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);
}
}
static void
mcp_cpg_deliver(cpg_handle_t handle,
const struct cpg_name *groupName,
uint32_t nodeid, uint32_t pid, void *msg, size_t msg_len)
{
xmlNode *xml = string2xml(msg);
const char *task = crm_element_value(xml, F_CRM_TASK);
crm_trace("Received %s %.200s", task, msg);
if (task == NULL && nodeid != local_nodeid) {
uint32_t procs = 0;
const char *uname = crm_element_value(xml, "uname");
crm_element_value_int(xml, "proclist", (int *)&procs);
/* crm_debug("Got proclist %.32x from %s", procs, uname); */
if (update_node_processes(nodeid, uname, procs)) {
update_process_clients(NULL);
}
} else if (crm_str_eq(task, CRM_OP_RM_NODE_CACHE, TRUE)) {
int id = 0;
const char *name = NULL;
crm_element_value_int(xml, XML_ATTR_ID, &id);
name = crm_element_value(xml, XML_ATTR_UNAME);
reap_crm_member(id, name);
}
}
static void
pcmk_cpg_deliver(cpg_handle_t handle,
const struct cpg_name *groupName,
uint32_t nodeid, uint32_t pid, void *msg, size_t msg_len)
{
if (nodeid != local_nodeid) {
uint32_t procs = 0;
xmlNode *xml = string2xml(msg);
const char *uname = crm_element_value(xml, "uname");
crm_element_value_int(xml, "proclist", (int *)&procs);
/* crm_debug("Got proclist %.32x from %s", procs, uname); */
if (update_node_processes(nodeid, uname, procs)) {
update_process_clients();
}
}
}
/*!
* \internal
* \brief Process a CPG message (process list or manual peer cache removal)
*
* \param[in] handle CPG connection (ignored)
* \param[in] groupName CPG group name (ignored)
* \param[in] nodeid ID of affected node
* \param[in] pid Process ID (ignored)
* \param[in] msg CPG XML message
* \param[in] msg_len Length of msg in bytes (ignored)
*/
static void
mcp_cpg_deliver(cpg_handle_t handle,
const struct cpg_name *groupName,
uint32_t nodeid, uint32_t pid, void *msg, size_t msg_len)
{
xmlNode *xml = string2xml(msg);
const char *task = crm_element_value(xml, F_CRM_TASK);
crm_trace("Received CPG message (%s): %.200s",
(task? task : "process list"), msg);
if (task == NULL) {
if (nodeid == local_nodeid) {
crm_info("Ignoring process list sent by peer for local node");
} else {
uint32_t procs = 0;
const char *uname = crm_element_value(xml, "uname");
crm_element_value_int(xml, "proclist", (int *)&procs);
if (update_node_processes(nodeid, uname, procs)) {
update_process_clients(NULL);
}
}
} else if (crm_str_eq(task, CRM_OP_RM_NODE_CACHE, TRUE)) {
int id = 0;
const char *name = NULL;
crm_element_value_int(xml, XML_ATTR_ID, &id);
name = crm_element_value(xml, XML_ATTR_UNAME);
reap_crm_member(id, name);
}
if (xml != NULL) {
free_xml(xml);
}
}
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 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 */
}
