gboolean
check_sbd_timeout(const char *value)
{
long st_timeout = value? crm_get_msec(value) : 0;
if (st_timeout <= 0) {
crm_debug("Watchdog may be enabled but stonith-watchdog-timeout is disabled (%s)",
value? value : "default");
} else if (pcmk_locate_sbd() == 0) {
do_crm_log_always(LOG_EMERG,
"Shutting down: stonith-watchdog-timeout configured (%s) but SBD not active",
value);
crm_exit(DAEMON_RESPAWN_STOP);
return FALSE;
} else {
long sbd_timeout = crm_get_sbd_timeout();
if (st_timeout < sbd_timeout) {
do_crm_log_always(LOG_EMERG,
"Shutting down: stonith-watchdog-timeout (%s) too short (must be >%ldms)",
value, sbd_timeout);
crm_exit(DAEMON_RESPAWN_STOP);
return FALSE;
}
crm_info("Watchdog configured with stonith-watchdog-timeout %s and SBD timeout %ldms",
value, sbd_timeout);
}
return TRUE;
}
void
pcmk_panic(const char *origin)
{
static struct qb_log_callsite *panic_cs = NULL;
if (panic_cs == NULL) {
panic_cs = qb_log_callsite_get(__func__, __FILE__, "panic-delay", LOG_TRACE, __LINE__, crm_trace_nonlog);
}
/* Ensure sbd_pid is set */
(void)pcmk_locate_sbd();
if (panic_cs && panic_cs->targets) {
/* getppid() == 1 means our original parent no longer exists */
do_crm_log_always(LOG_EMERG,
"Shutting down instead of panicking the node: origin=%s, sbd=%d, parent=%d",
origin, sbd_pid, getppid());
crm_exit(DAEMON_RESPAWN_STOP);
return;
}
if(sbd_pid > 1) {
do_crm_log_always(LOG_EMERG, "Signaling sbd(%d) to panic the system: %s", sbd_pid, origin);
pcmk_panic_sbd();
} else {
do_crm_log_always(LOG_EMERG, "Panicking the system directly: %s", origin);
pcmk_panic_local();
}
}
gboolean
check_sbd_timeout(const char *value)
{
long sbd_timeout = crm_get_sbd_timeout();
long st_timeout = crm_get_msec(value);
if(value == NULL || st_timeout <= 0) {
crm_notice("Watchdog may be enabled but stonith-watchdog-timeout is disabled: %s", value);
} else if(pcmk_locate_sbd() == 0) {
do_crm_log_always(LOG_EMERG, "Shutting down: stonith-watchdog-timeout is configured (%ldms) but SBD is not active", st_timeout);
crm_exit(DAEMON_RESPAWN_STOP);
return FALSE;
} else if(st_timeout < sbd_timeout) {
do_crm_log_always(LOG_EMERG, "Shutting down: stonith-watchdog-timeout (%ldms) is too short (must be greater than %ldms)",
st_timeout, sbd_timeout);
crm_exit(DAEMON_RESPAWN_STOP);
return FALSE;
}
crm_info("Watchdog functionality is consistent: %s delay exceeds timeout of %ldms", value, sbd_timeout);
return TRUE;
}
int
main(int argc, char **argv)
{
int rc;
int flag;
int argerr = 0;
int option_index = 0;
gboolean shutdown = FALSE;
uid_t pcmk_uid = 0;
gid_t pcmk_gid = 0;
struct rlimit cores;
crm_ipc_t *old_instance = NULL;
qb_ipcs_service_t *ipcs = NULL;
const char *facility = daemon_option("logfacility");
static crm_cluster_t cluster;
crm_log_preinit(NULL, argc, argv);
crm_set_options(NULL, "mode [options]", long_options, "Start/Stop Pacemaker\n");
mainloop_add_signal(SIGHUP, pcmk_ignore);
mainloop_add_signal(SIGQUIT, pcmk_sigquit);
while (1) {
flag = crm_get_option(argc, argv, &option_index);
if (flag == -1)
break;
switch (flag) {
case 'V':
crm_bump_log_level(argc, argv);
break;
case 'f':
/* Legacy */
break;
case 'p':
pid_file = optarg;
break;
case '$':
case '?':
crm_help(flag, EX_OK);
break;
case 'S':
shutdown = TRUE;
break;
case 'F':
printf("Pacemaker %s (Build: %s)\n Supporting v%s: %s\n", PACEMAKER_VERSION, BUILD_VERSION,
CRM_FEATURE_SET, CRM_FEATURES);
crm_exit(pcmk_ok);
default:
printf("Argument code 0%o (%c) is not (?yet?) supported\n", flag, flag);
++argerr;
break;
}
}
if (optind < argc) {
printf("non-option ARGV-elements: ");
while (optind < argc)
printf("%s ", argv[optind++]);
printf("\n");
}
if (argerr) {
crm_help('?', EX_USAGE);
}
setenv("LC_ALL", "C", 1);
setenv("HA_LOGD", "no", 1);
set_daemon_option("mcp", "true");
set_daemon_option("use_logd", "off");
crm_log_init(NULL, LOG_INFO, TRUE, FALSE, argc, argv, FALSE);
/* Restore the original facility so that mcp_read_config() does the right thing */
set_daemon_option("logfacility", facility);
crm_debug("Checking for old instances of %s", CRM_SYSTEM_MCP);
old_instance = crm_ipc_new(CRM_SYSTEM_MCP, 0);
crm_ipc_connect(old_instance);
if (shutdown) {
crm_debug("Terminating previous instance");
while (crm_ipc_connected(old_instance)) {
xmlNode *cmd =
create_request(CRM_OP_QUIT, NULL, NULL, CRM_SYSTEM_MCP, CRM_SYSTEM_MCP, NULL);
crm_debug(".");
crm_ipc_send(old_instance, cmd, 0, 0, NULL);
free_xml(cmd);
sleep(2);
}
crm_ipc_close(old_instance);
crm_ipc_destroy(old_instance);
crm_exit(pcmk_ok);
} else if (crm_ipc_connected(old_instance)) {
crm_ipc_close(old_instance);
crm_ipc_destroy(old_instance);
crm_err("Pacemaker is already active, aborting startup");
crm_exit(DAEMON_RESPAWN_STOP);
}
crm_ipc_close(old_instance);
crm_ipc_destroy(old_instance);
if (mcp_read_config() == FALSE) {
crm_notice("Could not obtain corosync config data, exiting");
crm_exit(ENODATA);
}
crm_notice("Starting Pacemaker %s (Build: %s): %s", PACEMAKER_VERSION, BUILD_VERSION, CRM_FEATURES);
mainloop = g_main_new(FALSE);
sysrq_init();
rc = getrlimit(RLIMIT_CORE, &cores);
if (rc < 0) {
crm_perror(LOG_ERR, "Cannot determine current maximum core size.");
} else {
if (cores.rlim_max == 0 && geteuid() == 0) {
cores.rlim_max = RLIM_INFINITY;
} else {
crm_info("Maximum core file size is: %lu", (unsigned long)cores.rlim_max);
}
cores.rlim_cur = cores.rlim_max;
rc = setrlimit(RLIMIT_CORE, &cores);
if (rc < 0) {
crm_perror(LOG_ERR,
"Core file generation will remain disabled."
" Core files are an important diagnositic tool,"
" please consider enabling them by default.");
}
#if 0
/* system() is not thread-safe, can't call from here
* Actually, its a pretty hacky way to try and achieve this anyway
*/
if (system("echo 1 > /proc/sys/kernel/core_uses_pid") != 0) {
crm_perror(LOG_ERR, "Could not enable /proc/sys/kernel/core_uses_pid");
}
#endif
}
rc = pcmk_ok;
if (crm_user_lookup(CRM_DAEMON_USER, &pcmk_uid, &pcmk_gid) < 0) {
crm_err("Cluster user %s does not exist, aborting Pacemaker startup", CRM_DAEMON_USER);
crm_exit(ENOKEY);
}
mkdir(CRM_STATE_DIR, 0750);
mcp_chown(CRM_STATE_DIR, pcmk_uid, pcmk_gid);
/* Used to store core files in */
crm_build_path(CRM_CORE_DIR, 0775);
mcp_chown(CRM_CORE_DIR, pcmk_uid, pcmk_gid);
/* Used to store blackbox dumps in */
crm_build_path(CRM_BLACKBOX_DIR, 0755);
mcp_chown(CRM_BLACKBOX_DIR, pcmk_uid, pcmk_gid);
/* Used to store policy engine inputs in */
crm_build_path(PE_STATE_DIR, 0755);
mcp_chown(PE_STATE_DIR, pcmk_uid, pcmk_gid);
/* Used to store the cluster configuration */
crm_build_path(CRM_CONFIG_DIR, 0755);
mcp_chown(CRM_CONFIG_DIR, pcmk_uid, pcmk_gid);
/* Resource agent paths are constructed by the lrmd */
ipcs = mainloop_add_ipc_server(CRM_SYSTEM_MCP, QB_IPC_NATIVE, &mcp_ipc_callbacks);
if (ipcs == NULL) {
crm_err("Couldn't start IPC server");
crm_exit(EIO);
}
/* Allows us to block shutdown */
if (cluster_connect_cfg(&local_nodeid) == FALSE) {
crm_err("Couldn't connect to Corosync's CFG service");
crm_exit(ENOPROTOOPT);
}
if(pcmk_locate_sbd() > 0) {
setenv("PCMK_watchdog", "true", 1);
} else {
setenv("PCMK_watchdog", "false", 1);
}
find_and_track_existing_processes();
cluster.destroy = mcp_cpg_destroy;
cluster.cpg.cpg_deliver_fn = mcp_cpg_deliver;
cluster.cpg.cpg_confchg_fn = mcp_cpg_membership;
crm_set_autoreap(FALSE);
if(cluster_connect_cpg(&cluster) == FALSE) {
crm_err("Couldn't connect to Corosync's CPG service");
rc = -ENOPROTOOPT;
}
if (rc == pcmk_ok && is_corosync_cluster()) {
/* Keep the membership list up-to-date for crm_node to query */
if(cluster_connect_quorum(mcp_quorum_callback, mcp_quorum_destroy) == FALSE) {
rc = -ENOTCONN;
}
}
#if SUPPORT_CMAN
if (rc == pcmk_ok && is_cman_cluster()) {
init_cman_connection(mcp_cman_dispatch, mcp_cman_destroy);
}
#endif
if(rc == pcmk_ok) {
local_name = get_local_node_name();
update_node_processes(local_nodeid, local_name, get_process_list());
mainloop_add_signal(SIGTERM, pcmk_shutdown);
mainloop_add_signal(SIGINT, pcmk_shutdown);
init_children_processes();
crm_info("Starting mainloop");
g_main_run(mainloop);
}
if (ipcs) {
crm_trace("Closing IPC server");
mainloop_del_ipc_server(ipcs);
ipcs = NULL;
}
g_main_destroy(mainloop);
cluster_disconnect_cpg(&cluster);
cluster_disconnect_cfg();
crm_info("Exiting %s", crm_system_name);
return crm_exit(rc);
}
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);
}
