/*!
* \internal
* \brief Return host's hash table entry (creating one if needed)
*
* \param[in] values Hash table of values
* \param[in] host Name of peer to look up
* \param[in] xml XML describing the attribute
*
* \return Pointer to new or existing hash table entry
*/
static attribute_value_t *
attrd_lookup_or_create_value(GHashTable *values, const char *host, xmlNode *xml)
{
attribute_value_t *v = g_hash_table_lookup(values, host);
int is_remote = 0;
crm_element_value_int(xml, F_ATTRD_IS_REMOTE, &is_remote);
if (is_remote) {
/* If we previously assumed this node was an unseen cluster node,
* remove its entry from the cluster peer cache.
*/
crm_node_t *dup = crm_find_peer(0, host);
if (dup && (dup->uuid == NULL)) {
reap_crm_member(0, host);
}
/* Ensure this host is in the remote peer cache */
crm_remote_peer_cache_add(host);
}
if (v == NULL) {
v = calloc(1, sizeof(attribute_value_t));
CRM_ASSERT(v != NULL);
v->nodename = strdup(host);
CRM_ASSERT(v->nodename != NULL);
v->is_remote = is_remote;
g_hash_table_replace(values, v->nodename, v);
}
return(v);
}
/*!
* \internal
* \brief Respond to a client peer-remove request (i.e. propagate to all peers)
*
* \param[in] client_name Name of client that made request (for log messages)
* \param[in] xml Root of request XML
*
* \return void
*/
void
attrd_client_peer_remove(const char *client_name, xmlNode *xml)
{
// Host and ID are not used in combination, rather host has precedence
const char *host = crm_element_value(xml, F_ATTRD_HOST);
char *host_alloc = NULL;
if (host == NULL) {
int nodeid;
crm_element_value_int(xml, F_ATTRD_HOST_ID, &nodeid);
if (nodeid > 0) {
crm_node_t *node = crm_find_peer(nodeid, NULL);
char *host_alloc = NULL;
if (node && node->uname) {
// Use cached name if available
host = node->uname;
} else {
// Otherwise ask cluster layer
host_alloc = get_node_name(nodeid);
host = host_alloc;
}
crm_xml_add(xml, F_ATTRD_HOST, host);
}
}
if (host) {
crm_info("Client %s is requesting all values for %s be removed",
client_name, host);
send_attrd_message(NULL, xml); /* ends up at attrd_peer_message() */
free(host_alloc);
} else {
crm_info("Ignoring request by client %s to remove all peer values without specifying peer",
client_name);
}
}
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)
{
int i;
gboolean found = FALSE;
static int counter = 0;
uint32_t local_nodeid = get_local_nodeid(handle);
for (i = 0; i < left_list_entries; i++) {
crm_node_t *peer = crm_find_peer(left_list[i].nodeid, NULL);
crm_info("Node %u left group %s (peer=%s, counter=%d.%d)",
left_list[i].nodeid, groupName->value,
(peer? peer->uname : "<none>"), counter, i);
if (peer) {
crm_update_peer_proc(__FUNCTION__, peer, crm_proc_cpg, OFFLINESTATUS);
}
}
for (i = 0; i < joined_list_entries; i++) {
crm_info("Node %u joined group %s (counter=%d.%d)",
joined_list[i].nodeid, groupName->value, counter, i);
}
for (i = 0; i < member_list_entries; i++) {
crm_node_t *peer = crm_get_peer(member_list[i].nodeid, NULL);
crm_info("Node %u still member of group %s (peer=%s, counter=%d.%d)",
member_list[i].nodeid, groupName->value,
(peer? peer->uname : "<none>"), counter, i);
/* Anyone that is sending us CPG messages must also be a _CPG_ member.
* But it's _not_ safe to assume it's in the quorum membership.
* We may have just found out it's dead and are processing the last couple of messages it sent
*/
peer = crm_update_peer_proc(__FUNCTION__, peer, crm_proc_cpg, ONLINESTATUS);
if(peer && peer->state && crm_is_peer_active(peer) == FALSE) {
time_t now = time(NULL);
/* Co-opt the otherwise unused votes field */
if(peer->votes == 0) {
peer->votes = now;
} else if(now > (60 + peer->votes)) {
/* On the otherhand, if we're still getting messages, at a certain point
* we need to acknowledge our internal cache is probably wrong
*
* Set the threshold to 1 minute
*/
crm_err("Node %s[%u] appears to be online even though we think it is dead", peer->uname, peer->id);
if (crm_update_peer_state(__FUNCTION__, peer, CRM_NODE_MEMBER, 0)) {
peer->votes = 0;
}
}
}
if (local_nodeid == member_list[i].nodeid) {
found = TRUE;
}
}
if (!found) {
crm_err("We're not part of CPG group '%s' anymore!", groupName->value);
cpg_evicted = TRUE;
}
counter++;
}
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, F_CRM_TASK);
/* Optimize this for the DC - it has the most to do */
if (op == NULL) {
crm_log_xml_err(stored_msg, "Bad message");
return I_NULL;
}
if (strcmp(op, CRM_OP_SHUTDOWN_REQ) == 0) {
const char *from = crm_element_value(stored_msg, F_CRM_HOST_FROM);
crm_node_t *node = crm_find_peer(0, from);
crm_update_peer_expected(__FUNCTION__, 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) {
const char *host_from = crm_element_value(stored_msg, F_CRM_HOST_FROM);
gboolean dc_match = safe_str_eq(host_from, fsa_our_dc);
if (is_set(fsa_input_register, R_SHUTDOWN)) {
crm_info("Shutting ourselves down (DC)");
return I_STOP;
} else if (dc_match) {
crm_err("We didnt ask to be shut down, yet our"
" TE is telling us too." " Better get out now!");
return I_TERMINATE;
} else if (fsa_state != S_STOPPING) {
crm_err("Another node is asking us to shutdown" " but we think we're ok.");
return I_ELECTION;
}
} else if (strcmp(op, CRM_OP_SHUTDOWN_REQ) == 0) {
/* a slave wants to shut down */
/* create cib fragment and add to message */
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, __FUNCTION__);
} else if (strcmp(op, CRM_OP_THROTTLE) == 0) {
throttle_update(stored_msg);
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, __FUNCTION__);
/* Sometimes we _must_ go into S_ELECTION */
if (fsa_state == S_HALT) {
crm_debug("Forcing an election from S_HALT");
return I_ELECTION;
#if 0
} else if (AM_I_DC) {
/* This is the old way of doing things but what is gained? */
return I_ELECTION;
#endif
}
} else if (strcmp(op, CRM_OP_JOIN_OFFER) == 0) {
crm_debug("Raising I_JOIN_OFFER: join-%s", crm_element_value(stored_msg, F_CRM_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, F_CRM_JOIN_ID));
return I_JOIN_RESULT;
} else if (strcmp(op, CRM_OP_LRM_DELETE) == 0
|| strcmp(op, CRM_OP_LRM_FAIL) == 0
|| strcmp(op, CRM_OP_LRM_REFRESH) == 0 || strcmp(op, CRM_OP_REPROBE) == 0) {
crm_xml_add(stored_msg, F_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_LOCAL_SHUTDOWN) == 0) {
crm_shutdown(SIGTERM);
/*return I_SHUTDOWN; */
return I_NULL;
/*========== (NOT_DC)-Only Actions ==========*/
} else if (AM_I_DC == FALSE && strcmp(op, CRM_OP_SHUTDOWN) == 0) {
const char *host_from = crm_element_value(stored_msg, F_CRM_HOST_FROM);
gboolean dc_match = safe_str_eq(host_from, fsa_our_dc);
if (dc_match || fsa_our_dc == NULL) {
if (is_set(fsa_input_register, R_SHUTDOWN) == FALSE) {
crm_err("We didn't ask to be shut down, yet our" " DC is telling us too.");
set_bit(fsa_input_register, R_STAYDOWN);
return I_STOP;
}
crm_info("Shutting down");
return I_STOP;
} else {
crm_warn("Discarding %s op from %s", op, host_from);
}
} else if (strcmp(op, CRM_OP_PING) == 0) {
/* eventually do some stuff to figure out
* if we /are/ ok
*/
const char *sys_to = crm_element_value(stored_msg, F_CRM_SYS_TO);
xmlNode *ping = create_xml_node(NULL, XML_CRM_TAG_PING);
crm_xml_add(ping, XML_PING_ATTR_STATUS, "ok");
crm_xml_add(ping, XML_PING_ATTR_SYSFROM, sys_to);
crm_xml_add(ping, "crmd_state", fsa_state2string(fsa_state));
/* Ok, so technically not so interesting, but CTS needs to see this */
crm_notice("Current ping state: %s", fsa_state2string(fsa_state));
msg = create_reply(stored_msg, ping);
if (msg) {
(void)relay_message(msg, TRUE);
}
free_xml(ping);
free_xml(msg);
} else if (strcmp(op, CRM_OP_RM_NODE_CACHE) == 0) {
int id = 0;
const char *name = NULL;
crm_element_value_int(stored_msg, XML_ATTR_ID, &id);
name = crm_element_value(stored_msg, XML_ATTR_UNAME);
if(cause == C_IPC_MESSAGE) {
msg = create_request(CRM_OP_RM_NODE_CACHE, NULL, NULL, CRM_SYSTEM_CRMD, CRM_SYSTEM_CRMD, NULL);
if (send_cluster_message(NULL, crm_msg_crmd, msg, TRUE) == FALSE) {
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);
}
free_xml(msg);
} else {
reap_crm_member(id, name);
}
} 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;
}
gboolean
relay_message(xmlNode * msg, gboolean originated_locally)
{
int dest = 1;
int is_for_dc = 0;
int is_for_dcib = 0;
int is_for_te = 0;
int is_for_crm = 0;
int is_for_cib = 0;
int is_local = 0;
gboolean processing_complete = FALSE;
const char *host_to = crm_element_value(msg, F_CRM_HOST_TO);
const char *sys_to = crm_element_value(msg, F_CRM_SYS_TO);
const char *sys_from = crm_element_value(msg, F_CRM_SYS_FROM);
const char *type = crm_element_value(msg, F_TYPE);
const char *msg_error = NULL;
crm_trace("Routing message %s", crm_element_value(msg, XML_ATTR_REFERENCE));
if (msg == NULL) {
msg_error = "Cannot route empty message";
} else if (safe_str_eq(CRM_OP_HELLO, crm_element_value(msg, F_CRM_TASK))) {
/* quietly ignore */
processing_complete = TRUE;
} else if (safe_str_neq(type, T_CRM)) {
msg_error = "Bad message type";
} else if (sys_to == NULL) {
msg_error = "Bad message destination: no subsystem";
}
if (msg_error != NULL) {
processing_complete = TRUE;
crm_err("%s", msg_error);
crm_log_xml_warn(msg, "bad msg");
}
if (processing_complete) {
return TRUE;
}
processing_complete = TRUE;
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);
is_local = 0;
if (host_to == NULL || strlen(host_to) == 0) {
if (is_for_dc || is_for_te) {
is_local = 0;
} else if (is_for_crm && originated_locally) {
is_local = 0;
} else {
is_local = 1;
}
} else if (safe_str_eq(fsa_our_uname, host_to)) {
is_local = 1;
}
if (is_for_dc || is_for_dcib || is_for_te) {
if (AM_I_DC && is_for_te) {
ROUTER_RESULT("Message result: Local relay");
send_msg_via_ipc(msg, sys_to);
} else if (AM_I_DC) {
ROUTER_RESULT("Message result: DC/CRMd process");
processing_complete = FALSE; /* more to be done by caller */
} else if (originated_locally && safe_str_neq(sys_from, CRM_SYSTEM_PENGINE)
&& safe_str_neq(sys_from, CRM_SYSTEM_TENGINE)) {
/* Neither the TE or PE should be sending messages
* to DC's on other nodes
*
* By definition, if we are no longer the DC, then
* the PE or TE's data should be discarded
*/
#if SUPPORT_COROSYNC
if (is_openais_cluster()) {
dest = text2msg_type(sys_to);
}
#endif
ROUTER_RESULT("Message result: External relay to DC");
send_cluster_message(host_to ? crm_get_peer(0, host_to) : NULL, dest, msg, TRUE);
} else {
/* discard */
ROUTER_RESULT("Message result: Discard, not DC");
}
} else if (is_local && (is_for_crm || is_for_cib)) {
ROUTER_RESULT("Message result: CRMd process");
processing_complete = FALSE; /* more to be done by caller */
} else if (is_local) {
ROUTER_RESULT("Message result: Local relay");
send_msg_via_ipc(msg, sys_to);
} else {
crm_node_t *node_to = NULL;
#if SUPPORT_COROSYNC
if (is_openais_cluster()) {
dest = text2msg_type(sys_to);
if (dest == crm_msg_none || dest > crm_msg_stonith_ng) {
dest = crm_msg_crmd;
}
}
#endif
if (host_to) {
node_to = crm_find_peer(0, host_to);
if (node_to == NULL) {
crm_err("Cannot route message to unknown node %s", host_to);
return TRUE;
}
}
ROUTER_RESULT("Message result: External relay");
send_cluster_message(host_to ? node_to : NULL, dest, msg, TRUE);
}
return processing_complete;
}
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(node->uuid && strcasecmp(node->uuid, uuid) == 0) {
if(node->uname) {
return node->uname;
}
break;
}
}
#if SUPPORT_COROSYNC
if (is_openais_cluster()) {
if (uname_is_uuid() == FALSE && is_corosync_cluster()) {
uint32_t id = crm_int_helper(uuid, NULL);
if(id != 0) {
node = crm_find_peer(id, NULL);
} else {
crm_err("Invalid node id: %s", uuid);
}
} else {
node = crm_find_peer(0, uuid);
}
if (node) {
crm_info("Setting uuid for node %s[%u] to '%s'", node->uname, node->id, uuid);
node->uuid = strdup(uuid);
if(node->uname) {
return node->uname;
}
}
return NULL;
}
#endif
#if SUPPORT_HEARTBEAT
if (is_heartbeat_cluster()) {
if (heartbeat_cluster != NULL) {
cl_uuid_t uuid_raw;
char *uuid_copy = strdup(uuid);
char *uname = malloc(MAX_NAME);
cl_uuid_parse(uuid_copy, &uuid_raw);
if (heartbeat_cluster->llc_ops->get_name_by_uuid(heartbeat_cluster, &uuid_raw, uname,
MAX_NAME) == HA_FAIL) {
crm_err("Could not calculate uname for %s", uuid);
} else {
node = crm_get_peer(0, uname);
}
free(uuid_copy);
free(uname);
}
if (node) {
crm_info("Setting uuid for node %s to '%s'", node->uname, uuid);
node->uuid = strdup(uuid);
if(node->uname) {
return node->uname;
}
}
return NULL;
}
#endif
return NULL;
}
