mainloop_io_t *
mainloop_add_ipc_client(const char *name, int priority, size_t max_size, void *userdata,
struct ipc_client_callbacks *callbacks)
{
mainloop_io_t *client = NULL;
crm_ipc_t *conn = crm_ipc_new(name, max_size);
if (conn && crm_ipc_connect(conn)) {
int32_t fd = crm_ipc_get_fd(conn);
client = mainloop_add_fd(name, priority, fd, userdata, NULL);
}
if (client == NULL) {
crm_perror(LOG_TRACE, "Connection to %s failed", name);
if (conn) {
crm_ipc_close(conn);
crm_ipc_destroy(conn);
}
return NULL;
}
client->ipc = conn;
client->destroy_fn = callbacks->destroy;
client->dispatch_fn_ipc = callbacks->dispatch;
return client;
}
/*!
* \internal
* \brief Send an operation to pacemaker-attrd via IPC
*
* \param[in] ipc Connection to pacemaker-attrd (or NULL to use a local connection)
* \param[in] attrd_op XML of pacemaker-attrd operation to send
*
* \return pcmk_ok on success, -errno otherwise
*/
static int
send_attrd_op(crm_ipc_t *ipc, xmlNode *attrd_op)
{
int rc = -ENOTCONN;
int max = 5;
static gboolean connected = TRUE;
static crm_ipc_t *local_ipc = NULL;
static enum crm_ipc_flags flags = crm_ipc_flags_none;
if (ipc == NULL && local_ipc == NULL) {
local_ipc = crm_ipc_new(T_ATTRD, 0);
flags |= crm_ipc_client_response;
connected = FALSE;
}
if (ipc == NULL) {
ipc = local_ipc;
}
while (max > 0) {
if (connected == FALSE) {
crm_info("Connecting to cluster... %d retries remaining", max);
connected = crm_ipc_connect(ipc);
}
if (connected) {
rc = crm_ipc_send(ipc, attrd_op, flags, 0, NULL);
} else {
crm_perror(LOG_INFO, "Connection to cluster attribute manager failed");
}
if (ipc != local_ipc) {
break;
} else if (rc > 0) {
break;
} else if (rc == -EAGAIN || rc == -EALREADY) {
sleep(5 - max);
max--;
} else {
crm_ipc_close(ipc);
connected = FALSE;
sleep(5 - max);
max--;
}
}
if (rc > 0) {
rc = pcmk_ok;
}
return rc;
}
int
main(int argc, char **argv)
{
int flag;
int argerr = 0;
gboolean allow_cores = TRUE;
crm_ipc_t *old_instance = NULL;
crm_system_name = CRM_SYSTEM_PENGINE;
mainloop_add_signal(SIGTERM, pengine_shutdown);
while ((flag = getopt(argc, argv, OPTARGS)) != EOF) {
switch (flag) {
case 'V':
crm_bump_log_level();
break;
case 'h': /* Help message */
usage(crm_system_name, EX_OK);
break;
case 'c':
allow_cores = TRUE;
break;
default:
++argerr;
break;
}
}
if (argc - optind == 1 && safe_str_eq("metadata", argv[optind])) {
pe_metadata();
return 0;
}
if (optind > argc) {
++argerr;
}
if (argerr) {
usage(crm_system_name, EX_USAGE);
}
crm_log_init(NULL, LOG_NOTICE, TRUE, FALSE, argc, argv, FALSE);
if (crm_is_writable(PE_STATE_DIR, NULL, CRM_DAEMON_USER, CRM_DAEMON_GROUP, FALSE) == FALSE) {
crm_err("Bad permissions on " PE_STATE_DIR ". Terminating");
fprintf(stderr, "ERROR: Bad permissions on " PE_STATE_DIR ". See logs for details\n");
fflush(stderr);
return 100;
}
/* find any previous instances and shut them down */
crm_debug("Checking for old instances of %s", CRM_SYSTEM_PENGINE);
old_instance = crm_ipc_new(CRM_SYSTEM_PENGINE, 0);
crm_ipc_connect(old_instance);
crm_debug("Terminating previous instance");
while (crm_ipc_connected(old_instance)) {
xmlNode *cmd = create_request(CRM_OP_QUIT, NULL, NULL, CRM_SYSTEM_PENGINE, CRM_SYSTEM_PENGINE, NULL);
crm_debug(".");
crm_ipc_send(old_instance, cmd, NULL, 0);
free_xml(cmd);
sleep(2);
}
crm_ipc_close(old_instance);
crm_ipc_destroy(old_instance);
crm_debug("Init server comms");
ipcs = mainloop_add_ipc_server(CRM_SYSTEM_PENGINE, QB_IPC_SHM, &ipc_callbacks);
if (ipcs == NULL) {
crm_err("Couldn't start IPC server");
return 1;
}
/* Create the mainloop and run it... */
crm_info("Starting %s", crm_system_name);
mainloop = g_main_new(FALSE);
g_main_run(mainloop);
#if HAVE_LIBXML2
crm_xml_cleanup();
#endif
crm_info("Exiting %s", crm_system_name);
qb_log_fini();
return 0;
}
int
main(int argc, char **argv)
{
int flag;
int index = 0;
int argerr = 0;
crm_ipc_t *old_instance = NULL;
crmd_mainloop = g_main_loop_new(NULL, FALSE);
crm_log_preinit(NULL, argc, argv);
crm_set_options(NULL, "[options]", long_options,
"Daemon for aggregating resource and node failures as well as co-ordinating the cluster's response");
while (1) {
flag = crm_get_option(argc, argv, &index);
if (flag == -1)
break;
switch (flag) {
case 'V':
crm_bump_log_level(argc, argv);
break;
case 'h': /* Help message */
crm_help(flag, CRM_EX_OK);
break;
default:
++argerr;
break;
}
}
if (argc - optind == 1 && safe_str_eq("metadata", argv[optind])) {
crmd_metadata();
return CRM_EX_OK;
} else if (argc - optind == 1 && safe_str_eq("version", argv[optind])) {
fprintf(stdout, "CRM Version: %s (%s)\n", PACEMAKER_VERSION, BUILD_VERSION);
return CRM_EX_OK;
}
crm_log_init(NULL, LOG_INFO, TRUE, FALSE, argc, argv, FALSE);
if (optind > argc) {
++argerr;
}
if (argerr) {
crm_help('?', CRM_EX_USAGE);
}
crm_notice("Starting Pacemaker controller");
old_instance = crm_ipc_new(CRM_SYSTEM_CRMD, 0);
if (crm_ipc_connect(old_instance)) {
/* IPC end-point already up */
crm_ipc_close(old_instance);
crm_ipc_destroy(old_instance);
crm_err("pacemaker-controld is already active, aborting startup");
crm_exit(CRM_EX_OK);
} else {
/* not up or not authentic, we'll proceed either way */
crm_ipc_destroy(old_instance);
old_instance = NULL;
}
if (pcmk__daemon_can_write(PE_STATE_DIR, NULL) == FALSE) {
crm_err("Terminating due to bad permissions on " PE_STATE_DIR);
fprintf(stderr,
"ERROR: Bad permissions on " PE_STATE_DIR " (see logs for details)\n");
fflush(stderr);
return CRM_EX_FATAL;
} else if (pcmk__daemon_can_write(CRM_CONFIG_DIR, NULL) == FALSE) {
crm_err("Terminating due to bad permissions on " CRM_CONFIG_DIR);
fprintf(stderr,
"ERROR: Bad permissions on " CRM_CONFIG_DIR " (see logs for details)\n");
fflush(stderr);
return CRM_EX_FATAL;
}
crmd_init();
return 0; // not reachable
}
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);
}
int
cib_native_signon_raw(cib_t * cib, const char *name, enum cib_conn_type type, int *async_fd)
{
int rc = pcmk_ok;
const char *channel = NULL;
cib_native_opaque_t *native = cib->variant_opaque;
static struct ipc_client_callbacks cib_callbacks = {
.dispatch = cib_native_dispatch_internal,
.destroy = cib_native_destroy
};
cib->call_timeout = MAX_IPC_DELAY;
if (type == cib_command) {
cib->state = cib_connected_command;
channel = cib_channel_rw;
} else if (type == cib_command_nonblocking) {
cib->state = cib_connected_command;
channel = cib_channel_shm;
} else if (type == cib_query) {
cib->state = cib_connected_query;
channel = cib_channel_ro;
} else {
return -ENOTCONN;
}
crm_trace("Connecting %s channel", channel);
if (async_fd != NULL) {
native->ipc = crm_ipc_new(channel, 0);
if (native->ipc && crm_ipc_connect(native->ipc)) {
*async_fd = crm_ipc_get_fd(native->ipc);
} else if (native->ipc) {
rc = -ENOTCONN;
}
} else {
native->source =
mainloop_add_ipc_client(channel, G_PRIORITY_HIGH, 512 * 1024 /* 512k */ , cib,
&cib_callbacks);
native->ipc = mainloop_get_ipc_client(native->source);
}
if (rc != pcmk_ok || native->ipc == NULL || crm_ipc_connected(native->ipc) == FALSE) {
crm_debug("Connection unsuccessful (%d %p)", rc, native->ipc);
rc = -ENOTCONN;
}
if (rc == pcmk_ok) {
xmlNode *reply = NULL;
xmlNode *hello = create_xml_node(NULL, "cib_command");
crm_xml_add(hello, F_TYPE, T_CIB);
crm_xml_add(hello, F_CIB_OPERATION, CRM_OP_REGISTER);
crm_xml_add(hello, F_CIB_CLIENTNAME, name);
crm_xml_add_int(hello, F_CIB_CALLOPTS, cib_sync_call);
if (crm_ipc_send(native->ipc, hello, crm_ipc_client_response, -1, &reply) > 0) {
const char *msg_type = crm_element_value(reply, F_CIB_OPERATION);
rc = pcmk_ok;
crm_log_xml_trace(reply, "reg-reply");
if (safe_str_neq(msg_type, CRM_OP_REGISTER)) {
crm_err("Invalid registration message: %s", msg_type);
rc = -EPROTO;
} else {
native->token = crm_element_value_copy(reply, F_CIB_CLIENTID);
if (native->token == NULL) {
rc = -EPROTO;
}
}
free_xml(reply);
} else {
rc = -ECOMM;
}
free_xml(hello);
}
if (rc == pcmk_ok) {
crm_debug("Connection to CIB successful");
return pcmk_ok;
}
crm_debug("Connection to CIB failed: %s", pcmk_strerror(rc));
cib_native_signoff(cib);
return rc;
}
int
main(int argc, char **argv)
{
int flag = 0;
int index = 0;
int argerr = 0;
crm_ipc_t *old_instance = NULL;
attrd_init_mainloop();
crm_log_preinit(NULL, argc, argv);
crm_set_options(NULL, "[options]", long_options,
"Daemon for aggregating and atomically storing node attribute updates into the CIB");
mainloop_add_signal(SIGTERM, attrd_shutdown);
while (1) {
flag = crm_get_option(argc, argv, &index);
if (flag == -1)
break;
switch (flag) {
case 'V':
crm_bump_log_level(argc, argv);
break;
case 'h': /* Help message */
crm_help(flag, CRM_EX_OK);
break;
default:
++argerr;
break;
}
}
if (optind > argc) {
++argerr;
}
if (argerr) {
crm_help('?', CRM_EX_USAGE);
}
crm_log_init(T_ATTRD, LOG_INFO, TRUE, FALSE, argc, argv, FALSE);
crm_notice("Starting Pacemaker node attribute manager");
old_instance = crm_ipc_new(T_ATTRD, 0);
if (crm_ipc_connect(old_instance)) {
/* IPC end-point already up */
crm_ipc_close(old_instance);
crm_ipc_destroy(old_instance);
crm_err("pacemaker-attrd is already active, aborting startup");
crm_exit(CRM_EX_OK);
} else {
/* not up or not authentic, we'll proceed either way */
crm_ipc_destroy(old_instance);
old_instance = NULL;
}
attributes = g_hash_table_new_full(crm_str_hash, g_str_equal, NULL, free_attribute);
/* Connect to the CIB before connecting to the cluster or listening for IPC.
* This allows us to assume the CIB is connected whenever we process a
* cluster or IPC message (which also avoids start-up race conditions).
*/
if (attrd_cib_connect(10) != pcmk_ok) {
attrd_exit_status = CRM_EX_FATAL;
goto done;
}
crm_info("CIB connection active");
if (attrd_cluster_connect() != pcmk_ok) {
attrd_exit_status = CRM_EX_FATAL;
goto done;
}
crm_info("Cluster connection active");
// Initialization that requires the cluster to be connected
attrd_election_init();
attrd_cib_init();
/* Set a private attribute for ourselves with the protocol version we
* support. This lets all nodes determine the minimum supported version
* across all nodes. It also ensures that the writer learns our node name,
* so it can send our attributes to the CIB.
*/
attrd_broadcast_protocol();
attrd_init_ipc(&ipcs, attrd_ipc_dispatch);
crm_notice("Pacemaker node attribute manager successfully started and accepting connections");
attrd_run_mainloop();
done:
crm_info("Shutting down attribute manager");
attrd_election_fini();
attrd_ipc_fini();
attrd_lrmd_disconnect();
attrd_cib_disconnect();
g_hash_table_destroy(attributes);
crm_exit(attrd_exit_status);
}
