void
cib_fencing_updated(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data)
{
if (rc < pcmk_ok) {
crm_err("Fencing update %d for %s: failed - %s (%d)",
call_id, (char *)user_data, pcmk_strerror(rc), rc);
crm_log_xml_warn(msg, "Failed update");
abort_transition(INFINITY, tg_shutdown, "CIB update failed", NULL);
} else {
crm_info("Fencing update %d for %s: complete", call_id, (char *)user_data);
}
}
/*!
* \brief Handle a timeout in node-to-node communication
*
* \param[in] data Pointer to action timer
*
* \return FALSE (indicating that source should be not be re-added)
*/
gboolean
action_timer_callback(gpointer data)
{
crm_action_timer_t *timer = NULL;
const char *task = NULL;
const char *on_node = NULL;
const char *via_node = NULL;
CRM_CHECK(data != NULL, return FALSE);
timer = (crm_action_timer_t *) data;
stop_te_timer(timer);
CRM_CHECK(timer->action != NULL, return FALSE);
task = crm_element_value(timer->action->xml, XML_LRM_ATTR_TASK);
on_node = crm_element_value(timer->action->xml, XML_LRM_ATTR_TARGET);
via_node = crm_element_value(timer->action->xml, XML_LRM_ATTR_ROUTER_NODE);
if (transition_graph->complete) {
crm_notice("Node %s did not send %s result (via %s) within %dms "
"(ignoring because transition not in progress)",
(on_node? on_node : ""), (task? task : "unknown action"),
(via_node? via_node : "controller"), timer->timeout);
} else {
/* fail the action */
crm_err("Node %s did not send %s result (via %s) within %dms "
"(action timeout plus cluster-delay)",
(on_node? on_node : ""), (task? task : "unknown action"),
(via_node? via_node : "controller"), timer->timeout);
print_action(LOG_ERR, "Aborting transition, action lost: ", timer->action);
timer->action->failed = TRUE;
te_action_confirmed(timer->action);
abort_transition(INFINITY, tg_restart, "Action lost", NULL);
update_graph(transition_graph, timer->action);
trigger_graph();
// Record timeout in the CIB if appropriate
if ((timer->action->type == action_type_rsc)
&& controld_action_is_recordable(task)) {
controld_record_action_timeout(timer->action);
}
}
return FALSE;
}
static void
process_cib_diff(xmlNode *cib, xmlNode *change, const char *op,
const char *xpath)
{
xmlNode *status = first_named_child(cib, XML_CIB_TAG_STATUS);
xmlNode *config = first_named_child(cib, XML_CIB_TAG_CONFIGURATION);
if (status) {
process_status_diff(status, change, op, xpath);
}
if (config) {
abort_transition(INFINITY, tg_restart,
"Non-status-only change", change);
}
}
void
send_stonith_update(stonith_ops_t * op)
{
enum cib_errors rc = cib_ok;
const char *target = op->node_name;
const char *uuid = op->node_uuid;
/* zero out the node-status & remove all LRM status info */
xmlNode *node_state = create_xml_node(NULL, XML_CIB_TAG_STATE);
CRM_CHECK(op->node_name != NULL, return);
CRM_CHECK(op->node_uuid != NULL, return);
crm_xml_add(node_state, XML_ATTR_UUID, uuid);
crm_xml_add(node_state, XML_ATTR_UNAME, target);
crm_xml_add(node_state, XML_CIB_ATTR_HASTATE, DEADSTATUS);
crm_xml_add(node_state, XML_CIB_ATTR_INCCM, XML_BOOLEAN_NO);
crm_xml_add(node_state, XML_CIB_ATTR_CRMDSTATE, OFFLINESTATUS);
crm_xml_add(node_state, XML_CIB_ATTR_JOINSTATE, CRMD_JOINSTATE_DOWN);
crm_xml_add(node_state, XML_CIB_ATTR_EXPSTATE, CRMD_JOINSTATE_DOWN);
crm_xml_add(node_state, XML_ATTR_ORIGIN, __FUNCTION__);
rc = fsa_cib_conn->cmds->update(
fsa_cib_conn, XML_CIB_TAG_STATUS, node_state,
cib_quorum_override|cib_scope_local|cib_can_create);
if(rc < cib_ok) {
crm_err("CIB update failed: %s", cib_error2string(rc));
abort_transition(
INFINITY, tg_shutdown, "CIB update failed", node_state);
} else {
/* delay processing the trigger until the update completes */
add_cib_op_callback(fsa_cib_conn, rc, FALSE, crm_strdup(target), cib_fencing_updated);
}
erase_status_tag(op->node_name, XML_CIB_TAG_LRM, cib_scope_local);
erase_status_tag(op->node_name, XML_TAG_TRANSIENT_NODEATTRS, cib_scope_local);
free_xml(node_state);
#if 0
/* Make sure the membership cache is accurate */
crm_update_peer(0, 0, 0, -1, 0, uuid, target, NULL, CRM_NODE_LOST);
#endif
return;
}
static void process_resource_updates(
const char *node, xmlNode *xml, xmlNode *change, const char *op, const char *xpath)
{
xmlNode *cIter = NULL;
xmlNode *rsc = NULL;
xmlNode *rsc_op = NULL;
int num_resources = 0;
if(xml == NULL) {
return;
} else if(strcmp((const char*)xml->name, XML_CIB_TAG_LRM) == 0) {
xml = first_named_child(xml, XML_LRM_TAG_RESOURCES);
crm_trace("Got %p in %s", xml, XML_CIB_TAG_LRM);
}
CRM_ASSERT(strcmp((const char*)xml->name, XML_LRM_TAG_RESOURCES) == 0);
for(cIter = xml->children; cIter; cIter = cIter->next) {
num_resources++;
}
if(num_resources > 1) {
/*
* Check for and fast-track the processing of LRM refreshes
* In large clusters this can result in _huge_ speedups
*
* Unfortunately we can only do so when there are no pending actions
* Otherwise we could miss updates we're waiting for and stall
*
*/
crm_debug("Detected LRM refresh - %d resources updated", num_resources);
crm_log_xml_trace(change, "lrm-refresh");
abort_transition(INFINITY, tg_restart, "LRM Refresh", NULL);
return;
}
for (rsc = __xml_first_child(xml); rsc != NULL; rsc = __xml_next(rsc)) {
crm_trace("Processing %s", ID(rsc));
for (rsc_op = __xml_first_child(rsc); rsc_op != NULL; rsc_op = __xml_next(rsc_op)) {
crm_trace("Processing %s", ID(rsc_op));
process_graph_event(rsc_op, node);
}
}
}
static gboolean
fail_incompletable_stonith(crm_graph_t * graph)
{
GListPtr lpc = NULL;
const char *task = NULL;
xmlNode *last_action = NULL;
if (graph == NULL) {
return FALSE;
}
for (lpc = graph->synapses; lpc != NULL; lpc = lpc->next) {
GListPtr lpc2 = NULL;
synapse_t *synapse = (synapse_t *) lpc->data;
if (synapse->confirmed) {
continue;
}
for (lpc2 = synapse->actions; lpc2 != NULL; lpc2 = lpc2->next) {
crm_action_t *action = (crm_action_t *) lpc2->data;
if (action->type != action_type_crm || action->confirmed) {
continue;
}
task = crm_element_value(action->xml, XML_LRM_ATTR_TASK);
if (task && safe_str_eq(task, CRM_OP_FENCE)) {
action->failed = TRUE;
last_action = action->xml;
update_graph(graph, action);
crm_notice("Failing action %d (%s): STONITHd terminated",
action->id, ID(action->xml));
}
}
}
if (last_action != NULL) {
crm_warn("STONITHd failure resulted in un-runnable actions");
abort_transition(INFINITY, tg_restart, "Stonith failure", last_action);
return TRUE;
}
return FALSE;
}
static void
process_resource_updates(const char *node, xmlNode *xml, xmlNode *change,
const char *op, const char *xpath)
{
xmlNode *rsc = NULL;
if (xml == NULL) {
return;
} else if (strcmp((const char*)xml->name, XML_CIB_TAG_LRM) == 0) {
xml = first_named_child(xml, XML_LRM_TAG_RESOURCES);
crm_trace("Got %p in %s", xml, XML_CIB_TAG_LRM);
}
CRM_ASSERT(strcmp((const char*)xml->name, XML_LRM_TAG_RESOURCES) == 0);
/*
* Updates by, or in response to, TE actions will never contain updates
* for more than one resource at a time, so such updates indicate an
* LRM refresh.
*
* In that case, start a new transition rather than check each result
* individually, which can result in _huge_ speedups in large clusters.
*
* Unfortunately, we can only do so when there are no pending actions.
* Otherwise, we could mistakenly throw away those results here, and
* the cluster will stall waiting for them and time out the operation.
*/
if ((transition_graph->pending == 0)
&& xml->children && xml->children->next) {
crm_log_xml_trace(change, "lrm-refresh");
abort_transition(INFINITY, tg_restart, "History refresh", NULL);
return;
}
for (rsc = __xml_first_child(xml); rsc != NULL; rsc = __xml_next(rsc)) {
crm_trace("Processing %s", ID(rsc));
process_lrm_resource_diff(rsc, node);
}
}
void
send_stonith_update(stonith_ops_t * op)
{
enum cib_errors rc = cib_ok;
const char *target = op->node_name;
const char *uuid = op->node_uuid;
/* zero out the node-status & remove all LRM status info */
crm_data_t *node_state = create_xml_node(NULL, XML_CIB_TAG_STATE);
CRM_CHECK(op->node_name != NULL, return);
CRM_CHECK(op->node_uuid != NULL, return);
crm_xml_add(node_state, XML_ATTR_UUID, uuid);
crm_xml_add(node_state, XML_ATTR_UNAME, target);
crm_xml_add(node_state, XML_CIB_ATTR_HASTATE, DEADSTATUS);
crm_xml_add(node_state, XML_CIB_ATTR_INCCM, XML_BOOLEAN_NO);
crm_xml_add(node_state, XML_CIB_ATTR_CRMDSTATE, OFFLINESTATUS);
crm_xml_add(node_state, XML_CIB_ATTR_JOINSTATE, CRMD_JOINSTATE_DOWN);
crm_xml_add(node_state, XML_CIB_ATTR_EXPSTATE, CRMD_JOINSTATE_DOWN);
crm_xml_add(node_state, XML_CIB_ATTR_REPLACE, XML_CIB_TAG_LRM);
crm_xml_add(node_state, XML_ATTR_ORIGIN, __FUNCTION__);
rc = te_cib_conn->cmds->update(
te_cib_conn, XML_CIB_TAG_STATUS, node_state, NULL,
cib_quorum_override|cib_scope_local);
if(rc < cib_ok) {
crm_err("CIB update failed: %s", cib_error2string(rc));
abort_transition(
INFINITY, tg_shutdown, "CIB update failed", node_state);
} else {
/* delay processing the trigger until the update completes */
add_cib_op_callback(rc, FALSE, NULL, cib_fencing_updated);
}
free_xml(node_state);
return;
}
static void
process_op_deletion(const char *xpath, xmlNode *change)
{
char *mutable_key = strdup(xpath);
char *key;
char *node_uuid;
crm_action_t *cancel = NULL;
// Extract the part of xpath between last pair of single quotes
key = strrchr(mutable_key, '\'');
if (key != NULL) {
*key = '\0';
key = strrchr(mutable_key, '\'');
}
if (key == NULL) {
crm_warn("Ignoring malformed CIB update (resource deletion of %s)",
xpath);
free(mutable_key);
return;
}
++key;
node_uuid = extract_node_uuid(xpath);
cancel = get_cancel_action(key, node_uuid);
if (cancel) {
crm_info("Cancellation of %s on %s confirmed (%d)",
key, node_uuid, cancel->id);
stop_te_timer(cancel->timer);
te_action_confirmed(cancel);
update_graph(transition_graph, cancel);
trigger_graph();
} else {
abort_transition(INFINITY, tg_restart, "Resource operation removal",
change);
}
free(mutable_key);
free(node_uuid);
}
/* A_TE_INVOKE, A_TE_CANCEL */
void
do_te_invoke(long long action,
enum crmd_fsa_cause cause,
enum crmd_fsa_state cur_state,
enum crmd_fsa_input current_input,
fsa_data_t *msg_data)
{
if(AM_I_DC == FALSE) {
crm_err("Not DC: No need to invoke the TE (anymore): %s",
fsa_action2string(action));
return;
} else if(fsa_state != S_TRANSITION_ENGINE && (action & A_TE_INVOKE)) {
crm_err("No need to invoke the TE (%s) in state %s",
fsa_action2string(action),
fsa_state2string(fsa_state));
return;
}
if(action & A_TE_CANCEL) {
crm_debug("Cancelling the transition: %s",
transition_graph->complete?"inactive":"active");
abort_transition(INFINITY, tg_restart, "Peer Cancelled", NULL);
if(transition_graph && transition_graph->complete == FALSE) {
crmd_fsa_stall(NULL);
}
} else if(action & A_TE_HALT) {
crm_debug("Halting the transition: %s",
transition_graph->complete?"inactive":"active");
abort_transition(INFINITY, tg_stop, "Peer Halt", NULL);
if(transition_graph && transition_graph->complete == FALSE) {
crmd_fsa_stall(NULL);
}
} else if(action & A_TE_INVOKE) {
const char *value = NULL;
xmlNode *graph_data = NULL;
ha_msg_input_t *input = fsa_typed_data(fsa_dt_ha_msg);
const char *ref = crm_element_value(input->msg, XML_ATTR_REFERENCE);
const char *graph_file = crm_element_value(input->msg, F_CRM_TGRAPH);
const char *graph_input = crm_element_value(input->msg, F_CRM_TGRAPH_INPUT);
if(graph_file == NULL && input->xml == NULL) {
crm_log_xml_err(input->msg, "Bad command");
register_fsa_error(C_FSA_INTERNAL, I_FAIL, NULL);
return;
}
if(transition_graph && transition_graph->complete == FALSE) {
crm_info("Another transition is already active");
abort_transition(INFINITY, tg_restart, "Transition Active", NULL);
return;
}
if(fsa_pe_ref == NULL || safe_str_neq(fsa_pe_ref, ref)) {
crm_info("Transition is redundant: %s vs. %s", crm_str(fsa_pe_ref), crm_str(ref));
abort_transition(INFINITY, tg_restart, "Transition Redundant", NULL);
}
graph_data = input->xml;
if(graph_data == NULL && graph_file != NULL) {
graph_data = filename2xml(graph_file);
}
if (is_timer_started(transition_timer)) {
crm_debug("The transitioner wait for a transition timer");
return;
}
CRM_CHECK(graph_data != NULL,
crm_err("Input raised by %s is invalid", msg_data->origin);
crm_log_xml_err(input->msg, "Bad command");
return);
destroy_graph(transition_graph);
transition_graph = unpack_graph(graph_data, graph_input);
CRM_CHECK(transition_graph != NULL, transition_graph = create_blank_graph(); return);
crm_info("Processing graph %d (ref=%s) derived from %s", transition_graph->id, ref, graph_input);
value = crm_element_value(graph_data, "failed-stop-offset");
if(value) {
crm_free(failed_stop_offset);
failed_stop_offset = crm_strdup(value);
}
value = crm_element_value(graph_data, "failed-start-offset");
if(value) {
crm_free(failed_start_offset);
failed_start_offset = crm_strdup(value);
}
trigger_graph();
print_graph(LOG_DEBUG_2, transition_graph);
if(graph_data != input->xml) {
free_xml(graph_data);
}
}
}
void
te_update_diff(const char *event, HA_Message *msg)
{
int rc = -1;
const char *op = NULL;
crm_data_t *diff = NULL;
crm_data_t *aborted = NULL;
const char *set_name = NULL;
int diff_add_updates = 0;
int diff_add_epoch = 0;
int diff_add_admin_epoch = 0;
int diff_del_updates = 0;
int diff_del_epoch = 0;
int diff_del_admin_epoch = 0;
if(msg == NULL) {
crm_err("NULL update");
return;
}
ha_msg_value_int(msg, F_CIB_RC, &rc);
op = cl_get_string(msg, F_CIB_OPERATION);
if(rc < cib_ok) {
crm_debug_2("Ignoring failed %s operation: %s",
op, cib_error2string(rc));
return;
}
diff = get_message_xml(msg, F_CIB_UPDATE_RESULT);
cib_diff_version_details(
diff,
&diff_add_admin_epoch, &diff_add_epoch, &diff_add_updates,
&diff_del_admin_epoch, &diff_del_epoch, &diff_del_updates);
crm_debug("Processing diff (%s): %d.%d.%d -> %d.%d.%d", op,
diff_del_admin_epoch,diff_del_epoch,diff_del_updates,
diff_add_admin_epoch,diff_add_epoch,diff_add_updates);
log_cib_diff(LOG_DEBUG_2, diff, op);
set_name = "diff-added";
if(diff != NULL) {
crm_data_t *section = NULL;
crm_data_t *change_set = find_xml_node(diff, set_name, FALSE);
change_set = find_xml_node(change_set, XML_TAG_CIB, FALSE);
if(change_set != NULL) {
crm_debug_2("Checking status changes");
section=get_object_root(XML_CIB_TAG_STATUS,change_set);
}
if(section != NULL) {
extract_event(section);
}
crm_debug_2("Checking change set: %s", set_name);
aborted = need_abort(change_set);
}
set_name = "diff-removed";
if(diff != NULL && aborted == NULL) {
crm_data_t *attrs = NULL;
crm_data_t *status = NULL;
crm_data_t *change_set = find_xml_node(diff, set_name, FALSE);
change_set = find_xml_node(change_set, XML_TAG_CIB, FALSE);
crm_debug_2("Checking change set: %s", set_name);
aborted = need_abort(change_set);
if(aborted == NULL && change_set != NULL) {
status = get_object_root(XML_CIB_TAG_STATUS, change_set);
xml_child_iter_filter(
status, node_state, XML_CIB_TAG_STATE,
attrs = find_xml_node(
node_state, XML_TAG_TRANSIENT_NODEATTRS, FALSE);
if(attrs != NULL) {
crm_info("Aborting on "XML_TAG_TRANSIENT_NODEATTRS" deletions");
abort_transition(INFINITY, tg_restart,
XML_TAG_TRANSIENT_NODEATTRS, attrs);
}
);
void
peer_update_callback(enum crm_status_type type, crm_node_t * node, const void *data)
{
uint32_t old = 0;
uint32_t changed = 0;
bool appeared = FALSE;
bool is_remote = is_set(node->flags, crm_remote_node);
const char *status = NULL;
/* Crmd waits to receive some information from the membership layer before
* declaring itself operational. If this is being called for a cluster node,
* indicate that we have it.
*/
if (!is_remote) {
set_bit(fsa_input_register, R_PEER_DATA);
}
if (node->uname == NULL) {
return;
}
switch (type) {
case crm_status_uname:
/* If we've never seen the node, then it also won't be in the status section */
crm_info("%s node %s is now %s",
(is_remote? "Remote" : "Cluster"),
node->uname, state_text(node->state));
return;
case crm_status_rstate:
case crm_status_nstate:
/* This callback should not be called unless the state actually
* changed, but here's a failsafe just in case.
*/
CRM_CHECK(safe_str_neq(data, node->state), return);
crm_info("%s node %s is now %s (was %s)",
(is_remote? "Remote" : "Cluster"),
node->uname, state_text(node->state), state_text(data));
if (safe_str_eq(CRM_NODE_MEMBER, node->state)) {
appeared = TRUE;
if (!is_remote) {
remove_stonith_cleanup(node->uname);
}
}
crmd_alert_node_event(node);
break;
case crm_status_processes:
if (data) {
old = *(const uint32_t *)data;
changed = node->processes ^ old;
}
status = (node->processes & proc_flags) ? ONLINESTATUS : OFFLINESTATUS;
crm_info("Client %s/%s now has status [%s] (DC=%s, changed=%6x)",
node->uname, peer2text(proc_flags), status,
AM_I_DC ? "true" : crm_str(fsa_our_dc), changed);
if ((changed & proc_flags) == 0) {
/* Peer process did not change */
crm_trace("No change %6x %6x %6x", old, node->processes, proc_flags);
return;
} else if (is_not_set(fsa_input_register, R_CIB_CONNECTED)) {
crm_trace("Not connected");
return;
} else if (fsa_state == S_STOPPING) {
crm_trace("Stopping");
return;
}
appeared = (node->processes & proc_flags) != 0;
if (safe_str_eq(node->uname, fsa_our_uname) && (node->processes & proc_flags) == 0) {
/* Did we get evicted? */
crm_notice("Our peer connection failed");
register_fsa_input(C_CRMD_STATUS_CALLBACK, I_ERROR, NULL);
} else if (safe_str_eq(node->uname, fsa_our_dc) && crm_is_peer_active(node) == FALSE) {
/* Did the DC leave us? */
crm_notice("Our peer on the DC (%s) is dead", fsa_our_dc);
register_fsa_input(C_CRMD_STATUS_CALLBACK, I_ELECTION, NULL);
/* @COMPAT DC < 1.1.13: If a DC shuts down normally, we don't
* want to fence it. Newer DCs will send their shutdown request
* to all peers, who will update the DC's expected state to
* down, thus avoiding fencing. We can safely erase the DC's
* transient attributes when it leaves in that case. However,
* the only way to avoid fencing older DCs is to leave the
* transient attributes intact until it rejoins.
*/
if (compare_version(fsa_our_dc_version, "3.0.9") > 0) {
erase_status_tag(node->uname, XML_TAG_TRANSIENT_NODEATTRS, cib_scope_local);
}
} else if(AM_I_DC && appeared == FALSE) {
crm_info("Peer %s left us", node->uname);
erase_status_tag(node->uname, XML_TAG_TRANSIENT_NODEATTRS, cib_scope_local);
}
break;
}
if (AM_I_DC) {
xmlNode *update = NULL;
int flags = node_update_peer;
gboolean alive = is_remote? appeared : crm_is_peer_active(node);
crm_action_t *down = match_down_event(node->uuid, appeared);
crm_trace("Alive=%d, appeared=%d, down=%d",
alive, appeared, (down? down->id : -1));
if (alive && type == crm_status_processes) {
register_fsa_input_before(C_FSA_INTERNAL, I_NODE_JOIN, NULL);
}
if (down) {
const char *task = crm_element_value(down->xml, XML_LRM_ATTR_TASK);
if (safe_str_eq(task, CRM_OP_FENCE)) {
/* tengine_stonith_callback() confirms fence actions */
crm_trace("Updating CIB %s stonithd reported fencing of %s complete",
(down->confirmed? "after" : "before"), node->uname);
} else if ((alive == FALSE) && safe_str_eq(task, CRM_OP_SHUTDOWN)) {
crm_notice("%s of peer %s is complete "CRM_XS" op=%d",
task, node->uname, down->id);
/* down->confirmed = TRUE; */
stop_te_timer(down->timer);
if (!is_remote) {
flags |= node_update_join | node_update_expected;
crmd_peer_down(node, FALSE);
check_join_state(fsa_state, __FUNCTION__);
}
update_graph(transition_graph, down);
trigger_graph();
} else {
crm_trace("Node %s is %salive, was expected to %s (op %d)",
node->uname, (alive? "" : "not "), task, down->id);
}
} else if (appeared == FALSE) {
crm_notice("Stonith/shutdown of %s not matched", node->uname);
if (!is_remote) {
crm_update_peer_join(__FUNCTION__, node, crm_join_none);
check_join_state(fsa_state, __FUNCTION__);
}
abort_transition(INFINITY, tg_restart, "Node failure", NULL);
fail_incompletable_actions(transition_graph, node->uuid);
} else {
crm_trace("Node %s came up, was not expected to be down",
node->uname);
}
if (is_remote) {
/* A pacemaker_remote node won't have its cluster status updated
* in the CIB by membership-layer callbacks, so do it here.
*/
flags |= node_update_cluster;
/* Trigger resource placement on newly integrated nodes */
if (appeared) {
abort_transition(INFINITY, tg_restart,
"pacemaker_remote node integrated", NULL);
}
}
/* Update the CIB node state */
update = create_node_state_update(node, flags, NULL, __FUNCTION__);
fsa_cib_anon_update(XML_CIB_TAG_STATUS, update,
cib_scope_local | cib_quorum_override | cib_can_create);
free_xml(update);
}
trigger_fsa(fsa_source);
}
static void
tengine_stonith_notify(stonith_t * st, stonith_event_t * st_event)
{
static char *client_id = NULL;
if(client_id == NULL) {
client_id = g_strdup_printf("%s.%d", crm_system_name, getpid());
}
if (st_event == NULL) {
crm_err("Notify data not found");
return;
}
if (st_event->result == pcmk_ok && crm_str_eq(st_event->target, fsa_our_uname, TRUE)) {
crm_err("We were alegedly just fenced by %s for %s!", st_event->executioner,
st_event->origin);
register_fsa_error_adv(C_FSA_INTERNAL, I_ERROR, NULL, NULL, __FUNCTION__);
return;
}
if (st_event->result == pcmk_ok &&
safe_str_eq(st_event->operation, T_STONITH_NOTIFY_FENCE)) {
reset_st_fail_count(st_event->target);
}
crm_notice("Peer %s was%s terminated (%s) by %s for %s: %s (ref=%s) by client %s",
st_event->target, st_event->result == pcmk_ok ? "" : " not",
st_event->action,
st_event->executioner ? st_event->executioner : "<anyone>",
st_event->origin, pcmk_strerror(st_event->result), st_event->id,
st_event->client_origin ? st_event->client_origin : "<unknown>");
#if SUPPORT_CMAN
if (st_event->result == pcmk_ok && is_cman_cluster()) {
int local_rc = 0;
char *target_copy = strdup(st_event->target);
/* In case fenced hasn't noticed yet
*
* Any fencing that has been inititated will be completed by way of the fence_pcmk redirect
*/
local_rc = fenced_external(target_copy);
if (local_rc != 0) {
crm_err("Could not notify CMAN that '%s' is now fenced: %d", st_event->target,
local_rc);
} else {
crm_notice("Notified CMAN that '%s' is now fenced", st_event->target);
}
free(target_copy);
}
#endif
if (st_event->result == pcmk_ok) {
const char *uuid = get_uuid(st_event->target);
gboolean we_are_executioner = safe_str_eq(st_event->executioner, fsa_our_uname);
crm_trace("target=%s dc=%s", st_event->target, fsa_our_dc);
if(AM_I_DC) {
/* The DC always sends updates */
send_stonith_update(NULL, st_event->target, uuid);
if (st_event->client_origin && safe_str_neq(st_event->client_origin, client_id)) {
/* Abort the current transition graph if it wasn't us
* that invoked stonith to fence someone
*/
crm_info("External fencing operation from %s fenced %s", st_event->client_origin, st_event->target);
abort_transition(INFINITY, tg_restart, "External Fencing Operation", NULL);
}
/* Assume it was our leader if we dont currently have one */
} else if (fsa_our_dc == NULL || safe_str_eq(fsa_our_dc, st_event->target)) {
crm_notice("Target %s our leader %s (recorded: %s)",
fsa_our_dc ? "was" : "may have been", st_event->target,
fsa_our_dc ? fsa_our_dc : "<unset>");
/* Given the CIB resyncing that occurs around elections,
* have one node update the CIB now and, if the new DC is different,
* have them do so too after the election
*/
if (we_are_executioner) {
send_stonith_update(NULL, st_event->target, uuid);
}
stonith_cleanup_list = g_list_append(stonith_cleanup_list, strdup(st_event->target));
}
}
}
static void
te_legacy_update_diff(const char *event, xmlNode * diff)
{
int lpc, max;
xmlXPathObject *xpathObj = NULL;
CRM_CHECK(diff != NULL, return);
xml_log_patchset(LOG_TRACE, __FUNCTION__, diff);
if (cib_config_changed(NULL, NULL, &diff)) {
abort_transition(INFINITY, tg_restart, "Non-status change", diff);
goto bail; /* configuration changed */
}
/* Tickets Attributes - Added/Updated */
xpathObj =
xpath_search(diff,
"//" F_CIB_UPDATE_RESULT "//" XML_TAG_DIFF_ADDED "//" XML_CIB_TAG_TICKETS);
if (numXpathResults(xpathObj) > 0) {
xmlNode *aborted = getXpathResult(xpathObj, 0);
abort_transition(INFINITY, tg_restart, "Ticket attribute: update", aborted);
goto bail;
}
freeXpathObject(xpathObj);
/* Tickets Attributes - Removed */
xpathObj =
xpath_search(diff,
"//" F_CIB_UPDATE_RESULT "//" XML_TAG_DIFF_REMOVED "//" XML_CIB_TAG_TICKETS);
if (numXpathResults(xpathObj) > 0) {
xmlNode *aborted = getXpathResult(xpathObj, 0);
abort_transition(INFINITY, tg_restart, "Ticket attribute: removal", aborted);
goto bail;
}
freeXpathObject(xpathObj);
/* Transient Attributes - Added/Updated */
xpathObj =
xpath_search(diff,
"//" F_CIB_UPDATE_RESULT "//" XML_TAG_DIFF_ADDED "//"
XML_TAG_TRANSIENT_NODEATTRS "//" XML_CIB_TAG_NVPAIR);
max = numXpathResults(xpathObj);
for (lpc = 0; lpc < max; lpc++) {
xmlNode *attr = getXpathResult(xpathObj, lpc);
const char *name = crm_element_value(attr, XML_NVPAIR_ATTR_NAME);
const char *value = NULL;
if (safe_str_eq(CRM_OP_PROBED, name)) {
value = crm_element_value(attr, XML_NVPAIR_ATTR_VALUE);
}
if (crm_is_true(value) == FALSE) {
abort_transition(INFINITY, tg_restart, "Transient attribute: update", attr);
crm_log_xml_trace(attr, "Abort");
goto bail;
}
}
freeXpathObject(xpathObj);
/* Transient Attributes - Removed */
xpathObj =
xpath_search(diff,
"//" F_CIB_UPDATE_RESULT "//" XML_TAG_DIFF_REMOVED "//"
XML_TAG_TRANSIENT_NODEATTRS);
if (numXpathResults(xpathObj) > 0) {
xmlNode *aborted = getXpathResult(xpathObj, 0);
abort_transition(INFINITY, tg_restart, "Transient attribute: removal", aborted);
goto bail;
}
freeXpathObject(xpathObj);
/*
* Check for and fast-track the processing of LRM refreshes
* In large clusters this can result in _huge_ speedups
*
* Unfortunately we can only do so when there are no pending actions
* Otherwise we could miss updates we're waiting for and stall
*
*/
xpathObj = NULL;
if (transition_graph->pending == 0) {
xpathObj =
xpath_search(diff,
"//" F_CIB_UPDATE_RESULT "//" XML_TAG_DIFF_ADDED "//"
XML_LRM_TAG_RESOURCE);
}
max = numXpathResults(xpathObj);
if (max > 1) {
/* Updates by, or in response to, TE actions will never contain updates
* for more than one resource at a time
*/
crm_debug("Detected LRM refresh - %d resources updated: Skipping all resource events", max);
crm_log_xml_trace(diff, "lrm-refresh");
abort_transition(INFINITY, tg_restart, "LRM Refresh", NULL);
goto bail;
}
freeXpathObject(xpathObj);
/* Process operation updates */
xpathObj =
xpath_search(diff,
"//" F_CIB_UPDATE_RESULT "//" XML_TAG_DIFF_ADDED "//" XML_LRM_TAG_RSC_OP);
if (numXpathResults(xpathObj)) {
/*
<status>
<node_state id="node1" state=CRMD_JOINSTATE_MEMBER exp_state="active">
<lrm>
<lrm_resources>
<rsc_state id="" rsc_id="rsc4" node_id="node1" rsc_state="stopped"/>
*/
int lpc = 0, max = numXpathResults(xpathObj);
for (lpc = 0; lpc < max; lpc++) {
xmlNode *rsc_op = getXpathResult(xpathObj, lpc);
const char *node = get_node_id(rsc_op);
process_graph_event(rsc_op, node);
}
}
freeXpathObject(xpathObj);
/* Detect deleted (as opposed to replaced or added) actions - eg. crm_resource -C */
xpathObj = xpath_search(diff, "//" XML_TAG_DIFF_REMOVED "//" XML_LRM_TAG_RSC_OP);
max = numXpathResults(xpathObj);
for (lpc = 0; lpc < max; lpc++) {
int path_max = 0;
const char *op_id = NULL;
char *rsc_op_xpath = NULL;
xmlXPathObject *op_match = NULL;
xmlNode *match = getXpathResult(xpathObj, lpc);
CRM_LOG_ASSERT(match != NULL);
if(match == NULL) { continue; };
op_id = ID(match);
path_max = strlen(rsc_op_template) + strlen(op_id) + 1;
rsc_op_xpath = calloc(1, path_max);
snprintf(rsc_op_xpath, path_max, rsc_op_template, op_id);
op_match = xpath_search(diff, rsc_op_xpath);
if (numXpathResults(op_match) == 0) {
/* Prevent false positives by matching cancelations too */
const char *node = get_node_id(match);
crm_action_t *cancelled = get_cancel_action(op_id, node);
if (cancelled == NULL) {
crm_debug("No match for deleted action %s (%s on %s)", rsc_op_xpath, op_id,
node);
abort_transition(INFINITY, tg_restart, "Resource op removal", match);
freeXpathObject(op_match);
free(rsc_op_xpath);
goto bail;
} else {
crm_debug("Deleted lrm_rsc_op %s on %s was for graph event %d",
op_id, node, cancelled->id);
}
}
freeXpathObject(op_match);
free(rsc_op_xpath);
}
bail:
freeXpathObject(xpathObj);
}
static void
tengine_stonith_notify(stonith_t * st, stonith_event_t * st_event)
{
if(te_client_id == NULL) {
te_client_id = crm_strdup_printf("%s.%d", crm_system_name, getpid());
}
if (st_event == NULL) {
crm_err("Notify data not found");
return;
}
crmd_notify_fencing_op(st_event);
if (st_event->result == pcmk_ok && safe_str_eq("on", st_event->action)) {
crm_notice("%s was successfully unfenced by %s (at the request of %s)",
st_event->target, st_event->executioner ? st_event->executioner : "<anyone>", st_event->origin);
/* TODO: Hook up st_event->device */
return;
} else if (safe_str_eq("on", st_event->action)) {
crm_err("Unfencing of %s by %s failed: %s (%d)",
st_event->target, st_event->executioner ? st_event->executioner : "<anyone>",
pcmk_strerror(st_event->result), st_event->result);
return;
} else if (st_event->result == pcmk_ok && crm_str_eq(st_event->target, fsa_our_uname, TRUE)) {
crm_crit("We were allegedly just fenced by %s for %s!",
st_event->executioner ? st_event->executioner : "<anyone>", st_event->origin); /* Dumps blackbox if enabled */
qb_log_fini(); /* Try to get the above log message to disk - somehow */
/* Get out ASAP and do not come back up.
*
* Triggering a reboot is also not the worst idea either since
* the rest of the cluster thinks we're safely down
*/
#ifdef RB_HALT_SYSTEM
reboot(RB_HALT_SYSTEM);
#endif
/*
* If reboot() fails or is not supported, coming back up will
* probably lead to a situation where the other nodes set our
* status to 'lost' because of the fencing callback and will
* discard subsequent election votes with:
*
* Election 87 (current: 5171, owner: 103): Processed vote from east-03 (Peer is not part of our cluster)
*
* So just stay dead, something is seriously messed up anyway.
*
*/
exit(100); /* None of our wrappers since we already called qb_log_fini() */
return;
}
if (st_event->result == pcmk_ok &&
safe_str_eq(st_event->operation, T_STONITH_NOTIFY_FENCE)) {
st_fail_count_reset(st_event->target);
}
crm_notice("Peer %s was%s terminated (%s) by %s for %s: %s (ref=%s) by client %s",
st_event->target, st_event->result == pcmk_ok ? "" : " not",
st_event->action,
st_event->executioner ? st_event->executioner : "<anyone>",
st_event->origin, pcmk_strerror(st_event->result), st_event->id,
st_event->client_origin ? st_event->client_origin : "<unknown>");
#if SUPPORT_CMAN
if (st_event->result == pcmk_ok && is_cman_cluster()) {
int local_rc = 0;
int confirm = 0;
char *target_copy = strdup(st_event->target);
/* In case fenced hasn't noticed yet
*
* Any fencing that has been inititated will be completed by way of the fence_pcmk redirect
*/
local_rc = fenced_external(target_copy);
if (local_rc != 0) {
crm_err("Could not notify CMAN that '%s' is now fenced: %d", st_event->target,
local_rc);
} else {
crm_notice("Notified CMAN that '%s' is now fenced", st_event->target);
}
/* In case fenced is already trying to shoot it */
confirm = open("/var/run/cluster/fenced_override", O_NONBLOCK|O_WRONLY);
if (confirm > 0) {
int ignore = 0;
int len = strlen(target_copy);
errno = 0;
local_rc = write(confirm, target_copy, len);
ignore = write(confirm, "\n", 1);
if(ignore < 0 && errno == EBADF) {
crm_trace("CMAN not expecting %s to be fenced (yet)", st_event->target);
} else if (local_rc < len) {
crm_perror(LOG_ERR, "Confirmation of CMAN fencing event for '%s' failed: %d", st_event->target, local_rc);
} else {
fsync(confirm);
crm_notice("Confirmed CMAN fencing event for '%s'", st_event->target);
}
close(confirm);
}
free(target_copy);
}
#endif
if (st_event->result == pcmk_ok) {
crm_node_t *peer = crm_find_peer_full(0, st_event->target, CRM_GET_PEER_REMOTE | CRM_GET_PEER_CLUSTER);
const char *uuid = NULL;
gboolean we_are_executioner = safe_str_eq(st_event->executioner, fsa_our_uname);
if (peer == NULL) {
return;
}
uuid = crm_peer_uuid(peer);
crm_trace("target=%s dc=%s", st_event->target, fsa_our_dc);
if(AM_I_DC) {
/* The DC always sends updates */
send_stonith_update(NULL, st_event->target, uuid);
if (st_event->client_origin && safe_str_neq(st_event->client_origin, te_client_id)) {
/* Abort the current transition graph if it wasn't us
* that invoked stonith to fence someone
*/
crm_info("External fencing operation from %s fenced %s", st_event->client_origin, st_event->target);
abort_transition(INFINITY, tg_restart, "External Fencing Operation", NULL);
}
/* Assume it was our leader if we dont currently have one */
} else if (fsa_our_dc == NULL || safe_str_eq(fsa_our_dc, st_event->target)) {
crm_notice("Target %s our leader %s (recorded: %s)",
fsa_our_dc ? "was" : "may have been", st_event->target,
fsa_our_dc ? fsa_our_dc : "<unset>");
/* Given the CIB resyncing that occurs around elections,
* have one node update the CIB now and, if the new DC is different,
* have them do so too after the election
*/
if (we_are_executioner) {
send_stonith_update(NULL, st_event->target, uuid);
}
stonith_cleanup_list = g_list_append(stonith_cleanup_list, strdup(st_event->target));
}
crmd_peer_down(peer, TRUE);
}
}
gboolean
process_graph_event(xmlNode * event, const char *event_node)
{
int rc = -1;
int status = -1;
int callid = -1;
int action = -1;
int target_rc = -1;
int transition_num = -1;
char *update_te_uuid = NULL;
gboolean stop_early = FALSE;
gboolean passed = FALSE;
const char *id = NULL;
const char *desc = NULL;
const char *magic = NULL;
CRM_ASSERT(event != NULL);
/*
<lrm_rsc_op id="rsc_east-05_last_0" operation_key="rsc_east-05_monitor_0" operation="monitor" crm-debug-origin="do_update_resource" crm_feature_set="3.0.6" transition-key="9:2:7:be2e97d9-05e2-439d-863e-48f7aecab2aa" transition-magic="0:7;9:2:7:be2e97d9-05e2-439d-863e-48f7aecab2aa" call-id="17" rc-code="7" op-status="0" interval="0" last-run="1355361636" last-rc-change="1355361636" exec-time="128" queue-time="0" op-digest="c81f5f40b1c9e859c992e800b1aa6972"/>
*/
id = crm_element_value(event, XML_LRM_ATTR_TASK_KEY);
crm_element_value_int(event, XML_LRM_ATTR_RC, &rc);
crm_element_value_int(event, XML_LRM_ATTR_OPSTATUS, &status);
crm_element_value_int(event, XML_LRM_ATTR_CALLID, &callid);
magic = crm_element_value(event, XML_ATTR_TRANSITION_KEY);
if (magic == NULL) {
/* non-change */
return FALSE;
}
if (decode_transition_key(magic, &update_te_uuid, &transition_num, &action, &target_rc) ==
FALSE) {
crm_err("Invalid event %s.%d detected: %s", id, callid, magic);
abort_transition(INFINITY, tg_restart, "Bad event", event);
return FALSE;
}
if (status == PCMK_LRM_OP_PENDING) {
goto bail;
}
if (transition_num == -1) {
desc = "initiated outside of the cluster";
abort_transition(INFINITY, tg_restart, "Unexpected event", event);
} else if (action < 0 || crm_str_eq(update_te_uuid, te_uuid, TRUE) == FALSE) {
desc = "initiated by a different node";
abort_transition(INFINITY, tg_restart, "Foreign event", event);
stop_early = TRUE; /* This could be an lrm status refresh */
} else if (transition_graph->id != transition_num) {
desc = "arrived really late";
abort_transition(INFINITY, tg_restart, "Old event", event);
stop_early = TRUE; /* This could be an lrm status refresh */
} else if (transition_graph->complete) {
desc = "arrived late";
abort_transition(INFINITY, tg_restart, "Inactive graph", event);
} else if (match_graph_event(action, event, event_node, status, rc, target_rc) < 0) {
desc = "unknown";
abort_transition(INFINITY, tg_restart, "Unknown event", event);
} else if (rc == target_rc) {
passed = TRUE;
crm_trace("Processed update to %s: %s", id, magic);
}
if (passed == FALSE) {
if (update_failcount(event, event_node, rc, target_rc, transition_num == -1)) {
/* Turns out this wasn't an lrm status refresh update aferall */
stop_early = FALSE;
desc = "failed";
}
crm_info("Detected action (%d.%d) %s.%d=%s: %s", transition_num, action, id, callid,
services_ocf_exitcode_str(rc), desc);
}
bail:
free(update_te_uuid);
return stop_early;
}
/*
* returns the ID of the action if a match is found
* returns -1 if a match was not found
* returns -2 if a match was found but the action failed (and was
* not allowed to)
*/
int
match_graph_event(int action_id, xmlNode * event, const char *event_node,
int op_status, int op_rc, int target_rc)
{
const char *target = NULL;
const char *allow_fail = NULL;
const char *this_event = NULL;
crm_action_t *action = NULL;
action = get_action(action_id, FALSE);
if (action == NULL) {
return -1;
}
op_status = status_from_rc(action, op_status, op_rc, target_rc);
if (op_status != PCMK_LRM_OP_DONE) {
update_failcount(event, event_node, op_rc, target_rc, FALSE);
}
/* Process OP status */
switch (op_status) {
case PCMK_LRM_OP_PENDING:
crm_debug("Ignoring pending operation");
return action->id;
break;
case PCMK_LRM_OP_DONE:
break;
case PCMK_LRM_OP_ERROR:
case PCMK_LRM_OP_TIMEOUT:
case PCMK_LRM_OP_NOTSUPPORTED:
action->failed = TRUE;
break;
case PCMK_LRM_OP_CANCELLED:
/* do nothing?? */
crm_err("Dont know what to do for cancelled ops yet");
break;
default:
action->failed = TRUE;
crm_err("Unsupported action result: %d", op_status);
}
/* stop this event's timer if it had one */
stop_te_timer(action->timer);
te_action_confirmed(action);
update_graph(transition_graph, action);
trigger_graph();
if (action->failed) {
allow_fail = crm_meta_value(action->params, XML_ATTR_TE_ALLOWFAIL);
if (crm_is_true(allow_fail)) {
action->failed = FALSE;
}
}
if (action->failed) {
abort_transition(action->synapse->priority + 1, tg_restart, "Event failed", event);
}
this_event = crm_element_value(event, XML_LRM_ATTR_TASK_KEY);
target = crm_element_value(action->xml, XML_LRM_ATTR_TARGET);
crm_info("Action %s (%d) confirmed on %s (rc=%d)",
crm_str(this_event), action->id, crm_str(target), op_status);
/* determine if this action affects a remote-node's online/offline status */
process_remote_node_action(action, event);
return action->id;
}
static void
te_update_diff_v1(const char *event, xmlNode *diff)
{
int lpc, max;
xmlXPathObject *xpathObj = NULL;
CRM_CHECK(diff != NULL, return);
xml_log_patchset(LOG_TRACE, __FUNCTION__, diff);
if (cib_config_changed(NULL, NULL, &diff)) {
abort_transition(INFINITY, tg_restart, "Non-status change", diff);
goto bail; /* configuration changed */
}
/* Tickets Attributes - Added/Updated */
xpathObj =
xpath_search(diff,
"//" F_CIB_UPDATE_RESULT "//" XML_TAG_DIFF_ADDED "//" XML_CIB_TAG_TICKETS);
if (numXpathResults(xpathObj) > 0) {
xmlNode *aborted = getXpathResult(xpathObj, 0);
abort_transition(INFINITY, tg_restart, "Ticket attribute: update", aborted);
goto bail;
}
freeXpathObject(xpathObj);
/* Tickets Attributes - Removed */
xpathObj =
xpath_search(diff,
"//" F_CIB_UPDATE_RESULT "//" XML_TAG_DIFF_REMOVED "//" XML_CIB_TAG_TICKETS);
if (numXpathResults(xpathObj) > 0) {
xmlNode *aborted = getXpathResult(xpathObj, 0);
abort_transition(INFINITY, tg_restart, "Ticket attribute: removal", aborted);
goto bail;
}
freeXpathObject(xpathObj);
/* Transient Attributes - Added/Updated */
xpathObj =
xpath_search(diff,
"//" F_CIB_UPDATE_RESULT "//" XML_TAG_DIFF_ADDED "//"
XML_TAG_TRANSIENT_NODEATTRS "//" XML_CIB_TAG_NVPAIR);
max = numXpathResults(xpathObj);
for (lpc = 0; lpc < max; lpc++) {
xmlNode *attr = getXpathResult(xpathObj, lpc);
const char *name = crm_element_value(attr, XML_NVPAIR_ATTR_NAME);
const char *value = NULL;
if (safe_str_eq(CRM_OP_PROBED, name)) {
value = crm_element_value(attr, XML_NVPAIR_ATTR_VALUE);
}
if (crm_is_true(value) == FALSE) {
abort_transition(INFINITY, tg_restart, "Transient attribute: update", attr);
crm_log_xml_trace(attr, "Abort");
goto bail;
}
}
freeXpathObject(xpathObj);
/* Transient Attributes - Removed */
xpathObj =
xpath_search(diff,
"//" F_CIB_UPDATE_RESULT "//" XML_TAG_DIFF_REMOVED "//"
XML_TAG_TRANSIENT_NODEATTRS);
if (numXpathResults(xpathObj) > 0) {
xmlNode *aborted = getXpathResult(xpathObj, 0);
abort_transition(INFINITY, tg_restart, "Transient attribute: removal", aborted);
goto bail;
}
freeXpathObject(xpathObj);
/*
* Updates by, or in response to, TE actions will never contain updates
* for more than one resource at a time, so such updates indicate an
* LRM refresh.
*
* In that case, start a new transition rather than check each result
* individually, which can result in _huge_ speedups in large clusters.
*
* Unfortunately, we can only do so when there are no pending actions.
* Otherwise, we could mistakenly throw away those results here, and
* the cluster will stall waiting for them and time out the operation.
*/
if (transition_graph->pending == 0) {
xpathObj = xpath_search(diff,
"//" F_CIB_UPDATE_RESULT
"//" XML_TAG_DIFF_ADDED
"//" XML_LRM_TAG_RESOURCE);
max = numXpathResults(xpathObj);
if (max > 1) {
crm_debug("Ignoring resource operation updates due to history refresh of %d resources",
max);
crm_log_xml_trace(diff, "lrm-refresh");
abort_transition(INFINITY, tg_restart, "History refresh", NULL);
goto bail;
}
freeXpathObject(xpathObj);
}
/* Process operation updates */
xpathObj =
xpath_search(diff,
"//" F_CIB_UPDATE_RESULT "//" XML_TAG_DIFF_ADDED "//" XML_LRM_TAG_RSC_OP);
max = numXpathResults(xpathObj);
if (max > 0) {
int lpc = 0;
for (lpc = 0; lpc < max; lpc++) {
xmlNode *rsc_op = getXpathResult(xpathObj, lpc);
const char *node = get_node_id(rsc_op);
process_graph_event(rsc_op, node);
}
}
freeXpathObject(xpathObj);
/* Detect deleted (as opposed to replaced or added) actions - eg. crm_resource -C */
xpathObj = xpath_search(diff, "//" XML_TAG_DIFF_REMOVED "//" XML_LRM_TAG_RSC_OP);
max = numXpathResults(xpathObj);
for (lpc = 0; lpc < max; lpc++) {
int path_max = 0;
const char *op_id = NULL;
char *rsc_op_xpath = NULL;
xmlXPathObject *op_match = NULL;
xmlNode *match = getXpathResult(xpathObj, lpc);
CRM_LOG_ASSERT(match != NULL);
if(match == NULL) { continue; };
op_id = ID(match);
path_max = strlen(RSC_OP_TEMPLATE) + strlen(op_id) + 1;
rsc_op_xpath = calloc(1, path_max);
snprintf(rsc_op_xpath, path_max, RSC_OP_TEMPLATE, op_id);
op_match = xpath_search(diff, rsc_op_xpath);
if (numXpathResults(op_match) == 0) {
/* Prevent false positives by matching cancelations too */
const char *node = get_node_id(match);
crm_action_t *cancelled = get_cancel_action(op_id, node);
if (cancelled == NULL) {
crm_debug("No match for deleted action %s (%s on %s)", rsc_op_xpath, op_id,
node);
abort_transition(INFINITY, tg_restart, "Resource op removal", match);
freeXpathObject(op_match);
free(rsc_op_xpath);
goto bail;
} else {
crm_debug("Deleted lrm_rsc_op %s on %s was for graph event %d",
op_id, node, cancelled->id);
}
}
freeXpathObject(op_match);
free(rsc_op_xpath);
}
bail:
freeXpathObject(xpathObj);
}
static void
te_update_diff_v2(xmlNode *diff)
{
crm_log_xml_trace(diff, "Patch:Raw");
for (xmlNode *change = __xml_first_child(diff); change != NULL;
change = __xml_next(change)) {
xmlNode *match = NULL;
const char *name = NULL;
const char *xpath = crm_element_value(change, XML_DIFF_PATH);
// Possible ops: create, modify, delete, move
const char *op = crm_element_value(change, XML_DIFF_OP);
// Ignore uninteresting updates
if (op == NULL) {
continue;
} else if (xpath == NULL) {
crm_trace("Ignoring %s change for version field", op);
continue;
} else if (strcmp(op, "move") == 0) {
crm_trace("Ignoring move change at %s", xpath);
continue;
}
// Find the result of create/modify ops
if (strcmp(op, "create") == 0) {
match = change->children;
} else if (strcmp(op, "modify") == 0) {
match = first_named_child(change, XML_DIFF_RESULT);
if(match) {
match = match->children;
}
} else if (strcmp(op, "delete") != 0) {
crm_warn("Ignoring malformed CIB update (%s operation on %s is unrecognized)",
op, xpath);
continue;
}
if (match) {
if (match->type == XML_COMMENT_NODE) {
crm_trace("Ignoring %s operation for comment at %s", op, xpath);
continue;
}
name = (const char *)match->name;
}
crm_trace("Handling %s operation for %s%s%s",
op, (xpath? xpath : "CIB"),
(name? " matched by " : ""), (name? name : ""));
if (strstr(xpath, "/" XML_TAG_CIB "/" XML_CIB_TAG_CONFIGURATION)) {
abort_transition(INFINITY, tg_restart, "Configuration change",
change);
break; // Won't be packaged with operation results we may be waiting for
} else if (strstr(xpath, "/" XML_CIB_TAG_TICKETS)
|| safe_str_eq(name, XML_CIB_TAG_TICKETS)) {
abort_transition(INFINITY, tg_restart, "Ticket attribute change", change);
break; // Won't be packaged with operation results we may be waiting for
} else if (strstr(xpath, "/" XML_TAG_TRANSIENT_NODEATTRS "[")
|| safe_str_eq(name, XML_TAG_TRANSIENT_NODEATTRS)) {
abort_unless_down(xpath, op, change, "Transient attribute change");
break; // Won't be packaged with operation results we may be waiting for
} else if (strcmp(op, "delete") == 0) {
process_delete_diff(xpath, op, change);
} else if (name == NULL) {
crm_warn("Ignoring malformed CIB update (%s at %s has no result)",
op, xpath);
} else if (strcmp(name, XML_TAG_CIB) == 0) {
process_cib_diff(match, change, op, xpath);
} else if (strcmp(name, XML_CIB_TAG_STATUS) == 0) {
process_status_diff(match, change, op, xpath);
} else if (strcmp(name, XML_CIB_TAG_STATE) == 0) {
process_node_state_diff(match, change, op, xpath);
} else if (strcmp(name, XML_CIB_TAG_LRM) == 0) {
process_resource_updates(ID(match), match, change, op, xpath);
} else if (strcmp(name, XML_LRM_TAG_RESOURCES) == 0) {
char *local_node = get_node_from_xpath(xpath);
process_resource_updates(local_node, match, change, op, xpath);
free(local_node);
} else if (strcmp(name, XML_LRM_TAG_RESOURCE) == 0) {
char *local_node = get_node_from_xpath(xpath);
process_lrm_resource_diff(match, local_node);
free(local_node);
} else if (strcmp(name, XML_LRM_TAG_RSC_OP) == 0) {
char *local_node = get_node_from_xpath(xpath);
process_graph_event(match, local_node);
free(local_node);
} else {
crm_warn("Ignoring malformed CIB update (%s at %s has unrecognized result %s)",
op, xpath, name);
}
}
}
static void
tengine_stonith_notify(stonith_t * st, stonith_event_t *st_event)
{
if (st_event == NULL) {
crm_err("Notify data not found");
return;
}
if (st_event->result == pcmk_ok && crm_str_eq(st_event->target, fsa_our_uname, TRUE)) {
crm_err("We were alegedly just fenced by %s for %s!", st_event->executioner, st_event->origin);
register_fsa_error_adv(C_FSA_INTERNAL, I_ERROR, NULL, NULL, __FUNCTION__);
return;
}
crm_notice("Peer %s was%s terminated (%s) by %s for %s: %s (ref=%s) by client %s",
st_event->target, st_event->result == pcmk_ok?"":" not",
st_event->operation,
st_event->executioner ? st_event->executioner : "<anyone>",
st_event->origin, pcmk_strerror(st_event->result), st_event->id,
st_event->client_origin ? st_event->client_origin : "<unknown>");
#if SUPPORT_CMAN
if (st_event->result == pcmk_ok && is_cman_cluster()) {
int local_rc = 0;
int confirm = 0;
char *target_copy = strdup(st_event->target);
/* In case fenced hasn't noticed yet */
local_rc = fenced_external(target_copy);
if (local_rc != 0) {
crm_err("Could not notify CMAN that '%s' is now fenced: %d", st_event->target, local_rc);
} else {
crm_notice("Notified CMAN that '%s' is now fenced", st_event->target);
}
/* In case fenced is already trying to shoot it */
confirm = open("/var/run/cluster/fenced_override", O_NONBLOCK|O_WRONLY);
if (confirm) {
int ignore = 0;
int len = strlen(target_copy);
errno = 0;
local_rc = write(confirm, target_copy, len);
ignore = write(confirm, "\n", 1);
if(errno == EBADF) {
crm_trace("CMAN not expecting %s to be fenced (yet)", st_event->target);
} else if (local_rc < len) {
crm_perror(LOG_ERR, "Confirmation of CMAN fencing event for '%s' failed: %d", st_event->target, local_rc);
} else {
fsync(confirm);
crm_notice("Confirmed CMAN fencing event for '%s'", st_event->target);
}
close(confirm);
}
}
#endif
if (st_event->result == pcmk_ok) {
gboolean we_are_executioner = safe_str_eq(st_event->executioner, fsa_our_uname);
crm_trace("target=%s dc=%s", st_event->target, fsa_our_dc);
if (fsa_our_dc == NULL || safe_str_eq(fsa_our_dc, st_event->target)) {
crm_notice("Target %s our leader %s (recorded: %s)",
fsa_our_dc?"was":"may have been", st_event->target, fsa_our_dc ? fsa_our_dc : "<unset>");
/* Given the CIB resyncing that occurs around elections,
* have one node update the CIB now and, if the new DC is different,
* have them do so too after the election
*/
if (we_are_executioner) {
const char *uuid = get_uuid(st_event->target);
send_stonith_update(NULL, st_event->target, uuid);
}
stonith_cleanup_list = g_list_append(stonith_cleanup_list, strdup(st_event->target));
} else if (AM_I_DC &&
st_event->client_origin &&
safe_str_neq(st_event->client_origin, crm_system_name)) {
const char *uuid = get_uuid(st_event->target);
/* If a remote process outside of pacemaker invoked stonith to
* fence someone, report the fencing result to the cib
* and abort the transition graph. */
crm_info("External fencing operation from %s fenced %s", st_event->client_origin, st_event->target);
send_stonith_update(NULL, st_event->target, uuid);
abort_transition(INFINITY, tg_restart, "External Fencing Operation", NULL);
}
}
}
gboolean tengine_shutdown(int nsig, gpointer unused)
{
shuttingdown = TRUE;
abort_transition(INFINITY, tg_shutdown, "Shutdown", NULL);
return TRUE;
}
/* A_DC_JOIN_OFFER_ONE */
void
do_dc_join_offer_one(long long action,
enum crmd_fsa_cause cause,
enum crmd_fsa_state cur_state,
enum crmd_fsa_input current_input, fsa_data_t * msg_data)
{
crm_node_t *member;
ha_msg_input_t *welcome = NULL;
const char *op = NULL;
const char *join_to = NULL;
if (msg_data->data) {
welcome = fsa_typed_data(fsa_dt_ha_msg);
} else {
crm_info("An unknown node joined - (re-)offer to any unconfirmed nodes");
g_hash_table_foreach(crm_peer_cache, join_make_offer, &member);
check_join_state(cur_state, __FUNCTION__);
return;
}
if (welcome == NULL) {
crm_err("Attempt to send welcome message without a message to reply to!");
return;
}
join_to = crm_element_value(welcome->msg, F_CRM_HOST_FROM);
if (join_to == NULL) {
crm_err("Attempt to send welcome message without a host to reply to!");
return;
}
member = crm_get_peer(0, join_to);
op = crm_element_value(welcome->msg, F_CRM_TASK);
if (join_to != NULL && (cur_state == S_INTEGRATION || cur_state == S_FINALIZE_JOIN)) {
/* note: it _is_ possible that a node will have been
* sick or starting up when the original offer was made.
* however, it will either re-announce itself in due course
* _or_ we can re-store the original offer on the client.
*/
crm_trace("(Re-)offering membership to %s...", join_to);
}
crm_info("join-%d: Processing %s request from %s in state %s",
current_join_id, op, join_to, fsa_state2string(cur_state));
crm_update_peer_join(__FUNCTION__, member, crm_join_none);
join_make_offer(NULL, member, NULL);
/* always offer to the DC (ourselves)
* this ensures the correct value for max_generation_from
*/
member = crm_get_peer(0, fsa_our_uname);
join_make_offer(NULL, member, NULL);
/* this was a genuine join request, cancel any existing
* transition and invoke the PE
*/
abort_transition(INFINITY, tg_restart, "Node join", NULL);
/* don't waste time by invoking the PE yet; */
crm_debug("Waiting on %d outstanding join acks for join-%d",
crmd_join_phase_count(crm_join_welcomed), current_join_id);
}
gboolean
process_graph_event(xmlNode * event, const char *event_node)
{
int rc = -1;
int status = -1;
int action = -1;
int target_rc = -1;
int transition_num = -1;
char *update_te_uuid = NULL;
gboolean stop_early = FALSE;
gboolean passed = FALSE;
const char *id = NULL;
const char *magic = NULL;
CRM_ASSERT(event != NULL);
id = crm_element_value(event, XML_LRM_ATTR_TASK_KEY);
magic = crm_element_value(event, XML_ATTR_TRANSITION_MAGIC);
if (magic == NULL) {
/* non-change */
return FALSE;
}
if(decode_transition_magic(magic, &update_te_uuid, &transition_num, &action,
&status, &rc, &target_rc) == FALSE) {
crm_err("Invalid event %s detected: %s", id, magic);
abort_transition(INFINITY, tg_restart, "Bad event", event);
return FALSE;
}
if (status == LRM_OP_PENDING) {
goto bail;
}
if (transition_num == -1) {
crm_err("Action %s (%s) initiated outside of a transition", id, magic);
abort_transition(INFINITY, tg_restart, "Unexpected event", event);
} else if (action < 0 || crm_str_eq(update_te_uuid, te_uuid, TRUE) == FALSE) {
crm_info("Action %s/%d (%s) initiated by a different transitioner", id, action, magic);
abort_transition(INFINITY, tg_restart, "Foreign event", event);
stop_early = TRUE; /* This could be an lrm status refresh */
} else if (transition_graph->id != transition_num) {
crm_info("Detected action %s from a different transition:"
" %d vs. %d", id, transition_num, transition_graph->id);
abort_transition(INFINITY, tg_restart, "Old event", event);
stop_early = TRUE; /* This could be an lrm status refresh */
} else if (transition_graph->complete) {
crm_info("Action %s arrived after a completed transition", id);
abort_transition(INFINITY, tg_restart, "Inactive graph", event);
} else if (match_graph_event(action, event, event_node, status, rc, target_rc) < 0) {
crm_err("Unknown graph action %s", id);
abort_transition(INFINITY, tg_restart, "Unknown event", event);
} else {
passed = TRUE;
crm_debug_2("Processed update to %s: %s", id, magic);
}
if (passed == FALSE) {
if (update_failcount(event, event_node, rc, target_rc, transition_num == -1)) {
/* Turns out this wasn't an lrm status refresh update aferall */
stop_early = FALSE;
}
}
bail:
crm_free(update_te_uuid);
return stop_early;
}
void
peer_update_callback(enum crm_status_type type, crm_node_t * node, const void *data)
{
uint32_t old = 0;
uint32_t changed = 0;
bool appeared = FALSE;
const char *status = NULL;
set_bit(fsa_input_register, R_PEER_DATA);
if (node->uname == NULL) {
return;
}
switch (type) {
case crm_status_uname:
/* If we've never seen the node, then it also wont be in the status section */
crm_info("%s is now %s", node->uname, node->state);
return;
case crm_status_rstate:
crm_info("Remote node %s is now %s (was %s)", node->uname, node->state, (const char *)data);
/* Keep going */
case crm_status_nstate:
crm_info("%s is now %s (was %s)", node->uname, node->state, (const char *)data);
if (safe_str_eq(data, node->state)) {
/* State did not change */
return;
} else if(safe_str_eq(CRM_NODE_MEMBER, node->state)) {
appeared = TRUE;
}
break;
case crm_status_processes:
if (data) {
old = *(const uint32_t *)data;
changed = node->processes ^ old;
}
/* crmd_proc_update(node, proc_flags); */
status = (node->processes & proc_flags) ? ONLINESTATUS : OFFLINESTATUS;
crm_info("Client %s/%s now has status [%s] (DC=%s, changed=%6x)",
node->uname, peer2text(proc_flags), status,
AM_I_DC ? "true" : crm_str(fsa_our_dc), changed);
if ((changed & proc_flags) == 0) {
/* Peer process did not change */
crm_trace("No change %6x %6x %6x", old, node->processes, proc_flags);
return;
} else if (is_not_set(fsa_input_register, R_CIB_CONNECTED)) {
crm_trace("Not connected");
return;
} else if (fsa_state == S_STOPPING) {
crm_trace("Stopping");
return;
}
appeared = (node->processes & proc_flags) != 0;
if (safe_str_eq(node->uname, fsa_our_uname) && (node->processes & proc_flags) == 0) {
/* Did we get evicted? */
crm_notice("Our peer connection failed");
register_fsa_input(C_CRMD_STATUS_CALLBACK, I_ERROR, NULL);
} else if (safe_str_eq(node->uname, fsa_our_dc) && crm_is_peer_active(node) == FALSE) {
/* Did the DC leave us? */
crm_notice("Our peer on the DC (%s) is dead", fsa_our_dc);
register_fsa_input(C_CRMD_STATUS_CALLBACK, I_ELECTION, NULL);
} else if(AM_I_DC && appeared == FALSE) {
crm_info("Peer %s left us", node->uname);
/* crm_update_peer_join(__FUNCTION__, node, crm_join_none); */
}
break;
}
if (AM_I_DC) {
xmlNode *update = NULL;
int flags = node_update_peer;
gboolean alive = crm_is_peer_active(node);
crm_action_t *down = match_down_event(0, node->uuid, NULL, appeared);
crm_trace("Alive=%d, appear=%d, down=%p", alive, appeared, down);
if (alive && type == crm_status_processes) {
register_fsa_input_before(C_FSA_INTERNAL, I_NODE_JOIN, NULL);
}
if (down) {
const char *task = crm_element_value(down->xml, XML_LRM_ATTR_TASK);
if (alive && safe_str_eq(task, CRM_OP_FENCE)) {
crm_info("Node return implies stonith of %s (action %d) completed", node->uname,
down->id);
erase_status_tag(node->uname, XML_CIB_TAG_LRM, cib_scope_local);
erase_status_tag(node->uname, XML_TAG_TRANSIENT_NODEATTRS, cib_scope_local);
/* down->confirmed = TRUE; Only stonith-ng returning should imply completion */
down->sent_update = TRUE; /* Prevent tengine_stonith_callback() from calling send_stonith_update() */
} else if (safe_str_eq(task, CRM_OP_FENCE)) {
crm_trace("Waiting for stonithd to report the fencing of %s is complete", node->uname); /* via tengine_stonith_callback() */
} else if (alive == FALSE) {
crm_notice("%s of %s (op %d) is complete", task, node->uname, down->id);
/* down->confirmed = TRUE; Only stonith-ng returning should imply completion */
stop_te_timer(down->timer);
flags |= node_update_join | node_update_expected;
crmd_peer_down(node, FALSE);
check_join_state(fsa_state, __FUNCTION__);
update_graph(transition_graph, down);
trigger_graph();
} else {
crm_trace("Other %p", down);
}
} else if (appeared == FALSE) {
crm_notice("Stonith/shutdown of %s not matched", node->uname);
crm_update_peer_join(__FUNCTION__, node, crm_join_none);
check_join_state(fsa_state, __FUNCTION__);
abort_transition(INFINITY, tg_restart, "Node failure", NULL);
fail_incompletable_actions(transition_graph, node->uuid);
} else {
crm_trace("Other %p", down);
}
update = do_update_node_cib(node, flags, NULL, __FUNCTION__);
fsa_cib_anon_update(XML_CIB_TAG_STATUS, update,
cib_scope_local | cib_quorum_override | cib_can_create);
free_xml(update);
}
trigger_fsa(fsa_source);
}
void
te_update_diff(const char *event, xmlNode * msg)
{
int rc = -EINVAL;
int format = 1;
xmlNode *change = NULL;
const char *op = NULL;
xmlNode *diff = NULL;
int p_add[] = { 0, 0, 0 };
int p_del[] = { 0, 0, 0 };
CRM_CHECK(msg != NULL, return);
crm_element_value_int(msg, F_CIB_RC, &rc);
if (transition_graph == NULL) {
crm_trace("No graph");
return;
} else if (rc < pcmk_ok) {
crm_trace("Filter rc=%d (%s)", rc, pcmk_strerror(rc));
return;
} else if (transition_graph->complete == TRUE
&& fsa_state != S_IDLE
&& fsa_state != S_TRANSITION_ENGINE && fsa_state != S_POLICY_ENGINE) {
crm_trace("Filter state=%s, complete=%d", fsa_state2string(fsa_state),
transition_graph->complete);
return;
}
op = crm_element_value(msg, F_CIB_OPERATION);
diff = get_message_xml(msg, F_CIB_UPDATE_RESULT);
xml_patch_versions(diff, p_add, p_del);
crm_debug("Processing (%s) diff: %d.%d.%d -> %d.%d.%d (%s)", op,
p_del[0], p_del[1], p_del[2], p_add[0], p_add[1], p_add[2],
fsa_state2string(fsa_state));
crm_element_value_int(diff, "format", &format);
switch(format) {
case 1:
te_legacy_update_diff(event, diff);
return;
case 2:
/* Cool, we know what to do here */
crm_log_xml_trace(diff, "Patch:Raw");
break;
default:
crm_warn("Unknown patch format: %d", format);
return;
}
for (change = __xml_first_child(diff); change != NULL; change = __xml_next(change)) {
const char *name = NULL;
const char *op = crm_element_value(change, XML_DIFF_OP);
const char *xpath = crm_element_value(change, XML_DIFF_PATH);
xmlNode *match = NULL;
const char *node = NULL;
if(op == NULL) {
continue;
} else if(strcmp(op, "create") == 0) {
match = change->children;
} else if(strcmp(op, "move") == 0) {
continue;
} else if(strcmp(op, "modify") == 0) {
match = first_named_child(change, XML_DIFF_RESULT);
if(match) {
match = match->children;
}
}
if(match) {
name = (const char *)match->name;
}
crm_trace("Handling %s operation for %s %p, %s", op, xpath, match, name);
if(xpath == NULL) {
/* Version field, ignore */
} else if(strstr(xpath, "/cib/configuration")) {
abort_transition(INFINITY, tg_restart, "Non-status change", change);
break; /* Wont be packaged with any resource operations we may be waiting for */
} else if(strstr(xpath, "/"XML_CIB_TAG_TICKETS) || safe_str_eq(name, XML_CIB_TAG_TICKETS)) {
abort_transition(INFINITY, tg_restart, "Ticket attribute change", change);
break; /* Wont be packaged with any resource operations we may be waiting for */
} else if(strstr(xpath, "/"XML_TAG_TRANSIENT_NODEATTRS"[") || safe_str_eq(name, XML_TAG_TRANSIENT_NODEATTRS)) {
abort_transition(INFINITY, tg_restart, "Transient attribute change", change);
break; /* Wont be packaged with any resource operations we may be waiting for */
} else if(strstr(xpath, "/"XML_LRM_TAG_RSC_OP"[") && safe_str_eq(op, "delete")) {
crm_action_t *cancel = NULL;
char *mutable_key = strdup(xpath);
char *mutable_node = strdup(xpath);
char *search = NULL;
const char *key = NULL;
const char *node_uuid = NULL;
search = strrchr(mutable_key, '\'');
search[0] = 0;
key = strrchr(mutable_key, '\'') + 1;
node_uuid = strstr(mutable_node, "node_state[@id=\'") + strlen("node_state[@id=\'");
search = strchr(node_uuid, '\'');
search[0] = 0;
cancel = get_cancel_action(key, node_uuid);
if (cancel == NULL) {
abort_transition(INFINITY, tg_restart, "Resource operation removal", change);
} else {
crm_info("Cancellation of %s on %s confirmed (%d)", key, node_uuid, cancel->id);
stop_te_timer(cancel->timer);
te_action_confirmed(cancel);
update_graph(transition_graph, cancel);
trigger_graph();
}
free(mutable_node);
free(mutable_key);
} else if(strstr(xpath, "/"XML_CIB_TAG_LRM"[") && safe_str_eq(op, "delete")) {
abort_transition(INFINITY, tg_restart, "Resource state removal", change);
} else if(strstr(xpath, "/"XML_CIB_TAG_STATE"[") && safe_str_eq(op, "delete")) {
abort_transition(INFINITY, tg_restart, "Node state removal", change);
} else if(name == NULL) {
crm_debug("No result for %s operation to %s", op, xpath);
CRM_ASSERT(strcmp(op, "delete") == 0 || strcmp(op, "move") == 0);
} else if(strcmp(name, XML_TAG_CIB) == 0) {
xmlNode *state = NULL;
xmlNode *status = first_named_child(match, XML_CIB_TAG_STATUS);
xmlNode *config = first_named_child(match, XML_CIB_TAG_CONFIGURATION);
for (state = __xml_first_child(status); state != NULL; state = __xml_next(state)) {
xmlNode *lrm = first_named_child(state, XML_CIB_TAG_LRM);
node = ID(state);
process_resource_updates(node, lrm, change, op, xpath);
}
if(config) {
abort_transition(INFINITY, tg_restart, "Non-status change", change);
}
} else if(strcmp(name, XML_CIB_TAG_STATUS) == 0) {
xmlNode *state = NULL;
for (state = __xml_first_child(match); state != NULL; state = __xml_next(state)) {
xmlNode *lrm = first_named_child(state, XML_CIB_TAG_LRM);
node = ID(state);
process_resource_updates(node, lrm, change, op, xpath);
}
} else if(strcmp(name, XML_CIB_TAG_STATE) == 0) {
xmlNode *lrm = first_named_child(match, XML_CIB_TAG_LRM);
node = ID(match);
process_resource_updates(node, lrm, change, op, xpath);
} else if(strcmp(name, XML_CIB_TAG_LRM) == 0) {
node = ID(match);
process_resource_updates(node, match, change, op, xpath);
} else if(strcmp(name, XML_LRM_TAG_RESOURCES) == 0) {
char *local_node = get_node_from_xpath(xpath);
process_resource_updates(local_node, match, change, op, xpath);
free(local_node);
} else if(strcmp(name, XML_LRM_TAG_RESOURCE) == 0) {
xmlNode *rsc_op;
char *local_node = get_node_from_xpath(xpath);
for (rsc_op = __xml_first_child(match); rsc_op != NULL; rsc_op = __xml_next(rsc_op)) {
process_graph_event(rsc_op, local_node);
}
free(local_node);
} else if(strcmp(name, XML_LRM_TAG_RSC_OP) == 0) {
char *local_node = get_node_from_xpath(xpath);
process_graph_event(match, local_node);
free(local_node);
} else {
crm_err("Ignoring %s operation for %s %p, %s", op, xpath, match, name);
}
}
}
gboolean
fail_incompletable_actions(crm_graph_t * graph, const char *down_node)
{
const char *target = NULL;
xmlNode *last_action = NULL;
GListPtr gIter = NULL;
GListPtr gIter2 = NULL;
if (graph == NULL || graph->complete) {
return FALSE;
}
gIter = graph->synapses;
for (; gIter != NULL; gIter = gIter->next) {
synapse_t *synapse = (synapse_t *) gIter->data;
if (synapse->confirmed) {
continue;
}
gIter2 = synapse->actions;
for (; gIter2 != NULL; gIter2 = gIter2->next) {
crm_action_t *action = (crm_action_t *) gIter2->data;
if (action->type == action_type_pseudo || action->confirmed) {
continue;
} else if (action->type == action_type_crm) {
const char *task = crm_element_value(action->xml, XML_LRM_ATTR_TASK);
if (safe_str_eq(task, CRM_OP_FENCE)) {
continue;
}
}
target = crm_element_value(action->xml, XML_LRM_ATTR_TARGET_UUID);
if (safe_str_eq(target, down_node)) {
action->failed = TRUE;
synapse->failed = TRUE;
last_action = action->xml;
stop_te_timer(action->timer);
update_graph(graph, action);
if (synapse->executed) {
crm_notice("Action %d (%s) was pending on %s (offline)",
action->id, ID(action->xml), down_node);
} else {
crm_notice("Action %d (%s) is scheduled for %s (offline)",
action->id, ID(action->xml), down_node);
}
}
}
}
if (last_action != NULL) {
crm_warn("Node %s shutdown resulted in un-runnable actions", down_node);
abort_transition(INFINITY, tg_restart, "Node failure", last_action);
return TRUE;
}
return FALSE;
}
void
te_update_diff(const char *event, xmlNode *msg)
{
int rc = -1;
const char *op = NULL;
xmlNode *diff = NULL;
xmlNode *cib_top = NULL;
xmlXPathObject *xpathObj = NULL;
int diff_add_updates = 0;
int diff_add_epoch = 0;
int diff_add_admin_epoch = 0;
int diff_del_updates = 0;
int diff_del_epoch = 0;
int diff_del_admin_epoch = 0;
CRM_CHECK(msg != NULL, return);
crm_element_value_int(msg, F_CIB_RC, &rc);
if(transition_graph == NULL) {
crm_debug_3("No graph");
return;
} else if(rc < cib_ok) {
crm_debug_3("Filter rc=%d (%s)", rc, cib_error2string(rc));
return;
} else if(transition_graph->complete == TRUE
&& fsa_state != S_IDLE
&& fsa_state != S_TRANSITION_ENGINE
&& fsa_state != S_POLICY_ENGINE) {
crm_debug_2("Filter state=%s, complete=%d", fsa_state2string(fsa_state), transition_graph->complete);
return;
}
op = crm_element_value(msg, F_CIB_OPERATION);
diff = get_message_xml(msg, F_CIB_UPDATE_RESULT);
cib_diff_version_details(
diff,
&diff_add_admin_epoch, &diff_add_epoch, &diff_add_updates,
&diff_del_admin_epoch, &diff_del_epoch, &diff_del_updates);
crm_debug("Processing diff (%s): %d.%d.%d -> %d.%d.%d (%s)", op,
diff_del_admin_epoch,diff_del_epoch,diff_del_updates,
diff_add_admin_epoch,diff_add_epoch,diff_add_updates,
fsa_state2string(fsa_state));
log_cib_diff(LOG_DEBUG_2, diff, op);
/* Process anything that was added */
cib_top = get_xpath_object("//"F_CIB_UPDATE_RESULT"//"XML_TAG_DIFF_ADDED"//"XML_TAG_CIB, diff, LOG_ERR);
if(need_abort(cib_top)) {
goto bail; /* configuration changed */
}
/* Process anything that was removed */
cib_top = get_xpath_object("//"F_CIB_UPDATE_RESULT"//"XML_TAG_DIFF_REMOVED"//"XML_TAG_CIB, diff, LOG_ERR);
if(need_abort(cib_top)) {
goto bail; /* configuration changed */
}
/* Transient Attributes - Added/Updated */
xpathObj = xpath_search(diff,"//"F_CIB_UPDATE_RESULT"//"XML_TAG_DIFF_ADDED"//"XML_TAG_TRANSIENT_NODEATTRS"//"XML_CIB_TAG_NVPAIR);
if(xpathObj && xpathObj->nodesetval->nodeNr > 0) {
int lpc;
for(lpc = 0; lpc < xpathObj->nodesetval->nodeNr; lpc++) {
xmlNode *attr = getXpathResult(xpathObj, lpc);
const char *name = crm_element_value(attr, XML_NVPAIR_ATTR_NAME);
const char *value = NULL;
if(safe_str_eq(CRM_OP_PROBED, name)) {
value = crm_element_value(attr, XML_NVPAIR_ATTR_VALUE);
}
if(crm_is_true(value) == FALSE) {
abort_transition(INFINITY, tg_restart, "Transient attribute: update", attr);
crm_log_xml_debug_2(attr, "Abort");
goto bail;
}
}
} else if(xpathObj) {
xmlXPathFreeObject(xpathObj);
}
/* Transient Attributes - Removed */
xpathObj = xpath_search(diff,"//"F_CIB_UPDATE_RESULT"//"XML_TAG_DIFF_REMOVED"//"XML_TAG_TRANSIENT_NODEATTRS);
if(xpathObj && xpathObj->nodesetval->nodeNr > 0) {
xmlNode *aborted = getXpathResult(xpathObj, 0);
abort_transition(INFINITY, tg_restart, "Transient attribute: removal", aborted);
goto bail;
} else if(xpathObj) {
xmlXPathFreeObject(xpathObj);
}
/* Check for node state updates... possibly from a shutdown we requested */
xpathObj = xpath_search(diff, "//"F_CIB_UPDATE_RESULT"//"XML_TAG_DIFF_ADDED"//"XML_CIB_TAG_STATE);
if(xpathObj) {
int lpc = 0, max = xpathObj->nodesetval->nodeNr;
for(lpc = 0; lpc < max; lpc++) {
xmlNode *node = getXpathResult(xpathObj, lpc);
const char *event_node = crm_element_value(node, XML_ATTR_ID);
const char *ccm_state = crm_element_value(node, XML_CIB_ATTR_INCCM);
const char *ha_state = crm_element_value(node, XML_CIB_ATTR_HASTATE);
const char *shutdown_s = crm_element_value(node, XML_CIB_ATTR_SHUTDOWN);
const char *crmd_state = crm_element_value(node, XML_CIB_ATTR_CRMDSTATE);
if(safe_str_eq(ccm_state, XML_BOOLEAN_FALSE)
|| safe_str_eq(ha_state, DEADSTATUS)
|| safe_str_eq(crmd_state, CRMD_JOINSTATE_DOWN)) {
crm_action_t *shutdown = match_down_event(0, event_node, NULL);
if(shutdown != NULL) {
const char *task = crm_element_value(shutdown->xml, XML_LRM_ATTR_TASK);
if(safe_str_neq(task, CRM_OP_FENCE)) {
/* Wait for stonithd to tell us it is complete via tengine_stonith_callback() */
update_graph(transition_graph, shutdown);
trigger_graph();
}
} else {
crm_info("Stonith/shutdown of %s not matched", event_node);
abort_transition(INFINITY, tg_restart, "Node failure", node);
}
fail_incompletable_actions(transition_graph, event_node);
}
if(shutdown_s) {
int shutdown = crm_parse_int(shutdown_s, NULL);
if(shutdown > 0) {
crm_info("Aborting on "XML_CIB_ATTR_SHUTDOWN" attribute for %s", event_node);
abort_transition(INFINITY, tg_restart, "Shutdown request", node);
}
}
}
xmlXPathFreeObject(xpathObj);
}
/*
* Check for and fast-track the processing of LRM refreshes
* In large clusters this can result in _huge_ speedups
*
* Unfortunately we can only do so when there are no pending actions
* Otherwise we could miss updates we're waiting for and stall
*
*/
xpathObj = NULL;
if(transition_graph->pending == 0) {
xpathObj = xpath_search(diff, "//"F_CIB_UPDATE_RESULT"//"XML_TAG_DIFF_ADDED"//"XML_LRM_TAG_RESOURCE);
}
if(xpathObj) {
int updates = xpathObj->nodesetval->nodeNr;
if(updates > 1) {
/* Updates by, or in response to, TE actions will never contain updates
* for more than one resource at a time
*/
crm_info("Detected LRM refresh - %d resources updated: Skipping all resource events", updates);
abort_transition(INFINITY, tg_restart, "LRM Refresh", diff);
goto bail;
}
xmlXPathFreeObject(xpathObj);
}
/* Process operation updates */
xpathObj = xpath_search(diff, "//"F_CIB_UPDATE_RESULT"//"XML_TAG_DIFF_ADDED"//"XML_LRM_TAG_RSC_OP);
if(xpathObj) {
process_resource_updates(xpathObj);
xmlXPathFreeObject(xpathObj);
}
/* Detect deleted (as opposed to replaced or added) actions - eg. crm_resource -C */
xpathObj = xpath_search(diff, "//"XML_TAG_DIFF_REMOVED"//"XML_LRM_TAG_RSC_OP);
if(xpathObj) {
int lpc = 0, max = xpathObj->nodesetval->nodeNr;
for(lpc = 0; lpc < max; lpc++) {
int max = 0;
const char *op_id = NULL;
char *rsc_op_xpath = NULL;
xmlXPathObject *op_match = NULL;
xmlNode *match = getXpathResult(xpathObj, lpc);
CRM_CHECK(match != NULL, continue);
op_id = ID(match);
max = strlen(rsc_op_template) + strlen(op_id) + 1;
crm_malloc0(rsc_op_xpath, max);
snprintf(rsc_op_xpath, max, rsc_op_template, op_id);
op_match = xpath_search(diff, rsc_op_xpath);
if(op_match == NULL || op_match->nodesetval->nodeNr == 0) {
/* Prevent false positives by matching cancelations too */
const char *node = get_node_id(match);
crm_action_t *cancelled = get_cancel_action(op_id, node);
if(cancelled == NULL) {
crm_debug("No match for deleted action %s (%s on %s)", rsc_op_xpath, op_id, node);
abort_transition(INFINITY, tg_restart, "Resource op removal", match);
if(op_match) {
xmlXPathFreeObject(op_match);
}
crm_free(rsc_op_xpath);
goto bail;
} else {
crm_debug("Deleted lrm_rsc_op %s on %s was for graph event %d",
op_id, node, cancelled->id);
}
}
if(op_match) {
xmlXPathFreeObject(op_match);
}
crm_free(rsc_op_xpath);
}
}
bail:
if(xpathObj) {
xmlXPathFreeObject(xpathObj);
}
}
void
te_update_diff(const char *event, xmlNode * msg)
{
int rc = -1;
const char *op = NULL;
xmlNode *diff = NULL;
xmlXPathObject *xpathObj = NULL;
int diff_add_updates = 0;
int diff_add_epoch = 0;
int diff_add_admin_epoch = 0;
int diff_del_updates = 0;
int diff_del_epoch = 0;
int diff_del_admin_epoch = 0;
CRM_CHECK(msg != NULL, return);
crm_element_value_int(msg, F_CIB_RC, &rc);
if (transition_graph == NULL) {
crm_trace("No graph");
return;
} else if (rc < pcmk_ok) {
crm_trace("Filter rc=%d (%s)", rc, pcmk_strerror(rc));
return;
} else if (transition_graph->complete == TRUE
&& fsa_state != S_IDLE
&& fsa_state != S_TRANSITION_ENGINE && fsa_state != S_POLICY_ENGINE) {
crm_trace("Filter state=%s, complete=%d", fsa_state2string(fsa_state),
transition_graph->complete);
return;
}
op = crm_element_value(msg, F_CIB_OPERATION);
diff = get_message_xml(msg, F_CIB_UPDATE_RESULT);
cib_diff_version_details(diff,
&diff_add_admin_epoch, &diff_add_epoch, &diff_add_updates,
&diff_del_admin_epoch, &diff_del_epoch, &diff_del_updates);
crm_debug("Processing diff (%s): %d.%d.%d -> %d.%d.%d (%s)", op,
diff_del_admin_epoch, diff_del_epoch, diff_del_updates,
diff_add_admin_epoch, diff_add_epoch, diff_add_updates, fsa_state2string(fsa_state));
log_cib_diff(LOG_DEBUG_2, diff, op);
if (cib_config_changed(NULL, NULL, &diff)) {
abort_transition(INFINITY, tg_restart, "Non-status change", diff);
goto bail; /* configuration changed */
}
/* Tickets Attributes - Added/Updated */
xpathObj =
xpath_search(diff,
"//" F_CIB_UPDATE_RESULT "//" XML_TAG_DIFF_ADDED "//" XML_CIB_TAG_TICKETS);
if (xpathObj && xpathObj->nodesetval->nodeNr > 0) {
xmlNode *aborted = getXpathResult(xpathObj, 0);
abort_transition(INFINITY, tg_restart, "Ticket attribute: update", aborted);
goto bail;
} else if (xpathObj) {
xmlXPathFreeObject(xpathObj);
}
/* Tickets Attributes - Removed */
xpathObj =
xpath_search(diff,
"//" F_CIB_UPDATE_RESULT "//" XML_TAG_DIFF_REMOVED "//" XML_CIB_TAG_TICKETS);
if (xpathObj && xpathObj->nodesetval->nodeNr > 0) {
xmlNode *aborted = getXpathResult(xpathObj, 0);
abort_transition(INFINITY, tg_restart, "Ticket attribute: removal", aborted);
goto bail;
} else if (xpathObj) {
xmlXPathFreeObject(xpathObj);
}
/* Transient Attributes - Added/Updated */
xpathObj =
xpath_search(diff,
"//" F_CIB_UPDATE_RESULT "//" XML_TAG_DIFF_ADDED "//"
XML_TAG_TRANSIENT_NODEATTRS "//" XML_CIB_TAG_NVPAIR);
if (xpathObj && xpathObj->nodesetval->nodeNr > 0) {
int lpc;
for (lpc = 0; lpc < xpathObj->nodesetval->nodeNr; lpc++) {
xmlNode *attr = getXpathResult(xpathObj, lpc);
const char *name = crm_element_value(attr, XML_NVPAIR_ATTR_NAME);
const char *value = NULL;
if (safe_str_eq(CRM_OP_PROBED, name)) {
value = crm_element_value(attr, XML_NVPAIR_ATTR_VALUE);
}
if (crm_is_true(value) == FALSE) {
abort_transition(INFINITY, tg_restart, "Transient attribute: update", attr);
crm_log_xml_trace(attr, "Abort");
goto bail;
}
}
} else if (xpathObj) {
xmlXPathFreeObject(xpathObj);
}
/* Transient Attributes - Removed */
xpathObj =
xpath_search(diff,
"//" F_CIB_UPDATE_RESULT "//" XML_TAG_DIFF_REMOVED "//"
XML_TAG_TRANSIENT_NODEATTRS);
if (xpathObj && xpathObj->nodesetval->nodeNr > 0) {
xmlNode *aborted = getXpathResult(xpathObj, 0);
abort_transition(INFINITY, tg_restart, "Transient attribute: removal", aborted);
goto bail;
} else if (xpathObj) {
xmlXPathFreeObject(xpathObj);
}
/*
* Check for and fast-track the processing of LRM refreshes
* In large clusters this can result in _huge_ speedups
*
* Unfortunately we can only do so when there are no pending actions
* Otherwise we could miss updates we're waiting for and stall
*
*/
xpathObj = NULL;
if (transition_graph->pending == 0) {
xpathObj =
xpath_search(diff,
"//" F_CIB_UPDATE_RESULT "//" XML_TAG_DIFF_ADDED "//"
XML_LRM_TAG_RESOURCE);
}
if (xpathObj) {
int updates = xpathObj->nodesetval->nodeNr;
if (updates > 1) {
/* Updates by, or in response to, TE actions will never contain updates
* for more than one resource at a time
*/
crm_debug("Detected LRM refresh - %d resources updated: Skipping all resource events",
updates);
crm_log_xml_trace(diff, "lrm-refresh");
abort_transition(INFINITY, tg_restart, "LRM Refresh", NULL);
goto bail;
}
xmlXPathFreeObject(xpathObj);
}
/* Process operation updates */
xpathObj =
xpath_search(diff,
"//" F_CIB_UPDATE_RESULT "//" XML_TAG_DIFF_ADDED "//" XML_LRM_TAG_RSC_OP);
if (xpathObj) {
process_resource_updates(xpathObj);
xmlXPathFreeObject(xpathObj);
}
/* Detect deleted (as opposed to replaced or added) actions - eg. crm_resource -C */
xpathObj = xpath_search(diff, "//" XML_TAG_DIFF_REMOVED "//" XML_LRM_TAG_RSC_OP);
if (xpathObj) {
int lpc = 0, max = xpathObj->nodesetval->nodeNr;
for (lpc = 0; lpc < max; lpc++) {
int max = 0;
const char *op_id = NULL;
char *rsc_op_xpath = NULL;
xmlXPathObject *op_match = NULL;
xmlNode *match = getXpathResult(xpathObj, lpc);
CRM_CHECK(match != NULL, continue);
op_id = ID(match);
max = strlen(rsc_op_template) + strlen(op_id) + 1;
rsc_op_xpath = calloc(1, max);
snprintf(rsc_op_xpath, max, rsc_op_template, op_id);
op_match = xpath_search(diff, rsc_op_xpath);
if (op_match == NULL || op_match->nodesetval->nodeNr == 0) {
/* Prevent false positives by matching cancelations too */
const char *node = get_node_id(match);
crm_action_t *cancelled = get_cancel_action(op_id, node);
if (cancelled == NULL) {
crm_debug("No match for deleted action %s (%s on %s)", rsc_op_xpath, op_id,
node);
abort_transition(INFINITY, tg_restart, "Resource op removal", match);
if (op_match) {
xmlXPathFreeObject(op_match);
}
free(rsc_op_xpath);
goto bail;
} else {
crm_debug("Deleted lrm_rsc_op %s on %s was for graph event %d",
op_id, node, cancelled->id);
}
}
if (op_match) {
xmlXPathFreeObject(op_match);
}
free(rsc_op_xpath);
}
}
bail:
if (xpathObj) {
xmlXPathFreeObject(xpathObj);
}
}
static void
tengine_stonith_notify(stonith_t * st, stonith_event_t * st_event)
{
if(te_client_id == NULL) {
te_client_id = crm_strdup_printf("%s.%lu", crm_system_name,
(unsigned long) getpid());
}
if (st_event == NULL) {
crm_err("Notify data not found");
return;
}
crmd_alert_fencing_op(st_event);
if (st_event->result == pcmk_ok && safe_str_eq("on", st_event->action)) {
crm_notice("%s was successfully unfenced by %s (at the request of %s)",
st_event->target, st_event->executioner ? st_event->executioner : "<anyone>", st_event->origin);
/* TODO: Hook up st_event->device */
return;
} else if (safe_str_eq("on", st_event->action)) {
crm_err("Unfencing of %s by %s failed: %s (%d)",
st_event->target, st_event->executioner ? st_event->executioner : "<anyone>",
pcmk_strerror(st_event->result), st_event->result);
return;
} else if (st_event->result == pcmk_ok && crm_str_eq(st_event->target, fsa_our_uname, TRUE)) {
crm_crit("We were allegedly just fenced by %s for %s!",
st_event->executioner ? st_event->executioner : "<anyone>", st_event->origin); /* Dumps blackbox if enabled */
qb_log_fini(); /* Try to get the above log message to disk - somehow */
/* Get out ASAP and do not come back up.
*
* Triggering a reboot is also not the worst idea either since
* the rest of the cluster thinks we're safely down
*/
#ifdef RB_HALT_SYSTEM
reboot(RB_HALT_SYSTEM);
#endif
/*
* If reboot() fails or is not supported, coming back up will
* probably lead to a situation where the other nodes set our
* status to 'lost' because of the fencing callback and will
* discard subsequent election votes with:
*
* Election 87 (current: 5171, owner: 103): Processed vote from east-03 (Peer is not part of our cluster)
*
* So just stay dead, something is seriously messed up anyway.
*
*/
exit(CRM_EX_FATAL); // None of our wrappers since we already called qb_log_fini()
return;
}
/* Update the count of stonith failures for this target, in case we become
* DC later. The current DC has already updated its fail count in
* tengine_stonith_callback().
*/
if (!AM_I_DC && safe_str_eq(st_event->operation, T_STONITH_NOTIFY_FENCE)) {
if (st_event->result == pcmk_ok) {
st_fail_count_reset(st_event->target);
} else {
st_fail_count_increment(st_event->target);
}
}
crm_notice("Peer %s was%s terminated (%s) by %s on behalf of %s: %s "
CRM_XS " initiator=%s ref=%s",
st_event->target, st_event->result == pcmk_ok ? "" : " not",
st_event->action,
st_event->executioner ? st_event->executioner : "<anyone>",
(st_event->client_origin? st_event->client_origin : "<unknown>"),
pcmk_strerror(st_event->result),
st_event->origin, st_event->id);
if (st_event->result == pcmk_ok) {
crm_node_t *peer = crm_find_peer_full(0, st_event->target, CRM_GET_PEER_ANY);
const char *uuid = NULL;
gboolean we_are_executioner = safe_str_eq(st_event->executioner, fsa_our_uname);
if (peer == NULL) {
return;
}
uuid = crm_peer_uuid(peer);
crm_trace("target=%s dc=%s", st_event->target, fsa_our_dc);
if(AM_I_DC) {
/* The DC always sends updates */
send_stonith_update(NULL, st_event->target, uuid);
/* @TODO Ideally, at this point, we'd check whether the fenced node
* hosted any guest nodes, and call remote_node_down() for them.
* Unfortunately, the controller doesn't have a simple, reliable way
* to map hosts to guests. It might be possible to track this in the
* peer cache via crm_remote_peer_cache_refresh(). For now, we rely
* on the PE creating fence pseudo-events for the guests.
*/
if (st_event->client_origin && safe_str_neq(st_event->client_origin, te_client_id)) {
/* Abort the current transition graph if it wasn't us
* that invoked stonith to fence someone
*/
crm_info("External fencing operation from %s fenced %s", st_event->client_origin, st_event->target);
abort_transition(INFINITY, tg_restart, "External Fencing Operation", NULL);
}
/* Assume it was our leader if we don't currently have one */
} else if (((fsa_our_dc == NULL) || safe_str_eq(fsa_our_dc, st_event->target))
&& !is_set(peer->flags, crm_remote_node)) {
crm_notice("Target %s our leader %s (recorded: %s)",
fsa_our_dc ? "was" : "may have been", st_event->target,
fsa_our_dc ? fsa_our_dc : "<unset>");
/* Given the CIB resyncing that occurs around elections,
* have one node update the CIB now and, if the new DC is different,
* have them do so too after the election
*/
if (we_are_executioner) {
send_stonith_update(NULL, st_event->target, uuid);
}
add_stonith_cleanup(st_event->target);
}
/* If the target is a remote node, and we host its connection,
* immediately fail all monitors so it can be recovered quickly.
* The connection won't necessarily drop when a remote node is fenced,
* so the failure might not otherwise be detected until the next poke.
*/
if (is_set(peer->flags, crm_remote_node)) {
remote_ra_fail(st_event->target);
}
crmd_peer_down(peer, TRUE);
}
}
