/* A_PE_START, A_PE_STOP, O_PE_RESTART */
void
do_pe_control(long long action,
enum crmd_fsa_cause cause,
enum crmd_fsa_state cur_state,
enum crmd_fsa_input current_input, fsa_data_t * msg_data)
{
if (action & A_PE_STOP) {
clear_bit(fsa_input_register, R_PE_REQUIRED);
pe_subsystem_free();
clear_bit(fsa_input_register, R_PE_CONNECTED);
}
if ((action & A_PE_START) && (is_set(fsa_input_register, R_PE_CONNECTED) == FALSE)) {
if (cur_state != S_STOPPING) {
set_bit(fsa_input_register, R_PE_REQUIRED);
if (pe_subsystem_new()) {
set_bit(fsa_input_register, R_PE_CONNECTED);
} else {
crm_warn("Could not connect to scheduler");
register_fsa_error(C_FSA_INTERNAL, I_FAIL, NULL);
}
} else {
crm_info("Ignoring request to connect to scheduler while shutting down");
}
}
}
static void
revision_check_callback(xmlNode *msg, int call_id, int rc,
xmlNode *output, void *user_data)
{
int cmp = -1;
const char *revision = NULL;
xmlNode *generation = NULL;
if(rc != cib_ok) {
fsa_data_t *msg_data = NULL;
register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL);
return;
}
generation = output;
CRM_CHECK(safe_str_eq(crm_element_name(generation), XML_TAG_CIB), crm_log_xml_err(output, __FUNCTION__); return);
crm_debug_3("Checking our feature revision is allowed: %s", CIB_FEATURE_SET);
revision = crm_element_value(generation, XML_ATTR_CRM_VERSION);
cmp = compare_version(revision, CRM_FEATURE_SET);
if(cmp > 0) {
crm_err("This build (%s) does not support the current"
" resource configuration", VERSION);
crm_err("We can support up to CRM feature set %s (current=%s)",
revision, CRM_FEATURE_SET);
crm_err("Shutting down the CRM");
/* go into a stall state */
register_fsa_error_adv(
C_FSA_INTERNAL, I_SHUTDOWN, NULL, NULL, __FUNCTION__);
return;
}
}
static void
node_list_update_callback(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data)
{
fsa_data_t *msg_data = NULL;
if(call_id < pcmk_ok) {
crm_err("Node list update failed: %s (%d)", pcmk_strerror(call_id), call_id);
crm_log_xml_debug(msg, "update:failed");
register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL);
} else if(rc < pcmk_ok) {
crm_err("Node update %d failed: %s (%d)", call_id, pcmk_strerror(rc), rc);
crm_log_xml_debug(msg, "update:failed");
register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL);
}
}
static void
feature_update_callback(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data)
{
if (rc != pcmk_ok) {
fsa_data_t *msg_data = NULL;
register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL);
}
}
static void
revision_check_callback(const HA_Message *msg, int call_id, int rc,
crm_data_t *output, void *user_data)
{
int cmp = -1;
const char *revision = NULL;
crm_data_t *generation = NULL;
#if CRM_DEPRECATED_SINCE_2_0_4
if(safe_str_eq(crm_element_name(output), XML_TAG_CIB)) {
generation = output;
} else {
generation = find_xml_node(output, XML_TAG_CIB, TRUE);
}
#else
generation = output;
CRM_DEV_ASSERT(safe_str_eq(crm_element_name(generation), XML_TAG_CIB));
#endif
if(rc != cib_ok) {
fsa_data_t *msg_data = NULL;
register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL);
return;
}
crm_debug_3("Checking our feature revision is allowed: %s",
CIB_FEATURE_SET);
revision = crm_element_value(generation, XML_ATTR_CIB_REVISION);
cmp = compare_version(revision, CIB_FEATURE_SET);
if(cmp > 0) {
crm_err("This build (%s) does not support the current"
" resource configuration", VERSION);
crm_err("We can support up to CIB feature set %s (current=%s)",
CIB_FEATURE_SET, revision);
crm_err("Shutting down the CRM");
/* go into a stall state */
register_fsa_error_adv(
C_FSA_INTERNAL, I_SHUTDOWN, NULL, NULL, __FUNCTION__);
return;
}
revision = crm_element_value(generation, XML_ATTR_CRM_VERSION);
cmp = compare_version(revision, CRM_FEATURE_SET);
if(cmp > 0) {
crm_err("This build (%s) does not support the current"
" resource configuration", VERSION);
crm_err("We can support up to CRM feature set %s (current=%s)",
revision, CRM_FEATURE_SET);
crm_err("Shutting down the CRM");
/* go into a stall state */
register_fsa_error_adv(
C_FSA_INTERNAL, I_SHUTDOWN, NULL, NULL, __FUNCTION__);
return;
}
}
static void
feature_update_callback(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data)
{
if (rc != pcmk_ok) {
fsa_data_t *msg_data = NULL;
crm_notice("Update failed: %s (%d)", pcmk_strerror(rc), rc);
register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL);
}
}
static void
crmd_node_update_complete(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data)
{
fsa_data_t *msg_data = NULL;
last_peer_update = 0;
if (rc == pcmk_ok) {
crm_trace("Node update %d complete", call_id);
} else if(call_id < pcmk_ok) {
crm_err("Node update failed: %s (%d)", pcmk_strerror(call_id), call_id);
crm_log_xml_debug(msg, "failed");
register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL);
} else {
crm_err("Node update %d failed: %s (%d)", call_id, pcmk_strerror(rc), rc);
crm_log_xml_debug(msg, "failed");
register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL);
}
}
static void
join_update_complete_callback(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data)
{
fsa_data_t *msg_data = NULL;
if (rc == pcmk_ok) {
crm_debug("Join update %d complete", call_id);
check_join_state(fsa_state, __FUNCTION__);
} else {
crm_err("Join update %d failed", call_id);
crm_log_xml_debug(msg, "failed");
register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL);
}
}
/* 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);
}
}
}
/* A_CIB_STOP, A_CIB_START, A_CIB_RESTART, */
void
do_cib_control(long long action,
enum crmd_fsa_cause cause,
enum crmd_fsa_state cur_state,
enum crmd_fsa_input current_input,
fsa_data_t *msg_data)
{
struct crm_subsystem_s *this_subsys = cib_subsystem;
long long stop_actions = A_CIB_STOP;
long long start_actions = A_CIB_START;
if(action & stop_actions) {
if (fsa_cib_conn->state != cib_disconnected && last_resource_update != 0) {
crm_info("Waiting for resource update %d to complete", last_resource_update);
crmd_fsa_stall(NULL);
return;
}
crm_info("Disconnecting CIB");
clear_bit_inplace(fsa_input_register, R_CIB_CONNECTED);
CRM_ASSERT(fsa_cib_conn != NULL);
fsa_cib_conn->cmds->del_notify_callback(
fsa_cib_conn, T_CIB_DIFF_NOTIFY, do_cib_updated);
if(fsa_cib_conn->state != cib_disconnected) {
fsa_cib_conn->cmds->set_slave(
fsa_cib_conn, cib_scope_local);
fsa_cib_conn->cmds->signoff(fsa_cib_conn);
}
}
if(action & start_actions) {
int rc = cib_ok;
CRM_ASSERT(fsa_cib_conn != NULL);
if(cur_state == S_STOPPING) {
crm_err("Ignoring request to start %s after shutdown",
this_subsys->name);
return;
}
rc = fsa_cib_conn->cmds->signon(
fsa_cib_conn, CRM_SYSTEM_CRMD, cib_command);
if(rc != cib_ok) {
/* a short wait that usually avoids stalling the FSA */
sleep(1);
rc = fsa_cib_conn->cmds->signon(
fsa_cib_conn, CRM_SYSTEM_CRMD, cib_command);
}
if(rc != cib_ok){
crm_info("Could not connect to the CIB service: %s", cib_error2string(rc));
} else if(cib_ok != fsa_cib_conn->cmds->set_connection_dnotify(
fsa_cib_conn, crmd_cib_connection_destroy)) {
crm_err("Could not set dnotify callback");
} else if(cib_ok != fsa_cib_conn->cmds->add_notify_callback(
fsa_cib_conn, T_CIB_REPLACE_NOTIFY,
do_cib_replaced)) {
crm_err("Could not set CIB notification callback");
} else if(cib_ok != fsa_cib_conn->cmds->add_notify_callback(
fsa_cib_conn, T_CIB_DIFF_NOTIFY, do_cib_updated)) {
crm_err("Could not set CIB notification callback");
} else {
set_bit_inplace(
fsa_input_register, R_CIB_CONNECTED);
}
if(is_set(fsa_input_register, R_CIB_CONNECTED) == FALSE) {
cib_retries++;
crm_warn("Couldn't complete CIB registration %d"
" times... pause and retry",
cib_retries);
if(cib_retries < 30) {
crm_timer_start(wait_timer);
crmd_fsa_stall(NULL);
} else {
crm_err("Could not complete CIB"
" registration %d times..."
" hard error", cib_retries);
register_fsa_error(
C_FSA_INTERNAL, I_ERROR, NULL);
}
} else {
int call_id = 0;
crm_info("CIB connection established");
call_id = fsa_cib_conn->cmds->query(
fsa_cib_conn, NULL, NULL, cib_scope_local);
add_cib_op_callback(fsa_cib_conn, call_id, FALSE, NULL,
revision_check_callback);
cib_retries = 0;
}
}
}
/* aka. this is notification that we have (or have not) been accepted */
void
do_cl_join_finalize_respond(long long action,
enum crmd_fsa_cause cause,
enum crmd_fsa_state cur_state,
enum crmd_fsa_input current_input, fsa_data_t * msg_data)
{
xmlNode *tmp1 = NULL;
gboolean was_nack = TRUE;
static gboolean first_join = TRUE;
ha_msg_input_t *input = fsa_typed_data(fsa_dt_ha_msg);
const char *start_state = daemon_option("node_start_state");
int join_id = -1;
const char *op = crm_element_value(input->msg, F_CRM_TASK);
const char *ack_nack = crm_element_value(input->msg, CRM_OP_JOIN_ACKNAK);
const char *welcome_from = crm_element_value(input->msg, F_CRM_HOST_FROM);
if (safe_str_neq(op, CRM_OP_JOIN_ACKNAK)) {
crm_trace("Ignoring op=%s message", op);
return;
}
/* calculate if it was an ack or a nack */
if (crm_is_true(ack_nack)) {
was_nack = FALSE;
}
crm_element_value_int(input->msg, F_CRM_JOIN_ID, &join_id);
if (was_nack) {
crm_err("Shutting down because cluster join with leader %s failed "
CRM_XS" join-%d NACK'd", welcome_from, join_id);
register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL);
return;
}
if (AM_I_DC == FALSE && safe_str_eq(welcome_from, fsa_our_uname)) {
crm_warn("Discarding our own welcome - we're no longer the DC");
return;
}
if (update_dc(input->msg) == FALSE) {
crm_warn("Discarding %s from node %s (expected from %s)",
op, welcome_from, fsa_our_dc);
return;
}
update_dc_expected(input->msg);
/* send our status section to the DC */
tmp1 = do_lrm_query(TRUE, fsa_our_uname);
if (tmp1 != NULL) {
xmlNode *reply = create_request(CRM_OP_JOIN_CONFIRM, tmp1, fsa_our_dc,
CRM_SYSTEM_DC, CRM_SYSTEM_CRMD, NULL);
crm_xml_add_int(reply, F_CRM_JOIN_ID, join_id);
crm_debug("Confirming join-%d: sending local operation history to %s",
join_id, fsa_our_dc);
/*
* If this is the node's first join since the crmd started on it, clear
* any previous transient node attributes, to handle the case where
* the node restarted so quickly that the cluster layer didn't notice.
*
* Do not remove the resources though, they'll be cleaned up in
* do_dc_join_ack(). Removing them here creates a race condition if the
* crmd is being recovered. Instead of a list of active resources from
* the lrmd, we may end up with a blank status section. If we are _NOT_
* lucky, we will probe for the "wrong" instance of anonymous clones and
* end up with multiple active instances on the machine.
*/
if (first_join && is_not_set(fsa_input_register, R_SHUTDOWN)) {
first_join = FALSE;
erase_status_tag(fsa_our_uname, XML_TAG_TRANSIENT_NODEATTRS, 0);
update_attrd(fsa_our_uname, "terminate", NULL, NULL, FALSE);
update_attrd(fsa_our_uname, XML_CIB_ATTR_SHUTDOWN, "0", NULL, FALSE);
if (start_state) {
set_join_state(start_state);
}
}
send_cluster_message(crm_get_peer(0, fsa_our_dc), crm_msg_crmd, reply, TRUE);
free_xml(reply);
if (AM_I_DC == FALSE) {
register_fsa_input_adv(cause, I_NOT_DC, NULL, A_NOTHING, TRUE, __FUNCTION__);
update_attrd(NULL, NULL, NULL, NULL, FALSE);
}
free_xml(tmp1);
} else {
crm_err("Could not confirm join-%d with %s: Local operation history failed",
join_id, fsa_our_dc);
register_fsa_error(C_FSA_INTERNAL, I_FAIL, NULL);
}
}
/* A_CIB_INVOKE, A_CIB_BUMPGEN, A_UPDATE_NODESTATUS */
void
do_cib_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)
{
HA_Message *answer = NULL;
ha_msg_input_t *cib_msg = fsa_typed_data(fsa_dt_ha_msg);
const char *sys_from = cl_get_string(cib_msg->msg, F_CRM_SYS_FROM);
if(fsa_cib_conn->state == cib_disconnected) {
if(cur_state != S_STOPPING) {
crm_err("CIB is disconnected");
crm_log_message_adv(LOG_WARNING, "CIB Input", cib_msg->msg);
return;
}
crm_info("CIB is disconnected");
crm_log_message_adv(LOG_DEBUG, "CIB Input", cib_msg->msg);
return;
}
if(action & A_CIB_INVOKE) {
if(safe_str_eq(sys_from, CRM_SYSTEM_CRMD)) {
action = A_CIB_INVOKE_LOCAL;
} else if(safe_str_eq(sys_from, CRM_SYSTEM_DC)) {
action = A_CIB_INVOKE_LOCAL;
}
}
if(action & A_CIB_INVOKE || action & A_CIB_INVOKE_LOCAL) {
int call_options = 0;
enum cib_errors rc = cib_ok;
crm_data_t *cib_frag = NULL;
const char *section = NULL;
const char *op = cl_get_string(cib_msg->msg, F_CRM_TASK);
section = cl_get_string(cib_msg->msg, F_CIB_SECTION);
ha_msg_value_int(cib_msg->msg, F_CIB_CALLOPTS, &call_options);
crm_log_message(LOG_MSG, cib_msg->msg);
crm_log_xml_debug_3(cib_msg->xml, "[CIB update]");
if(op == NULL) {
crm_err("Invalid CIB Message");
register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL);
return;
}
cib_frag = NULL;
rc = fsa_cib_conn->cmds->variant_op(
fsa_cib_conn, op, NULL, section,
cib_msg->xml, &cib_frag, call_options);
if(rc < cib_ok || (action & A_CIB_INVOKE)) {
answer = create_reply(cib_msg->msg, cib_frag);
ha_msg_add(answer,XML_ATTR_RESULT,cib_error2string(rc));
}
if(action & A_CIB_INVOKE) {
if(relay_message(answer, TRUE) == FALSE) {
crm_err("Confused what to do with cib result");
crm_log_message(LOG_ERR, answer);
crm_msg_del(answer);
register_fsa_input(C_FSA_INTERNAL, I_ERROR, NULL);
return;
}
} else if(rc < cib_ok) {
ha_msg_input_t *input = NULL;
crm_err("Internal CRM/CIB command from %s() failed: %s",
msg_data->origin, cib_error2string(rc));
crm_log_message_adv(LOG_WARNING, "CIB Input", cib_msg->msg);
crm_log_message_adv(LOG_WARNING, "CIB Reply", answer);
input = new_ha_msg_input(answer);
register_fsa_input(C_FSA_INTERNAL, I_ERROR, input);
crm_msg_del(answer);
delete_ha_msg_input(input);
}
} else {
crm_err("Unexpected action %s in %s",
fsa_action2string(action), __FUNCTION__);
}
}
/* aka. this is notification that we have (or have not) been accepted */
void
do_cl_join_finalize_respond(long long action,
enum crmd_fsa_cause cause,
enum crmd_fsa_state cur_state,
enum crmd_fsa_input current_input, fsa_data_t * msg_data)
{
xmlNode *tmp1 = NULL;
gboolean was_nack = TRUE;
static gboolean first_join = TRUE;
ha_msg_input_t *input = fsa_typed_data(fsa_dt_ha_msg);
int join_id = -1;
const char *op = crm_element_value(input->msg, F_CRM_TASK);
const char *ack_nack = crm_element_value(input->msg, CRM_OP_JOIN_ACKNAK);
const char *welcome_from = crm_element_value(input->msg, F_CRM_HOST_FROM);
if (safe_str_neq(op, CRM_OP_JOIN_ACKNAK)) {
crm_trace("Ignoring op=%s message", op);
return;
}
/* calculate if it was an ack or a nack */
if (crm_is_true(ack_nack)) {
was_nack = FALSE;
}
crm_element_value_int(input->msg, F_CRM_JOIN_ID, &join_id);
if (was_nack) {
crm_err("Join (join-%d) with leader %s failed (NACK'd): Shutting down",
join_id, welcome_from);
register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL);
return;
}
if (AM_I_DC == FALSE && safe_str_eq(welcome_from, fsa_our_uname)) {
crm_warn("Discarding our own welcome - we're no longer the DC");
return;
}
if (update_dc(input->msg) == FALSE) {
crm_warn("Discarding %s from %s (expected %s)", op, welcome_from, fsa_our_dc);
return;
}
/* send our status section to the DC */
crm_debug("Confirming join join-%d: %s", join_id, crm_element_value(input->msg, F_CRM_TASK));
tmp1 = do_lrm_query(TRUE, fsa_our_uname);
if (tmp1 != NULL) {
xmlNode *reply = create_request(CRM_OP_JOIN_CONFIRM, tmp1, fsa_our_dc,
CRM_SYSTEM_DC, CRM_SYSTEM_CRMD, NULL);
crm_xml_add_int(reply, F_CRM_JOIN_ID, join_id);
crm_debug("join-%d: Join complete."
" Sending local LRM status to %s", join_id, fsa_our_dc);
if (first_join) {
first_join = FALSE;
/*
* Clear any previous transient node attribute and lrm operations
*
* Corosync has a nasty habit of not being able to tell if a
* node is returning or didn't leave in the first place.
* This confuses Pacemaker because it never gets a "node up"
* event which is normally used to clean up the status section.
*
* Do not remove the resources though, they'll be cleaned up in
* do_dc_join_ack(). Removing them here creates a race
* condition if the crmd is being recovered.
* Instead of a list of active resources from the lrmd
* we may end up with a blank status section.
* If we are _NOT_ lucky, we will probe for the "wrong" instance
* of anonymous clones and end up with multiple active
* instances on the machine.
*/
erase_status_tag(fsa_our_uname, XML_TAG_TRANSIENT_NODEATTRS, 0);
/* Just in case attrd was still around too */
if (is_not_set(fsa_input_register, R_SHUTDOWN)) {
update_attrd(fsa_our_uname, "terminate", NULL, NULL, FALSE);
update_attrd(fsa_our_uname, XML_CIB_ATTR_SHUTDOWN, "0", NULL, FALSE);
}
}
send_cluster_message(crm_get_peer(0, fsa_our_dc), crm_msg_crmd, reply, TRUE);
free_xml(reply);
if (AM_I_DC == FALSE) {
register_fsa_input_adv(cause, I_NOT_DC, NULL, A_NOTHING, TRUE, __FUNCTION__);
update_attrd(NULL, NULL, NULL, NULL, FALSE);
}
free_xml(tmp1);
} else {
crm_err("Could not send our LRM state to the DC");
register_fsa_error(C_FSA_INTERNAL, I_FAIL, NULL);
}
}
