diff --git a/daemons/pacemakerd/pacemakerd.c b/daemons/pacemakerd/pacemakerd.c index 2aa1090149..642f5c4f1c 100644 --- a/daemons/pacemakerd/pacemakerd.c +++ b/daemons/pacemakerd/pacemakerd.c @@ -1,1442 +1,1445 @@ /* * Copyright 2010-2020 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include "pacemakerd.h" #include #include #include #include #include #include #include #include #include #include #include /* indirectly: CRM_EX_* */ #include /* cib_channel_ro */ #include #include #include #include #include #include /* PCMK__SPECIAL_PID*, ... */ #ifdef SUPPORT_COROSYNC #include #endif #include #include static gboolean pcmk_quorate = FALSE; static gboolean fatal_error = FALSE; static GMainLoop *mainloop = NULL; static bool global_keep_tracking = false; #define PCMK_PROCESS_CHECK_INTERVAL 5 static const char *local_name = NULL; static uint32_t local_nodeid = 0; static crm_trigger_t *shutdown_trigger = NULL; static const char *pid_file = PCMK_RUN_DIR "/pacemaker.pid"; typedef struct pcmk_child_s { - int pid; + pid_t pid; long flag; int start_seq; int respawn_count; gboolean respawn; const char *name; const char *uid; const char *command; const char *endpoint; /* IPC server name */ gboolean active_before_startup; } pcmk_child_t; /* Index into the array below */ #define PCMK_CHILD_CONTROLD 3 static pcmk_child_t pcmk_children[] = { { 0, crm_proc_none, 0, 0, FALSE, "none", NULL, NULL }, { 0, crm_proc_execd, 3, 0, TRUE, "pacemaker-execd", NULL, CRM_DAEMON_DIR "/pacemaker-execd", CRM_SYSTEM_LRMD }, { 0, crm_proc_based, 1, 0, TRUE, "pacemaker-based", CRM_DAEMON_USER, CRM_DAEMON_DIR "/pacemaker-based", CIB_CHANNEL_RO }, { 0, crm_proc_controld, 6, 0, TRUE, "pacemaker-controld", CRM_DAEMON_USER, CRM_DAEMON_DIR "/pacemaker-controld", CRM_SYSTEM_CRMD }, { 0, crm_proc_attrd, 4, 0, TRUE, "pacemaker-attrd", CRM_DAEMON_USER, CRM_DAEMON_DIR "/pacemaker-attrd", T_ATTRD }, { 0, crm_proc_schedulerd, 5, 0, TRUE, "pacemaker-schedulerd", CRM_DAEMON_USER, CRM_DAEMON_DIR "/pacemaker-schedulerd", CRM_SYSTEM_PENGINE }, { 0, crm_proc_fenced, 2, 0, TRUE, "pacemaker-fenced", NULL, CRM_DAEMON_DIR "/pacemaker-fenced", "stonith-ng" }, }; static gboolean check_active_before_startup_processes(gpointer user_data); static int child_liveness(pcmk_child_t *child); static gboolean start_child(pcmk_child_t * child); static gboolean update_node_processes(uint32_t id, const char *uname, uint32_t procs); void update_process_clients(pcmk__client_t *client); static uint32_t get_process_list(void) { int lpc = 0; uint32_t procs = crm_get_cluster_proc(); for (lpc = 0; lpc < SIZEOF(pcmk_children); lpc++) { if (pcmk_children[lpc].pid != 0) { procs |= pcmk_children[lpc].flag; } } return procs; } 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) { /* resume step-wise shutdown (returned TRUE yields no parallelizing) */ mainloop_set_trigger(shutdown_trigger); /* intended to speed up propagating expected lay-off of the daemons? */ update_node_processes(local_nodeid, NULL, get_process_list()); } else if (!child->respawn) { /* nothing to do */ } else if (crm_is_true(getenv("PCMK_fail_fast"))) { crm_err("Rebooting system because of %s", child->name); pcmk_panic(__FUNCTION__); } else if (child_liveness(child) == pcmk_rc_ok) { crm_warn("One-off suppressing strict respawning of a child process %s," " appears alright per %s IPC end-point", child->name, child->endpoint); /* need to monitor how it evolves, and start new process if badly */ child->active_before_startup = TRUE; if (!global_keep_tracking) { global_keep_tracking = true; g_timeout_add_seconds(PCMK_PROCESS_CHECK_INTERVAL, check_active_before_startup_processes, NULL); } } else { crm_notice("Respawning failed child process: %s", child->name); start_child(child); } } static void pcmk_exit_with_cluster(int exitcode) { #ifdef SUPPORT_COROSYNC corosync_cfg_handle_t cfg_handle; cs_error_t err; if (exitcode == CRM_EX_FATAL) { crm_info("Asking Corosync to shut down"); err = corosync_cfg_initialize(&cfg_handle, NULL); if (err != CS_OK) { crm_warn("Unable to open handle to corosync to close it down. err=%d", err); } err = corosync_cfg_try_shutdown(cfg_handle, COROSYNC_CFG_SHUTDOWN_FLAG_IMMEDIATE); if (err != CS_OK) { crm_warn("Corosync shutdown failed. err=%d", err); } corosync_cfg_finalize(cfg_handle); } #endif crm_exit(exitcode); } static void pcmk_child_exit(mainloop_child_t * p, pid_t pid, int core, int signo, int exitcode) { pcmk_child_t *child = mainloop_child_userdata(p); const char *name = mainloop_child_name(p); if (signo) { do_crm_log(((signo == SIGKILL)? LOG_WARNING : LOG_ERR), "%s[%d] terminated with signal %d (core=%d)", name, pid, signo, core); } else { switch(exitcode) { case CRM_EX_OK: crm_info("%s[%d] exited with status %d (%s)", name, pid, exitcode, crm_exit_str(exitcode)); break; case CRM_EX_FATAL: crm_warn("Shutting cluster down because %s[%d] had fatal failure", name, pid); child->respawn = FALSE; fatal_error = TRUE; pcmk_shutdown(SIGTERM); break; case CRM_EX_PANIC: do_crm_log_always(LOG_EMERG, "%s[%d] instructed the machine to reset", name, pid); child->respawn = FALSE; fatal_error = TRUE; pcmk_panic(__FUNCTION__); pcmk_shutdown(SIGTERM); break; default: crm_err("%s[%d] exited with status %d (%s)", name, pid, exitcode, crm_exit_str(exitcode)); break; } } pcmk_process_exit(child); } static gboolean stop_child(pcmk_child_t * child, int signal) { if (signal == 0) { signal = SIGTERM; } /* why to skip PID of 1? - FreeBSD ~ how untrackable process behind IPC is masqueraded as - elsewhere: how "init" task is designated; in particular, in systemd arrangement of socket-based activation, this is pretty real */ if (child->command == NULL || child->pid == PCMK__SPECIAL_PID) { crm_debug("Nothing to do for child \"%s\" (process %lld)", child->name, (long long) PCMK__SPECIAL_PID_AS_0(child->pid)); return TRUE; } if (child->pid <= 0) { crm_trace("Client %s not running", child->name); return TRUE; } errno = 0; if (kill(child->pid, signal) == 0) { - crm_notice("Stopping %s "CRM_XS" sent signal %d to process %d", - child->name, signal, child->pid); + crm_notice("Stopping %s "CRM_XS" sent signal %d to process %lld", + child->name, signal, (long long) child->pid); } else { - crm_perror(LOG_ERR, "Could not stop %s (process %d) with signal %d", - child->name, child->pid, signal); + crm_err("Could not stop %s (process %lld) with signal %d: %s", + child->name, (long long) child->pid, signal, strerror(errno)); } return TRUE; } static char *opts_default[] = { NULL, NULL }; static char *opts_vgrind[] = { NULL, NULL, NULL, NULL, NULL }; /* TODO once libqb is taught to juggle with IPC end-points carried over as bare file descriptor (https://github.com/ClusterLabs/libqb/issues/325) it shall hand over these descriptors here if/once they are successfully pre-opened in (presumably) child_liveness(), to avoid any remaining room for races */ static gboolean start_child(pcmk_child_t * child) { uid_t uid = 0; gid_t gid = 0; 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"); 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, + crm_info("Forked child %lld for process %s%s", + (long long) 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 arrays */ 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) { // Whether we need root group access to talk to cluster layer bool need_root_group = TRUE; if (is_corosync_cluster()) { /* Corosync clusters can drop root group access, because we set * uidgid.gid.${gid}=1 via CMAP, which allows these processes to * connect to corosync. */ need_root_group = FALSE; } // Drop root group access if not needed if (!need_root_group && (setgid(gid) < 0)) { crm_perror(LOG_ERR, "Could not set group to %d", gid); } /* Initialize supplementary groups to only those always granted to * the user, plus haclient (so we can access IPC). */ 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); } pcmk__close_fds_in_child(true); (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(CRM_EX_FATAL); } return TRUE; /* never reached */ } static gboolean escalate_shutdown(gpointer data) { pcmk_child_t *child = data; if (child->pid == PCMK__SPECIAL_PID) { pcmk_process_exit(child); - } else if (child->pid) { + } else if (child->pid != 0) { /* Use SIGSEGV instead of SIGKILL to create a core so we can see what it was up to */ crm_err("Child %s not terminating in a timely manner, forcing", child->name); stop_child(child, SIGSEGV); } return FALSE; } #define SHUTDOWN_ESCALATION_PERIOD 180000 /* 3m */ static gboolean pcmk_shutdown_worker(gpointer user_data) { static int phase = 0; static time_t next_log = 0; static int max = SIZEOF(pcmk_children); int lpc = 0; if (phase == 0) { crm_notice("Shutting down Pacemaker"); phase = max; } for (; phase > 0; phase--) { /* Don't stop anything with start_seq < 1 */ for (lpc = max - 1; lpc >= 0; lpc--) { pcmk_child_t *child = &(pcmk_children[lpc]); if (phase != child->start_seq) { continue; } - if (child->pid) { + if (child->pid != 0) { time_t now = time(NULL); if (child->respawn) { if (child->pid == PCMK__SPECIAL_PID) { crm_warn("The process behind %s IPC cannot be" " terminated, so either wait the graceful" " period of %ld s for its native termination" " if it vitally depends on some other daemons" " going down in a controlled way already," " or locate and kill the correct %s process" " on your own; set PCMK_fail_fast=1 to avoid" " this altogether next time around", child->name, (long) SHUTDOWN_ESCALATION_PERIOD, child->command); } next_log = now + 30; child->respawn = FALSE; stop_child(child, SIGTERM); if (phase < pcmk_children[PCMK_CHILD_CONTROLD].start_seq) { g_timeout_add(SHUTDOWN_ESCALATION_PERIOD, escalate_shutdown, child); } } else if (now >= next_log) { next_log = now + 30; crm_notice("Still waiting for %s to terminate " - CRM_XS " pid=%d seq=%d", - child->name, child->pid, child->start_seq); + CRM_XS " pid=%lld seq=%d", + child->name, (long long) child->pid, + child->start_seq); } return TRUE; } /* cleanup */ crm_debug("%s confirmed stopped", child->name); child->pid = 0; } } /* send_cluster_id(); */ crm_notice("Shutdown complete"); { const char *delay = pcmk__env_option("shutdown_delay"); if(delay) { sync(); sleep(crm_get_msec(delay) / 1000); } } g_main_loop_quit(mainloop); if (fatal_error) { crm_notice("Shutting down and staying down after fatal error"); pcmk_exit_with_cluster(CRM_EX_FATAL); } return TRUE; } static void pcmk_ignore(int nsig) { crm_info("Ignoring signal %s (%d)", strsignal(nsig), nsig); } static void pcmk_sigquit(int nsig) { pcmk_panic(__FUNCTION__); } void pcmk_shutdown(int nsig) { if (shutdown_trigger == NULL) { shutdown_trigger = mainloop_add_trigger(G_PRIORITY_HIGH, pcmk_shutdown_worker, NULL); } mainloop_set_trigger(shutdown_trigger); } static int32_t pcmk_ipc_accept(qb_ipcs_connection_t * c, uid_t uid, gid_t gid) { crm_trace("Connection %p", c); if (pcmk__new_client(c, uid, gid) == NULL) { return -EIO; } return 0; } static void pcmk_ipc_created(qb_ipcs_connection_t * c) { crm_trace("Connection %p", c); } /* Exit code means? */ static int32_t pcmk_ipc_dispatch(qb_ipcs_connection_t * qbc, void *data, size_t size) { uint32_t id = 0; uint32_t flags = 0; const char *task = NULL; pcmk__client_t *c = pcmk__find_client(qbc); xmlNode *msg = pcmk__client_data2xml(c, data, size, &id, &flags); pcmk__ipc_send_ack(c, id, flags, "ack"); if (msg == NULL) { return 0; } task = crm_element_value(msg, F_CRM_TASK); if (crm_str_eq(task, CRM_OP_QUIT, TRUE)) { /* Time to quit */ crm_notice("Shutting down in response to ticket %s (%s)", crm_element_value(msg, F_CRM_REFERENCE), crm_element_value(msg, F_CRM_ORIGIN)); pcmk_shutdown(15); } else if (crm_str_eq(task, CRM_OP_RM_NODE_CACHE, TRUE)) { /* Send to everyone */ struct iovec *iov; int id = 0; const char *name = NULL; crm_element_value_int(msg, XML_ATTR_ID, &id); name = crm_element_value(msg, XML_ATTR_UNAME); crm_notice("Instructing peers to remove references to node %s/%u", name, id); iov = calloc(1, sizeof(struct iovec)); iov->iov_base = dump_xml_unformatted(msg); iov->iov_len = 1 + strlen(iov->iov_base); send_cpg_iov(iov); } else { update_process_clients(c); } free_xml(msg); return 0; } /* Error code means? */ static int32_t pcmk_ipc_closed(qb_ipcs_connection_t * c) { pcmk__client_t *client = pcmk__find_client(c); if (client == NULL) { return 0; } crm_trace("Connection %p", c); pcmk__free_client(client); return 0; } static void pcmk_ipc_destroy(qb_ipcs_connection_t * c) { crm_trace("Connection %p", c); pcmk_ipc_closed(c); } struct qb_ipcs_service_handlers mcp_ipc_callbacks = { .connection_accept = pcmk_ipc_accept, .connection_created = pcmk_ipc_created, .msg_process = pcmk_ipc_dispatch, .connection_closed = pcmk_ipc_closed, .connection_destroyed = pcmk_ipc_destroy }; static void send_xml_to_client(gpointer key, gpointer value, gpointer user_data) { pcmk__ipc_send_xml((pcmk__client_t *) value, 0, (xmlNode *) user_data, crm_ipc_server_event); } /*! * \internal * \brief Send an XML message with process list of all known peers to client(s) * * \param[in] client Send message to this client, or all clients if NULL */ void update_process_clients(pcmk__client_t *client) { GHashTableIter iter; crm_node_t *node = NULL; xmlNode *update = create_xml_node(NULL, "nodes"); if (is_corosync_cluster()) { crm_xml_add_int(update, "quorate", pcmk_quorate); } g_hash_table_iter_init(&iter, crm_peer_cache); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) & node)) { xmlNode *xml = create_xml_node(update, "node"); crm_xml_add_int(xml, "id", node->id); crm_xml_add(xml, "uname", node->uname); crm_xml_add(xml, "state", node->state); crm_xml_add_int(xml, "processes", node->processes); } if(client) { crm_trace("Sending process list to client %s", client->id); send_xml_to_client(NULL, client, update); } else { crm_trace("Sending process list to %d clients", pcmk__ipc_client_count()); pcmk__foreach_ipc_client(send_xml_to_client, update); } free_xml(update); } /*! * \internal * \brief Send a CPG message with local node's process list to all peers */ static void update_process_peers(void) { /* Do nothing for corosync-2 based clusters */ struct iovec *iov = calloc(1, sizeof(struct iovec)); CRM_ASSERT(iov); if (local_name) { iov->iov_base = crm_strdup_printf("", local_name, get_process_list()); } else { iov->iov_base = crm_strdup_printf("", get_process_list()); } iov->iov_len = strlen(iov->iov_base) + 1; crm_trace("Sending %s", (char*) iov->iov_base); send_cpg_iov(iov); } /*! * \internal * \brief Update a node's process list, notifying clients and peers if needed * * \param[in] id Node ID of affected node * \param[in] uname Uname of affected node * \param[in] procs Affected node's process list mask * * \return TRUE if the process list changed, FALSE otherwise */ static gboolean update_node_processes(uint32_t id, const char *uname, uint32_t procs) { gboolean changed = FALSE; crm_node_t *node = crm_get_peer(id, uname); if (procs != 0) { if (procs != node->processes) { crm_debug("Node %s now has process list: %.32x (was %.32x)", node->uname, procs, node->processes); node->processes = procs; changed = TRUE; /* If local node's processes have changed, notify clients/peers */ if (id == local_nodeid) { update_process_clients(NULL); update_process_peers(); } } else { crm_trace("Node %s still has process list: %.32x", node->uname, procs); } } return changed; } /* *INDENT-OFF* */ static struct crm_option long_options[] = { /* Top-level Options */ {"help", 0, 0, '?', "\tThis text"}, {"version", 0, 0, '$', "\tVersion information" }, {"verbose", 0, 0, 'V', "\tIncrease debug output"}, {"shutdown", 0, 0, 'S', "\tInstruct Pacemaker to shutdown on this machine"}, {"features", 0, 0, 'F', "\tDisplay the full version and list of features Pacemaker was built with"}, {"-spacer-", 1, 0, '-', "\nAdditional Options:"}, {"foreground", 0, 0, 'f', "\t(Ignored) Pacemaker always runs in the foreground"}, {"pid-file", 1, 0, 'p', "\t(Ignored) Daemon pid file location"}, {"standby", 0, 0, 's', "\tStart node in standby state"}, {NULL, 0, 0, 0} }; /* *INDENT-ON* */ static void mcp_chown(const char *path, uid_t uid, gid_t gid) { int rc = chown(path, uid, gid); if (rc < 0) { crm_warn("Cannot change the ownership of %s to user %s and gid %d: %s", path, CRM_DAEMON_USER, gid, pcmk_strerror(errno)); } } /*! * \internal * \brief Check the liveness of the child based on IPC name and PID if tracked * * \param[inout] child Child tracked data * * \return Standard Pacemaker return code * * \note Return codes of particular interest include pcmk_rc_ipc_unresponsive * indicating that no trace of IPC liveness was detected, * pcmk_rc_ipc_unauthorized indicating that the IPC endpoint is blocked by * an unauthorized process, and pcmk_rc_ipc_pid_only indicating that * the child is up by PID but not IPC end-point (possibly starting). * \note This function doesn't modify any of \p child members but \c pid, * and is not actively toying with processes as such but invoking * \c stop_child in one particular case (there's for some reason * a different authentic holder of the IPC end-point). */ static int -child_liveness(pcmk_child_t *child) { +child_liveness(pcmk_child_t *child) +{ uid_t cl_uid = 0; gid_t cl_gid = 0; const uid_t root_uid = 0; const gid_t root_gid = 0; const uid_t *ref_uid; const gid_t *ref_gid; int rc = pcmk_rc_ipc_unresponsive; pid_t ipc_pid = 0; if (child->endpoint == NULL && (child->pid <= 0 || child->pid == PCMK__SPECIAL_PID)) { crm_err("Cannot track child %s for missing both API end-point and PID", child->name); rc = EINVAL; // Misuse of function when child is not trackable } else if (child->endpoint != NULL) { int legacy_rc = pcmk_ok; if (child->uid == NULL) { ref_uid = &root_uid; ref_gid = &root_gid; } else { ref_uid = &cl_uid; ref_gid = &cl_gid; legacy_rc = pcmk_daemon_user(&cl_uid, &cl_gid); } if (legacy_rc < 0) { rc = pcmk_legacy2rc(legacy_rc); crm_err("Could not find user and group IDs for user %s: %s " CRM_XS " rc=%d", CRM_DAEMON_USER, pcmk_rc_str(rc), rc); } else { rc = pcmk__ipc_is_authentic_process_active(child->endpoint, *ref_uid, *ref_gid, &ipc_pid); if ((rc == pcmk_rc_ok) || (rc == pcmk_rc_ipc_unresponsive)) { if (child->pid <= 0) { /* If rc is pcmk_rc_ok, ipc_pid is nonzero and this * initializes a new child. If rc is * pcmk_rc_ipc_unresponsive, ipc_pid is zero, and we will * investigate further. */ child->pid = ipc_pid; } else if ((ipc_pid != 0) && (child->pid != ipc_pid)) { /* An unexpected (but authorized) process is responding to * IPC. Investigate further. */ rc = pcmk_rc_ipc_unresponsive; } } } } if (rc == pcmk_rc_ipc_unresponsive) { /* If we get here, a child without IPC is being tracked, no IPC liveness * has been detected, or IPC liveness has been detected with an * unexpected (but authorized) process. This is safe on FreeBSD since * the only change possible from a proper child's PID into "special" PID * of 1 behind more loosely related process. */ - int ret = crm_pid_active(child->pid, child->name); + int ret = pcmk__pid_active(child->pid, child->name); - if (ipc_pid && (ret != 1 + if (ipc_pid && ((ret != pcmk_rc_ok) || ipc_pid == PCMK__SPECIAL_PID - || crm_pid_active(ipc_pid, child->name) == 1)) { + || (pcmk__pid_active(ipc_pid, + child->name) == pcmk_rc_ok))) { /* An unexpected (but authorized) process was detected at the IPC * endpoint, and either it is active, or the child we're tracking is * not. */ - if (ret == 1) { + if (ret == pcmk_rc_ok) { /* The child we're tracking is active. Kill it, and adopt the * detected process. This assumes that our children don't fork * (thus getting a different PID owning the IPC), but rather the * tracking got out of sync because of some means external to * Pacemaker, and adopting the detected process is better than * killing it and possibly having to spawn a new child. */ /* not possessing IPC, afterall (what about corosync CPG?) */ stop_child(child, SIGKILL); } rc = pcmk_rc_ok; child->pid = ipc_pid; - } else if (ret == 1) { + } else if (ret == pcmk_rc_ok) { // Our tracked child's PID was found active, but not its IPC rc = pcmk_rc_ipc_pid_only; - } else if (!child->pid && ret == -1) { - // FreeBSD can return -1 + } else if ((child->pid == 0) && (ret == EINVAL)) { + // FreeBSD can return EINVAL rc = pcmk_rc_ipc_unresponsive; } else { - // Map crm_pid_active() result to standard rc switch (ret) { - case -1: - rc = EINVAL; - break; - case -2: + case EACCES: rc = pcmk_rc_ipc_unauthorized; break; - case 0: + case ESRCH: rc = pcmk_rc_ipc_unresponsive; break; + default: + rc = ret; + break; } } } return rc; } static gboolean check_active_before_startup_processes(gpointer user_data) { int start_seq = 1, lpc = 0; static int max = SIZEOF(pcmk_children); gboolean keep_tracking = FALSE; for (start_seq = 1; start_seq < max; start_seq++) { for (lpc = 0; lpc < max; lpc++) { if (pcmk_children[lpc].active_before_startup == FALSE) { /* we are already tracking it as a child process. */ continue; } else if (start_seq != pcmk_children[lpc].start_seq) { continue; } else { int rc = child_liveness(&pcmk_children[lpc]); switch (rc) { case pcmk_rc_ok: break; case pcmk_rc_ipc_unresponsive: case pcmk_rc_ipc_pid_only: // This case: it was previously OK if (pcmk_children[lpc].respawn == TRUE) { - crm_err("%s[%d] terminated%s", pcmk_children[lpc].name, - PCMK__SPECIAL_PID_AS_0(pcmk_children[lpc].pid), + crm_err("%s[%lld] terminated%s", pcmk_children[lpc].name, + (long long) PCMK__SPECIAL_PID_AS_0(pcmk_children[lpc].pid), (rc == pcmk_rc_ipc_pid_only)? " as IPC server" : ""); } else { /* orderly shutdown */ - crm_notice("%s[%d] terminated%s", pcmk_children[lpc].name, - PCMK__SPECIAL_PID_AS_0(pcmk_children[lpc].pid), + crm_notice("%s[%lld] terminated%s", pcmk_children[lpc].name, + (long long) PCMK__SPECIAL_PID_AS_0(pcmk_children[lpc].pid), (rc == pcmk_rc_ipc_pid_only)? " as IPC server" : ""); } pcmk_process_exit(&(pcmk_children[lpc])); continue; default: crm_exit(CRM_EX_FATAL); break; /* static analysis/noreturn */ } } /* at least one of the processes found at startup * is still going, so keep this recurring timer around */ keep_tracking = TRUE; } } global_keep_tracking = keep_tracking; return keep_tracking; } /*! * \internal * \brief Initial one-off check of the pre-existing "child" processes * * With "child" process, we mean the subdaemon that defines an API end-point * (all of them do as of the comment) -- the possible complement is skipped * as it is deemed it has no such shared resources to cause conflicts about, * hence it can presumably be started anew without hesitation. * If that won't hold true in the future, the concept of a shared resource * will have to be generalized beyond the API end-point. * * For boundary cases that the "child" is still starting (IPC end-point is yet * to be witnessed), or more rarely (practically FreeBSD only), when there's * a pre-existing "untrackable" authentic process, we give the situation some * time to possibly unfold in the right direction, meaning that said socket * will appear or the unattainable process will disappear per the observable * IPC, respectively. * * \return Standard Pacemaker return code * * \note Since this gets run at the very start, \c respawn_count fields * for particular children get temporarily overloaded with "rounds * of waiting" tracking, restored once we are about to finish with * success (i.e. returning value >=0) and will remain unrestored * otherwise. One way to suppress liveness detection logic for * particular child is to set the said value to a negative number. */ #define WAIT_TRIES 4 /* together with interleaved sleeps, worst case ~ 1s */ static int find_and_track_existing_processes(void) { bool tracking = false; bool wait_in_progress; int rc; size_t i, rounds; for (rounds = 1; rounds <= WAIT_TRIES; rounds++) { wait_in_progress = false; for (i = 0; i < SIZEOF(pcmk_children); i++) { if ((pcmk_children[i].endpoint == NULL) || (pcmk_children[i].respawn_count < 0)) { continue; } rc = child_liveness(&pcmk_children[i]); if (rc == pcmk_rc_ipc_unresponsive) { /* As a speculation, don't give up if there are more rounds to * come for other reasons, but don't artificially wait just * because of this, since we would preferably start ASAP. */ continue; } pcmk_children[i].respawn_count = rounds; switch (rc) { case pcmk_rc_ok: if (pcmk_children[i].pid == PCMK__SPECIAL_PID) { if (crm_is_true(getenv("PCMK_fail_fast"))) { crm_crit("Cannot reliably track pre-existing" " authentic process behind %s IPC on this" " platform and PCMK_fail_fast requested", pcmk_children[i].endpoint); return EOPNOTSUPP; } else if (pcmk_children[i].respawn_count == WAIT_TRIES) { crm_notice("Assuming pre-existing authentic, though" " on this platform untrackable, process" " behind %s IPC is stable (was in %d" " previous samples) so rather than" " bailing out (PCMK_fail_fast not" " requested), we just switch to a less" " optimal IPC liveness monitoring" " (not very suitable for heavy load)", pcmk_children[i].name, WAIT_TRIES - 1); crm_warn("The process behind %s IPC cannot be" " terminated, so the overall shutdown" " will get delayed implicitly (%ld s)," " which serves as a graceful period for" " its native termination if it vitally" " depends on some other daemons going" " down in a controlled way already", pcmk_children[i].name, (long) SHUTDOWN_ESCALATION_PERIOD); } else { wait_in_progress = true; crm_warn("Cannot reliably track pre-existing" " authentic process behind %s IPC on this" " platform, can still disappear in %d" " attempt(s)", pcmk_children[i].endpoint, WAIT_TRIES - pcmk_children[i].respawn_count); continue; } } crm_notice("Tracking existing %s process (pid=%lld)", pcmk_children[i].name, (long long) PCMK__SPECIAL_PID_AS_0( pcmk_children[i].pid)); pcmk_children[i].respawn_count = -1; /* 0~keep watching */ pcmk_children[i].active_before_startup = TRUE; tracking = true; break; case pcmk_rc_ipc_pid_only: if (pcmk_children[i].respawn_count == WAIT_TRIES) { crm_crit("%s IPC end-point for existing authentic" " process %lld did not (re)appear", pcmk_children[i].endpoint, (long long) PCMK__SPECIAL_PID_AS_0( pcmk_children[i].pid)); return rc; } wait_in_progress = true; crm_warn("Cannot find %s IPC end-point for existing" " authentic process %lld, can still (re)appear" " in %d attempts (?)", pcmk_children[i].endpoint, (long long) PCMK__SPECIAL_PID_AS_0( pcmk_children[i].pid), WAIT_TRIES - pcmk_children[i].respawn_count); continue; default: crm_crit("Checked liveness of %s: %s " CRM_XS " rc=%d", pcmk_children[i].name, pcmk_rc_str(rc), rc); return rc; } } if (!wait_in_progress) { break; } (void) poll(NULL, 0, 250); /* a bit for changes to possibly happen */ } for (i = 0; i < SIZEOF(pcmk_children); i++) { pcmk_children[i].respawn_count = 0; /* restore pristine state */ } if (tracking) { g_timeout_add_seconds(PCMK_PROCESS_CHECK_INTERVAL, check_active_before_startup_processes, NULL); } return pcmk_rc_ok; } static void init_children_processes(void) { int start_seq = 1, lpc = 0; static int max = SIZEOF(pcmk_children); /* start any children that have not been detected */ for (start_seq = 1; start_seq < max; start_seq++) { /* don't start anything with start_seq < 1 */ for (lpc = 0; lpc < max; lpc++) { - if (pcmk_children[lpc].pid) { + if (pcmk_children[lpc].pid != 0) { /* we are already tracking it */ continue; } if (start_seq == pcmk_children[lpc].start_seq) { start_child(&(pcmk_children[lpc])); } } } /* From this point on, any daemons being started will be due to * respawning rather than node start. * * This may be useful for the daemons to know */ setenv("PCMK_respawned", "true", 1); } static void mcp_cpg_destroy(gpointer user_data) { crm_crit("Lost connection to cluster layer, shutting down"); crm_exit(CRM_EX_DISCONNECT); } /*! * \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"), (char*)msg); if (task == NULL) { if (nodeid == local_nodeid) { crm_debug("Ignoring message with local node's process list"); } 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); } } static void mcp_cpg_membership(cpg_handle_t handle, const struct cpg_name *groupName, const struct cpg_address *member_list, size_t member_list_entries, const struct cpg_address *left_list, size_t left_list_entries, const struct cpg_address *joined_list, size_t joined_list_entries) { /* Update peer cache if needed */ pcmk_cpg_membership(handle, groupName, member_list, member_list_entries, left_list, left_list_entries, joined_list, joined_list_entries); /* Always broadcast our own presence after any membership change */ update_process_peers(); } static gboolean mcp_quorum_callback(unsigned long long seq, gboolean quorate) { pcmk_quorate = quorate; return TRUE; } static void mcp_quorum_destroy(gpointer user_data) { crm_info("connection lost"); } 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; 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 's': pcmk__set_env_option("node_start_state", "standby"); break; case '$': case '?': crm_help(flag, CRM_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(CRM_EX_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('?', CRM_EX_USAGE); } setenv("LC_ALL", "C", 1); pcmk__set_env_option("mcp", "true"); crm_log_init(NULL, LOG_INFO, TRUE, FALSE, argc, argv, FALSE); crm_debug("Checking for existing Pacemaker instance"); old_instance = crm_ipc_new(CRM_SYSTEM_MCP, 0); (void) crm_ipc_connect(old_instance); if (shutdown) { crm_debug("Shutting down existing Pacemaker instance by request"); 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(CRM_EX_OK); } else if (crm_ipc_connected(old_instance)) { crm_ipc_close(old_instance); crm_ipc_destroy(old_instance); crm_err("Aborting start-up because active Pacemaker instance found"); crm_exit(CRM_EX_FATAL); } 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(CRM_EX_UNAVAILABLE); } // OCF shell functions and cluster-glue need facility under different name { const char *facility = pcmk__env_option("logfacility"); if (facility && safe_str_neq(facility, "none")) { setenv("HA_LOGFACILITY", facility, 1); } } crm_notice("Starting Pacemaker %s "CRM_XS" build=%s features:%s", PACEMAKER_VERSION, BUILD_VERSION, CRM_FEATURES); mainloop = g_main_loop_new(NULL, FALSE); 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 diagnostic tool, so" " please consider enabling them by default."); } } if (pcmk_daemon_user(&pcmk_uid, &pcmk_gid) < 0) { crm_err("Cluster user %s does not exist, aborting Pacemaker startup", CRM_DAEMON_USER); crm_exit(CRM_EX_NOUSER); } // Used by some resource agents if ((mkdir(CRM_STATE_DIR, 0750) < 0) && (errno != EEXIST)) { crm_warn("Could not create " CRM_STATE_DIR ": %s", pcmk_strerror(errno)); } else { mcp_chown(CRM_STATE_DIR, pcmk_uid, pcmk_gid); } /* Used to store core/blackbox/scheduler/cib files in */ crm_build_path(CRM_PACEMAKER_DIR, 0750); mcp_chown(CRM_PACEMAKER_DIR, pcmk_uid, pcmk_gid); /* Used to store core files in */ crm_build_path(CRM_CORE_DIR, 0750); mcp_chown(CRM_CORE_DIR, pcmk_uid, pcmk_gid); /* Used to store blackbox dumps in */ crm_build_path(CRM_BLACKBOX_DIR, 0750); mcp_chown(CRM_BLACKBOX_DIR, pcmk_uid, pcmk_gid); // Used to store scheduler inputs in crm_build_path(PE_STATE_DIR, 0750); mcp_chown(PE_STATE_DIR, pcmk_uid, pcmk_gid); /* Used to store the cluster configuration */ crm_build_path(CRM_CONFIG_DIR, 0750); mcp_chown(CRM_CONFIG_DIR, pcmk_uid, pcmk_gid); // Don't build CRM_RSCTMP_DIR, pacemaker-execd will do it 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(CRM_EX_OSERR); } /* Allows us to block shutdown */ if (cluster_connect_cfg(&local_nodeid) == FALSE) { crm_err("Couldn't connect to Corosync's CFG service"); crm_exit(CRM_EX_PROTOCOL); } if(pcmk_locate_sbd() > 0) { setenv("PCMK_watchdog", "true", 1); } else { setenv("PCMK_watchdog", "false", 1); } switch (find_and_track_existing_processes()) { case pcmk_rc_ok: break; case pcmk_rc_ipc_unauthorized: crm_exit(CRM_EX_CANTCREAT); default: crm_exit(CRM_EX_FATAL); }; 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); rc = pcmk_ok; if (cluster_connect_cpg(&cluster) == FALSE) { crm_err("Couldn't connect to Corosync's CPG service"); rc = -ENOPROTOOPT; } else if (cluster_connect_quorum(mcp_quorum_callback, mcp_quorum_destroy) == FALSE) { rc = -ENOTCONN; } else { 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_notice("Pacemaker daemon successfully started and accepting connections"); g_main_loop_run(mainloop); } if (ipcs) { crm_trace("Closing IPC server"); mainloop_del_ipc_server(ipcs); ipcs = NULL; } g_main_loop_unref(mainloop); cluster_disconnect_cpg(&cluster); cluster_disconnect_cfg(); crm_exit(crm_errno2exit(rc)); } diff --git a/include/crm/common/internal.h b/include/crm/common/internal.h index 8537e3e0af..bb5219ce02 100644 --- a/include/crm/common/internal.h +++ b/include/crm/common/internal.h @@ -1,220 +1,217 @@ /* * Copyright 2015-2020 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #ifndef CRM_COMMON_INTERNAL__H #define CRM_COMMON_INTERNAL__H #include /* for gboolean */ #include /* for struct dirent */ #include /* for getpid() */ #include /* for bool */ -#include /* for uid_t and gid_t */ +#include // uid_t, gid_t, pid_t #include #if SUPPORT_CIBSECRETS // Internal CIB utilities (from cib_secrets.c) */ int pcmk__substitute_secrets(const char *rsc_id, GHashTable *params); #endif /* internal I/O utilities (from io.c) */ int pcmk__real_path(const char *path, char **resolved_path); char *pcmk__series_filename(const char *directory, const char *series, int sequence, bool bzip); int pcmk__read_series_sequence(const char *directory, const char *series, unsigned int *seq); void pcmk__write_series_sequence(const char *directory, const char *series, unsigned int sequence, int max); int pcmk__chown_series_sequence(const char *directory, const char *series, uid_t uid, gid_t gid); bool pcmk__daemon_can_write(const char *dir, const char *file); void pcmk__sync_directory(const char *name); int pcmk__file_contents(const char *filename, char **contents); int pcmk__write_sync(int fd, const char *contents); int pcmk__set_nonblocking(int fd); const char *pcmk__get_tmpdir(void); void pcmk__close_fds_in_child(bool); /* internal procfs utilities (from procfs.c) */ int crm_procfs_process_info(struct dirent *entry, char *name, int *pid); int crm_procfs_pid_of(const char *name); unsigned int crm_procfs_num_cores(void); /* internal XML schema functions (from xml.c) */ void crm_schema_init(void); void crm_schema_cleanup(void); /* internal functions related to process IDs (from pid.c) */ /*! * \internal - * \brief Detect if process per PID and optionally exe path (component) exists + * \brief Check whether process exists (by PID and optionally executable path) * - * \param[in] pid PID of process assumed alive, disproving of which to try - * \param[in] daemon exe path (component) to possibly match with procfs entry - * - * \return -1 on invalid PID specification, -2 when the calling process has no - * (is refused an) ability to (dis)prove the predicate, - * 0 if the negation of the predicate is confirmed (check-through-kill - * indicates so, or the subsequent check-through-procfs-match on - * \p daemon when provided and procfs available at the standard path), - * 1 if it cannot be disproved (reliably [modulo race conditions] - * when \p daemon provided, procfs available at the standard path - * and the calling process has permissions to access the respective - * procfs location, less so otherwise, since mere check-through-kill - * is exercised without powers to exclude PID recycled in the interim). + * \param[in] pid PID of process to check + * \param[in] daemon If not NULL, path component to match with procfs entry * + * \return Standard Pacemaker return code + * \note Particular return codes of interest include pcmk_rc_ok for alive, + * ESRCH for process is not alive (verified by kill and/or executable path + * match), EACCES for caller unable or not allowed to check. A result of + * "alive" is less reliable when \p daemon is not provided or procfs is + * not available, since there is no guarantee that the PID has not been + * recycled for another process. * \note This function cannot be used to verify \e authenticity of the process. */ -int crm_pid_active(long pid, const char *daemon); +int pcmk__pid_active(pid_t pid, const char *daemon); -long crm_pidfile_inuse(const char *filename, long mypid, const char *daemon); -long crm_read_pidfile(const char *filename); -int crm_lock_pidfile(const char *filename, const char *name); +int pcmk__read_pidfile(const char *filename, pid_t *pid); +int pcmk__pidfile_matches(const char *filename, pid_t expected_pid, + const char *expected_name, pid_t *pid); +int pcmk__lock_pidfile(const char *filename, const char *name); /* interal functions related to resource operations (from operations.c) */ char *pcmk__op_key(const char *rsc_id, const char *op_type, guint interval_ms); char *generate_notify_key(const char *rsc_id, const char *notify_type, const char *op_type); char *generate_transition_key(int action, int transition_id, int target_rc, const char *node); void filter_action_parameters(xmlNode *param_set, const char *version); // miscellaneous utilities (from utils.c) const char *pcmk_message_name(const char *name); /* internal generic string functions (from strings.c) */ long long crm_int_helper(const char *text, char **end_text); guint crm_parse_ms(const char *text); bool crm_starts_with(const char *str, const char *prefix); gboolean crm_ends_with(const char *s, const char *match); gboolean crm_ends_with_ext(const char *s, const char *match); char *add_list_element(char *list, const char *value); bool crm_compress_string(const char *data, int length, int max, char **result, unsigned int *result_len); gint crm_alpha_sort(gconstpointer a, gconstpointer b); /* Correctly displaying singular or plural is complicated; consider "1 node has" * vs. "2 nodes have". A flexible solution is to pluralize entire strings, e.g. * * if (a == 1) { * crm_info("singular message"): * } else { * crm_info("plural message"); * } * * though even that's not sufficient for all languages besides English (if we * ever desire to do translations of output and log messages). But the following * convenience macros are "good enough" and more concise for many cases. */ /* Example: * crm_info("Found %d %s", nentries, * pcmk__plural_alt(nentries, "entry", "entries")); */ #define pcmk__plural_alt(i, s1, s2) (((i) == 1)? (s1) : (s2)) // Example: crm_info("Found %d node%s", nnodes, pcmk__plural_s(nnodes)); #define pcmk__plural_s(i) pcmk__plural_alt(i, "", "s") static inline char * crm_concat(const char *prefix, const char *suffix, char join) { CRM_ASSERT(prefix && suffix); return crm_strdup_printf("%s%c%s", prefix, join, suffix); } static inline int crm_strlen_zero(const char *s) { return !s || *s == '\0'; } static inline char * crm_getpid_s() { return crm_strdup_printf("%lu", (unsigned long) getpid()); } // More efficient than g_list_length(list) == 1 static inline bool pcmk__list_of_1(GList *list) { return list && (list->next == NULL); } // More efficient than g_list_length(list) > 1 static inline bool pcmk__list_of_multiple(GList *list) { return list && (list->next != NULL); } /* convenience functions for failure-related node attributes */ #define CRM_FAIL_COUNT_PREFIX "fail-count" #define CRM_LAST_FAILURE_PREFIX "last-failure" /*! * \internal * \brief Generate a failure-related node attribute name for a resource * * \param[in] prefix Start of attribute name * \param[in] rsc_id Resource name * \param[in] op Operation name * \param[in] interval_ms Operation interval * * \return Newly allocated string with attribute name * * \note Failure attributes are named like PREFIX-RSC#OP_INTERVAL (for example, * "fail-count-myrsc#monitor_30000"). The '#' is used because it is not * a valid character in a resource ID, to reliably distinguish where the * operation name begins. The '_' is used simply to be more comparable to * action labels like "myrsc_monitor_30000". */ static inline char * crm_fail_attr_name(const char *prefix, const char *rsc_id, const char *op, guint interval_ms) { CRM_CHECK(prefix && rsc_id && op, return NULL); return crm_strdup_printf("%s-%s#%s_%u", prefix, rsc_id, op, interval_ms); } static inline char * crm_failcount_name(const char *rsc_id, const char *op, guint interval_ms) { return crm_fail_attr_name(CRM_FAIL_COUNT_PREFIX, rsc_id, op, interval_ms); } static inline char * crm_lastfailure_name(const char *rsc_id, const char *op, guint interval_ms) { return crm_fail_attr_name(CRM_LAST_FAILURE_PREFIX, rsc_id, op, interval_ms); } #endif /* CRM_COMMON_INTERNAL__H */ diff --git a/lib/common/pid.c b/lib/common/pid.c index 59f64dd51a..90a24a699c 100644 --- a/lib/common/pid.c +++ b/lib/common/pid.c @@ -1,201 +1,247 @@ /* * Copyright 2004-2020 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #include #ifndef _GNU_SOURCE # define _GNU_SOURCE #endif #include #include #include #include int -crm_pid_active(long pid, const char *daemon) +pcmk__pid_active(pid_t pid, const char *daemon) { - static int last_asked_pid = 0; /* log spam prevention */ + static pid_t last_asked_pid = 0; /* log spam prevention */ #if SUPPORT_PROCFS static int have_proc_pid = 0; #else static int have_proc_pid = -1; #endif int rc = 0; + bool no_name_check = ((daemon == NULL) || (have_proc_pid == -1)); if (have_proc_pid == 0) { /* evaluation of /proc/PID/exe applicability via self-introspection */ char proc_path[PATH_MAX], exe_path[PATH_MAX]; - snprintf(proc_path, sizeof(proc_path), "/proc/%lu/exe", - (long unsigned int) getpid()); + snprintf(proc_path, sizeof(proc_path), "/proc/%lld/exe", + (long long) getpid()); have_proc_pid = 1; if (readlink(proc_path, exe_path, sizeof(exe_path) - 1) < 0) { have_proc_pid = -1; } } if (pid <= 0) { - return -1; + return EINVAL; + } - } else if ((rc = kill(pid, 0)) < 0 && errno == ESRCH) { - return 0; /* no such PID detected */ + rc = kill(pid, 0); + if ((rc < 0) && (errno == ESRCH)) { + return ESRCH; /* no such PID detected */ - } else if (rc < 0 && (daemon == NULL || have_proc_pid == -1)) { + } else if ((rc < 0) && no_name_check) { + rc = errno; if (last_asked_pid != pid) { - crm_info("Cannot examine PID %ld: %s", pid, strerror(errno)); + crm_info("Cannot examine PID %lld: %s", + (long long) pid, strerror(errno)); last_asked_pid = pid; } - return -2; /* errno != ESRCH */ + return rc; /* errno != ESRCH */ - } else if (rc == 0 && (daemon == NULL || have_proc_pid == -1)) { - return 1; /* kill as the only indicator, cannot double check */ + } else if ((rc == 0) && no_name_check) { + return pcmk_rc_ok; /* kill as the only indicator, cannot double check */ } else if (daemon != NULL) { /* make sure PID hasn't been reused by another process XXX: might still be just a zombie, which could confuse decisions */ bool checked_through_kill = (rc == 0); char proc_path[PATH_MAX], exe_path[PATH_MAX], myexe_path[PATH_MAX]; - snprintf(proc_path, sizeof(proc_path), "/proc/%ld/exe", pid); + snprintf(proc_path, sizeof(proc_path), "/proc/%lld/exe", + (long long) pid); rc = readlink(proc_path, exe_path, sizeof(exe_path) - 1); - if ((rc < 0) && (errno == EACCES)) { + if (rc < 0) { if (last_asked_pid != pid) { - crm_info("Could not read from %s: %s", proc_path, - strerror(errno)); + crm_err("Could not read from %s: %s " CRM_XS " errno=%d", + proc_path, strerror(errno), errno); last_asked_pid = pid; } - return checked_through_kill ? 1 : -2; - } else if (rc < 0) { - if (last_asked_pid != pid) { - crm_err("Could not read from %s: %s (%d)", proc_path, - strerror(errno), errno); - last_asked_pid = pid; + if ((errno == EACCES) && checked_through_kill) { + // Trust kill result, can't double-check via path + return pcmk_rc_ok; + } else if (errno == EACCES) { + return EACCES; + } else { + return ESRCH; /* most likely errno == ENOENT */ } - return 0; /* most likely errno == ENOENT */ } exe_path[rc] = '\0'; if (daemon[0] != '/') { rc = snprintf(myexe_path, sizeof(myexe_path), CRM_DAEMON_DIR"/%s", daemon); } else { rc = snprintf(myexe_path, sizeof(myexe_path), "%s", daemon); } if (rc > 0 && rc < sizeof(myexe_path) && !strcmp(exe_path, myexe_path)) { - return 1; + return pcmk_rc_ok; } } - return 0; + return ESRCH; } #define LOCKSTRLEN 11 -long -crm_read_pidfile(const char *filename) +/*! + * \internal + * \brief Read a process ID from a file + * + * \param[in] filename Process ID file to read + * \param[out] pid Where to put PID that was read + * + * \return Standard Pacemaker return code + */ +int +pcmk__read_pidfile(const char *filename, pid_t *pid) { int fd; struct stat sbuf; - long pid = -ENOENT; + int rc = pcmk_rc_unknown_format; + long long pid_read = 0; char buf[LOCKSTRLEN + 1]; + CRM_CHECK((filename != NULL) && (pid != NULL), return EINVAL); + fd = open(filename, O_RDONLY); if (fd < 0) { - goto bail; + return errno; } if ((fstat(fd, &sbuf) >= 0) && (sbuf.st_size < LOCKSTRLEN)) { sleep(2); /* if someone was about to create one, * give'm a sec to do so */ } if (read(fd, buf, sizeof(buf)) < 1) { + rc = errno; goto bail; } - if (sscanf(buf, "%ld", &pid) > 0) { - if (pid <= 0) { - pid = -ESRCH; + if (sscanf(buf, "%lld", &pid_read) > 0) { + if (pid_read <= 0) { + rc = ESRCH; } else { - crm_trace("Got pid %lu from %s\n", pid, filename); + rc = pcmk_rc_ok; + *pid = (pid_t) pid_read; + crm_trace("Read pid %lld from %s", pid_read, filename); } } bail: - if (fd >= 0) { - close(fd); - } - return pid; + close(fd); + return rc; } -long -crm_pidfile_inuse(const char *filename, long mypid, const char *daemon) +/*! + * \internal + * \brief Check whether a process from a PID file matches expected values + * + * \param[in] filename Path of PID file + * \param[in] expected_pid If positive, compare to this PID + * \param[in] expected_name If not NULL, the PID from the PID file is valid + * only if it is active as a process with this name + * \param[out] pid If not NULL, store PID found in PID file here + * + * \return Standard Pacemaker return code + */ +int +pcmk__pidfile_matches(const char *filename, pid_t expected_pid, + const char *expected_name, pid_t *pid) { - long pid = crm_read_pidfile(filename); + pid_t pidfile_pid = 0; + int rc = pcmk__read_pidfile(filename, &pidfile_pid); - if (pid < 2) { - // Invalid pid - pid = -ENOENT; + if (pid) { + *pid = pidfile_pid; + } + + if (rc != pcmk_rc_ok) { + // Error reading PID file or invalid contents unlink(filename); + rc = ENOENT; - } else if (mypid && (pid == mypid)) { - // In use by us - pid = pcmk_ok; + } else if ((expected_pid > 0) && (pidfile_pid == expected_pid)) { + // PID in file matches what was expected + rc = pcmk_rc_ok; - } else if (crm_pid_active(pid, daemon) == FALSE) { + } else if (pcmk__pid_active(pidfile_pid, expected_name) == ESRCH) { // Contains a stale value unlink(filename); - pid = -ENOENT; + rc = ENOENT; - } else if (mypid && (pid != mypid)) { + } else if ((expected_pid > 0) && (pidfile_pid != expected_pid)) { // Locked by existing process - pid = -EEXIST; + rc = EEXIST; } - return pid; + return rc; } +/*! + * \internal + * \brief Create a PID file for the current process (if not already existent) + * + * \param[in] filename Name of PID file to create + * \param[in] name Name of current process + * + * \return Standard Pacemaker return code + */ int -crm_lock_pidfile(const char *filename, const char *name) +pcmk__lock_pidfile(const char *filename, const char *name) { - long mypid = 0; + pid_t mypid = getpid(); int fd = 0; int rc = 0; char buf[LOCKSTRLEN + 2]; - mypid = (unsigned long) getpid(); - - rc = crm_pidfile_inuse(filename, 0, name); - if (rc == -ENOENT) { - // Exists, but the process is not active - - } else if (rc != pcmk_ok) { + rc = pcmk__pidfile_matches(filename, 0, name, NULL); + if ((rc != pcmk_rc_ok) && (rc != ENOENT)) { // Locked by existing process return rc; } fd = open(filename, O_CREAT | O_WRONLY | O_EXCL, 0644); if (fd < 0) { - return -errno; + return errno; } - snprintf(buf, sizeof(buf), "%*ld\n", LOCKSTRLEN - 1, mypid); + snprintf(buf, sizeof(buf), "%*lld\n", LOCKSTRLEN - 1, (long long) mypid); rc = write(fd, buf, LOCKSTRLEN); close(fd); if (rc != LOCKSTRLEN) { crm_perror(LOG_ERR, "Incomplete write to %s", filename); - return -errno; + return errno; } - return crm_pidfile_inuse(filename, mypid, name); + rc = pcmk__pidfile_matches(filename, mypid, name, NULL); + if (rc != pcmk_rc_ok) { + // Something is really wrong -- maybe I/O error on read back? + unlink(filename); + } + return rc; } diff --git a/lib/common/procfs.c b/lib/common/procfs.c index 026876960e..891dc87561 100644 --- a/lib/common/procfs.c +++ b/lib/common/procfs.c @@ -1,164 +1,166 @@ /* - * Copyright 2015-2018 Andrew Beekhof + * Copyright 2015-2020 the Pacemaker project contributors + * + * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #include #ifndef _GNU_SOURCE # define _GNU_SOURCE #endif #include #include #include #include #include #include #include /*! * \internal * \brief Get process ID and name associated with a /proc directory entry * * \param[in] entry Directory entry (must be result of readdir() on /proc) * \param[out] name If not NULL, a char[16] to hold the process name * \param[out] pid If not NULL, will be set to process ID of entry * * \return 0 on success, -1 if entry is not for a process or info not found * * \note This should be called only on Linux systems, as not all systems that * support /proc store process names and IDs in the same way. The kernel * limits the process name to the first 15 characters (plus terminator). * It would be nice if there were a public kernel API constant for that * limit, but there isn't. */ int crm_procfs_process_info(struct dirent *entry, char *name, int *pid) { int fd, local_pid; FILE *file; struct stat statbuf; char procpath[128] = { 0 }; /* We're only interested in entries whose name is a PID, * so skip anything non-numeric or that is too long. * * 114 = 128 - strlen("/proc/") - strlen("/status") - 1 */ local_pid = atoi(entry->d_name); if ((local_pid <= 0) || (strlen(entry->d_name) > 114)) { return -1; } if (pid) { *pid = local_pid; } /* Get this entry's file information */ strcpy(procpath, "/proc/"); strcat(procpath, entry->d_name); fd = open(procpath, O_RDONLY); if (fd < 0 ) { return -1; } if (fstat(fd, &statbuf) < 0) { close(fd); return -1; } close(fd); /* We're only interested in subdirectories */ if (!S_ISDIR(statbuf.st_mode)) { return -1; } /* Read the first entry ("Name:") from the process's status file. * We could handle the valgrind case if we parsed the cmdline file * instead, but that's more of a pain than it's worth. */ if (name != NULL) { strcat(procpath, "/status"); file = fopen(procpath, "r"); if (!file) { return -1; } if (fscanf(file, "Name:\t%15[^\n]", name) != 1) { fclose(file); return -1; } name[15] = 0; fclose(file); } return 0; } /*! * \internal * \brief Return process ID of a named process * * \param[in] name Process name (as used in /proc/.../status) * * \return Process ID of named process if running, 0 otherwise * * \note This will return 0 if the process is being run via valgrind. * This should be called only on Linux systems. */ int crm_procfs_pid_of(const char *name) { DIR *dp; struct dirent *entry; int pid = 0; char entry_name[64] = { 0 }; dp = opendir("/proc"); if (dp == NULL) { crm_notice("Can not read /proc directory to track existing components"); return 0; } while ((entry = readdir(dp)) != NULL) { if ((crm_procfs_process_info(entry, entry_name, &pid) == 0) && safe_str_eq(entry_name, name) - && (crm_pid_active(pid, NULL) == 1)) { + && (pcmk__pid_active((pid_t) pid, NULL) == pcmk_rc_ok)) { crm_info("Found %s active as process %d", name, pid); break; } pid = 0; } closedir(dp); return pid; } /*! * \internal * \brief Calculate number of logical CPU cores from procfs * * \return Number of cores (or 1 if unable to determine) */ unsigned int crm_procfs_num_cores(void) { int cores = 0; FILE *stream = NULL; /* Parse /proc/stat instead of /proc/cpuinfo because it's smaller */ stream = fopen("/proc/stat", "r"); if (stream == NULL) { crm_perror(LOG_INFO, "Could not open /proc/stat"); } else { char buffer[2048]; while (fgets(buffer, sizeof(buffer), stream)) { if (crm_starts_with(buffer, "cpu") && isdigit(buffer[3])) { ++cores; } } fclose(stream); } return cores? cores : 1; } diff --git a/lib/common/utils.c b/lib/common/utils.c index 19fc0a3f75..09656e9dc5 100644 --- a/lib/common/utils.c +++ b/lib/common/utils.c @@ -1,1243 +1,1246 @@ /* - * Copyright 2004-2019 the Pacemaker project contributors + * Copyright 2004-2020 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #include #include #ifndef _GNU_SOURCE # define _GNU_SOURCE #endif #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifndef MAXLINE # define MAXLINE 512 #endif #ifdef HAVE_GETOPT_H # include #endif #ifndef PW_BUFFER_LEN # define PW_BUFFER_LEN 500 #endif CRM_TRACE_INIT_DATA(common); gboolean crm_config_error = FALSE; gboolean crm_config_warning = FALSE; char *crm_system_name = NULL; int node_score_red = 0; int node_score_green = 0; int node_score_yellow = 0; static struct crm_option *crm_long_options = NULL; static const char *crm_app_description = NULL; static char *crm_short_options = NULL; static const char *crm_app_usage = NULL; gboolean check_time(const char *value) { if (crm_get_msec(value) < 5000) { return FALSE; } return TRUE; } gboolean check_timer(const char *value) { if (crm_get_msec(value) < 0) { return FALSE; } return TRUE; } gboolean check_boolean(const char *value) { int tmp = FALSE; if (crm_str_to_boolean(value, &tmp) != 1) { return FALSE; } return TRUE; } gboolean check_number(const char *value) { errno = 0; if (value == NULL) { return FALSE; } else if (safe_str_eq(value, CRM_MINUS_INFINITY_S)) { } else if (safe_str_eq(value, CRM_INFINITY_S)) { } else { crm_int_helper(value, NULL); } if (errno != 0) { return FALSE; } return TRUE; } gboolean check_positive_number(const char* value) { if (safe_str_eq(value, CRM_INFINITY_S) || (crm_int_helper(value, NULL))) { return TRUE; } return FALSE; } gboolean check_quorum(const char *value) { if (safe_str_eq(value, "stop")) { return TRUE; } else if (safe_str_eq(value, "freeze")) { return TRUE; } else if (safe_str_eq(value, "ignore")) { return TRUE; } else if (safe_str_eq(value, "suicide")) { return TRUE; } return FALSE; } gboolean check_script(const char *value) { struct stat st; if(safe_str_eq(value, "/dev/null")) { return TRUE; } if(stat(value, &st) != 0) { crm_err("Script %s does not exist", value); return FALSE; } if(S_ISREG(st.st_mode) == 0) { crm_err("Script %s is not a regular file", value); return FALSE; } if( (st.st_mode & (S_IXUSR | S_IXGRP )) == 0) { crm_err("Script %s is not executable", value); return FALSE; } return TRUE; } gboolean check_utilization(const char *value) { char *end = NULL; long number = strtol(value, &end, 10); if(end && end[0] != '%') { return FALSE; } else if(number < 0) { return FALSE; } return TRUE; } void crm_args_fini() { free(crm_short_options); crm_short_options = NULL; } int char2score(const char *score) { int score_f = 0; if (score == NULL) { } else if (safe_str_eq(score, CRM_MINUS_INFINITY_S)) { score_f = -CRM_SCORE_INFINITY; } else if (safe_str_eq(score, CRM_INFINITY_S)) { score_f = CRM_SCORE_INFINITY; } else if (safe_str_eq(score, CRM_PLUS_INFINITY_S)) { score_f = CRM_SCORE_INFINITY; } else if (safe_str_eq(score, "red")) { score_f = node_score_red; } else if (safe_str_eq(score, "yellow")) { score_f = node_score_yellow; } else if (safe_str_eq(score, "green")) { score_f = node_score_green; } else { score_f = crm_parse_int(score, NULL); if (score_f > 0 && score_f > CRM_SCORE_INFINITY) { score_f = CRM_SCORE_INFINITY; } else if (score_f < 0 && score_f < -CRM_SCORE_INFINITY) { score_f = -CRM_SCORE_INFINITY; } } return score_f; } char * score2char_stack(int score, char *buf, size_t len) { if (score >= CRM_SCORE_INFINITY) { strncpy(buf, CRM_INFINITY_S, 9); } else if (score <= -CRM_SCORE_INFINITY) { strncpy(buf, CRM_MINUS_INFINITY_S , 10); } else { return crm_itoa_stack(score, buf, len); } return buf; } char * score2char(int score) { if (score >= CRM_SCORE_INFINITY) { return strdup(CRM_INFINITY_S); } else if (score <= -CRM_SCORE_INFINITY) { return strdup(CRM_MINUS_INFINITY_S); } return crm_itoa(score); } const char * cluster_option(GHashTable * options, gboolean(*validate) (const char *), const char *name, const char *old_name, const char *def_value) { const char *value = NULL; char *new_value = NULL; CRM_ASSERT(name != NULL); if (options) { value = g_hash_table_lookup(options, name); if ((value == NULL) && old_name) { value = g_hash_table_lookup(options, old_name); if (value != NULL) { crm_config_warn("Support for legacy name '%s' for cluster option '%s'" " is deprecated and will be removed in a future release", old_name, name); // Inserting copy with current name ensures we only warn once new_value = strdup(value); g_hash_table_insert(options, strdup(name), new_value); value = new_value; } } if (value && validate && (validate(value) == FALSE)) { crm_config_err("Resetting cluster option '%s' to default: value '%s' is invalid", name, value); value = NULL; } if (value) { return value; } } // No value found, use default value = def_value; if (value == NULL) { crm_trace("No value or default provided for cluster option '%s'", name); return NULL; } if (validate) { CRM_CHECK(validate(value) != FALSE, crm_err("Bug: default value for cluster option '%s' is invalid", name); return NULL); } crm_trace("Using default value '%s' for cluster option '%s'", value, name); if (options) { new_value = strdup(value); g_hash_table_insert(options, strdup(name), new_value); value = new_value; } return value; } const char * get_cluster_pref(GHashTable * options, pe_cluster_option * option_list, int len, const char *name) { const char *value = NULL; for (int lpc = 0; lpc < len; lpc++) { if (safe_str_eq(name, option_list[lpc].name)) { value = cluster_option(options, option_list[lpc].is_valid, option_list[lpc].name, option_list[lpc].alt_name, option_list[lpc].default_value); return value; } } CRM_CHECK(FALSE, crm_err("Bug: looking for unknown option '%s'", name)); return NULL; } void config_metadata(const char *name, const char *version, const char *desc_short, const char *desc_long, pe_cluster_option * option_list, int len) { int lpc = 0; fprintf(stdout, "" "\n" "\n" " %s\n" " %s\n" " %s\n" " \n", name, version, desc_long, desc_short); for (lpc = 0; lpc < len; lpc++) { if (option_list[lpc].description_long == NULL && option_list[lpc].description_short == NULL) { continue; } fprintf(stdout, " \n" " %s\n" " \n" " %s%s%s\n" " \n", option_list[lpc].name, option_list[lpc].description_short, option_list[lpc].type, option_list[lpc].default_value, option_list[lpc].description_long ? option_list[lpc]. description_long : option_list[lpc].description_short, option_list[lpc].values ? " Allowed values: " : "", option_list[lpc].values ? option_list[lpc].values : ""); } fprintf(stdout, " \n\n"); } void verify_all_options(GHashTable * options, pe_cluster_option * option_list, int len) { int lpc = 0; for (lpc = 0; lpc < len; lpc++) { cluster_option(options, option_list[lpc].is_valid, option_list[lpc].name, option_list[lpc].alt_name, option_list[lpc].default_value); } } char * generate_hash_key(const char *crm_msg_reference, const char *sys) { char *hash_key = crm_concat(sys ? sys : "none", crm_msg_reference, '_'); crm_trace("created hash key: (%s)", hash_key); return hash_key; } int crm_user_lookup(const char *name, uid_t * uid, gid_t * gid) { int rc = pcmk_ok; char *buffer = NULL; struct passwd pwd; struct passwd *pwentry = NULL; buffer = calloc(1, PW_BUFFER_LEN); if (buffer == NULL) { return -ENOMEM; } rc = getpwnam_r(name, &pwd, buffer, PW_BUFFER_LEN, &pwentry); if (pwentry) { if (uid) { *uid = pwentry->pw_uid; } if (gid) { *gid = pwentry->pw_gid; } crm_trace("User %s has uid=%d gid=%d", name, pwentry->pw_uid, pwentry->pw_gid); } else { rc = rc? -rc : -EINVAL; crm_info("User %s lookup: %s", name, pcmk_strerror(rc)); } free(buffer); return rc; } /*! * \brief Get user and group IDs of pacemaker daemon user * * \param[out] uid If non-NULL, where to store daemon user ID * \param[out] gid If non-NULL, where to store daemon group ID * * \return pcmk_ok on success, -errno otherwise */ int pcmk_daemon_user(uid_t *uid, gid_t *gid) { static uid_t daemon_uid; static gid_t daemon_gid; static bool found = false; int rc = pcmk_err_generic; if (!found) { rc = crm_user_lookup(CRM_DAEMON_USER, &daemon_uid, &daemon_gid); if (rc == pcmk_ok) { found = true; } } if (found) { if (uid) { *uid = daemon_uid; } if (gid) { *gid = daemon_gid; } } return rc; } static int crm_version_helper(const char *text, const char **end_text) { int atoi_result = -1; CRM_ASSERT(end_text != NULL); errno = 0; if (text != NULL && text[0] != 0) { /* seemingly sacrificing const-correctness -- because while strtol doesn't modify the input, it doesn't want to artificially taint the "end_text" pointer-to-pointer-to-first-char-in-string with constness in case the input wasn't actually constant -- by semantic definition not a single character will get modified so it shall be perfectly safe to make compiler happy with dropping "const" qualifier here */ atoi_result = (int) strtol(text, (char **) end_text, 10); if (errno == EINVAL) { crm_err("Conversion of '%s' %c failed", text, text[0]); atoi_result = -1; } } return atoi_result; } /* * version1 < version2 : -1 * version1 = version2 : 0 * version1 > version2 : 1 */ int compare_version(const char *version1, const char *version2) { int rc = 0; int lpc = 0; const char *ver1_iter, *ver2_iter; if (version1 == NULL && version2 == NULL) { return 0; } else if (version1 == NULL) { return -1; } else if (version2 == NULL) { return 1; } ver1_iter = version1; ver2_iter = version2; while (1) { int digit1 = 0; int digit2 = 0; lpc++; if (ver1_iter == ver2_iter) { break; } if (ver1_iter != NULL) { digit1 = crm_version_helper(ver1_iter, &ver1_iter); } if (ver2_iter != NULL) { digit2 = crm_version_helper(ver2_iter, &ver2_iter); } if (digit1 < digit2) { rc = -1; break; } else if (digit1 > digit2) { rc = 1; break; } if (ver1_iter != NULL && *ver1_iter == '.') { ver1_iter++; } if (ver1_iter != NULL && *ver1_iter == '\0') { ver1_iter = NULL; } if (ver2_iter != NULL && *ver2_iter == '.') { ver2_iter++; } if (ver2_iter != NULL && *ver2_iter == 0) { ver2_iter = NULL; } } if (rc == 0) { crm_trace("%s == %s (%d)", version1, version2, lpc); } else if (rc < 0) { crm_trace("%s < %s (%d)", version1, version2, lpc); } else if (rc > 0) { crm_trace("%s > %s (%d)", version1, version2, lpc); } return rc; } gboolean do_stderr = FALSE; #ifndef NUMCHARS # define NUMCHARS "0123456789." #endif #ifndef WHITESPACE # define WHITESPACE " \t\n\r\f" #endif guint crm_parse_interval_spec(const char *input) { long long msec = 0; if (input == NULL) { return 0; } else if (input[0] != 'P') { long long tmp = crm_get_msec(input); if(tmp > 0) { msec = tmp; } } else { crm_time_t *period_s = crm_time_parse_duration(input); if (period_s) { msec = 1000 * crm_time_get_seconds(period_s); crm_time_free(period_s); } else { // Reason why not valid has already been logged crm_warn("Using 0 instead of '%s'", input); } } return (msec <= 0)? 0 : ((msec >= G_MAXUINT)? G_MAXUINT : (guint) msec); } long long crm_get_msec(const char *input) { const char *cp = input; const char *units; long long multiplier = 1000; long long divisor = 1; long long msec = -1; char *end_text = NULL; /* double dret; */ if (input == NULL) { return msec; } cp += strspn(cp, WHITESPACE); units = cp + strspn(cp, NUMCHARS); units += strspn(units, WHITESPACE); if (strchr(NUMCHARS, *cp) == NULL) { return msec; } if (strncasecmp(units, "ms", 2) == 0 || strncasecmp(units, "msec", 4) == 0) { multiplier = 1; divisor = 1; } else if (strncasecmp(units, "us", 2) == 0 || strncasecmp(units, "usec", 4) == 0) { multiplier = 1; divisor = 1000; } else if (strncasecmp(units, "s", 1) == 0 || strncasecmp(units, "sec", 3) == 0) { multiplier = 1000; divisor = 1; } else if (strncasecmp(units, "m", 1) == 0 || strncasecmp(units, "min", 3) == 0) { multiplier = 60 * 1000; divisor = 1; } else if (strncasecmp(units, "h", 1) == 0 || strncasecmp(units, "hr", 2) == 0) { multiplier = 60 * 60 * 1000; divisor = 1; } else if (*units != EOS && *units != '\n' && *units != '\r') { return msec; } msec = crm_int_helper(cp, &end_text); if (msec > LLONG_MAX/multiplier) { /* arithmetics overflow while multiplier/divisor mutually exclusive */ return LLONG_MAX; } msec *= multiplier; msec /= divisor; /* dret += 0.5; */ /* msec = (long long)dret; */ return msec; } extern bool crm_is_daemon; /* coverity[+kill] */ void crm_abort(const char *file, const char *function, int line, const char *assert_condition, gboolean do_core, gboolean do_fork) { int rc = 0; int pid = 0; int status = 0; /* Implied by the parent's error logging below */ /* crm_write_blackbox(0); */ if(crm_is_daemon == FALSE) { /* This is a command line tool - do not fork */ /* crm_add_logfile(NULL); * Record it to a file? */ crm_enable_stderr(TRUE); /* Make sure stderr is enabled so we can tell the caller */ do_fork = FALSE; /* Just crash if needed */ } if (do_core == FALSE) { crm_err("%s: Triggered assert at %s:%d : %s", function, file, line, assert_condition); return; } else if (do_fork) { pid = fork(); } else { crm_err("%s: Triggered fatal assert at %s:%d : %s", function, file, line, assert_condition); } if (pid == -1) { crm_crit("%s: Cannot create core for non-fatal assert at %s:%d : %s", function, file, line, assert_condition); return; } else if(pid == 0) { /* Child process */ abort(); return; } /* Parent process */ crm_err("%s: Forked child %d to record non-fatal assert at %s:%d : %s", function, pid, file, line, assert_condition); crm_write_blackbox(SIGTRAP, NULL); do { rc = waitpid(pid, &status, 0); if(rc == pid) { return; /* Job done */ } } while(errno == EINTR); if (errno == ECHILD) { /* crm_mon does this */ crm_trace("Cannot wait on forked child %d - SIGCHLD is probably set to SIG_IGN", pid); return; } crm_perror(LOG_ERR, "Cannot wait on forked child %d", pid); } void crm_make_daemon(const char *name, gboolean daemonize, const char *pidfile) { int rc; - long pid; + pid_t pid; const char *devnull = "/dev/null"; if (daemonize == FALSE) { return; } /* Check before we even try... */ - rc = crm_pidfile_inuse(pidfile, 1, name); - if(rc < pcmk_ok && rc != -ENOENT) { - pid = crm_read_pidfile(pidfile); - crm_err("%s: already running [pid %ld in %s]", name, pid, pidfile); - printf("%s: already running [pid %ld in %s]\n", name, pid, pidfile); + rc = pcmk__pidfile_matches(pidfile, 1, name, &pid); + if ((rc != pcmk_rc_ok) && (rc != ENOENT)) { + crm_err("%s: already running [pid %lld in %s]", + name, (long long) pid, pidfile); + printf("%s: already running [pid %lld in %s]\n", + name, (long long) pid, pidfile); crm_exit(CRM_EX_ERROR); } pid = fork(); if (pid < 0) { fprintf(stderr, "%s: could not start daemon\n", name); crm_perror(LOG_ERR, "fork"); crm_exit(CRM_EX_OSERR); } else if (pid > 0) { crm_exit(CRM_EX_OK); } - rc = crm_lock_pidfile(pidfile, name); - if(rc < pcmk_ok) { - crm_err("Could not lock '%s' for %s: %s (%d)", pidfile, name, pcmk_strerror(rc), rc); - printf("Could not lock '%s' for %s: %s (%d)\n", pidfile, name, pcmk_strerror(rc), rc); + rc = pcmk__lock_pidfile(pidfile, name); + if (rc != pcmk_rc_ok) { + crm_err("Could not lock '%s' for %s: %s " CRM_XS " rc=%d", + pidfile, name, pcmk_rc_str(rc), rc); + printf("Could not lock '%s' for %s: %s (%d)\n", + pidfile, name, pcmk_rc_str(rc), rc); crm_exit(CRM_EX_ERROR); } umask(S_IWGRP | S_IWOTH | S_IROTH); close(STDIN_FILENO); (void)open(devnull, O_RDONLY); /* Stdin: fd 0 */ close(STDOUT_FILENO); (void)open(devnull, O_WRONLY); /* Stdout: fd 1 */ close(STDERR_FILENO); (void)open(devnull, O_WRONLY); /* Stderr: fd 2 */ } char * crm_meta_name(const char *field) { int lpc = 0; int max = 0; char *crm_name = NULL; CRM_CHECK(field != NULL, return NULL); crm_name = crm_concat(CRM_META, field, '_'); /* Massage the names so they can be used as shell variables */ max = strlen(crm_name); for (; lpc < max; lpc++) { switch (crm_name[lpc]) { case '-': crm_name[lpc] = '_'; break; } } return crm_name; } const char * crm_meta_value(GHashTable * hash, const char *field) { char *key = NULL; const char *value = NULL; key = crm_meta_name(field); if (key) { value = g_hash_table_lookup(hash, key); free(key); } return value; } static struct option * crm_create_long_opts(struct crm_option *long_options) { struct option *long_opts = NULL; #ifdef HAVE_GETOPT_H int index = 0, lpc = 0; /* * A previous, possibly poor, choice of '?' as the short form of --help * means that getopt_long() returns '?' for both --help and for "unknown option" * * This dummy entry allows us to differentiate between the two in crm_get_option() * and exit with the correct error code */ long_opts = realloc_safe(long_opts, (index + 1) * sizeof(struct option)); long_opts[index].name = "__dummmy__"; long_opts[index].has_arg = 0; long_opts[index].flag = 0; long_opts[index].val = '_'; index++; for (lpc = 0; long_options[lpc].name != NULL; lpc++) { if (long_options[lpc].name[0] == '-') { continue; } long_opts = realloc_safe(long_opts, (index + 1) * sizeof(struct option)); /*fprintf(stderr, "Creating %d %s = %c\n", index, * long_options[lpc].name, long_options[lpc].val); */ long_opts[index].name = long_options[lpc].name; long_opts[index].has_arg = long_options[lpc].has_arg; long_opts[index].flag = long_options[lpc].flag; long_opts[index].val = long_options[lpc].val; index++; } /* Now create the list terminator */ long_opts = realloc_safe(long_opts, (index + 1) * sizeof(struct option)); long_opts[index].name = NULL; long_opts[index].has_arg = 0; long_opts[index].flag = 0; long_opts[index].val = 0; #endif return long_opts; } void crm_set_options(const char *short_options, const char *app_usage, struct crm_option *long_options, const char *app_desc) { if (short_options) { crm_short_options = strdup(short_options); } else if (long_options) { int lpc = 0; int opt_string_len = 0; char *local_short_options = NULL; for (lpc = 0; long_options[lpc].name != NULL; lpc++) { if (long_options[lpc].val && long_options[lpc].val != '-' && long_options[lpc].val < UCHAR_MAX) { local_short_options = realloc_safe(local_short_options, opt_string_len + 4); local_short_options[opt_string_len++] = long_options[lpc].val; /* getopt(3) says: Two colons mean an option takes an optional arg; */ if (long_options[lpc].has_arg == optional_argument) { local_short_options[opt_string_len++] = ':'; } if (long_options[lpc].has_arg >= required_argument) { local_short_options[opt_string_len++] = ':'; } local_short_options[opt_string_len] = 0; } } crm_short_options = local_short_options; crm_trace("Generated short option string: '%s'", local_short_options); } if (long_options) { crm_long_options = long_options; } if (app_desc) { crm_app_description = app_desc; } if (app_usage) { crm_app_usage = app_usage; } } int crm_get_option(int argc, char **argv, int *index) { return crm_get_option_long(argc, argv, index, NULL); } int crm_get_option_long(int argc, char **argv, int *index, const char **longname) { #ifdef HAVE_GETOPT_H static struct option *long_opts = NULL; if (long_opts == NULL && crm_long_options) { long_opts = crm_create_long_opts(crm_long_options); } *index = 0; if (long_opts) { int flag = getopt_long(argc, argv, crm_short_options, long_opts, index); switch (flag) { case 0: if (long_opts[*index].val) { return long_opts[*index].val; } else if (longname) { *longname = long_opts[*index].name; } else { crm_notice("Unhandled option --%s", long_opts[*index].name); return flag; } case -1: /* End of option processing */ break; case ':': crm_trace("Missing argument"); crm_help('?', CRM_EX_USAGE); break; case '?': crm_help('?', (*index? CRM_EX_OK : CRM_EX_USAGE)); break; } return flag; } #endif if (crm_short_options) { return getopt(argc, argv, crm_short_options); } return -1; } void crm_help(char cmd, crm_exit_t exit_code) { int i = 0; FILE *stream = (exit_code ? stderr : stdout); if (cmd == 'v' || cmd == '$') { fprintf(stream, "Pacemaker %s\n", PACEMAKER_VERSION); fprintf(stream, "Written by Andrew Beekhof\n"); goto out; } if (cmd == '!') { fprintf(stream, "Pacemaker %s (Build: %s): %s\n", PACEMAKER_VERSION, BUILD_VERSION, CRM_FEATURES); goto out; } fprintf(stream, "%s - %s\n", crm_system_name, crm_app_description); if (crm_app_usage) { fprintf(stream, "Usage: %s %s\n", crm_system_name, crm_app_usage); } if (crm_long_options) { fprintf(stream, "Options:\n"); for (i = 0; crm_long_options[i].name != NULL; i++) { if (crm_long_options[i].flags & pcmk_option_hidden) { } else if (crm_long_options[i].flags & pcmk_option_paragraph) { fprintf(stream, "%s\n\n", crm_long_options[i].desc); } else if (crm_long_options[i].flags & pcmk_option_example) { fprintf(stream, "\t#%s\n\n", crm_long_options[i].desc); } else if (crm_long_options[i].val == '-' && crm_long_options[i].desc) { fprintf(stream, "%s\n", crm_long_options[i].desc); } else { /* is val printable as char ? */ if (crm_long_options[i].val && crm_long_options[i].val <= UCHAR_MAX) { fprintf(stream, " -%c,", crm_long_options[i].val); } else { fputs(" ", stream); } fprintf(stream, " --%s%s\t%s\n", crm_long_options[i].name, crm_long_options[i].has_arg == optional_argument ? "[=value]" : crm_long_options[i].has_arg == required_argument ? "=value" : "", crm_long_options[i].desc ? crm_long_options[i].desc : ""); } } } else if (crm_short_options) { fprintf(stream, "Usage: %s - %s\n", crm_system_name, crm_app_description); for (i = 0; crm_short_options[i] != 0; i++) { int has_arg = no_argument /* 0 */; if (crm_short_options[i + 1] == ':') { if (crm_short_options[i + 2] == ':') has_arg = optional_argument /* 2 */; else has_arg = required_argument /* 1 */; } fprintf(stream, " -%c %s\n", crm_short_options[i], has_arg == optional_argument ? "[value]" : has_arg == required_argument ? "{value}" : ""); i += has_arg; } } fprintf(stream, "\nReport bugs to %s\n", PACKAGE_BUGREPORT); out: crm_exit(exit_code); while(1); // above does not return } void cib_ipc_servers_init(qb_ipcs_service_t **ipcs_ro, qb_ipcs_service_t **ipcs_rw, qb_ipcs_service_t **ipcs_shm, struct qb_ipcs_service_handlers *ro_cb, struct qb_ipcs_service_handlers *rw_cb) { *ipcs_ro = mainloop_add_ipc_server(CIB_CHANNEL_RO, QB_IPC_NATIVE, ro_cb); *ipcs_rw = mainloop_add_ipc_server(CIB_CHANNEL_RW, QB_IPC_NATIVE, rw_cb); *ipcs_shm = mainloop_add_ipc_server(CIB_CHANNEL_SHM, QB_IPC_SHM, rw_cb); if (*ipcs_ro == NULL || *ipcs_rw == NULL || *ipcs_shm == NULL) { crm_err("Failed to create the CIB manager: exiting and inhibiting respawn"); crm_warn("Verify pacemaker and pacemaker_remote are not both enabled"); crm_exit(CRM_EX_FATAL); } } void cib_ipc_servers_destroy(qb_ipcs_service_t *ipcs_ro, qb_ipcs_service_t *ipcs_rw, qb_ipcs_service_t *ipcs_shm) { qb_ipcs_destroy(ipcs_ro); qb_ipcs_destroy(ipcs_rw); qb_ipcs_destroy(ipcs_shm); } qb_ipcs_service_t * crmd_ipc_server_init(struct qb_ipcs_service_handlers *cb) { return mainloop_add_ipc_server(CRM_SYSTEM_CRMD, QB_IPC_NATIVE, cb); } void attrd_ipc_server_init(qb_ipcs_service_t **ipcs, struct qb_ipcs_service_handlers *cb) { *ipcs = mainloop_add_ipc_server(T_ATTRD, QB_IPC_NATIVE, cb); if (*ipcs == NULL) { crm_err("Failed to create pacemaker-attrd server: exiting and inhibiting respawn"); crm_warn("Verify pacemaker and pacemaker_remote are not both enabled."); crm_exit(CRM_EX_FATAL); } } void stonith_ipc_server_init(qb_ipcs_service_t **ipcs, struct qb_ipcs_service_handlers *cb) { *ipcs = mainloop_add_ipc_server_with_prio("stonith-ng", QB_IPC_NATIVE, cb, QB_LOOP_HIGH); if (*ipcs == NULL) { crm_err("Failed to create fencer: exiting and inhibiting respawn."); crm_warn("Verify pacemaker and pacemaker_remote are not both enabled."); crm_exit(CRM_EX_FATAL); } } void * find_library_function(void **handle, const char *lib, const char *fn, gboolean fatal) { char *error; void *a_function; if (*handle == NULL) { *handle = dlopen(lib, RTLD_LAZY); } if (!(*handle)) { crm_err("%sCould not open %s: %s", fatal ? "Fatal: " : "", lib, dlerror()); if (fatal) { crm_exit(CRM_EX_FATAL); } return NULL; } a_function = dlsym(*handle, fn); if (a_function == NULL) { error = dlerror(); crm_err("%sCould not find %s in %s: %s", fatal ? "Fatal: " : "", fn, lib, error); if (fatal) { crm_exit(CRM_EX_FATAL); } } return a_function; } #ifdef HAVE_UUID_UUID_H # include #endif char * crm_generate_uuid(void) { unsigned char uuid[16]; char *buffer = malloc(37); /* Including NUL byte */ uuid_generate(uuid); uuid_unparse(uuid, buffer); return buffer; } /*! * \brief Get name to be used as identifier for cluster messages * * \param[in] name Actual system name to check * * \return Non-NULL cluster message identifier corresponding to name * * \note The Pacemaker daemons were renamed in version 2.0.0, but the old names * must continue to be used as the identifier for cluster messages, so * that mixed-version clusters are possible during a rolling upgrade. */ const char * pcmk_message_name(const char *name) { if (name == NULL) { return "unknown"; } else if (!strcmp(name, "pacemaker-attrd")) { return "attrd"; } else if (!strcmp(name, "pacemaker-based")) { return CRM_SYSTEM_CIB; } else if (!strcmp(name, "pacemaker-controld")) { return CRM_SYSTEM_CRMD; } else if (!strcmp(name, "pacemaker-execd")) { return CRM_SYSTEM_LRMD; } else if (!strcmp(name, "pacemaker-fenced")) { return "stonith-ng"; } else if (!strcmp(name, "pacemaker-schedulerd")) { return CRM_SYSTEM_PENGINE; } else { return name; } } /*! * \brief Check whether a string represents a cluster daemon name * * \param[in] name String to check * * \return TRUE if name is standard client name used by daemons, FALSE otherwise */ bool crm_is_daemon_name(const char *name) { name = pcmk_message_name(name); return (!strcmp(name, CRM_SYSTEM_CRMD) || !strcmp(name, CRM_SYSTEM_STONITHD) || !strcmp(name, "stonith-ng") || !strcmp(name, "attrd") || !strcmp(name, CRM_SYSTEM_CIB) || !strcmp(name, CRM_SYSTEM_MCP) || !strcmp(name, CRM_SYSTEM_DC) || !strcmp(name, CRM_SYSTEM_TENGINE) || !strcmp(name, CRM_SYSTEM_LRMD)); } #include char * crm_md5sum(const char *buffer) { int lpc = 0, len = 0; char *digest = NULL; unsigned char raw_digest[MD5_DIGEST_SIZE]; if (buffer == NULL) { buffer = ""; } len = strlen(buffer); crm_trace("Beginning digest of %d bytes", len); digest = malloc(2 * MD5_DIGEST_SIZE + 1); if(digest) { md5_buffer(buffer, len, raw_digest); for (lpc = 0; lpc < MD5_DIGEST_SIZE; lpc++) { sprintf(digest + (2 * lpc), "%02x", raw_digest[lpc]); } digest[(2 * MD5_DIGEST_SIZE)] = 0; crm_trace("Digest %s.", digest); } else { crm_err("Could not create digest"); } return digest; } #ifdef HAVE_GNUTLS_GNUTLS_H void crm_gnutls_global_init(void) { signal(SIGPIPE, SIG_IGN); gnutls_global_init(); } #endif /*! * \brief Get the local hostname * * \return Newly allocated string with name, or NULL (and set errno) on error */ char * pcmk_hostname() { struct utsname hostinfo; return (uname(&hostinfo) < 0)? NULL : strdup(hostinfo.nodename); } diff --git a/lib/common/watchdog.c b/lib/common/watchdog.c index c73c8ab58f..23155c5245 100644 --- a/lib/common/watchdog.c +++ b/lib/common/watchdog.c @@ -1,269 +1,274 @@ /* * Copyright 2013-2019 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #include #include #ifdef _POSIX_MEMLOCK # include #endif -static int sbd_pid = 0; +static pid_t sbd_pid = 0; enum pcmk_panic_flags { pcmk_panic_none = 0x00, pcmk_panic_delay = 0x01, pcmk_panic_kdump = 0x02, pcmk_panic_shutdown = 0x04, }; static void sysrq_trigger(char t) { #if SUPPORT_PROCFS FILE *procf; // Root can always write here, regardless of kernel.sysrq value procf = fopen("/proc/sysrq-trigger", "a"); if (!procf) { crm_perror(LOG_WARNING, "Opening sysrq-trigger failed"); return; } crm_info("sysrq-trigger: %c", t); fprintf(procf, "%c\n", t); fclose(procf); #endif // SUPPORT_PROCFS return; } static void pcmk_panic_local(void) { int rc = pcmk_ok; uid_t uid = geteuid(); pid_t ppid = getppid(); if(uid != 0 && ppid > 1) { /* We're a non-root pacemaker daemon (pacemaker-based, * pacemaker-controld, pacemaker-schedulerd, pacemaker-attrd, etc.) with * the original pacemakerd parent. * * Of these, only the controller is likely to be initiating resets. */ do_crm_log_always(LOG_EMERG, "Signaling parent %d to panic", ppid); crm_exit(CRM_EX_PANIC); return; } else if (uid != 0) { #if SUPPORT_PROCFS /* * No permissions, and no pacemakerd parent to escalate to. * Track down the new pacemakerd process and send a signal instead. */ union sigval signal_value; memset(&signal_value, 0, sizeof(signal_value)); ppid = crm_procfs_pid_of("pacemakerd"); do_crm_log_always(LOG_EMERG, "Signaling pacemakerd(%d) to panic", ppid); if(ppid > 1 && sigqueue(ppid, SIGQUIT, signal_value) < 0) { crm_perror(LOG_EMERG, "Cannot signal pacemakerd(%d) to panic", ppid); } #endif // SUPPORT_PROCFS /* The best we can do now is die */ crm_exit(CRM_EX_PANIC); return; } /* We're either pacemakerd, or a pacemaker daemon running as root */ if (safe_str_eq("crash", getenv("PCMK_panic_action"))) { sysrq_trigger('c'); } else { sysrq_trigger('b'); } /* reboot(RB_HALT_SYSTEM); rc = errno; */ reboot(RB_AUTOBOOT); rc = errno; do_crm_log_always(LOG_EMERG, "Reboot failed, escalating to %d: %s (%d)", ppid, pcmk_strerror(rc), rc); if(ppid > 1) { /* child daemon */ exit(CRM_EX_PANIC); } else { /* pacemakerd or orphan child */ exit(CRM_EX_FATAL); } } static void pcmk_panic_sbd(void) { union sigval signal_value; pid_t ppid = getppid(); - do_crm_log_always(LOG_EMERG, "Signaling sbd(%d) to panic", sbd_pid); + do_crm_log_always(LOG_EMERG, "Signaling sbd[%lld] to panic", + (long long) sbd_pid); memset(&signal_value, 0, sizeof(signal_value)); /* TODO: Arrange for a slightly less brutal option? */ if(sigqueue(sbd_pid, SIGKILL, signal_value) < 0) { - crm_perror(LOG_EMERG, "Cannot signal SBD(%d) to terminate", sbd_pid); + crm_perror(LOG_EMERG, "Cannot signal sbd[%lld] to terminate", + (long long) sbd_pid); pcmk_panic_local(); } if(ppid > 1) { /* child daemon */ exit(CRM_EX_PANIC); } else { /* pacemakerd or orphan child */ exit(CRM_EX_FATAL); } } 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()); + "Shutting down instead of panicking the node " + CRM_XS " origin=%s sbd=%lld parent=%d", + origin, (long long) sbd_pid, getppid()); crm_exit(CRM_EX_FATAL); return; } if(sbd_pid > 1) { - do_crm_log_always(LOG_EMERG, "Signaling sbd(%d) to panic the system: %s", sbd_pid, origin); + do_crm_log_always(LOG_EMERG, "Signaling sbd[%lld] to panic the system: %s", + (long long) sbd_pid, origin); pcmk_panic_sbd(); } else { do_crm_log_always(LOG_EMERG, "Panicking the system directly: %s", origin); pcmk_panic_local(); } } pid_t pcmk_locate_sbd(void) { char *pidfile = NULL; char *sbd_path = NULL; + int rc; if(sbd_pid > 1) { return sbd_pid; } /* Look for the pid file */ pidfile = crm_strdup_printf(PCMK_RUN_DIR "/sbd.pid"); sbd_path = crm_strdup_printf("%s/sbd", SBIN_DIR); /* Read the pid file */ - CRM_ASSERT(pidfile); - - sbd_pid = crm_pidfile_inuse(pidfile, 0, sbd_path); - if(sbd_pid > 0) { - crm_trace("SBD detected at pid=%d (file)", sbd_pid); + rc = pcmk__pidfile_matches(pidfile, 0, sbd_path, &sbd_pid); + if (rc == pcmk_rc_ok) { + crm_trace("SBD detected at pid %lld (via PID file %s)", + (long long) sbd_pid, pidfile); #if SUPPORT_PROCFS } else { /* Fall back to /proc for systems that support it */ - sbd_pid = crm_procfs_pid_of("sbd"); - crm_trace("SBD detected at pid=%d (proc)", sbd_pid); + sbd_pid = (pid_t) crm_procfs_pid_of("sbd"); + crm_trace("SBD detected at pid %lld (via procfs)", + (long long) sbd_pid); #endif // SUPPORT_PROCFS } if(sbd_pid < 0) { sbd_pid = 0; crm_trace("SBD not detected"); } free(pidfile); free(sbd_path); return sbd_pid; } long crm_get_sbd_timeout(void) { static long sbd_timeout = -2; if (sbd_timeout == -2) { sbd_timeout = crm_get_msec(getenv("SBD_WATCHDOG_TIMEOUT")); } return sbd_timeout; } long crm_auto_watchdog_timeout() { long sbd_timeout = crm_get_sbd_timeout(); return (sbd_timeout <= 0)? 0 : (2 * sbd_timeout); } gboolean check_sbd_timeout(const char *value) { long st_timeout = value? crm_get_msec(value) : 0; if (st_timeout < 0) { st_timeout = crm_auto_watchdog_timeout(); crm_debug("Using calculated value %ld for stonith-watchdog-timeout (%s)", st_timeout, value); } 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? value : "auto")); crm_exit(CRM_EX_FATAL); 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(CRM_EX_FATAL); return FALSE; } crm_info("Watchdog configured with stonith-watchdog-timeout %s and SBD timeout %ldms", value, sbd_timeout); } return TRUE; }