bool can_interleave_actions(pe_action_t *first, pe_action_t *then)
{
bool interleave = FALSE;
resource_t *rsc = NULL;
const char *interleave_s = NULL;
if(first->rsc == NULL || then->rsc == NULL) {
crm_trace("Not interleaving %s with %s (both must be resources)", first->uuid, then->uuid);
return FALSE;
} else if(first->rsc == then->rsc) {
crm_trace("Not interleaving %s with %s (must belong to different resources)", first->uuid, then->uuid);
return FALSE;
} else if(first->rsc->variant < pe_clone || then->rsc->variant < pe_clone) {
crm_trace("Not interleaving %s with %s (both sides must be clones, masters, or bundles)", first->uuid, then->uuid);
return FALSE;
}
if (crm_ends_with(then->uuid, "_stop_0") || crm_ends_with(then->uuid, "_demote_0")) {
rsc = first->rsc;
} else {
rsc = then->rsc;
}
interleave_s = g_hash_table_lookup(rsc->meta, XML_RSC_ATTR_INTERLEAVE);
interleave = crm_is_true(interleave_s);
crm_trace("Interleave %s -> %s: %s (based on %s)",
first->uuid, then->uuid, interleave ? "yes" : "no", rsc->id);
return interleave;
}
static void
pcmk_process_exit(pcmk_child_t * child)
{
child->pid = 0;
child->active_before_startup = FALSE;
/* Broadcast the fact that one of our processes died ASAP
*
* Try to get some logging of the cause out first though
* because we're probably about to get fenced
*
* Potentially do this only if respawn_count > N
* to allow for local recovery
*/
update_node_processes(local_nodeid, NULL, get_process_list());
child->respawn_count += 1;
if (child->respawn_count > MAX_RESPAWN) {
crm_err("Child respawn count exceeded by %s", child->name);
child->respawn = FALSE;
}
if (shutdown_trigger) {
mainloop_set_trigger(shutdown_trigger);
update_node_processes(local_nodeid, NULL, get_process_list());
} else if (child->respawn && crm_is_true(getenv("PCMK_fail_fast"))) {
crm_err("Rebooting system because of %s", child->name);
pcmk_panic(__FUNCTION__);
} else if (child->respawn) {
crm_notice("Respawning failed child process: %s", child->name);
start_child(child);
}
}
/*!
* \internal
* \brief Remember if DC is shutting down as we join
*
* If we're joining while the current DC is shutting down, update its expected
* state, so we don't fence it if we become the new DC. (We weren't a peer
* when it broadcast its shutdown request.)
*
* \param[in] msg A join message from the DC
*/
static void
update_dc_expected(xmlNode *msg)
{
if (fsa_our_dc && crm_is_true(crm_element_value(msg, F_CRM_DC_LEAVING))) {
crm_node_t *dc_node = crm_get_peer(0, fsa_our_dc);
crm_update_peer_expected(__FUNCTION__, dc_node, CRMD_JOINSTATE_DOWN);
}
}
/*
* Unpack everything
* At the end you'll have:
* - A list of nodes
* - A list of resources (each with any dependencies on other resources)
* - A list of constraints between resources and nodes
* - A list of constraints between start/stop actions
* - A list of nodes that need to be stonith'd
* - A list of nodes that need to be shutdown
* - A list of the possible stop/start actions (without dependencies)
*/
gboolean
cluster_status(pe_working_set_t *data_set)
{
xmlNode * config = get_object_root(
XML_CIB_TAG_CRMCONFIG, data_set->input);
xmlNode * cib_nodes = get_object_root(
XML_CIB_TAG_NODES, data_set->input);
xmlNode * cib_resources = get_object_root(
XML_CIB_TAG_RESOURCES, data_set->input);
xmlNode * cib_status = get_object_root(
XML_CIB_TAG_STATUS, data_set->input);
const char *value = crm_element_value(
data_set->input, XML_ATTR_HAVE_QUORUM);
crm_debug_3("Beginning unpack");
/* reset remaining global variables */
if(data_set->input == NULL) {
return FALSE;
}
if(data_set->now == NULL) {
data_set->now = new_ha_date(TRUE);
}
if(data_set->input != NULL
&& crm_element_value(data_set->input, XML_ATTR_DC_UUID) != NULL) {
/* this should always be present */
data_set->dc_uuid = crm_element_value_copy(
data_set->input, XML_ATTR_DC_UUID);
}
clear_bit_inplace(data_set->flags, pe_flag_have_quorum);
if(crm_is_true(value)) {
set_bit_inplace(data_set->flags, pe_flag_have_quorum);
}
data_set->op_defaults = get_object_root(XML_CIB_TAG_OPCONFIG, data_set->input);
data_set->rsc_defaults = get_object_root(XML_CIB_TAG_RSCCONFIG, data_set->input);
unpack_config(config, data_set);
if(is_set(data_set->flags, pe_flag_have_quorum) == FALSE
&& data_set->no_quorum_policy != no_quorum_ignore) {
crm_warn("We do not have quorum"
" - fencing and resource management disabled");
}
unpack_nodes(cib_nodes, data_set);
unpack_resources(cib_resources, data_set);
unpack_status(cib_status, data_set);
return TRUE;
}
static gboolean
te_crm_command(crm_graph_t *graph, crm_action_t *action)
{
char *value = NULL;
char *counter = NULL;
HA_Message *cmd = NULL;
const char *id = NULL;
const char *task = NULL;
const char *on_node = NULL;
gboolean ret = TRUE;
id = ID(action->xml);
task = crm_element_value(action->xml, XML_LRM_ATTR_TASK);
on_node = crm_element_value(action->xml, XML_LRM_ATTR_TARGET);
CRM_CHECK(on_node != NULL && strlen(on_node) != 0,
te_log_action(LOG_ERR, "Corrupted command (id=%s) %s: no node",
crm_str(id), crm_str(task));
return FALSE);
te_log_action(LOG_INFO, "Executing crm-event (%s): %s on %s",
crm_str(id), crm_str(task), on_node);
cmd = create_request(task, NULL, on_node, CRM_SYSTEM_CRMD,
CRM_SYSTEM_TENGINE, NULL);
counter = generate_transition_key(
transition_graph->id, action->id, te_uuid);
crm_xml_add(cmd, XML_ATTR_TRANSITION_KEY, counter);
ret = send_ipc_message(crm_ch, cmd);
crm_free(counter);
crm_msg_del(cmd);
value = g_hash_table_lookup(action->params, crm_meta_name(XML_ATTR_TE_NOWAIT));
if(ret == FALSE) {
crm_err("Action %d failed: send", action->id);
return FALSE;
} else if(crm_is_true(value)) {
crm_info("Skipping wait for %d", action->id);
action->confirmed = TRUE;
update_graph(graph, action);
trigger_graph();
} else if(ret && action->timeout > 0) {
crm_debug("Setting timer for action %d",action->id);
action->timer->reason = timeout_action_warn;
te_start_action_timer(action);
}
return TRUE;
}
static gboolean
daemon_option_enabled(const char *daemon, const char *option)
{
const char *value = daemon_option(option);
if (value != NULL && crm_is_true(value)) {
return TRUE;
} else if (value != NULL && strstr(value, daemon)) {
return TRUE;
}
return FALSE;
}
/*!
* \internal
* \brief Return node status based on a CIB status entry
*
* \param[in] node_state XML of node state
*
* \return CRM_NODE_LOST if XML_NODE_IN_CLUSTER is false in node_state,
* CRM_NODE_MEMBER otherwise
* \note Unlike most boolean XML attributes, this one defaults to true, for
* backward compatibility with older controllers that don't set it.
*/
static const char *
remote_state_from_cib(xmlNode *node_state)
{
const char *status;
status = crm_element_value(node_state, XML_NODE_IN_CLUSTER);
if (status && !crm_is_true(status)) {
status = CRM_NODE_LOST;
} else {
status = CRM_NODE_MEMBER;
}
return status;
}
static gboolean uname_is_uuid(void)
{
static const char *uuid_pref = NULL;
if(uuid_pref == NULL) {
uuid_pref = getenv("PCMK_uname_is_uuid");
}
if(uuid_pref == NULL) {
/* true is legacy mode */
uuid_pref = "false";
}
return crm_is_true(uuid_pref);
}
void
group_create_actions(resource_t * rsc, pe_working_set_t * data_set)
{
action_t *op = NULL;
const char *value = NULL;
GListPtr gIter = rsc->children;
crm_debug_2("Creating actions for %s", rsc->id);
for (; gIter != NULL; gIter = gIter->next) {
resource_t *child_rsc = (resource_t *) gIter->data;
child_rsc->cmds->create_actions(child_rsc, data_set);
group_update_pseudo_status(rsc, child_rsc);
}
op = start_action(rsc, NULL, TRUE /* !group_data->child_starting */ );
set_bit_inplace(op->flags, pe_action_pseudo | pe_action_runnable);
op = custom_action(rsc, started_key(rsc),
RSC_STARTED, NULL, TRUE /* !group_data->child_starting */ , TRUE, data_set);
set_bit_inplace(op->flags, pe_action_pseudo | pe_action_runnable);
op = stop_action(rsc, NULL, TRUE /* !group_data->child_stopping */ );
set_bit_inplace(op->flags, pe_action_pseudo | pe_action_runnable);
op = custom_action(rsc, stopped_key(rsc),
RSC_STOPPED, NULL, TRUE /* !group_data->child_stopping */ , TRUE, data_set);
set_bit_inplace(op->flags, pe_action_pseudo | pe_action_runnable);
value = g_hash_table_lookup(rsc->meta, "stateful");
if (crm_is_true(value)) {
op = custom_action(rsc, demote_key(rsc), RSC_DEMOTE, NULL, TRUE, TRUE, data_set);
set_bit_inplace(op->flags, pe_action_pseudo);
set_bit_inplace(op->flags, pe_action_runnable);
op = custom_action(rsc, demoted_key(rsc), RSC_DEMOTED, NULL, TRUE, TRUE, data_set);
set_bit_inplace(op->flags, pe_action_pseudo);
set_bit_inplace(op->flags, pe_action_runnable);
op = custom_action(rsc, promote_key(rsc), RSC_PROMOTE, NULL, TRUE, TRUE, data_set);
set_bit_inplace(op->flags, pe_action_pseudo);
set_bit_inplace(op->flags, pe_action_runnable);
op = custom_action(rsc, promoted_key(rsc), RSC_PROMOTED, NULL, TRUE, TRUE, data_set);
set_bit_inplace(op->flags, pe_action_pseudo);
set_bit_inplace(op->flags, pe_action_runnable);
}
}
/*!
* \brief Handle a CRM_OP_REMOTE_STATE message by updating remote peer cache
*
* \param[in] msg Message XML
*
* \return Next FSA input
*/
static enum crmd_fsa_input
handle_remote_state(xmlNode *msg)
{
const char *remote_uname = ID(msg);
const char *remote_is_up = crm_element_value(msg, XML_NODE_IN_CLUSTER);
crm_node_t *remote_peer;
CRM_CHECK(remote_uname && remote_is_up, return I_NULL);
remote_peer = crm_remote_peer_get(remote_uname);
CRM_CHECK(remote_peer, return I_NULL);
crm_update_peer_state(__FUNCTION__, remote_peer,
crm_is_true(remote_is_up)?
CRM_NODE_MEMBER : CRM_NODE_LOST, 0);
return I_NULL;
}
cib_t *
cib_new(void)
{
const char *value = getenv("CIB_shadow");
if (value && value[0] != 0) {
return cib_shadow_new(value);
}
value = getenv("CIB_file");
if (value) {
return cib_file_new(value);
}
value = getenv("CIB_port");
if (value) {
gboolean encrypted = TRUE;
int port = crm_parse_int(value, NULL);
const char *server = getenv("CIB_server");
const char *user = getenv("CIB_user");
const char *pass = getenv("CIB_passwd");
value = getenv("CIB_encrypted");
if (value && crm_is_true(value) == FALSE) {
crm_info("Disabling TLS");
encrypted = FALSE;
}
if (user == NULL) {
user = CRM_DAEMON_USER;
crm_info("Defaulting to user: %s", user);
}
if (server == NULL) {
server = "localhost";
crm_info("Defaulting to localhost");
}
return cib_remote_new(server, user, pass, port, encrypted);
}
return cib_native_new();
}
static int
cib_prepare_diff(xmlNode * request, xmlNode ** data, const char **section)
{
xmlNode *input_fragment = NULL;
const char *update = crm_element_value(request, F_CIB_GLOBAL_UPDATE);
*data = NULL;
*section = NULL;
if (crm_is_true(update)) {
input_fragment = get_message_xml(request, F_CIB_UPDATE_DIFF);
} else {
input_fragment = get_message_xml(request, F_CIB_CALLDATA);
}
CRM_CHECK(input_fragment != NULL, crm_log_xml_warn(request, "no input"));
*data = cib_prepare_common(input_fragment, NULL);
return pcmk_ok;
}
gboolean
is_node_online(xmlNode * node_state)
{
const char *uname = crm_element_value(node_state, XML_ATTR_UNAME);
const char *join_state = crm_element_value(node_state, XML_NODE_JOIN_STATE);
const char *exp_state = crm_element_value(node_state, XML_NODE_EXPECTED);
const char *crm_state = crm_element_value(node_state, XML_NODE_IS_PEER);
const char *ccm_state = crm_element_value(node_state, XML_NODE_IN_CLUSTER);
if (safe_str_neq(join_state, CRMD_JOINSTATE_DOWN)
&& crm_is_true(ccm_state)
&& safe_str_eq(crm_state, "online")) {
crm_trace("Node %s is online", uname);
return TRUE;
}
crm_trace("Node %s: ccm=%s join=%s exp=%s crm=%s",
uname, crm_str(ccm_state),
crm_str(join_state), crm_str(exp_state), crm_str(crm_state));
crm_trace("Node %s is offline", uname);
return FALSE;
}
gboolean
is_node_online(xmlNode * node_state)
{
const char *uname = crm_element_value(node_state, XML_ATTR_UNAME);
const char *join_state = crm_element_value(node_state, XML_CIB_ATTR_JOINSTATE);
const char *exp_state = crm_element_value(node_state, XML_CIB_ATTR_EXPSTATE);
const char *crm_state = crm_element_value(node_state, XML_CIB_ATTR_CRMDSTATE);
const char *ha_state = crm_element_value(node_state, XML_CIB_ATTR_HASTATE);
const char *ccm_state = crm_element_value(node_state, XML_CIB_ATTR_INCCM);
if (safe_str_neq(join_state, CRMD_JOINSTATE_DOWN)
&& (ha_state == NULL || safe_str_eq(ha_state, "active"))
&& crm_is_true(ccm_state)
&& safe_str_eq(crm_state, "online")) {
crm_trace("Node %s is online", uname);
return TRUE;
}
crm_trace("Node %s: ha=%s ccm=%s join=%s exp=%s crm=%s",
uname, crm_str(ha_state), crm_str(ccm_state),
crm_str(join_state), crm_str(exp_state), crm_str(crm_state));
crm_trace("Node %s is offline", uname);
return FALSE;
}
gboolean
read_config(void)
{
int rc = CS_OK;
int retries = 0;
gboolean have_log = FALSE;
char *logging_debug = NULL;
char *logging_logfile = NULL;
char *logging_to_logfile = NULL;
char *logging_to_syslog = NULL;
char *logging_syslog_facility = NULL;
enum cluster_type_e stack = pcmk_cluster_unknown;
#if HAVE_CONFDB
char *value = NULL;
confdb_handle_t config;
confdb_handle_t top_handle = 0;
hdb_handle_t local_handle;
static confdb_callbacks_t callbacks = { };
do {
rc = confdb_initialize(&config, &callbacks);
if(rc < 0) {
retries++;
printf("Connection setup failed: %d. Retrying in %ds\n", rc, retries);
sleep(retries);
} else {
break;
}
} while(retries < 5);
#elif HAVE_CMAP
cmap_handle_t local_handle;
/* There can be only one (possibility if confdb isn't around) */
do {
rc = cmap_initialize(&local_handle);
if(rc < 0) {
retries++;
printf("Connection setup failed: %s. Retrying in %ds\n",
cs_strerror(rc), retries);
sleep(retries);
} else {
break;
}
} while(retries < 5);
#endif
if (rc != CS_OK) {
printf("Could not initialize Cluster Configuration Database API instance, error %d\n", rc);
return FALSE;
}
stack = get_cluster_type();
crm_info("Reading configure for stack: %s", name_for_cluster_type(stack));
/* =::=::= Should we be here =::=::= */
if (stack == pcmk_cluster_corosync) {
setenv("HA_cluster_type", "corosync", 1);
setenv("HA_quorum_type", "corosync", 1);
#if HAVE_CONFDB
} else if (stack == pcmk_cluster_cman) {
setenv("HA_cluster_type", "cman", 1);
setenv("HA_quorum_type", "cman", 1);
enable_crmd_as_root(TRUE);
use_cman = TRUE;
} else if (stack == pcmk_cluster_classic_ais) {
setenv("HA_cluster_type", "openais", 1);
setenv("HA_quorum_type", "pcmk", 1);
/* Look for a service block to indicate our plugin is loaded */
top_handle = config_find_init(config);
local_handle = config_find_next(config, "service", top_handle);
while (local_handle) {
crm_free(value);
get_config_opt(config, local_handle, "name", &value, NULL);
if (safe_str_eq("pacemaker", value)) {
crm_free(value);
get_config_opt(config, local_handle, "ver", &value, "0");
if (safe_str_eq(value, "1")) {
crm_free(value);
get_config_opt(config, local_handle, "use_logd", &value, "no");
setenv("HA_use_logd", value, 1);
setenv("HA_LOGD", value, 1);
crm_free(value);
get_config_opt(config, local_handle, "use_mgmtd", &value, "no");
enable_mgmtd(crm_is_true(value));
} else {
crm_err("We can only start Pacemaker from init if using version 1"
" of the Pacemaker plugin for Corosync. Terminating.");
exit(100);
}
break;
}
local_handle = config_find_next(config, "service", top_handle);
}
crm_free(value);
#endif
} else {
crm_err("Unsupported stack type: %s", name_for_cluster_type(stack));
return FALSE;
}
#if HAVE_CONFDB
top_handle = config_find_init(config);
local_handle = config_find_next(config, "logging", top_handle);
get_config_opt(config, local_handle, "debug", &logging_debug, "off");
get_config_opt(config, local_handle, "logfile", &logging_logfile, "/var/log/pacemaker");
get_config_opt(config, local_handle, "to_logfile", &logging_to_logfile, "off");
get_config_opt(config, local_handle, "to_syslog", &logging_to_syslog, "on");
get_config_opt(config, local_handle, "syslog_facility", &logging_syslog_facility, "daemon");
confdb_finalize(config);
#elif HAVE_CMAP
/* =::=::= Logging =::=::= */
get_config_opt(local_handle, "logging.debug", &logging_debug, "off");
get_config_opt(local_handle, "logging.logfile", &logging_logfile, "/var/log/pacemaker");
get_config_opt(local_handle, "logging.to_logfile", &logging_to_logfile, "off");
get_config_opt(local_handle, "logging.to_syslog", &logging_to_syslog, "on");
get_config_opt(local_handle, "logging.syslog_facility", &logging_syslog_facility, "daemon");
cmap_finalize(local_handle);
#endif
if (crm_is_true(logging_debug)) {
setenv("HA_debug", "1", 1);
if(get_crm_log_level() < LOG_DEBUG) {
set_crm_log_level(LOG_DEBUG);
}
} else {
setenv("HA_debug", "0", 1);
}
if (crm_is_true(logging_to_logfile)) {
struct stat parent;
uid_t pcmk_uid = 0;
uid_t pcmk_gid = getegid();
FILE *logfile = NULL;
char *parent_dir = NULL;
struct passwd *pcmk_user = getpwnam(CRM_DAEMON_USER);
if(pcmk_user) {
pcmk_uid = pcmk_user->pw_uid;
} else {
crm_err("User %s does not exist. Terminating", CRM_DAEMON_USER);
exit(100);
}
parent_dir = dirname(strdup(logging_logfile));
rc = stat(parent_dir, &parent);
if (rc != 0) {
crm_err("Directory '%s' does not exist for logfile '%s'", parent_dir, logging_logfile);
} else if (parent.st_uid == pcmk_uid && (parent.st_mode & (S_IRUSR | S_IWUSR))) {
/* all good - user */
logfile = fopen(logging_logfile, "a");
} else if (parent.st_gid == pcmk_gid && (parent.st_mode & S_IXGRP)) {
/* all good - group */
logfile = fopen(logging_logfile, "a");
} else {
crm_err
("Daemons running as %s do not have permission to access '%s'. Logging to '%s' is disabled",
CRM_DAEMON_USER, parent_dir, logging_logfile);
crm_err
("Either %s must be owned by %s with rwx permissions, or have rwx permissions for group %d",
parent_dir, CRM_DAEMON_USER, pcmk_gid);
}
if (logfile) {
int logfd = fileno(logfile);
setenv("HA_debugfile", logging_logfile, 1);
setenv("HA_DEBUGLOG", logging_logfile, 1);
setenv("HA_LOGFILE", logging_logfile, 1);
/* Ensure the file has the correct permissions */
rc = fchown(logfd, pcmk_uid, pcmk_gid);
if(rc < 0) {
crm_warn("Cannot change the ownership of %s to user %s and gid %d",
logging_logfile, CRM_DAEMON_USER, pcmk_gid);
}
rc = fchmod(logfd, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
if(rc < 0) {
crm_warn("Cannot change the mode of %s to rw-rw----", logging_logfile);
}
have_log = TRUE;
fprintf(logfile, "Set r/w permissions for uid=%d, gid=%d on %s\n",
pcmk_uid, pcmk_gid, logging_logfile);
if(fflush(logfile) < 0 || fsync(logfd) < 0) {
crm_err("Couldn't write out logfile: %s", logging_logfile);
have_log = FALSE;
}
fclose(logfile);
} else {
crm_err("Couldn't create logfile: %s", logging_logfile);
}
crm_free(parent_dir);
}
if (have_log && crm_is_true(logging_to_syslog) == FALSE) {
crm_info("User configured file based logging and explicitly disabled syslog.");
free(logging_syslog_facility);
logging_syslog_facility = NULL;
} else {
if (crm_is_true(logging_to_syslog) == FALSE) {
crm_err
("Please enable some sort of logging, either 'to_logfile: on' or 'to_syslog: on'.");
crm_err("If you use file logging, be sure to also define a value for 'logfile'");
}
}
if(logging_syslog_facility) {
setenv("HA_logfacility", logging_syslog_facility, 1);
setenv("HA_LOGFACILITY", logging_syslog_facility, 1);
} else {
unsetenv("HA_logfacility");
unsetenv("HA_LOGFACILITY");
}
free(logging_debug);
free(logging_logfile);
free(logging_to_logfile);
free(logging_to_syslog);
free(logging_syslog_facility);
return TRUE;
}
static gboolean
te_rsc_command(crm_graph_t * graph, crm_action_t * action)
{
/* never overwrite stop actions in the CIB with
* anything other than completed results
*
* Writing pending stops makes it look like the
* resource is running again
*/
xmlNode *cmd = NULL;
xmlNode *rsc_op = NULL;
gboolean rc = TRUE;
gboolean no_wait = FALSE;
gboolean is_local = FALSE;
char *counter = NULL;
const char *task = NULL;
const char *value = NULL;
const char *on_node = NULL;
const char *task_uuid = NULL;
CRM_ASSERT(action != NULL);
CRM_ASSERT(action->xml != NULL);
action->executed = FALSE;
on_node = crm_element_value(action->xml, XML_LRM_ATTR_TARGET);
CRM_CHECK(on_node != NULL && strlen(on_node) != 0,
te_log_action(LOG_ERR, "Corrupted command(id=%s) %s: no node",
ID(action->xml), crm_str(task));
return FALSE);
rsc_op = action->xml;
task = crm_element_value(rsc_op, XML_LRM_ATTR_TASK);
task_uuid = crm_element_value(action->xml, XML_LRM_ATTR_TASK_KEY);
on_node = crm_element_value(rsc_op, XML_LRM_ATTR_TARGET);
counter =
generate_transition_key(transition_graph->id, action->id, get_target_rc(action), te_uuid);
crm_xml_add(rsc_op, XML_ATTR_TRANSITION_KEY, counter);
if (safe_str_eq(on_node, fsa_our_uname)) {
is_local = TRUE;
}
value = crm_meta_value(action->params, XML_ATTR_TE_NOWAIT);
if (crm_is_true(value)) {
no_wait = TRUE;
}
crm_info("Initiating action %d: %s %s on %s%s%s",
action->id, task, task_uuid, on_node,
is_local ? " (local)" : "", no_wait ? " - no waiting" : "");
cmd = create_request(CRM_OP_INVOKE_LRM, rsc_op, on_node,
CRM_SYSTEM_LRMD, CRM_SYSTEM_TENGINE, NULL);
if (is_local) {
/* shortcut local resource commands */
ha_msg_input_t data = {
.msg = cmd,
.xml = rsc_op,
};
fsa_data_t msg = {
.id = 0,
.data = &data,
.data_type = fsa_dt_ha_msg,
.fsa_input = I_NULL,
.fsa_cause = C_FSA_INTERNAL,
.actions = A_LRM_INVOKE,
.origin = __FUNCTION__,
};
do_lrm_invoke(A_LRM_INVOKE, C_FSA_INTERNAL, fsa_state, I_NULL, &msg);
} else {
rc = send_cluster_message(on_node, crm_msg_lrmd, cmd, TRUE);
}
crm_free(counter);
free_xml(cmd);
action->executed = TRUE;
if (rc == FALSE) {
crm_err("Action %d failed: send", action->id);
return FALSE;
} else if (no_wait) {
action->confirmed = TRUE;
update_graph(transition_graph, action);
trigger_graph();
} else {
if (action->timeout <= 0) {
crm_err("Action %d: %s %s on %s had an invalid timeout (%dms). Using %dms instead",
action->id, task, task_uuid, on_node, action->timeout, graph->network_delay);
action->timeout = graph->network_delay;
}
te_start_action_timer(graph, action);
}
value = crm_meta_value(action->params, XML_OP_ATTR_PENDING);
if (crm_is_true(value)) {
/* write a "pending" entry to the CIB, inhibit notification */
crm_info("Recording pending op %s in the CIB", task_uuid);
cib_action_update(action, LRM_OP_PENDING, EXECRA_STATUS_UNKNOWN);
}
return TRUE;
}
static gboolean
te_crm_command(crm_graph_t * graph, crm_action_t * action)
{
char *counter = NULL;
xmlNode *cmd = NULL;
gboolean is_local = FALSE;
const char *id = NULL;
const char *task = NULL;
const char *value = NULL;
const char *on_node = NULL;
gboolean rc = TRUE;
gboolean no_wait = FALSE;
id = ID(action->xml);
task = crm_element_value(action->xml, XML_LRM_ATTR_TASK);
on_node = crm_element_value(action->xml, XML_LRM_ATTR_TARGET);
CRM_CHECK(on_node != NULL && strlen(on_node) != 0,
te_log_action(LOG_ERR, "Corrupted command (id=%s) %s: no node",
crm_str(id), crm_str(task));
return FALSE);
te_log_action(LOG_INFO, "Executing crm-event (%s): %s on %s%s%s",
crm_str(id), crm_str(task), on_node,
is_local ? " (local)" : "", no_wait ? " - no waiting" : "");
if (safe_str_eq(on_node, fsa_our_uname)) {
is_local = TRUE;
}
value = crm_meta_value(action->params, XML_ATTR_TE_NOWAIT);
if (crm_is_true(value)) {
no_wait = TRUE;
}
if (is_local && safe_str_eq(task, CRM_OP_SHUTDOWN)) {
/* defer until everything else completes */
te_log_action(LOG_INFO, "crm-event (%s) is a local shutdown", crm_str(id));
graph->completion_action = tg_shutdown;
graph->abort_reason = "local shutdown";
action->confirmed = TRUE;
update_graph(graph, action);
trigger_graph();
return TRUE;
}
cmd = create_request(task, action->xml, on_node, CRM_SYSTEM_CRMD, CRM_SYSTEM_TENGINE, NULL);
counter =
generate_transition_key(transition_graph->id, action->id, get_target_rc(action), te_uuid);
crm_xml_add(cmd, XML_ATTR_TRANSITION_KEY, counter);
rc = send_cluster_message(on_node, crm_msg_crmd, cmd, TRUE);
crm_free(counter);
free_xml(cmd);
if (rc == FALSE) {
crm_err("Action %d failed: send", action->id);
return FALSE;
} else if (no_wait) {
action->confirmed = TRUE;
update_graph(graph, action);
trigger_graph();
} else {
if (action->timeout <= 0) {
crm_err("Action %d: %s on %s had an invalid timeout (%dms). Using %dms instead",
action->id, task, on_node, action->timeout, graph->network_delay);
action->timeout = graph->network_delay;
}
te_start_action_timer(graph, action);
}
return TRUE;
}
/*
* 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;
}
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);
}
}
/*
* Unpack everything
* At the end you'll have:
* - A list of nodes
* - A list of resources (each with any dependencies on other resources)
* - A list of constraints between resources and nodes
* - A list of constraints between start/stop actions
* - A list of nodes that need to be stonith'd
* - A list of nodes that need to be shutdown
* - A list of the possible stop/start actions (without dependencies)
*/
gboolean
cluster_status(pe_working_set_t * data_set)
{
xmlNode *config = get_xpath_object("//"XML_CIB_TAG_CRMCONFIG, data_set->input, LOG_TRACE);
xmlNode *cib_nodes = get_xpath_object("//"XML_CIB_TAG_NODES, data_set->input, LOG_TRACE);
xmlNode *cib_resources = get_xpath_object("//"XML_CIB_TAG_RESOURCES, data_set->input, LOG_TRACE);
xmlNode *cib_status = get_xpath_object("//"XML_CIB_TAG_STATUS, data_set->input, LOG_TRACE);
xmlNode *cib_tags = get_xpath_object("//"XML_CIB_TAG_TAGS, data_set->input, LOG_TRACE);
const char *value = crm_element_value(data_set->input, XML_ATTR_HAVE_QUORUM);
crm_trace("Beginning unpack");
pe_dataset = data_set;
/* reset remaining global variables */
data_set->failed = create_xml_node(NULL, "failed-ops");
if (data_set->input == NULL) {
return FALSE;
}
if (data_set->now == NULL) {
data_set->now = crm_time_new(NULL);
}
if (data_set->dc_uuid == NULL
&& data_set->input != NULL
&& crm_element_value(data_set->input, XML_ATTR_DC_UUID) != NULL) {
/* this should always be present */
data_set->dc_uuid = crm_element_value_copy(data_set->input, XML_ATTR_DC_UUID);
}
clear_bit(data_set->flags, pe_flag_have_quorum);
if (crm_is_true(value)) {
set_bit(data_set->flags, pe_flag_have_quorum);
}
data_set->op_defaults = get_xpath_object("//"XML_CIB_TAG_OPCONFIG, data_set->input, LOG_TRACE);
data_set->rsc_defaults = get_xpath_object("//"XML_CIB_TAG_RSCCONFIG, data_set->input, LOG_TRACE);
unpack_config(config, data_set);
if (is_not_set(data_set->flags, pe_flag_quick_location)
&& is_not_set(data_set->flags, pe_flag_have_quorum)
&& data_set->no_quorum_policy != no_quorum_ignore) {
crm_warn("Fencing and resource management disabled due to lack of quorum");
}
unpack_nodes(cib_nodes, data_set);
if(is_not_set(data_set->flags, pe_flag_quick_location)) {
unpack_remote_nodes(cib_resources, data_set);
}
unpack_resources(cib_resources, data_set);
unpack_tags(cib_tags, data_set);
if(is_not_set(data_set->flags, pe_flag_quick_location)) {
unpack_status(cib_status, data_set);
}
set_bit(data_set->flags, pe_flag_have_status);
return TRUE;
}
gboolean
read_config(void)
{
int rc = CS_OK;
int retries = 0;
gboolean have_log = FALSE;
char *logging_debug = NULL;
char *logging_logfile = NULL;
char *logging_to_logfile = NULL;
char *logging_to_syslog = NULL;
char *logging_syslog_facility = NULL;
enum cluster_type_e stack = pcmk_cluster_unknown;
#if HAVE_CONFDB
char *value = NULL;
confdb_handle_t config;
confdb_handle_t top_handle = 0;
hdb_handle_t local_handle;
static confdb_callbacks_t callbacks = { };
do {
rc = confdb_initialize(&config, &callbacks);
if(rc != CS_OK) {
retries++;
printf("Connection setup failed: %d. Retrying in %ds\n", rc, retries);
sleep(retries);
} else {
break;
}
} while(retries < 5);
#elif HAVE_CMAP
cmap_handle_t local_handle;
/* There can be only one (possibility if confdb isn't around) */
do {
rc = cmap_initialize(&local_handle);
if(rc != CS_OK) {
retries++;
printf("API connection setup failed: %s. Retrying in %ds\n",
cs_strerror(rc), retries);
crm_info("API connection setup failed: %s. Retrying in %ds",
cs_strerror(rc), retries);
sleep(retries);
} else {
break;
}
} while(retries < 5);
#endif
if (rc != CS_OK) {
printf("Could not connect to Cluster Configuration Database API, error %d\n", rc);
crm_warn("Could not connect to Cluster Configuration Database API, error %d", rc);
return FALSE;
}
stack = get_cluster_type();
crm_info("Reading configure for stack: %s", name_for_cluster_type(stack));
/* =::=::= Should we be here =::=::= */
if (stack == pcmk_cluster_corosync) {
setenv("HA_cluster_type", "corosync", 1);
setenv("HA_quorum_type", "corosync", 1);
#if HAVE_CONFDB
} else if (stack == pcmk_cluster_cman) {
setenv("HA_cluster_type", "cman", 1);
setenv("HA_quorum_type", "cman", 1);
enable_crmd_as_root(TRUE);
use_cman = TRUE;
} else if (stack == pcmk_cluster_classic_ais) {
setenv("HA_cluster_type", "openais", 1);
setenv("HA_quorum_type", "pcmk", 1);
/* Look for a service block to indicate our plugin is loaded */
top_handle = config_find_init(config);
local_handle = config_find_next(config, "service", top_handle);
while (local_handle) {
get_config_opt(config, local_handle, "name", &value, NULL);
if (safe_str_eq("pacemaker", value)) {
get_config_opt(config, local_handle, "ver", &value, "0");
if (safe_str_eq(value, "1")) {
get_config_opt(config, local_handle, "use_logd", &value, "no");
setenv("HA_use_logd", value, 1);
setenv("HA_LOGD", value, 1);
get_config_opt(config, local_handle, "use_mgmtd", &value, "no");
enable_mgmtd(crm_is_true(value));
} else {
crm_err("We can only start Pacemaker from init if using version 1"
" of the Pacemaker plugin for Corosync. Terminating.");
exit(100);
}
break;
}
local_handle = config_find_next(config, "service", top_handle);
}
free(value);
#endif
} else {
crm_err("Unsupported stack type: %s", name_for_cluster_type(stack));
return FALSE;
}
#if HAVE_CONFDB
top_handle = config_find_init(config);
local_handle = config_find_next(config, "logging", top_handle);
get_config_opt(config, local_handle, "debug", &logging_debug, "off");
get_config_opt(config, local_handle, "logfile", &logging_logfile, "/var/log/pacemaker");
get_config_opt(config, local_handle, "to_logfile", &logging_to_logfile, "off");
get_config_opt(config, local_handle, "to_syslog", &logging_to_syslog, "on");
get_config_opt(config, local_handle, "syslog_facility", &logging_syslog_facility, "daemon");
confdb_finalize(config);
#elif HAVE_CMAP
/* =::=::= Logging =::=::= */
get_config_opt(local_handle, "logging.debug", &logging_debug, "off");
get_config_opt(local_handle, "logging.logfile", &logging_logfile, "/var/log/pacemaker");
get_config_opt(local_handle, "logging.to_logfile", &logging_to_logfile, "off");
get_config_opt(local_handle, "logging.to_syslog", &logging_to_syslog, "on");
get_config_opt(local_handle, "logging.syslog_facility", &logging_syslog_facility, "daemon");
cmap_finalize(local_handle);
#endif
if (crm_is_true(logging_debug)) {
setenv("HA_debug", "1", 1);
if(get_crm_log_level() < LOG_DEBUG) {
set_crm_log_level(LOG_DEBUG);
}
} else {
setenv("HA_debug", "0", 1);
}
if (crm_is_true(logging_to_logfile)) {
if(crm_add_logfile(logging_logfile)) {
setenv("HA_debugfile", logging_logfile, 1);
setenv("HA_DEBUGLOG", logging_logfile, 1);
setenv("HA_LOGFILE", logging_logfile, 1);
have_log = TRUE;
} else {
crm_err("Couldn't create logfile: %s", logging_logfile);
}
}
if (have_log && crm_is_true(logging_to_syslog) == FALSE) {
crm_info("User configured file based logging and explicitly disabled syslog.");
free(logging_syslog_facility);
logging_syslog_facility = NULL;
} else {
if (crm_is_true(logging_to_syslog) == FALSE) {
crm_err
("Please enable some sort of logging, either 'to_logfile: on' or 'to_syslog: on'.");
crm_err("If you use file logging, be sure to also define a value for 'logfile'");
}
}
if(logging_syslog_facility) {
setenv("HA_logfacility", logging_syslog_facility, 1);
setenv("HA_LOGFACILITY", logging_syslog_facility, 1);
} else {
unsetenv("HA_logfacility");
unsetenv("HA_LOGFACILITY");
}
free(logging_debug);
free(logging_logfile);
free(logging_to_logfile);
free(logging_to_syslog);
free(logging_syslog_facility);
return TRUE;
}
int
cib_process_command(xmlNode * request, xmlNode ** reply, xmlNode ** cib_diff, gboolean privileged)
{
xmlNode *input = NULL;
xmlNode *output = NULL;
xmlNode *result_cib = NULL;
xmlNode *current_cib = NULL;
#if ENABLE_ACL
xmlNode *filtered_current_cib = NULL;
#endif
int call_type = 0;
int call_options = 0;
int log_level = LOG_DEBUG_4;
const char *op = NULL;
const char *section = NULL;
int rc = pcmk_ok;
int rc2 = pcmk_ok;
gboolean send_r_notify = FALSE;
gboolean global_update = FALSE;
gboolean config_changed = FALSE;
gboolean manage_counters = TRUE;
CRM_ASSERT(cib_status == pcmk_ok);
*reply = NULL;
*cib_diff = NULL;
current_cib = the_cib;
/* Start processing the request... */
op = crm_element_value(request, F_CIB_OPERATION);
crm_element_value_int(request, F_CIB_CALLOPTS, &call_options);
rc = cib_get_operation_id(op, &call_type);
if (rc == pcmk_ok && privileged == FALSE) {
rc = cib_op_can_run(call_type, call_options, privileged, global_update);
}
rc2 = cib_op_prepare(call_type, request, &input, §ion);
if (rc == pcmk_ok) {
rc = rc2;
}
if (rc != pcmk_ok) {
crm_trace("Call setup failed: %s", pcmk_strerror(rc));
goto done;
} else if (cib_op_modifies(call_type) == FALSE) {
#if ENABLE_ACL
if (acl_enabled(config_hash) == FALSE
|| acl_filter_cib(request, current_cib, current_cib, &filtered_current_cib) == FALSE) {
rc = cib_perform_op(op, call_options, cib_op_func(call_type), TRUE,
section, request, input, FALSE, &config_changed,
current_cib, &result_cib, NULL, &output);
} else if (filtered_current_cib == NULL) {
crm_debug("Pre-filtered the entire cib");
rc = -EACCES;
} else {
crm_debug("Pre-filtered the queried cib according to the ACLs");
rc = cib_perform_op(op, call_options, cib_op_func(call_type), TRUE,
section, request, input, FALSE, &config_changed,
filtered_current_cib, &result_cib, NULL, &output);
}
#else
rc = cib_perform_op(op, call_options, cib_op_func(call_type), TRUE,
section, request, input, FALSE, &config_changed,
current_cib, &result_cib, NULL, &output);
#endif
CRM_CHECK(result_cib == NULL, free_xml(result_cib));
goto done;
}
/* Handle a valid write action */
global_update = crm_is_true(crm_element_value(request, F_CIB_GLOBAL_UPDATE));
if (global_update) {
manage_counters = FALSE;
call_options |= cib_force_diff;
CRM_CHECK(call_type == 3 || call_type == 4, crm_err("Call type: %d", call_type);
crm_log_xml_err(request, "bad op"));
}
#ifdef SUPPORT_PRENOTIFY
if ((call_options & cib_inhibit_notify) == 0) {
cib_pre_notify(call_options, op, the_cib, input);
}
#endif
if (rc == pcmk_ok) {
if (call_options & cib_inhibit_bcast) {
/* skip */
crm_trace("Skipping update: inhibit broadcast");
manage_counters = FALSE;
}
rc = cib_perform_op(op, call_options, cib_op_func(call_type), FALSE,
section, request, input, manage_counters, &config_changed,
current_cib, &result_cib, cib_diff, &output);
#if ENABLE_ACL
if (acl_enabled(config_hash) == TRUE
&& acl_check_diff(request, current_cib, result_cib, *cib_diff) == FALSE) {
rc = -EACCES;
}
#endif
if (rc == pcmk_ok && config_changed) {
time_t now;
char *now_str = NULL;
const char *validation = crm_element_value(result_cib, XML_ATTR_VALIDATION);
if (validation) {
int current_version = get_schema_version(validation);
int support_version = get_schema_version("pacemaker-1.1");
/* Once the later schemas support the "update-*" attributes, change "==" to ">=" -- Changed */
if (current_version >= support_version) {
const char *origin = crm_element_value(request, F_ORIG);
crm_xml_replace(result_cib, XML_ATTR_UPDATE_ORIG,
origin ? origin : cib_our_uname);
crm_xml_replace(result_cib, XML_ATTR_UPDATE_CLIENT,
crm_element_value(request, F_CIB_CLIENTNAME));
#if ENABLE_ACL
crm_xml_replace(result_cib, XML_ATTR_UPDATE_USER,
crm_element_value(request, F_CIB_USER));
#endif
}
}
now = time(NULL);
now_str = ctime(&now);
now_str[24] = EOS; /* replace the newline */
crm_xml_replace(result_cib, XML_CIB_ATTR_WRITTEN, now_str);
}
if (manage_counters == FALSE) {
config_changed = cib_config_changed(current_cib, result_cib, cib_diff);
}
/* Always write to disk for replace ops,
* this negates the need to detect ordering changes
*/
if (config_changed == FALSE && crm_str_eq(CIB_OP_REPLACE, op, TRUE)) {
config_changed = TRUE;
}
}
if (rc == pcmk_ok && (call_options & cib_dryrun) == 0) {
rc = activateCibXml(result_cib, config_changed, op);
if (rc == pcmk_ok && cib_internal_config_changed(*cib_diff)) {
cib_read_config(config_hash, result_cib);
}
if (crm_str_eq(CIB_OP_REPLACE, op, TRUE)) {
if (section == NULL) {
send_r_notify = TRUE;
} else if (safe_str_eq(section, XML_TAG_CIB)) {
send_r_notify = TRUE;
} else if (safe_str_eq(section, XML_CIB_TAG_NODES)) {
send_r_notify = TRUE;
} else if (safe_str_eq(section, XML_CIB_TAG_STATUS)) {
send_r_notify = TRUE;
}
} else if (crm_str_eq(CIB_OP_ERASE, op, TRUE)) {
send_r_notify = TRUE;
}
} else if (rc == -pcmk_err_dtd_validation) {
if (output != NULL) {
crm_log_xml_info(output, "cib:output");
free_xml(output);
}
#if ENABLE_ACL
{
xmlNode *filtered_result_cib = NULL;
if (acl_enabled(config_hash) == FALSE
|| acl_filter_cib(request, current_cib, result_cib,
&filtered_result_cib) == FALSE) {
output = result_cib;
} else {
crm_debug("Filtered the result cib for output according to the ACLs");
output = filtered_result_cib;
if (result_cib != NULL) {
free_xml(result_cib);
}
}
}
#else
output = result_cib;
#endif
} else {
free_xml(result_cib);
}
if ((call_options & cib_inhibit_notify) == 0) {
const char *call_id = crm_element_value(request, F_CIB_CALLID);
const char *client = crm_element_value(request, F_CIB_CLIENTNAME);
#ifdef SUPPORT_POSTNOTIFY
cib_post_notify(call_options, op, input, rc, the_cib);
#endif
cib_diff_notify(call_options, client, call_id, op, input, rc, *cib_diff);
}
if (send_r_notify) {
const char *origin = crm_element_value(request, F_ORIG);
cib_replace_notify(origin, the_cib, rc, *cib_diff);
}
if (rc != pcmk_ok) {
log_level = LOG_DEBUG_4;
if (rc == -pcmk_err_dtd_validation && global_update) {
log_level = LOG_WARNING;
crm_log_xml_info(input, "cib:global_update");
}
} else if (config_changed) {
log_level = LOG_DEBUG_3;
if (cib_is_master) {
log_level = LOG_NOTICE;
}
} else if (cib_is_master) {
log_level = LOG_DEBUG_2;
}
log_xml_diff(log_level, *cib_diff, "cib:diff");
done:
if ((call_options & cib_discard_reply) == 0) {
*reply = cib_construct_reply(request, output, rc);
crm_log_xml_trace(*reply, "cib:reply");
}
#if ENABLE_ACL
if (filtered_current_cib != NULL) {
free_xml(filtered_current_cib);
}
#endif
if (call_type >= 0) {
cib_op_cleanup(call_type, call_options, &input, &output);
}
return rc;
}
void
cib_process_request(xmlNode * request, gboolean force_synchronous, gboolean privileged,
gboolean from_peer, cib_client_t * cib_client)
{
int call_type = 0;
int call_options = 0;
gboolean process = TRUE;
gboolean is_update = TRUE;
gboolean needs_reply = TRUE;
gboolean local_notify = FALSE;
gboolean needs_forward = FALSE;
gboolean global_update = crm_is_true(crm_element_value(request, F_CIB_GLOBAL_UPDATE));
xmlNode *op_reply = NULL;
xmlNode *result_diff = NULL;
int rc = pcmk_ok;
const char *op = crm_element_value(request, F_CIB_OPERATION);
const char *originator = crm_element_value(request, F_ORIG);
const char *host = crm_element_value(request, F_CIB_HOST);
const char *client_id = crm_element_value(request, F_CIB_CLIENTID);
crm_trace("%s Processing msg %s", cib_our_uname, crm_element_value(request, F_SEQ));
cib_num_ops++;
if (cib_num_ops == 0) {
cib_num_fail = 0;
cib_num_local = 0;
cib_num_updates = 0;
crm_info("Stats wrapped around");
}
if (host != NULL && strlen(host) == 0) {
host = NULL;
}
crm_element_value_int(request, F_CIB_CALLOPTS, &call_options);
if (force_synchronous) {
call_options |= cib_sync_call;
}
crm_trace("Processing %s message (%s) for %s...",
from_peer ? "peer" : "local",
from_peer ? originator : cib_our_uname, host ? host : "master");
rc = cib_get_operation_id(op, &call_type);
if (rc != pcmk_ok) {
/* TODO: construct error reply? */
crm_err("Pre-processing of command failed: %s", pcmk_strerror(rc));
return;
}
is_update = cib_op_modifies(call_type);
if (is_update) {
cib_num_updates++;
}
if (from_peer == FALSE) {
parse_local_options(cib_client, call_type, call_options, host, op,
&local_notify, &needs_reply, &process, &needs_forward);
} else if (parse_peer_options(call_type, request, &local_notify,
&needs_reply, &process, &needs_forward) == FALSE) {
return;
}
crm_trace("Finished determining processing actions");
if (call_options & cib_discard_reply) {
needs_reply = is_update;
local_notify = FALSE;
}
if (needs_forward) {
forward_request(request, cib_client, call_options);
return;
}
if (cib_status != pcmk_ok) {
rc = cib_status;
crm_err("Operation ignored, cluster configuration is invalid."
" Please repair and restart: %s", pcmk_strerror(cib_status));
op_reply = cib_construct_reply(request, the_cib, cib_status);
} else if (process) {
int level = LOG_INFO;
const char *section = crm_element_value(request, F_CIB_SECTION);
cib_num_local++;
rc = cib_process_command(request, &op_reply, &result_diff, privileged);
if (global_update) {
switch (rc) {
case pcmk_ok:
case -pcmk_err_old_data:
case -pcmk_err_diff_resync:
case -pcmk_err_diff_failed:
level = LOG_DEBUG_2;
break;
default:
level = LOG_ERR;
}
} else if (safe_str_eq(op, CIB_OP_QUERY)) {
level = LOG_DEBUG_2;
} else if (rc != pcmk_ok) {
cib_num_fail++;
level = LOG_WARNING;
} else if (safe_str_eq(op, CIB_OP_SLAVE)) {
level = LOG_DEBUG_2;
} else if (safe_str_eq(section, XML_CIB_TAG_STATUS)) {
level = LOG_DEBUG_2;
}
do_crm_log_unlikely(level,
"Operation complete: op %s for section %s (origin=%s/%s/%s, version=%s.%s.%s): %s (rc=%d)",
op, section ? section : "'all'", originator ? originator : "local",
crm_element_value(request, F_CIB_CLIENTNAME), crm_element_value(request,
F_CIB_CALLID),
the_cib ? crm_element_value(the_cib, XML_ATTR_GENERATION_ADMIN) : "0",
the_cib ? crm_element_value(the_cib, XML_ATTR_GENERATION) : "0",
the_cib ? crm_element_value(the_cib, XML_ATTR_NUMUPDATES) : "0",
pcmk_strerror(rc), rc);
if (op_reply == NULL && (needs_reply || local_notify)) {
crm_err("Unexpected NULL reply to message");
crm_log_xml_err(request, "null reply");
needs_reply = FALSE;
local_notify = FALSE;
}
}
crm_trace("processing response cases %.16x %.16x", call_options, cib_sync_call);
/* from now on we are the server */
if (needs_reply == FALSE || stand_alone) {
/* nothing more to do...
* this was a non-originating slave update
*/
crm_trace("Completed slave update");
} else if (rc == pcmk_ok && result_diff != NULL && !(call_options & cib_inhibit_bcast)) {
gboolean broadcast = FALSE;
cib_local_bcast_num++;
crm_xml_add_int(request, F_CIB_LOCAL_NOTIFY_ID, cib_local_bcast_num);
broadcast = send_peer_reply(request, result_diff, originator, TRUE);
if (broadcast &&
client_id &&
local_notify &&
op_reply) {
/* If we have been asked to sync the reply,
* and a bcast msg has gone out, we queue the local notify
* until we know the bcast message has been received */
local_notify = FALSE;
queue_local_notify(op_reply, client_id, (call_options & cib_sync_call), from_peer);
op_reply = NULL; /* the reply is queued, so don't free here */
}
} else if (call_options & cib_discard_reply) {
crm_trace("Caller isn't interested in reply");
} else if (from_peer) {
if (is_update == FALSE || result_diff == NULL) {
crm_trace("Request not broadcast: R/O call");
} else if (call_options & cib_inhibit_bcast) {
crm_trace("Request not broadcast: inhibited");
} else if (rc != pcmk_ok) {
crm_trace("Request not broadcast: call failed: %s", pcmk_strerror(rc));
} else {
crm_trace("Directing reply to %s", originator);
}
send_peer_reply(op_reply, result_diff, originator, FALSE);
}
if (local_notify && client_id) {
if (process == FALSE) {
do_local_notify(request, client_id, call_options & cib_sync_call, from_peer);
} else {
do_local_notify(op_reply, client_id, call_options & cib_sync_call, from_peer);
}
}
free_xml(op_reply);
free_xml(result_diff);
return;
}
static gboolean
parse_peer_options(int call_type, xmlNode * request,
gboolean * local_notify, gboolean * needs_reply, gboolean * process,
gboolean * needs_forward)
{
const char *op = NULL;
const char *host = NULL;
const char *delegated = NULL;
const char *originator = crm_element_value(request, F_ORIG);
const char *reply_to = crm_element_value(request, F_CIB_ISREPLY);
const char *update = crm_element_value(request, F_CIB_GLOBAL_UPDATE);
gboolean is_reply = safe_str_eq(reply_to, cib_our_uname);
if (crm_is_true(update)) {
*needs_reply = FALSE;
if (is_reply) {
*local_notify = TRUE;
crm_trace("Processing global/peer update from %s"
" that originated from us", originator);
} else {
crm_trace("Processing global/peer update from %s", originator);
}
return TRUE;
}
host = crm_element_value(request, F_CIB_HOST);
if (host != NULL && safe_str_eq(host, cib_our_uname)) {
crm_trace("Processing request sent to us from %s", originator);
return TRUE;
} else if (host == NULL && cib_is_master == TRUE) {
crm_trace("Processing request sent to master instance from %s", originator);
return TRUE;
}
op = crm_element_value(request, F_CIB_OPERATION);
if(safe_str_eq(op, "cib_shutdown_req")) {
/* Always process these */
*local_notify = FALSE;
if(reply_to == NULL || is_reply) {
*process = TRUE;
}
if(is_reply) {
*needs_reply = FALSE;
}
return *process;
}
if (is_reply) {
crm_trace("Forward reply sent from %s to local clients", originator);
*process = FALSE;
*needs_reply = FALSE;
*local_notify = TRUE;
return TRUE;
}
delegated = crm_element_value(request, F_CIB_DELEGATED);
if (delegated != NULL) {
crm_trace("Ignoring msg for master instance");
} else if (host != NULL) {
/* this is for a specific instance and we're not it */
crm_trace("Ignoring msg for instance on %s", crm_str(host));
} else if (reply_to == NULL && cib_is_master == FALSE) {
/* this is for the master instance and we're not it */
crm_trace("Ignoring reply to %s", crm_str(reply_to));
} else if (safe_str_eq(op, "cib_shutdown_req")) {
if (reply_to != NULL) {
crm_debug("Processing %s from %s", op, host);
*needs_reply = FALSE;
} else {
crm_debug("Processing %s reply from %s", op, host);
}
return TRUE;
} else {
crm_err("Nothing for us to do?");
crm_log_xml_err(request, "Peer[inbound]");
}
return FALSE;
}
void
send_rsc_command(crm_action_t *action)
{
HA_Message *cmd = NULL;
crm_data_t *rsc_op = NULL;
char *counter = NULL;
const char *task = NULL;
const char *value = NULL;
const char *on_node = NULL;
const char *task_uuid = NULL;
CRM_ASSERT(action != NULL);
CRM_ASSERT(action->xml != NULL);
rsc_op = action->xml;
task = crm_element_value(rsc_op, XML_LRM_ATTR_TASK);
task_uuid = crm_element_value(action->xml, XML_LRM_ATTR_TASK_KEY);
on_node = crm_element_value(rsc_op, XML_LRM_ATTR_TARGET);
counter = generate_transition_key(
transition_graph->id, action->id, te_uuid);
crm_xml_add(rsc_op, XML_ATTR_TRANSITION_KEY, counter);
crm_info("Initiating action %d: %s on %s",
action->id, task_uuid, on_node);
crm_free(counter);
if(rsc_op != NULL) {
crm_log_xml_debug_2(rsc_op, "Performing");
}
cmd = create_request(CRM_OP_INVOKE_LRM, rsc_op, on_node,
CRM_SYSTEM_LRMD, CRM_SYSTEM_TENGINE, NULL);
#if 1
send_ipc_message(crm_ch, cmd);
#else
/* test the TE timer/recovery code */
if((action->id % 11) == 0) {
crm_err("Faking lost action %d: %s", action->id, task_uuid);
} else {
send_ipc_message(crm_ch, cmd);
}
#endif
crm_msg_del(cmd);
action->executed = TRUE;
value = g_hash_table_lookup(action->params, crm_meta_name(XML_ATTR_TE_NOWAIT));
if(crm_is_true(value)) {
crm_debug("Skipping wait for %d", action->id);
action->confirmed = TRUE;
update_graph(transition_graph, action);
trigger_graph();
} else if(action->timeout > 0) {
int action_timeout = (2 * action->timeout) + transition_graph->network_delay;
crm_debug_3("Setting timer for action %s", task_uuid);
if(transition_graph->transition_timeout < action_timeout) {
crm_debug("Action %d:"
" Increasing transition %d timeout to %d (2*%d + %d)",
action->id, transition_graph->id, action_timeout,
action->timeout, transition_graph->network_delay);
transition_graph->transition_timeout = action_timeout;
}
te_start_action_timer(action);
}
}
static gboolean
start_child(pcmk_child_t * child)
{
int lpc = 0;
uid_t uid = 0;
gid_t gid = 0;
struct rlimit oflimits;
gboolean use_valgrind = FALSE;
gboolean use_callgrind = FALSE;
const char *devnull = "/dev/null";
const char *env_valgrind = getenv("PCMK_valgrind_enabled");
const char *env_callgrind = getenv("PCMK_callgrind_enabled");
enum cluster_type_e stack = get_cluster_type();
child->active_before_startup = FALSE;
if (child->command == NULL) {
crm_info("Nothing to do for child \"%s\"", child->name);
return TRUE;
}
if (env_callgrind != NULL && crm_is_true(env_callgrind)) {
use_callgrind = TRUE;
use_valgrind = TRUE;
} else if (env_callgrind != NULL && strstr(env_callgrind, child->name)) {
use_callgrind = TRUE;
use_valgrind = TRUE;
} else if (env_valgrind != NULL && crm_is_true(env_valgrind)) {
use_valgrind = TRUE;
} else if (env_valgrind != NULL && strstr(env_valgrind, child->name)) {
use_valgrind = TRUE;
}
if (use_valgrind && strlen(VALGRIND_BIN) == 0) {
crm_warn("Cannot enable valgrind for %s:"
" The location of the valgrind binary is unknown", child->name);
use_valgrind = FALSE;
}
if (child->uid) {
if (crm_user_lookup(child->uid, &uid, &gid) < 0) {
crm_err("Invalid user (%s) for %s: not found", child->uid, child->name);
return FALSE;
}
crm_info("Using uid=%u and group=%u for process %s", uid, gid, child->name);
}
child->pid = fork();
CRM_ASSERT(child->pid != -1);
if (child->pid > 0) {
/* parent */
mainloop_child_add(child->pid, 0, child->name, child, pcmk_child_exit);
crm_info("Forked child %d for process %s%s", child->pid, child->name,
use_valgrind ? " (valgrind enabled: " VALGRIND_BIN ")" : "");
update_node_processes(local_nodeid, NULL, get_process_list());
return TRUE;
} else {
/* Start a new session */
(void)setsid();
/* Setup the two alternate arg arrarys */
opts_vgrind[0] = strdup(VALGRIND_BIN);
if (use_callgrind) {
opts_vgrind[1] = strdup("--tool=callgrind");
opts_vgrind[2] = strdup("--callgrind-out-file=" CRM_STATE_DIR "/callgrind.out.%p");
opts_vgrind[3] = strdup(child->command);
opts_vgrind[4] = NULL;
} else {
opts_vgrind[1] = strdup(child->command);
opts_vgrind[2] = NULL;
opts_vgrind[3] = NULL;
opts_vgrind[4] = NULL;
}
opts_default[0] = strdup(child->command);;
if(gid) {
if(stack == pcmk_cluster_corosync) {
/* Drop root privileges completely
*
* We can do this because we set uidgid.gid.${gid}=1
* via CMAP which allows these processes to connect to
* corosync
*/
if (setgid(gid) < 0) {
crm_perror(LOG_ERR, "Could not set group to %d", gid);
}
/* Keep the root group (so we can access corosync), but add the haclient group (so we can access ipc) */
} else if (initgroups(child->uid, gid) < 0) {
crm_err("Cannot initialize groups for %s: %s (%d)", child->uid, pcmk_strerror(errno), errno);
}
}
if (uid && setuid(uid) < 0) {
crm_perror(LOG_ERR, "Could not set user to %d (%s)", uid, child->uid);
}
/* Close all open file descriptors */
getrlimit(RLIMIT_NOFILE, &oflimits);
for (lpc = 0; lpc < oflimits.rlim_cur; lpc++) {
close(lpc);
}
(void)open(devnull, O_RDONLY); /* Stdin: fd 0 */
(void)open(devnull, O_WRONLY); /* Stdout: fd 1 */
(void)open(devnull, O_WRONLY); /* Stderr: fd 2 */
if (use_valgrind) {
(void)execvp(VALGRIND_BIN, opts_vgrind);
} else {
(void)execvp(child->command, opts_default);
}
crm_perror(LOG_ERR, "FATAL: Cannot exec %s", child->command);
crm_exit(DAEMON_RESPAWN_STOP);
}
return TRUE; /* never reached */
}
/* 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);
}
}
/*!
* \internal
* \brief Handle fence-history messages (either from API or coming in as
* broadcasts
*
* \param[in] msg Request message
* \param[in] output In case of a request from the API used to craft
* a reply from
* \param[in] remote_peer
* \param[in] options call-options from the request
*
* \return always success as there is actully nothing that can go really wrong
*/
int
stonith_fence_history(xmlNode *msg, xmlNode **output,
const char *remote_peer, int options)
{
int rc = 0;
const char *target = NULL;
xmlNode *dev = get_xpath_object("//@" F_STONITH_TARGET, msg, LOG_TRACE);
xmlNode *out_history = NULL;
if (dev) {
target = crm_element_value(dev, F_STONITH_TARGET);
if (target && (options & st_opt_cs_nodeid)) {
int nodeid = crm_atoi(target, NULL);
crm_node_t *node = crm_find_known_peer_full(nodeid, NULL, CRM_GET_PEER_ANY);
if (node) {
target = node->uname;
}
}
}
if (options & st_opt_cleanup) {
crm_trace("Cleaning up operations on %s in %p", target,
stonith_remote_op_list);
stonith_fence_history_cleanup(target,
crm_element_value(msg, F_STONITH_CALLID) != NULL);
} else if (options & st_opt_broadcast) {
if (crm_element_value(msg, F_STONITH_CALLID)) {
/* this is coming from the stonith-API
*
* craft a broadcast with node's history
* so that every node can merge and broadcast
* what it has on top
*/
out_history = stonith_local_history_diff(NULL, TRUE, NULL);
crm_trace("Broadcasting history to peers");
stonith_send_broadcast_history(out_history,
st_opt_broadcast | st_opt_discard_reply,
NULL);
} else if (remote_peer &&
!safe_str_eq(remote_peer, stonith_our_uname)) {
xmlNode *history =
get_xpath_object("//" F_STONITH_HISTORY_LIST, msg, LOG_TRACE);
GHashTable *received_history =
history?stonith_xml_history_to_list(history):NULL;
/* either a broadcast created directly upon stonith-API request
* or a diff as response to such a thing
*
* in both cases it may have a history or not
* if we have differential data
* merge in what we've received and stop
* otherwise broadcast what we have on top
* marking as differential and merge in afterwards
*/
if (!history ||
!crm_is_true(crm_element_value(history,
F_STONITH_DIFFERENTIAL))) {
out_history =
stonith_local_history_diff(received_history, TRUE, NULL);
if (out_history) {
crm_trace("Broadcasting history-diff to peers");
crm_xml_add(out_history, F_STONITH_DIFFERENTIAL,
XML_BOOLEAN_TRUE);
stonith_send_broadcast_history(out_history,
st_opt_broadcast | st_opt_discard_reply,
NULL);
} else {
crm_trace("History-diff is empty - skip broadcast");
}
}
stonith_merge_in_history_list(received_history);
} else {
crm_trace("Skipping history-query-broadcast (%s%s)"
" we sent ourselves",
remote_peer?"remote-peer=":"local-ipc",
remote_peer?remote_peer:"");
}
} else {
/* plain history request */
crm_trace("Looking for operations on %s in %p", target,
stonith_remote_op_list);
*output = stonith_local_history_diff(NULL, FALSE, target);
}
free_xml(out_history);
return rc;
}
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);
}
}
gboolean
mcp_read_config(void)
{
int rc = CS_OK;
int retries = 0;
cmap_handle_t local_handle;
uint64_t config = 0;
// There can be only one possibility
do {
rc = cmap_initialize(&local_handle);
if (rc != CS_OK) {
retries++;
printf("cmap connection setup failed: %s. Retrying in %ds\n", cs_strerror(rc), retries);
crm_info("cmap connection setup failed: %s. Retrying in %ds", cs_strerror(rc), retries);
sleep(retries);
} else {
break;
}
} while (retries < 5);
if (rc != CS_OK) {
printf("Could not connect to Cluster Configuration Database API, error %d\n", rc);
crm_warn("Could not connect to Cluster Configuration Database API, error %d", rc);
return FALSE;
}
stack = get_cluster_type();
crm_info("Reading configure for stack: %s", name_for_cluster_type(stack));
/* =::=::= Should we be here =::=::= */
if (stack == pcmk_cluster_corosync) {
set_daemon_option("cluster_type", "corosync");
set_daemon_option("quorum_type", "corosync");
} else {
crm_err("Unsupported stack type: %s", name_for_cluster_type(stack));
return FALSE;
}
/* =::=::= Logging =::=::= */
if (daemon_option("debug")) {
/* Syslog logging is already setup by crm_log_init() */
} else {
/* Check corosync */
char *debug_enabled = NULL;
get_config_opt(config, local_handle, "logging.debug", &debug_enabled, "off");
if (crm_is_true(debug_enabled)) {
set_daemon_option("debug", "1");
if (get_crm_log_level() < LOG_DEBUG) {
set_crm_log_level(LOG_DEBUG);
}
} else {
set_daemon_option("debug", "0");
}
free(debug_enabled);
}
if(local_handle){
gid_t gid = 0;
if (crm_user_lookup(CRM_DAEMON_USER, NULL, &gid) < 0) {
crm_warn("Could not authorize group with corosync " CRM_XS
" No group found for user %s", CRM_DAEMON_USER);
} else {
char key[PATH_MAX];
snprintf(key, PATH_MAX, "uidgid.gid.%u", gid);
rc = cmap_set_uint8(local_handle, key, 1);
if (rc != CS_OK) {
crm_warn("Could not authorize group with corosync "CRM_XS
" group=%u rc=%d (%s)", gid, rc, ais_error2text(rc));
}
}
}
cmap_finalize(local_handle);
return TRUE;
}