", caller);
} else if (fsa_data->data == NULL) {
crm_err("%s: No message data available. Origin: %s", caller, fsa_data->origin);
} else if (fsa_data->data_type != a_type) {
crm_crit("%s: Message data was the wrong type! %d vs. requested=%d. Origin: %s",
caller, fsa_data->data_type, a_type, fsa_data->origin);
CRM_ASSERT(fsa_data->data_type == a_type);
} else {
ret_val = fsa_data->data;
}
return ret_val;
}
/* A_MSG_ROUTE */
void
do_msg_route(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)
{
ha_msg_input_t *input = fsa_typed_data(fsa_dt_ha_msg);
route_message(msg_data->fsa_cause, input->msg);
}
void
route_message(enum crmd_fsa_cause cause, xmlNode * input)
{
ha_msg_input_t fsa_input;
enum crmd_fsa_input result = I_NULL;
fsa_input.msg = input;
CRM_CHECK(cause == C_IPC_MESSAGE || cause == C_HA_MESSAGE, return);
/* try passing the buck first */
if (relay_message(input, cause == C_IPC_MESSAGE)) {
return;
}
/* handle locally */
result = handle_message(input, cause);
/* done or process later? */
switch (result) {
case I_NULL:
case I_CIB_OP:
case I_ROUTER:
case I_NODE_JOIN:
case I_JOIN_REQUEST:
case I_JOIN_RESULT:
break;
default:
/* Defering local processing of message */
register_fsa_input_later(cause, result, &fsa_input);
return;
}
if (result != I_NULL) {
/* add to the front of the queue */
register_fsa_input(cause, result, &fsa_input);
}
}
gboolean
relay_message(xmlNode * msg, gboolean originated_locally)
{
enum crm_ais_msg_types dest = crm_msg_none;
bool is_for_dc = false;
bool is_for_dcib = false;
bool is_for_te = false;
bool is_for_crm = false;
bool is_for_cib = false;
bool is_local = false;
bool broadcast = false;
const char *host_to = NULL;
const char *sys_to = NULL;
const char *sys_from = NULL;
const char *type = NULL;
const char *task = NULL;
const char *ref = NULL;
pcmk__node_status_t *node_to = NULL;
CRM_CHECK(msg != NULL, return TRUE);
host_to = crm_element_value(msg, PCMK__XA_CRM_HOST_TO);
sys_to = crm_element_value(msg, PCMK__XA_CRM_SYS_TO);
sys_from = crm_element_value(msg, PCMK__XA_CRM_SYS_FROM);
type = crm_element_value(msg, PCMK__XA_T);
task = crm_element_value(msg, PCMK__XA_CRM_TASK);
ref = crm_element_value(msg, PCMK_XA_REFERENCE);
broadcast = pcmk__str_empty(host_to);
if (ref == NULL) {
ref = "without reference ID";
}
if (pcmk__str_eq(task, CRM_OP_HELLO, pcmk__str_casei)) {
crm_trace("Received hello %s from %s (no processing needed)",
ref, pcmk__s(sys_from, "unidentified source"));
crm_log_xml_trace(msg, "hello");
return TRUE;
}
// Require message type (set by create_request())
if (!pcmk__str_eq(type, PCMK__VALUE_CRMD, pcmk__str_none)) {
crm_warn("Ignoring invalid message %s with type '%s' "
"(not '" PCMK__VALUE_CRMD "')",
ref, pcmk__s(type, ""));
crm_log_xml_trace(msg, "ignored");
return TRUE;
}
// Require a destination subsystem (also set by create_request())
if (sys_to == NULL) {
crm_warn("Ignoring invalid message %s with no " PCMK__XA_CRM_SYS_TO,
ref);
crm_log_xml_trace(msg, "ignored");
return TRUE;
}
// Get the message type appropriate to the destination subsystem
if (pcmk_get_cluster_layer() == pcmk_cluster_layer_corosync) {
dest = pcmk__cluster_parse_msg_type(sys_to);
if (dest == crm_msg_none) {
/* Unrecognized value, use a sane default
*
* @TODO Maybe we should bail instead
*/
dest = crm_msg_crmd;
}
}
is_for_dc = (strcasecmp(CRM_SYSTEM_DC, sys_to) == 0);
is_for_dcib = (strcasecmp(CRM_SYSTEM_DCIB, sys_to) == 0);
is_for_te = (strcasecmp(CRM_SYSTEM_TENGINE, sys_to) == 0);
is_for_cib = (strcasecmp(CRM_SYSTEM_CIB, sys_to) == 0);
is_for_crm = (strcasecmp(CRM_SYSTEM_CRMD, sys_to) == 0);
// Check whether message should be processed locally
is_local = false;
if (broadcast) {
if (is_for_dc || is_for_te) {
is_local = false;
} else if (is_for_crm) {
if (pcmk__strcase_any_of(task, CRM_OP_NODE_INFO,
PCMK__CONTROLD_CMD_NODES, NULL)) {
/* Node info requests do not specify a host, which is normally
* treated as "all hosts", because the whole point is that the
* client may not know the local node name. Always handle these
* requests locally.
*/
is_local = true;
} else {
is_local = !originated_locally;
}
} else {
is_local = true;
}
} else if (pcmk__str_eq(controld_globals.our_nodename, host_to,
pcmk__str_casei)) {
is_local = true;
} else if (is_for_crm && pcmk__str_eq(task, CRM_OP_LRM_DELETE, pcmk__str_casei)) {
xmlNode *wrapper = pcmk__xe_first_child(msg, PCMK__XE_CRM_XML, NULL,
NULL);
xmlNode *msg_data = pcmk__xe_first_child(wrapper, NULL, NULL, NULL);
const char *mode = crm_element_value(msg_data, PCMK__XA_MODE);
if (pcmk__str_eq(mode, PCMK__VALUE_CIB, pcmk__str_none)) {
// Local delete of an offline node's resource history
is_local = true;
}
}
// Check whether message should be relayed
if (is_for_dc || is_for_dcib || is_for_te) {
if (AM_I_DC) {
if (is_for_te) {
crm_trace("Route message %s locally as transition request",
ref);
crm_log_xml_trace(msg, sys_to);
send_msg_via_ipc(msg, sys_to);
return TRUE; // No further processing of message is needed
}
crm_trace("Route message %s locally as DC request", ref);
return FALSE; // More to be done by caller
}
if (originated_locally
&& !pcmk__strcase_any_of(sys_from, CRM_SYSTEM_PENGINE,
CRM_SYSTEM_TENGINE, NULL)) {
crm_trace("Relay message %s to DC (via %s)",
ref, pcmk__s(host_to, "broadcast"));
crm_log_xml_trace(msg, "relayed");
if (!broadcast) {
node_to = pcmk__get_node(0, host_to, NULL,
pcmk__node_search_cluster_member);
}
pcmk__cluster_send_message(node_to, dest, msg);
return TRUE;
}
/* Transition engine and scheduler messages are sent only to the DC on
* the same node. If we are no longer the DC, discard this message.
*/
crm_trace("Ignoring message %s because we are no longer DC", ref);
crm_log_xml_trace(msg, "ignored");
return TRUE; // No further processing of message is needed
}
if (is_local) {
if (is_for_crm || is_for_cib) {
crm_trace("Route message %s locally as controller request", ref);
return FALSE; // More to be done by caller
}
crm_trace("Relay message %s locally to %s", ref, sys_to);
crm_log_xml_trace(msg, "IPC-relay");
send_msg_via_ipc(msg, sys_to);
return TRUE;
}
if (!broadcast) {
node_to = pcmk__search_node_caches(0, host_to,
pcmk__node_search_cluster_member);
if (node_to == NULL) {
crm_warn("Ignoring message %s because node %s is unknown",
ref, host_to);
crm_log_xml_trace(msg, "ignored");
return TRUE;
}
}
crm_trace("Relay message %s to %s",
ref, pcmk__s(host_to, "all peers"));
crm_log_xml_trace(msg, "relayed");
pcmk__cluster_send_message(node_to, dest, msg);
return TRUE;
}
// Return true if field contains a positive integer
static bool
authorize_version(xmlNode *message_data, const char *field,
const char *client_name, const char *ref, const char *uuid)
{
const char *version = crm_element_value(message_data, field);
long long version_num;
if ((pcmk__scan_ll(version, &version_num, -1LL) != pcmk_rc_ok)
|| (version_num < 0LL)) {
crm_warn("Rejected IPC hello from %s: '%s' is not a valid protocol %s "
QB_XS " ref=%s uuid=%s",
client_name, ((version == NULL)? "" : version),
field, (ref? ref : "none"), uuid);
return false;
}
return true;
}
/*!
* \internal
* \brief Check whether a client IPC message is acceptable
*
* If a given client IPC message is a hello, "authorize" it by ensuring it has
* valid information such as a protocol version, and return false indicating
* that nothing further needs to be done with the message. If the message is not
* a hello, just return true to indicate it needs further processing.
*
* \param[in] client_msg XML of IPC message
* \param[in,out] curr_client If IPC is not proxied, client that sent message
* \param[in] proxy_session If IPC is proxied, the session ID
*
* \return true if message needs further processing, false if it doesn't
*/
bool
controld_authorize_ipc_message(const xmlNode *client_msg, pcmk__client_t *curr_client,
const char *proxy_session)
{
xmlNode *wrapper = NULL;
xmlNode *message_data = NULL;
const char *client_name = NULL;
const char *op = crm_element_value(client_msg, PCMK__XA_CRM_TASK);
const char *ref = crm_element_value(client_msg, PCMK_XA_REFERENCE);
const char *uuid = (curr_client? curr_client->id : proxy_session);
if (uuid == NULL) {
crm_warn("IPC message from client rejected: No client identifier "
QB_XS " ref=%s", (ref? ref : "none"));
goto rejected;
}
if (!pcmk__str_eq(CRM_OP_HELLO, op, pcmk__str_casei)) {
// Only hello messages need to be authorized
return true;
}
wrapper = pcmk__xe_first_child(client_msg, PCMK__XE_CRM_XML, NULL, NULL);
message_data = pcmk__xe_first_child(wrapper, NULL, NULL, NULL);
client_name = crm_element_value(message_data, PCMK__XA_CLIENT_NAME);
if (pcmk__str_empty(client_name)) {
crm_warn("IPC hello from client rejected: No client name",
QB_XS " ref=%s uuid=%s", (ref? ref : "none"), uuid);
goto rejected;
}
if (!authorize_version(message_data, PCMK__XA_MAJOR_VERSION, client_name,
ref, uuid)) {
goto rejected;
}
if (!authorize_version(message_data, PCMK__XA_MINOR_VERSION, client_name,
ref, uuid)) {
goto rejected;
}
crm_trace("Validated IPC hello from client %s", client_name);
crm_log_xml_trace(client_msg, "hello");
if (curr_client) {
curr_client->userdata = pcmk__str_copy(client_name);
}
controld_trigger_fsa();
return false;
rejected:
crm_log_xml_trace(client_msg, "rejected");
if (curr_client) {
qb_ipcs_disconnect(curr_client->ipcs);
}
return false;
}
static enum crmd_fsa_input
handle_message(xmlNode *msg, enum crmd_fsa_cause cause)
{
const char *type = NULL;
CRM_CHECK(msg != NULL, return I_NULL);
type = crm_element_value(msg, PCMK__XA_SUBT);
if (pcmk__str_eq(type, PCMK__VALUE_REQUEST, pcmk__str_none)) {
return handle_request(msg, cause);
}
if (pcmk__str_eq(type, PCMK__VALUE_RESPONSE, pcmk__str_none)) {
handle_response(msg);
return I_NULL;
}
crm_warn("Ignoring message with unknown " PCMK__XA_SUBT" '%s'",
pcmk__s(type, ""));
crm_log_xml_trace(msg, "bad");
return I_NULL;
}
static enum crmd_fsa_input
handle_failcount_op(xmlNode * stored_msg)
{
const char *rsc = NULL;
const char *uname = NULL;
const char *op = NULL;
char *interval_spec = NULL;
guint interval_ms = 0;
gboolean is_remote_node = FALSE;
xmlNode *wrapper = pcmk__xe_first_child(stored_msg, PCMK__XE_CRM_XML, NULL,
NULL);
xmlNode *xml_op = pcmk__xe_first_child(wrapper, NULL, NULL, NULL);
if (xml_op) {
xmlNode *xml_rsc = pcmk__xe_first_child(xml_op, PCMK_XE_PRIMITIVE, NULL,
NULL);
xmlNode *xml_attrs = pcmk__xe_first_child(xml_op, PCMK__XE_ATTRIBUTES,
NULL, NULL);
if (xml_rsc) {
rsc = pcmk__xe_id(xml_rsc);
}
if (xml_attrs) {
op = crm_element_value(xml_attrs,
CRM_META "_" PCMK__META_CLEAR_FAILURE_OP);
crm_element_value_ms(xml_attrs,
CRM_META "_" PCMK__META_CLEAR_FAILURE_INTERVAL,
&interval_ms);
}
}
uname = crm_element_value(xml_op, PCMK__META_ON_NODE);
if ((rsc == NULL) || (uname == NULL)) {
crm_log_xml_warn(stored_msg, "invalid failcount op");
return I_NULL;
}
if (crm_element_value(xml_op, PCMK__XA_ROUTER_NODE)) {
is_remote_node = TRUE;
}
crm_debug("Clearing failures for %s-interval %s on %s "
"from attribute manager, CIB, and executor state",
pcmk__readable_interval(interval_ms), rsc, uname);
if (interval_ms) {
interval_spec = crm_strdup_printf("%ums", interval_ms);
}
update_attrd_clear_failures(uname, rsc, op, interval_spec, is_remote_node);
free(interval_spec);
controld_cib_delete_last_failure(rsc, uname, op, interval_ms);
lrm_clear_last_failure(rsc, uname, op, interval_ms);
return I_NULL;
}
static enum crmd_fsa_input
handle_lrm_delete(xmlNode *stored_msg)
{
const char *mode = NULL;
xmlNode *wrapper = pcmk__xe_first_child(stored_msg, PCMK__XE_CRM_XML, NULL,
NULL);
xmlNode *msg_data = pcmk__xe_first_child(wrapper, NULL, NULL, NULL);
CRM_CHECK(msg_data != NULL, return I_NULL);
/* CRM_OP_LRM_DELETE has two distinct modes. The default behavior is to
* relay the operation to the affected node, which will unregister the
* resource from the local executor, clear the resource's history from the
* CIB, and do some bookkeeping in the controller.
*
* However, if the affected node is offline, the client will specify
* mode=PCMK__VALUE_CIB which means the controller receiving the operation
* should clear the resource's history from the CIB and nothing else. This
* is used to clear shutdown locks.
*/
mode = crm_element_value(msg_data, PCMK__XA_MODE);
if (!pcmk__str_eq(mode, PCMK__VALUE_CIB, pcmk__str_none)) {
// Relay to affected node
crm_xml_add(stored_msg, PCMK__XA_CRM_SYS_TO, CRM_SYSTEM_LRMD);
return I_ROUTER;
} else {
// Delete CIB history locally (compare with do_lrm_delete())
const char *from_sys = NULL;
const char *user_name = NULL;
const char *rsc_id = NULL;
const char *node = NULL;
xmlNode *rsc_xml = NULL;
int rc = pcmk_rc_ok;
rsc_xml = pcmk__xe_first_child(msg_data, PCMK_XE_PRIMITIVE, NULL, NULL);
CRM_CHECK(rsc_xml != NULL, return I_NULL);
rsc_id = pcmk__xe_id(rsc_xml);
from_sys = crm_element_value(stored_msg, PCMK__XA_CRM_SYS_FROM);
node = crm_element_value(msg_data, PCMK__META_ON_NODE);
user_name = pcmk__update_acl_user(stored_msg, PCMK__XA_CRM_USER, NULL);
crm_debug("Handling " CRM_OP_LRM_DELETE " for %s on %s locally%s%s "
"(clearing CIB resource history only)", rsc_id, node,
(user_name? " for user " : ""), (user_name? user_name : ""));
rc = controld_delete_resource_history(rsc_id, node, user_name,
cib_dryrun|cib_sync_call);
if (rc == pcmk_rc_ok) {
rc = controld_delete_resource_history(rsc_id, node, user_name,
crmd_cib_smart_opt());
}
/* Notify client. Also notify tengine if mode=PCMK__VALUE_CIB and
* op=CRM_OP_LRM_DELETE.
*/
if (from_sys) {
lrmd_event_data_t *op = NULL;
const char *from_host = crm_element_value(stored_msg, PCMK__XA_SRC);
const char *transition;
if (strcmp(from_sys, CRM_SYSTEM_TENGINE)) {
transition = crm_element_value(msg_data,
PCMK__XA_TRANSITION_KEY);
} else {
transition = crm_element_value(stored_msg,
PCMK__XA_TRANSITION_KEY);
}
crm_info("Notifying %s on %s that %s was%s deleted",
from_sys, (from_host? from_host : "local node"), rsc_id,
((rc == pcmk_rc_ok)? "" : " not"));
op = lrmd_new_event(rsc_id, PCMK_ACTION_DELETE, 0);
op->type = lrmd_event_exec_complete;
op->user_data = pcmk__str_copy(pcmk__s(transition, FAKE_TE_ID));
op->params = pcmk__strkey_table(free, free);
pcmk__insert_dup(op->params, PCMK_XA_CRM_FEATURE_SET,
CRM_FEATURE_SET);
controld_rc2event(op, 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);
}
return I_NULL;
}
}
/*!
* \brief Handle a CRM_OP_REMOTE_STATE message by updating remote peer cache
*
* \param[in] msg Message XML
*
* \return Next FSA input
*/
static enum crmd_fsa_input
handle_remote_state(const xmlNode *msg)
{
const char *conn_host = NULL;
const char *remote_uname = pcmk__xe_id(msg);
pcmk__node_status_t *remote_peer;
bool remote_is_up = false;
int rc = pcmk_rc_ok;
rc = pcmk__xe_get_bool_attr(msg, PCMK__XA_IN_CCM, &remote_is_up);
CRM_CHECK(remote_uname && rc == pcmk_rc_ok, return I_NULL);
remote_peer = pcmk__cluster_lookup_remote_node(remote_uname);
CRM_CHECK(remote_peer, return I_NULL);
pcmk__update_peer_state(__func__, remote_peer,
remote_is_up ? CRM_NODE_MEMBER : CRM_NODE_LOST,
0);
conn_host = crm_element_value(msg, PCMK__XA_CONNECTION_HOST);
if (conn_host) {
pcmk__str_update(&remote_peer->conn_host, conn_host);
} else if (remote_peer->conn_host) {
free(remote_peer->conn_host);
remote_peer->conn_host = NULL;
}
return I_NULL;
}
/*!
* \brief Handle a CRM_OP_PING message
*
* \param[in] msg Message XML
*
* \return Next FSA input
*/
static enum crmd_fsa_input
handle_ping(const xmlNode *msg)
{
const char *value = NULL;
xmlNode *ping = NULL;
xmlNode *reply = NULL;
// Build reply
ping = pcmk__xe_create(NULL, PCMK__XE_PING_RESPONSE);
value = crm_element_value(msg, PCMK__XA_CRM_SYS_TO);
crm_xml_add(ping, PCMK__XA_CRM_SUBSYSTEM, value);
// Add controller state
value = fsa_state2string(controld_globals.fsa_state);
crm_xml_add(ping, PCMK__XA_CRMD_STATE, value);
crm_notice("Current ping state: %s", value); // CTS needs this
// Add controller health
// @TODO maybe do some checks to determine meaningful status
crm_xml_add(ping, PCMK_XA_RESULT, "ok");
// Send reply
reply = create_reply(msg, ping);
pcmk__xml_free(ping);
if (reply != NULL) {
(void) relay_message(reply, TRUE);
pcmk__xml_free(reply);
}
// Nothing further to do
return I_NULL;
}
/*!
* \brief Handle a PCMK__CONTROLD_CMD_NODES message
*
* \param[in] request Message XML
*
* \return Next FSA input
*/
static enum crmd_fsa_input
handle_node_list(const xmlNode *request)
{
GHashTableIter iter;
pcmk__node_status_t *node = NULL;
xmlNode *reply = NULL;
xmlNode *reply_data = NULL;
// Create message data for reply
reply_data = pcmk__xe_create(NULL, PCMK_XE_NODES);
g_hash_table_iter_init(&iter, crm_peer_cache);
while (g_hash_table_iter_next(&iter, NULL, (gpointer *) & node)) {
xmlNode *xml = pcmk__xe_create(reply_data, PCMK_XE_NODE);
crm_xml_add_ll(xml, PCMK_XA_ID,
(long long) node->cluster_layer_id); // uint32_t
crm_xml_add(xml, PCMK_XA_UNAME, node->name);
crm_xml_add(xml, PCMK__XA_IN_CCM, node->state);
}
// Create and send reply
reply = create_reply(request, reply_data);
pcmk__xml_free(reply_data);
if (reply) {
(void) relay_message(reply, TRUE);
pcmk__xml_free(reply);
}
// Nothing further to do
return I_NULL;
}
/*!
* \brief Handle a CRM_OP_NODE_INFO request
*
* \param[in] msg Message XML
*
* \return Next FSA input
*/
static enum crmd_fsa_input
handle_node_info_request(const xmlNode *msg)
{
const char *value = NULL;
pcmk__node_status_t *node = NULL;
int node_id = 0;
xmlNode *reply = NULL;
xmlNode *reply_data = NULL;
// Build reply
reply_data = pcmk__xe_create(NULL, PCMK_XE_NODE);
crm_xml_add(reply_data, PCMK__XA_CRM_SUBSYSTEM, CRM_SYSTEM_CRMD);
// Add whether current partition has quorum
pcmk__xe_set_bool_attr(reply_data, PCMK_XA_HAVE_QUORUM,
pcmk_is_set(controld_globals.flags,
controld_has_quorum));
// Check whether client requested node info by ID and/or name
crm_element_value_int(msg, PCMK_XA_ID, &node_id);
if (node_id < 0) {
node_id = 0;
}
value = crm_element_value(msg, PCMK_XA_UNAME);
// Default to local node if none given
if ((node_id == 0) && (value == NULL)) {
value = controld_globals.our_nodename;
}
node = pcmk__search_node_caches(node_id, value, pcmk__node_search_any);
if (node) {
- crm_xml_add(reply_data, PCMK_XA_ID, node->uuid);
+ crm_xml_add(reply_data, PCMK_XA_ID, node->xml_id);
crm_xml_add(reply_data, PCMK_XA_UNAME, node->name);
crm_xml_add(reply_data, PCMK_XA_CRMD, node->state);
pcmk__xe_set_bool_attr(reply_data, PCMK_XA_REMOTE_NODE,
pcmk_is_set(node->flags, crm_remote_node));
}
// Send reply
reply = create_reply(msg, reply_data);
pcmk__xml_free(reply_data);
if (reply != NULL) {
(void) relay_message(reply, TRUE);
pcmk__xml_free(reply);
}
// Nothing further to do
return I_NULL;
}
static void
verify_feature_set(xmlNode *msg)
{
const char *dc_version = crm_element_value(msg, PCMK_XA_CRM_FEATURE_SET);
if (dc_version == NULL) {
/* All we really know is that the DC feature set is older than 3.1.0,
* but that's also all that really matters.
*/
dc_version = "3.0.14";
}
if (feature_set_compatible(dc_version, CRM_FEATURE_SET)) {
crm_trace("Local feature set (%s) is compatible with DC's (%s)",
CRM_FEATURE_SET, dc_version);
} else {
crm_err("Local feature set (%s) is incompatible with DC's (%s)",
CRM_FEATURE_SET, dc_version);
// Nothing is likely to improve without administrator involvement
controld_set_fsa_input_flags(R_STAYDOWN);
crmd_exit(CRM_EX_FATAL);
}
}
// DC gets own shutdown all-clear
static enum crmd_fsa_input
handle_shutdown_self_ack(xmlNode *stored_msg)
{
const char *host_from = crm_element_value(stored_msg, PCMK__XA_SRC);
if (pcmk_is_set(controld_globals.fsa_input_register, R_SHUTDOWN)) {
// The expected case -- we initiated own shutdown sequence
crm_info("Shutting down controller");
return I_STOP;
}
if (pcmk__str_eq(host_from, controld_globals.dc_name, pcmk__str_casei)) {
// Must be logic error -- DC confirming its own unrequested shutdown
crm_err("Shutting down controller immediately due to "
"unexpected shutdown confirmation");
return I_TERMINATE;
}
if (controld_globals.fsa_state != S_STOPPING) {
// Shouldn't happen -- non-DC confirming unrequested shutdown
crm_err("Starting new DC election because %s is "
"confirming shutdown we did not request",
(host_from? host_from : "another node"));
return I_ELECTION;
}
// Shouldn't happen, but we are already stopping anyway
crm_debug("Ignoring unexpected shutdown confirmation from %s",
(host_from? host_from : "another node"));
return I_NULL;
}
// Non-DC gets shutdown all-clear from DC
static enum crmd_fsa_input
handle_shutdown_ack(xmlNode *stored_msg)
{
const char *host_from = crm_element_value(stored_msg, PCMK__XA_SRC);
if (host_from == NULL) {
crm_warn("Ignoring shutdown request without origin specified");
return I_NULL;
}
if (pcmk__str_eq(host_from, controld_globals.dc_name,
pcmk__str_null_matches|pcmk__str_casei)) {
if (pcmk_is_set(controld_globals.fsa_input_register, R_SHUTDOWN)) {
crm_info("Shutting down controller after confirmation from %s",
host_from);
} else {
crm_err("Shutting down controller after unexpected "
"shutdown request from %s", host_from);
controld_set_fsa_input_flags(R_STAYDOWN);
}
return I_STOP;
}
crm_warn("Ignoring shutdown request from %s because DC is %s",
host_from, controld_globals.dc_name);
return I_NULL;
}
static enum crmd_fsa_input
handle_request(xmlNode *stored_msg, enum crmd_fsa_cause cause)
{
xmlNode *msg = NULL;
const char *op = crm_element_value(stored_msg, PCMK__XA_CRM_TASK);
/* Optimize this for the DC - it has the most to do */
crm_log_xml_trace(stored_msg, "request");
if (op == NULL) {
crm_warn("Ignoring request without " PCMK__XA_CRM_TASK);
return I_NULL;
}
if (strcmp(op, CRM_OP_SHUTDOWN_REQ) == 0) {
const char *from = crm_element_value(stored_msg, PCMK__XA_SRC);
pcmk__node_status_t *node =
pcmk__search_node_caches(0, from, pcmk__node_search_cluster_member);
pcmk__update_peer_expected(__func__, node, CRMD_JOINSTATE_DOWN);
if(AM_I_DC == FALSE) {
return I_NULL; /* Done */
}
}
/*========== DC-Only Actions ==========*/
if (AM_I_DC) {
if (strcmp(op, CRM_OP_JOIN_ANNOUNCE) == 0) {
return I_NODE_JOIN;
} else if (strcmp(op, CRM_OP_JOIN_REQUEST) == 0) {
return I_JOIN_REQUEST;
} else if (strcmp(op, CRM_OP_JOIN_CONFIRM) == 0) {
return I_JOIN_RESULT;
} else if (strcmp(op, CRM_OP_SHUTDOWN) == 0) {
return handle_shutdown_self_ack(stored_msg);
} else if (strcmp(op, CRM_OP_SHUTDOWN_REQ) == 0) {
// Another controller wants to shut down its node
return handle_shutdown_request(stored_msg);
}
}
/*========== common actions ==========*/
if (strcmp(op, CRM_OP_NOVOTE) == 0) {
ha_msg_input_t fsa_input;
fsa_input.msg = stored_msg;
register_fsa_input_adv(C_HA_MESSAGE, I_NULL, &fsa_input,
A_ELECTION_COUNT | A_ELECTION_CHECK, FALSE,
__func__);
} else if (strcmp(op, CRM_OP_REMOTE_STATE) == 0) {
/* a remote connection host is letting us know the node state */
return handle_remote_state(stored_msg);
} else if (strcmp(op, CRM_OP_THROTTLE) == 0) {
throttle_update(stored_msg);
if (AM_I_DC && (controld_globals.transition_graph != NULL)
&& !controld_globals.transition_graph->complete) {
crm_debug("The throttle changed. Trigger a graph.");
trigger_graph();
}
return I_NULL;
} else if (strcmp(op, CRM_OP_CLEAR_FAILCOUNT) == 0) {
return handle_failcount_op(stored_msg);
} else if (strcmp(op, CRM_OP_VOTE) == 0) {
/* count the vote and decide what to do after that */
ha_msg_input_t fsa_input;
fsa_input.msg = stored_msg;
register_fsa_input_adv(C_HA_MESSAGE, I_NULL, &fsa_input,
A_ELECTION_COUNT | A_ELECTION_CHECK, FALSE,
__func__);
/* Sometimes we _must_ go into S_ELECTION */
if (controld_globals.fsa_state == S_HALT) {
crm_debug("Forcing an election from S_HALT");
return I_ELECTION;
}
} else if (strcmp(op, CRM_OP_JOIN_OFFER) == 0) {
verify_feature_set(stored_msg);
crm_debug("Raising I_JOIN_OFFER: join-%s",
crm_element_value(stored_msg, PCMK__XA_JOIN_ID));
return I_JOIN_OFFER;
} else if (strcmp(op, CRM_OP_JOIN_ACKNAK) == 0) {
crm_debug("Raising I_JOIN_RESULT: join-%s",
crm_element_value(stored_msg, PCMK__XA_JOIN_ID));
return I_JOIN_RESULT;
} else if (strcmp(op, CRM_OP_LRM_DELETE) == 0) {
return handle_lrm_delete(stored_msg);
} else if ((strcmp(op, CRM_OP_LRM_FAIL) == 0)
|| (strcmp(op, CRM_OP_LRM_REFRESH) == 0) // @COMPAT
|| (strcmp(op, CRM_OP_REPROBE) == 0)) {
crm_xml_add(stored_msg, PCMK__XA_CRM_SYS_TO, CRM_SYSTEM_LRMD);
return I_ROUTER;
} else if (strcmp(op, CRM_OP_NOOP) == 0) {
return I_NULL;
} else if (strcmp(op, CRM_OP_PING) == 0) {
return handle_ping(stored_msg);
} else if (strcmp(op, CRM_OP_NODE_INFO) == 0) {
return handle_node_info_request(stored_msg);
} else if (strcmp(op, CRM_OP_RM_NODE_CACHE) == 0) {
int id = 0;
const char *name = NULL;
crm_element_value_int(stored_msg, PCMK_XA_ID, &id);
name = crm_element_value(stored_msg, PCMK_XA_UNAME);
if(cause == C_IPC_MESSAGE) {
msg = create_request(CRM_OP_RM_NODE_CACHE, NULL, NULL, CRM_SYSTEM_CRMD, CRM_SYSTEM_CRMD, NULL);
if (!pcmk__cluster_send_message(NULL, crm_msg_crmd, msg)) {
crm_err("Could not instruct peers to remove references to node %s/%u", name, id);
} else {
crm_notice("Instructing peers to remove references to node %s/%u", name, id);
}
pcmk__xml_free(msg);
} else {
pcmk__cluster_forget_cluster_node(id, name);
/* If we're forgetting this node, also forget any failures to fence
* it, so we don't carry that over to any node added later with the
* same name.
*/
st_fail_count_reset(name);
}
} else if (strcmp(op, CRM_OP_MAINTENANCE_NODES) == 0) {
xmlNode *wrapper = pcmk__xe_first_child(stored_msg, PCMK__XE_CRM_XML,
NULL, NULL);
xmlNode *xml = pcmk__xe_first_child(wrapper, NULL, NULL, NULL);
remote_ra_process_maintenance_nodes(xml);
} else if (strcmp(op, PCMK__CONTROLD_CMD_NODES) == 0) {
return handle_node_list(stored_msg);
/*========== (NOT_DC)-Only Actions ==========*/
} else if (!AM_I_DC) {
if (strcmp(op, CRM_OP_SHUTDOWN) == 0) {
return handle_shutdown_ack(stored_msg);
}
} else {
crm_err("Unexpected request (%s) sent to %s", op, AM_I_DC ? "the DC" : "non-DC node");
crm_log_xml_err(stored_msg, "Unexpected");
}
return I_NULL;
}
static void
handle_response(xmlNode *stored_msg)
{
const char *op = crm_element_value(stored_msg, PCMK__XA_CRM_TASK);
crm_log_xml_trace(stored_msg, "reply");
if (op == NULL) {
crm_warn("Ignoring reply without " PCMK__XA_CRM_TASK);
} else if (AM_I_DC && strcmp(op, CRM_OP_PECALC) == 0) {
// Check whether scheduler answer been superseded by subsequent request
const char *msg_ref = crm_element_value(stored_msg, PCMK_XA_REFERENCE);
if (msg_ref == NULL) {
crm_err("%s - Ignoring calculation with no reference", op);
} else if (pcmk__str_eq(msg_ref, controld_globals.fsa_pe_ref,
pcmk__str_none)) {
ha_msg_input_t fsa_input;
controld_stop_sched_timer();
fsa_input.msg = stored_msg;
register_fsa_input_later(C_IPC_MESSAGE, I_PE_SUCCESS, &fsa_input);
} else {
crm_info("%s calculation %s is obsolete", op, msg_ref);
}
} else if (strcmp(op, CRM_OP_VOTE) == 0
|| strcmp(op, CRM_OP_SHUTDOWN_REQ) == 0 || strcmp(op, CRM_OP_SHUTDOWN) == 0) {
} else {
const char *host_from = crm_element_value(stored_msg, PCMK__XA_SRC);
crm_err("Unexpected response (op=%s, src=%s) sent to the %s",
op, host_from, AM_I_DC ? "DC" : "controller");
}
}
static enum crmd_fsa_input
handle_shutdown_request(xmlNode * stored_msg)
{
/* handle here to avoid potential version issues
* where the shutdown message/procedure may have
* been changed in later versions.
*
* This way the DC is always in control of the shutdown
*/
char *now_s = NULL;
const char *host_from = crm_element_value(stored_msg, PCMK__XA_SRC);
if (host_from == NULL) {
/* we're shutting down and the DC */
host_from = controld_globals.our_nodename;
}
crm_info("Creating shutdown request for %s (state=%s)", host_from,
fsa_state2string(controld_globals.fsa_state));
crm_log_xml_trace(stored_msg, "message");
now_s = pcmk__ttoa(time(NULL));
update_attrd(host_from, PCMK__NODE_ATTR_SHUTDOWN, now_s, NULL, FALSE);
free(now_s);
/* will be picked up by the TE as long as its running */
return I_NULL;
}
static void
send_msg_via_ipc(xmlNode * msg, const char *sys)
{
pcmk__client_t *client_channel = NULL;
CRM_CHECK(sys != NULL, return);
client_channel = pcmk__find_client_by_id(sys);
if (crm_element_value(msg, PCMK__XA_SRC) == NULL) {
crm_xml_add(msg, PCMK__XA_SRC, controld_globals.our_nodename);
}
if (client_channel != NULL) {
/* Transient clients such as crmadmin */
pcmk__ipc_send_xml(client_channel, 0, msg, crm_ipc_server_event);
} else if (pcmk__str_eq(sys, CRM_SYSTEM_TENGINE, pcmk__str_none)) {
xmlNode *wrapper = pcmk__xe_first_child(msg, PCMK__XE_CRM_XML, NULL,
NULL);
xmlNode *data = pcmk__xe_first_child(wrapper, NULL, NULL, NULL);
process_te_message(msg, data);
} else if (pcmk__str_eq(sys, CRM_SYSTEM_LRMD, pcmk__str_none)) {
fsa_data_t fsa_data;
ha_msg_input_t fsa_input;
xmlNode *wrapper = NULL;
fsa_input.msg = msg;
wrapper = pcmk__xe_first_child(msg, PCMK__XE_CRM_XML, NULL, NULL);
fsa_input.xml = pcmk__xe_first_child(wrapper, NULL, NULL, NULL);
fsa_data.id = 0;
fsa_data.actions = 0;
fsa_data.data = &fsa_input;
fsa_data.fsa_input = I_MESSAGE;
fsa_data.fsa_cause = C_IPC_MESSAGE;
fsa_data.origin = __func__;
fsa_data.data_type = fsa_dt_ha_msg;
do_lrm_invoke(A_LRM_INVOKE, C_IPC_MESSAGE, controld_globals.fsa_state,
I_MESSAGE, &fsa_data);
} else if (crmd_is_proxy_session(sys)) {
crmd_proxy_send(sys, msg);
} else {
crm_info("Received invalid request: unknown subsystem '%s'", sys);
}
}
void
delete_ha_msg_input(ha_msg_input_t * orig)
{
if (orig == NULL) {
return;
}
pcmk__xml_free(orig->msg);
free(orig);
}
/*!
* \internal
* \brief Notify the cluster of a remote node state change
*
* \param[in] node_name Node's name
* \param[in] node_up true if node is up, false if down
*/
void
broadcast_remote_state_message(const char *node_name, bool node_up)
{
xmlNode *msg = create_request(CRM_OP_REMOTE_STATE, NULL, NULL,
CRM_SYSTEM_CRMD, CRM_SYSTEM_CRMD, NULL);
crm_info("Notifying cluster of Pacemaker Remote node %s %s",
node_name, node_up? "coming up" : "going down");
crm_xml_add(msg, PCMK_XA_ID, node_name);
pcmk__xe_set_bool_attr(msg, PCMK__XA_IN_CCM, node_up);
if (node_up) {
crm_xml_add(msg, PCMK__XA_CONNECTION_HOST,
controld_globals.our_nodename);
}
pcmk__cluster_send_message(NULL, crm_msg_crmd, msg);
pcmk__xml_free(msg);
}
diff --git a/daemons/controld/controld_remote_ra.c b/daemons/controld/controld_remote_ra.c
index 04fa3b4337..9a483a06e4 100644
--- a/daemons/controld/controld_remote_ra.c
+++ b/daemons/controld/controld_remote_ra.c
@@ -1,1474 +1,1474 @@
/*
* Copyright 2013-2024 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 <crm_internal.h>
#include <crm/crm.h>
#include <crm/common/xml.h>
#include <crm/common/xml_internal.h>
#include <crm/lrmd.h>
#include <crm/lrmd_internal.h>
#include <crm/services.h>
#include <pacemaker-controld.h>
#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;
int remaining_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 = lrm_state_find(cmd->rsc_id);
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 = lrm_state_find(cmd->rsc_id);
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)
{
bool purge = true;
pcmk__node_status_t *conn_node = NULL;
lrm_state_t *connection_rsc = NULL;
if (!node->conn_host) {
return purge;
}
/* 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 purge;
}
/* 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 = lrm_state_find(node->name);
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) {
purge = false;
}
}
return purge;
}
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);
/* Purge node from attrd's memory */
if (purge) {
update_attrd_remote_node_removed(node->name, NULL);
}
controld_delete_node_state(node->name, section, 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, return);
purge_remote_node_attrs(call_opt, node);
pcmk__update_peer_state(__func__, node, CRM_NODE_MEMBER, 0);
/* Apply any start state that we were given from the environment on the
* remote node.
*/
connection_rsc = lrm_state_find(node->name);
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->uuid, true);
+ 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, 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);
/* 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.
*/
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);
}
/* 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, CRM_NODE_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, 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, CRM_NODE_MEMBER, 0);
} else if (pcmk__str_eq(cmd->action, PCMK_ACTION_STOP, pcmk__str_casei)) {
lrm_state_t *lrm_state = lrm_state_find(cmd->rsc_id);
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);
}
static void
update_remaining_timeout(remote_ra_cmd_t * cmd)
{
cmd->remaining_timeout = ((cmd->timeout / 1000) - (time(NULL) - cmd->start_time)) * 1000;
}
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;
if (!ra_data || !ra_data->cur_cmd) {
return FALSE;
}
cmd = ra_data->cur_cmd;
if (!pcmk__strcase_any_of(cmd->action, PCMK_ACTION_START,
PCMK_ACTION_MIGRATE_FROM, NULL)) {
return FALSE;
}
update_remaining_timeout(cmd);
if (cmd->remaining_timeout > 0) {
rc = handle_remote_ra_start(lrm_state, cmd, cmd->remaining_timeout);
} 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 = lrm_state_find(cmd->rsc_id);
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 = lrm_state_find(cmd->rsc_id);
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) {
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 = lrm_state_find(controld_globals.our_nodename);
}
CRM_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_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,
services_ocf_exitcode_str(op->rc), op->rc,
pcmk_exec_status_str(op->op_status), op->op_status);
lrm_state = lrm_state_find(op->remote_nodename);
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__strcase_any_of(cmd->action, PCMK_ACTION_START,
PCMK_ACTION_MIGRATE_FROM, NULL)) {
if (op->connection_rc < 0) {
update_remaining_timeout(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 (cmd->remaining_timeout > 3000) {
crm_trace("rescheduling start, remaining timeout %d", cmd->remaining_timeout);
g_timeout_add(1000, retry_start_cmd_cb, lrm_state);
return;
} else {
crm_trace("can't reschedule start, remaining timeout too small %d",
cmd->remaining_timeout);
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 = g_timeout_add(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 if ((op->type == lrmd_event_new_client)
&& pcmk__str_eq(cmd->action, PCMK_ACTION_STOP,
pcmk__str_casei)) {
handle_remote_ra_stop(lrm_state, cmd);
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;
CRM_ASSERT(lrm_state);
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 = g_timeout_add(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 the takeover event to occur. If the
* takeover event does not occur during the wait period, that's fine.
* It just means that the remote-node's lrm_status section is going to get
* cleared which will require all the resources running in the remote-node
* to be explicitly re-detected via probe actions. If the takeover does occur
* successfully, then we can leave the status section intact. */
cmd->takeover_timeout_id = g_timeout_add((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;
}
if ((id != NULL) && (lrm_state_find(id) != NULL)
&& !pcmk__str_eq(id, controld_globals.our_nodename, pcmk__str_casei)) {
return TRUE;
}
return FALSE;
}
lrmd_rsc_info_t *
remote_ra_get_rsc_info(lrm_state_t * lrm_state, const char *rsc_id)
{
lrmd_rsc_info_t *info = NULL;
if ((lrm_state_find(rsc_id))) {
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;
connection_rsc = lrm_state_find(rsc_id);
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 = lrm_state_find(rsc_id);
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 = g_timeout_add(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 = lrm_state_find(node_name);
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:
*
* <pseudo_event id="103" operation="stonith" operation_key="stonith-lxc1-off"
* on_node="lxc1" on_node_uuid="lxc1">
* <attributes CRM_meta_on_node="lxc1" CRM_meta_on_node_uuid="lxc1"
* CRM_meta_stonith_action="off" crm_feature_set="3.0.12"/>
* <downed>
* <node id="lxc1"/>
* </downed>
* </pseudo_event>
*/
#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)
{
xmlXPathObjectPtr search = xpath_search(xml, XPATH_PSEUDO_FENCE);
if (numXpathResults(search) == 1) {
xmlNode *result = getXpathResult(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);
}
}
}
freeXpathObject(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, 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)
{
xmlXPathObjectPtr search = xpath_search(xml, XPATH_PSEUDO_MAINTENANCE);
if (numXpathResults(search) == 1) {
xmlNode *node;
int cnt = 0, cnt_remote = 0;
for (node = pcmk__xe_first_child(getXpathResult(search, 0),
PCMK_XE_NODE, NULL, NULL);
node != NULL; node = pcmk__xe_next_same(node)) {
lrm_state_t *lrm_state = lrm_state_find(pcmk__xe_id(node));
cnt++;
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);
}
freeXpathObject(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_te_events.c b/daemons/controld/controld_te_events.c
index 1ce0f82609..d2b36c1785 100644
--- a/daemons/controld/controld_te_events.c
+++ b/daemons/controld/controld_te_events.c
@@ -1,613 +1,613 @@
/*
* Copyright 2004-2024 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 <crm_internal.h>
#include <sys/param.h>
#include <crm/crm.h>
#include <crm/cib.h>
#include <crm/common/xml.h>
#include <pacemaker-controld.h>
#include <crm/common/attrs_internal.h>
#include <crm/common/ipc_attrd_internal.h>
/*!
* \internal
* \brief Action numbers of outside events processed in current update diff
*
* This table is to be used as a set. It should be empty when the transitioner
* begins processing a CIB update diff. It ensures that if there are multiple
* events (for example, "_last_0" and "_last_failure_0") for the same action,
* only one of them updates the failcount. Events that originate outside the
* cluster can't be confirmed, since they're not in the transition graph.
*/
static GHashTable *outside_events = NULL;
/*!
* \internal
* \brief Empty the hash table containing action numbers of outside events
*/
void
controld_remove_all_outside_events(void)
{
if (outside_events != NULL) {
g_hash_table_remove_all(outside_events);
}
}
/*!
* \internal
* \brief Destroy the hash table containing action numbers of outside events
*/
void
controld_destroy_outside_events_table(void)
{
if (outside_events != NULL) {
g_hash_table_destroy(outside_events);
outside_events = NULL;
}
}
/*!
* \internal
* \brief Add an outside event's action number to a set
*
* \return Standard Pacemaker return code. Specifically, \p pcmk_rc_ok if the
* event was not already in the set, or \p pcmk_rc_already otherwise.
*/
static int
record_outside_event(gint action_num)
{
if (outside_events == NULL) {
outside_events = g_hash_table_new(NULL, NULL);
}
if (g_hash_table_add(outside_events, GINT_TO_POINTER(action_num))) {
return pcmk_rc_ok;
}
return pcmk_rc_already;
}
gboolean
fail_incompletable_actions(pcmk__graph_t *graph, const char *down_node)
{
const char *target_uuid = NULL;
const char *router = NULL;
const char *router_uuid = NULL;
xmlNode *last_action = NULL;
GList *gIter = NULL;
GList *gIter2 = NULL;
if (graph == NULL || graph->complete) {
return FALSE;
}
gIter = graph->synapses;
for (; gIter != NULL; gIter = gIter->next) {
pcmk__graph_synapse_t *synapse = (pcmk__graph_synapse_t *) gIter->data;
if (pcmk_any_flags_set(synapse->flags, pcmk__synapse_confirmed|pcmk__synapse_failed)) {
/* We've already been here */
continue;
}
gIter2 = synapse->actions;
for (; gIter2 != NULL; gIter2 = gIter2->next) {
pcmk__graph_action_t *action = (pcmk__graph_action_t *) gIter2->data;
if ((action->type == pcmk__pseudo_graph_action)
|| pcmk_is_set(action->flags, pcmk__graph_action_confirmed)) {
continue;
} else if (action->type == pcmk__cluster_graph_action) {
const char *task = crm_element_value(action->xml,
PCMK_XA_OPERATION);
if (pcmk__str_eq(task, PCMK_ACTION_STONITH, pcmk__str_casei)) {
continue;
}
}
target_uuid = crm_element_value(action->xml,
PCMK__META_ON_NODE_UUID);
router = crm_element_value(action->xml, PCMK__XA_ROUTER_NODE);
if (router) {
const pcmk__node_status_t *node =
pcmk__get_node(0, router, NULL,
pcmk__node_search_cluster_member);
- if (node) {
- router_uuid = node->uuid;
+ if (node != NULL) {
+ router_uuid = node->xml_id;
}
}
if (pcmk__str_eq(target_uuid, down_node, pcmk__str_casei) || pcmk__str_eq(router_uuid, down_node, pcmk__str_casei)) {
pcmk__set_graph_action_flags(action, pcmk__graph_action_failed);
pcmk__set_synapse_flags(synapse, pcmk__synapse_failed);
last_action = action->xml;
stop_te_timer(action);
pcmk__update_graph(graph, action);
if (pcmk_is_set(synapse->flags, pcmk__synapse_executed)) {
crm_notice("Action %d (%s) was pending on %s (offline)",
action->id,
crm_element_value(action->xml,
PCMK__XA_OPERATION_KEY),
down_node);
} else {
crm_info("Action %d (%s) is scheduled for %s (offline)",
action->id,
crm_element_value(action->xml, PCMK__XA_OPERATION_KEY),
down_node);
}
}
}
}
if (last_action != NULL) {
crm_info("Node %s shutdown resulted in un-runnable actions", down_node);
abort_transition(PCMK_SCORE_INFINITY, pcmk__graph_restart,
"Node failure", last_action);
return TRUE;
}
return FALSE;
}
/*!
* \internal
* \brief Update failure-related node attributes if warranted
*
* \param[in] event XML describing operation that (maybe) failed
* \param[in] event_node_uuid Node that event occurred on
* \param[in] rc Actual operation return code
* \param[in] target_rc Expected operation return code
* \param[in] do_update If TRUE, do update regardless of operation type
* \param[in] ignore_failures If TRUE, update last failure but not fail count
*
* \return TRUE if this was not a direct nack, success or lrm status refresh
*/
static gboolean
update_failcount(const xmlNode *event, const char *event_node_uuid, int rc,
int target_rc, gboolean do_update, gboolean ignore_failures)
{
guint interval_ms = 0;
char *task = NULL;
char *rsc_id = NULL;
const char *value = NULL;
const char *id = crm_element_value(event, PCMK__XA_OPERATION_KEY);
const char *on_uname = pcmk__node_name_from_uuid(event_node_uuid);
const char *origin = crm_element_value(event, PCMK_XA_CRM_DEBUG_ORIGIN);
// Nothing needs to be done for success or status refresh
if (rc == target_rc) {
return FALSE;
} else if (pcmk__str_eq(origin, "build_active_RAs", pcmk__str_casei)) {
crm_debug("No update for %s (rc=%d) on %s: Old failure from lrm status refresh",
id, rc, on_uname);
return FALSE;
}
/* Sanity check */
CRM_CHECK(on_uname != NULL, return TRUE);
CRM_CHECK(parse_op_key(id, &rsc_id, &task, &interval_ms),
crm_err("Couldn't parse: %s", pcmk__xe_id(event)); goto bail);
/* Decide whether update is necessary and what value to use */
if ((interval_ms > 0)
|| pcmk__str_eq(task, PCMK_ACTION_PROMOTE, pcmk__str_none)
|| pcmk__str_eq(task, PCMK_ACTION_DEMOTE, pcmk__str_none)) {
do_update = TRUE;
} else if (pcmk__str_eq(task, PCMK_ACTION_START, pcmk__str_none)) {
do_update = TRUE;
value = pcmk__s(controld_globals.transition_graph->failed_start_offset,
PCMK_VALUE_INFINITY);
} else if (pcmk__str_eq(task, PCMK_ACTION_STOP, pcmk__str_none)) {
do_update = TRUE;
value = pcmk__s(controld_globals.transition_graph->failed_stop_offset,
PCMK_VALUE_INFINITY);
}
if (do_update) {
pcmk__attrd_query_pair_t *fail_pair = NULL;
pcmk__attrd_query_pair_t *last_pair = NULL;
char *fail_name = NULL;
char *last_name = NULL;
GList *attrs = NULL;
uint32_t opts = pcmk__node_attr_none;
char *now = pcmk__ttoa(time(NULL));
// Fail count will be either incremented or set to infinity
if (!pcmk_str_is_infinity(value)) {
value = PCMK_XA_VALUE "++";
}
if (g_hash_table_lookup(crm_remote_peer_cache, event_node_uuid)) {
opts |= pcmk__node_attr_remote;
}
crm_info("Updating %s for %s on %s after failed %s: rc=%d (update=%s, time=%s)",
(ignore_failures? "last failure" : "failcount"),
rsc_id, on_uname, task, rc, value, now);
/* Update the fail count, if we're not ignoring failures */
if (!ignore_failures) {
fail_pair = pcmk__assert_alloc(1, sizeof(pcmk__attrd_query_pair_t));
fail_name = pcmk__failcount_name(rsc_id, task, interval_ms);
fail_pair->name = fail_name;
fail_pair->value = value;
fail_pair->node = on_uname;
attrs = g_list_prepend(attrs, fail_pair);
}
/* Update the last failure time (even if we're ignoring failures,
* so that failure can still be detected and shown, e.g. by crm_mon)
*/
last_pair = pcmk__assert_alloc(1, sizeof(pcmk__attrd_query_pair_t));
last_name = pcmk__lastfailure_name(rsc_id, task, interval_ms);
last_pair->name = last_name;
last_pair->value = now;
last_pair->node = on_uname;
attrs = g_list_prepend(attrs, last_pair);
update_attrd_list(attrs, opts);
free(fail_name);
free(fail_pair);
free(last_name);
free(last_pair);
g_list_free(attrs);
free(now);
}
bail:
free(rsc_id);
free(task);
return TRUE;
}
pcmk__graph_action_t *
controld_get_action(int id)
{
for (GList *item = controld_globals.transition_graph->synapses;
item != NULL; item = item->next) {
pcmk__graph_synapse_t *synapse = (pcmk__graph_synapse_t *) item->data;
for (GList *item2 = synapse->actions; item2; item2 = item2->next) {
pcmk__graph_action_t *action = (pcmk__graph_action_t *) item2->data;
if (action->id == id) {
return action;
}
}
}
return NULL;
}
pcmk__graph_action_t *
get_cancel_action(const char *id, const char *node)
{
GList *gIter = NULL;
GList *gIter2 = NULL;
gIter = controld_globals.transition_graph->synapses;
for (; gIter != NULL; gIter = gIter->next) {
pcmk__graph_synapse_t *synapse = (pcmk__graph_synapse_t *) gIter->data;
gIter2 = synapse->actions;
for (; gIter2 != NULL; gIter2 = gIter2->next) {
const char *task = NULL;
const char *target = NULL;
pcmk__graph_action_t *action = (pcmk__graph_action_t *) gIter2->data;
task = crm_element_value(action->xml, PCMK_XA_OPERATION);
if (!pcmk__str_eq(PCMK_ACTION_CANCEL, task, pcmk__str_casei)) {
continue;
}
task = crm_element_value(action->xml, PCMK__XA_OPERATION_KEY);
if (!pcmk__str_eq(task, id, pcmk__str_casei)) {
crm_trace("Wrong key %s for %s on %s", task, id, node);
continue;
}
target = crm_element_value(action->xml, PCMK__META_ON_NODE_UUID);
if (node && !pcmk__str_eq(target, node, pcmk__str_casei)) {
crm_trace("Wrong node %s for %s on %s", target, id, node);
continue;
}
crm_trace("Found %s on %s", id, node);
return action;
}
}
return NULL;
}
bool
confirm_cancel_action(const char *id, const char *node_id)
{
const char *op_key = NULL;
const char *node_name = NULL;
pcmk__graph_action_t *cancel = get_cancel_action(id, node_id);
if (cancel == NULL) {
return FALSE;
}
op_key = crm_element_value(cancel->xml, PCMK__XA_OPERATION_KEY);
node_name = crm_element_value(cancel->xml, PCMK__META_ON_NODE);
stop_te_timer(cancel);
te_action_confirmed(cancel, controld_globals.transition_graph);
crm_info("Cancellation of %s on %s confirmed (action %d)",
op_key, node_name, cancel->id);
return TRUE;
}
/* downed nodes are listed like: <downed> <node id="UUID1" /> ... </downed> */
#define XPATH_DOWNED "//" PCMK__XE_DOWNED \
"/" PCMK_XE_NODE "[@" PCMK_XA_ID "='%s']"
/*!
* \brief Find a transition event that would have made a specified node down
*
* \param[in] target UUID of node to match
*
* \return Matching event if found, NULL otherwise
*/
pcmk__graph_action_t *
match_down_event(const char *target)
{
pcmk__graph_action_t *match = NULL;
xmlXPathObjectPtr xpath_ret = NULL;
GList *gIter, *gIter2;
char *xpath = crm_strdup_printf(XPATH_DOWNED, target);
for (gIter = controld_globals.transition_graph->synapses;
gIter != NULL && match == NULL;
gIter = gIter->next) {
for (gIter2 = ((pcmk__graph_synapse_t * ) gIter->data)->actions;
gIter2 != NULL && match == NULL;
gIter2 = gIter2->next) {
match = (pcmk__graph_action_t *) gIter2->data;
if (pcmk_is_set(match->flags, pcmk__graph_action_executed)) {
xpath_ret = xpath_search(match->xml, xpath);
if (numXpathResults(xpath_ret) < 1) {
match = NULL;
}
freeXpathObject(xpath_ret);
} else {
// Only actions that were actually started can match
match = NULL;
}
}
}
free(xpath);
if (match != NULL) {
crm_debug("Shutdown action %d (%s) found for node %s", match->id,
crm_element_value(match->xml, PCMK__XA_OPERATION_KEY),
target);
} else {
crm_debug("No reason to expect node %s to be down", target);
}
return match;
}
void
process_graph_event(xmlNode *event, const char *event_node)
{
int rc = -1; // Actual result
int target_rc = -1; // Expected result
int status = -1; // Executor status
int callid = -1; // Executor call ID
int transition_num = -1; // Transition number
int action_num = -1; // Action number within transition
char *update_te_uuid = NULL;
bool ignore_failures = FALSE;
const char *id = NULL;
const char *desc = NULL;
const char *magic = NULL;
const char *uname = NULL;
CRM_ASSERT(event != NULL);
/*
<lrm_rsc_op id="rsc_east-05_last_0" operation_key="rsc_east-05_monitor_0" operation="monitor" crm-debug-origin="do_update_resource" crm_feature_set="3.0.6" transition-key="9:2:7:be2e97d9-05e2-439d-863e-48f7aecab2aa" transition-magic="0:7;9:2:7:be2e97d9-05e2-439d-863e-48f7aecab2aa" call-id="17" rc-code="7" op-status="0" interval="0" last-rc-change="1355361636" exec-time="128" queue-time="0" op-digest="c81f5f40b1c9e859c992e800b1aa6972"/>
*/
magic = crm_element_value(event, PCMK__XA_TRANSITION_KEY);
if (magic == NULL) {
/* non-change */
return;
}
crm_element_value_int(event, PCMK__XA_OP_STATUS, &status);
if (status == PCMK_EXEC_PENDING) {
return;
}
id = crm_element_value(event, PCMK__XA_OPERATION_KEY);
crm_element_value_int(event, PCMK__XA_RC_CODE, &rc);
crm_element_value_int(event, PCMK__XA_CALL_ID, &callid);
rc = pcmk__effective_rc(rc);
if (decode_transition_key(magic, &update_te_uuid, &transition_num,
&action_num, &target_rc) == FALSE) {
// decode_transition_key() already logged the bad key
crm_err("Can't process action %s result: Incompatible versions? "
QB_XS " call-id=%d", id, callid);
abort_transition(PCMK_SCORE_INFINITY, pcmk__graph_restart,
"Bad event", event);
return;
}
if (transition_num == -1) {
// E.g. crm_resource --fail
if (record_outside_event(action_num) != pcmk_rc_ok) {
crm_debug("Outside event with transition key '%s' has already been "
"processed", magic);
goto bail;
}
desc = "initiated outside of the cluster";
abort_transition(PCMK_SCORE_INFINITY, pcmk__graph_restart,
"Unexpected event", event);
} else if ((action_num < 0)
|| !pcmk__str_eq(update_te_uuid, controld_globals.te_uuid,
pcmk__str_none)) {
desc = "initiated by a different DC";
abort_transition(PCMK_SCORE_INFINITY, pcmk__graph_restart,
"Foreign event", event);
} else if ((controld_globals.transition_graph->id != transition_num)
|| controld_globals.transition_graph->complete) {
// Action is not from currently active transition
guint interval_ms = 0;
if (parse_op_key(id, NULL, NULL, &interval_ms)
&& (interval_ms != 0)) {
/* Recurring actions have the transition number they were first
* scheduled in.
*/
if (status == PCMK_EXEC_CANCELLED) {
confirm_cancel_action(id, get_node_id(event));
goto bail;
}
desc = "arrived after initial scheduling";
abort_transition(PCMK_SCORE_INFINITY, pcmk__graph_restart,
"Change in recurring result", event);
} else if (controld_globals.transition_graph->id != transition_num) {
desc = "arrived really late";
abort_transition(PCMK_SCORE_INFINITY, pcmk__graph_restart,
"Old event", event);
} else {
desc = "arrived late";
abort_transition(PCMK_SCORE_INFINITY, pcmk__graph_restart,
"Inactive graph", event);
}
} else {
// Event is result of an action from currently active transition
pcmk__graph_action_t *action = controld_get_action(action_num);
if (action == NULL) {
// Should never happen
desc = "unknown";
abort_transition(PCMK_SCORE_INFINITY, pcmk__graph_restart,
"Unknown event", event);
} else if (pcmk_is_set(action->flags, pcmk__graph_action_confirmed)) {
/* Nothing further needs to be done if the action has already been
* confirmed. This can happen e.g. when processing both an
* "xxx_last_0" or "xxx_last_failure_0" record as well as the main
* history record, which would otherwise result in incorrectly
* bumping the fail count twice.
*/
crm_log_xml_debug(event, "Event already confirmed:");
goto bail;
} else {
/* An action result needs to be confirmed.
* (This is the only case where desc == NULL.)
*/
if (pcmk__str_eq(crm_meta_value(action->params, PCMK_META_ON_FAIL),
PCMK_VALUE_IGNORE, pcmk__str_casei)) {
ignore_failures = TRUE;
} else if (rc != target_rc) {
pcmk__set_graph_action_flags(action, pcmk__graph_action_failed);
}
stop_te_timer(action);
te_action_confirmed(action, controld_globals.transition_graph);
if (pcmk_is_set(action->flags, pcmk__graph_action_failed)) {
abort_transition(action->synapse->priority + 1,
pcmk__graph_restart, "Event failed", event);
}
}
}
if (id == NULL) {
id = "unknown action";
}
uname = crm_element_value(event, PCMK__META_ON_NODE);
if (uname == NULL) {
uname = "unknown node";
}
if (status == PCMK_EXEC_INVALID) {
// We couldn't attempt the action
crm_info("Transition %d action %d (%s on %s): %s",
transition_num, action_num, id, uname,
pcmk_exec_status_str(status));
} else if (desc && update_failcount(event, event_node, rc, target_rc,
(transition_num == -1), FALSE)) {
crm_notice("Transition %d action %d (%s on %s): expected '%s' but got '%s' "
QB_XS " target-rc=%d rc=%d call-id=%d event='%s'",
transition_num, action_num, id, uname,
services_ocf_exitcode_str(target_rc),
services_ocf_exitcode_str(rc),
target_rc, rc, callid, desc);
} else if (desc) {
crm_info("Transition %d action %d (%s on %s): %s "
QB_XS " rc=%d target-rc=%d call-id=%d",
transition_num, action_num, id, uname,
desc, rc, target_rc, callid);
} else if (rc == target_rc) {
crm_info("Transition %d action %d (%s on %s) confirmed: %s "
QB_XS " rc=%d call-id=%d",
transition_num, action_num, id, uname,
services_ocf_exitcode_str(rc), rc, callid);
} else {
update_failcount(event, event_node, rc, target_rc,
(transition_num == -1), ignore_failures);
crm_notice("Transition %d action %d (%s on %s): expected '%s' but got '%s' "
QB_XS " target-rc=%d rc=%d call-id=%d",
transition_num, action_num, id, uname,
services_ocf_exitcode_str(target_rc),
services_ocf_exitcode_str(rc),
target_rc, rc, callid);
}
bail:
free(update_te_uuid);
}
diff --git a/daemons/controld/controld_te_utils.c b/daemons/controld/controld_te_utils.c
index 451691aec9..2afa50a08b 100644
--- a/daemons/controld/controld_te_utils.c
+++ b/daemons/controld/controld_te_utils.c
@@ -1,507 +1,508 @@
/*
* Copyright 2004-2024 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 <crm_internal.h>
#include <crm/crm.h>
#include <crm/common/xml.h>
#include <pacemaker-controld.h>
//! Triggers transition graph processing
static crm_trigger_t *transition_trigger = NULL;
static GHashTable *node_pending_timers = NULL;
gboolean
stop_te_timer(pcmk__graph_action_t *action)
{
if (action == NULL) {
return FALSE;
}
if (action->timer != 0) {
crm_trace("Stopping action timer");
g_source_remove(action->timer);
action->timer = 0;
} else {
crm_trace("Action timer was already stopped");
return FALSE;
}
return TRUE;
}
static gboolean
te_graph_trigger(gpointer user_data)
{
if (controld_globals.transition_graph == NULL) {
crm_debug("Nothing to do");
return TRUE;
}
crm_trace("Invoking graph %d in state %s",
controld_globals.transition_graph->id,
fsa_state2string(controld_globals.fsa_state));
switch (controld_globals.fsa_state) {
case S_STARTING:
case S_PENDING:
case S_NOT_DC:
case S_HALT:
case S_ILLEGAL:
case S_STOPPING:
case S_TERMINATE:
return TRUE;
default:
break;
}
if (!controld_globals.transition_graph->complete) {
enum pcmk__graph_status graph_rc;
int orig_limit = controld_globals.transition_graph->batch_limit;
int throttled_limit = throttle_get_total_job_limit(orig_limit);
controld_globals.transition_graph->batch_limit = throttled_limit;
graph_rc = pcmk__execute_graph(controld_globals.transition_graph);
controld_globals.transition_graph->batch_limit = orig_limit;
if (graph_rc == pcmk__graph_active) {
crm_trace("Transition not yet complete");
return TRUE;
} else if (graph_rc == pcmk__graph_pending) {
crm_trace("Transition not yet complete - no actions fired");
return TRUE;
}
if (graph_rc != pcmk__graph_complete) {
crm_warn("Transition failed: %s",
pcmk__graph_status2text(graph_rc));
pcmk__log_graph(LOG_NOTICE, controld_globals.transition_graph);
}
}
crm_debug("Transition %d is now complete",
controld_globals.transition_graph->id);
controld_globals.transition_graph->complete = true;
notify_crmd(controld_globals.transition_graph);
return TRUE;
}
/*!
* \internal
* \brief Initialize transition trigger
*/
void
controld_init_transition_trigger(void)
{
transition_trigger = mainloop_add_trigger(G_PRIORITY_LOW, te_graph_trigger,
NULL);
}
/*!
* \internal
* \brief Destroy transition trigger
*/
void
controld_destroy_transition_trigger(void)
{
mainloop_destroy_trigger(transition_trigger);
transition_trigger = NULL;
}
void
controld_trigger_graph_as(const char *fn, int line)
{
crm_trace("%s:%d - Triggered graph processing", fn, line);
mainloop_set_trigger(transition_trigger);
}
static struct abort_timer_s {
bool aborted;
guint id;
int priority;
enum pcmk__graph_next action;
const char *text;
} abort_timer = { 0, };
static gboolean
abort_timer_popped(gpointer data)
{
struct abort_timer_s *abort_timer = (struct abort_timer_s *) data;
if (AM_I_DC && (abort_timer->aborted == FALSE)) {
abort_transition(abort_timer->priority, abort_timer->action,
abort_timer->text, NULL);
}
abort_timer->id = 0;
return FALSE; // do not immediately reschedule timer
}
/*!
* \internal
* \brief Abort transition after delay, if not already aborted in that time
*
* \param[in] abort_text Must be literal string
*/
void
abort_after_delay(int abort_priority, enum pcmk__graph_next abort_action,
const char *abort_text, guint delay_ms)
{
if (abort_timer.id) {
// Timer already in progress, stop and reschedule
g_source_remove(abort_timer.id);
}
abort_timer.aborted = FALSE;
abort_timer.priority = abort_priority;
abort_timer.action = abort_action;
abort_timer.text = abort_text;
abort_timer.id = g_timeout_add(delay_ms, abort_timer_popped, &abort_timer);
}
static void
free_node_pending_timer(gpointer data)
{
struct abort_timer_s *node_pending_timer = (struct abort_timer_s *) data;
if (node_pending_timer->id != 0) {
g_source_remove(node_pending_timer->id);
node_pending_timer->id = 0;
}
free(node_pending_timer);
}
static gboolean
node_pending_timer_popped(gpointer key)
{
struct abort_timer_s *node_pending_timer = NULL;
if (node_pending_timers == NULL) {
return FALSE;
}
node_pending_timer = g_hash_table_lookup(node_pending_timers, key);
if (node_pending_timer == NULL) {
return FALSE;
}
crm_warn("Node with " PCMK_XA_ID " '%s' pending timed out (%us) "
"on joining the process group",
(const char *) key, controld_globals.node_pending_timeout);
if (controld_globals.node_pending_timeout > 0) {
abort_timer_popped(node_pending_timer);
}
g_hash_table_remove(node_pending_timers, key);
return FALSE; // do not reschedule timer
}
static void
init_node_pending_timer(const pcmk__node_status_t *node, guint timeout)
{
struct abort_timer_s *node_pending_timer = NULL;
char *key = NULL;
- if (node->uuid == NULL) {
+ if (node->xml_id == NULL) {
return;
}
if (node_pending_timers == NULL) {
node_pending_timers = pcmk__strikey_table(free,
free_node_pending_timer);
// The timer is somehow already existing
- } else if (g_hash_table_lookup(node_pending_timers, node->uuid) != NULL) {
+ } else if (g_hash_table_lookup(node_pending_timers, node->xml_id) != NULL) {
return;
}
crm_notice("Waiting for pending %s with " PCMK_XA_ID " '%s' "
"to join the process group (timeout=%us)",
- pcmk__s(node->name, "node"), node->uuid,
+ pcmk__s(node->name, "node"), node->xml_id,
controld_globals.node_pending_timeout);
- key = pcmk__str_copy(node->uuid);
+ key = pcmk__str_copy(node->xml_id);
node_pending_timer = pcmk__assert_alloc(1, sizeof(struct abort_timer_s));
node_pending_timer->aborted = FALSE;
node_pending_timer->priority = PCMK_SCORE_INFINITY;
node_pending_timer->action = pcmk__graph_restart;
node_pending_timer->text = "Node pending timed out";
g_hash_table_replace(node_pending_timers, key, node_pending_timer);
node_pending_timer->id = g_timeout_add_seconds(timeout,
node_pending_timer_popped,
key);
CRM_ASSERT(node_pending_timer->id != 0);
}
static void
remove_node_pending_timer(const char *node_uuid)
{
if (node_pending_timers == NULL) {
return;
}
g_hash_table_remove(node_pending_timers, node_uuid);
}
void
controld_node_pending_timer(const pcmk__node_status_t *node)
{
long long remaining_timeout = 0;
/* If the node is not an active cluster node, is leaving the cluster, or is
* already part of CPG, or PCMK_OPT_NODE_PENDING_TIMEOUT is disabled, free
* any node pending timer for it.
*/
if (pcmk_is_set(node->flags, crm_remote_node)
|| (node->when_member <= 1) || (node->when_online > 0)
|| (controld_globals.node_pending_timeout == 0)) {
- remove_node_pending_timer(node->uuid);
+
+ remove_node_pending_timer(node->xml_id);
return;
}
// Node is a cluster member but offline in CPG
remaining_timeout = node->when_member - time(NULL)
+ controld_globals.node_pending_timeout;
/* It already passed node pending timeout somehow.
* Free any node pending timer of it.
*/
if (remaining_timeout <= 0) {
- remove_node_pending_timer(node->uuid);
+ remove_node_pending_timer(node->xml_id);
return;
}
init_node_pending_timer(node, remaining_timeout);
}
void
controld_free_node_pending_timers(void)
{
if (node_pending_timers == NULL) {
return;
}
g_hash_table_destroy(node_pending_timers);
node_pending_timers = NULL;
}
static const char *
abort2text(enum pcmk__graph_next abort_action)
{
switch (abort_action) {
case pcmk__graph_done: return "done";
case pcmk__graph_wait: return "stop";
case pcmk__graph_restart: return "restart";
case pcmk__graph_shutdown: return "shutdown";
}
return "unknown";
}
static bool
update_abort_priority(pcmk__graph_t *graph, int priority,
enum pcmk__graph_next action, const char *abort_reason)
{
bool change = FALSE;
if (graph == NULL) {
return change;
}
if (graph->abort_priority < priority) {
crm_debug("Abort priority upgraded from %d to %d", graph->abort_priority, priority);
graph->abort_priority = priority;
if (graph->abort_reason != NULL) {
crm_debug("'%s' abort superseded by %s", graph->abort_reason, abort_reason);
}
graph->abort_reason = abort_reason;
change = TRUE;
}
if (graph->completion_action < action) {
crm_debug("Abort action %s superseded by %s: %s",
abort2text(graph->completion_action), abort2text(action), abort_reason);
graph->completion_action = action;
change = TRUE;
}
return change;
}
void
abort_transition_graph(int abort_priority, enum pcmk__graph_next abort_action,
const char *abort_text, const xmlNode *reason,
const char *fn, int line)
{
int add[] = { 0, 0, 0 };
int del[] = { 0, 0, 0 };
int level = LOG_INFO;
const xmlNode *diff = NULL;
const xmlNode *change = NULL;
CRM_CHECK(controld_globals.transition_graph != NULL, return);
switch (controld_globals.fsa_state) {
case S_STARTING:
case S_PENDING:
case S_NOT_DC:
case S_HALT:
case S_ILLEGAL:
case S_STOPPING:
case S_TERMINATE:
crm_info("Abort %s suppressed: state=%s (%scomplete)",
abort_text, fsa_state2string(controld_globals.fsa_state),
(controld_globals.transition_graph->complete? "" : "in"));
return;
default:
break;
}
abort_timer.aborted = TRUE;
controld_expect_sched_reply(NULL);
if (!controld_globals.transition_graph->complete
&& update_abort_priority(controld_globals.transition_graph,
abort_priority, abort_action,
abort_text)) {
level = LOG_NOTICE;
}
if (reason != NULL) {
const xmlNode *search = NULL;
for(search = reason; search; search = search->parent) {
if (pcmk__xe_is(search, PCMK_XE_DIFF)) {
diff = search;
break;
}
}
if(diff) {
xml_patch_versions(diff, add, del);
for(search = reason; search; search = search->parent) {
if (pcmk__xe_is(search, PCMK_XE_CHANGE)) {
change = search;
break;
}
}
}
}
if (reason == NULL) {
do_crm_log(level,
"Transition %d aborted: %s " QB_XS " source=%s:%d "
"complete=%s", controld_globals.transition_graph->id,
abort_text, fn, line,
pcmk__btoa(controld_globals.transition_graph->complete));
} else if(change == NULL) {
GString *local_path = pcmk__element_xpath(reason);
CRM_ASSERT(local_path != NULL);
do_crm_log(level, "Transition %d aborted by %s.%s: %s "
QB_XS " cib=%d.%d.%d source=%s:%d path=%s complete=%s",
controld_globals.transition_graph->id, reason->name,
pcmk__xe_id(reason), abort_text, add[0], add[1], add[2], fn,
line, (const char *) local_path->str,
pcmk__btoa(controld_globals.transition_graph->complete));
g_string_free(local_path, TRUE);
} else {
const char *op = crm_element_value(change, PCMK_XA_OPERATION);
const char *path = crm_element_value(change, PCMK_XA_PATH);
if(change == reason) {
if (strcmp(op, PCMK_VALUE_CREATE) == 0) {
reason = reason->children;
} else if (strcmp(op, PCMK_VALUE_MODIFY) == 0) {
reason = pcmk__xe_first_child(reason, PCMK_XE_CHANGE_RESULT,
NULL, NULL);
if(reason) {
reason = reason->children;
}
}
CRM_CHECK(reason != NULL, goto done);
}
if (strcmp(op, PCMK_VALUE_DELETE) == 0) {
const char *shortpath = strrchr(path, '/');
do_crm_log(level, "Transition %d aborted by deletion of %s: %s "
QB_XS " cib=%d.%d.%d source=%s:%d path=%s complete=%s",
controld_globals.transition_graph->id,
(shortpath? (shortpath + 1) : path), abort_text,
add[0], add[1], add[2], fn, line, path,
pcmk__btoa(controld_globals.transition_graph->complete));
} else if (pcmk__xe_is(reason, PCMK_XE_NVPAIR)) {
do_crm_log(level, "Transition %d aborted by %s doing %s %s=%s: %s "
QB_XS " cib=%d.%d.%d source=%s:%d path=%s complete=%s",
controld_globals.transition_graph->id,
crm_element_value(reason, PCMK_XA_ID), op,
crm_element_value(reason, PCMK_XA_NAME),
crm_element_value(reason, PCMK_XA_VALUE),
abort_text, add[0], add[1], add[2], fn, line, path,
pcmk__btoa(controld_globals.transition_graph->complete));
} else if (pcmk__xe_is(reason, PCMK__XE_LRM_RSC_OP)) {
const char *magic = crm_element_value(reason,
PCMK__XA_TRANSITION_MAGIC);
do_crm_log(level, "Transition %d aborted by operation %s '%s' on %s: %s "
QB_XS " magic=%s cib=%d.%d.%d source=%s:%d complete=%s",
controld_globals.transition_graph->id,
crm_element_value(reason, PCMK__XA_OPERATION_KEY), op,
crm_element_value(reason, PCMK__META_ON_NODE),
abort_text,
magic, add[0], add[1], add[2], fn, line,
pcmk__btoa(controld_globals.transition_graph->complete));
} else if (pcmk__str_any_of((const char *) reason->name,
PCMK__XE_NODE_STATE, PCMK_XE_NODE, NULL)) {
const char *uname = pcmk__node_name_from_uuid(pcmk__xe_id(reason));
do_crm_log(level, "Transition %d aborted by %s '%s' on %s: %s "
QB_XS " cib=%d.%d.%d source=%s:%d complete=%s",
controld_globals.transition_graph->id,
reason->name, op, pcmk__s(uname, pcmk__xe_id(reason)),
abort_text, add[0], add[1], add[2], fn, line,
pcmk__btoa(controld_globals.transition_graph->complete));
} else {
const char *id = pcmk__xe_id(reason);
do_crm_log(level, "Transition %d aborted by %s.%s '%s': %s "
QB_XS " cib=%d.%d.%d source=%s:%d path=%s complete=%s",
controld_globals.transition_graph->id,
reason->name, pcmk__s(id, ""), pcmk__s(op, "change"),
abort_text, add[0], add[1], add[2], fn, line, path,
pcmk__btoa(controld_globals.transition_graph->complete));
}
}
done:
if (controld_globals.transition_graph->complete) {
if (controld_get_period_transition_timer() > 0) {
controld_stop_transition_timer();
controld_start_transition_timer();
} else {
register_fsa_input(C_FSA_INTERNAL, I_PE_CALC, NULL);
}
return;
}
trigger_graph();
}
diff --git a/include/crm/cluster/internal.h b/include/crm/cluster/internal.h
index 620d5016c1..2d7656ea61 100644
--- a/include/crm/cluster/internal.h
+++ b/include/crm/cluster/internal.h
@@ -1,258 +1,257 @@
/*
* Copyright 2004-2024 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__CRM_CLUSTER_INTERNAL__H
#define PCMK__CRM_CLUSTER_INTERNAL__H
#include <stdbool.h>
#include <stdint.h> // uint32_t, uint64_t
#include <glib.h> // gboolean
#include <crm/cluster.h>
#ifdef __cplusplus
extern "C" {
#endif
enum crm_proc_flag {
/* @COMPAT When pcmk__node_status_t:processes is made internal, we can merge
* this into node flags or turn it into a boolean. Until then, in theory
* something could depend on these particular numeric values.
*/
crm_proc_none = 0x00000001,
// Cluster layers
crm_proc_cpg = 0x04000000,
};
// Used with node cache search functions
enum pcmk__node_search_flags {
//! Does not affect search
pcmk__node_search_none = 0,
//! Search for cluster nodes from membership cache
pcmk__node_search_cluster_member = (1 << 0),
//! Search for remote nodes
pcmk__node_search_remote = (1 << 1),
//! Search for cluster member nodes and remote nodes
pcmk__node_search_any = pcmk__node_search_cluster_member
|pcmk__node_search_remote,
//! Search for cluster nodes from CIB (as of last cache refresh)
pcmk__node_search_cluster_cib = (1 << 2),
};
//! Node status data (may be a cluster node or a Pacemaker Remote node)
typedef struct pcmk__node_status {
//! Node name as known to cluster layer, or Pacemaker Remote node name
char *name;
/* @COMPAT This is less than ideal since the value is not a valid XML ID
* (for Corosync, it's the string equivalent of the node's numeric node ID,
* but XML IDs can't start with a number) and the three elements should have
* different IDs.
*
* Ideally, we would use something like node-NODEID, node_state-NODEID, and
* transient_attributes-NODEID as the element IDs. Unfortunately changing it
* would be impractical due to backward compatibility; older nodes in a
* rolling upgrade will always write and expect the value in the old format.
- *
- * This is also named poorly, since the value is not a UUID, but at least
- * that can be changed at an API compatibility break.
*/
- /*! Value of the PCMK_XA_ID XML attribute to use with the node's
+
+ /*!
+ * Value of the PCMK_XA_ID XML attribute to use with the node's
* PCMK_XE_NODE, PCMK_XE_NODE_STATE, and PCMK_XE_TRANSIENT_ATTRIBUTES
* XML elements in the CIB
*/
- char *uuid;
+ char *xml_id;
char *state; // @TODO change to enum
uint64_t flags; // Bitmask of crm_node_flags
uint64_t last_seen; // Only needed by cluster nodes
uint32_t processes; // @TODO most not needed, merge into flags
/* @TODO When we can break public API compatibility, we can make the rest of
* these members separate structs and use void *cluster_data and
* void *user_data here instead, to abstract the cluster layer further.
*/
// Only used by controller
enum crm_join_phase join;
char *expected;
time_t peer_lost;
char *conn_host;
time_t when_member; // Since when node has been a cluster member
time_t when_online; // Since when peer has been online in CPG
/* @TODO The following are currently needed only by the Corosync stack.
* Eventually consider moving them to a cluster-layer-specific data object.
*/
uint32_t cluster_layer_id; //! Cluster-layer numeric node ID
time_t when_lost; //! When CPG membership was last lost
} pcmk__node_status_t;
/*!
* \internal
* \brief Return the process bit corresponding to the current cluster stack
*
* \return Process flag if detectable, otherwise 0
*/
static inline uint32_t
crm_get_cluster_proc(void)
{
switch (pcmk_get_cluster_layer()) {
case pcmk_cluster_layer_corosync:
return crm_proc_cpg;
default:
break;
}
return crm_proc_none;
}
/*!
* \internal
* \brief Get log-friendly string description of a Corosync return code
*
* \param[in] error Corosync return code
*
* \return Log-friendly string description corresponding to \p error
*/
static inline const char *
pcmk__cs_err_str(int error)
{
# if SUPPORT_COROSYNC
switch (error) {
case CS_OK: return "OK";
case CS_ERR_LIBRARY: return "Library error";
case CS_ERR_VERSION: return "Version error";
case CS_ERR_INIT: return "Initialization error";
case CS_ERR_TIMEOUT: return "Timeout";
case CS_ERR_TRY_AGAIN: return "Try again";
case CS_ERR_INVALID_PARAM: return "Invalid parameter";
case CS_ERR_NO_MEMORY: return "No memory";
case CS_ERR_BAD_HANDLE: return "Bad handle";
case CS_ERR_BUSY: return "Busy";
case CS_ERR_ACCESS: return "Access error";
case CS_ERR_NOT_EXIST: return "Doesn't exist";
case CS_ERR_NAME_TOO_LONG: return "Name too long";
case CS_ERR_EXIST: return "Exists";
case CS_ERR_NO_SPACE: return "No space";
case CS_ERR_INTERRUPT: return "Interrupt";
case CS_ERR_NAME_NOT_FOUND: return "Name not found";
case CS_ERR_NO_RESOURCES: return "No resources";
case CS_ERR_NOT_SUPPORTED: return "Not supported";
case CS_ERR_BAD_OPERATION: return "Bad operation";
case CS_ERR_FAILED_OPERATION: return "Failed operation";
case CS_ERR_MESSAGE_ERROR: return "Message error";
case CS_ERR_QUEUE_FULL: return "Queue full";
case CS_ERR_QUEUE_NOT_AVAILABLE: return "Queue not available";
case CS_ERR_BAD_FLAGS: return "Bad flags";
case CS_ERR_TOO_BIG: return "Too big";
case CS_ERR_NO_SECTIONS: return "No sections";
}
# endif
return "Corosync error";
}
# if SUPPORT_COROSYNC
#if 0
/* This is the new way to do it, but we still support all Corosync 2 versions,
* and this isn't always available. A better alternative here would be to check
* for support in the configure script and enable this conditionally.
*/
#define pcmk__init_cmap(handle) cmap_initialize_map((handle), CMAP_MAP_ICMAP)
#else
#define pcmk__init_cmap(handle) cmap_initialize(handle)
#endif
char *pcmk__corosync_cluster_name(void);
bool pcmk__corosync_add_nodes(xmlNode *xml_parent);
void pcmk__cpg_confchg_cb(cpg_handle_t handle,
const struct cpg_name *group_name,
const struct cpg_address *member_list,
size_t member_list_entries,
const struct cpg_address *left_list,
size_t left_list_entries,
const struct cpg_address *joined_list,
size_t joined_list_entries);
char *pcmk__cpg_message_data(cpg_handle_t handle, uint32_t sender_id,
uint32_t pid, void *content, uint32_t *kind,
const char **from);
# endif
const char *pcmk__cluster_node_uuid(pcmk__node_status_t *node);
char *pcmk__cluster_node_name(uint32_t nodeid);
const char *pcmk__cluster_local_node_name(void);
const char *pcmk__node_name_from_uuid(const char *uuid);
pcmk__node_status_t *crm_update_peer_proc(const char *source,
pcmk__node_status_t *peer,
uint32_t flag, const char *status);
pcmk__node_status_t *pcmk__update_peer_state(const char *source,
pcmk__node_status_t *node,
const char *state,
uint64_t membership);
void pcmk__update_peer_expected(const char *source, pcmk__node_status_t *node,
const char *expected);
void pcmk__reap_unseen_nodes(uint64_t ring_id);
void pcmk__corosync_quorum_connect(gboolean (*dispatch)(unsigned long long,
gboolean),
void (*destroy) (gpointer));
enum crm_ais_msg_types pcmk__cluster_parse_msg_type(const char *text);
bool pcmk__cluster_send_message(const pcmk__node_status_t *node,
enum crm_ais_msg_types service,
const xmlNode *data);
// Membership
bool pcmk__cluster_has_quorum(void);
void pcmk__cluster_init_node_caches(void);
void pcmk__cluster_destroy_node_caches(void);
void pcmk__cluster_set_autoreap(bool enable);
void pcmk__cluster_set_status_callback(void (*dispatch)(enum crm_status_type,
pcmk__node_status_t *,
const void *));
bool pcmk__cluster_is_node_active(const pcmk__node_status_t *node);
unsigned int pcmk__cluster_num_active_nodes(void);
unsigned int pcmk__cluster_num_remote_nodes(void);
pcmk__node_status_t *pcmk__cluster_lookup_remote_node(const char *node_name);
void pcmk__cluster_forget_cluster_node(uint32_t id, const char *node_name);
void pcmk__cluster_forget_remote_node(const char *node_name);
pcmk__node_status_t *pcmk__search_node_caches(unsigned int id,
const char *uname,
uint32_t flags);
void pcmk__purge_node_from_cache(const char *node_name, uint32_t node_id);
void pcmk__refresh_node_caches_from_cib(xmlNode *cib);
pcmk__node_status_t *pcmk__get_node(unsigned int id, const char *uname,
const char *uuid, uint32_t flags);
#ifdef __cplusplus
}
#endif
#endif // PCMK__CRM_CLUSTER_INTERNAL__H
diff --git a/lib/cluster/cluster.c b/lib/cluster/cluster.c
index f2b8372abe..5cd6263d40 100644
--- a/lib/cluster/cluster.c
+++ b/lib/cluster/cluster.c
@@ -1,495 +1,495 @@
/*
* Copyright 2004-2024 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.
*/
#include <crm_internal.h>
#include <dlfcn.h>
#include <inttypes.h> // PRIu32
#include <stdbool.h>
#include <stdio.h>
#include <unistd.h>
#include <string.h>
#include <stdlib.h>
#include <time.h>
#include <sys/param.h>
#include <sys/types.h>
#include <sys/utsname.h> // uname()
#include <glib.h> // gboolean
#include <crm/crm.h>
#include <crm/common/ipc.h>
#include <crm/common/xml.h>
#include <crm/cluster/internal.h>
#include "crmcluster_private.h"
CRM_TRACE_INIT_DATA(cluster);
/*!
* \internal
* \brief Get the message type equivalent of a string
*
* \param[in] text String of message type
*
* \return Message type equivalent of \p text
*/
enum crm_ais_msg_types
pcmk__cluster_parse_msg_type(const char *text)
{
CRM_CHECK(text != NULL, return crm_msg_none);
text = pcmk__message_name(text);
if (pcmk__str_eq(text, "ais", pcmk__str_none)) {
return crm_msg_ais;
}
if (pcmk__str_eq(text, CRM_SYSTEM_CIB, pcmk__str_none)) {
return crm_msg_cib;
}
if (pcmk__str_any_of(text, CRM_SYSTEM_CRMD, CRM_SYSTEM_DC, NULL)) {
return crm_msg_crmd;
}
if (pcmk__str_eq(text, CRM_SYSTEM_TENGINE, pcmk__str_none)) {
return crm_msg_te;
}
if (pcmk__str_eq(text, CRM_SYSTEM_PENGINE, pcmk__str_none)) {
return crm_msg_pe;
}
if (pcmk__str_eq(text, CRM_SYSTEM_LRMD, pcmk__str_none)) {
return crm_msg_lrmd;
}
if (pcmk__str_eq(text, CRM_SYSTEM_STONITHD, pcmk__str_none)) {
return crm_msg_stonithd;
}
if (pcmk__str_eq(text, "stonith-ng", pcmk__str_none)) {
return crm_msg_stonith_ng;
}
if (pcmk__str_eq(text, "attrd", pcmk__str_none)) {
return crm_msg_attrd;
}
return crm_msg_none;
}
/*!
* \internal
* \brief Get a node's cluster-layer UUID, setting it if not already set
*
* \param[in,out] node Node to check
*
* \return Cluster-layer node UUID of \p node, or \c NULL if unknown
*/
const char *
pcmk__cluster_node_uuid(pcmk__node_status_t *node)
{
const enum pcmk_cluster_layer cluster_layer = pcmk_get_cluster_layer();
if (node == NULL) {
return NULL;
}
- if (node->uuid != NULL) {
- return node->uuid;
+ if (node->xml_id != NULL) {
+ return node->xml_id;
}
switch (cluster_layer) {
#if SUPPORT_COROSYNC
case pcmk_cluster_layer_corosync:
- node->uuid = pcmk__corosync_uuid(node);
- return node->uuid;
+ node->xml_id = pcmk__corosync_uuid(node);
+ return node->xml_id;
#endif // SUPPORT_COROSYNC
default:
crm_err("Unsupported cluster layer %s",
pcmk_cluster_layer_text(cluster_layer));
return NULL;
}
}
/*!
* \internal
* \brief Connect to the cluster layer
*
* \param[in,out] cluster Initialized cluster object to connect
*
* \return Standard Pacemaker return code
*/
int
pcmk_cluster_connect(pcmk_cluster_t *cluster)
{
const enum pcmk_cluster_layer cluster_layer = pcmk_get_cluster_layer();
const char *cluster_layer_s = pcmk_cluster_layer_text(cluster_layer);
// cts-lab looks for this message
crm_notice("Connecting to %s cluster layer", cluster_layer_s);
switch (cluster_layer) {
#if SUPPORT_COROSYNC
case pcmk_cluster_layer_corosync:
return pcmk__corosync_connect(cluster);
#endif // SUPPORT_COROSYNC
default:
break;
}
crm_err("Failed to connect to unsupported cluster layer %s",
cluster_layer_s);
return EPROTONOSUPPORT;
}
/*!
* \brief Disconnect from the cluster layer
*
* \param[in,out] cluster Cluster object to disconnect
*
* \return Standard Pacemaker return code
*/
int
pcmk_cluster_disconnect(pcmk_cluster_t *cluster)
{
const enum pcmk_cluster_layer cluster_layer = pcmk_get_cluster_layer();
const char *cluster_layer_s = pcmk_cluster_layer_text(cluster_layer);
crm_info("Disconnecting from %s cluster layer", cluster_layer_s);
switch (cluster_layer) {
#if SUPPORT_COROSYNC
case pcmk_cluster_layer_corosync:
pcmk__corosync_disconnect(cluster);
pcmk__cluster_destroy_node_caches();
return pcmk_rc_ok;
#endif // SUPPORT_COROSYNC
default:
break;
}
crm_err("Failed to disconnect from unsupported cluster layer %s",
cluster_layer_s);
return EPROTONOSUPPORT;
}
/*!
* \brief Allocate a new \p pcmk_cluster_t object
*
* \return A newly allocated \p pcmk_cluster_t object (guaranteed not \c NULL)
* \note The caller is responsible for freeing the return value using
* \p pcmk_cluster_free().
*/
pcmk_cluster_t *
pcmk_cluster_new(void)
{
return (pcmk_cluster_t *) pcmk__assert_alloc(1, sizeof(pcmk_cluster_t));
}
/*!
* \brief Free a \p pcmk_cluster_t object and its dynamically allocated members
*
* \param[in,out] cluster Cluster object to free
*/
void
pcmk_cluster_free(pcmk_cluster_t *cluster)
{
if (cluster == NULL) {
return;
}
free(cluster->uuid);
free(cluster->uname);
free(cluster);
}
/*!
* \brief Set the destroy function for a cluster object
*
* \param[in,out] cluster Cluster object
* \param[in] fn Destroy function to set
*
* \return Standard Pacemaker return code
*/
int
pcmk_cluster_set_destroy_fn(pcmk_cluster_t *cluster, void (*fn)(gpointer))
{
if (cluster == NULL) {
return EINVAL;
}
cluster->destroy = fn;
return pcmk_rc_ok;
}
/*!
* \internal
* \brief Send an XML message via the cluster messaging layer
*
* \param[in] node Cluster node to send message to
* \param[in] service Message type to use in message host info
* \param[in] data XML message to send
*
* \return \c true on success, or \c false otherwise
*/
bool
pcmk__cluster_send_message(const pcmk__node_status_t *node,
enum crm_ais_msg_types service, const xmlNode *data)
{
// @TODO Return standard Pacemaker return code
switch (pcmk_get_cluster_layer()) {
#if SUPPORT_COROSYNC
case pcmk_cluster_layer_corosync:
return pcmk__cpg_send_xml(data, node, service);
#endif // SUPPORT_COROSYNC
default:
break;
}
return false;
}
/*!
* \internal
* \brief Get the node name corresponding to a cluster-layer node ID
*
* Get the node name from the cluster layer if possible. Otherwise, if for the
* local node, call \c uname() and get the \c nodename member from the
* <tt>struct utsname</tt> object.
*
* \param[in] nodeid Node ID to check (or 0 for the local node)
*
* \return Node name corresponding to \p nodeid
*
* \note This will fatally exit if \c uname() fails to get the local node name
* or we run out of memory.
* \note The caller is responsible for freeing the return value using \c free().
*/
char *
pcmk__cluster_node_name(uint32_t nodeid)
{
const enum pcmk_cluster_layer cluster_layer = pcmk_get_cluster_layer();
const char *cluster_layer_s = pcmk_cluster_layer_text(cluster_layer);
switch (cluster_layer) {
#if SUPPORT_COROSYNC
case pcmk_cluster_layer_corosync:
return pcmk__corosync_name(0, nodeid);
#else
break;
#endif // SUPPORT_COROSYNC
default:
crm_err("Unsupported cluster layer: %s", cluster_layer_s);
break;
}
if (nodeid == 0) {
struct utsname hostinfo;
crm_notice("Could not get local node name from %s cluster layer, "
"defaulting to local hostname",
cluster_layer_s);
if (uname(&hostinfo) < 0) {
// @TODO Maybe let the caller decide what to do
crm_err("Failed to get the local hostname");
crm_exit(CRM_EX_FATAL);
}
return pcmk__str_copy(hostinfo.nodename);
}
crm_notice("Could not obtain a node name for node with "
PCMK_XA_ID "=" PRIu32,
nodeid);
return NULL;
}
/*!
* \internal
* \brief Get the local node's cluster-layer node name
*
* If getting the node name from the cluster layer is impossible, call
* \c uname() and get the \c nodename member from the <tt>struct utsname</tt>
* object.
*
* \return Local node's name
*
* \note This will fatally exit if \c uname() fails to get the local node name
* or we run out of memory.
*/
const char *
pcmk__cluster_local_node_name(void)
{
// @TODO Refactor to avoid trivially leaking name at exit
static char *name = NULL;
if (name == NULL) {
name = pcmk__cluster_node_name(0);
}
return name;
}
/*!
* \internal
* \brief Get the node name corresonding to a node UUID
*
* Look for the UUID in both the remote node cache and the cluster member cache.
*
* \param[in] uuid UUID to search for
*
* \return Node name corresponding to \p uuid if found, or \c NULL otherwise
*/
const char *
pcmk__node_name_from_uuid(const char *uuid)
{
/* @TODO There are too many functions in libcrmcluster that look up a node
* from the node caches (possibly creating a cache entry if none exists).
* There are at least the following:
* * pcmk__cluster_lookup_remote_node()
* * pcmk__get_node()
* * pcmk__node_name_from_uuid()
* * pcmk__search_node_caches()
*
* There's a lot of duplication among them, but they all do slightly
* different things. We should try to clean them up and consolidate them to
* the extent possible, likely with new helper functions.
*/
GHashTableIter iter;
pcmk__node_status_t *node = NULL;
CRM_CHECK(uuid != NULL, return NULL);
// Remote nodes have the same uname and uuid
if (g_hash_table_lookup(crm_remote_peer_cache, uuid)) {
return uuid;
}
g_hash_table_iter_init(&iter, crm_peer_cache);
while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &node)) {
- if (pcmk__str_eq(node->uuid, uuid, pcmk__str_casei)) {
+ if (pcmk__str_eq(node->xml_id, uuid, pcmk__str_casei)) {
return node->name;
}
}
return NULL;
}
/*!
* \brief Get a log-friendly string equivalent of a cluster layer
*
* \param[in] layer Cluster layer
*
* \return Log-friendly string corresponding to \p layer
*/
const char *
pcmk_cluster_layer_text(enum pcmk_cluster_layer layer)
{
switch (layer) {
case pcmk_cluster_layer_corosync:
return "corosync";
case pcmk_cluster_layer_unknown:
return "unknown";
case pcmk_cluster_layer_invalid:
return "invalid";
default:
crm_err("Invalid cluster layer: %d", layer);
return "invalid";
}
}
/*!
* \brief Get and validate the local cluster layer
*
* If a cluster layer is not configured via the \c PCMK__ENV_CLUSTER_TYPE local
* option, this will try to detect an active cluster from among the supported
* cluster layers.
*
* \return Local cluster layer
*
* \note This will fatally exit if the configured cluster layer is invalid.
*/
enum pcmk_cluster_layer
pcmk_get_cluster_layer(void)
{
static enum pcmk_cluster_layer cluster_layer = pcmk_cluster_layer_unknown;
const char *cluster = NULL;
// Cluster layer is stable once set
if (cluster_layer != pcmk_cluster_layer_unknown) {
return cluster_layer;
}
cluster = pcmk__env_option(PCMK__ENV_CLUSTER_TYPE);
if (cluster != NULL) {
crm_info("Verifying configured cluster layer '%s'", cluster);
cluster_layer = pcmk_cluster_layer_invalid;
#if SUPPORT_COROSYNC
if (pcmk__str_eq(cluster, PCMK_VALUE_COROSYNC, pcmk__str_casei)) {
cluster_layer = pcmk_cluster_layer_corosync;
}
#endif // SUPPORT_COROSYNC
if (cluster_layer == pcmk_cluster_layer_invalid) {
crm_notice("This installation does not support the '%s' cluster "
"infrastructure: terminating",
cluster);
crm_exit(CRM_EX_FATAL);
}
crm_info("Assuming an active '%s' cluster", cluster);
} else {
// Nothing configured, so test supported cluster layers
#if SUPPORT_COROSYNC
crm_debug("Testing with Corosync");
if (pcmk__corosync_is_active()) {
cluster_layer = pcmk_cluster_layer_corosync;
}
#endif // SUPPORT_COROSYNC
if (cluster_layer == pcmk_cluster_layer_unknown) {
crm_notice("Could not determine the current cluster layer");
} else {
crm_info("Detected an active '%s' cluster",
pcmk_cluster_layer_text(cluster_layer));
}
}
return cluster_layer;
}
// Deprecated functions kept only for backward API compatibility
// LCOV_EXCL_START
#include <crm/cluster/compat.h>
gboolean
crm_cluster_connect(pcmk_cluster_t *cluster)
{
return pcmk_cluster_connect(cluster) == pcmk_rc_ok;
}
const char *
name_for_cluster_type(enum cluster_type_e type)
{
switch (type) {
case pcmk_cluster_corosync:
return "corosync";
case pcmk_cluster_unknown:
return "unknown";
case pcmk_cluster_invalid:
return "invalid";
}
crm_err("Invalid cluster type: %d", type);
return "invalid";
}
enum cluster_type_e
get_cluster_type(void)
{
return (enum cluster_type_e) pcmk_get_cluster_layer();
}
// LCOV_EXCL_STOP
// End deprecated API
diff --git a/lib/cluster/election.c b/lib/cluster/election.c
index b8921218ef..9ebe7c7e9f 100644
--- a/lib/cluster/election.c
+++ b/lib/cluster/election.c
@@ -1,727 +1,727 @@
/*
* Copyright 2004-2024 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.
*/
#include <crm_internal.h>
#include <sys/time.h>
#include <sys/resource.h>
#include <crm/common/xml.h>
#include <crm/common/mainloop.h>
#include <crm/cluster/internal.h>
#include <crm/cluster/election_internal.h>
#include <crm/crm.h>
#define STORM_INTERVAL 2 /* in seconds */
struct election_s {
enum election_result state;
guint count; // How many times local node has voted
char *name; // Descriptive name for this election
char *uname; // Local node's name
GSourceFunc cb; // Function to call if election is won
GHashTable *voted; // Key = node name, value = how node voted
mainloop_timer_t *timeout; // When to abort if all votes not received
int election_wins; // Track wins, for storm detection
bool wrote_blackbox; // Write a storm blackbox at most once
time_t expires; // When storm detection period ends
time_t last_election_loss; // When dampening period ends
};
static void
election_complete(election_t *e)
{
e->state = election_won;
if (e->cb != NULL) {
e->cb(e);
}
election_reset(e);
}
static gboolean
election_timer_cb(gpointer user_data)
{
election_t *e = user_data;
crm_info("%s timed out, declaring local node as winner", e->name);
election_complete(e);
return FALSE;
}
/*!
* \brief Get current state of an election
*
* \param[in] e Election object
*
* \return Current state of \e
*/
enum election_result
election_state(const election_t *e)
{
return (e == NULL)? election_error : e->state;
}
/*!
* \brief Create a new election object
*
* Every node that wishes to participate in an election must create an election
* object. Typically, this should be done once, at start-up. A caller should
* only create a single election object.
*
* \param[in] name Label for election (for logging)
* \param[in] uname Local node's name
* \param[in] period_ms How long to wait for all peers to vote
* \param[in] cb Function to call if local node wins election
*
* \return Newly allocated election object on success, NULL on error
* \note The caller is responsible for freeing the returned value using
* election_fini().
*/
election_t *
election_init(const char *name, const char *uname, guint period_ms, GSourceFunc cb)
{
election_t *e = NULL;
static guint count = 0;
CRM_CHECK(uname != NULL, return NULL);
e = calloc(1, sizeof(election_t));
if (e == NULL) {
crm_perror(LOG_CRIT, "Cannot create election");
return NULL;
}
e->uname = strdup(uname);
if (e->uname == NULL) {
crm_perror(LOG_CRIT, "Cannot create election");
free(e);
return NULL;
}
e->name = name? crm_strdup_printf("election-%s", name)
: crm_strdup_printf("election-%u", count++);
e->cb = cb;
e->timeout = mainloop_timer_add(e->name, period_ms, FALSE,
election_timer_cb, e);
crm_trace("Created %s", e->name);
return e;
}
/*!
* \brief Disregard any previous vote by specified peer
*
* This discards any recorded vote from a specified peer. Election users should
* call this whenever a voting peer becomes inactive.
*
* \param[in,out] e Election object
* \param[in] uname Name of peer to disregard
*/
void
election_remove(election_t *e, const char *uname)
{
if ((e != NULL) && (uname != NULL) && (e->voted != NULL)) {
crm_trace("Discarding %s (no-)vote from lost peer %s", e->name, uname);
g_hash_table_remove(e->voted, uname);
}
}
/*!
* \brief Stop election timer and disregard all votes
*
* \param[in,out] e Election object
*/
void
election_reset(election_t *e)
{
if (e != NULL) {
crm_trace("Resetting election %s", e->name);
mainloop_timer_stop(e->timeout);
if (e->voted) {
crm_trace("Destroying voted cache with %d members", g_hash_table_size(e->voted));
g_hash_table_destroy(e->voted);
e->voted = NULL;
}
}
}
/*!
* \brief Free an election object
*
* Free all memory associated with an election object, stopping its
* election timer (if running).
*
* \param[in,out] e Election object
*/
void
election_fini(election_t *e)
{
if (e != NULL) {
election_reset(e);
crm_trace("Destroying %s", e->name);
mainloop_timer_del(e->timeout);
free(e->uname);
free(e->name);
free(e);
}
}
static void
election_timeout_start(election_t *e)
{
if (e != NULL) {
mainloop_timer_start(e->timeout);
}
}
/*!
* \brief Stop an election's timer, if running
*
* \param[in,out] e Election object
*/
void
election_timeout_stop(election_t *e)
{
if (e != NULL) {
mainloop_timer_stop(e->timeout);
}
}
/*!
* \brief Change an election's timeout (restarting timer if running)
*
* \param[in,out] e Election object
* \param[in] period New timeout
*/
void
election_timeout_set_period(election_t *e, guint period)
{
if (e != NULL) {
mainloop_timer_set_period(e->timeout, period);
} else {
crm_err("No election defined");
}
}
static int
get_uptime(struct timeval *output)
{
static time_t expires = 0;
static struct rusage info;
time_t tm_now = time(NULL);
if (expires < tm_now) {
int rc = 0;
info.ru_utime.tv_sec = 0;
info.ru_utime.tv_usec = 0;
rc = getrusage(RUSAGE_SELF, &info);
output->tv_sec = 0;
output->tv_usec = 0;
if (rc < 0) {
crm_perror(LOG_ERR, "Could not calculate the current uptime");
expires = 0;
return -1;
}
crm_debug("Current CPU usage is: %lds, %ldus", (long)info.ru_utime.tv_sec,
(long)info.ru_utime.tv_usec);
}
expires = tm_now + STORM_INTERVAL; /* N seconds after the last _access_ */
output->tv_sec = info.ru_utime.tv_sec;
output->tv_usec = info.ru_utime.tv_usec;
return 1;
}
static int
compare_age(struct timeval your_age)
{
struct timeval our_age;
get_uptime(&our_age); /* If an error occurred, our_age will be compared as {0,0} */
if (our_age.tv_sec > your_age.tv_sec) {
crm_debug("Win: %ld vs %ld (seconds)", (long)our_age.tv_sec, (long)your_age.tv_sec);
return 1;
} else if (our_age.tv_sec < your_age.tv_sec) {
crm_debug("Lose: %ld vs %ld (seconds)", (long)our_age.tv_sec, (long)your_age.tv_sec);
return -1;
} else if (our_age.tv_usec > your_age.tv_usec) {
crm_debug("Win: %ld.%06ld vs %ld.%06ld (usec)",
(long)our_age.tv_sec, (long)our_age.tv_usec, (long)your_age.tv_sec, (long)your_age.tv_usec);
return 1;
} else if (our_age.tv_usec < your_age.tv_usec) {
crm_debug("Lose: %ld.%06ld vs %ld.%06ld (usec)",
(long)our_age.tv_sec, (long)our_age.tv_usec, (long)your_age.tv_sec, (long)your_age.tv_usec);
return -1;
}
return 0;
}
/*!
* \brief Start a new election by offering local node's candidacy
*
* Broadcast a "vote" election message containing the local node's ID,
* (incremented) election counter, and uptime, and start the election timer.
*
* \param[in,out] e Election object
*
* \note Any nodes agreeing to the candidacy will send a "no-vote" reply, and if
* all active peers do so, or if the election times out, the local node
* wins the election. (If we lose to any peer vote, we will stop the
* timer, so a timeout means we did not lose -- either some peer did not
* vote, or we did not call election_check() in time.)
*/
void
election_vote(election_t *e)
{
struct timeval age;
xmlNode *vote = NULL;
pcmk__node_status_t *our_node = NULL;
if (e == NULL) {
crm_trace("Election vote requested, but no election available");
return;
}
our_node = pcmk__get_node(0, e->uname, NULL,
pcmk__node_search_cluster_member);
if (!pcmk__cluster_is_node_active(our_node)) {
crm_trace("Cannot vote in %s yet: local node not connected to cluster",
e->name);
return;
}
election_reset(e);
e->state = election_in_progress;
vote = create_request(CRM_OP_VOTE, NULL, NULL, CRM_SYSTEM_CRMD, CRM_SYSTEM_CRMD, NULL);
e->count++;
- crm_xml_add(vote, PCMK__XA_ELECTION_OWNER, our_node->uuid);
+ crm_xml_add(vote, PCMK__XA_ELECTION_OWNER, our_node->xml_id);
crm_xml_add_int(vote, PCMK__XA_ELECTION_ID, e->count);
// Warning: PCMK__XA_ELECTION_AGE_NANO_SEC value is actually microseconds
get_uptime(&age);
crm_xml_add_timeval(vote, PCMK__XA_ELECTION_AGE_SEC,
PCMK__XA_ELECTION_AGE_NANO_SEC, &age);
pcmk__cluster_send_message(NULL, crm_msg_crmd, vote);
pcmk__xml_free(vote);
crm_debug("Started %s round %d", e->name, e->count);
election_timeout_start(e);
return;
}
/*!
* \brief Check whether local node has won an election
*
* If all known peers have sent no-vote messages, stop the election timer, set
* the election state to won, and call any registered win callback.
*
* \param[in,out] e Election object
*
* \return TRUE if local node has won, FALSE otherwise
* \note If all known peers have sent no-vote messages, but the election owner
* does not call this function, the election will not be won (and the
* callback will not be called) until the election times out.
* \note This should be called when election_count_vote() returns
* \c election_in_progress.
*/
bool
election_check(election_t *e)
{
int voted_size = 0;
int num_members = 0;
if (e == NULL) {
crm_trace("Election check requested, but no election available");
return FALSE;
}
if (e->voted == NULL) {
crm_trace("%s check requested, but no votes received yet", e->name);
return FALSE;
}
voted_size = g_hash_table_size(e->voted);
num_members = pcmk__cluster_num_active_nodes();
/* in the case of #voted > #members, it is better to
* wait for the timeout and give the cluster time to
* stabilize
*/
if (voted_size >= num_members) {
/* we won and everyone has voted */
election_timeout_stop(e);
if (voted_size > num_members) {
GHashTableIter gIter;
const pcmk__node_status_t *node = NULL;
char *key = NULL;
crm_warn("Received too many votes in %s", e->name);
g_hash_table_iter_init(&gIter, crm_peer_cache);
while (g_hash_table_iter_next(&gIter, NULL, (gpointer *) & node)) {
if (pcmk__cluster_is_node_active(node)) {
crm_warn("* expected vote: %s", node->name);
}
}
g_hash_table_iter_init(&gIter, e->voted);
while (g_hash_table_iter_next(&gIter, (gpointer *) & key, NULL)) {
crm_warn("* actual vote: %s", key);
}
}
crm_info("%s won by local node", e->name);
election_complete(e);
return TRUE;
} else {
crm_debug("%s still waiting on %d of %d votes",
e->name, num_members - voted_size, num_members);
}
return FALSE;
}
#define LOSS_DAMPEN 2 /* in seconds */
struct vote {
const char *op;
const char *from;
const char *version;
const char *election_owner;
int election_id;
struct timeval age;
};
/*!
* \brief Unpack an election message
*
* \param[in] e Election object (for logging only)
* \param[in] message Election message XML
* \param[out] vote Parsed fields from message
*
* \return TRUE if election message and election are valid, FALSE otherwise
* \note The parsed struct's pointer members are valid only for the lifetime of
* the message argument.
*/
static bool
parse_election_message(const election_t *e, const xmlNode *message,
struct vote *vote)
{
CRM_CHECK(message && vote, return FALSE);
vote->election_id = -1;
vote->age.tv_sec = -1;
vote->age.tv_usec = -1;
vote->op = crm_element_value(message, PCMK__XA_CRM_TASK);
vote->from = crm_element_value(message, PCMK__XA_SRC);
vote->version = crm_element_value(message, PCMK_XA_VERSION);
vote->election_owner = crm_element_value(message, PCMK__XA_ELECTION_OWNER);
crm_element_value_int(message, PCMK__XA_ELECTION_ID, &(vote->election_id));
if ((vote->op == NULL) || (vote->from == NULL) || (vote->version == NULL)
|| (vote->election_owner == NULL) || (vote->election_id < 0)) {
crm_warn("Invalid %s message from %s in %s ",
(vote->op? vote->op : "election"),
(vote->from? vote->from : "unspecified node"),
(e? e->name : "election"));
return FALSE;
}
// Op-specific validation
if (pcmk__str_eq(vote->op, CRM_OP_VOTE, pcmk__str_none)) {
/* Only vote ops have uptime.
Warning: PCMK__XA_ELECTION_AGE_NANO_SEC value is in microseconds.
*/
crm_element_value_timeval(message, PCMK__XA_ELECTION_AGE_SEC,
PCMK__XA_ELECTION_AGE_NANO_SEC, &(vote->age));
if ((vote->age.tv_sec < 0) || (vote->age.tv_usec < 0)) {
crm_warn("Cannot count %s %s from %s because it is missing uptime",
(e? e->name : "election"), vote->op, vote->from);
return FALSE;
}
} else if (!pcmk__str_eq(vote->op, CRM_OP_NOVOTE, pcmk__str_none)) {
crm_info("Cannot process %s message from %s because %s is not a known election op",
(e? e->name : "election"), vote->from, vote->op);
return FALSE;
}
// Election validation
if (e == NULL) {
crm_info("Cannot count %s from %s because no election available",
vote->op, vote->from);
return FALSE;
}
/* If the membership cache is NULL, we REALLY shouldn't be voting --
* the question is how we managed to get here.
*/
if (crm_peer_cache == NULL) {
crm_info("Cannot count %s %s from %s because no peer information available",
e->name, vote->op, vote->from);
return FALSE;
}
return TRUE;
}
static void
record_vote(election_t *e, struct vote *vote)
{
CRM_ASSERT(e && vote && vote->from && vote->op);
if (e->voted == NULL) {
e->voted = pcmk__strkey_table(free, free);
}
pcmk__insert_dup(e->voted, vote->from, vote->op);
}
static void
send_no_vote(pcmk__node_status_t *peer, struct vote *vote)
{
// @TODO probably shouldn't hardcode CRM_SYSTEM_CRMD and crm_msg_crmd
xmlNode *novote = create_request(CRM_OP_NOVOTE, NULL, vote->from,
CRM_SYSTEM_CRMD, CRM_SYSTEM_CRMD, NULL);
crm_xml_add(novote, PCMK__XA_ELECTION_OWNER, vote->election_owner);
crm_xml_add_int(novote, PCMK__XA_ELECTION_ID, vote->election_id);
pcmk__cluster_send_message(peer, crm_msg_crmd, novote);
pcmk__xml_free(novote);
}
/*!
* \brief Process an election message (vote or no-vote) from a peer
*
* \param[in,out] e Election object
* \param[in] message Election message XML from peer
* \param[in] can_win Whether local node is eligible to win
*
* \return Election state after new vote is considered
* \note If the peer message is a vote, and we prefer the peer to win, this will
* send a no-vote reply to the peer.
* \note The situations "we lost to this vote" from "this is a late no-vote
* after we've already lost" both return election_lost. If a caller needs
* to distinguish them, it should save the current state before calling
* this function, and then compare the result.
*/
enum election_result
election_count_vote(election_t *e, const xmlNode *message, bool can_win)
{
int log_level = LOG_INFO;
gboolean done = FALSE;
gboolean we_lose = FALSE;
const char *reason = "unknown";
bool we_are_owner = FALSE;
pcmk__node_status_t *our_node = NULL;
pcmk__node_status_t *your_node = NULL;
time_t tm_now = time(NULL);
struct vote vote;
CRM_CHECK(message != NULL, return election_error);
if (parse_election_message(e, message, &vote) == FALSE) {
return election_error;
}
your_node = pcmk__get_node(0, vote.from, NULL,
pcmk__node_search_cluster_member);
our_node = pcmk__get_node(0, e->uname, NULL,
pcmk__node_search_cluster_member);
we_are_owner = (our_node != NULL)
- && pcmk__str_eq(our_node->uuid, vote.election_owner,
+ && pcmk__str_eq(our_node->xml_id, vote.election_owner,
pcmk__str_none);
if (!can_win) {
reason = "Not eligible";
we_lose = TRUE;
} else if (!pcmk__cluster_is_node_active(our_node)) {
reason = "We are not part of the cluster";
log_level = LOG_ERR;
we_lose = TRUE;
} else if (we_are_owner && (vote.election_id != e->count)) {
log_level = LOG_TRACE;
reason = "Superseded";
done = TRUE;
} else if (!pcmk__cluster_is_node_active(your_node)) {
/* Possibly we cached the message in the FSA queue at a point that it wasn't */
reason = "Peer is not part of our cluster";
log_level = LOG_WARNING;
done = TRUE;
} else if (pcmk__str_eq(vote.op, CRM_OP_NOVOTE, pcmk__str_none)
|| pcmk__str_eq(vote.from, e->uname, pcmk__str_none)) {
/* Receiving our own broadcast vote, or a no-vote from peer, is a vote
* for us to win
*/
if (!we_are_owner) {
crm_warn("Cannot count %s round %d %s from %s because we are not election owner (%s)",
e->name, vote.election_id, vote.op, vote.from,
vote.election_owner);
return election_error;
}
if (e->state != election_in_progress) {
// Should only happen if we already lost
crm_debug("Not counting %s round %d %s from %s because no election in progress",
e->name, vote.election_id, vote.op, vote.from);
return e->state;
}
record_vote(e, &vote);
reason = "Recorded";
done = TRUE;
} else {
// A peer vote requires a comparison to determine which node is better
int age_result = compare_age(vote.age);
int version_result = compare_version(vote.version, CRM_FEATURE_SET);
if (version_result < 0) {
reason = "Version";
we_lose = TRUE;
} else if (version_result > 0) {
reason = "Version";
} else if (age_result < 0) {
reason = "Uptime";
we_lose = TRUE;
} else if (age_result > 0) {
reason = "Uptime";
} else if (strcasecmp(e->uname, vote.from) > 0) {
reason = "Host name";
we_lose = TRUE;
} else {
reason = "Host name";
}
}
if (e->expires < tm_now) {
e->election_wins = 0;
e->expires = tm_now + STORM_INTERVAL;
} else if (done == FALSE && we_lose == FALSE) {
int peers = 1 + g_hash_table_size(crm_peer_cache);
/* If every node has to vote down every other node, thats N*(N-1) total elections
* Allow some leeway before _really_ complaining
*/
e->election_wins++;
if (e->election_wins > (peers * peers)) {
crm_warn("%s election storm detected: %d wins in %d seconds",
e->name, e->election_wins, STORM_INTERVAL);
e->election_wins = 0;
e->expires = tm_now + STORM_INTERVAL;
if (e->wrote_blackbox == FALSE) {
/* It's questionable whether a black box (from every node in the
* cluster) would be truly helpful in diagnosing an election
* storm. It's also highly doubtful a production environment
* would get multiple election storms from distinct causes, so
* saving one blackbox per process lifetime should be
* sufficient. Alternatives would be to save a timestamp of the
* last blackbox write instead of a boolean, and write a new one
* if some amount of time has passed; or to save a storm count,
* write a blackbox on every Nth occurrence.
*/
crm_write_blackbox(0, NULL);
e->wrote_blackbox = TRUE;
}
}
}
if (done) {
do_crm_log(log_level + 1,
"Processed %s round %d %s (current round %d) from %s (%s)",
e->name, vote.election_id, vote.op, e->count, vote.from,
reason);
return e->state;
} else if (we_lose == FALSE) {
/* We track the time of the last election loss to implement an election
* dampening period, reducing the likelihood of an election storm. If
* this node has lost within the dampening period, don't start a new
* election, even if we win against a peer's vote -- the peer we lost to
* should win again.
*
* @TODO This has a problem case: if an election winner immediately
* leaves the cluster, and a new election is immediately called, all
* nodes could lose, with no new winner elected. The ideal solution
* would be to tie the election structure with the peer caches, which
* would allow us to clear the dampening when the previous winner
* leaves (and would allow other improvements as well).
*/
if ((e->last_election_loss == 0)
|| ((tm_now - e->last_election_loss) > (time_t) LOSS_DAMPEN)) {
do_crm_log(log_level, "%s round %d (owner node ID %s) pass: %s from %s (%s)",
e->name, vote.election_id, vote.election_owner, vote.op,
vote.from, reason);
e->last_election_loss = 0;
election_timeout_stop(e);
/* Start a new election by voting down this, and other, peers */
e->state = election_start;
return e->state;
} else {
char *loss_time = ctime(&e->last_election_loss);
if (loss_time) {
// Show only HH:MM:SS
loss_time += 11;
loss_time[8] = '\0';
}
crm_info("Ignoring %s round %d (owner node ID %s) pass vs %s because we lost less than %ds ago at %s",
e->name, vote.election_id, vote.election_owner, vote.from,
LOSS_DAMPEN, (loss_time? loss_time : "unknown"));
}
}
e->last_election_loss = tm_now;
do_crm_log(log_level, "%s round %d (owner node ID %s) lost: %s from %s (%s)",
e->name, vote.election_id, vote.election_owner, vote.op,
vote.from, reason);
election_reset(e);
send_no_vote(your_node, &vote);
e->state = election_lost;
return e->state;
}
/*!
* \brief Reset any election dampening currently in effect
*
* \param[in,out] e Election object to clear
*/
void
election_clear_dampening(election_t *e)
{
e->last_election_loss = 0;
}
diff --git a/lib/cluster/membership.c b/lib/cluster/membership.c
index 929ff3d36b..8335bf7c91 100644
--- a/lib/cluster/membership.c
+++ b/lib/cluster/membership.c
@@ -1,1502 +1,1502 @@
/*
* Copyright 2004-2024 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.
*/
#include <crm_internal.h>
#ifndef _GNU_SOURCE
# define _GNU_SOURCE
#endif
#include <inttypes.h> // PRIu32
#include <stdbool.h> // bool
#include <stdio.h>
#include <string.h>
#include <sys/param.h>
#include <sys/types.h>
#include <unistd.h>
#include <glib.h>
#include <crm/common/ipc.h>
#include <crm/common/xml_internal.h>
#include <crm/cluster/internal.h>
#include <crm/common/xml.h>
#include <crm/stonith-ng.h>
#include "crmcluster_private.h"
/* The peer cache remembers cluster nodes that have been seen.
* This is managed mostly automatically by libcluster, based on
* cluster membership events.
*
* Because cluster nodes can have conflicting names or UUIDs,
* the hash table key is a uniquely generated ID.
*
* @COMPAT When this is internal, rename to cluster_node_member_cache and make
* static.
*/
GHashTable *crm_peer_cache = NULL;
/*
* The remote peer cache tracks pacemaker_remote nodes. While the
* value has the same type as the peer cache's, it is tracked separately for
* three reasons: pacemaker_remote nodes can't have conflicting names or UUIDs,
* so the name (which is also the UUID) is used as the hash table key; there
* is no equivalent of membership events, so management is not automatic; and
* most users of the peer cache need to exclude pacemaker_remote nodes.
*
* That said, using a single cache would be more logical and less error-prone,
* so it would be a good idea to merge them one day.
*
* libcluster provides two avenues for populating the cache:
* pcmk__cluster_lookup_remote_node() and pcmk__cluster_forget_remote_node()
* directly manage it, while refresh_remote_nodes() populates it via the CIB.
*/
GHashTable *crm_remote_peer_cache = NULL;
/*
* The CIB cluster node cache tracks cluster nodes that have been seen in
* the CIB. It is useful mainly when a caller needs to know about a node that
* may no longer be in the membership, but doesn't want to add the node to the
* main peer cache tables.
*/
static GHashTable *cluster_node_cib_cache = NULL;
unsigned long long crm_peer_seq = 0;
static bool autoreap = true;
static bool has_quorum = false;
// Flag setting and clearing for pcmk__node_status_t:flags
#define set_peer_flags(peer, flags_to_set) do { \
(peer)->flags = pcmk__set_flags_as(__func__, __LINE__, LOG_TRACE, \
"Peer", (peer)->name, \
(peer)->flags, (flags_to_set), \
#flags_to_set); \
} while (0)
#define clear_peer_flags(peer, flags_to_clear) do { \
(peer)->flags = pcmk__clear_flags_as(__func__, __LINE__, \
LOG_TRACE, \
"Peer", (peer)->name, \
(peer)->flags, (flags_to_clear), \
#flags_to_clear); \
} while (0)
static void update_peer_uname(pcmk__node_status_t *node, const char *uname);
static pcmk__node_status_t *find_cib_cluster_node(const char *id,
const char *uname);
/*!
* \internal
* \brief Check whether the cluster currently has quorum
*
* \return \c true if the cluster has quorum, or \c false otherwise
*/
bool
pcmk__cluster_has_quorum(void)
{
return has_quorum;
}
/*!
* \internal
* \brief Set whether the cluster currently has quorum
*
* \param[in] quorate \c true if the cluster has quorum, or \c false otherwise
*/
void
pcmk__cluster_set_quorum(bool quorate)
{
has_quorum = quorate;
}
/*!
* \internal
* \brief Get the number of Pacemaker Remote nodes that have been seen
*
* \return Number of cached Pacemaker Remote nodes
*/
unsigned int
pcmk__cluster_num_remote_nodes(void)
{
if (crm_remote_peer_cache == NULL) {
return 0U;
}
return g_hash_table_size(crm_remote_peer_cache);
}
/*!
* \internal
* \brief Get a remote node cache entry, creating it if necessary
*
* \param[in] node_name Name of remote node
*
* \return Cache entry for node on success, or \c NULL (and set \c errno)
* otherwise
*
* \note When creating a new entry, this will leave the node state undetermined.
* The caller should also call \c pcmk__update_peer_state() if the state
* is known.
* \note Because this can add and remove cache entries, callers should not
* assume any previously obtained cache entry pointers remain valid.
*/
pcmk__node_status_t *
pcmk__cluster_lookup_remote_node(const char *node_name)
{
pcmk__node_status_t *node = NULL;
char *node_name_copy = NULL;
if (node_name == NULL) {
errno = EINVAL;
return NULL;
}
/* It's theoretically possible that the node was added to the cluster peer
* cache before it was known to be a Pacemaker Remote node. Remove that
* entry unless it has a node ID, which means the name actually is
* associated with a cluster node. (@TODO return an error in that case?)
*/
node = pcmk__search_node_caches(0, node_name,
pcmk__node_search_cluster_member);
- if ((node != NULL) && (node->uuid == NULL)) {
+ if ((node != NULL) && (node->xml_id == NULL)) {
/* node_name could be a pointer into the cache entry being removed, so
* reassign it to a copy before the original gets freed
*/
node_name_copy = strdup(node_name);
if (node_name_copy == NULL) {
errno = ENOMEM;
return NULL;
}
node_name = node_name_copy;
pcmk__cluster_forget_cluster_node(0, node_name);
}
/* Return existing cache entry if one exists */
node = g_hash_table_lookup(crm_remote_peer_cache, node_name);
if (node) {
free(node_name_copy);
return node;
}
/* Allocate a new entry */
node = calloc(1, sizeof(pcmk__node_status_t));
if (node == NULL) {
free(node_name_copy);
return NULL;
}
/* Populate the essential information */
set_peer_flags(node, crm_remote_node);
- node->uuid = strdup(node_name);
- if (node->uuid == NULL) {
+ node->xml_id = strdup(node_name);
+ if (node->xml_id == NULL) {
free(node);
errno = ENOMEM;
free(node_name_copy);
return NULL;
}
/* Add the new entry to the cache */
- g_hash_table_replace(crm_remote_peer_cache, node->uuid, node);
+ g_hash_table_replace(crm_remote_peer_cache, node->xml_id, node);
crm_trace("added %s to remote cache", node_name);
/* Update the entry's uname, ensuring peer status callbacks are called */
update_peer_uname(node, node_name);
free(node_name_copy);
return node;
}
/*!
* \internal
* \brief Remove a node from the Pacemaker Remote node cache
*
* \param[in] node_name Name of node to remove from cache
*
* \note The caller must be careful not to use \p node_name after calling this
* function if it might be a pointer into the cache entry being removed.
*/
void
pcmk__cluster_forget_remote_node(const char *node_name)
{
/* Do a lookup first, because node_name could be a pointer within the entry
* being removed -- we can't log it *after* removing it.
*/
if (g_hash_table_lookup(crm_remote_peer_cache, node_name) != NULL) {
crm_trace("Removing %s from Pacemaker Remote node cache", node_name);
g_hash_table_remove(crm_remote_peer_cache, node_name);
}
}
/*!
* \internal
* \brief Return node status based on a CIB status entry
*
* \param[in] node_state XML of node state
*
* \return \c CRM_NODE_MEMBER if \c PCMK__XA_IN_CCM is true in
* \c PCMK__XE_NODE_STATE, or \c CRM_NODE_LOST otherwise
*/
static const char *
remote_state_from_cib(const xmlNode *node_state)
{
bool in_ccm = false;
if ((pcmk__xe_get_bool_attr(node_state, PCMK__XA_IN_CCM,
&in_ccm) == pcmk_rc_ok) && in_ccm) {
return CRM_NODE_MEMBER;
}
return CRM_NODE_LOST;
}
/* user data for looping through remote node xpath searches */
struct refresh_data {
const char *field; /* XML attribute to check for node name */
gboolean has_state; /* whether to update node state based on XML */
};
/*!
* \internal
* \brief Process one pacemaker_remote node xpath search result
*
* \param[in] result XML search result
* \param[in] user_data what to look for in the XML
*/
static void
remote_cache_refresh_helper(xmlNode *result, void *user_data)
{
const struct refresh_data *data = user_data;
const char *remote = crm_element_value(result, data->field);
const char *state = NULL;
pcmk__node_status_t *node;
CRM_CHECK(remote != NULL, return);
/* Determine node's state, if the result has it */
if (data->has_state) {
state = remote_state_from_cib(result);
}
/* Check whether cache already has entry for node */
node = g_hash_table_lookup(crm_remote_peer_cache, remote);
if (node == NULL) {
/* Node is not in cache, so add a new entry for it */
node = pcmk__cluster_lookup_remote_node(remote);
CRM_ASSERT(node);
if (state) {
pcmk__update_peer_state(__func__, node, state, 0);
}
} else if (pcmk_is_set(node->flags, crm_node_dirty)) {
/* Node is in cache and hasn't been updated already, so mark it clean */
clear_peer_flags(node, crm_node_dirty);
if (state) {
pcmk__update_peer_state(__func__, node, state, 0);
}
}
}
static void
mark_dirty(gpointer key, gpointer value, gpointer user_data)
{
set_peer_flags((pcmk__node_status_t *) value, crm_node_dirty);
}
static gboolean
is_dirty(gpointer key, gpointer value, gpointer user_data)
{
return pcmk_is_set(((pcmk__node_status_t*)value)->flags, crm_node_dirty);
}
/*!
* \internal
* \brief Repopulate the remote node cache based on CIB XML
*
* \param[in] cib CIB XML to parse
*/
static void
refresh_remote_nodes(xmlNode *cib)
{
struct refresh_data data;
pcmk__cluster_init_node_caches();
/* First, we mark all existing cache entries as dirty,
* so that later we can remove any that weren't in the CIB.
* We don't empty the cache, because we need to detect changes in state.
*/
g_hash_table_foreach(crm_remote_peer_cache, mark_dirty, NULL);
/* Look for guest nodes and remote nodes in the status section */
data.field = PCMK_XA_ID;
data.has_state = TRUE;
crm_foreach_xpath_result(cib, PCMK__XP_REMOTE_NODE_STATUS,
remote_cache_refresh_helper, &data);
/* Look for guest nodes and remote nodes in the configuration section,
* because they may have just been added and not have a status entry yet.
* In that case, the cached node state will be left NULL, so that the
* peer status callback isn't called until we're sure the node started
* successfully.
*/
data.field = PCMK_XA_VALUE;
data.has_state = FALSE;
crm_foreach_xpath_result(cib, PCMK__XP_GUEST_NODE_CONFIG,
remote_cache_refresh_helper, &data);
data.field = PCMK_XA_ID;
data.has_state = FALSE;
crm_foreach_xpath_result(cib, PCMK__XP_REMOTE_NODE_CONFIG,
remote_cache_refresh_helper, &data);
/* Remove all old cache entries that weren't seen in the CIB */
g_hash_table_foreach_remove(crm_remote_peer_cache, is_dirty, NULL);
}
/*!
* \internal
* \brief Check whether a node is an active cluster node
*
* Remote nodes are never considered active. This guarantees that they can never
* become DC.
*
* \param[in] node Node to check
*
* \return \c true if the node is an active cluster node, or \c false otherwise
*/
bool
pcmk__cluster_is_node_active(const pcmk__node_status_t *node)
{
const enum pcmk_cluster_layer cluster_layer = pcmk_get_cluster_layer();
if ((node == NULL) || pcmk_is_set(node->flags, crm_remote_node)) {
return false;
}
switch (cluster_layer) {
case pcmk_cluster_layer_corosync:
#if SUPPORT_COROSYNC
return pcmk__corosync_is_peer_active(node);
#else
break;
#endif // SUPPORT_COROSYNC
default:
break;
}
crm_err("Unhandled cluster layer: %s",
pcmk_cluster_layer_text(cluster_layer));
return false;
}
/*!
* \internal
* \brief Check if a node's entry should be removed from the cluster node cache
*
* A node should be removed from the cache if it's inactive and matches another
* \c pcmk__node_status_t (the search object). The node is considered a
* mismatch if any of the following are true:
* * The search object is \c NULL.
* * The search object has an ID set and the cached node's ID does not match it.
* * The search object does not have an ID set, and the cached node's name does
* not match the search node's name. (If both names are \c NULL, it's a
* match.)
*
* Otherwise, the node is considered a match.
*
* Note that if the search object has both an ID and a name set, the name is
* ignored for matching purposes.
*
* \param[in] key Ignored
* \param[in] value \c pcmk__node_status_t object from cluster node cache
* \param[in] user_data \c pcmk__node_status_t object to match against (search
* object)
*
* \return \c TRUE if the node entry should be removed from \c crm_peer_cache,
* or \c FALSE otherwise
*/
static gboolean
should_forget_cluster_node(gpointer key, gpointer value, gpointer user_data)
{
pcmk__node_status_t *node = value;
pcmk__node_status_t *search = user_data;
if (search == NULL) {
return FALSE;
}
if ((search->cluster_layer_id != 0)
&& (node->cluster_layer_id != search->cluster_layer_id)) {
return FALSE;
}
if ((search->cluster_layer_id == 0)
&& !pcmk__str_eq(node->name, search->name, pcmk__str_casei)) {
// @TODO Consider name even if ID is set?
return FALSE;
}
if (pcmk__cluster_is_node_active(value)) {
return FALSE;
}
crm_info("Removing node with name %s and cluster layer ID " PRIu32
" from membership cache",
pcmk__s(node->name, "(unknown)"), node->cluster_layer_id);
return TRUE;
}
/*!
* \internal
* \brief Remove one or more inactive nodes from the cluster node cache
*
* All inactive nodes matching \p id and \p node_name as described in
* \c should_forget_cluster_node documentation are removed from the cache.
*
* If \p id is 0 and \p node_name is \c NULL, all inactive nodes are removed
* from the cache regardless of ID and name. This differs from clearing the
* cache, in that entries for active nodes are preserved.
*
* \param[in] id ID of node to remove from cache (0 to ignore)
* \param[in] node_name Name of node to remove from cache (ignored if \p id is
* nonzero)
*
* \note \p node_name is not modified directly, but it will be freed if it's a
* pointer into a cache entry that is removed.
*/
void
pcmk__cluster_forget_cluster_node(uint32_t id, const char *node_name)
{
pcmk__node_status_t search = { 0, };
char *criterion = NULL; // For logging
guint matches = 0;
if (crm_peer_cache == NULL) {
crm_trace("Membership cache not initialized, ignoring removal request");
return;
}
search.cluster_layer_id = id;
search.name = pcmk__str_copy(node_name); // May log after original freed
if (id > 0) {
criterion = crm_strdup_printf("cluster layer ID %" PRIu32, id);
} else if (node_name != NULL) {
criterion = crm_strdup_printf("name %s", node_name);
}
matches = g_hash_table_foreach_remove(crm_peer_cache,
should_forget_cluster_node, &search);
if (matches > 0) {
if (criterion != NULL) {
crm_notice("Removed %u inactive node%s with %s from the membership "
"cache",
matches, pcmk__plural_s(matches), criterion);
} else {
crm_notice("Removed all (%u) inactive cluster nodes from the "
"membership cache",
matches);
}
} else {
crm_info("No inactive cluster nodes%s%s to remove from the membership "
"cache",
((criterion != NULL)? " with " : ""), pcmk__s(criterion, ""));
}
free(search.name);
free(criterion);
}
static void
count_peer(gpointer key, gpointer value, gpointer user_data)
{
unsigned int *count = user_data;
pcmk__node_status_t *node = value;
if (pcmk__cluster_is_node_active(node)) {
*count = *count + 1;
}
}
/*!
* \internal
* \brief Get the number of active cluster nodes that have been seen
*
* Remote nodes are never considered active. This guarantees that they can never
* become DC.
*
* \return Number of active nodes in the cluster node cache
*/
unsigned int
pcmk__cluster_num_active_nodes(void)
{
unsigned int count = 0;
if (crm_peer_cache != NULL) {
g_hash_table_foreach(crm_peer_cache, count_peer, &count);
}
return count;
}
static void
destroy_crm_node(gpointer data)
{
pcmk__node_status_t *node = data;
crm_trace("Destroying entry for node %" PRIu32 ": %s",
node->cluster_layer_id, node->name);
free(node->name);
free(node->state);
- free(node->uuid);
+ free(node->xml_id);
free(node->expected);
free(node->conn_host);
free(node);
}
/*!
* \internal
* \brief Initialize node caches
*/
void
pcmk__cluster_init_node_caches(void)
{
if (crm_peer_cache == NULL) {
crm_peer_cache = pcmk__strikey_table(free, destroy_crm_node);
}
if (crm_remote_peer_cache == NULL) {
crm_remote_peer_cache = pcmk__strikey_table(NULL, destroy_crm_node);
}
if (cluster_node_cib_cache == NULL) {
cluster_node_cib_cache = pcmk__strikey_table(free, destroy_crm_node);
}
}
/*!
* \internal
* \brief Initialize node caches
*/
void
pcmk__cluster_destroy_node_caches(void)
{
if (crm_peer_cache != NULL) {
crm_trace("Destroying peer cache with %d members",
g_hash_table_size(crm_peer_cache));
g_hash_table_destroy(crm_peer_cache);
crm_peer_cache = NULL;
}
if (crm_remote_peer_cache != NULL) {
crm_trace("Destroying remote peer cache with %d members",
pcmk__cluster_num_remote_nodes());
g_hash_table_destroy(crm_remote_peer_cache);
crm_remote_peer_cache = NULL;
}
if (cluster_node_cib_cache != NULL) {
crm_trace("Destroying configured cluster node cache with %d members",
g_hash_table_size(cluster_node_cib_cache));
g_hash_table_destroy(cluster_node_cib_cache);
cluster_node_cib_cache = NULL;
}
}
static void (*peer_status_callback)(enum crm_status_type, pcmk__node_status_t *,
const void *) = NULL;
/*!
* \internal
* \brief Set a client function that will be called after peer status changes
*
* \param[in] dispatch Pointer to function to use as callback
*
* \note Client callbacks should do only client-specific handling. Callbacks
* must not add or remove entries in the peer caches.
*/
void
pcmk__cluster_set_status_callback(void (*dispatch)(enum crm_status_type,
pcmk__node_status_t *,
const void *))
{
// @TODO Improve documentation of peer_status_callback
peer_status_callback = dispatch;
}
/*!
* \internal
* \brief Tell the library whether to automatically reap lost nodes
*
* If \c true (the default), calling \c crm_update_peer_proc() will also update
* the peer state to \c CRM_NODE_MEMBER or \c CRM_NODE_LOST, and updating the
* peer state will reap peers whose state changes to anything other than
* \c CRM_NODE_MEMBER.
*
* Callers should leave this enabled unless they plan to manage the cache
* separately on their own.
*
* \param[in] enable \c true to enable automatic reaping, \c false to disable
*/
void
pcmk__cluster_set_autoreap(bool enable)
{
autoreap = enable;
}
static void
dump_peer_hash(int level, const char *caller)
{
GHashTableIter iter;
const char *id = NULL;
pcmk__node_status_t *node = NULL;
g_hash_table_iter_init(&iter, crm_peer_cache);
while (g_hash_table_iter_next(&iter, (gpointer *) &id, (gpointer *) &node)) {
do_crm_log(level, "%s: Node %" PRIu32 "/%s = %p - %s",
caller, node->cluster_layer_id, node->name, node, id);
}
}
static gboolean
hash_find_by_data(gpointer key, gpointer value, gpointer user_data)
{
return value == user_data;
}
/*!
* \internal
* \brief Search cluster member node cache
*
* \param[in] id If not 0, cluster node ID to search for
* \param[in] uname If not NULL, node name to search for
* \param[in] uuid If not NULL while id is 0, node UUID instead of cluster
* node ID to search for
*
* \return Cluster node cache entry if found, otherwise NULL
*/
static pcmk__node_status_t *
search_cluster_member_cache(unsigned int id, const char *uname,
const char *uuid)
{
GHashTableIter iter;
pcmk__node_status_t *node = NULL;
pcmk__node_status_t *by_id = NULL;
pcmk__node_status_t *by_name = NULL;
CRM_ASSERT(id > 0 || uname != NULL);
pcmk__cluster_init_node_caches();
if (uname != NULL) {
g_hash_table_iter_init(&iter, crm_peer_cache);
while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &node)) {
if (pcmk__str_eq(node->name, uname, pcmk__str_casei)) {
- crm_trace("Name match: %s = %p", node->name, node);
+ crm_trace("Name match: %s", node->name);
by_name = node;
break;
}
}
}
if (id > 0) {
g_hash_table_iter_init(&iter, crm_peer_cache);
while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &node)) {
if (node->cluster_layer_id == id) {
crm_trace("ID match: %" PRIu32, node->cluster_layer_id);
by_id = node;
break;
}
}
} else if (uuid != NULL) {
g_hash_table_iter_init(&iter, crm_peer_cache);
while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &node)) {
- if (pcmk__str_eq(node->uuid, uuid, pcmk__str_casei)) {
- crm_trace("UUID match: %s = %p", node->uuid, node);
+ if (pcmk__str_eq(node->xml_id, uuid, pcmk__str_casei)) {
+ crm_trace("UUID match: %s", node->xml_id);
by_id = node;
break;
}
}
}
node = by_id; /* Good default */
if(by_id == by_name) {
/* Nothing to do if they match (both NULL counts) */
crm_trace("Consistent: %p for %u/%s", by_id, id, uname);
} else if(by_id == NULL && by_name) {
crm_trace("Only one: %p for %u/%s", by_name, id, uname);
if (id && by_name->cluster_layer_id) {
dump_peer_hash(LOG_WARNING, __func__);
crm_crit("Nodes %u and %" PRIu32 " share the same name '%s'",
id, by_name->cluster_layer_id, uname);
node = NULL; /* Create a new one */
} else {
node = by_name;
}
} else if(by_name == NULL && by_id) {
crm_trace("Only one: %p for %u/%s", by_id, id, uname);
if ((uname != NULL) && (by_id->name != NULL)) {
dump_peer_hash(LOG_WARNING, __func__);
crm_crit("Nodes '%s' and '%s' share the same cluster nodeid %u: "
"assuming '%s' is correct",
uname, by_id->name, id, uname);
}
} else if ((uname != NULL) && (by_id->name != NULL)) {
if (pcmk__str_eq(uname, by_id->name, pcmk__str_casei)) {
crm_notice("Node '%s' has changed its cluster layer ID "
"from %" PRIu32 " to %" PRIu32,
by_id->name, by_name->cluster_layer_id,
by_id->cluster_layer_id);
g_hash_table_foreach_remove(crm_peer_cache, hash_find_by_data, by_name);
} else {
crm_warn("Nodes '%s' and '%s' share the same cluster nodeid: %u %s",
by_id->name, by_name->name, id, uname);
dump_peer_hash(LOG_INFO, __func__);
crm_abort(__FILE__, __func__, __LINE__, "member weirdness", TRUE,
TRUE);
}
} else if ((id > 0) && (by_name->cluster_layer_id > 0)) {
crm_warn("Nodes %" PRIu32 " and %" PRIu32 " share the same name: '%s'",
by_id->cluster_layer_id, by_name->cluster_layer_id, uname);
} else {
/* Simple merge */
/* Only corosync-based clusters use node IDs. The functions that call
* pcmk__update_peer_state() and crm_update_peer_proc() only know
* nodeid, so 'by_id' is authoritative when merging.
*/
dump_peer_hash(LOG_DEBUG, __func__);
crm_info("Merging %p into %p", by_name, by_id);
g_hash_table_foreach_remove(crm_peer_cache, hash_find_by_data, by_name);
}
return node;
}
/*!
* \internal
* \brief Search caches for a node (cluster or Pacemaker Remote)
*
* \param[in] id If not 0, cluster node ID to search for
* \param[in] uname If not NULL, node name to search for
* \param[in] flags Group of enum pcmk__node_search_flags
*
* \return Node cache entry if found, otherwise NULL
*/
pcmk__node_status_t *
pcmk__search_node_caches(unsigned int id, const char *uname, uint32_t flags)
{
pcmk__node_status_t *node = NULL;
CRM_ASSERT(id > 0 || uname != NULL);
pcmk__cluster_init_node_caches();
if ((uname != NULL) && pcmk_is_set(flags, pcmk__node_search_remote)) {
node = g_hash_table_lookup(crm_remote_peer_cache, uname);
}
if ((node == NULL)
&& pcmk_is_set(flags, pcmk__node_search_cluster_member)) {
node = search_cluster_member_cache(id, uname, NULL);
}
if ((node == NULL) && pcmk_is_set(flags, pcmk__node_search_cluster_cib)) {
char *id_str = (id == 0)? NULL : crm_strdup_printf("%u", id);
node = find_cib_cluster_node(id_str, uname);
free(id_str);
}
return node;
}
/*!
* \internal
* \brief Purge a node from cache (both cluster and Pacemaker Remote)
*
* \param[in] node_name If not NULL, purge only nodes with this name
* \param[in] node_id If not 0, purge cluster nodes only if they have this ID
*
* \note If \p node_name is NULL and \p node_id is 0, no nodes will be purged.
* If \p node_name is not NULL and \p node_id is not 0, Pacemaker Remote
* nodes that match \p node_name will be purged, and cluster nodes that
* match both \p node_name and \p node_id will be purged.
* \note The caller must be careful not to use \p node_name after calling this
* function if it might be a pointer into a cache entry being removed.
*/
void
pcmk__purge_node_from_cache(const char *node_name, uint32_t node_id)
{
char *node_name_copy = NULL;
if ((node_name == NULL) && (node_id == 0U)) {
return;
}
// Purge from Pacemaker Remote node cache
if ((node_name != NULL)
&& (g_hash_table_lookup(crm_remote_peer_cache, node_name) != NULL)) {
/* node_name could be a pointer into the cache entry being purged,
* so reassign it to a copy before the original gets freed
*/
node_name_copy = pcmk__str_copy(node_name);
node_name = node_name_copy;
crm_trace("Purging %s from Pacemaker Remote node cache", node_name);
g_hash_table_remove(crm_remote_peer_cache, node_name);
}
pcmk__cluster_forget_cluster_node(node_id, node_name);
free(node_name_copy);
}
#if SUPPORT_COROSYNC
static guint
remove_conflicting_peer(pcmk__node_status_t *node)
{
int matches = 0;
GHashTableIter iter;
pcmk__node_status_t *existing_node = NULL;
if ((node->cluster_layer_id == 0) || (node->name == NULL)) {
return 0;
}
if (!pcmk__corosync_has_nodelist()) {
return 0;
}
g_hash_table_iter_init(&iter, crm_peer_cache);
while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &existing_node)) {
if ((existing_node->cluster_layer_id > 0)
&& (existing_node->cluster_layer_id != node->cluster_layer_id)
&& pcmk__str_eq(existing_node->name, node->name, pcmk__str_casei)) {
if (pcmk__cluster_is_node_active(existing_node)) {
continue;
}
crm_warn("Removing cached offline node %" PRIu32 "/%s which has "
"conflicting name with %" PRIu32,
existing_node->cluster_layer_id, existing_node->name,
node->cluster_layer_id);
g_hash_table_iter_remove(&iter);
matches++;
}
}
return matches;
}
#endif
/*!
* \internal
* \brief Get a cluster node cache entry, possibly creating one if not found
*
* If \c pcmk__node_search_cluster_member is set in \p flags, the return value
* is guaranteed not to be \c NULL. A new cache entry is created if one does not
* already exist.
*
* \param[in] id If not 0, cluster node ID to search for
* \param[in] uname If not NULL, node name to search for
* \param[in] uuid If not NULL while id is 0, node UUID instead of cluster
* node ID to search for
* \param[in] flags Group of enum pcmk__node_search_flags
*
* \return (Possibly newly created) cluster node cache entry
*/
/* coverity[-alloc] Memory is referenced in one or both hashtables */
pcmk__node_status_t *
pcmk__get_node(unsigned int id, const char *uname, const char *uuid,
uint32_t flags)
{
pcmk__node_status_t *node = NULL;
char *uname_lookup = NULL;
CRM_ASSERT(id > 0 || uname != NULL);
pcmk__cluster_init_node_caches();
// Check the Pacemaker Remote node cache first
if (pcmk_is_set(flags, pcmk__node_search_remote)) {
node = g_hash_table_lookup(crm_remote_peer_cache, uname);
if (node != NULL) {
return node;
}
}
if (!pcmk_is_set(flags, pcmk__node_search_cluster_member)) {
return NULL;
}
node = search_cluster_member_cache(id, uname, uuid);
/* if uname wasn't provided, and find_peer did not turn up a uname based on id.
* we need to do a lookup of the node name using the id in the cluster membership. */
if ((uname == NULL) && ((node == NULL) || (node->name == NULL))) {
uname_lookup = pcmk__cluster_node_name(id);
}
if (uname_lookup) {
uname = uname_lookup;
crm_trace("Inferred a name of '%s' for node %u", uname, id);
/* try to turn up the node one more time now that we know the uname. */
if (node == NULL) {
node = search_cluster_member_cache(id, uname, uuid);
}
}
if (node == NULL) {
char *uniqueid = crm_generate_uuid();
node = pcmk__assert_alloc(1, sizeof(pcmk__node_status_t));
crm_info("Created entry %s/%p for node %s/%u (%d total)",
uniqueid, node, uname, id, 1 + g_hash_table_size(crm_peer_cache));
g_hash_table_replace(crm_peer_cache, uniqueid, node);
}
if ((id > 0) && (uname != NULL)
&& ((node->cluster_layer_id == 0) || (node->name == NULL))) {
crm_info("Node %u is now known as %s", id, uname);
}
if ((id > 0) && (node->cluster_layer_id == 0)) {
node->cluster_layer_id = id;
}
if ((uname != NULL) && (node->name == NULL)) {
update_peer_uname(node, uname);
}
- if(node->uuid == NULL) {
+ if (node->xml_id == NULL) {
if (uuid == NULL) {
uuid = pcmk__cluster_node_uuid(node);
}
if (uuid) {
crm_info("Node %u has uuid %s", id, uuid);
} else {
crm_info("Cannot obtain a UUID for node %u/%s", id, node->name);
}
}
free(uname_lookup);
return node;
}
/*!
* \internal
* \brief Update a node's uname
*
* \param[in,out] node Node object to update
* \param[in] uname New name to set
*
* \note This function should not be called within a peer cache iteration,
* because in some cases it can remove conflicting cache entries,
* which would invalidate the iterator.
*/
static void
update_peer_uname(pcmk__node_status_t *node, const char *uname)
{
CRM_CHECK(uname != NULL,
crm_err("Bug: can't update node name without name"); return);
CRM_CHECK(node != NULL,
crm_err("Bug: can't update node name to %s without node", uname);
return);
if (pcmk__str_eq(uname, node->name, pcmk__str_casei)) {
crm_debug("Node name '%s' did not change", uname);
return;
}
for (const char *c = uname; *c; ++c) {
if ((*c >= 'A') && (*c <= 'Z')) {
crm_warn("Node names with capitals are discouraged, consider changing '%s'",
uname);
break;
}
}
pcmk__str_update(&node->name, uname);
if (peer_status_callback != NULL) {
peer_status_callback(crm_status_uname, node, NULL);
}
#if SUPPORT_COROSYNC
if ((pcmk_get_cluster_layer() == pcmk_cluster_layer_corosync)
&& !pcmk_is_set(node->flags, crm_remote_node)) {
remove_conflicting_peer(node);
}
#endif
}
/*!
* \internal
* \brief Get log-friendly string equivalent of a process flag
*
* \param[in] proc Process flag
*
* \return Log-friendly string equivalent of \p proc
*/
static inline const char *
proc2text(enum crm_proc_flag proc)
{
const char *text = "unknown";
switch (proc) {
case crm_proc_none:
text = "none";
break;
case crm_proc_cpg:
text = "corosync-cpg";
break;
}
return text;
}
/*!
* \internal
* \brief Update a node's process information (and potentially state)
*
* \param[in] source Caller's function name (for log messages)
* \param[in,out] node Node object to update
* \param[in] flag Bitmask of new process information
* \param[in] status node status (online, offline, etc.)
*
* \return NULL if any node was reaped from peer caches, value of node otherwise
*
* \note If this function returns NULL, the supplied node object was likely
* freed and should not be used again. This function should not be
* called within a cache iteration if reaping is possible, otherwise
* reaping could invalidate the iterator.
*/
pcmk__node_status_t *
crm_update_peer_proc(const char *source, pcmk__node_status_t *node,
uint32_t flag, const char *status)
{
uint32_t last = 0;
gboolean changed = FALSE;
CRM_CHECK(node != NULL, crm_err("%s: Could not set %s to %s for NULL",
source, proc2text(flag), status);
return NULL);
/* Pacemaker doesn't spawn processes on remote nodes */
if (pcmk_is_set(node->flags, crm_remote_node)) {
return node;
}
last = node->processes;
if (status == NULL) {
node->processes = flag;
if (node->processes != last) {
changed = TRUE;
}
} else if (pcmk__str_eq(status, PCMK_VALUE_ONLINE, pcmk__str_casei)) {
if ((node->processes & flag) != flag) {
node->processes = pcmk__set_flags_as(__func__, __LINE__,
LOG_TRACE, "Peer process",
node->name, node->processes,
flag, "processes");
changed = TRUE;
}
} else if (node->processes & flag) {
node->processes = pcmk__clear_flags_as(__func__, __LINE__,
LOG_TRACE, "Peer process",
node->name, node->processes,
flag, "processes");
changed = TRUE;
}
if (changed) {
if (status == NULL && flag <= crm_proc_none) {
crm_info("%s: Node %s[%" PRIu32 "] - all processes are now offline",
source, node->name, node->cluster_layer_id);
} else {
crm_info("%s: Node %s[%" PRIu32 "] - %s is now %s",
source, node->name, node->cluster_layer_id,
proc2text(flag), status);
}
if (pcmk_is_set(node->processes, crm_get_cluster_proc())) {
node->when_online = time(NULL);
} else {
node->when_online = 0;
}
/* Call the client callback first, then update the peer state,
* in case the node will be reaped
*/
if (peer_status_callback != NULL) {
peer_status_callback(crm_status_processes, node, &last);
}
/* The client callback shouldn't touch the peer caches,
* but as a safety net, bail if the peer cache was destroyed.
*/
if (crm_peer_cache == NULL) {
return NULL;
}
if (autoreap) {
const char *peer_state = NULL;
if (pcmk_is_set(node->processes, crm_get_cluster_proc())) {
peer_state = CRM_NODE_MEMBER;
} else {
peer_state = CRM_NODE_LOST;
}
node = pcmk__update_peer_state(__func__, node, peer_state, 0);
}
} else {
crm_trace("%s: Node %s[%" PRIu32 "] - %s is unchanged (%s)",
source, node->name, node->cluster_layer_id, proc2text(flag),
status);
}
return node;
}
/*!
* \internal
* \brief Update a cluster node cache entry's expected join state
*
* \param[in] source Caller's function name (for logging)
* \param[in,out] node Node to update
* \param[in] expected Node's new join state
*/
void
pcmk__update_peer_expected(const char *source, pcmk__node_status_t *node,
const char *expected)
{
char *last = NULL;
gboolean changed = FALSE;
CRM_CHECK(node != NULL, crm_err("%s: Could not set 'expected' to %s", source, expected);
return);
/* Remote nodes don't participate in joins */
if (pcmk_is_set(node->flags, crm_remote_node)) {
return;
}
last = node->expected;
if (expected != NULL && !pcmk__str_eq(node->expected, expected, pcmk__str_casei)) {
node->expected = strdup(expected);
changed = TRUE;
}
if (changed) {
crm_info("%s: Node %s[%" PRIu32 "] - expected state is now %s (was %s)",
source, node->name, node->cluster_layer_id, expected, last);
free(last);
} else {
crm_trace("%s: Node %s[%" PRIu32 "] - expected state is unchanged (%s)",
source, node->name, node->cluster_layer_id, expected);
}
}
/*!
* \internal
* \brief Update a node's state and membership information
*
* \param[in] source Caller's function name (for log messages)
* \param[in,out] node Node object to update
* \param[in] state Node's new state
* \param[in] membership Node's new membership ID
* \param[in,out] iter If not NULL, pointer to node's peer cache iterator
*
* \return NULL if any node was reaped, value of node otherwise
*
* \note If this function returns NULL, the supplied node object was likely
* freed and should not be used again. This function may be called from
* within a peer cache iteration if the iterator is supplied.
*/
static pcmk__node_status_t *
update_peer_state_iter(const char *source, pcmk__node_status_t *node,
const char *state, uint64_t membership,
GHashTableIter *iter)
{
gboolean is_member;
CRM_CHECK(node != NULL,
crm_err("Could not set state for unknown host to %s"
QB_XS " source=%s", state, source);
return NULL);
is_member = pcmk__str_eq(state, CRM_NODE_MEMBER, pcmk__str_casei);
if (is_member) {
node->when_lost = 0;
if (membership) {
node->last_seen = membership;
}
}
if (state && !pcmk__str_eq(node->state, state, pcmk__str_casei)) {
char *last = node->state;
if (is_member) {
node->when_member = time(NULL);
} else {
node->when_member = 0;
}
node->state = strdup(state);
crm_notice("Node %s state is now %s " QB_XS
" nodeid=%" PRIu32 " previous=%s source=%s",
node->name, state, node->cluster_layer_id,
pcmk__s(last, "unknown"), source);
if (peer_status_callback != NULL) {
peer_status_callback(crm_status_nstate, node, last);
}
free(last);
if (autoreap && !is_member
&& !pcmk_is_set(node->flags, crm_remote_node)) {
/* We only autoreap from the peer cache, not the remote peer cache,
* because the latter should be managed only by
* refresh_remote_nodes().
*/
if(iter) {
crm_notice("Purged 1 peer with cluster layer ID=" PRIu32
"and/or name=%s from the membership cache",
node->cluster_layer_id, node->name);
g_hash_table_iter_remove(iter);
} else {
pcmk__cluster_forget_cluster_node(node->cluster_layer_id,
node->name);
}
node = NULL;
}
} else {
crm_trace("Node %s state is unchanged (%s) " QB_XS
" nodeid=%" PRIu32 " source=%s",
node->name, state, node->cluster_layer_id, source);
}
return node;
}
/*!
* \brief Update a node's state and membership information
*
* \param[in] source Caller's function name (for log messages)
* \param[in,out] node Node object to update
* \param[in] state Node's new state
* \param[in] membership Node's new membership ID
*
* \return NULL if any node was reaped, value of node otherwise
*
* \note If this function returns NULL, the supplied node object was likely
* freed and should not be used again. This function should not be
* called within a cache iteration if reaping is possible,
* otherwise reaping could invalidate the iterator.
*/
pcmk__node_status_t *
pcmk__update_peer_state(const char *source, pcmk__node_status_t *node,
const char *state, uint64_t membership)
{
return update_peer_state_iter(source, node, state, membership, NULL);
}
/*!
* \internal
* \brief Reap all nodes from cache whose membership information does not match
*
* \param[in] membership Membership ID of nodes to keep
*/
void
pcmk__reap_unseen_nodes(uint64_t membership)
{
GHashTableIter iter;
pcmk__node_status_t *node = NULL;
crm_trace("Reaping unseen nodes...");
g_hash_table_iter_init(&iter, crm_peer_cache);
while (g_hash_table_iter_next(&iter, NULL, (gpointer *)&node)) {
if (node->last_seen != membership) {
if (node->state) {
/*
* Calling update_peer_state_iter() allows us to
* remove the node from crm_peer_cache without
* invalidating our iterator
*/
update_peer_state_iter(__func__, node, CRM_NODE_LOST,
membership, &iter);
} else {
crm_info("State of node %s[%" PRIu32 "] is still unknown",
node->name, node->cluster_layer_id);
}
}
}
}
static pcmk__node_status_t *
find_cib_cluster_node(const char *id, const char *uname)
{
GHashTableIter iter;
pcmk__node_status_t *node = NULL;
pcmk__node_status_t *by_id = NULL;
pcmk__node_status_t *by_name = NULL;
if (uname) {
g_hash_table_iter_init(&iter, cluster_node_cib_cache);
while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &node)) {
if (pcmk__str_eq(node->name, uname, pcmk__str_casei)) {
crm_trace("Name match: %s = %p", node->name, node);
by_name = node;
break;
}
}
}
if (id) {
g_hash_table_iter_init(&iter, cluster_node_cib_cache);
while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &node)) {
- if(strcasecmp(node->uuid, id) == 0) {
+ if (pcmk__str_eq(node->xml_id, id, pcmk__str_casei)) {
crm_trace("ID match: %s= %p", id, node);
by_id = node;
break;
}
}
}
node = by_id; /* Good default */
if (by_id == by_name) {
/* Nothing to do if they match (both NULL counts) */
crm_trace("Consistent: %p for %s/%s", by_id, id, uname);
} else if (by_id == NULL && by_name) {
crm_trace("Only one: %p for %s/%s", by_name, id, uname);
if (id) {
node = NULL;
} else {
node = by_name;
}
} else if (by_name == NULL && by_id) {
crm_trace("Only one: %p for %s/%s", by_id, id, uname);
if (uname) {
node = NULL;
}
} else if ((uname != NULL) && (by_id->name != NULL)
&& pcmk__str_eq(uname, by_id->name, pcmk__str_casei)) {
/* Multiple nodes have the same uname in the CIB.
* Return by_id. */
- } else if (id && by_name->uuid
- && pcmk__str_eq(id, by_name->uuid, pcmk__str_casei)) {
+ } else if ((id != NULL) && (by_name->xml_id != NULL)
+ && pcmk__str_eq(id, by_name->xml_id, pcmk__str_casei)) {
/* Multiple nodes have the same id in the CIB.
* Return by_name. */
node = by_name;
} else {
node = NULL;
}
if (node == NULL) {
crm_debug("Couldn't find node%s%s%s%s",
id? " " : "",
id? id : "",
uname? " with name " : "",
uname? uname : "");
}
return node;
}
static void
cluster_node_cib_cache_refresh_helper(xmlNode *xml_node, void *user_data)
{
const char *id = crm_element_value(xml_node, PCMK_XA_ID);
const char *uname = crm_element_value(xml_node, PCMK_XA_UNAME);
pcmk__node_status_t * node = NULL;
CRM_CHECK(id != NULL && uname !=NULL, return);
node = find_cib_cluster_node(id, uname);
if (node == NULL) {
char *uniqueid = crm_generate_uuid();
node = pcmk__assert_alloc(1, sizeof(pcmk__node_status_t));
node->name = pcmk__str_copy(uname);
- node->uuid = pcmk__str_copy(id);
+ node->xml_id = pcmk__str_copy(id);
g_hash_table_replace(cluster_node_cib_cache, uniqueid, node);
} else if (pcmk_is_set(node->flags, crm_node_dirty)) {
pcmk__str_update(&node->name, uname);
/* Node is in cache and hasn't been updated already, so mark it clean */
clear_peer_flags(node, crm_node_dirty);
}
}
static void
refresh_cluster_node_cib_cache(xmlNode *cib)
{
pcmk__cluster_init_node_caches();
g_hash_table_foreach(cluster_node_cib_cache, mark_dirty, NULL);
crm_foreach_xpath_result(cib, PCMK__XP_MEMBER_NODE_CONFIG,
cluster_node_cib_cache_refresh_helper, NULL);
// Remove all old cache entries that weren't seen in the CIB
g_hash_table_foreach_remove(cluster_node_cib_cache, is_dirty, NULL);
}
void
pcmk__refresh_node_caches_from_cib(xmlNode *cib)
{
refresh_remote_nodes(cib);
refresh_cluster_node_cib_cache(cib);
}
// Deprecated functions kept only for backward API compatibility
// LCOV_EXCL_START
#include <crm/cluster/compat.h>
void
crm_peer_init(void)
{
pcmk__cluster_init_node_caches();
}
// LCOV_EXCL_STOP
// End deprecated API

File Metadata

Mime Type
text/x-diff
Expires
Thu, Jul 10, 3:19 AM (17 h, 18 m)
Storage Engine
blob
Storage Format
Raw Data
Storage Handle
1988588
Default Alt Text
(473 KB)

Event Timeline