diff --git a/daemons/controld/controld_join_dc.c b/daemons/controld/controld_join_dc.c
index 1118d71a23..98053a9760 100644
--- a/daemons/controld/controld_join_dc.c
+++ b/daemons/controld/controld_join_dc.c
@@ -1,1031 +1,1031 @@
/*
* Copyright 2004-2023 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/msg_xml.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, crm_node_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 "
CRM_XS " nodeid=%u source=%s",
node->uname, current_join_id, crm_join_phase_str(last),
node->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) "
CRM_XS " nodeid=%u source=%s",
node->uname, current_join_id, crm_join_phase_str(phase),
crm_join_phase_str(last), node->id, source);
} else {
crm_warn("Rejecting join-%d phase update for node %s because "
"can't go from %s to %s " CRM_XS " nodeid=%u source=%s",
current_join_id, node->uname, crm_join_phase_str(last),
crm_join_phase_str(phase), node->id, source);
}
}
static void
start_join_round(void)
{
GHashTableIter iter;
crm_node_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) {
free_xml(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, F_CRM_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, F_CRM_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;
crm_node_t *member = (crm_node_t *)value;
CRM_ASSERT(member != NULL);
if (crm_is_peer_active(member) == FALSE) {
crm_info("Not making join-%d offer to inactive node %s",
current_join_id,
(member->uname? member->uname : "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->uname == NULL) {
crm_info("Not making join-%d offer to node uuid %s with unknown name",
current_join_id, member->uuid);
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->uname,
crm_join_phase_str(member->join));
return;
}
crm_update_peer_join(__func__, (crm_node_t*)member, crm_join_none);
offer = create_dc_message(CRM_OP_JOIN_OFFER, member->uname);
// Advertise our feature set so the joining node can bail if not compatible
crm_xml_add(offer, XML_ATTR_CRM_VERSION, CRM_FEATURE_SET);
crm_info("Sending join-%d offer to %s", current_join_id, member->uname);
send_cluster_message(member, crm_msg_crmd, offer, TRUE);
free_xml(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)
{
crm_node_t *member;
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, F_CRM_HOST_FROM);
if (join_to == NULL) {
crm_err("Can't make join-%d offer to unknown node", current_join_id);
return;
}
member = crm_get_peer(0, join_to);
/* 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 = crm_get_peer(0, controld_globals.our_nodename);
join_make_offer(NULL, member, NULL);
}
/* This was a genuine join request; cancel any existing transition and
* invoke the scheduler.
*/
abort_transition(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, F_CRM_HOST_FROM);
const char *ref = crm_element_value(join_ack->msg, F_CRM_REFERENCE);
const char *join_version = crm_element_value(join_ack->msg,
XML_ATTR_CRM_VERSION);
crm_node_t *join_node = NULL;
if (join_from == NULL) {
crm_err("Ignoring invalid join request without node name");
return;
}
join_node = crm_get_peer(0, join_from);
crm_element_value_int(join_ack->msg, F_CRM_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[] = {
XML_ATTR_GENERATION_ADMIN,
XML_ATTR_GENERATION,
XML_ATTR_NUMUPDATES,
};
for (lpc = 0; cmp == 0 && lpc < PCMK__NELEM(attributes); lpc++) {
cmp = compare_int_fields(max_generation_xml, generation, attributes[lpc]);
}
}
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 " CRM_XS " ref=%s",
join_id, join_from, value, ref);
ack_nack_bool = FALSE;
} else if (!crm_is_peer_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 "
CRM_XS " ref=%s", join_id, join_from, ref);
} else {
crm_err("Rejecting join-%d request from inactive node %s "
CRM_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 " CRM_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) " CRM_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,
XML_ATTR_VALIDATION);
if (get_schema_version(validation) < 0) {
crm_err("Rejecting join-%d request from %s (with first CIB "
"generation) due to unknown schema version %s "
CRM_XS " ref=%s",
join_id, join_from, validation, ref);
ack_nack_bool = FALSE;
} else {
crm_debug("Accepting join-%d request from %s (with first CIB "
"generation) " CRM_XS " ref=%s",
join_id, join_from, ref);
max_generation_xml = copy_xml(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,
XML_ATTR_VALIDATION);
if (get_schema_version(validation) < 0) {
crm_err("Rejecting join-%d request from %s (with better CIB "
"generation than current best from %s) due to unknown "
"schema version %s " CRM_XS " ref=%s",
join_id, join_from, max_generation_from, validation, ref);
ack_nack_bool = FALSE;
} else {
crm_debug("Accepting join-%d request from %s (with better CIB "
"generation than current best from %s) " CRM_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");
free_xml(max_generation_xml);
max_generation_xml = copy_xml(join_ack->xml);
pcmk__str_update(&max_generation_from, join_from);
}
} else {
crm_debug("Accepting join-%d request from %s " CRM_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
pcmk__str_update(&sync_from, 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
pcmk__str_update(&sync_from, 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;
free_xml(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, F_CRM_TASK);
char *join_from = crm_element_value_copy(join_ack->msg, F_CRM_HOST_FROM);
crm_node_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, F_CRM_JOIN_ID, &join_id) != 0) {
crm_warn("Ignoring join confirmation from %s without valid join ID",
join_from);
goto done;
}
peer = crm_get_peer(0, join_from);
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, crm_join_phase_str(peer->join),
crm_join_phase_str(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_scope_local
|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, XML_CIB_TAG_STATUS, state,
cib_scope_local|cib_can_create|cib_transaction);
free_xml(execd_state);
// Commit the transaction
rc = cib->cmds->end_transaction(cib, true, cib_scope_local);
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;
crm_node_t *join_node = value;
const char *join_to = join_node->uname;
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,
crm_join_phase_str(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 = create_xml_node(NULL, XML_CIB_TAG_NODE);
- set_uuid(tmp1, XML_ATTR_ID, join_node);
+ crm_xml_add(tmp1, XML_ATTR_ID, crm_peer_uuid(join_node));
crm_xml_add(tmp1, XML_ATTR_UNAME, join_to);
fsa_cib_anon_update(XML_CIB_TAG_NODES, tmp1);
free_xml(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 = crm_get_peer(0, join_to);
if (!crm_is_peer_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 (crm_remote_peer_cache_size() != 0) {
GHashTableIter iter;
crm_node_t *node = NULL;
xmlNode *remotes = create_xml_node(acknak, XML_CIB_TAG_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 = create_xml_node(remotes, XML_CIB_TAG_NODE);
pcmk__xe_set_props(remote,
XML_ATTR_ID, node->uname,
XML_CIB_TAG_STATE, node->state,
PCMK__XA_CONN_HOST, node->conn_host,
NULL);
}
}
}
send_cluster_message(join_node, crm_msg_crmd, acknak, TRUE);
free_xml(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 "
CRM_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 "
CRM_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 "
CRM_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 "
CRM_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 "
CRM_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 "
CRM_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 " CRM_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(crm_have_quorum, TRUE);
}
int crmd_join_phase_count(enum crm_join_phase phase)
{
int count = 0;
crm_node_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)
{
crm_node_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->uname,
crm_join_phase_str(peer->join));
}
}
diff --git a/daemons/controld/controld_membership.c b/daemons/controld/controld_membership.c
index 9eb20afc7a..f25d1e9e6e 100644
--- a/daemons/controld/controld_membership.c
+++ b/daemons/controld/controld_membership.c
@@ -1,471 +1,469 @@
/*
* Copyright 2004-2023 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/msg_xml.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);
static void
reap_dead_nodes(gpointer key, gpointer value, gpointer user_data)
{
crm_node_t *node = value;
if (crm_is_peer_active(node) == FALSE) {
crm_update_peer_join(__func__, node, crm_join_none);
if(node && node->uname) {
if (pcmk__str_eq(controld_globals.our_nodename, node->uname,
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->uname, controld_globals.dc_name,
pcmk__str_casei)) {
crm_warn("Our DC node (%s) left the cluster", node->uname);
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)) {
fail_incompletable_actions(controld_globals.transition_graph,
node->uuid);
}
}
}
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);
send_cluster_message(NULL, crm_msg_crmd, no_op, FALSE);
free_xml(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(crm_node_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->uname);
return NULL;
}
node_state = create_xml_node(parent, XML_CIB_TAG_STATE);
if (pcmk_is_set(node->flags, crm_remote_node)) {
pcmk__xe_set_bool_attr(node_state, XML_NODE_IS_REMOTE, true);
}
- set_uuid(node_state, XML_ATTR_ID, node);
-
- if (crm_element_value(node_state, XML_ATTR_ID) == NULL) {
- crm_info("Node update for %s cancelled: no id", node->uname);
+ if (crm_xml_add(node_state, XML_ATTR_ID, crm_peer_uuid(node)) == NULL) {
+ crm_info("Node update for %s cancelled: no ID", node->uname);
free_xml(node_state);
return NULL;
}
crm_xml_add(node_state, XML_ATTR_UNAME, node->uname);
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 = OFFLINESTATUS;
if (pcmk_is_set(node->processes, crm_get_cluster_proc())) {
value = ONLINESTATUS;
}
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, XML_ATTR_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, XML_CIB_TAG_NODE)) {
node_xml = output;
} else {
node_xml = pcmk__xml_first_child(output);
}
for (; node_xml != NULL; node_xml = pcmk__xml_next(node_xml)) {
const char *node_uuid = NULL;
const char *node_uname = NULL;
GHashTableIter iter;
crm_node_t *node = NULL;
gboolean known = FALSE;
if (!pcmk__xe_is(node_xml, XML_CIB_TAG_NODE)) {
continue;
}
node_uuid = crm_element_value(node_xml, XML_ATTR_ID);
node_uname = crm_element_value(node_xml, XML_ATTR_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->uuid
&& pcmk__str_eq(node->uuid, node_uuid, pcmk__str_casei)
&& node->uname
&& pcmk__str_eq(node->uname, 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, XML_CIB_TAG_NODES,
node_xml, cib_scope_local);
fsa_register_cib_callback(delete_call_id, strdup(node_uuid),
remove_conflicting_node_callback);
node_state_xml = create_xml_node(NULL, XML_CIB_TAG_STATE);
crm_xml_add(node_state_xml, XML_ATTR_ID, node_uuid);
crm_xml_add(node_state_xml, XML_ATTR_UNAME, node_uname);
delete_call_id = cib_conn->cmds->remove(cib_conn,
XML_CIB_TAG_STATUS,
node_state_xml,
cib_scope_local);
fsa_register_cib_callback(delete_call_id, strdup(node_uuid),
remove_conflicting_node_callback);
free_xml(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 = create_xml_node(NULL, XML_CIB_TAG_NODES);
#if SUPPORT_COROSYNC
if (!pcmk_is_set(flags, node_update_quick) && is_corosync_cluster()) {
from_hashtable = pcmk__corosync_add_nodes(node_list);
}
#endif
if (from_hashtable) {
GHashTableIter iter;
crm_node_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->uname != NULL)) {
crm_trace("Creating node entry for %s/%s", node->uname, node->uuid);
if (xpath == NULL) {
xpath = g_string_sized_new(512);
} else {
g_string_truncate(xpath, 0);
}
/* We need both to be valid */
new_node = create_xml_node(node_list, XML_CIB_TAG_NODE);
crm_xml_add(new_node, XML_ATTR_ID, node->uuid);
crm_xml_add(new_node, XML_ATTR_UNAME, node->uname);
/* Search and remove unknown nodes with the conflicting uname from CIB */
pcmk__g_strcat(xpath,
"/" XML_TAG_CIB "/" XML_CIB_TAG_CONFIGURATION
"/" XML_CIB_TAG_NODES "/" XML_CIB_TAG_NODE
"[@" XML_ATTR_UNAME "='", node->uname, "']"
"[@" XML_ATTR_ID "!='", node->uuid, "']", NULL);
call_id = cib_conn->cmds->query(cib_conn,
(const char *) xpath->str,
NULL,
cib_scope_local|cib_xpath);
fsa_register_cib_callback(call_id, strdup(node->uuid),
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(XML_CIB_TAG_NODES, node_list, cib_scope_local,
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;
crm_node_t *node = NULL;
free_xml(node_list);
node_list = create_xml_node(NULL, XML_CIB_TAG_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(XML_CIB_TAG_STATUS, node_list, cib_scope_local,
crmd_node_update_complete);
}
free_xml(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 = create_xml_node(NULL, XML_TAG_CIB);
crm_xml_add_int(update, XML_ATTR_HAVE_QUORUM, quorum);
crm_xml_add(update, XML_ATTR_DC_UUID, controld_globals.our_uuid);
crm_debug("Updating quorum status to %s", pcmk__btoa(quorum));
controld_update_cib(XML_TAG_CIB, update, cib_scope_local,
cib_quorum_update_complete);
free_xml(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(INFINITY, pcmk__graph_restart, "Quorum gained",
5000);
} else {
abort_transition(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/include/crm/cluster.h b/include/crm/cluster.h
index 9717b826d8..b61fd70b04 100644
--- a/include/crm/cluster.h
+++ b/include/crm/cluster.h
@@ -1,239 +1,238 @@
/*
* Copyright 2004-2023 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__H
# define PCMK__CRM_CLUSTER__H
# include <stdint.h> // uint32_t, uint64_t
# include <glib.h> // gboolean, GHashTable
# include <libxml/tree.h> // xmlNode
# include <crm/common/xml.h>
# include <crm/common/util.h>
#ifdef __cplusplus
extern "C" {
#endif
# if SUPPORT_COROSYNC
# include <corosync/cpg.h>
# endif
extern gboolean crm_have_quorum;
extern GHashTable *crm_peer_cache;
extern GHashTable *crm_remote_peer_cache;
extern unsigned long long crm_peer_seq;
#define CRM_NODE_LOST "lost"
#define CRM_NODE_MEMBER "member"
enum crm_join_phase {
/* @COMPAT: crm_join_nack_quiet can be replaced by crm_node_t:user_data
* at a compatibility break.
*/
//! Not allowed to join, but don't send a nack message
crm_join_nack_quiet = -2,
crm_join_nack = -1,
crm_join_none = 0,
crm_join_welcomed = 1,
crm_join_integrated = 2,
crm_join_finalized = 3,
crm_join_confirmed = 4,
};
enum crm_node_flags {
/* node is not a cluster node and should not be considered for cluster membership */
crm_remote_node = 0x0001,
/* node's cache entry is dirty */
crm_node_dirty = 0x0010,
};
typedef struct crm_peer_node_s {
char *uname; // Node name as known to cluster
char *uuid; // Node UUID to ensure uniqueness
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.
*/
// Currently only needed by corosync stack
uint32_t id; // Node ID
time_t when_lost; // When CPG membership was last lost
// 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
} crm_node_t;
void crm_peer_init(void);
void crm_peer_destroy(void);
typedef struct crm_cluster_s {
char *uuid;
char *uname;
uint32_t nodeid;
void (*destroy) (gpointer);
# if SUPPORT_COROSYNC
/* @TODO When we can break public API compatibility, make these members a
* separate struct and use void *cluster_data here instead, to abstract the
* cluster layer further.
*/
struct cpg_name group;
cpg_callbacks_t cpg;
cpg_handle_t cpg_handle;
# endif
} crm_cluster_t;
gboolean crm_cluster_connect(crm_cluster_t *cluster);
void crm_cluster_disconnect(crm_cluster_t *cluster);
crm_cluster_t *pcmk_cluster_new(void);
void pcmk_cluster_free(crm_cluster_t *cluster);
enum crm_ais_msg_class {
crm_class_cluster = 0,
};
enum crm_ais_msg_types {
crm_msg_none = 0,
crm_msg_ais = 1,
crm_msg_lrmd = 2,
crm_msg_cib = 3,
crm_msg_crmd = 4,
crm_msg_attrd = 5,
crm_msg_stonithd = 6,
crm_msg_te = 7,
crm_msg_pe = 8,
crm_msg_stonith_ng = 9,
};
/* used with crm_get_peer_full */
enum crm_get_peer_flags {
CRM_GET_PEER_CLUSTER = 0x0001,
CRM_GET_PEER_REMOTE = 0x0002,
CRM_GET_PEER_ANY = CRM_GET_PEER_CLUSTER|CRM_GET_PEER_REMOTE,
};
gboolean send_cluster_message(const crm_node_t *node,
enum crm_ais_msg_types service,
const xmlNode *data, gboolean ordered);
int crm_remote_peer_cache_size(void);
/* Initialize and refresh the remote peer cache from a cib config */
void crm_remote_peer_cache_refresh(xmlNode *cib);
crm_node_t *crm_remote_peer_get(const char *node_name);
void crm_remote_peer_cache_remove(const char *node_name);
/* allows filtering of remote and cluster nodes using crm_get_peer_flags */
crm_node_t *crm_get_peer_full(unsigned int id, const char *uname, int flags);
/* only searches cluster nodes */
crm_node_t *crm_get_peer(unsigned int id, const char *uname);
guint crm_active_peers(void);
gboolean crm_is_peer_active(const crm_node_t * node);
guint reap_crm_member(uint32_t id, const char *name);
# if SUPPORT_COROSYNC
uint32_t get_local_nodeid(cpg_handle_t handle);
gboolean cluster_connect_cpg(crm_cluster_t *cluster);
void cluster_disconnect_cpg(crm_cluster_t * cluster);
void pcmk_cpg_membership(cpg_handle_t handle,
const struct cpg_name *groupName,
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);
gboolean crm_is_corosync_peer_active(const crm_node_t * node);
gboolean send_cluster_text(enum crm_ais_msg_class msg_class, const char *data,
gboolean local, const crm_node_t *node,
enum crm_ais_msg_types dest);
char *pcmk_message_common_cs(cpg_handle_t handle, uint32_t nodeid, uint32_t pid, void *msg,
uint32_t *kind, const char **from);
# endif
const char *crm_peer_uuid(crm_node_t *node);
const char *crm_peer_uname(const char *uuid);
-void set_uuid(xmlNode *xml, const char *attr, crm_node_t *node);
enum crm_status_type {
crm_status_uname,
crm_status_nstate,
crm_status_processes,
};
enum crm_ais_msg_types text2msg_type(const char *text);
void crm_set_status_callback(void (*dispatch) (enum crm_status_type, crm_node_t *, const void *));
void crm_set_autoreap(gboolean autoreap);
enum cluster_type_e {
pcmk_cluster_unknown = 0x0001,
pcmk_cluster_invalid = 0x0002,
// 0x0004 was heartbeat
// 0x0010 was corosync 1 with plugin
pcmk_cluster_corosync = 0x0020,
// 0x0040 was corosync 1 with CMAN
};
enum cluster_type_e get_cluster_type(void);
const char *name_for_cluster_type(enum cluster_type_e type);
gboolean is_corosync_cluster(void);
const char *get_local_node_name(void);
char *get_node_name(uint32_t nodeid);
/*!
* \brief Get log-friendly string equivalent of a join phase
*
* \param[in] phase Join phase
*
* \return Log-friendly string equivalent of \p phase
*/
static inline const char *
crm_join_phase_str(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";
}
}
#if !defined(PCMK_ALLOW_DEPRECATED) || (PCMK_ALLOW_DEPRECATED == 1)
#include <crm/cluster/compat.h>
#endif
#ifdef __cplusplus
}
#endif
#endif
diff --git a/include/crm/cluster/compat.h b/include/crm/cluster/compat.h
index 9bf14eefcb..89a03fd09d 100644
--- a/include/crm/cluster/compat.h
+++ b/include/crm/cluster/compat.h
@@ -1,37 +1,43 @@
/*
- * Copyright 2004-2021 the Pacemaker project contributors
+ * Copyright 2004-2023 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_COMPAT__H
# define PCMK__CRM_CLUSTER_COMPAT__H
+#include <libxml/tree.h> // xmlNode
+#include <crm/cluster.h> // crm_node_t
+
#ifdef __cplusplus
extern "C" {
#endif
/**
* \file
* \brief Deprecated Pacemaker cluster API
* \ingroup cluster
* \deprecated Do not include this header directly. The cluster APIs in this
* header, and the header itself, will be removed in a future
* release.
*/
// \deprecated Use stonith_api_kick() from libstonithd instead
int crm_terminate_member(int nodeid, const char *uname, void *unused);
// \deprecated Use stonith_api_kick() from libstonithd instead
int crm_terminate_member_no_mainloop(int nodeid, const char *uname,
int *connection);
+// \deprecated Use crm_xml_add(xml, attr, crm_peer_uuid(node)) instead
+void set_uuid(xmlNode *xml, const char *attr, crm_node_t *node);
+
#ifdef __cplusplus
}
#endif
#endif // PCMK_CLUSTER_COMPAT__H
diff --git a/lib/cluster/cluster.c b/lib/cluster/cluster.c
index 009239ad10..f2cd428b57 100644
--- a/lib/cluster/cluster.c
+++ b/lib/cluster/cluster.c
@@ -1,405 +1,406 @@
/*
* Copyright 2004-2023 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 <stdio.h>
#include <unistd.h>
#include <string.h>
#include <stdlib.h>
#include <time.h>
#include <sys/param.h>
#include <sys/types.h>
#include <crm/crm.h>
#include <crm/msg_xml.h>
#include <crm/common/ipc.h>
#include <crm/cluster/internal.h>
#include "crmcluster_private.h"
CRM_TRACE_INIT_DATA(cluster);
/*!
* \brief Get (and set if needed) a node's UUID
*
* \param[in,out] peer Node to check
*
* \return Node UUID of \p peer, or NULL if unknown
*/
const char *
crm_peer_uuid(crm_node_t *peer)
{
char *uuid = NULL;
// Check simple cases first, to avoid any calls that might block
if (peer == NULL) {
return NULL;
}
if (peer->uuid != NULL) {
return peer->uuid;
}
switch (get_cluster_type()) {
case pcmk_cluster_corosync:
#if SUPPORT_COROSYNC
uuid = pcmk__corosync_uuid(peer);
#endif
break;
case pcmk_cluster_unknown:
case pcmk_cluster_invalid:
crm_err("Unsupported cluster type");
break;
}
peer->uuid = uuid;
return peer->uuid;
}
/*!
* \brief Connect to the cluster layer
*
* \param[in,out] Initialized cluster object to connect
*
* \return TRUE on success, otherwise FALSE
*/
gboolean
crm_cluster_connect(crm_cluster_t *cluster)
{
enum cluster_type_e type = get_cluster_type();
crm_notice("Connecting to %s cluster infrastructure",
name_for_cluster_type(type));
switch (type) {
case pcmk_cluster_corosync:
#if SUPPORT_COROSYNC
crm_peer_init();
return pcmk__corosync_connect(cluster);
#else
break;
#endif // SUPPORT_COROSYNC
default:
break;
}
return FALSE;
}
/*!
* \brief Disconnect from the cluster layer
*
* \param[in,out] cluster Cluster object to disconnect
*/
void
crm_cluster_disconnect(crm_cluster_t *cluster)
{
enum cluster_type_e type = get_cluster_type();
crm_info("Disconnecting from %s cluster infrastructure",
name_for_cluster_type(type));
switch (type) {
case pcmk_cluster_corosync:
#if SUPPORT_COROSYNC
crm_peer_destroy();
pcmk__corosync_disconnect(cluster);
#endif // SUPPORT_COROSYNC
break;
default:
break;
}
}
/*!
* \brief Allocate a new \p crm_cluster_t object
*
* \return A newly allocated \p crm_cluster_t object (guaranteed not \p NULL)
* \note The caller is responsible for freeing the return value using
* \p pcmk_cluster_free().
*/
crm_cluster_t *
pcmk_cluster_new(void)
{
crm_cluster_t *cluster = calloc(1, sizeof(crm_cluster_t));
CRM_ASSERT(cluster != NULL);
return cluster;
}
/*!
* \brief Free a \p crm_cluster_t object and its dynamically allocated members
*
* \param[in,out] cluster Cluster object to free
*/
void
pcmk_cluster_free(crm_cluster_t *cluster)
{
if (cluster == NULL) {
return;
}
free(cluster->uuid);
free(cluster->uname);
free(cluster);
}
/*!
* \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
* \param[in] ordered Ignored for currently supported messaging layers
*
* \return TRUE on success, otherwise FALSE
*/
gboolean
send_cluster_message(const crm_node_t *node, enum crm_ais_msg_types service,
const xmlNode *data, gboolean ordered)
{
switch (get_cluster_type()) {
case pcmk_cluster_corosync:
#if SUPPORT_COROSYNC
return pcmk__cpg_send_xml(data, node, service);
#endif
break;
default:
break;
}
return FALSE;
}
/*!
* \brief Get the local node's name
*
* \return Local node's name
* \note This will fatally exit if local node name cannot be known.
*/
const char *
get_local_node_name(void)
{
static char *name = NULL;
if (name == NULL) {
name = get_node_name(0);
}
return name;
}
/*!
* \brief Get the node name corresponding to a cluster node ID
*
* \param[in] nodeid Node ID to check (or 0 for local node)
*
* \return Node name corresponding to \p nodeid
* \note This will fatally exit if \p nodeid is 0 and local node name cannot be
* known.
*/
char *
get_node_name(uint32_t nodeid)
{
char *name = NULL;
enum cluster_type_e stack = get_cluster_type();
switch (stack) {
case pcmk_cluster_corosync:
#if SUPPORT_COROSYNC
name = pcmk__corosync_name(0, nodeid);
break;
#endif // SUPPORT_COROSYNC
default:
crm_err("Unknown cluster type: %s (%d)", name_for_cluster_type(stack), stack);
}
if ((name == NULL) && (nodeid == 0)) {
name = pcmk_hostname();
if (name == NULL) {
// @TODO Maybe let the caller decide what to do
crm_err("Could not obtain the local %s node name",
name_for_cluster_type(stack));
crm_exit(CRM_EX_FATAL);
}
crm_notice("Defaulting to uname -n for the local %s node name",
name_for_cluster_type(stack));
}
if (name == NULL) {
crm_notice("Could not obtain a node name for %s node with id %u",
name_for_cluster_type(stack), nodeid);
}
return name;
}
/*!
* \brief Get the node name corresponding to a node UUID
*
* \param[in] uuid UUID of desired node
*
* \return name of desired node
*
* \note This relies on the remote peer cache being populated with all
* remote nodes in the cluster, so callers should maintain that cache.
*/
const char *
crm_peer_uname(const char *uuid)
{
GHashTableIter iter;
crm_node_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;
}
/* avoid blocking calls where possible */
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 (node->uname != NULL) {
return node->uname;
}
break;
}
}
node = NULL;
if (is_corosync_cluster()) {
long long id;
if ((pcmk__scan_ll(uuid, &id, 0LL) != pcmk_rc_ok)
|| (id < 1LL) || (id > UINT32_MAX)) {
crm_err("Invalid Corosync node ID '%s'", uuid);
return NULL;
}
node = pcmk__search_cluster_node_cache((uint32_t) id, NULL, NULL);
if (node != NULL) {
crm_info("Setting uuid for node %s[%u] to %s",
node->uname, node->id, uuid);
node->uuid = strdup(uuid);
return node->uname;
}
return NULL;
}
return NULL;
}
-/*!
- * \brief Add a node's UUID as an XML attribute
- *
- * \param[in,out] xml XML element to add UUID to
- * \param[in] attr XML attribute name to set
- * \param[in,out] node Node whose UUID should be used as attribute value
- */
-void
-set_uuid(xmlNode *xml, const char *attr, crm_node_t *node)
-{
- crm_xml_add(xml, attr, crm_peer_uuid(node));
-}
-
/*!
* \brief Get a log-friendly string equivalent of a cluster type
*
* \param[in] type Cluster type
*
* \return Log-friendly string corresponding to \p type
*/
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";
}
/*!
* \brief Get (and validate) the local cluster type
*
* \return Local cluster type
* \note This will fatally exit if the local cluster type is invalid.
*/
enum cluster_type_e
get_cluster_type(void)
{
bool detected = false;
const char *cluster = NULL;
static enum cluster_type_e cluster_type = pcmk_cluster_unknown;
/* Return the previous calculation, if any */
if (cluster_type != pcmk_cluster_unknown) {
return cluster_type;
}
cluster = pcmk__env_option(PCMK__ENV_CLUSTER_TYPE);
#if SUPPORT_COROSYNC
/* If nothing is defined in the environment, try corosync (if supported) */
if (cluster == NULL) {
crm_debug("Testing with Corosync");
cluster_type = pcmk__corosync_detect();
if (cluster_type != pcmk_cluster_unknown) {
detected = true;
goto done;
}
}
#endif
/* Something was defined in the environment, test it against what we support */
crm_info("Verifying cluster type: '%s'",
((cluster == NULL)? "-unspecified-" : cluster));
if (cluster == NULL) {
#if SUPPORT_COROSYNC
} else if (pcmk__str_eq(cluster, "corosync", pcmk__str_casei)) {
cluster_type = pcmk_cluster_corosync;
#endif
} else {
cluster_type = pcmk_cluster_invalid;
goto done; /* Keep the compiler happy when no stacks are supported */
}
done:
if (cluster_type == pcmk_cluster_unknown) {
crm_notice("Could not determine the current cluster type");
} else if (cluster_type == pcmk_cluster_invalid) {
crm_notice("This installation does not support the '%s' cluster infrastructure: terminating.",
cluster);
crm_exit(CRM_EX_FATAL);
} else {
crm_info("%s an active '%s' cluster",
(detected? "Detected" : "Assuming"),
name_for_cluster_type(cluster_type));
}
return cluster_type;
}
/*!
* \brief Check whether the local cluster is a Corosync cluster
*
* \return TRUE if the local cluster is a Corosync cluster, otherwise FALSE
*/
gboolean
is_corosync_cluster(void)
{
return get_cluster_type() == pcmk_cluster_corosync;
}
+
+// Deprecated functions kept only for backward API compatibility
+// LCOV_EXCL_START
+
+#include <crm/cluster/compat.h>
+
+void
+set_uuid(xmlNode *xml, const char *attr, crm_node_t *node)
+{
+ crm_xml_add(xml, attr, crm_peer_uuid(node));
+}
+
+// LCOV_EXCL_STOP
+// End deprecated API