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;
}
}
void
finalize_sync_callback(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data)
{
CRM_LOG_ASSERT(-EPERM != rc);
clear_bit(fsa_input_register, R_CIB_ASKED);
if (rc != pcmk_ok) {
do_crm_log((rc == -pcmk_err_old_data ? LOG_WARNING : LOG_ERR),
"Sync from %s failed: %s", (char *)user_data, pcmk_strerror(rc));
/* restart the whole join process */
register_fsa_error_adv(C_FSA_INTERNAL, I_ELECTION_DC, NULL, NULL, __FUNCTION__);
} else if (AM_I_DC && fsa_state == S_FINALIZE_JOIN) {
set_bit(fsa_input_register, R_HAVE_CIB);
clear_bit(fsa_input_register, R_CIB_ASKED);
/* make sure dc_uuid is re-set to us */
if (check_join_state(fsa_state, __FUNCTION__) == FALSE) {
crm_debug("Notifying %d clients of join-%d results",
crmd_join_phase_count(crm_join_integrated), current_join_id);
g_hash_table_foreach(crm_peer_cache, finalize_join_for, NULL);
}
} else {
crm_debug("No longer the DC in S_FINALIZE_JOIN: %s/%s",
AM_I_DC ? "DC" : "CRMd", fsa_state2string(fsa_state));
}
}
/*
welcomed_nodesハッシュテーブルのサイズが０(すべての認識済みのノードからCRM_OP_JOIN_REQUESTを受信した)の場合
*/
void
do_dc_join_finalize(long long action,
enum crmd_fsa_cause cause,
enum crmd_fsa_state cur_state,
enum crmd_fsa_input current_input,
fsa_data_t *msg_data)
{
char *sync_from = NULL;
enum cib_errors rc = cib_ok;
/* This we can do straight away and avoid clients timing us out
* while we compute the latest CIB
*/
crm_debug("Finializing join-%d for %d clients",
current_join_id, g_hash_table_size(integrated_nodes));
if(g_hash_table_size(integrated_nodes) == 0) {
/* integrated_nodesハッシュテーブルにまだ、welcom_nodesハッシュテーブルからデータが１つも移送されていない場合 */
/* は、I_ELECTION_DCへ戻る */
/* 認識済みのメンバーから、どのノードからもCRM_OP_JOIN_REQUESTを受信し終わっていない状態ということになる */
/* If we don't even have ourself, start again */
register_fsa_error_adv(
C_FSA_INTERNAL, I_ELECTION_DC, NULL, NULL, __FUNCTION__);
return;
}
clear_bit_inplace(fsa_input_register, R_HAVE_CIB);
if(max_generation_from == NULL
|| safe_str_eq(max_generation_from, fsa_our_uname)){
set_bit_inplace(fsa_input_register, R_HAVE_CIB);
}
if(is_set(fsa_input_register, R_IN_TRANSITION)) {
crm_warn("join-%d: We are still in a transition."
" Delaying until the TE completes.", current_join_id);
crmd_fsa_stall(NULL);
return;
}
if(is_set(fsa_input_register, R_HAVE_CIB) == FALSE) {
/* ask for the agreed best CIB */
sync_from = crm_strdup(max_generation_from);
crm_log_xml_debug(max_generation_xml, "Requesting version");
set_bit_inplace(fsa_input_register, R_CIB_ASKED);
} else {
/* Send _our_ CIB out to everyone */
sync_from = crm_strdup(fsa_our_uname);
}
crm_info("join-%d: Syncing the CIB from %s to the rest of the cluster",
current_join_id, sync_from);
/* CIBにsync_from処理を実行する */
rc = fsa_cib_conn->cmds->sync_from(
fsa_cib_conn, sync_from, NULL,cib_quorum_override);
/* sync_from処理のコールバックをセットする */
/* コールバック内からCRM_OP_JOIN_ACKNAKメッセージを送信する */
fsa_cib_conn->cmds->register_callback(
fsa_cib_conn, rc, 60, FALSE, sync_from,
"finalize_sync_callback", finalize_sync_callback);
}
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
save_cib_contents(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data)
{
char *id = user_data;
register_fsa_error_adv(C_FSA_INTERNAL, I_ERROR, NULL, NULL, __FUNCTION__);
CRM_CHECK(id != NULL, return);
if (rc == cib_ok) {
int len = 15;
char *filename = NULL;
len += strlen(id);
len += strlen(PE_STATE_DIR);
crm_malloc0(filename, len);
CRM_CHECK(filename != NULL, return);
sprintf(filename, PE_STATE_DIR "/pe-core-%s.bz2", id);
if (write_xml_file(output, filename, TRUE) < 0) {
crm_err("Could not save CIB contents after PE crash to %s", filename);
} else {
crm_notice("Saved CIB contents after PE crash to %s", filename);
}
crm_free(filename);
}
gboolean
crm_timer_popped(gpointer data)
{
fsa_timer_t *timer = (fsa_timer_t *) data;
if (timer == wait_timer
|| timer == recheck_timer
|| timer == transition_timer || timer == finalization_timer || timer == election_trigger) {
crm_info("%s (%s) just popped (%dms)",
get_timer_desc(timer), fsa_input2string(timer->fsa_input), timer->period_ms);
timer->counter++;
} else {
crm_err("%s (%s) just popped in state %s! (%dms)",
get_timer_desc(timer), fsa_input2string(timer->fsa_input),
fsa_state2string(fsa_state), timer->period_ms);
}
if (timer == election_trigger && election_trigger->counter > 5) {
crm_notice("We appear to be in an election loop, something may be wrong");
crm_write_blackbox(0, NULL);
election_trigger->counter = 0;
}
if (timer->repeat == FALSE) {
crm_timer_stop(timer); /* make it _not_ go off again */
}
if (timer->fsa_input == I_INTEGRATED) {
crm_info("Welcomed: %d, Integrated: %d",
crmd_join_phase_count(crm_join_welcomed),
crmd_join_phase_count(crm_join_integrated));
if (crmd_join_phase_count(crm_join_welcomed) == 0) {
/* If we don't even have ourself, start again */
register_fsa_error_adv(C_FSA_INTERNAL, I_ELECTION, NULL, NULL, __FUNCTION__);
} else {
register_fsa_input_before(C_TIMER_POPPED, timer->fsa_input, NULL);
}
} else if (timer == recheck_timer && fsa_state != S_IDLE) {
crm_debug("Discarding %s event in state: %s",
fsa_input2string(timer->fsa_input), fsa_state2string(fsa_state));
} else if (timer == finalization_timer && fsa_state != S_FINALIZE_JOIN) {
crm_debug("Discarding %s event in state: %s",
fsa_input2string(timer->fsa_input), fsa_state2string(fsa_state));
} else if (timer->fsa_input != I_NULL) {
register_fsa_input(C_TIMER_POPPED, timer->fsa_input, NULL);
}
crm_trace("Triggering FSA: %s", __FUNCTION__);
mainloop_set_trigger(fsa_source);
return TRUE;
}
/* A_DC_JOIN_FINALIZE */
void
do_dc_join_finalize(long long action,
enum crmd_fsa_cause cause,
enum crmd_fsa_state cur_state,
enum crmd_fsa_input current_input, fsa_data_t * msg_data)
{
char *sync_from = NULL;
int rc = pcmk_ok;
/* This we can do straight away and avoid clients timing us out
* while we compute the latest CIB
*/
crm_debug("Finializing join-%d for %d clients",
current_join_id, g_hash_table_size(integrated_nodes));
if (g_hash_table_size(integrated_nodes) == 0) {
/* If we don't even have ourself, start again */
register_fsa_error_adv(C_FSA_INTERNAL, I_ELECTION_DC, NULL, NULL, __FUNCTION__);
return;
}
clear_bit_inplace(fsa_input_register, R_HAVE_CIB);
if (max_generation_from == NULL || safe_str_eq(max_generation_from, fsa_our_uname)) {
set_bit_inplace(fsa_input_register, R_HAVE_CIB);
}
if (is_set(fsa_input_register, R_IN_TRANSITION)) {
crm_warn("join-%d: We are still in a transition."
" Delaying until the TE completes.", current_join_id);
crmd_fsa_stall(NULL);
return;
}
if (is_set(fsa_input_register, R_HAVE_CIB) == FALSE) {
/* ask for the agreed best CIB */
sync_from = strdup(max_generation_from);
crm_log_xml_debug(max_generation_xml, "Requesting version");
set_bit_inplace(fsa_input_register, R_CIB_ASKED);
} else {
/* Send _our_ CIB out to everyone */
sync_from = strdup(fsa_our_uname);
}
crm_info("join-%d: Syncing the CIB from %s to the rest of the cluster",
current_join_id, sync_from);
rc = fsa_cib_conn->cmds->sync_from(fsa_cib_conn, sync_from, NULL, cib_quorum_override);
fsa_cib_conn->cmds->register_callback(fsa_cib_conn, rc, 60, FALSE, sync_from,
"finalize_sync_callback", finalize_sync_callback);
}
gboolean
crm_timer_popped(gpointer data)
{
fsa_timer_t *timer = (fsa_timer_t *) data;
if (timer == wait_timer
|| timer == recheck_timer
|| timer == transition_timer || timer == finalization_timer || timer == election_trigger) {
crm_info("%s (%s) just popped (%dms)",
get_timer_desc(timer), fsa_input2string(timer->fsa_input), timer->period_ms);
} else {
crm_err("%s (%s) just popped in state %s! (%dms)",
get_timer_desc(timer), fsa_input2string(timer->fsa_input),
fsa_state2string(fsa_state), timer->period_ms);
}
if (timer->repeat == FALSE) {
crm_timer_stop(timer); /* make it _not_ go off again */
}
if (timer->fsa_input == I_INTEGRATED) {
crm_info("Welcomed: %d, Integrated: %d",
g_hash_table_size(welcomed_nodes), g_hash_table_size(integrated_nodes));
if (g_hash_table_size(welcomed_nodes) == 0) {
/* If we don't even have ourself, start again */
register_fsa_error_adv(C_FSA_INTERNAL, I_ELECTION, NULL, NULL, __FUNCTION__);
} else {
register_fsa_input_before(C_TIMER_POPPED, timer->fsa_input, NULL);
}
} else if (timer == recheck_timer && fsa_state != S_IDLE) {
crm_debug("Discarding %s event in state: %s",
fsa_input2string(timer->fsa_input), fsa_state2string(fsa_state));
} else if (timer == finalization_timer && fsa_state != S_FINALIZE_JOIN) {
crm_debug("Discarding %s event in state: %s",
fsa_input2string(timer->fsa_input), fsa_state2string(fsa_state));
} else if (timer->fsa_input != I_NULL) {
register_fsa_input(C_TIMER_POPPED, timer->fsa_input, NULL);
}
crm_trace("Triggering FSA: %s", __FUNCTION__);
mainloop_set_trigger(fsa_source);
return TRUE;
}
void
join_query_callback(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data)
{
xmlNode *local_cib = NULL;
char *join_id = user_data;
xmlNode *generation = create_xml_node(NULL, XML_CIB_TAG_GENERATION_TUPPLE);
CRM_LOG_ASSERT(join_id != NULL);
query_call_id = 0;
if (rc == pcmk_ok) {
local_cib = output;
CRM_LOG_ASSERT(safe_str_eq(crm_element_name(local_cib), XML_TAG_CIB));
}
if (local_cib != NULL) {
xmlNode *reply = NULL;
crm_debug("Respond to join offer join-%s", join_id);
crm_debug("Acknowledging %s as our DC", fsa_our_dc);
copy_in_properties(generation, local_cib);
reply = create_request(CRM_OP_JOIN_REQUEST, generation, fsa_our_dc,
CRM_SYSTEM_DC, CRM_SYSTEM_CRMD, NULL);
crm_xml_add(reply, F_CRM_JOIN_ID, join_id);
if (fsa_our_dc) {
send_cluster_message(crm_get_peer(0, fsa_our_dc), crm_msg_crmd, reply, TRUE);
} else {
crm_warn("No DC for join-%s", join_id);
send_cluster_message(NULL, crm_msg_crmd, reply, TRUE);
}
free_xml(reply);
} else {
crm_err("Could not retrieve Generation to attach to our"
" join acknowledgement: %s", pcmk_strerror(rc));
register_fsa_error_adv(C_FSA_INTERNAL, I_ERROR, NULL, NULL, __FUNCTION__);
}
free(join_id);
free_xml(generation);
}
/*
CIBへのsync_from処理のコールバック処理
CRM_OP_JOIN_ACKNAKメッセージを送信する
*/
void
finalize_sync_callback(xmlNode *msg, int call_id, int rc,
xmlNode *output, void *user_data)
{
CRM_DEV_ASSERT(cib_not_master != rc);
clear_bit_inplace(fsa_input_register, R_CIB_ASKED);
if(rc != cib_ok) {
do_crm_log((rc==cib_old_data?LOG_WARNING:LOG_ERR),
"Sync from %s resulted in an error: %s",
(char*)user_data, cib_error2string(rc));
/* restart the whole join process */
/* コールバックがcib_okではない場合は、I_ELECTION_DCに戻る */
register_fsa_error_adv(
C_FSA_INTERNAL, I_ELECTION_DC,NULL,NULL,__FUNCTION__);
} else if(AM_I_DC && fsa_state == S_FINALIZE_JOIN) {
/* 自ノードがDCノードで、fsa_stateがS_FINALIZE_JOINの場合 */
set_bit_inplace(fsa_input_register, R_HAVE_CIB);
clear_bit_inplace(fsa_input_register, R_CIB_ASKED);
/* make sure dc_uuid is re-set to us */
/* JOIN状態チェック処理 */
if(check_join_state(fsa_state, __FUNCTION__) == FALSE) {
crm_debug("Notifying %d clients of join-%d results",
g_hash_table_size(integrated_nodes), current_join_id);
/*
ノードをCIBのnodeエントリに追加して、CRM_OP_JOIN_ACKNAKメッセージを送信する
また、クラスタメンバーとして認識したノードは、finalized_nodesハッシュテーブルに追加する
そして、integrated_nodesハッシュテーブルから削除する
*/
g_hash_table_foreach_remove(
integrated_nodes, finalize_join_for, NULL);
}
} else {
crm_debug("No longer the DC in S_FINALIZE_JOIN: %s/%s",
AM_I_DC?"DC":"CRMd", fsa_state2string(fsa_state));
}
crm_free(user_data);
}
void
crmd_ha_msg_filter(xmlNode * msg)
{
if (AM_I_DC) {
const char *sys_from = crm_element_value(msg, F_CRM_SYS_FROM);
if (safe_str_eq(sys_from, CRM_SYSTEM_DC)) {
const char *from = crm_element_value(msg, F_ORIG);
if (safe_str_neq(from, fsa_our_uname)) {
int level = LOG_INFO;
const char *op = crm_element_value(msg, F_CRM_TASK);
/* make sure the election happens NOW */
if (fsa_state != S_ELECTION) {
ha_msg_input_t new_input;
level = LOG_WARNING;
new_input.msg = msg;
register_fsa_error_adv(C_FSA_INTERNAL, I_ELECTION, NULL, &new_input,
__FUNCTION__);
}
do_crm_log(level, "Another DC detected: %s (op=%s)", from, op);
goto done;
}
}
} else {
const char *sys_to = crm_element_value(msg, F_CRM_SYS_TO);
if (safe_str_eq(sys_to, CRM_SYSTEM_DC)) {
return;
}
}
/* crm_log_xml_trace("HA[inbound]", msg); */
route_message(C_HA_MESSAGE, msg);
done:
trigger_fsa(fsa_source);
}
void
join_query_callback(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data)
{
char *join_id = user_data;
xmlNode *generation = create_xml_node(NULL, XML_CIB_TAG_GENERATION_TUPPLE);
CRM_LOG_ASSERT(join_id != NULL);
if (query_call_id != call_id) {
crm_trace("Query %d superceeded", call_id);
goto done;
}
query_call_id = 0;
if(rc != pcmk_ok || output == NULL) {
crm_err("Could not retrieve version details for join-%s: %s (%d)",
join_id, pcmk_strerror(rc), rc);
register_fsa_error_adv(C_FSA_INTERNAL, I_ERROR, NULL, NULL, __FUNCTION__);
} else if (fsa_our_dc == NULL) {
crm_debug("Membership is in flux, not continuing join-%s", join_id);
} else {
xmlNode *reply = NULL;
crm_debug("Respond to join offer join-%s from %s", join_id, fsa_our_dc);
copy_in_properties(generation, output);
reply = create_request(CRM_OP_JOIN_REQUEST, generation, fsa_our_dc,
CRM_SYSTEM_DC, CRM_SYSTEM_CRMD, NULL);
crm_xml_add(reply, F_CRM_JOIN_ID, join_id);
send_cluster_message(crm_get_peer(0, fsa_our_dc), crm_msg_crmd, reply, TRUE);
free_xml(reply);
}
done:
free_xml(generation);
free(join_id);
}
/*!
* \internal
* \brief Save CIB query result to file, raising FSA error
*
* \param[in] msg Ignored
* \param[in] call_id Call ID of CIB query
* \param[in] rc Return code of CIB query
* \param[in] output Result of CIB query
* \param[in] user_data Unique identifier for filename (will be freed)
*
* \note This is intended to be called after a scheduler connection fails.
*/
static void
save_cib_contents(xmlNode *msg, int call_id, int rc, xmlNode *output,
void *user_data)
{
char *id = user_data;
register_fsa_error_adv(C_FSA_INTERNAL, I_ERROR, NULL, NULL, __FUNCTION__);
CRM_CHECK(id != NULL, return);
if (rc == pcmk_ok) {
char *filename = crm_strdup_printf(PE_STATE_DIR "/pe-core-%s.bz2", id);
if (write_xml_file(output, filename, TRUE) < 0) {
crm_err("Could not save Cluster Information Base to %s after scheduler crash",
filename);
} else {
crm_notice("Saved Cluster Information Base to %s after scheduler crash",
filename);
}
free(filename);
}
}
static void
do_pe_invoke_callback(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data)
{
xmlNode *cmd = NULL;
pid_t watchdog = pcmk_locate_sbd();
if (rc != pcmk_ok) {
crm_err("Could not retrieve the Cluster Information Base: %s "
CRM_XS " rc=%d call=%d", pcmk_strerror(rc), rc, call_id);
register_fsa_error_adv(C_FSA_INTERNAL, I_ERROR, NULL, NULL, __FUNCTION__);
return;
} else if (call_id != fsa_pe_query) {
crm_trace("Skipping superseded CIB query: %d (current=%d)", call_id, fsa_pe_query);
return;
} else if (AM_I_DC == FALSE || is_set(fsa_input_register, R_PE_CONNECTED) == FALSE) {
crm_debug("No need to invoke the scheduler anymore");
return;
} else if (fsa_state != S_POLICY_ENGINE) {
crm_debug("Discarding scheduler request in state: %s",
fsa_state2string(fsa_state));
return;
/* this callback counts as 1 */
} else if (num_cib_op_callbacks() > 1) {
crm_debug("Re-asking for the CIB: %d other peer updates still pending",
(num_cib_op_callbacks() - 1));
sleep(1);
register_fsa_action(A_PE_INVOKE);
return;
} else if (fsa_state != S_POLICY_ENGINE) {
crm_err("Invoking scheduler in state: %s", fsa_state2string(fsa_state));
return;
}
CRM_LOG_ASSERT(output != NULL);
// Refresh the remote node cache when the scheduler is invoked
crm_remote_peer_cache_refresh(output);
crm_xml_add(output, XML_ATTR_DC_UUID, fsa_our_uuid);
crm_xml_add_int(output, XML_ATTR_HAVE_QUORUM, fsa_has_quorum);
force_local_option(output, XML_ATTR_HAVE_WATCHDOG, watchdog?"true":"false");
if (ever_had_quorum && crm_have_quorum == FALSE) {
crm_xml_add_int(output, XML_ATTR_QUORUM_PANIC, 1);
}
cmd = create_request(CRM_OP_PECALC, output, NULL, CRM_SYSTEM_PENGINE, CRM_SYSTEM_DC, NULL);
free(fsa_pe_ref);
fsa_pe_ref = crm_element_value_copy(cmd, XML_ATTR_REFERENCE);
rc = pe_subsystem_send(cmd);
if (rc < 0) {
crm_err("Could not contact the scheduler: %s " CRM_XS " rc=%d",
pcmk_strerror(rc), rc);
register_fsa_error_adv(C_FSA_INTERNAL, I_ERROR, NULL, NULL, __FUNCTION__);
}
crm_debug("Invoking the scheduler: query=%d, ref=%s, seq=%llu, quorate=%d",
fsa_pe_query, fsa_pe_ref, crm_peer_seq, fsa_has_quorum);
free_xml(cmd);
}
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);
}
}
}
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
tengine_stonith_notify(stonith_t * st, const char *event, xmlNode * msg)
{
int rc = -99;
const char *origin = NULL;
const char *target = NULL;
const char *executioner = NULL;
xmlNode *action = get_xpath_object("//st-data", msg, LOG_ERR);
if (action == NULL) {
crm_log_xml(LOG_ERR, "Notify data not found", msg);
return;
}
crm_log_xml(LOG_DEBUG, "stonith_notify", msg);
crm_element_value_int(msg, F_STONITH_RC, &rc);
origin = crm_element_value(action, F_STONITH_ORIGIN);
target = crm_element_value(action, F_STONITH_TARGET);
executioner = crm_element_value(action, F_STONITH_DELEGATE);
if (rc == stonith_ok && crm_str_eq(target, fsa_our_uname, TRUE)) {
crm_err("We were alegedly just fenced by %s for %s!", executioner, origin);
register_fsa_error_adv(C_FSA_INTERNAL, I_ERROR, NULL, NULL, __FUNCTION__);
} else if (rc == stonith_ok) {
crm_info("Peer %s was terminated (%s) by %s for %s (ref=%s): %s",
target,
crm_element_value(action, F_STONITH_OPERATION),
executioner, origin,
crm_element_value(action, F_STONITH_REMOTE), stonith_error2string(rc));
} else {
crm_err("Peer %s could not be terminated (%s) by %s for %s (ref=%s): %s",
target,
crm_element_value(action, F_STONITH_OPERATION),
executioner ? executioner : "<anyone>", origin,
crm_element_value(action, F_STONITH_REMOTE), stonith_error2string(rc));
}
#ifdef SUPPORT_CMAN
if (rc == stonith_ok && is_cman_cluster()) {
int local_rc = 0;
int confirm = 0;
char *target_copy = crm_strdup(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", target, local_rc);
} else {
crm_notice("Notified CMAN that '%s' is now fenced", target);
}
/* In case fenced is already trying to shoot it */
confirm = open("/var/run/cluster/fenced_override", O_NONBLOCK|O_WRONLY);
if (confirm) {
int len = strlen(target_copy);
errno = 0;
local_rc = write(confirm, target_copy, len);
write(confirm, "\n", 1);
if(errno == EBADF) {
crm_trace("CMAN not expecting %s to be fenced (yet)", target);
} else if (local_rc < len) {
crm_perror(LOG_ERR, "Confirmation of CMAN fencing event for '%s' failed: %d", target, local_rc);
} else {
fsync(confirm);
crm_notice("Confirmed CMAN fencing event for '%s'", target);
}
close(confirm);
}
}
#endif
if (rc == stonith_ok && safe_str_eq(target, origin)) {
if (fsa_our_dc == NULL || safe_str_eq(fsa_our_dc, target)) {
crm_notice("Target was our leader %s (recorded: %s)",
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 (safe_str_eq(executioner, fsa_our_uname)) {
const char *uuid = get_uuid(target);
send_stonith_update(NULL, target, uuid);
} else {
stonith_cleanup_list = g_list_append(stonith_cleanup_list, crm_strdup(target));
}
}
}
}
