diff --git a/daemons/attrd/attrd_corosync.c b/daemons/attrd/attrd_corosync.c
index 9602c684ae..c9096e2241 100644
--- a/daemons/attrd/attrd_corosync.c
+++ b/daemons/attrd/attrd_corosync.c
@@ -1,598 +1,598 @@
/*
* 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/msg_xml.h>
#include "pacemaker-attrd.h"
static xmlNode *
attrd_confirmation(int callid)
{
xmlNode *node = create_xml_node(NULL, __func__);
crm_xml_add(node, PCMK__XA_T, PCMK__VALUE_ATTRD);
crm_xml_add(node, PCMK__XA_SRC, get_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(crm_node_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->uname,
.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->uname);
attrd_send_message(peer, reply, false);
free_xml(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_message_common_cs(handle, nodeid, pid, msg, &kind, &from);
if(data == NULL) {
return;
}
if (kind == crm_class_cluster) {
xml = string2xml(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),
xml);
}
free_xml(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 Override an attribute sync with a local value
*
* Broadcast the local node's value for an attribute that's different 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.
*
* \param[in] a Attribute entry to override
*
* \return Local instance of attribute value
*/
static attribute_value_t *
broadcast_local_value(const attribute_t *a)
{
attribute_value_t *v = g_hash_table_lookup(a->values, attrd_cluster->uname);
xmlNode *sync = create_xml_node(NULL, __func__);
crm_xml_add(sync, PCMK__XA_TASK, PCMK__ATTRD_CMD_SYNC_RESPONSE);
attrd_add_value_xml(sync, a, v, false);
attrd_send_message(NULL, sync, false);
free_xml(sync);
return v;
}
#define state_text(state) pcmk__s((state), "in unknown state")
/*!
* \internal
* \brief Return a node's value from hash table (creating one if needed)
*
* \param[in,out] values Hash table of values
* \param[in] node_name Name of node 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 *node_name,
const xmlNode *xml)
{
attribute_value_t *v = g_hash_table_lookup(values, node_name);
int is_remote = 0;
if (v == NULL) {
v = calloc(1, sizeof(attribute_value_t));
CRM_ASSERT(v != NULL);
pcmk__str_update(&v->nodename, node_name);
g_hash_table_replace(values, v->nodename, v);
}
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(crm_remote_peer_get(node_name) != NULL);
}
return(v);
}
static void
attrd_peer_change_cb(enum crm_status_type kind, crm_node_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->uname, 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->uname, (gone? "no longer" : "now"));
break;
case crm_status_nstate:
crm_debug("%s node %s is now %s (was %s)",
(is_remote? "Remote" : "Cluster"),
peer->uname, 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, NULL);
}
} else {
// Remove all attribute values associated with lost nodes
attrd_peer_remove(peer->uname, 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->uname);
attrd_do_not_expect_from_peer(peer->uname);
}
}
static void
record_peer_nodeid(attribute_value_t *v, const char *host)
{
crm_node_t *known_peer = pcmk__get_node(v->nodeid, host, NULL,
pcmk__node_search_cluster);
crm_trace("Learned %s has node id %s", known_peer->uname, known_peer->uuid);
if (attrd_election_won()) {
attrd_write_attributes(attrd_write_changed);
}
}
static void
update_attr_on_host(attribute_t *a, const crm_node_t *peer, const xmlNode *xml,
const char *attr, const char *value, const char *host,
bool filter, int is_force_write)
{
attribute_value_t *v = NULL;
v = attrd_lookup_or_create_value(a->values, host, xml);
if (filter && !pcmk__str_eq(v->current, value, pcmk__str_casei)
&& pcmk__str_eq(host, attrd_cluster->uname, pcmk__str_casei)) {
crm_notice("%s[%s]: local value '%s' takes priority over '%s' from %s",
attr, host, v->current, value, peer->uname);
v = broadcast_local_value(a);
} else if (!pcmk__str_eq(v->current, value, pcmk__str_casei)) {
crm_notice("Setting %s[%s]%s%s: %s -> %s "
CRM_XS " from %s with %s write delay",
attr, host, a->set_type ? " in " : "",
pcmk__s(a->set_type, ""), pcmk__s(v->current, "(unset)"),
pcmk__s(value, "(unset)"), peer->uname,
(a->timeout_ms == 0)? "no" : pcmk__readable_interval(a->timeout_ms));
pcmk__str_update(&v->current, value);
a->changed = true;
if (pcmk__str_eq(host, attrd_cluster->uname, pcmk__str_casei)
- && pcmk__str_eq(attr, XML_CIB_ATTR_SHUTDOWN, pcmk__str_none)) {
+ && 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("Delayed write out (%dms) for %s", a->timeout_ms, attr);
mainloop_timer_start(a->timer);
} else {
attrd_write_or_elect_attribute(a);
}
} else {
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("Unchanged %s[%s] from %s is %s(Set the forced write flag)",
attr, host, peer->uname, value);
a->force_write = TRUE;
} else {
crm_trace("Unchanged %s[%s] from %s is %s", attr, host, peer->uname, value);
}
}
// 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_NODE_ID,
(int*)&v->nodeid) == 0) && (v->nodeid > 0)) {
record_peer_nodeid(v, host);
}
}
static void
attrd_peer_update_one(const crm_node_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_NODE_NAME);
int is_force_write = 0;
if (attr == NULL) {
crm_warn("Could not update attribute: peer did not specify name");
return;
}
crm_element_value_int(xml, PCMK__XA_ATTR_FORCE, &is_force_write);
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);
xml_remove_prop(xml, PCMK__XA_ATTR_NODE_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, is_force_write);
}
} else {
// Update attribute value for the given host
update_attr_on_host(a, peer, xml, attr, value, host, filter, is_force_write);
}
/* 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->uname, 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)) {
if (sync == NULL) {
sync = create_xml_node(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);
free_xml(sync);
}
}
int
attrd_cluster_connect(void)
{
attrd_cluster = pcmk_cluster_new();
attrd_cluster->destroy = attrd_cpg_destroy;
attrd_cluster->cpg.cpg_deliver_fn = attrd_cpg_dispatch;
attrd_cluster->cpg.cpg_confchg_fn = pcmk_cpg_membership;
crm_set_status_callback(&attrd_peer_change_cb);
if (crm_cluster_connect(attrd_cluster) == FALSE) {
crm_err("Cluster connection failed");
return -ENOTCONN;
}
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_NODE_NAME);
const char *op = crm_element_value(xml, PCMK__XA_ATTR_OPERATION);
const char *interval_spec = crm_element_value(xml, PCMK__XA_ATTR_INTERVAL);
guint interval_ms = 0U;
char *attr = NULL;
GHashTableIter iter;
regex_t regex;
crm_node_t *peer = pcmk__get_node(0, request->peer, NULL,
pcmk__node_search_cluster);
pcmk_parse_interval_spec(interval_spec, &interval_ms);
if (attrd_failure_regex(®ex, rsc, op, interval_ms) != pcmk_ok) {
crm_info("Ignoring invalid request to clear failures for %s",
pcmk__s(rsc, "all resources"));
return;
}
crm_xml_add(xml, PCMK__XA_TASK, PCMK__ATTRD_CMD_UPDATE);
/* Make sure value is not set, so we delete */
xml_remove_prop(xml, PCMK__XA_ATTR_VALUE);
g_hash_table_iter_init(&iter, attributes);
while (g_hash_table_iter_next(&iter, (gpointer *) &attr, NULL)) {
if (regexec(®ex, attr, 0, NULL, 0) == 0) {
crm_trace("Matched %s when clearing %s",
attr, pcmk__s(rsc, "all resources"));
crm_xml_add(xml, PCMK__XA_ATTR_NAME, attr);
attrd_peer_update(peer, xml, host, false);
}
}
regfree(®ex);
}
/*!
* \internal
* \brief Load attributes from a peer sync response
*
* \param[in] peer Peer that sent clear request
* \param[in] peer_won Whether peer is the attribute writer
* \param[in,out] xml Request XML
*/
void
attrd_peer_sync_response(const crm_node_t *peer, bool peer_won, xmlNode *xml)
{
crm_info("Processing " PCMK__ATTRD_CMD_SYNC_RESPONSE " from %s",
peer->uname);
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__xml_first_child(xml); child != NULL;
child = pcmk__xml_next(child)) {
attrd_peer_update(peer, child,
crm_element_value(child, PCMK__XA_ATTR_NODE_NAME),
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 "
CRM_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);
}
}
void
attrd_peer_sync(crm_node_t *peer, xmlNode *xml)
{
GHashTableIter aIter;
GHashTableIter vIter;
attribute_t *a = NULL;
attribute_value_t *v = NULL;
xmlNode *sync = create_xml_node(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, v->current, peer?peer->uname:"everyone");
attrd_add_value_xml(sync, a, v, false);
}
}
crm_debug("Syncing values to %s", peer?peer->uname:"everyone");
attrd_send_message(peer, sync, false);
free_xml(sync);
}
void
attrd_peer_update(const crm_node_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 = first_named_child(xml, PCMK_XE_OP); child != NULL;
child = crm_next_same_xml(child)) {
attrd_copy_xml_attributes(xml, child);
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_execd_state.c b/daemons/controld/controld_execd_state.c
index 9d746645b5..f4423e0ac9 100644
--- a/daemons/controld/controld_execd_state.c
+++ b/daemons/controld/controld_execd_state.c
@@ -1,817 +1,818 @@
/*
* Copyright 2012-2024 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 <crm/crm.h>
#include <crm/msg_xml.h>
#include <crm/common/iso8601.h>
#include <crm/pengine/rules.h>
#include <crm/pengine/rules_internal.h>
#include <crm/lrmd_internal.h>
#include <pacemaker-internal.h>
#include <pacemaker-controld.h>
static GHashTable *lrm_state_table = NULL;
extern GHashTable *proxy_table;
int lrmd_internal_proxy_send(lrmd_t * lrmd, xmlNode *msg);
void lrmd_internal_set_proxy_callback(lrmd_t * lrmd, void *userdata, void (*callback)(lrmd_t *lrmd, void *userdata, xmlNode *msg));
static void
free_rsc_info(gpointer value)
{
lrmd_rsc_info_t *rsc_info = value;
lrmd_free_rsc_info(rsc_info);
}
static void
free_deletion_op(gpointer value)
{
struct pending_deletion_op_s *op = value;
free(op->rsc);
delete_ha_msg_input(op->input);
free(op);
}
static void
free_recurring_op(gpointer value)
{
active_op_t *op = value;
free(op->user_data);
free(op->rsc_id);
free(op->op_type);
free(op->op_key);
if (op->params) {
g_hash_table_destroy(op->params);
}
free(op);
}
static gboolean
fail_pending_op(gpointer key, gpointer value, gpointer user_data)
{
lrmd_event_data_t event = { 0, };
lrm_state_t *lrm_state = user_data;
active_op_t *op = value;
crm_trace("Pre-emptively failing " PCMK__OP_FMT " on %s (call=%s, %s)",
op->rsc_id, op->op_type, op->interval_ms,
lrm_state->node_name, (char*)key, op->user_data);
event.type = lrmd_event_exec_complete;
event.rsc_id = op->rsc_id;
event.op_type = op->op_type;
event.user_data = op->user_data;
event.timeout = 0;
event.interval_ms = op->interval_ms;
lrmd__set_result(&event, PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_NOT_CONNECTED,
"Action was pending when executor connection was dropped");
event.t_run = (unsigned int) op->start_time;
event.t_rcchange = (unsigned int) op->start_time;
event.call_id = op->call_id;
event.remote_nodename = lrm_state->node_name;
event.params = op->params;
process_lrm_event(lrm_state, &event, op, NULL);
lrmd__reset_result(&event);
return TRUE;
}
gboolean
lrm_state_is_local(lrm_state_t *lrm_state)
{
return (lrm_state != NULL)
&& pcmk__str_eq(lrm_state->node_name, controld_globals.our_nodename,
pcmk__str_casei);
}
/*!
* \internal
* \brief Create executor state entry for a node and add it to the state table
*
* \param[in] node_name Node to create entry for
*
* \return Newly allocated executor state object initialized for \p node_name
*/
static lrm_state_t *
lrm_state_create(const char *node_name)
{
lrm_state_t *state = NULL;
if (!node_name) {
crm_err("No node name given for lrm state object");
return NULL;
}
state = calloc(1, sizeof(lrm_state_t));
if (!state) {
return NULL;
}
state->node_name = strdup(node_name);
state->rsc_info_cache = pcmk__strkey_table(NULL, free_rsc_info);
state->deletion_ops = pcmk__strkey_table(free, free_deletion_op);
state->active_ops = pcmk__strkey_table(free, free_recurring_op);
state->resource_history = pcmk__strkey_table(NULL, history_free);
state->metadata_cache = metadata_cache_new();
g_hash_table_insert(lrm_state_table, (char *)state->node_name, state);
return state;
}
static gboolean
remote_proxy_remove_by_node(gpointer key, gpointer value, gpointer user_data)
{
remote_proxy_t *proxy = value;
const char *node_name = user_data;
if (pcmk__str_eq(node_name, proxy->node_name, pcmk__str_casei)) {
return TRUE;
}
return FALSE;
}
static remote_proxy_t *
find_connected_proxy_by_node(const char * node_name)
{
GHashTableIter gIter;
remote_proxy_t *proxy = NULL;
CRM_CHECK(proxy_table != NULL, return NULL);
g_hash_table_iter_init(&gIter, proxy_table);
while (g_hash_table_iter_next(&gIter, NULL, (gpointer *) &proxy)) {
if (proxy->source
&& pcmk__str_eq(node_name, proxy->node_name, pcmk__str_casei)) {
return proxy;
}
}
return NULL;
}
static void
remote_proxy_disconnect_by_node(const char * node_name)
{
remote_proxy_t *proxy = NULL;
CRM_CHECK(proxy_table != NULL, return);
while ((proxy = find_connected_proxy_by_node(node_name)) != NULL) {
/* mainloop_del_ipc_client() eventually calls remote_proxy_disconnected()
* , which removes the entry from proxy_table.
* Do not do this in a g_hash_table_iter_next() loop. */
if (proxy->source) {
mainloop_del_ipc_client(proxy->source);
}
}
return;
}
static void
internal_lrm_state_destroy(gpointer data)
{
lrm_state_t *lrm_state = data;
if (!lrm_state) {
return;
}
/* Rather than directly remove the recorded proxy entries from proxy_table,
* make sure any connected proxies get disconnected. So that
* remote_proxy_disconnected() will be called and as well remove the
* entries from proxy_table.
*/
remote_proxy_disconnect_by_node(lrm_state->node_name);
crm_trace("Destroying proxy table %s with %u members",
lrm_state->node_name, g_hash_table_size(proxy_table));
// Just in case there's still any leftovers in proxy_table
g_hash_table_foreach_remove(proxy_table, remote_proxy_remove_by_node, (char *) lrm_state->node_name);
remote_ra_cleanup(lrm_state);
lrmd_api_delete(lrm_state->conn);
if (lrm_state->rsc_info_cache) {
crm_trace("Destroying rsc info cache with %u members",
g_hash_table_size(lrm_state->rsc_info_cache));
g_hash_table_destroy(lrm_state->rsc_info_cache);
}
if (lrm_state->resource_history) {
crm_trace("Destroying history op cache with %u members",
g_hash_table_size(lrm_state->resource_history));
g_hash_table_destroy(lrm_state->resource_history);
}
if (lrm_state->deletion_ops) {
crm_trace("Destroying deletion op cache with %u members",
g_hash_table_size(lrm_state->deletion_ops));
g_hash_table_destroy(lrm_state->deletion_ops);
}
if (lrm_state->active_ops != NULL) {
crm_trace("Destroying pending op cache with %u members",
g_hash_table_size(lrm_state->active_ops));
g_hash_table_destroy(lrm_state->active_ops);
}
metadata_cache_free(lrm_state->metadata_cache);
free((char *)lrm_state->node_name);
free(lrm_state);
}
void
lrm_state_reset_tables(lrm_state_t * lrm_state, gboolean reset_metadata)
{
if (lrm_state->resource_history) {
crm_trace("Resetting resource history cache with %u members",
g_hash_table_size(lrm_state->resource_history));
g_hash_table_remove_all(lrm_state->resource_history);
}
if (lrm_state->deletion_ops) {
crm_trace("Resetting deletion operations cache with %u members",
g_hash_table_size(lrm_state->deletion_ops));
g_hash_table_remove_all(lrm_state->deletion_ops);
}
if (lrm_state->active_ops != NULL) {
crm_trace("Resetting active operations cache with %u members",
g_hash_table_size(lrm_state->active_ops));
g_hash_table_remove_all(lrm_state->active_ops);
}
if (lrm_state->rsc_info_cache) {
crm_trace("Resetting resource information cache with %u members",
g_hash_table_size(lrm_state->rsc_info_cache));
g_hash_table_remove_all(lrm_state->rsc_info_cache);
}
if (reset_metadata) {
metadata_cache_reset(lrm_state->metadata_cache);
}
}
gboolean
lrm_state_init_local(void)
{
if (lrm_state_table) {
return TRUE;
}
lrm_state_table = pcmk__strikey_table(NULL, internal_lrm_state_destroy);
if (!lrm_state_table) {
return FALSE;
}
proxy_table = pcmk__strikey_table(NULL, remote_proxy_free);
if (!proxy_table) {
g_hash_table_destroy(lrm_state_table);
lrm_state_table = NULL;
return FALSE;
}
return TRUE;
}
void
lrm_state_destroy_all(void)
{
if (lrm_state_table) {
crm_trace("Destroying state table with %u members",
g_hash_table_size(lrm_state_table));
g_hash_table_destroy(lrm_state_table); lrm_state_table = NULL;
}
if(proxy_table) {
crm_trace("Destroying proxy table with %u members",
g_hash_table_size(proxy_table));
g_hash_table_destroy(proxy_table); proxy_table = NULL;
}
}
lrm_state_t *
lrm_state_find(const char *node_name)
{
if ((node_name == NULL) || (lrm_state_table == NULL)) {
return NULL;
}
return g_hash_table_lookup(lrm_state_table, node_name);
}
lrm_state_t *
lrm_state_find_or_create(const char *node_name)
{
lrm_state_t *lrm_state;
CRM_CHECK(lrm_state_table != NULL, return NULL);
lrm_state = g_hash_table_lookup(lrm_state_table, node_name);
if (!lrm_state) {
lrm_state = lrm_state_create(node_name);
}
return lrm_state;
}
GList *
lrm_state_get_list(void)
{
if (lrm_state_table == NULL) {
return NULL;
}
return g_hash_table_get_values(lrm_state_table);
}
void
lrm_state_disconnect_only(lrm_state_t * lrm_state)
{
int removed = 0;
if (!lrm_state->conn) {
return;
}
crm_trace("Disconnecting %s", lrm_state->node_name);
remote_proxy_disconnect_by_node(lrm_state->node_name);
((lrmd_t *) lrm_state->conn)->cmds->disconnect(lrm_state->conn);
if (!pcmk_is_set(controld_globals.fsa_input_register, R_SHUTDOWN)) {
removed = g_hash_table_foreach_remove(lrm_state->active_ops,
fail_pending_op, lrm_state);
crm_trace("Synthesized %d operation failures for %s", removed, lrm_state->node_name);
}
}
void
lrm_state_disconnect(lrm_state_t * lrm_state)
{
if (!lrm_state->conn) {
return;
}
lrm_state_disconnect_only(lrm_state);
lrmd_api_delete(lrm_state->conn);
lrm_state->conn = NULL;
}
int
lrm_state_is_connected(lrm_state_t * lrm_state)
{
if (!lrm_state->conn) {
return FALSE;
}
return ((lrmd_t *) lrm_state->conn)->cmds->is_connected(lrm_state->conn);
}
int
lrm_state_poke_connection(lrm_state_t * lrm_state)
{
if (!lrm_state->conn) {
return -ENOTCONN;
}
return ((lrmd_t *) lrm_state->conn)->cmds->poke_connection(lrm_state->conn);
}
// \return Standard Pacemaker return code
int
controld_connect_local_executor(lrm_state_t *lrm_state)
{
int rc = pcmk_rc_ok;
if (lrm_state->conn == NULL) {
lrmd_t *api = NULL;
rc = lrmd__new(&api, NULL, NULL, 0);
if (rc != pcmk_rc_ok) {
return rc;
}
api->cmds->set_callback(api, lrm_op_callback);
lrm_state->conn = api;
}
rc = ((lrmd_t *) lrm_state->conn)->cmds->connect(lrm_state->conn,
CRM_SYSTEM_CRMD, NULL);
rc = pcmk_legacy2rc(rc);
if (rc == pcmk_rc_ok) {
lrm_state->num_lrm_register_fails = 0;
} else {
lrm_state->num_lrm_register_fails++;
}
return rc;
}
static remote_proxy_t *
crmd_remote_proxy_new(lrmd_t *lrmd, const char *node_name, const char *session_id, const char *channel)
{
struct ipc_client_callbacks proxy_callbacks = {
.dispatch = remote_proxy_dispatch,
.destroy = remote_proxy_disconnected
};
remote_proxy_t *proxy = remote_proxy_new(lrmd, &proxy_callbacks, node_name,
session_id, channel);
return proxy;
}
gboolean
crmd_is_proxy_session(const char *session)
{
return g_hash_table_lookup(proxy_table, session) ? TRUE : FALSE;
}
void
crmd_proxy_send(const char *session, xmlNode *msg)
{
remote_proxy_t *proxy = g_hash_table_lookup(proxy_table, session);
lrm_state_t *lrm_state = NULL;
if (!proxy) {
return;
}
crm_log_xml_trace(msg, "to-proxy");
lrm_state = lrm_state_find(proxy->node_name);
if (lrm_state) {
crm_trace("Sending event to %.8s on %s", proxy->session_id, proxy->node_name);
remote_proxy_relay_event(proxy, msg);
}
}
static void
crmd_proxy_dispatch(const char *session, xmlNode *msg)
{
crm_trace("Processing proxied IPC message from session %s", session);
crm_log_xml_trace(msg, "controller[inbound]");
crm_xml_add(msg, PCMK__XA_CRM_SYS_FROM, session);
if (controld_authorize_ipc_message(msg, NULL, session)) {
route_message(C_IPC_MESSAGE, msg);
}
controld_trigger_fsa();
}
static void
remote_config_check(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data)
{
if (rc != pcmk_ok) {
crm_err("Query resulted in an error: %s", pcmk_strerror(rc));
if (rc == -EACCES || rc == -pcmk_err_schema_validation) {
crm_err("The cluster is mis-configured - shutting down and staying down");
}
} else {
lrmd_t * lrmd = (lrmd_t *)user_data;
crm_time_t *now = crm_time_new(NULL);
GHashTable *config_hash = pcmk__strkey_table(free, free);
crm_debug("Call %d : Parsing CIB options", call_id);
pe_unpack_nvpairs(output, output, PCMK_XE_CLUSTER_PROPERTY_SET, NULL,
config_hash, PCMK_VALUE_CIB_BOOTSTRAP_OPTIONS, FALSE,
now, NULL);
/* Now send it to the remote peer */
lrmd__validate_remote_settings(lrmd, config_hash);
g_hash_table_destroy(config_hash);
crm_time_free(now);
}
}
static void
crmd_remote_proxy_cb(lrmd_t *lrmd, void *userdata, xmlNode *msg)
{
lrm_state_t *lrm_state = userdata;
const char *session = crm_element_value(msg, F_LRMD_IPC_SESSION);
remote_proxy_t *proxy = g_hash_table_lookup(proxy_table, session);
const char *op = crm_element_value(msg, F_LRMD_IPC_OP);
if (pcmk__str_eq(op, LRMD_IPC_OP_NEW, pcmk__str_casei)) {
const char *channel = crm_element_value(msg, F_LRMD_IPC_IPC_SERVER);
proxy = crmd_remote_proxy_new(lrmd, lrm_state->node_name, session, channel);
if (!remote_ra_controlling_guest(lrm_state)) {
if (proxy != NULL) {
cib_t *cib_conn = controld_globals.cib_conn;
/* Look up PCMK_OPT_STONITH_WATCHDOG_TIMEOUT and send to the
* remote peer for validation
*/
int rc = cib_conn->cmds->query(cib_conn, PCMK_XE_CRM_CONFIG,
NULL, cib_scope_local);
cib_conn->cmds->register_callback_full(cib_conn, rc, 10, FALSE,
lrmd,
"remote_config_check",
remote_config_check,
NULL);
}
} else {
crm_debug("Skipping remote_config_check for guest-nodes");
}
} else if (pcmk__str_eq(op, LRMD_IPC_OP_SHUTDOWN_REQ, pcmk__str_casei)) {
char *now_s = NULL;
crm_notice("%s requested shutdown of its remote connection",
lrm_state->node_name);
if (!remote_ra_is_in_maintenance(lrm_state)) {
now_s = pcmk__ttoa(time(NULL));
- update_attrd(lrm_state->node_name, XML_CIB_ATTR_SHUTDOWN, now_s, NULL, TRUE);
+ update_attrd(lrm_state->node_name, PCMK__NODE_ATTR_SHUTDOWN, now_s,
+ NULL, TRUE);
free(now_s);
remote_proxy_ack_shutdown(lrmd);
crm_warn("Reconnection attempts to %s may result in failures that must be cleared",
lrm_state->node_name);
} else {
remote_proxy_nack_shutdown(lrmd);
crm_notice("Remote resource for %s is not managed so no ordered shutdown happening",
lrm_state->node_name);
}
return;
} else if (pcmk__str_eq(op, LRMD_IPC_OP_REQUEST, pcmk__str_casei) && proxy && proxy->is_local) {
/* This is for the controller, which we are, so don't try
* to send to ourselves over IPC -- do it directly.
*/
int flags = 0;
xmlNode *request = get_message_xml(msg, F_LRMD_IPC_MSG);
CRM_CHECK(request != NULL, return);
CRM_CHECK(lrm_state->node_name, return);
crm_xml_add(request, PCMK_XE_ACL_ROLE, "pacemaker-remote");
pcmk__update_acl_user(request, F_LRMD_IPC_USER, lrm_state->node_name);
/* Pacemaker Remote nodes don't know their own names (as known to the
* cluster). When getting a node info request with no name or ID, add
* the name, so we don't return info for ourselves instead of the
* Pacemaker Remote node.
*/
if (pcmk__str_eq(crm_element_value(request, PCMK__XA_CRM_TASK),
CRM_OP_NODE_INFO, pcmk__str_none)) {
int node_id = 0;
crm_element_value_int(request, PCMK_XA_ID, &node_id);
if ((node_id <= 0)
&& (crm_element_value(request, PCMK_XA_UNAME) == NULL)) {
crm_xml_add(request, PCMK_XA_UNAME, lrm_state->node_name);
}
}
crmd_proxy_dispatch(session, request);
crm_element_value_int(msg, F_LRMD_IPC_MSG_FLAGS, &flags);
if (flags & crm_ipc_client_response) {
int msg_id = 0;
xmlNode *op_reply = create_xml_node(NULL, "ack");
crm_xml_add(op_reply, "function", __func__);
crm_xml_add_int(op_reply, "line", __LINE__);
crm_element_value_int(msg, F_LRMD_IPC_MSG_ID, &msg_id);
remote_proxy_relay_response(proxy, op_reply, msg_id);
free_xml(op_reply);
}
} else {
remote_proxy_cb(lrmd, lrm_state->node_name, msg);
}
}
// \return Standard Pacemaker return code
int
controld_connect_remote_executor(lrm_state_t *lrm_state, const char *server,
int port, int timeout_ms)
{
int rc = pcmk_rc_ok;
if (lrm_state->conn == NULL) {
lrmd_t *api = NULL;
rc = lrmd__new(&api, lrm_state->node_name, server, port);
if (rc != pcmk_rc_ok) {
crm_warn("Pacemaker Remote connection to %s:%s failed: %s "
CRM_XS " rc=%d", server, port, pcmk_rc_str(rc), rc);
return rc;
}
lrm_state->conn = api;
api->cmds->set_callback(api, remote_lrm_op_callback);
lrmd_internal_set_proxy_callback(api, lrm_state, crmd_remote_proxy_cb);
}
crm_trace("Initiating remote connection to %s:%d with timeout %dms",
server, port, timeout_ms);
rc = ((lrmd_t *) lrm_state->conn)->cmds->connect_async(lrm_state->conn,
lrm_state->node_name,
timeout_ms);
if (rc == pcmk_ok) {
lrm_state->num_lrm_register_fails = 0;
} else {
lrm_state->num_lrm_register_fails++; // Ignored for remote connections
}
return pcmk_legacy2rc(rc);
}
int
lrm_state_get_metadata(lrm_state_t * lrm_state,
const char *class,
const char *provider,
const char *agent, char **output, enum lrmd_call_options options)
{
lrmd_key_value_t *params = NULL;
if (!lrm_state->conn) {
return -ENOTCONN;
}
/* Add the node name to the environment, as is done with normal resource
* action calls. Meta-data calls shouldn't need it, but some agents are
* written with an ocf_local_nodename call at the beginning regardless of
* action. Without the environment variable, the agent would try to contact
* the controller to get the node name -- but the controller would be
* blocking on the synchronous meta-data call.
*
* At this point, we have to assume that agents are unlikely to make other
* calls that require the controller, such as crm_node --quorum or
* --cluster-id.
*
* @TODO Make meta-data calls asynchronous. (This will be part of a larger
* project to make meta-data calls via the executor rather than directly.)
*/
params = lrmd_key_value_add(params, CRM_META "_" PCMK__META_ON_NODE,
lrm_state->node_name);
return ((lrmd_t *) lrm_state->conn)->cmds->get_metadata_params(lrm_state->conn,
class, provider, agent, output, options, params);
}
int
lrm_state_cancel(lrm_state_t *lrm_state, const char *rsc_id, const char *action,
guint interval_ms)
{
if (!lrm_state->conn) {
return -ENOTCONN;
}
/* Figure out a way to make this async?
* NOTICE: Currently it's synced and directly acknowledged in do_lrm_invoke(). */
if (is_remote_lrmd_ra(NULL, NULL, rsc_id)) {
return remote_ra_cancel(lrm_state, rsc_id, action, interval_ms);
}
return ((lrmd_t *) lrm_state->conn)->cmds->cancel(lrm_state->conn, rsc_id,
action, interval_ms);
}
lrmd_rsc_info_t *
lrm_state_get_rsc_info(lrm_state_t * lrm_state, const char *rsc_id, enum lrmd_call_options options)
{
lrmd_rsc_info_t *rsc = NULL;
if (!lrm_state->conn) {
return NULL;
}
if (is_remote_lrmd_ra(NULL, NULL, rsc_id)) {
return remote_ra_get_rsc_info(lrm_state, rsc_id);
}
rsc = g_hash_table_lookup(lrm_state->rsc_info_cache, rsc_id);
if (rsc == NULL) {
/* only contact the lrmd if we don't already have a cached rsc info */
rsc = ((lrmd_t *) lrm_state->conn)->cmds->get_rsc_info(lrm_state->conn, rsc_id, options);
if (rsc == NULL) {
return NULL;
}
/* cache the result */
g_hash_table_insert(lrm_state->rsc_info_cache, rsc->id, rsc);
}
return lrmd_copy_rsc_info(rsc);
}
/*!
* \internal
* \brief Initiate a resource agent action
*
* \param[in,out] lrm_state Executor state object
* \param[in] rsc_id ID of resource for action
* \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 ms) before initiating action
* \param[in] parameters Hash table of resource parameters
* \param[out] call_id Where to store call ID on success
*
* \return Standard Pacemaker return code
*/
int
controld_execute_resource_agent(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, GHashTable *parameters,
int *call_id)
{
int rc = pcmk_rc_ok;
lrmd_key_value_t *params = NULL;
if (lrm_state->conn == NULL) {
return ENOTCONN;
}
// Convert parameters from hash table to list
if (parameters != NULL) {
const char *key = NULL;
const char *value = NULL;
GHashTableIter iter;
g_hash_table_iter_init(&iter, parameters);
while (g_hash_table_iter_next(&iter, (gpointer *) &key,
(gpointer *) &value)) {
params = lrmd_key_value_add(params, key, value);
}
}
if (is_remote_lrmd_ra(NULL, NULL, rsc_id)) {
rc = controld_execute_remote_agent(lrm_state, rsc_id, action,
userdata, interval_ms, timeout_ms,
start_delay_ms, params, call_id);
} else {
rc = ((lrmd_t *) lrm_state->conn)->cmds->exec(lrm_state->conn, rsc_id,
action, userdata,
interval_ms, timeout_ms,
start_delay_ms,
lrmd_opt_notify_changes_only,
params);
if (rc < 0) {
rc = pcmk_legacy2rc(rc);
} else {
*call_id = rc;
rc = pcmk_rc_ok;
}
}
return rc;
}
int
lrm_state_register_rsc(lrm_state_t * lrm_state,
const char *rsc_id,
const char *class,
const char *provider, const char *agent, enum lrmd_call_options options)
{
lrmd_t *conn = (lrmd_t *) lrm_state->conn;
if (conn == NULL) {
return -ENOTCONN;
}
if (is_remote_lrmd_ra(agent, provider, NULL)) {
return lrm_state_find_or_create(rsc_id)? pcmk_ok : -EINVAL;
}
/* @TODO Implement an asynchronous version of this (currently a blocking
* call to the lrmd).
*/
return conn->cmds->register_rsc(lrm_state->conn, rsc_id, class, provider,
agent, options);
}
int
lrm_state_unregister_rsc(lrm_state_t * lrm_state,
const char *rsc_id, enum lrmd_call_options options)
{
if (!lrm_state->conn) {
return -ENOTCONN;
}
if (is_remote_lrmd_ra(NULL, NULL, rsc_id)) {
g_hash_table_remove(lrm_state_table, rsc_id);
return pcmk_ok;
}
g_hash_table_remove(lrm_state->rsc_info_cache, rsc_id);
/* @TODO Optimize this ... this function is a blocking round trip from
* client to daemon. The controld_execd_state.c code path that uses this
* function should always treat it as an async operation. The executor API
* should make an async version available.
*/
return ((lrmd_t *) lrm_state->conn)->cmds->unregister_rsc(lrm_state->conn, rsc_id, options);
}
diff --git a/daemons/controld/controld_messages.c b/daemons/controld/controld_messages.c
index 4b1b1669da..c8f6c4a1b2 100644
--- a/daemons/controld/controld_messages.c
+++ b/daemons/controld/controld_messages.c
@@ -1,1337 +1,1337 @@
/*
* 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/msg_xml.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 = calloc(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)
{
ha_msg_input_t *copy = calloc(1, sizeof(ha_msg_input_t));
CRM_ASSERT(copy != NULL);
copy->msg = (orig && orig->msg)? copy_xml(orig->msg) : NULL;
copy->xml = get_message_xml(copy->msg, PCMK__XE_CRM_XML);
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;
free_xml(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_ais;
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;
crm_node_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 (is_corosync_cluster()) {
dest = text2msg_type(sys_to);
if ((dest < crm_msg_ais) || (dest > crm_msg_stonith_ng)) {
/* 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 *msg_data = get_message_xml(msg, PCMK__XE_CRM_XML);
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);
}
send_cluster_message(node_to, dest, msg, TRUE);
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);
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");
send_cluster_message(node_to, dest, msg, TRUE);
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 "
CRM_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 *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 "
CRM_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;
}
message_data = get_message_xml(client_msg, PCMK__XE_CRM_XML);
client_name = crm_element_value(message_data, "client_name");
if (pcmk__str_empty(client_name)) {
crm_warn("IPC hello from client rejected: No client name",
CRM_XS " ref=%s uuid=%s", (ref? ref : "none"), uuid);
goto rejected;
}
if (!authorize_version(message_data, "major_version", client_name, ref,
uuid)) {
goto rejected;
}
if (!authorize_version(message_data, "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 = strdup(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 *xml_op = get_message_xml(stored_msg, PCMK__XE_CRM_XML);
if (xml_op) {
xmlNode *xml_rsc = first_named_child(xml_op, PCMK_XE_PRIMITIVE);
xmlNode *xml_attrs = first_named_child(xml_op, PCMK__XE_ATTRIBUTES);
if (xml_rsc) {
rsc = 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 *msg_data = get_message_xml(stored_msg, PCMK__XE_CRM_XML);
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 = first_named_child(msg_data, PCMK_XE_PRIMITIVE);
CRM_CHECK(rsc_xml != NULL, return I_NULL);
rsc_id = 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 = strdup(transition? transition : FAKE_TE_ID);
op->params = pcmk__strkey_table(free, free);
g_hash_table_insert(op->params, strdup(PCMK_XA_CRM_FEATURE_SET),
strdup(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 = ID(msg);
crm_node_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 = crm_remote_peer_get(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_CONN_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 = create_xml_node(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);
free_xml(ping);
if (reply != NULL) {
(void) relay_message(reply, TRUE);
free_xml(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;
crm_node_t *node = NULL;
xmlNode *reply = NULL;
xmlNode *reply_data = NULL;
// Create message data for reply
reply_data = create_xml_node(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 = create_xml_node(reply_data, PCMK_XE_NODE);
crm_xml_add_ll(xml, PCMK_XA_ID, (long long) node->id); // uint32_t
crm_xml_add(xml, PCMK_XA_UNAME, node->uname);
crm_xml_add(xml, PCMK__XA_IN_CCM, node->state);
}
// Create and send reply
reply = create_reply(request, reply_data);
free_xml(reply_data);
if (reply) {
(void) relay_message(reply, TRUE);
free_xml(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;
crm_node_t *node = NULL;
int node_id = 0;
xmlNode *reply = NULL;
xmlNode *reply_data = NULL;
// Build reply
reply_data = create_xml_node(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_UNAME, node->uname);
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);
free_xml(reply_data);
if (reply != NULL) {
(void) relay_message(reply, TRUE);
free_xml(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);
crm_node_t *node = pcmk__search_node_caches(0, from,
pcmk__node_search_cluster);
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 (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);
/* 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 *xml = get_message_xml(stored_msg, PCMK__XE_CRM_XML);
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, XML_CIB_ATTR_SHUTDOWN, now_s, NULL, FALSE);
+ 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 *data = get_message_xml(msg, PCMK__XE_CRM_XML);
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;
fsa_input.msg = msg;
fsa_input.xml = get_message_xml(msg, PCMK__XE_CRM_XML);
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;
}
free_xml(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_CONN_HOST, controld_globals.our_nodename);
}
send_cluster_message(NULL, crm_msg_crmd, msg, TRUE);
free_xml(msg);
}
diff --git a/include/crm/common/Makefile.am b/include/crm/common/Makefile.am
index 28bec45f5c..ce0b881301 100644
--- a/include/crm/common/Makefile.am
+++ b/include/crm/common/Makefile.am
@@ -1,46 +1,47 @@
#
-# Copyright 2004-2023 the Pacemaker project contributors
+# 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.
#
MAINTAINERCLEANFILES = Makefile.in
headerdir=$(pkgincludedir)/crm/common
header_HEADERS = acl.h \
actions.h \
agents.h \
agents_compat.h \
cib.h \
ipc.h \
ipc_controld.h \
ipc_pacemakerd.h \
ipc_schedulerd.h \
iso8601.h \
logging.h \
logging_compat.h \
mainloop.h \
mainloop_compat.h \
nodes.h \
+ nodes_internal.h \
nvpair.h \
options.h \
output.h \
resources.h \
results.h \
results_compat.h \
roles.h \
rules.h \
scheduler.h \
scheduler_types.h \
tags.h \
tickets.h \
util.h \
util_compat.h \
xml.h \
xml_compat.h
noinst_HEADERS = $(wildcard *internal.h)
diff --git a/include/crm/common/nodes_internal.h b/include/crm/common/nodes_internal.h
new file mode 100644
index 0000000000..b1695b0352
--- /dev/null
+++ b/include/crm/common/nodes_internal.h
@@ -0,0 +1,20 @@
+/*
+ * Copyright 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__NODES_INTERNAL__H
+# define PCMK__NODES_INTERNAL__H
+
+/*
+ * Special node attributes
+ */
+
+#define PCMK__NODE_ATTR_SHUTDOWN "shutdown"
+
+
+#endif // PCMK__NODES_INTERNAL__H
diff --git a/include/crm/common/scheduler_internal.h b/include/crm/common/scheduler_internal.h
index a2dff7c439..886f015264 100644
--- a/include/crm/common/scheduler_internal.h
+++ b/include/crm/common/scheduler_internal.h
@@ -1,171 +1,172 @@
/*
- * Copyright 2004-2023 the Pacemaker project contributors
+ * 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_COMMON_SCHEDULER_INTERNAL__H
# define PCMK__CRM_COMMON_SCHEDULER_INTERNAL__H
#include <crm/common/action_relation_internal.h>
#include <crm/common/actions_internal.h>
#include <crm/common/bundles_internal.h>
#include <crm/common/clone_internal.h>
#include <crm/common/digests_internal.h>
#include <crm/common/failcounts_internal.h>
#include <crm/common/group_internal.h>
#include <crm/common/history_internal.h>
#include <crm/common/location_internal.h>
+#include <crm/common/nodes_internal.h>
#include <crm/common/roles_internal.h>
#ifdef __cplusplus
extern "C" {
#endif
/* Some warnings are too noisy when logged every time a give function is called
* (for example, using a deprecated feature). As an alternative, we allow
* warnings to be logged once per scheduler sequence (transition). Each of those
* warnings needs a flag defined here.
*/
enum pcmk__sched_warnings {
pcmk__wo_blind = (1 << 0),
pcmk__wo_restart_type = (1 << 1),
pcmk__wo_role_after = (1 << 2),
pcmk__wo_poweroff = (1 << 3),
pcmk__wo_require_all = (1 << 4),
pcmk__wo_order_score = (1 << 5),
pcmk__wo_neg_threshold = (1 << 6),
pcmk__wo_remove_after = (1 << 7),
pcmk__wo_ping_node = (1 << 8),
pcmk__wo_order_inst = (1 << 9),
pcmk__wo_coloc_inst = (1 << 10),
pcmk__wo_group_order = (1 << 11),
pcmk__wo_group_coloc = (1 << 12),
pcmk__wo_upstart = (1 << 13),
pcmk__wo_nagios = (1 << 14),
pcmk__wo_set_ordering = (1 << 15),
};
enum pcmk__check_parameters {
/* Clear fail count if parameters changed for un-expired start or monitor
* last_failure.
*/
pcmk__check_last_failure,
/* Clear fail count if parameters changed for start, monitor, promote, or
* migrate_from actions for active resources.
*/
pcmk__check_active,
};
// Group of enum pcmk__sched_warnings flags for warnings we want to log once
extern uint32_t pcmk__warnings;
/*!
* \internal
* \brief Log a resource-tagged message at info severity
*
* \param[in] rsc Tag message with this resource's ID
* \param[in] fmt... printf(3)-style format and arguments
*/
#define pcmk__rsc_info(rsc, fmt, args...) \
crm_log_tag(LOG_INFO, ((rsc) == NULL)? "<NULL>" : (rsc)->id, (fmt), ##args)
/*!
* \internal
* \brief Log a resource-tagged message at debug severity
*
* \param[in] rsc Tag message with this resource's ID
* \param[in] fmt... printf(3)-style format and arguments
*/
#define pcmk__rsc_debug(rsc, fmt, args...) \
crm_log_tag(LOG_DEBUG, ((rsc) == NULL)? "<NULL>" : (rsc)->id, (fmt), ##args)
/*!
* \internal
* \brief Log a resource-tagged message at trace severity
*
* \param[in] rsc Tag message with this resource's ID
* \param[in] fmt... printf(3)-style format and arguments
*/
#define pcmk__rsc_trace(rsc, fmt, args...) \
crm_log_tag(LOG_TRACE, ((rsc) == NULL)? "<NULL>" : (rsc)->id, (fmt), ##args)
/*!
* \internal
* \brief Log an error and remember that current scheduler input has errors
*
* \param[in] fmt... printf(3)-style format and arguments
*/
#define pcmk__sched_err(fmt...) do { \
was_processing_error = TRUE; \
crm_err(fmt); \
} while (0)
/*!
* \internal
* \brief Log a warning and remember that current scheduler input has warnings
*
* \param[in] fmt... printf(3)-style format and arguments
*/
#define pcmk__sched_warn(fmt...) do { \
was_processing_warning = TRUE; \
crm_warn(fmt); \
} while (0)
/*!
* \internal
* \brief Log a warning once per scheduler run
*
* \param[in] wo_flag enum pcmk__sched_warnings value for this warning
* \param[in] fmt... printf(3)-style format and arguments
*/
#define pcmk__warn_once(wo_flag, fmt...) do { \
if (!pcmk_is_set(pcmk__warnings, wo_flag)) { \
if (wo_flag == pcmk__wo_blind) { \
crm_warn(fmt); \
} else { \
pcmk__config_warn(fmt); \
} \
pcmk__warnings = pcmk__set_flags_as(__func__, __LINE__, \
LOG_TRACE, \
"Warn-once", "logging", \
pcmk__warnings, \
(wo_flag), #wo_flag); \
} \
} while (0)
/*!
* \internal
* \brief Set scheduler flags
*
* \param[in,out] scheduler Scheduler data
* \param[in] flags_to_set Group of enum pcmk_scheduler_flags to set
*/
#define pcmk__set_scheduler_flags(scheduler, flags_to_set) do { \
(scheduler)->flags = pcmk__set_flags_as(__func__, __LINE__, \
LOG_TRACE, "Scheduler", crm_system_name, \
(scheduler)->flags, (flags_to_set), #flags_to_set); \
} while (0)
/*!
* \internal
* \brief Clear scheduler flags
*
* \param[in,out] scheduler Scheduler data
* \param[in] flags_to_clear Group of enum pcmk_scheduler_flags to clear
*/
#define pcmk__clear_scheduler_flags(scheduler, flags_to_clear) do { \
(scheduler)->flags = pcmk__clear_flags_as(__func__, __LINE__, \
LOG_TRACE, "Scheduler", crm_system_name, \
(scheduler)->flags, (flags_to_clear), #flags_to_clear); \
} while (0)
#ifdef __cplusplus
}
#endif
#endif // PCMK__CRM_COMMON_SCHEDULER_INTERNAL__H
diff --git a/include/crm_internal.h b/include/crm_internal.h
index b7a6514a34..11cfde214c 100644
--- a/include/crm_internal.h
+++ b/include/crm_internal.h
@@ -1,246 +1,247 @@
/*
* Copyright 2006-2024 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 CRM_INTERNAL__H
# define CRM_INTERNAL__H
# ifndef PCMK__CONFIG_H
# define PCMK__CONFIG_H
# include <config.h>
# endif
# include <portability.h>
/* Our minimum glib dependency is 2.42. Define that as both the minimum and
* maximum glib APIs that are allowed (i.e. APIs that were already deprecated
* in 2.42, and APIs introduced after 2.42, cannot be used by Pacemaker code).
*/
#define GLIB_VERSION_MIN_REQUIRED GLIB_VERSION_2_42
#define GLIB_VERSION_MAX_ALLOWED GLIB_VERSION_2_42
# include <glib.h>
# include <stdbool.h>
# include <libxml/tree.h>
/* Public API headers can guard including deprecated API headers with this
* symbol, thus preventing internal code (which includes this header) from using
* deprecated APIs, while still allowing external code to use them by default.
*/
#define PCMK_ALLOW_DEPRECATED 0
# include <crm/lrmd.h>
# include <crm/common/logging.h>
# include <crm/common/logging_internal.h>
# include <crm/common/ipc_internal.h>
# include <crm/common/options_internal.h>
# include <crm/common/output_internal.h>
+# include <crm/common/scheduler_internal.h>
# include <crm/common/xml_internal.h>
# include <crm/common/xml_names_internal.h>
# include <crm/common/internal.h>
# include <locale.h>
# include <gettext.h>
#define N_(String) (String)
#ifdef ENABLE_NLS
# define _(String) gettext(String)
#else
# define _(String) (String)
#endif
/*
* XML element names used only by internal code
*/
#define PCMK__XE_ATTRIBUTES "attributes"
#define PCMK__XE_CRM_EVENT "crm_event"
#define PCMK__XE_CRM_XML "crm_xml"
#define PCMK__XE_DOWNED "downed"
#define PCMK__XE_FAILED "failed"
#define PCMK__XE_FAILED_UPDATE "failed_update"
#define PCMK__XE_GENERATION_TUPLE "generation_tuple"
#define PCMK__XE_LRM "lrm"
#define PCMK__XE_LRM_RESOURCE "lrm_resource"
#define PCMK__XE_LRM_RESOURCES "lrm_resources"
#define PCMK__XE_LRM_RSC_OP "lrm_rsc_op"
#define PCMK__XE_MAINTENANCE "maintenance"
#define PCMK__XE_NODE_STATE "node_state"
#define PCMK__XE_OPTIONS "options"
#define PCMK__XE_PARAM "param"
#define PCMK__XE_PING_RESPONSE "ping_response"
#define PCMK__XE_PSEUDO_EVENT "pseudo_event"
#define PCMK__XE_RSC_OP "rsc_op"
#define PCMK__XE_SHUTDOWN "shutdown"
#define PCMK__XE_TICKET_STATE "ticket_state"
#define PCMK__XE_TRANSIENT_ATTRIBUTES "transient_attributes"
#define PCMK__XE_TRANSITION_GRAPH "transition_graph"
// @COMPAT Deprecated since 1.1.12
#define PCMK__XE_ACL_USER "acl_user"
/* @TODO Deprecate somehow. It's undocumented and behaves the same as
* PCMK__XE_CIB in places where it's recognized.
*/
#define PCMK__XE_ALL "all"
// @COMPAT Deprecated since 2.1.7
#define PCMK__XE_DIFF_ADDED "diff-added"
// @COMPAT Deprecated since 2.1.7
#define PCMK__XE_DIFF_REMOVED "diff-removed"
/* @COMPAT Deprecated since 2.0.0; alias for <clone> with PCMK_META_PROMOTABLE
* set to "true"
*/
#define PCMK__XE_PROMOTABLE_LEGACY "master"
// @COMPAT Deprecated since 1.1.12
#define PCMK__XE_ROLE_REF "role_ref"
/*
* XML attribute names used only by internal code
*/
#define PCMK__XA_ATTR_DAMPENING "attr_dampening"
#define PCMK__XA_ATTR_FORCE "attrd_is_force_write"
#define PCMK__XA_ATTR_INTERVAL "attr_clear_interval"
#define PCMK__XA_ATTR_IS_PRIVATE "attr_is_private"
#define PCMK__XA_ATTR_IS_REMOTE "attr_is_remote"
#define PCMK__XA_ATTR_NAME "attr_name"
#define PCMK__XA_ATTR_NODE_ID "attr_host_id"
#define PCMK__XA_ATTR_NODE_NAME "attr_host"
#define PCMK__XA_ATTR_OPERATION "attr_clear_operation"
#define PCMK__XA_ATTR_PATTERN "attr_regex"
#define PCMK__XA_ATTR_RESOURCE "attr_resource"
#define PCMK__XA_ATTR_SECTION "attr_section"
#define PCMK__XA_ATTR_SET "attr_set"
#define PCMK__XA_ATTR_SET_TYPE "attr_set_type"
#define PCMK__XA_ATTR_SYNC_POINT "attr_sync_point"
#define PCMK__XA_ATTR_USER "attr_user"
#define PCMK__XA_ATTR_UUID "attr_key"
#define PCMK__XA_ATTR_VALUE "attr_value"
#define PCMK__XA_ATTR_VERSION "attr_version"
#define PCMK__XA_ATTR_WRITER "attr_writer"
#define PCMK__XA_CALL_ID "call-id"
#define PCMK__XA_CONFIG_ERRORS "config-errors"
#define PCMK__XA_CONFIG_WARNINGS "config-warnings"
#define PCMK__XA_CONFIRM "confirm"
#define PCMK__XA_CONN_HOST "connection_host"
#define PCMK__XA_CRMD "crmd"
#define PCMK__XA_CRMD_STATE "crmd_state"
#define PCMK__XA_CRM_HOST_TO "crm_host_to"
#define PCMK__XA_CRM_LIMIT_MAX "crm-limit-max"
#define PCMK__XA_CRM_LIMIT_MODE "crm-limit-mode"
#define PCMK__XA_CRM_SUBSYSTEM "crm_subsystem"
#define PCMK__XA_CRM_SYS_FROM "crm_sys_from"
#define PCMK__XA_CRM_SYS_TO "crm_sys_to"
#define PCMK__XA_CRM_TASK "crm_task"
#define PCMK__XA_CRM_TGRAPH_IN "crm-tgraph-in"
#define PCMK__XA_CRM_USER "crm_user"
#define PCMK__XA_DC_LEAVING "dc-leaving"
#define PCMK__XA_DIGEST "digest"
#define PCMK__XA_ELECTION_AGE_SEC "election-age-sec"
#define PCMK__XA_ELECTION_AGE_NANO_SEC "election-age-nano-sec"
#define PCMK__XA_ELECTION_ID "election-id"
#define PCMK__XA_ELECTION_OWNER "election-owner"
#define PCMK__XA_EXPECTED "expected"
#define PCMK__XA_FILE "file"
#define PCMK__XA_GRAPH_ERRORS "graph-errors"
#define PCMK__XA_GRAPH_WARNINGS "graph-warnings"
#define PCMK__XA_IN_CCM "in_ccm"
#define PCMK__XA_JOIN "join"
#define PCMK__XA_JOIN_ID "join_id"
#define PCMK__XA_LONG_ID "long-id"
#define PCMK__XA_MODE "mode"
#define PCMK__XA_NODE_FENCED "node_fenced"
#define PCMK__XA_NODE_IN_MAINTENANCE "node_in_maintenance"
#define PCMK__XA_NODE_START_STATE "node_start_state"
#define PCMK__XA_NODE_STATE "node_state"
#define PCMK__XA_OP_DIGEST "op-digest"
#define PCMK__XA_OP_FORCE_RESTART "op-force-restart"
#define PCMK__XA_OP_RESTART_DIGEST "op-restart-digest"
#define PCMK__XA_OP_SECURE_DIGEST "op-secure-digest"
#define PCMK__XA_OP_SECURE_PARAMS "op-secure-params"
#define PCMK__XA_OP_STATUS "op-status"
#define PCMK__XA_PACEMAKERD_STATE "pacemakerd_state"
#define PCMK__XA_PRIORITY "priority"
#define PCMK__XA_RC_CODE "rc-code"
#define PCMK__XA_REAP "reap"
#define PCMK__XA_REMOTE_NODE "remote_node"
/* Actions to be executed on Pacemaker Remote nodes are routed through the
* controller on the cluster node hosting the remote connection. That cluster
* node is considered the router node for the action.
*/
#define PCMK__XA_ROUTER_NODE "router_node"
#define PCMK__XA_RSC_ID "rsc-id"
#define PCMK__XA_SCHEMA "schema"
#define PCMK__XA_SCHEMAS "schemas"
#define PCMK__XA_SRC "src"
#define PCMK__XA_SUBT "subt" // subtype
#define PCMK__XA_T "t" // type
#define PCMK__XA_TASK "task"
#define PCMK__XA_TRANSITION_KEY "transition-key"
#define PCMK__XA_TRANSITION_MAGIC "transition-magic"
#define PCMK__XA_UPTIME "uptime"
// @COMPAT Used only with v1 patchsets
#define PCMK__XA_CRM_DIFF_MARKER "__crm_diff_marker__"
// @COMPAT Deprecated since 2.1.5
#define PCMK__XA_FIRST_INSTANCE "first-instance"
// @COMPAT Deprecated since 1.1.12
#define PCMK__XA_REF "ref"
// @COMPAT Deprecated since 2.1.6
#define PCMK__XA_REPLACE "replace"
// @COMPAT Deprecated since 2.1.5
#define PCMK__XA_RSC_INSTANCE "rsc-instance"
// @COMPAT Deprecated since 1.1.12
#define PCMK__XA_TAG "tag"
// @COMPAT Deprecated since 2.1.5
#define PCMK__XA_THEN_INSTANCE "then-instance"
// @COMPAT Deprecated since 2.1.5
#define PCMK__XA_WITH_RSC_INSTANCE "with-rsc-instance"
/*
* IPC service names that are only used internally
*/
# define PCMK__SERVER_BASED_RO "cib_ro"
# define PCMK__SERVER_BASED_RW "cib_rw"
# define PCMK__SERVER_BASED_SHM "cib_shm"
/*
* IPC commands that can be sent to Pacemaker daemons
*/
#define PCMK__ATTRD_CMD_PEER_REMOVE "peer-remove"
#define PCMK__ATTRD_CMD_UPDATE "update"
#define PCMK__ATTRD_CMD_UPDATE_BOTH "update-both"
#define PCMK__ATTRD_CMD_UPDATE_DELAY "update-delay"
#define PCMK__ATTRD_CMD_QUERY "query"
#define PCMK__ATTRD_CMD_REFRESH "refresh"
#define PCMK__ATTRD_CMD_FLUSH "flush"
#define PCMK__ATTRD_CMD_SYNC "sync"
#define PCMK__ATTRD_CMD_SYNC_RESPONSE "sync-response"
#define PCMK__ATTRD_CMD_CLEAR_FAILURE "clear-failure"
#define PCMK__ATTRD_CMD_CONFIRM "confirm"
#define PCMK__CONTROLD_CMD_NODES "list-nodes"
#endif /* CRM_INTERNAL__H */
diff --git a/lib/pacemaker/pcmk_sched_primitive.c b/lib/pacemaker/pcmk_sched_primitive.c
index afaaf0f7b5..749cabc206 100644
--- a/lib/pacemaker/pcmk_sched_primitive.c
+++ b/lib/pacemaker/pcmk_sched_primitive.c
@@ -1,1676 +1,1677 @@
/*
* 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 <stdbool.h>
#include <stdint.h> // uint8_t, uint32_t
#include <crm/msg_xml.h>
#include <pacemaker-internal.h>
#include "libpacemaker_private.h"
static void stop_resource(pcmk_resource_t *rsc, pcmk_node_t *node,
bool optional);
static void start_resource(pcmk_resource_t *rsc, pcmk_node_t *node,
bool optional);
static void demote_resource(pcmk_resource_t *rsc, pcmk_node_t *node,
bool optional);
static void promote_resource(pcmk_resource_t *rsc, pcmk_node_t *node,
bool optional);
static void assert_role_error(pcmk_resource_t *rsc, pcmk_node_t *node,
bool optional);
#define RSC_ROLE_MAX (pcmk_role_promoted + 1)
static enum rsc_role_e rsc_state_matrix[RSC_ROLE_MAX][RSC_ROLE_MAX] = {
/* This array lists the immediate next role when transitioning from one role
* to a target role. For example, when going from Stopped to Promoted, the
* next role is Unpromoted, because the resource must be started before it
* can be promoted. The current state then becomes Started, which is fed
* into this array again, giving a next role of Promoted.
*
* Current role Immediate next role Final target role
* ------------ ------------------- -----------------
*/
/* Unknown */ { pcmk_role_unknown, /* Unknown */
pcmk_role_stopped, /* Stopped */
pcmk_role_stopped, /* Started */
pcmk_role_stopped, /* Unpromoted */
pcmk_role_stopped, /* Promoted */
},
/* Stopped */ { pcmk_role_stopped, /* Unknown */
pcmk_role_stopped, /* Stopped */
pcmk_role_started, /* Started */
pcmk_role_unpromoted, /* Unpromoted */
pcmk_role_unpromoted, /* Promoted */
},
/* Started */ { pcmk_role_stopped, /* Unknown */
pcmk_role_stopped, /* Stopped */
pcmk_role_started, /* Started */
pcmk_role_unpromoted, /* Unpromoted */
pcmk_role_promoted, /* Promoted */
},
/* Unpromoted */ { pcmk_role_stopped, /* Unknown */
pcmk_role_stopped, /* Stopped */
pcmk_role_stopped, /* Started */
pcmk_role_unpromoted, /* Unpromoted */
pcmk_role_promoted, /* Promoted */
},
/* Promoted */ { pcmk_role_stopped, /* Unknown */
pcmk_role_unpromoted, /* Stopped */
pcmk_role_unpromoted, /* Started */
pcmk_role_unpromoted, /* Unpromoted */
pcmk_role_promoted, /* Promoted */
},
};
/*!
* \internal
* \brief Function to schedule actions needed for a role change
*
* \param[in,out] rsc Resource whose role is changing
* \param[in,out] node Node where resource will be in its next role
* \param[in] optional Whether scheduled actions should be optional
*/
typedef void (*rsc_transition_fn)(pcmk_resource_t *rsc, pcmk_node_t *node,
bool optional);
static rsc_transition_fn rsc_action_matrix[RSC_ROLE_MAX][RSC_ROLE_MAX] = {
/* This array lists the function needed to transition directly from one role
* to another. NULL indicates that nothing is needed.
*
* Current role Transition function Next role
* ------------ ------------------- ----------
*/
/* Unknown */ { assert_role_error, /* Unknown */
stop_resource, /* Stopped */
assert_role_error, /* Started */
assert_role_error, /* Unpromoted */
assert_role_error, /* Promoted */
},
/* Stopped */ { assert_role_error, /* Unknown */
NULL, /* Stopped */
start_resource, /* Started */
start_resource, /* Unpromoted */
assert_role_error, /* Promoted */
},
/* Started */ { assert_role_error, /* Unknown */
stop_resource, /* Stopped */
NULL, /* Started */
NULL, /* Unpromoted */
promote_resource, /* Promoted */
},
/* Unpromoted */ { assert_role_error, /* Unknown */
stop_resource, /* Stopped */
stop_resource, /* Started */
NULL, /* Unpromoted */
promote_resource, /* Promoted */
},
/* Promoted */ { assert_role_error, /* Unknown */
demote_resource, /* Stopped */
demote_resource, /* Started */
demote_resource, /* Unpromoted */
NULL, /* Promoted */
},
};
/*!
* \internal
* \brief Get a list of a resource's allowed nodes sorted by node score
*
* \param[in] rsc Resource to check
*
* \return List of allowed nodes sorted by node score
*/
static GList *
sorted_allowed_nodes(const pcmk_resource_t *rsc)
{
if (rsc->allowed_nodes != NULL) {
GList *nodes = g_hash_table_get_values(rsc->allowed_nodes);
if (nodes != NULL) {
return pcmk__sort_nodes(nodes, pcmk__current_node(rsc));
}
}
return NULL;
}
/*!
* \internal
* \brief Assign a resource to its best allowed node, if possible
*
* \param[in,out] rsc Resource to choose a node for
* \param[in] prefer If not \c NULL, prefer this node when all else
* equal
* \param[in] stop_if_fail If \c true and \p rsc can't be assigned to a
* node, set next role to stopped and update
* existing actions
*
* \return true if \p rsc could be assigned to a node, otherwise false
*
* \note If \p stop_if_fail is \c false, then \c pcmk__unassign_resource() can
* completely undo the assignment. A successful assignment can be either
* undone or left alone as final. A failed assignment has the same effect
* as calling pcmk__unassign_resource(); there are no side effects on
* roles or actions.
*/
static bool
assign_best_node(pcmk_resource_t *rsc, const pcmk_node_t *prefer,
bool stop_if_fail)
{
GList *nodes = NULL;
pcmk_node_t *chosen = NULL;
pcmk_node_t *best = NULL;
const pcmk_node_t *most_free_node = pcmk__ban_insufficient_capacity(rsc);
if (prefer == NULL) {
prefer = most_free_node;
}
if (!pcmk_is_set(rsc->flags, pcmk_rsc_unassigned)) {
// We've already finished assignment of resources to nodes
return rsc->allocated_to != NULL;
}
// Sort allowed nodes by score
nodes = sorted_allowed_nodes(rsc);
if (nodes != NULL) {
best = (pcmk_node_t *) nodes->data; // First node has best score
}
if ((prefer != NULL) && (nodes != NULL)) {
// Get the allowed node version of prefer
chosen = g_hash_table_lookup(rsc->allowed_nodes, prefer->details->id);
if (chosen == NULL) {
pcmk__rsc_trace(rsc, "Preferred node %s for %s was unknown",
pcmk__node_name(prefer), rsc->id);
/* Favor the preferred node as long as its score is at least as good as
* the best allowed node's.
*
* An alternative would be to favor the preferred node even if the best
* node is better, when the best node's score is less than INFINITY.
*/
} else if (chosen->weight < best->weight) {
pcmk__rsc_trace(rsc, "Preferred node %s for %s was unsuitable",
pcmk__node_name(chosen), rsc->id);
chosen = NULL;
} else if (!pcmk__node_available(chosen, true, false)) {
pcmk__rsc_trace(rsc, "Preferred node %s for %s was unavailable",
pcmk__node_name(chosen), rsc->id);
chosen = NULL;
} else {
pcmk__rsc_trace(rsc,
"Chose preferred node %s for %s "
"(ignoring %d candidates)",
pcmk__node_name(chosen), rsc->id,
g_list_length(nodes));
}
}
if ((chosen == NULL) && (best != NULL)) {
/* Either there is no preferred node, or the preferred node is not
* suitable, but another node is allowed to run the resource.
*/
chosen = best;
if (!pe_rsc_is_unique_clone(rsc->parent)
&& (chosen->weight > 0) // Zero not acceptable
&& pcmk__node_available(chosen, false, false)) {
/* If the resource is already running on a node, prefer that node if
* it is just as good as the chosen node.
*
* We don't do this for unique clone instances, because
* pcmk__assign_instances() has already assigned instances to their
* running nodes when appropriate, and if we get here, we don't want
* remaining unassigned instances to prefer a node that's already
* running another instance.
*/
pcmk_node_t *running = pcmk__current_node(rsc);
if (running == NULL) {
// Nothing to do
} else if (!pcmk__node_available(running, true, false)) {
pcmk__rsc_trace(rsc,
"Current node for %s (%s) can't run resources",
rsc->id, pcmk__node_name(running));
} else {
int nodes_with_best_score = 1;
for (GList *iter = nodes->next; iter; iter = iter->next) {
pcmk_node_t *allowed = (pcmk_node_t *) iter->data;
if (allowed->weight != chosen->weight) {
// The nodes are sorted by score, so no more are equal
break;
}
if (pcmk__same_node(allowed, running)) {
// Scores are equal, so prefer the current node
chosen = allowed;
}
nodes_with_best_score++;
}
if (nodes_with_best_score > 1) {
uint8_t log_level = LOG_INFO;
if (chosen->weight >= INFINITY) {
log_level = LOG_WARNING;
}
do_crm_log(log_level,
"Chose %s for %s from %d nodes with score %s",
pcmk__node_name(chosen), rsc->id,
nodes_with_best_score,
pcmk_readable_score(chosen->weight));
}
}
}
pcmk__rsc_trace(rsc, "Chose %s for %s from %d candidates",
pcmk__node_name(chosen), rsc->id, g_list_length(nodes));
}
pcmk__assign_resource(rsc, chosen, false, stop_if_fail);
g_list_free(nodes);
return rsc->allocated_to != NULL;
}
/*!
* \internal
* \brief Apply a "this with" colocation to a node's allowed node scores
*
* \param[in,out] colocation Colocation to apply
* \param[in,out] rsc Resource being assigned
*/
static void
apply_this_with(pcmk__colocation_t *colocation, pcmk_resource_t *rsc)
{
GHashTable *archive = NULL;
pcmk_resource_t *other = colocation->primary;
// In certain cases, we will need to revert the node scores
if ((colocation->dependent_role >= pcmk_role_promoted)
|| ((colocation->score < 0) && (colocation->score > -INFINITY))) {
archive = pcmk__copy_node_table(rsc->allowed_nodes);
}
if (pcmk_is_set(other->flags, pcmk_rsc_unassigned)) {
pcmk__rsc_trace(rsc,
"%s: Assigning colocation %s primary %s first"
"(score=%d role=%s)",
rsc->id, colocation->id, other->id,
colocation->score,
pcmk_role_text(colocation->dependent_role));
other->cmds->assign(other, NULL, true);
}
// Apply the colocation score to this resource's allowed node scores
rsc->cmds->apply_coloc_score(rsc, other, colocation, true);
if ((archive != NULL)
&& !pcmk__any_node_available(rsc->allowed_nodes)) {
pcmk__rsc_info(rsc,
"%s: Reverting scores from colocation with %s "
"because no nodes allowed",
rsc->id, other->id);
g_hash_table_destroy(rsc->allowed_nodes);
rsc->allowed_nodes = archive;
archive = NULL;
}
if (archive != NULL) {
g_hash_table_destroy(archive);
}
}
/*!
* \internal
* \brief Update a Pacemaker Remote node once its connection has been assigned
*
* \param[in] connection Connection resource that has been assigned
*/
static void
remote_connection_assigned(const pcmk_resource_t *connection)
{
pcmk_node_t *remote_node = pe_find_node(connection->cluster->nodes,
connection->id);
CRM_CHECK(remote_node != NULL, return);
if ((connection->allocated_to != NULL)
&& (connection->next_role != pcmk_role_stopped)) {
crm_trace("Pacemaker Remote node %s will be online",
remote_node->details->id);
remote_node->details->online = TRUE;
if (remote_node->details->unseen) {
// Avoid unnecessary fence, since we will attempt connection
remote_node->details->unclean = FALSE;
}
} else {
crm_trace("Pacemaker Remote node %s will be shut down "
"(%sassigned connection's next role is %s)",
remote_node->details->id,
((connection->allocated_to == NULL)? "un" : ""),
pcmk_role_text(connection->next_role));
remote_node->details->shutdown = TRUE;
}
}
/*!
* \internal
* \brief Assign a primitive resource to a node
*
* \param[in,out] rsc Resource to assign to a node
* \param[in] prefer Node to prefer, if all else is equal
* \param[in] stop_if_fail If \c true and \p rsc can't be assigned to a
* node, set next role to stopped and update
* existing actions
*
* \return Node that \p rsc is assigned to, if assigned entirely to one node
*
* \note If \p stop_if_fail is \c false, then \c pcmk__unassign_resource() can
* completely undo the assignment. A successful assignment can be either
* undone or left alone as final. A failed assignment has the same effect
* as calling pcmk__unassign_resource(); there are no side effects on
* roles or actions.
*/
pcmk_node_t *
pcmk__primitive_assign(pcmk_resource_t *rsc, const pcmk_node_t *prefer,
bool stop_if_fail)
{
GList *this_with_colocations = NULL;
GList *with_this_colocations = NULL;
GList *iter = NULL;
pcmk__colocation_t *colocation = NULL;
CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_primitive));
// Never assign a child without parent being assigned first
if ((rsc->parent != NULL)
&& !pcmk_is_set(rsc->parent->flags, pcmk_rsc_assigning)) {
pcmk__rsc_debug(rsc, "%s: Assigning parent %s first",
rsc->id, rsc->parent->id);
rsc->parent->cmds->assign(rsc->parent, prefer, stop_if_fail);
}
if (!pcmk_is_set(rsc->flags, pcmk_rsc_unassigned)) {
// Assignment has already been done
const char *node_name = "no node";
if (rsc->allocated_to != NULL) {
node_name = pcmk__node_name(rsc->allocated_to);
}
pcmk__rsc_debug(rsc, "%s: pre-assigned to %s", rsc->id, node_name);
return rsc->allocated_to;
}
// Ensure we detect assignment loops
if (pcmk_is_set(rsc->flags, pcmk_rsc_assigning)) {
pcmk__rsc_debug(rsc, "Breaking assignment loop involving %s", rsc->id);
return NULL;
}
pcmk__set_rsc_flags(rsc, pcmk_rsc_assigning);
pe__show_node_scores(true, rsc, "Pre-assignment", rsc->allowed_nodes,
rsc->cluster);
this_with_colocations = pcmk__this_with_colocations(rsc);
with_this_colocations = pcmk__with_this_colocations(rsc);
// Apply mandatory colocations first, to satisfy as many as possible
for (iter = this_with_colocations; iter != NULL; iter = iter->next) {
colocation = iter->data;
if ((colocation->score <= -CRM_SCORE_INFINITY)
|| (colocation->score >= CRM_SCORE_INFINITY)) {
apply_this_with(colocation, rsc);
}
}
for (iter = with_this_colocations; iter != NULL; iter = iter->next) {
colocation = iter->data;
if ((colocation->score <= -CRM_SCORE_INFINITY)
|| (colocation->score >= CRM_SCORE_INFINITY)) {
pcmk__add_dependent_scores(colocation, rsc);
}
}
pe__show_node_scores(true, rsc, "Mandatory-colocations",
rsc->allowed_nodes, rsc->cluster);
// Then apply optional colocations
for (iter = this_with_colocations; iter != NULL; iter = iter->next) {
colocation = iter->data;
if ((colocation->score > -CRM_SCORE_INFINITY)
&& (colocation->score < CRM_SCORE_INFINITY)) {
apply_this_with(colocation, rsc);
}
}
for (iter = with_this_colocations; iter != NULL; iter = iter->next) {
colocation = iter->data;
if ((colocation->score > -CRM_SCORE_INFINITY)
&& (colocation->score < CRM_SCORE_INFINITY)) {
pcmk__add_dependent_scores(colocation, rsc);
}
}
g_list_free(this_with_colocations);
g_list_free(with_this_colocations);
if (rsc->next_role == pcmk_role_stopped) {
pcmk__rsc_trace(rsc,
"Banning %s from all nodes because it will be stopped",
rsc->id);
resource_location(rsc, NULL, -INFINITY, PCMK_META_TARGET_ROLE,
rsc->cluster);
} else if ((rsc->next_role > rsc->role)
&& !pcmk_is_set(rsc->cluster->flags, pcmk_sched_quorate)
&& (rsc->cluster->no_quorum_policy == pcmk_no_quorum_freeze)) {
crm_notice("Resource %s cannot be elevated from %s to %s due to "
PCMK_OPT_NO_QUORUM_POLICY "=freeze",
rsc->id, pcmk_role_text(rsc->role),
pcmk_role_text(rsc->next_role));
pe__set_next_role(rsc, rsc->role, PCMK_OPT_NO_QUORUM_POLICY "=freeze");
}
pe__show_node_scores(!pcmk_is_set(rsc->cluster->flags,
pcmk_sched_output_scores),
rsc, __func__, rsc->allowed_nodes, rsc->cluster);
// Unmanage resource if fencing is enabled but no device is configured
if (pcmk_is_set(rsc->cluster->flags, pcmk_sched_fencing_enabled)
&& !pcmk_is_set(rsc->cluster->flags, pcmk_sched_have_fencing)) {
pcmk__clear_rsc_flags(rsc, pcmk_rsc_managed);
}
if (!pcmk_is_set(rsc->flags, pcmk_rsc_managed)) {
// Unmanaged resources stay on their current node
const char *reason = NULL;
pcmk_node_t *assign_to = NULL;
pe__set_next_role(rsc, rsc->role, "unmanaged");
assign_to = pcmk__current_node(rsc);
if (assign_to == NULL) {
reason = "inactive";
} else if (rsc->role == pcmk_role_promoted) {
reason = "promoted";
} else if (pcmk_is_set(rsc->flags, pcmk_rsc_failed)) {
reason = "failed";
} else {
reason = "active";
}
pcmk__rsc_info(rsc, "Unmanaged resource %s assigned to %s: %s", rsc->id,
(assign_to? assign_to->details->uname : "no node"),
reason);
pcmk__assign_resource(rsc, assign_to, true, stop_if_fail);
} else if (pcmk_is_set(rsc->cluster->flags, pcmk_sched_stop_all)) {
// Must stop at some point, but be consistent with stop_if_fail
if (stop_if_fail) {
pcmk__rsc_debug(rsc,
"Forcing %s to stop: " PCMK_OPT_STOP_ALL_RESOURCES,
rsc->id);
}
pcmk__assign_resource(rsc, NULL, true, stop_if_fail);
} else if (!assign_best_node(rsc, prefer, stop_if_fail)) {
// Assignment failed
if (!pcmk_is_set(rsc->flags, pcmk_rsc_removed)) {
pcmk__rsc_info(rsc, "Resource %s cannot run anywhere", rsc->id);
} else if ((rsc->running_on != NULL) && stop_if_fail) {
pcmk__rsc_info(rsc, "Stopping orphan resource %s", rsc->id);
}
}
pcmk__clear_rsc_flags(rsc, pcmk_rsc_assigning);
if (rsc->is_remote_node) {
remote_connection_assigned(rsc);
}
return rsc->allocated_to;
}
/*!
* \internal
* \brief Schedule actions to bring resource down and back to current role
*
* \param[in,out] rsc Resource to restart
* \param[in,out] current Node that resource should be brought down on
* \param[in] need_stop Whether the resource must be stopped
* \param[in] need_promote Whether the resource must be promoted
*
* \return Role that resource would have after scheduled actions are taken
*/
static void
schedule_restart_actions(pcmk_resource_t *rsc, pcmk_node_t *current,
bool need_stop, bool need_promote)
{
enum rsc_role_e role = rsc->role;
enum rsc_role_e next_role;
rsc_transition_fn fn = NULL;
pcmk__set_rsc_flags(rsc, pcmk_rsc_restarting);
// Bring resource down to a stop on its current node
while (role != pcmk_role_stopped) {
next_role = rsc_state_matrix[role][pcmk_role_stopped];
pcmk__rsc_trace(rsc, "Creating %s action to take %s down from %s to %s",
(need_stop? "required" : "optional"), rsc->id,
pcmk_role_text(role), pcmk_role_text(next_role));
fn = rsc_action_matrix[role][next_role];
if (fn == NULL) {
break;
}
fn(rsc, current, !need_stop);
role = next_role;
}
// Bring resource up to its next role on its next node
while ((rsc->role <= rsc->next_role) && (role != rsc->role)
&& !pcmk_is_set(rsc->flags, pcmk_rsc_blocked)) {
bool required = need_stop;
next_role = rsc_state_matrix[role][rsc->role];
if ((next_role == pcmk_role_promoted) && need_promote) {
required = true;
}
pcmk__rsc_trace(rsc, "Creating %s action to take %s up from %s to %s",
(required? "required" : "optional"), rsc->id,
pcmk_role_text(role), pcmk_role_text(next_role));
fn = rsc_action_matrix[role][next_role];
if (fn == NULL) {
break;
}
fn(rsc, rsc->allocated_to, !required);
role = next_role;
}
pcmk__clear_rsc_flags(rsc, pcmk_rsc_restarting);
}
/*!
* \internal
* \brief If a resource's next role is not explicitly specified, set a default
*
* \param[in,out] rsc Resource to set next role for
*
* \return "explicit" if next role was explicitly set, otherwise "implicit"
*/
static const char *
set_default_next_role(pcmk_resource_t *rsc)
{
if (rsc->next_role != pcmk_role_unknown) {
return "explicit";
}
if (rsc->allocated_to == NULL) {
pe__set_next_role(rsc, pcmk_role_stopped, "assignment");
} else {
pe__set_next_role(rsc, pcmk_role_started, "assignment");
}
return "implicit";
}
/*!
* \internal
* \brief Create an action to represent an already pending start
*
* \param[in,out] rsc Resource to create start action for
*/
static void
create_pending_start(pcmk_resource_t *rsc)
{
pcmk_action_t *start = NULL;
pcmk__rsc_trace(rsc,
"Creating action for %s to represent already pending start",
rsc->id);
start = start_action(rsc, rsc->allocated_to, TRUE);
pcmk__set_action_flags(start, pcmk_action_always_in_graph);
}
/*!
* \internal
* \brief Schedule actions needed to take a resource to its next role
*
* \param[in,out] rsc Resource to schedule actions for
*/
static void
schedule_role_transition_actions(pcmk_resource_t *rsc)
{
enum rsc_role_e role = rsc->role;
while (role != rsc->next_role) {
enum rsc_role_e next_role = rsc_state_matrix[role][rsc->next_role];
rsc_transition_fn fn = NULL;
pcmk__rsc_trace(rsc,
"Creating action to take %s from %s to %s "
"(ending at %s)",
rsc->id, pcmk_role_text(role),
pcmk_role_text(next_role),
pcmk_role_text(rsc->next_role));
fn = rsc_action_matrix[role][next_role];
if (fn == NULL) {
break;
}
fn(rsc, rsc->allocated_to, false);
role = next_role;
}
}
/*!
* \internal
* \brief Create all actions needed for a given primitive resource
*
* \param[in,out] rsc Primitive resource to create actions for
*/
void
pcmk__primitive_create_actions(pcmk_resource_t *rsc)
{
bool need_stop = false;
bool need_promote = false;
bool is_moving = false;
bool allow_migrate = false;
bool multiply_active = false;
pcmk_node_t *current = NULL;
unsigned int num_all_active = 0;
unsigned int num_clean_active = 0;
const char *next_role_source = NULL;
CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_primitive));
next_role_source = set_default_next_role(rsc);
pcmk__rsc_trace(rsc,
"Creating all actions for %s transition from %s to %s "
"(%s) on %s",
rsc->id, pcmk_role_text(rsc->role),
pcmk_role_text(rsc->next_role), next_role_source,
pcmk__node_name(rsc->allocated_to));
current = rsc->fns->active_node(rsc, &num_all_active, &num_clean_active);
g_list_foreach(rsc->dangling_migrations, pcmk__abort_dangling_migration,
rsc);
if ((current != NULL) && (rsc->allocated_to != NULL)
&& !pcmk__same_node(current, rsc->allocated_to)
&& (rsc->next_role >= pcmk_role_started)) {
pcmk__rsc_trace(rsc, "Moving %s from %s to %s",
rsc->id, pcmk__node_name(current),
pcmk__node_name(rsc->allocated_to));
is_moving = true;
allow_migrate = pcmk__rsc_can_migrate(rsc, current);
// This is needed even if migrating (though I'm not sure why ...)
need_stop = true;
}
// Check whether resource is partially migrated and/or multiply active
if ((rsc->partial_migration_source != NULL)
&& (rsc->partial_migration_target != NULL)
&& allow_migrate && (num_all_active == 2)
&& pcmk__same_node(current, rsc->partial_migration_source)
&& pcmk__same_node(rsc->allocated_to, rsc->partial_migration_target)) {
/* A partial migration is in progress, and the migration target remains
* the same as when the migration began.
*/
pcmk__rsc_trace(rsc,
"Partial migration of %s from %s to %s will continue",
rsc->id, pcmk__node_name(rsc->partial_migration_source),
pcmk__node_name(rsc->partial_migration_target));
} else if ((rsc->partial_migration_source != NULL)
|| (rsc->partial_migration_target != NULL)) {
// A partial migration is in progress but can't be continued
if (num_all_active > 2) {
// The resource is migrating *and* multiply active!
crm_notice("Forcing recovery of %s because it is migrating "
"from %s to %s and possibly active elsewhere",
rsc->id, pcmk__node_name(rsc->partial_migration_source),
pcmk__node_name(rsc->partial_migration_target));
} else {
// The migration source or target isn't available
crm_notice("Forcing recovery of %s because it can no longer "
"migrate from %s to %s",
rsc->id, pcmk__node_name(rsc->partial_migration_source),
pcmk__node_name(rsc->partial_migration_target));
}
need_stop = true;
rsc->partial_migration_source = rsc->partial_migration_target = NULL;
allow_migrate = false;
} else if (pcmk_is_set(rsc->flags, pcmk_rsc_needs_fencing)) {
multiply_active = (num_all_active > 1);
} else {
/* If a resource has "requires" set to nothing or quorum, don't consider
* it active on unclean nodes (similar to how all resources behave when
* PCMK_OPT_STONITH_ENABLED is false). We can start such resources
* elsewhere before fencing completes, and if we considered the resource
* active on the failed node, we would attempt recovery for being active
* on multiple nodes.
*/
multiply_active = (num_clean_active > 1);
}
if (multiply_active) {
const char *class = crm_element_value(rsc->xml, PCMK_XA_CLASS);
// Resource was (possibly) incorrectly multiply active
pcmk__sched_err("%s resource %s might be active on %u nodes (%s)",
pcmk__s(class, "Untyped"), rsc->id, num_all_active,
pcmk_multiply_active_text(rsc->recovery_type));
crm_notice("See https://wiki.clusterlabs.org/wiki/FAQ"
"#Resource_is_Too_Active for more information");
switch (rsc->recovery_type) {
case pcmk_multiply_active_restart:
need_stop = true;
break;
case pcmk_multiply_active_unexpected:
need_stop = true; // stop_resource() will skip expected node
pcmk__set_rsc_flags(rsc, pcmk_rsc_stop_unexpected);
break;
default:
break;
}
} else {
pcmk__clear_rsc_flags(rsc, pcmk_rsc_stop_unexpected);
}
if (pcmk_is_set(rsc->flags, pcmk_rsc_start_pending)) {
create_pending_start(rsc);
}
if (is_moving) {
// Remaining tests are only for resources staying where they are
} else if (pcmk_is_set(rsc->flags, pcmk_rsc_failed)) {
if (pcmk_is_set(rsc->flags, pcmk_rsc_stop_if_failed)) {
need_stop = true;
pcmk__rsc_trace(rsc, "Recovering %s", rsc->id);
} else {
pcmk__rsc_trace(rsc, "Recovering %s by demotion", rsc->id);
if (rsc->next_role == pcmk_role_promoted) {
need_promote = true;
}
}
} else if (pcmk_is_set(rsc->flags, pcmk_rsc_blocked)) {
pcmk__rsc_trace(rsc, "Blocking further actions on %s", rsc->id);
need_stop = true;
} else if ((rsc->role > pcmk_role_started) && (current != NULL)
&& (rsc->allocated_to != NULL)) {
pcmk_action_t *start = NULL;
pcmk__rsc_trace(rsc, "Creating start action for promoted resource %s",
rsc->id);
start = start_action(rsc, rsc->allocated_to, TRUE);
if (!pcmk_is_set(start->flags, pcmk_action_optional)) {
// Recovery of a promoted resource
pcmk__rsc_trace(rsc, "%s restart is required for recovery", rsc->id);
need_stop = true;
}
}
// Create any actions needed to bring resource down and back up to same role
schedule_restart_actions(rsc, current, need_stop, need_promote);
// Create any actions needed to take resource from this role to the next
schedule_role_transition_actions(rsc);
pcmk__create_recurring_actions(rsc);
if (allow_migrate) {
pcmk__create_migration_actions(rsc, current);
}
}
/*!
* \internal
* \brief Ban a resource from any allowed nodes that are Pacemaker Remote nodes
*
* \param[in] rsc Resource to check
*/
static void
rsc_avoids_remote_nodes(const pcmk_resource_t *rsc)
{
GHashTableIter iter;
pcmk_node_t *node = NULL;
g_hash_table_iter_init(&iter, rsc->allowed_nodes);
while (g_hash_table_iter_next(&iter, NULL, (void **) &node)) {
if (node->details->remote_rsc != NULL) {
node->weight = -INFINITY;
}
}
}
/*!
* \internal
* \brief Return allowed nodes as (possibly sorted) list
*
* Convert a resource's hash table of allowed nodes to a list. If printing to
* stdout, sort the list, to keep action ID numbers consistent for regression
* test output (while avoiding the performance hit on a live cluster).
*
* \param[in] rsc Resource to check for allowed nodes
*
* \return List of resource's allowed nodes
* \note Callers should take care not to rely on the list being sorted.
*/
static GList *
allowed_nodes_as_list(const pcmk_resource_t *rsc)
{
GList *allowed_nodes = NULL;
if (rsc->allowed_nodes) {
allowed_nodes = g_hash_table_get_values(rsc->allowed_nodes);
}
if (!pcmk__is_daemon) {
allowed_nodes = g_list_sort(allowed_nodes, pe__cmp_node_name);
}
return allowed_nodes;
}
/*!
* \internal
* \brief Create implicit constraints needed for a primitive resource
*
* \param[in,out] rsc Primitive resource to create implicit constraints for
*/
void
pcmk__primitive_internal_constraints(pcmk_resource_t *rsc)
{
GList *allowed_nodes = NULL;
bool check_unfencing = false;
bool check_utilization = false;
CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_primitive));
if (!pcmk_is_set(rsc->flags, pcmk_rsc_managed)) {
pcmk__rsc_trace(rsc,
"Skipping implicit constraints for unmanaged resource "
"%s", rsc->id);
return;
}
// Whether resource requires unfencing
check_unfencing = !pcmk_is_set(rsc->flags, pcmk_rsc_fence_device)
&& pcmk_is_set(rsc->cluster->flags,
pcmk_sched_enable_unfencing)
&& pcmk_is_set(rsc->flags, pcmk_rsc_needs_unfencing);
// Whether a non-default placement strategy is used
check_utilization = (g_hash_table_size(rsc->utilization) > 0)
&& !pcmk__str_eq(rsc->cluster->placement_strategy,
"default", pcmk__str_casei);
// Order stops before starts (i.e. restart)
pcmk__new_ordering(rsc, pcmk__op_key(rsc->id, PCMK_ACTION_STOP, 0), NULL,
rsc, pcmk__op_key(rsc->id, PCMK_ACTION_START, 0), NULL,
pcmk__ar_ordered
|pcmk__ar_first_implies_then
|pcmk__ar_intermediate_stop,
rsc->cluster);
// Promotable ordering: demote before stop, start before promote
if (pcmk_is_set(pe__const_top_resource(rsc, false)->flags,
pcmk_rsc_promotable)
|| (rsc->role > pcmk_role_unpromoted)) {
pcmk__new_ordering(rsc, pcmk__op_key(rsc->id, PCMK_ACTION_DEMOTE, 0),
NULL,
rsc, pcmk__op_key(rsc->id, PCMK_ACTION_STOP, 0),
NULL,
pcmk__ar_promoted_then_implies_first, rsc->cluster);
pcmk__new_ordering(rsc, pcmk__op_key(rsc->id, PCMK_ACTION_START, 0),
NULL,
rsc, pcmk__op_key(rsc->id, PCMK_ACTION_PROMOTE, 0),
NULL,
pcmk__ar_unrunnable_first_blocks, rsc->cluster);
}
// Don't clear resource history if probing on same node
pcmk__new_ordering(rsc, pcmk__op_key(rsc->id, PCMK_ACTION_LRM_DELETE, 0),
NULL, rsc,
pcmk__op_key(rsc->id, PCMK_ACTION_MONITOR, 0),
NULL,
pcmk__ar_if_on_same_node|pcmk__ar_then_cancels_first,
rsc->cluster);
// Certain checks need allowed nodes
if (check_unfencing || check_utilization || (rsc->container != NULL)) {
allowed_nodes = allowed_nodes_as_list(rsc);
}
if (check_unfencing) {
g_list_foreach(allowed_nodes, pcmk__order_restart_vs_unfence, rsc);
}
if (check_utilization) {
pcmk__create_utilization_constraints(rsc, allowed_nodes);
}
if (rsc->container != NULL) {
pcmk_resource_t *remote_rsc = NULL;
if (rsc->is_remote_node) {
// rsc is the implicit remote connection for a guest or bundle node
/* Guest resources are not allowed to run on Pacemaker Remote nodes,
* to avoid nesting remotes. However, bundles are allowed.
*/
if (!pcmk_is_set(rsc->flags, pcmk_rsc_remote_nesting_allowed)) {
rsc_avoids_remote_nodes(rsc->container);
}
/* If someone cleans up a guest or bundle node's container, we will
* likely schedule a (re-)probe of the container and recovery of the
* connection. Order the connection stop after the container probe,
* so that if we detect the container running, we will trigger a new
* transition and avoid the unnecessary recovery.
*/
pcmk__order_resource_actions(rsc->container, PCMK_ACTION_MONITOR,
rsc, PCMK_ACTION_STOP,
pcmk__ar_ordered);
/* A user can specify that a resource must start on a Pacemaker Remote
* node by explicitly configuring it with the container=NODENAME
* meta-attribute. This is of questionable merit, since location
* constraints can accomplish the same thing. But we support it, so here
* we check whether a resource (that is not itself a remote connection)
* has container set to a remote node or guest node resource.
*/
} else if (rsc->container->is_remote_node) {
remote_rsc = rsc->container;
} else {
remote_rsc = pe__resource_contains_guest_node(rsc->cluster,
rsc->container);
}
if (remote_rsc != NULL) {
/* Force the resource on the Pacemaker Remote node instead of
* colocating the resource with the container resource.
*/
for (GList *item = allowed_nodes; item; item = item->next) {
pcmk_node_t *node = item->data;
if (node->details->remote_rsc != remote_rsc) {
node->weight = -INFINITY;
}
}
} else {
/* This resource is either a filler for a container that does NOT
* represent a Pacemaker Remote node, or a Pacemaker Remote
* connection resource for a guest node or bundle.
*/
int score;
crm_trace("Order and colocate %s relative to its container %s",
rsc->id, rsc->container->id);
pcmk__new_ordering(rsc->container,
pcmk__op_key(rsc->container->id,
PCMK_ACTION_START, 0),
NULL, rsc,
pcmk__op_key(rsc->id, PCMK_ACTION_START, 0),
NULL,
pcmk__ar_first_implies_then
|pcmk__ar_unrunnable_first_blocks,
rsc->cluster);
pcmk__new_ordering(rsc,
pcmk__op_key(rsc->id, PCMK_ACTION_STOP, 0),
NULL,
rsc->container,
pcmk__op_key(rsc->container->id,
PCMK_ACTION_STOP, 0),
NULL, pcmk__ar_then_implies_first, rsc->cluster);
if (pcmk_is_set(rsc->flags, pcmk_rsc_remote_nesting_allowed)) {
score = 10000; /* Highly preferred but not essential */
} else {
score = INFINITY; /* Force them to run on the same host */
}
pcmk__new_colocation("#resource-with-container", NULL, score, rsc,
rsc->container, NULL, NULL,
pcmk__coloc_influence);
}
}
if (rsc->is_remote_node
|| pcmk_is_set(rsc->flags, pcmk_rsc_fence_device)) {
/* Remote connections and fencing devices are not allowed to run on
* Pacemaker Remote nodes
*/
rsc_avoids_remote_nodes(rsc);
}
g_list_free(allowed_nodes);
}
/*!
* \internal
* \brief Apply a colocation's score to node scores or resource priority
*
* Given a colocation constraint, apply its score to the dependent's
* allowed node scores (if we are still placing resources) or priority (if
* we are choosing promotable clone instance roles).
*
* \param[in,out] dependent Dependent resource in colocation
* \param[in] primary Primary resource in colocation
* \param[in] colocation Colocation constraint to apply
* \param[in] for_dependent true if called on behalf of dependent
*/
void
pcmk__primitive_apply_coloc_score(pcmk_resource_t *dependent,
const pcmk_resource_t *primary,
const pcmk__colocation_t *colocation,
bool for_dependent)
{
enum pcmk__coloc_affects filter_results;
CRM_ASSERT((dependent != NULL) && (primary != NULL)
&& (colocation != NULL));
if (for_dependent) {
// Always process on behalf of primary resource
primary->cmds->apply_coloc_score(dependent, primary, colocation, false);
return;
}
filter_results = pcmk__colocation_affects(dependent, primary, colocation,
false);
pcmk__rsc_trace(dependent, "%s %s with %s (%s, score=%d, filter=%d)",
((colocation->score > 0)? "Colocating" : "Anti-colocating"),
dependent->id, primary->id, colocation->id,
colocation->score,
filter_results);
switch (filter_results) {
case pcmk__coloc_affects_role:
pcmk__apply_coloc_to_priority(dependent, primary, colocation);
break;
case pcmk__coloc_affects_location:
pcmk__apply_coloc_to_scores(dependent, primary, colocation);
break;
default: // pcmk__coloc_affects_nothing
return;
}
}
/* Primitive implementation of
* pcmk_assignment_methods_t:with_this_colocations()
*/
void
pcmk__with_primitive_colocations(const pcmk_resource_t *rsc,
const pcmk_resource_t *orig_rsc, GList **list)
{
CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_primitive)
&& (list != NULL));
if (rsc == orig_rsc) {
/* For the resource itself, add all of its own colocations and relevant
* colocations from its parent (if any).
*/
pcmk__add_with_this_list(list, rsc->rsc_cons_lhs, orig_rsc);
if (rsc->parent != NULL) {
rsc->parent->cmds->with_this_colocations(rsc->parent, orig_rsc, list);
}
} else {
// For an ancestor, add only explicitly configured constraints
for (GList *iter = rsc->rsc_cons_lhs; iter != NULL; iter = iter->next) {
pcmk__colocation_t *colocation = iter->data;
if (pcmk_is_set(colocation->flags, pcmk__coloc_explicit)) {
pcmk__add_with_this(list, colocation, orig_rsc);
}
}
}
}
/* Primitive implementation of
* pcmk_assignment_methods_t:this_with_colocations()
*/
void
pcmk__primitive_with_colocations(const pcmk_resource_t *rsc,
const pcmk_resource_t *orig_rsc, GList **list)
{
CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_primitive)
&& (list != NULL));
if (rsc == orig_rsc) {
/* For the resource itself, add all of its own colocations and relevant
* colocations from its parent (if any).
*/
pcmk__add_this_with_list(list, rsc->rsc_cons, orig_rsc);
if (rsc->parent != NULL) {
rsc->parent->cmds->this_with_colocations(rsc->parent, orig_rsc, list);
}
} else {
// For an ancestor, add only explicitly configured constraints
for (GList *iter = rsc->rsc_cons; iter != NULL; iter = iter->next) {
pcmk__colocation_t *colocation = iter->data;
if (pcmk_is_set(colocation->flags, pcmk__coloc_explicit)) {
pcmk__add_this_with(list, colocation, orig_rsc);
}
}
}
}
/*!
* \internal
* \brief Return action flags for a given primitive resource action
*
* \param[in,out] action Action to get flags for
* \param[in] node If not NULL, limit effects to this node (ignored)
*
* \return Flags appropriate to \p action on \p node
*/
uint32_t
pcmk__primitive_action_flags(pcmk_action_t *action, const pcmk_node_t *node)
{
CRM_ASSERT(action != NULL);
return (uint32_t) action->flags;
}
/*!
* \internal
* \brief Check whether a node is a multiply active resource's expected node
*
* \param[in] rsc Resource to check
* \param[in] node Node to check
*
* \return \c true if \p rsc is multiply active with
* \c PCMK_META_MULTIPLE_ACTIVE set to \c stop_unexpected, and \p node
* is the node where it will remain active
* \note This assumes that the resource's next role cannot be changed to stopped
* after this is called, which should be reasonable if status has already
* been unpacked and resources have been assigned to nodes.
*/
static bool
is_expected_node(const pcmk_resource_t *rsc, const pcmk_node_t *node)
{
return pcmk_all_flags_set(rsc->flags,
pcmk_rsc_stop_unexpected|pcmk_rsc_restarting)
&& (rsc->next_role > pcmk_role_stopped)
&& pcmk__same_node(rsc->allocated_to, node);
}
/*!
* \internal
* \brief Schedule actions needed to stop a resource wherever it is active
*
* \param[in,out] rsc Resource being stopped
* \param[in] node Node where resource is being stopped (ignored)
* \param[in] optional Whether actions should be optional
*/
static void
stop_resource(pcmk_resource_t *rsc, pcmk_node_t *node, bool optional)
{
for (GList *iter = rsc->running_on; iter != NULL; iter = iter->next) {
pcmk_node_t *current = (pcmk_node_t *) iter->data;
pcmk_action_t *stop = NULL;
if (is_expected_node(rsc, current)) {
/* We are scheduling restart actions for a multiply active resource
* with PCMK_META_MULTIPLE_ACTIVE=stop_unexpected, and this is where
* it should not be stopped.
*/
pcmk__rsc_trace(rsc,
"Skipping stop of multiply active resource %s "
"on expected node %s",
rsc->id, pcmk__node_name(current));
continue;
}
if (rsc->partial_migration_target != NULL) {
// Continue migration if node originally was and remains target
if (pcmk__same_node(current, rsc->partial_migration_target)
&& pcmk__same_node(current, rsc->allocated_to)) {
pcmk__rsc_trace(rsc,
"Skipping stop of %s on %s "
"because partial migration there will continue",
rsc->id, pcmk__node_name(current));
continue;
} else {
pcmk__rsc_trace(rsc,
"Forcing stop of %s on %s "
"because migration target changed",
rsc->id, pcmk__node_name(current));
optional = false;
}
}
pcmk__rsc_trace(rsc, "Scheduling stop of %s on %s",
rsc->id, pcmk__node_name(current));
stop = stop_action(rsc, current, optional);
if (rsc->allocated_to == NULL) {
pe_action_set_reason(stop, "node availability", true);
} else if (pcmk_all_flags_set(rsc->flags, pcmk_rsc_restarting
|pcmk_rsc_stop_unexpected)) {
/* We are stopping a multiply active resource on a node that is
* not its expected node, and we are still scheduling restart
* actions, so the stop is for being multiply active.
*/
pe_action_set_reason(stop, "being multiply active", true);
}
if (!pcmk_is_set(rsc->flags, pcmk_rsc_managed)) {
pcmk__clear_action_flags(stop, pcmk_action_runnable);
}
if (pcmk_is_set(rsc->cluster->flags, pcmk_sched_remove_after_stop)) {
pcmk__schedule_cleanup(rsc, current, optional);
}
if (pcmk_is_set(rsc->flags, pcmk_rsc_needs_unfencing)) {
pcmk_action_t *unfence = pe_fence_op(current, PCMK_ACTION_ON, true,
NULL, false, rsc->cluster);
order_actions(stop, unfence, pcmk__ar_then_implies_first);
if (!pcmk__node_unfenced(current)) {
pcmk__sched_err("Stopping %s until %s can be unfenced",
rsc->id, pcmk__node_name(current));
}
}
}
}
/*!
* \internal
* \brief Schedule actions needed to start a resource on a node
*
* \param[in,out] rsc Resource being started
* \param[in,out] node Node where resource should be started
* \param[in] optional Whether actions should be optional
*/
static void
start_resource(pcmk_resource_t *rsc, pcmk_node_t *node, bool optional)
{
pcmk_action_t *start = NULL;
CRM_ASSERT(node != NULL);
pcmk__rsc_trace(rsc, "Scheduling %s start of %s on %s (score %d)",
(optional? "optional" : "required"), rsc->id,
pcmk__node_name(node), node->weight);
start = start_action(rsc, node, TRUE);
pcmk__order_vs_unfence(rsc, node, start, pcmk__ar_first_implies_then);
if (pcmk_is_set(start->flags, pcmk_action_runnable) && !optional) {
pcmk__clear_action_flags(start, pcmk_action_optional);
}
if (is_expected_node(rsc, node)) {
/* This could be a problem if the start becomes necessary for other
* reasons later.
*/
pcmk__rsc_trace(rsc,
"Start of multiply active resouce %s "
"on expected node %s will be a pseudo-action",
rsc->id, pcmk__node_name(node));
pcmk__set_action_flags(start, pcmk_action_pseudo);
}
}
/*!
* \internal
* \brief Schedule actions needed to promote a resource on a node
*
* \param[in,out] rsc Resource being promoted
* \param[in] node Node where resource should be promoted
* \param[in] optional Whether actions should be optional
*/
static void
promote_resource(pcmk_resource_t *rsc, pcmk_node_t *node, bool optional)
{
GList *iter = NULL;
GList *action_list = NULL;
bool runnable = true;
CRM_ASSERT(node != NULL);
// Any start must be runnable for promotion to be runnable
action_list = pe__resource_actions(rsc, node, PCMK_ACTION_START, true);
for (iter = action_list; iter != NULL; iter = iter->next) {
pcmk_action_t *start = (pcmk_action_t *) iter->data;
if (!pcmk_is_set(start->flags, pcmk_action_runnable)) {
runnable = false;
}
}
g_list_free(action_list);
if (runnable) {
pcmk_action_t *promote = promote_action(rsc, node, optional);
pcmk__rsc_trace(rsc, "Scheduling %s promotion of %s on %s",
(optional? "optional" : "required"), rsc->id,
pcmk__node_name(node));
if (is_expected_node(rsc, node)) {
/* This could be a problem if the promote becomes necessary for
* other reasons later.
*/
pcmk__rsc_trace(rsc,
"Promotion of multiply active resouce %s "
"on expected node %s will be a pseudo-action",
rsc->id, pcmk__node_name(node));
pcmk__set_action_flags(promote, pcmk_action_pseudo);
}
} else {
pcmk__rsc_trace(rsc, "Not promoting %s on %s: start unrunnable",
rsc->id, pcmk__node_name(node));
action_list = pe__resource_actions(rsc, node, PCMK_ACTION_PROMOTE,
true);
for (iter = action_list; iter != NULL; iter = iter->next) {
pcmk_action_t *promote = (pcmk_action_t *) iter->data;
pcmk__clear_action_flags(promote, pcmk_action_runnable);
}
g_list_free(action_list);
}
}
/*!
* \internal
* \brief Schedule actions needed to demote a resource wherever it is active
*
* \param[in,out] rsc Resource being demoted
* \param[in] node Node where resource should be demoted (ignored)
* \param[in] optional Whether actions should be optional
*/
static void
demote_resource(pcmk_resource_t *rsc, pcmk_node_t *node, bool optional)
{
/* Since this will only be called for a primitive (possibly as an instance
* of a collective resource), the resource is multiply active if it is
* running on more than one node, so we want to demote on all of them as
* part of recovery, regardless of which one is the desired node.
*/
for (GList *iter = rsc->running_on; iter != NULL; iter = iter->next) {
pcmk_node_t *current = (pcmk_node_t *) iter->data;
if (is_expected_node(rsc, current)) {
pcmk__rsc_trace(rsc,
"Skipping demote of multiply active resource %s "
"on expected node %s",
rsc->id, pcmk__node_name(current));
} else {
pcmk__rsc_trace(rsc, "Scheduling %s demotion of %s on %s",
(optional? "optional" : "required"), rsc->id,
pcmk__node_name(current));
demote_action(rsc, current, optional);
}
}
}
static void
assert_role_error(pcmk_resource_t *rsc, pcmk_node_t *node, bool optional)
{
CRM_ASSERT(false);
}
/*!
* \internal
* \brief Schedule cleanup of a resource
*
* \param[in,out] rsc Resource to clean up
* \param[in] node Node to clean up on
* \param[in] optional Whether clean-up should be optional
*/
void
pcmk__schedule_cleanup(pcmk_resource_t *rsc, const pcmk_node_t *node,
bool optional)
{
/* If the cleanup is required, its orderings are optional, because they're
* relevant only if both actions are required. Conversely, if the cleanup is
* optional, the orderings make the then action required if the first action
* becomes required.
*/
uint32_t flag = optional? pcmk__ar_first_implies_then : pcmk__ar_ordered;
CRM_CHECK((rsc != NULL) && (node != NULL), return);
if (pcmk_is_set(rsc->flags, pcmk_rsc_failed)) {
pcmk__rsc_trace(rsc, "Skipping clean-up of %s on %s: resource failed",
rsc->id, pcmk__node_name(node));
return;
}
if (node->details->unclean || !node->details->online) {
pcmk__rsc_trace(rsc, "Skipping clean-up of %s on %s: node unavailable",
rsc->id, pcmk__node_name(node));
return;
}
crm_notice("Scheduling clean-up of %s on %s",
rsc->id, pcmk__node_name(node));
delete_action(rsc, node, optional);
// stop -> clean-up -> start
pcmk__order_resource_actions(rsc, PCMK_ACTION_STOP,
rsc, PCMK_ACTION_DELETE, flag);
pcmk__order_resource_actions(rsc, PCMK_ACTION_DELETE,
rsc, PCMK_ACTION_START, flag);
}
/*!
* \internal
* \brief Add primitive meta-attributes relevant to graph actions to XML
*
* \param[in] rsc Primitive resource whose meta-attributes should be added
* \param[in,out] xml Transition graph action attributes XML to add to
*/
void
pcmk__primitive_add_graph_meta(const pcmk_resource_t *rsc, xmlNode *xml)
{
char *name = NULL;
char *value = NULL;
const pcmk_resource_t *parent = NULL;
CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_primitive)
&& (xml != NULL));
/* Clone instance numbers get set internally as meta-attributes, and are
* needed in the transition graph (for example, to tell unique clone
* instances apart).
*/
value = g_hash_table_lookup(rsc->meta, PCMK__META_CLONE);
if (value != NULL) {
name = crm_meta_name(PCMK__META_CLONE);
crm_xml_add(xml, name, value);
free(name);
}
// Not sure if this one is really needed ...
value = g_hash_table_lookup(rsc->meta, PCMK_META_REMOTE_NODE);
if (value != NULL) {
name = crm_meta_name(PCMK_META_REMOTE_NODE);
crm_xml_add(xml, name, value);
free(name);
}
/* The container meta-attribute can be set on the primitive itself or one of
* its parents (for example, a group inside a container resource), so check
* them all, and keep the highest one found.
*/
for (parent = rsc; parent != NULL; parent = parent->parent) {
if (parent->container != NULL) {
crm_xml_add(xml, CRM_META "_" PCMK__META_CONTAINER,
parent->container->id);
}
}
/* Bundle replica children will get their external-ip set internally as a
* meta-attribute. The graph action needs it, but under a different naming
* convention than other meta-attributes.
*/
value = g_hash_table_lookup(rsc->meta, "external-ip");
if (value != NULL) {
crm_xml_add(xml, "pcmk_external_ip", value);
}
}
// Primitive implementation of pcmk_assignment_methods_t:add_utilization()
void
pcmk__primitive_add_utilization(const pcmk_resource_t *rsc,
const pcmk_resource_t *orig_rsc,
GList *all_rscs, GHashTable *utilization)
{
CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_primitive)
&& (orig_rsc != NULL) && (utilization != NULL));
if (!pcmk_is_set(rsc->flags, pcmk_rsc_unassigned)) {
return;
}
pcmk__rsc_trace(orig_rsc,
"%s: Adding primitive %s as colocated utilization",
orig_rsc->id, rsc->id);
pcmk__release_node_capacity(utilization, rsc);
}
/*!
* \internal
* \brief Get epoch time of node's shutdown attribute (or now if none)
*
* \param[in,out] node Node to check
*
* \return Epoch time corresponding to shutdown attribute if set or now if not
*/
static time_t
shutdown_time(pcmk_node_t *node)
{
- const char *shutdown = pe_node_attribute_raw(node, XML_CIB_ATTR_SHUTDOWN);
+ const char *shutdown = pe_node_attribute_raw(node,
+ PCMK__NODE_ATTR_SHUTDOWN);
time_t result = 0;
if (shutdown != NULL) {
long long result_ll;
if (pcmk__scan_ll(shutdown, &result_ll, 0LL) == pcmk_rc_ok) {
result = (time_t) result_ll;
}
}
return (result == 0)? get_effective_time(node->details->data_set) : result;
}
/*!
* \internal
* \brief Ban a resource from a node if it's not locked to the node
*
* \param[in] data Node to check
* \param[in,out] user_data Resource to check
*/
static void
ban_if_not_locked(gpointer data, gpointer user_data)
{
const pcmk_node_t *node = (const pcmk_node_t *) data;
pcmk_resource_t *rsc = (pcmk_resource_t *) user_data;
if (strcmp(node->details->uname, rsc->lock_node->details->uname) != 0) {
resource_location(rsc, node, -CRM_SCORE_INFINITY,
PCMK_OPT_SHUTDOWN_LOCK, rsc->cluster);
}
}
// Primitive implementation of pcmk_assignment_methods_t:shutdown_lock()
void
pcmk__primitive_shutdown_lock(pcmk_resource_t *rsc)
{
const char *class = NULL;
CRM_ASSERT((rsc != NULL) && (rsc->variant == pcmk_rsc_variant_primitive));
class = crm_element_value(rsc->xml, PCMK_XA_CLASS);
// Fence devices and remote connections can't be locked
if (pcmk__str_eq(class, PCMK_RESOURCE_CLASS_STONITH, pcmk__str_null_matches)
|| pe__resource_is_remote_conn(rsc)) {
return;
}
if (rsc->lock_node != NULL) {
// The lock was obtained from resource history
if (rsc->running_on != NULL) {
/* The resource was started elsewhere even though it is now
* considered locked. This shouldn't be possible, but as a
* failsafe, we don't want to disturb the resource now.
*/
pcmk__rsc_info(rsc,
"Cancelling shutdown lock "
"because %s is already active", rsc->id);
pe__clear_resource_history(rsc, rsc->lock_node);
rsc->lock_node = NULL;
rsc->lock_time = 0;
}
// Only a resource active on exactly one node can be locked
} else if (pcmk__list_of_1(rsc->running_on)) {
pcmk_node_t *node = rsc->running_on->data;
if (node->details->shutdown) {
if (node->details->unclean) {
pcmk__rsc_debug(rsc,
"Not locking %s to unclean %s for shutdown",
rsc->id, pcmk__node_name(node));
} else {
rsc->lock_node = node;
rsc->lock_time = shutdown_time(node);
}
}
}
if (rsc->lock_node == NULL) {
// No lock needed
return;
}
if (rsc->cluster->shutdown_lock > 0) {
time_t lock_expiration = rsc->lock_time + rsc->cluster->shutdown_lock;
pcmk__rsc_info(rsc, "Locking %s to %s due to shutdown (expires @%lld)",
rsc->id, pcmk__node_name(rsc->lock_node),
(long long) lock_expiration);
pe__update_recheck_time(++lock_expiration, rsc->cluster,
"shutdown lock expiration");
} else {
pcmk__rsc_info(rsc, "Locking %s to %s due to shutdown",
rsc->id, pcmk__node_name(rsc->lock_node));
}
// If resource is locked to one node, ban it from all other nodes
g_list_foreach(rsc->cluster->nodes, ban_if_not_locked, rsc);
}
diff --git a/lib/pengine/utils.c b/lib/pengine/utils.c
index 6016f30f9f..8cf9e4629f 100644
--- a/lib/pengine/utils.c
+++ b/lib/pengine/utils.c
@@ -1,914 +1,915 @@
/*
* 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 <glib.h>
#include <stdbool.h>
#include <crm/crm.h>
#include <crm/msg_xml.h>
#include <crm/pengine/rules.h>
#include <crm/pengine/internal.h>
#include "pe_status_private.h"
extern bool pcmk__is_daemon;
gboolean ghash_free_str_str(gpointer key, gpointer value, gpointer user_data);
/*!
* \internal
* \brief Check whether we can fence a particular node
*
* \param[in] scheduler Scheduler data
* \param[in] node Name of node to check
*
* \return true if node can be fenced, false otherwise
*/
bool
pe_can_fence(const pcmk_scheduler_t *scheduler, const pcmk_node_t *node)
{
if (pe__is_guest_node(node)) {
/* Guest nodes are fenced by stopping their container resource. We can
* do that if the container's host is either online or fenceable.
*/
pcmk_resource_t *rsc = node->details->remote_rsc->container;
for (GList *n = rsc->running_on; n != NULL; n = n->next) {
pcmk_node_t *container_node = n->data;
if (!container_node->details->online
&& !pe_can_fence(scheduler, container_node)) {
return false;
}
}
return true;
} else if (!pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled)) {
return false; /* Turned off */
} else if (!pcmk_is_set(scheduler->flags, pcmk_sched_have_fencing)) {
return false; /* No devices */
} else if (pcmk_is_set(scheduler->flags, pcmk_sched_quorate)) {
return true;
} else if (scheduler->no_quorum_policy == pcmk_no_quorum_ignore) {
return true;
} else if(node == NULL) {
return false;
} else if(node->details->online) {
crm_notice("We can fence %s without quorum because they're in our membership",
pcmk__node_name(node));
return true;
}
crm_trace("Cannot fence %s", pcmk__node_name(node));
return false;
}
/*!
* \internal
* \brief Copy a node object
*
* \param[in] this_node Node object to copy
*
* \return Newly allocated shallow copy of this_node
* \note This function asserts on errors and is guaranteed to return non-NULL.
*/
pcmk_node_t *
pe__copy_node(const pcmk_node_t *this_node)
{
pcmk_node_t *new_node = NULL;
CRM_ASSERT(this_node != NULL);
new_node = calloc(1, sizeof(pcmk_node_t));
CRM_ASSERT(new_node != NULL);
new_node->rsc_discover_mode = this_node->rsc_discover_mode;
new_node->weight = this_node->weight;
new_node->fixed = this_node->fixed; // @COMPAT deprecated and unused
new_node->count = this_node->count;
new_node->details = this_node->details;
return new_node;
}
/*!
* \internal
* \brief Create a node hash table from a node list
*
* \param[in] list Node list
*
* \return Hash table equivalent of node list
*/
GHashTable *
pe__node_list2table(const GList *list)
{
GHashTable *result = NULL;
result = pcmk__strkey_table(NULL, free);
for (const GList *gIter = list; gIter != NULL; gIter = gIter->next) {
pcmk_node_t *new_node = NULL;
new_node = pe__copy_node((const pcmk_node_t *) gIter->data);
g_hash_table_insert(result, (gpointer) new_node->details->id, new_node);
}
return result;
}
/*!
* \internal
* \brief Compare two nodes by name, with numeric portions sorted numerically
*
* Sort two node names case-insensitively like strcasecmp(), but with any
* numeric portions of the name sorted numerically. For example, "node10" will
* sort higher than "node9" but lower than "remotenode9".
*
* \param[in] a First node to compare (can be \c NULL)
* \param[in] b Second node to compare (can be \c NULL)
*
* \retval -1 \c a comes before \c b (or \c a is \c NULL and \c b is not)
* \retval 0 \c a and \c b are equal (or both are \c NULL)
* \retval 1 \c a comes after \c b (or \c b is \c NULL and \c a is not)
*/
gint
pe__cmp_node_name(gconstpointer a, gconstpointer b)
{
const pcmk_node_t *node1 = (const pcmk_node_t *) a;
const pcmk_node_t *node2 = (const pcmk_node_t *) b;
if ((node1 == NULL) && (node2 == NULL)) {
return 0;
}
if (node1 == NULL) {
return -1;
}
if (node2 == NULL) {
return 1;
}
return pcmk__numeric_strcasecmp(node1->details->uname,
node2->details->uname);
}
/*!
* \internal
* \brief Output node weights to stdout
*
* \param[in] rsc Use allowed nodes for this resource
* \param[in] comment Text description to prefix lines with
* \param[in] nodes If rsc is not specified, use these nodes
* \param[in,out] scheduler Scheduler data
*/
static void
pe__output_node_weights(const pcmk_resource_t *rsc, const char *comment,
GHashTable *nodes, pcmk_scheduler_t *scheduler)
{
pcmk__output_t *out = scheduler->priv;
// Sort the nodes so the output is consistent for regression tests
GList *list = g_list_sort(g_hash_table_get_values(nodes),
pe__cmp_node_name);
for (const GList *gIter = list; gIter != NULL; gIter = gIter->next) {
const pcmk_node_t *node = (const pcmk_node_t *) gIter->data;
out->message(out, "node-weight", rsc, comment, node->details->uname,
pcmk_readable_score(node->weight));
}
g_list_free(list);
}
/*!
* \internal
* \brief Log node weights at trace level
*
* \param[in] file Caller's filename
* \param[in] function Caller's function name
* \param[in] line Caller's line number
* \param[in] rsc If not NULL, include this resource's ID in logs
* \param[in] comment Text description to prefix lines with
* \param[in] nodes Nodes whose scores should be logged
*/
static void
pe__log_node_weights(const char *file, const char *function, int line,
const pcmk_resource_t *rsc, const char *comment,
GHashTable *nodes)
{
GHashTableIter iter;
pcmk_node_t *node = NULL;
// Don't waste time if we're not tracing at this point
pcmk__if_tracing({}, return);
g_hash_table_iter_init(&iter, nodes);
while (g_hash_table_iter_next(&iter, NULL, (void **) &node)) {
if (rsc) {
qb_log_from_external_source(function, file,
"%s: %s allocation score on %s: %s",
LOG_TRACE, line, 0,
comment, rsc->id,
pcmk__node_name(node),
pcmk_readable_score(node->weight));
} else {
qb_log_from_external_source(function, file, "%s: %s = %s",
LOG_TRACE, line, 0,
comment, pcmk__node_name(node),
pcmk_readable_score(node->weight));
}
}
}
/*!
* \internal
* \brief Log or output node weights
*
* \param[in] file Caller's filename
* \param[in] function Caller's function name
* \param[in] line Caller's line number
* \param[in] to_log Log if true, otherwise output
* \param[in] rsc If not NULL, use this resource's ID in logs,
* and show scores recursively for any children
* \param[in] comment Text description to prefix lines with
* \param[in] nodes Nodes whose scores should be shown
* \param[in,out] scheduler Scheduler data
*/
void
pe__show_node_scores_as(const char *file, const char *function, int line,
bool to_log, const pcmk_resource_t *rsc,
const char *comment, GHashTable *nodes,
pcmk_scheduler_t *scheduler)
{
if ((rsc != NULL) && pcmk_is_set(rsc->flags, pcmk_rsc_removed)) {
// Don't show allocation scores for orphans
return;
}
if (nodes == NULL) {
// Nothing to show
return;
}
if (to_log) {
pe__log_node_weights(file, function, line, rsc, comment, nodes);
} else {
pe__output_node_weights(rsc, comment, nodes, scheduler);
}
// If this resource has children, repeat recursively for each
if (rsc && rsc->children) {
for (GList *gIter = rsc->children; gIter != NULL; gIter = gIter->next) {
pcmk_resource_t *child = (pcmk_resource_t *) gIter->data;
pe__show_node_scores_as(file, function, line, to_log, child,
comment, child->allowed_nodes, scheduler);
}
}
}
/*!
* \internal
* \brief Compare two resources by priority
*
* \param[in] a First resource to compare (can be \c NULL)
* \param[in] b Second resource to compare (can be \c NULL)
*
* \retval -1 \c a->priority > \c b->priority (or \c b is \c NULL and \c a is
* not)
* \retval 0 \c a->priority == \c b->priority (or both \c a and \c b are
* \c NULL)
* \retval 1 \c a->priority < \c b->priority (or \c a is \c NULL and \c b is
* not)
*/
gint
pe__cmp_rsc_priority(gconstpointer a, gconstpointer b)
{
const pcmk_resource_t *resource1 = (const pcmk_resource_t *)a;
const pcmk_resource_t *resource2 = (const pcmk_resource_t *)b;
if (a == NULL && b == NULL) {
return 0;
}
if (a == NULL) {
return 1;
}
if (b == NULL) {
return -1;
}
if (resource1->priority > resource2->priority) {
return -1;
}
if (resource1->priority < resource2->priority) {
return 1;
}
return 0;
}
static void
resource_node_score(pcmk_resource_t *rsc, const pcmk_node_t *node, int score,
const char *tag)
{
pcmk_node_t *match = NULL;
if ((rsc->exclusive_discover
|| (node->rsc_discover_mode == pcmk_probe_never))
&& pcmk__str_eq(tag, "symmetric_default", pcmk__str_casei)) {
/* This string comparision may be fragile, but exclusive resources and
* exclusive nodes should not have the symmetric_default constraint
* applied to them.
*/
return;
} else if (rsc->children) {
GList *gIter = rsc->children;
for (; gIter != NULL; gIter = gIter->next) {
pcmk_resource_t *child_rsc = (pcmk_resource_t *) gIter->data;
resource_node_score(child_rsc, node, score, tag);
}
}
match = g_hash_table_lookup(rsc->allowed_nodes, node->details->id);
if (match == NULL) {
match = pe__copy_node(node);
g_hash_table_insert(rsc->allowed_nodes, (gpointer) match->details->id, match);
}
match->weight = pcmk__add_scores(match->weight, score);
pcmk__rsc_trace(rsc,
"Enabling %s preference (%s) for %s on %s (now %s)",
tag, pcmk_readable_score(score), rsc->id,
pcmk__node_name(node), pcmk_readable_score(match->weight));
}
void
resource_location(pcmk_resource_t *rsc, const pcmk_node_t *node, int score,
const char *tag, pcmk_scheduler_t *scheduler)
{
if (node != NULL) {
resource_node_score(rsc, node, score, tag);
} else if (scheduler != NULL) {
GList *gIter = scheduler->nodes;
for (; gIter != NULL; gIter = gIter->next) {
pcmk_node_t *node_iter = (pcmk_node_t *) gIter->data;
resource_node_score(rsc, node_iter, score, tag);
}
} else {
GHashTableIter iter;
pcmk_node_t *node_iter = NULL;
g_hash_table_iter_init(&iter, rsc->allowed_nodes);
while (g_hash_table_iter_next(&iter, NULL, (void **)&node_iter)) {
resource_node_score(rsc, node_iter, score, tag);
}
}
if (node == NULL && score == -INFINITY) {
if (rsc->allocated_to) {
crm_info("Deallocating %s from %s",
rsc->id, pcmk__node_name(rsc->allocated_to));
free(rsc->allocated_to);
rsc->allocated_to = NULL;
}
}
}
time_t
get_effective_time(pcmk_scheduler_t *scheduler)
{
if(scheduler) {
if (scheduler->now == NULL) {
crm_trace("Recording a new 'now'");
scheduler->now = crm_time_new(NULL);
}
return crm_time_get_seconds_since_epoch(scheduler->now);
}
crm_trace("Defaulting to 'now'");
return time(NULL);
}
gboolean
get_target_role(const pcmk_resource_t *rsc, enum rsc_role_e *role)
{
enum rsc_role_e local_role = pcmk_role_unknown;
const char *value = g_hash_table_lookup(rsc->meta, PCMK_META_TARGET_ROLE);
CRM_CHECK(role != NULL, return FALSE);
if (pcmk__str_eq(value, "started", pcmk__str_null_matches | pcmk__str_casei)
|| pcmk__str_eq("default", value, pcmk__str_casei)) {
return FALSE;
}
local_role = pcmk_parse_role(value);
if (local_role == pcmk_role_unknown) {
pcmk__config_err("Ignoring '" PCMK_META_TARGET_ROLE "' for %s "
"because '%s' is not valid", rsc->id, value);
return FALSE;
} else if (local_role > pcmk_role_started) {
if (pcmk_is_set(pe__const_top_resource(rsc, false)->flags,
pcmk_rsc_promotable)) {
if (local_role > pcmk_role_unpromoted) {
/* This is what we'd do anyway, just leave the default to avoid messing up the placement algorithm */
return FALSE;
}
} else {
pcmk__config_err("Ignoring '" PCMK_META_TARGET_ROLE "' for %s "
"because '%s' only makes sense for promotable "
"clones", rsc->id, value);
return FALSE;
}
}
*role = local_role;
return TRUE;
}
gboolean
order_actions(pcmk_action_t *lh_action, pcmk_action_t *rh_action,
uint32_t flags)
{
GList *gIter = NULL;
pcmk__related_action_t *wrapper = NULL;
GList *list = NULL;
if (flags == pcmk__ar_none) {
return FALSE;
}
if (lh_action == NULL || rh_action == NULL) {
return FALSE;
}
crm_trace("Creating action wrappers for ordering: %s then %s",
lh_action->uuid, rh_action->uuid);
/* Ensure we never create a dependency on ourselves... it's happened */
CRM_ASSERT(lh_action != rh_action);
/* Filter dups, otherwise update_action_states() has too much work to do */
gIter = lh_action->actions_after;
for (; gIter != NULL; gIter = gIter->next) {
pcmk__related_action_t *after = gIter->data;
if (after->action == rh_action && (after->type & flags)) {
return FALSE;
}
}
wrapper = calloc(1, sizeof(pcmk__related_action_t));
wrapper->action = rh_action;
wrapper->type = flags;
list = lh_action->actions_after;
list = g_list_prepend(list, wrapper);
lh_action->actions_after = list;
wrapper = calloc(1, sizeof(pcmk__related_action_t));
wrapper->action = lh_action;
wrapper->type = flags;
list = rh_action->actions_before;
list = g_list_prepend(list, wrapper);
rh_action->actions_before = list;
return TRUE;
}
void
destroy_ticket(gpointer data)
{
pcmk_ticket_t *ticket = data;
if (ticket->state) {
g_hash_table_destroy(ticket->state);
}
free(ticket->id);
free(ticket);
}
pcmk_ticket_t *
ticket_new(const char *ticket_id, pcmk_scheduler_t *scheduler)
{
pcmk_ticket_t *ticket = NULL;
if (pcmk__str_empty(ticket_id)) {
return NULL;
}
if (scheduler->tickets == NULL) {
scheduler->tickets = pcmk__strkey_table(free, destroy_ticket);
}
ticket = g_hash_table_lookup(scheduler->tickets, ticket_id);
if (ticket == NULL) {
ticket = calloc(1, sizeof(pcmk_ticket_t));
if (ticket == NULL) {
pcmk__sched_err("Cannot allocate ticket '%s'", ticket_id);
return NULL;
}
crm_trace("Creaing ticket entry for %s", ticket_id);
ticket->id = strdup(ticket_id);
ticket->granted = FALSE;
ticket->last_granted = -1;
ticket->standby = FALSE;
ticket->state = pcmk__strkey_table(free, free);
g_hash_table_insert(scheduler->tickets, strdup(ticket->id), ticket);
}
return ticket;
}
const char *
rsc_printable_id(const pcmk_resource_t *rsc)
{
return pcmk_is_set(rsc->flags, pcmk_rsc_unique)? rsc->id : ID(rsc->xml);
}
void
pe__clear_resource_flags_recursive(pcmk_resource_t *rsc, uint64_t flags)
{
pcmk__clear_rsc_flags(rsc, flags);
for (GList *gIter = rsc->children; gIter != NULL; gIter = gIter->next) {
pe__clear_resource_flags_recursive((pcmk_resource_t *) gIter->data,
flags);
}
}
void
pe__clear_resource_flags_on_all(pcmk_scheduler_t *scheduler, uint64_t flag)
{
for (GList *lpc = scheduler->resources; lpc != NULL; lpc = lpc->next) {
pcmk_resource_t *r = (pcmk_resource_t *) lpc->data;
pe__clear_resource_flags_recursive(r, flag);
}
}
void
pe__set_resource_flags_recursive(pcmk_resource_t *rsc, uint64_t flags)
{
pcmk__set_rsc_flags(rsc, flags);
for (GList *gIter = rsc->children; gIter != NULL; gIter = gIter->next) {
pe__set_resource_flags_recursive((pcmk_resource_t *) gIter->data,
flags);
}
}
void
trigger_unfencing(pcmk_resource_t *rsc, pcmk_node_t *node, const char *reason,
pcmk_action_t *dependency, pcmk_scheduler_t *scheduler)
{
if (!pcmk_is_set(scheduler->flags, pcmk_sched_enable_unfencing)) {
/* No resources require it */
return;
} else if ((rsc != NULL)
&& !pcmk_is_set(rsc->flags, pcmk_rsc_fence_device)) {
/* Wasn't a stonith device */
return;
} else if(node
&& node->details->online
&& node->details->unclean == FALSE
&& node->details->shutdown == FALSE) {
pcmk_action_t *unfence = pe_fence_op(node, PCMK_ACTION_ON, FALSE,
reason, FALSE, scheduler);
if(dependency) {
order_actions(unfence, dependency, pcmk__ar_ordered);
}
} else if(rsc) {
GHashTableIter iter;
g_hash_table_iter_init(&iter, rsc->allowed_nodes);
while (g_hash_table_iter_next(&iter, NULL, (void **)&node)) {
if(node->details->online && node->details->unclean == FALSE && node->details->shutdown == FALSE) {
trigger_unfencing(rsc, node, reason, dependency, scheduler);
}
}
}
}
gboolean
add_tag_ref(GHashTable * tags, const char * tag_name, const char * obj_ref)
{
pcmk_tag_t *tag = NULL;
GList *gIter = NULL;
gboolean is_existing = FALSE;
CRM_CHECK(tags && tag_name && obj_ref, return FALSE);
tag = g_hash_table_lookup(tags, tag_name);
if (tag == NULL) {
tag = calloc(1, sizeof(pcmk_tag_t));
if (tag == NULL) {
pcmk__sched_err("Could not allocate memory for tag %s", tag_name);
return FALSE;
}
tag->id = strdup(tag_name);
tag->refs = NULL;
g_hash_table_insert(tags, strdup(tag_name), tag);
}
for (gIter = tag->refs; gIter != NULL; gIter = gIter->next) {
const char *existing_ref = (const char *) gIter->data;
if (pcmk__str_eq(existing_ref, obj_ref, pcmk__str_none)){
is_existing = TRUE;
break;
}
}
if (is_existing == FALSE) {
tag->refs = g_list_append(tag->refs, strdup(obj_ref));
crm_trace("Added: tag=%s ref=%s", tag->id, obj_ref);
}
return TRUE;
}
/*!
* \internal
* \brief Check whether shutdown has been requested for a node
*
* \param[in] node Node to check
*
* \return TRUE if node has shutdown attribute set and nonzero, FALSE otherwise
* \note This differs from simply using node->details->shutdown in that it can
* be used before that has been determined (and in fact to determine it),
* and it can also be used to distinguish requested shutdown from implicit
* shutdown of remote nodes by virtue of their connection stopping.
*/
bool
pe__shutdown_requested(const pcmk_node_t *node)
{
- const char *shutdown = pe_node_attribute_raw(node, XML_CIB_ATTR_SHUTDOWN);
+ const char *shutdown = pe_node_attribute_raw(node,
+ PCMK__NODE_ATTR_SHUTDOWN);
return !pcmk__str_eq(shutdown, "0", pcmk__str_null_matches);
}
/*!
* \internal
* \brief Update "recheck by" time in scheduler data
*
* \param[in] recheck Epoch time when recheck should happen
* \param[in,out] scheduler Scheduler data
* \param[in] reason What time is being updated for (for logs)
*/
void
pe__update_recheck_time(time_t recheck, pcmk_scheduler_t *scheduler,
const char *reason)
{
if ((recheck > get_effective_time(scheduler))
&& ((scheduler->recheck_by == 0)
|| (scheduler->recheck_by > recheck))) {
scheduler->recheck_by = recheck;
crm_debug("Updated next scheduler recheck to %s for %s",
pcmk__trim(ctime(&recheck)), reason);
}
}
/*!
* \internal
* \brief Extract nvpair blocks contained by a CIB XML element into a hash table
*
* \param[in] xml_obj XML element containing blocks of nvpair elements
* \param[in] set_name If not NULL, only use blocks of this element
* \param[in] rule_data Matching parameters to use when unpacking
* \param[out] hash Where to store extracted name/value pairs
* \param[in] always_first If not NULL, process block with this ID first
* \param[in] overwrite Whether to replace existing values with same name
* \param[in,out] scheduler Scheduler data containing \p xml_obj
*/
void
pe__unpack_dataset_nvpairs(const xmlNode *xml_obj, const char *set_name,
const pe_rule_eval_data_t *rule_data,
GHashTable *hash, const char *always_first,
gboolean overwrite, pcmk_scheduler_t *scheduler)
{
crm_time_t *next_change = crm_time_new_undefined();
pe_eval_nvpairs(scheduler->input, xml_obj, set_name, rule_data, hash,
always_first, overwrite, next_change);
if (crm_time_is_defined(next_change)) {
time_t recheck = (time_t) crm_time_get_seconds_since_epoch(next_change);
pe__update_recheck_time(recheck, scheduler, "rule evaluation");
}
crm_time_free(next_change);
}
bool
pe__resource_is_disabled(const pcmk_resource_t *rsc)
{
const char *target_role = NULL;
CRM_CHECK(rsc != NULL, return false);
target_role = g_hash_table_lookup(rsc->meta, PCMK_META_TARGET_ROLE);
if (target_role) {
// If invalid, we've already logged an error when unpacking
enum rsc_role_e target_role_e = pcmk_parse_role(target_role);
if ((target_role_e == pcmk_role_stopped)
|| ((target_role_e == pcmk_role_unpromoted)
&& pcmk_is_set(pe__const_top_resource(rsc, false)->flags,
pcmk_rsc_promotable))) {
return true;
}
}
return false;
}
/*!
* \internal
* \brief Check whether a resource is running only on given node
*
* \param[in] rsc Resource to check
* \param[in] node Node to check
*
* \return true if \p rsc is running only on \p node, otherwise false
*/
bool
pe__rsc_running_on_only(const pcmk_resource_t *rsc, const pcmk_node_t *node)
{
return (rsc != NULL) && pcmk__list_of_1(rsc->running_on)
&& pcmk__same_node((const pcmk_node_t *) rsc->running_on->data,
node);
}
bool
pe__rsc_running_on_any(pcmk_resource_t *rsc, GList *node_list)
{
for (GList *ele = rsc->running_on; ele; ele = ele->next) {
pcmk_node_t *node = (pcmk_node_t *) ele->data;
if (pcmk__str_in_list(node->details->uname, node_list,
pcmk__str_star_matches|pcmk__str_casei)) {
return true;
}
}
return false;
}
bool
pcmk__rsc_filtered_by_node(pcmk_resource_t *rsc, GList *only_node)
{
return (rsc->fns->active(rsc, FALSE) && !pe__rsc_running_on_any(rsc, only_node));
}
GList *
pe__filter_rsc_list(GList *rscs, GList *filter)
{
GList *retval = NULL;
for (GList *gIter = rscs; gIter; gIter = gIter->next) {
pcmk_resource_t *rsc = (pcmk_resource_t *) gIter->data;
/* I think the second condition is safe here for all callers of this
* function. If not, it needs to move into pe__node_text.
*/
if (pcmk__str_in_list(rsc_printable_id(rsc), filter, pcmk__str_star_matches) ||
(rsc->parent && pcmk__str_in_list(rsc_printable_id(rsc->parent), filter, pcmk__str_star_matches))) {
retval = g_list_prepend(retval, rsc);
}
}
return retval;
}
GList *
pe__build_node_name_list(pcmk_scheduler_t *scheduler, const char *s)
{
GList *nodes = NULL;
if (pcmk__str_eq(s, "*", pcmk__str_null_matches)) {
/* Nothing was given so return a list of all node names. Or, '*' was
* given. This would normally fall into the pe__unames_with_tag branch
* where it will return an empty list. Catch it here instead.
*/
nodes = g_list_prepend(nodes, strdup("*"));
} else {
pcmk_node_t *node = pe_find_node(scheduler->nodes, s);
if (node) {
/* The given string was a valid uname for a node. Return a
* singleton list containing just that uname.
*/
nodes = g_list_prepend(nodes, strdup(s));
} else {
/* The given string was not a valid uname. It's either a tag or
* it's a typo or something. In the first case, we'll return a
* list of all the unames of the nodes with the given tag. In the
* second case, we'll return a NULL pointer and nothing will
* get displayed.
*/
nodes = pe__unames_with_tag(scheduler, s);
}
}
return nodes;
}
GList *
pe__build_rsc_list(pcmk_scheduler_t *scheduler, const char *s)
{
GList *resources = NULL;
if (pcmk__str_eq(s, "*", pcmk__str_null_matches)) {
resources = g_list_prepend(resources, strdup("*"));
} else {
const uint32_t flags = pcmk_rsc_match_history|pcmk_rsc_match_basename;
pcmk_resource_t *rsc = pe_find_resource_with_flags(scheduler->resources,
s, flags);
if (rsc) {
/* A colon in the name we were given means we're being asked to filter
* on a specific instance of a cloned resource. Put that exact string
* into the filter list. Otherwise, use the printable ID of whatever
* resource was found that matches what was asked for.
*/
if (strstr(s, ":") != NULL) {
resources = g_list_prepend(resources, strdup(rsc->id));
} else {
resources = g_list_prepend(resources, strdup(rsc_printable_id(rsc)));
}
} else {
/* The given string was not a valid resource name. It's a tag or a
* typo or something. See pe__build_node_name_list() for more
* detail.
*/
resources = pe__rscs_with_tag(scheduler, s);
}
}
return resources;
}
xmlNode *
pe__failed_probe_for_rsc(const pcmk_resource_t *rsc, const char *name)
{
const pcmk_resource_t *parent = pe__const_top_resource(rsc, false);
const char *rsc_id = rsc->id;
if (parent->variant == pcmk_rsc_variant_clone) {
rsc_id = pe__clone_child_id(parent);
}
for (xmlNode *xml_op = pcmk__xml_first_child(rsc->cluster->failed); xml_op != NULL;
xml_op = pcmk__xml_next(xml_op)) {
const char *value = NULL;
char *op_id = NULL;
/* This resource operation is not a failed probe. */
if (!pcmk_xe_mask_probe_failure(xml_op)) {
continue;
}
/* This resource operation was not run on the given node. Note that if name is
* NULL, this will always succeed.
*/
value = crm_element_value(xml_op, PCMK__META_ON_NODE);
if (value == NULL || !pcmk__str_eq(value, name, pcmk__str_casei|pcmk__str_null_matches)) {
continue;
}
if (!parse_op_key(pcmk__xe_history_key(xml_op), &op_id, NULL, NULL)) {
continue; // This history entry is missing an operation key
}
/* This resource operation's ID does not match the rsc_id we are looking for. */
if (!pcmk__str_eq(op_id, rsc_id, pcmk__str_none)) {
free(op_id);
continue;
}
free(op_id);
return xml_op;
}
return NULL;
}