diff --git a/daemons/pacemakerd/pcmkd_subdaemons.c b/daemons/pacemakerd/pcmkd_subdaemons.c index 15ef6a4977..3294ee7c70 100644 --- a/daemons/pacemakerd/pcmkd_subdaemons.c +++ b/daemons/pacemakerd/pcmkd_subdaemons.c @@ -1,944 +1,941 @@ /* * Copyright 2010-2025 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" #if SUPPORT_COROSYNC #include "pcmkd_corosync.h" #endif #include #include #include #include #include #include #include #include #include #include #include #include enum child_daemon_flags { child_none = 0, child_respawn = 1 << 0, child_needs_cluster = 1 << 1, child_needs_retry = 1 << 2, child_active_before_startup = 1 << 3, child_shutting_down = 1 << 4, }; typedef struct pcmk_child_s { enum pcmk_ipc_server server; pid_t pid; int respawn_count; const char *uid; int check_count; uint32_t flags; } pcmk_child_t; #define PCMK_PROCESS_CHECK_INTERVAL 1000 /* 1s */ #define PCMK_PROCESS_CHECK_RETRIES 5 #define SHUTDOWN_ESCALATION_PERIOD 180000 /* 3m */ /* Index into the array below */ #define PCMK_CHILD_CONTROLD 5 static pcmk_child_t pcmk_children[] = { { pcmk_ipc_based, 0, 0, CRM_DAEMON_USER, 0, child_respawn | child_needs_cluster }, { pcmk_ipc_fenced, 0, 0, NULL, 0, child_respawn | child_needs_cluster }, { pcmk_ipc_execd, 0, 0, NULL, 0, child_respawn }, { pcmk_ipc_attrd, 0, 0, CRM_DAEMON_USER, 0, child_respawn | child_needs_cluster }, { pcmk_ipc_schedulerd, 0, 0, CRM_DAEMON_USER, 0, child_respawn }, { pcmk_ipc_controld, 0, 0, CRM_DAEMON_USER, 0, child_respawn | child_needs_cluster }, }; crm_trigger_t *shutdown_trigger = NULL; crm_trigger_t *startup_trigger = NULL; time_t subdaemon_check_progress = 0; // Whether we need root group access to talk to cluster layer static bool need_root_group = true; /* When contacted via pacemakerd-api by a client having sbd in * the name we assume it is sbd-daemon which wants to know * if pacemakerd shutdown gracefully. * Thus when everything is shutdown properly pacemakerd * waits till it has reported the graceful completion of * shutdown to sbd and just when sbd-client closes the * connection we can assume that the report has arrived * properly so that pacemakerd can finally exit. * Following two variables are used to track that handshake. */ unsigned int shutdown_complete_state_reported_to = 0; bool shutdown_complete_state_reported_client_closed = false; /* state we report when asked via pacemakerd-api status-ping */ const char *pacemakerd_state = PCMK__VALUE_INIT; bool running_with_sbd = false; GMainLoop *mainloop = NULL; static bool fatal_error = false; static int child_liveness(pcmk_child_t *child); static gboolean escalate_shutdown(gpointer data); static int start_child(pcmk_child_t * child); static void pcmk_child_exit(mainloop_child_t * p, pid_t pid, int core, int signo, int exitcode); static void pcmk_process_exit(pcmk_child_t * child); static gboolean pcmk_shutdown_worker(gpointer user_data); static void stop_child(pcmk_child_t *child, int signal); static void for_each_child(void (*fn)(pcmk_child_t *child)) { for (int i = 0; i < PCMK__NELEM(pcmk_children); i++) { fn(&pcmk_children[i]); } } /*! * \internal * \brief Get path to subdaemon executable * * \param[in] subdaemon Subdaemon to get path for * * \return Newly allocated string with path to subdaemon executable * \note It is the caller's responsibility to free() the return value */ static inline char * subdaemon_path(pcmk_child_t *subdaemon) { return crm_strdup_printf(CRM_DAEMON_DIR "/%s", pcmk__server_name(subdaemon->server)); } static bool pcmkd_cluster_connected(void) { #if SUPPORT_COROSYNC return pcmkd_corosync_connected(); #else return true; #endif } static gboolean check_next_subdaemon(gpointer user_data) { static int next_child = 0; pcmk_child_t *child = &(pcmk_children[next_child]); const char *name = pcmk__server_name(child->server); const long long pid = PCMK__SPECIAL_PID_AS_0(child->pid); int rc = child_liveness(child); crm_trace("Checked subdaemon %s[%lld]: %s (%d)", name, pid, pcmk_rc_str(rc), rc); switch (rc) { case pcmk_rc_ok: child->check_count = 0; subdaemon_check_progress = time(NULL); break; case pcmk_rc_ipc_pid_only: // Child was previously OK if (pcmk_is_set(child->flags, child_shutting_down)) { crm_notice("Subdaemon %s[%lld] has stopped accepting IPC " "connections during shutdown", name, pid); } else if (++(child->check_count) >= PCMK_PROCESS_CHECK_RETRIES) { // cts-lab looks for this message crm_crit("Subdaemon %s[%lld] is unresponsive to IPC " "after %d attempt%s and will now be killed", name, pid, child->check_count, pcmk__plural_s(child->check_count)); stop_child(child, SIGKILL); if (pcmk_is_set(child->flags, child_respawn)) { // Respawn limit hasn't been reached, so retry another round child->check_count = 0; } } else { crm_notice("Subdaemon %s[%lld] is unresponsive to IPC " "after %d attempt%s (will recheck later)", name, pid, child->check_count, pcmk__plural_s(child->check_count)); if (pcmk_is_set(child->flags, child_respawn)) { /* as long as the respawn-limit isn't reached and we haven't run out of connect retries we account this as progress we are willing to tell to sbd */ subdaemon_check_progress = time(NULL); } } /* go to the next child and see if we can make progress there */ break; case pcmk_rc_ipc_unresponsive: if (!pcmk_is_set(child->flags, child_respawn)) { /* if a subdaemon is down and we don't want it to be restarted this is a success during shutdown. if it isn't restarted anymore due to MAX_RESPAWN it is rather no success. */ if (child->respawn_count <= MAX_RESPAWN) { subdaemon_check_progress = time(NULL); } } if (!pcmk_is_set(child->flags, child_active_before_startup)) { crm_trace("Subdaemon %s[%lld] terminated", name, pid); break; } if (pcmk_is_set(child->flags, child_respawn)) { // cts-lab looks for this message crm_err("Subdaemon %s[%lld] terminated", name, pid); } else { /* orderly shutdown */ crm_notice("Subdaemon %s[%lld] terminated", name, pid); } pcmk_process_exit(child); break; default: crm_exit(CRM_EX_FATAL); break; /* static analysis/noreturn */ } if (++next_child >= PCMK__NELEM(pcmk_children)) { next_child = 0; } return G_SOURCE_CONTINUE; } 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 != 0) { /* Use SIGSEGV instead of SIGKILL to create a core so we can see what it was up to */ crm_err("Subdaemon %s not terminating in a timely manner, forcing", pcmk__server_name(child->server)); stop_child(child, SIGSEGV); } return G_SOURCE_REMOVE; } 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) { // cts-lab looks for this message do_crm_log(((signo == SIGKILL)? LOG_WARNING : LOG_ERR), "%s[%d] terminated with signal %d (%s)%s", name, pid, signo, strsignal(signo), (core? " and dumped core" : "")); pcmk_process_exit(child); return; } 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->flags &= ~child_respawn; fatal_error = true; pcmk_shutdown(SIGTERM); break; case CRM_EX_PANIC: { char *msg = NULL; child->flags &= ~child_respawn; fatal_error = true; msg = crm_strdup_printf("Subdaemon %s[%d] requested panic", name, pid); pcmk__panic(msg); // Should never get here free(msg); pcmk_shutdown(SIGTERM); } break; default: // cts-lab looks for this message crm_err("%s[%d] exited with status %d (%s)", name, pid, exitcode, crm_exit_str(exitcode)); break; } pcmk_process_exit(child); } static void pcmk_process_exit(pcmk_child_t * child) { const char *name = pcmk__server_name(child->server); child->pid = 0; child->flags &= ~child_active_before_startup; child->check_count = 0; child->respawn_count += 1; if (child->respawn_count > MAX_RESPAWN) { crm_err("Subdaemon %s exceeded maximum respawn count", name); child->flags &= ~child_respawn; } if (shutdown_trigger) { /* resume step-wise shutdown (returned TRUE yields no parallelizing) */ mainloop_set_trigger(shutdown_trigger); } else if (!pcmk_is_set(child->flags, child_respawn)) { /* nothing to do */ } else if (crm_is_true(pcmk__env_option(PCMK__ENV_FAIL_FAST))) { pcmk__panic("Subdaemon failed"); } else if (child_liveness(child) == pcmk_rc_ok) { crm_warn("Not respawning subdaemon %s because IPC endpoint %s is OK", name, pcmk__server_ipc_name(child->server)); } else if (pcmk_is_set(child->flags, child_needs_cluster) && !pcmkd_cluster_connected()) { crm_notice("Not respawning subdaemon %s until cluster returns", name); child->flags |= child_needs_retry; } else { // cts-lab looks for this message crm_notice("Respawning subdaemon %s after unexpected exit", name); start_child(child); } } static gboolean pcmk_shutdown_worker(gpointer user_data) { static int phase = PCMK__NELEM(pcmk_children) - 1; static time_t next_log = 0; if (phase == PCMK__NELEM(pcmk_children) - 1) { crm_notice("Shutting down Pacemaker"); pacemakerd_state = PCMK__VALUE_SHUTTING_DOWN; } for (; phase >= 0; phase--) { pcmk_child_t *child = &(pcmk_children[phase]); const char *name = pcmk__server_name(child->server); time_t now = 0; if (child->pid == 0) { /* cleanup */ crm_debug("Subdaemon %s confirmed stopped", name); child->pid = 0; continue; } now = time(NULL); if (pcmk_is_set(child->flags, child_respawn)) { if (child->pid == PCMK__SPECIAL_PID) { crm_warn("Subdaemon %s cannot be terminated (shutdown " "will be escalated after %ld seconds if it does " "not terminate on its own; set PCMK_" PCMK__ENV_FAIL_FAST "=1 to exit immediately " "instead)", name, (long) SHUTDOWN_ESCALATION_PERIOD); } next_log = now + 30; child->flags &= ~child_respawn; stop_child(child, SIGTERM); if (phase < PCMK_CHILD_CONTROLD) { pcmk__create_timer(SHUTDOWN_ESCALATION_PERIOD, escalate_shutdown, child); } } else if (now >= next_log) { next_log = now + 30; crm_notice("Still waiting for subdaemon %s to terminate " QB_XS " pid=%lld", name, (long long) child->pid); } return G_SOURCE_CONTINUE; } crm_notice("Shutdown complete"); pacemakerd_state = PCMK__VALUE_SHUTDOWN_COMPLETE; if (!fatal_error && running_with_sbd && pcmk__get_sbd_sync_resource_startup() && !shutdown_complete_state_reported_client_closed) { crm_notice("Waiting for SBD to pick up shutdown-complete-state."); return G_SOURCE_CONTINUE; } g_main_loop_quit(mainloop); if (fatal_error) { crm_notice("Shutting down and staying down after fatal error"); #if SUPPORT_COROSYNC pcmkd_shutdown_corosync(); #endif crm_exit(CRM_EX_FATAL); } return G_SOURCE_CONTINUE; } /* 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 */ // \return Standard Pacemaker return code static int start_child(pcmk_child_t * child) { uid_t uid = 0; gid_t gid = 0; bool use_valgrind = false; bool use_callgrind = false; const char *name = pcmk__server_name(child->server); const char *env_valgrind = pcmk__env_option(PCMK__ENV_VALGRIND_ENABLED); const char *env_callgrind = pcmk__env_option(PCMK__ENV_CALLGRIND_ENABLED); child->flags &= ~(child_active_before_startup | child_shutting_down); child->check_count = 0; if (env_callgrind != NULL && crm_is_true(env_callgrind)) { use_callgrind = true; use_valgrind = true; } else if ((env_callgrind != NULL) && (strstr(env_callgrind, name) != NULL)) { 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, name) != NULL)) { use_valgrind = true; } if (use_valgrind && strlen(PCMK__VALGRIND_EXEC) == 0) { crm_warn("Cannot enable valgrind for subdaemon %s: valgrind not found", name); use_valgrind = false; } if ((child->uid != NULL) && (crm_user_lookup(child->uid, &uid, &gid) < 0)) { crm_err("Invalid user (%s) for subdaemon %s: not found", child->uid, name); return EACCES; } child->pid = fork(); pcmk__assert(child->pid != -1); if (child->pid > 0) { /* parent */ mainloop_child_add(child->pid, 0, name, child, pcmk_child_exit); if (use_valgrind) { crm_info("Forked process %lld using user %lu (%s) and group %lu " "for subdaemon %s (valgrind enabled: %s)", (long long) child->pid, (unsigned long) uid, pcmk__s(child->uid, "root"), (unsigned long) gid, name, PCMK__VALGRIND_EXEC); } else { crm_info("Forked process %lld using user %lu (%s) and group %lu " "for subdaemon %s", (long long) child->pid, (unsigned long) uid, pcmk__s(child->uid, "root"), (unsigned long) gid, name); } return pcmk_rc_ok; } else { char *path = subdaemon_path(child); /* Start a new session */ setsid(); if(gid) { // Drop root group access if not needed if (!need_root_group && (setgid(gid) < 0)) { crm_warn("Could not set subdaemon %s group to %lu: %s", name, (unsigned long) gid, strerror(errno)); } /* 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 system groups for subdaemon %s: %s " QB_XS " errno=%d", name, pcmk_rc_str(errno), errno); } } if (uid && setuid(uid) < 0) { crm_warn("Could not set subdaemon %s user to %s: %s " QB_XS " uid=%lu errno=%d", name, strerror(errno), child->uid, (unsigned long) uid, errno); } - pcmk__close_fds_in_child(true); - - pcmk__open_devnull(O_RDONLY); // stdin (fd 0) - pcmk__open_devnull(O_WRONLY); // stdout (fd 1) - pcmk__open_devnull(O_WRONLY); // stderr (fd 2) + pcmk__close_fds_in_child(); + pcmk__null_std_streams(); if (use_callgrind) { char *out_file = pcmk__str_copy("--callgrind-out-file=" CRM_STATE_DIR "/callgrind.opt.%p"); execlp(PCMK__VALGRIND_EXEC, PCMK__VALGRIND_EXEC, "--tool=callgrind", out_file, path, (char *) NULL); free(out_file); } else if (use_valgrind) { execlp(PCMK__VALGRIND_EXEC, PCMK__VALGRIND_EXEC, path, (char *) NULL); } else { execlp(path, path, (char *) NULL); } free(path); crm_crit("Could not execute subdaemon %s: %s", name, strerror(errno)); crm_exit(CRM_EX_FATAL); } return pcmk_rc_ok; /* never reached */ } /*! * \internal * \brief Check the liveness of the child based on IPC name and PID if tracked * * \param[in,out] 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) { 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; const char *name = pcmk__server_name(child->server); int rc = pcmk_rc_ipc_unresponsive; int pid_active = pcmk_rc_ok; int legacy_rc = pcmk_ok; pid_t ipc_pid = 0; 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 " QB_XS " rc=%d", CRM_DAEMON_USER, pcmk_rc_str(rc), rc); } else { const char *ipc_name = pcmk__server_ipc_name(child->server); rc = pcmk__ipc_is_authentic_process_active(ipc_name, *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) { return rc; } /* 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. */ pid_active = pcmk__pid_active(child->pid, name); if ((ipc_pid != 0) && ((pid_active != pcmk_rc_ok) || (ipc_pid == PCMK__SPECIAL_PID) || (pcmk__pid_active(ipc_pid, 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 (pid_active == 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 (pid_active == 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 == 0) && (pid_active == EINVAL)) { // FreeBSD can return EINVAL rc = pcmk_rc_ipc_unresponsive; } else if (pid_active == EACCES) { rc = pcmk_rc_ipc_unauthorized; } else if (pid_active == ESRCH) { rc = pcmk_rc_ipc_unresponsive; } else { rc = pid_active; } return rc; } static void reset_respawn_count(pcmk_child_t *child) { /* Restore pristine state */ child->respawn_count = 0; } #define WAIT_TRIES 4 /* together with interleaved sleeps, worst case ~ 1s */ static int child_up_but_no_ipc(pcmk_child_t *child) { const char *ipc_name = pcmk__server_ipc_name(child->server); if (child->respawn_count == WAIT_TRIES) { crm_crit("%s IPC endpoint for existing process %lld did not (re)appear", ipc_name, (long long) PCMK__SPECIAL_PID_AS_0(child->pid)); return pcmk_rc_ipc_pid_only; } crm_warn("Cannot find %s IPC endpoint for existing process %ld, could still " "reappear in %d attempts", ipc_name, (long long) PCMK__SPECIAL_PID_AS_0(child->pid), WAIT_TRIES - child->respawn_count); return EAGAIN; } static int child_alive(pcmk_child_t *child) { const char *name = pcmk__server_name(child->server); if (child->pid == PCMK__SPECIAL_PID) { if (crm_is_true(pcmk__env_option(PCMK__ENV_FAIL_FAST))) { crm_crit("Cannot track pre-existing process for %s IPC on this " "platform and PCMK_" PCMK__ENV_FAIL_FAST " requested", name); return EOPNOTSUPP; } else if (child->respawn_count == WAIT_TRIES) { /* Because PCMK__ENV_FAIL_FAST wasn't requested, we can't bail * out. Instead, switch to IPC liveness monitoring which is not * very suitable for heavy system load. */ crm_notice("Cannot track pre-existing process for %s IPC on this " "platform but assuming it is stable and using liveness " "monitoring", name); crm_warn("The process for %s IPC cannot be terminated, so " "shutdown will be delayed by %d s to allow time for it " "to terminate on its own", name, SHUTDOWN_ESCALATION_PERIOD); } else { crm_warn("Cannot track pre-existing process for %s IPC on this " "platform; checking %d more times", name, WAIT_TRIES - child->respawn_count); return EAGAIN; } } crm_notice("Tracking existing %s process (pid=%lld)", name, (long long) PCMK__SPECIAL_PID_AS_0(child->pid)); child->respawn_count = -1; /* 0~keep watching */ child->flags |= child_active_before_startup; return pcmk_rc_ok; } static int find_and_track_child(pcmk_child_t *child, int rounds, bool *wait_in_progress) { int rc = pcmk_rc_ok; const char *name = pcmk__server_name(child->server); if (child->respawn_count < 0) { return EAGAIN; } rc = child_liveness(child); 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. */ return EAGAIN; } child->respawn_count = rounds; if (rc == pcmk_rc_ok) { rc = child_alive(child); if (rc == EAGAIN) { *wait_in_progress = true; } } else if (rc == pcmk_rc_ipc_pid_only) { rc = child_up_but_no_ipc(child); if (rc == EAGAIN) { *wait_in_progress = true; } } else { crm_crit("Checked liveness of %s: %s " QB_XS " rc=%d", name, pcmk_rc_str(rc), rc); } return rc; } /*! * \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. */ int find_and_track_existing_processes(void) { bool wait_in_progress; size_t i, rounds; for (rounds = 1; rounds <= WAIT_TRIES; rounds++) { wait_in_progress = false; for (i = 0; i < PCMK__NELEM(pcmk_children); i++) { int rc = find_and_track_child(&pcmk_children[i], rounds, &wait_in_progress); if (rc == pcmk_rc_ok) { break; } else if (rc != EAGAIN) { return rc; } } if (!wait_in_progress) { break; } pcmk__sleep_ms(250); // Wait a bit for changes to possibly happen } for_each_child(reset_respawn_count); pcmk__create_timer(PCMK_PROCESS_CHECK_INTERVAL, check_next_subdaemon, NULL); return pcmk_rc_ok; } static void start_subdaemon(pcmk_child_t *child) { if (child->pid != 0) { /* We are already tracking this process */ return; } start_child(child); } gboolean init_children_processes(void *user_data) { if (pcmk_get_cluster_layer() == pcmk_cluster_layer_corosync) { /* 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; } /* start any children that have not been detected */ for_each_child(start_subdaemon); /* 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 */ pcmk__set_env_option(PCMK__ENV_RESPAWNED, PCMK_VALUE_TRUE, false); pacemakerd_state = PCMK__VALUE_RUNNING; return G_SOURCE_CONTINUE; } 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 void restart_subdaemon(pcmk_child_t *child) { if (!pcmk_is_set(child->flags, child_needs_retry) || child->pid != 0) { return; } crm_notice("Respawning cluster-based subdaemon %s", pcmk__server_name(child->server)); if (start_child(child)) { child->flags &= ~child_needs_retry; } } void restart_cluster_subdaemons(void) { for_each_child(restart_subdaemon); } static void stop_child(pcmk_child_t *child, int signal) { const char *name = pcmk__server_name(child->server); 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->pid == PCMK__SPECIAL_PID) { crm_debug("Nothing to do to stop subdaemon %s[%lld]", name, (long long) PCMK__SPECIAL_PID_AS_0(child->pid)); return; } if (child->pid <= 0) { crm_trace("Nothing to do to stop subdaemon %s: Not running", name); return; } errno = 0; if (kill(child->pid, signal) == 0) { crm_notice("Stopping subdaemon %s " QB_XS " via signal %d to process %lld", name, signal, (long long) child->pid); child->flags |= child_shutting_down; } else { crm_err("Could not stop subdaemon %s[%lld] with signal %d: %s", name, (long long) child->pid, signal, strerror(errno)); } } diff --git a/include/crm/common/cmdline_internal.h b/include/crm/common/cmdline_internal.h index 069bbb45da..d60ba7fca1 100644 --- a/include/crm/common/cmdline_internal.h +++ b/include/crm/common/cmdline_internal.h @@ -1,190 +1,168 @@ /* - * Copyright 2019-2024 the Pacemaker project contributors + * Copyright 2019-2025 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 PCMK__CRM_COMMON_CMDLINE_INTERNAL__H #define PCMK__CRM_COMMON_CMDLINE_INTERNAL__H #include #ifdef __cplusplus extern "C" { #endif typedef struct { char *summary; char *output_as_descr; gboolean version; gboolean quiet; unsigned int verbosity; char *output_ty; char *output_dest; } pcmk__common_args_t; /*! * \internal * \brief Allocate a new common args object * * \param[in] summary Summary description of tool for man page * * \return Newly allocated common args object * \note This function will immediately exit the program if memory allocation * fails, since the intent is to call it at the very beginning of a * program, before logging has been set up. */ pcmk__common_args_t * pcmk__new_common_args(const char *summary); /*! * \internal * \brief Create and return a GOptionContext containing the command line options * supported by all tools. * * \note Formatted output options will be added unless fmts is NULL. This allows * for using this function in tools that have not yet been converted to * formatted output. It should not be NULL in any tool that calls * pcmk__register_formats() as that function adds its own command line * options. * * \param[in,out] common_args A ::pcmk__common_args_t structure where the * results of handling command options will be written. * \param[in] fmts The help string for which formats are supported. * \param[in,out] output_group A ::GOptionGroup that formatted output related * command line arguments should be added to. * \param[in] param_string A string describing any remaining command line * arguments. */ GOptionContext * pcmk__build_arg_context(pcmk__common_args_t *common_args, const char *fmts, GOptionGroup **output_group, const char *param_string); /*! * \internal * \brief Clean up after pcmk__build_arg_context(). This should be called * instead of ::g_option_context_free at program termination. * * \param[in,out] context Argument context to free */ void pcmk__free_arg_context(GOptionContext *context); /*! * \internal * \brief Add options to the main application options * * \param[in,out] context Argument context to add options to * \param[in] entries Option entries to add * * \note This is simply a convenience wrapper to reduce duplication */ void pcmk__add_main_args(GOptionContext *context, const GOptionEntry entries[]); /*! * \internal * \brief Add an option group to an argument context * * \param[in,out] context Argument context to add group to * \param[in] name Option group name (to be used in --help-NAME) * \param[in] header Header for --help-NAME output * \param[in] desc Short description for --help-NAME option * \param[in] entries Array of options in group * * \note This is simply a convenience wrapper to reduce duplication */ void pcmk__add_arg_group(GOptionContext *context, const char *name, const char *header, const char *desc, const GOptionEntry entries[]); /*! * \internal * \brief Prepare the command line for being added to a pcmk__output_t as the * request * * This performs various transformations on the command line arguments, such * as surrounding arguments containing spaces with quotes and escaping any * single quotes in the string. * * \param[in,out] argv Command line (typically from pcmk__cmdline_preproc()) */ gchar *pcmk__quote_cmdline(gchar **argv); /*! * \internal * \brief Pre-process command line arguments to preserve compatibility with * getopt behavior. * * getopt and glib have slightly different behavior when it comes to processing * single command line arguments. getopt allows this: -x, while glib will * try to handle like it is additional single letter arguments. glib * prefers -x instead. * * This function scans argv, looking for any single letter command line options * (indicated by the 'special' parameter). When one is found, everything after * that argument to the next whitespace is converted into its own value. Single * letter command line options can come in a group after a single dash, but * this function will expand each group into many arguments. * * Long options and anything after "--" is preserved. The result of this function * can then be passed to ::g_option_context_parse_strv for actual processing. * * In pseudocode, this: * * pcmk__cmdline_preproc(4, ["-XbA", "--blah=foo", "-aF", "-Fval", "--", "--extra", "-args"], "aF") * * Would be turned into this: * * ["-X", "-b", "-A", "--blah=foo", "-a", "F", "-F", "val", "--", "--extra", "-args"] * * This function does not modify argv, and the return value is built of copies * of all the command line arguments. It is up to the caller to free this memory * after use. * * \note This function calls g_set_prgname assuming it wasn't previously set and * assuming argv is not NULL. It is not safe to call g_set_prgname more * than once so clients should not do so after calling this function. * * \param[in] argv The command line arguments. * \param[in] special Single-letter command line arguments that take a value. * These letters will all have pre-processing applied. * * \note @COMPAT We should drop this at a new major or minor release series * after we no longer support building with Booth <1.2, which invokes * Pacemaker CLI tools using the getopt syntax. */ gchar ** pcmk__cmdline_preproc(char *const *argv, const char *special); -/*! - * \internal - * \brief Process extra arguments as if they were provided by the user on the - * command line. - * - * \param[in,out] context The command line option processing context. - * \param[out] error A place for errors to be collected. - * \param[in] format The command line to be processed, potentially with - * format specifiers. - * \param[in] ... Arguments to be formatted. - * - * \note The first item in the list of arguments must be the name of the - * program, exactly as if the format string were coming from the - * command line. Otherwise, the first argument will be ignored. - * - * \return TRUE if processing succeeded, or FALSE otherwise. If FALSE, error - * should be checked and displayed to the user. - */ -G_GNUC_PRINTF(3, 4) -gboolean -pcmk__force_args(GOptionContext *context, GError **error, const char *format, ...); - #ifdef __cplusplus } #endif #endif // PCMK__CRM_COMMON_CMDLINE_INTERNAL__H diff --git a/include/crm/common/internal.h b/include/crm/common/internal.h index c333787d4a..0bc787eff2 100644 --- a/include/crm/common/internal.h +++ b/include/crm/common/internal.h @@ -1,347 +1,342 @@ /* * Copyright 2015-2025 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 PCMK__CRM_COMMON_INTERNAL__H #define PCMK__CRM_COMMON_INTERNAL__H #include // pid_t, getpid() #include // bool #include // uint8_t, uint64_t #include // PRIu64 #include // guint, GList, GHashTable #include // xmlNode #include // do_crm_log_unlikely(), etc. #include // mainloop_io_t, struct ipc_client_callbacks #include // crm_strdup_printf() #include #include #include #include #include #include #include #include #include #include #include #ifdef __cplusplus extern "C" { #endif /* This says whether the current application is a Pacemaker daemon or not, * and is used to change default logging settings such as whether to log to * stderr, etc., as well as a few other details such as whether blackbox signal * handling is enabled. * * It is set when logging is initialized, and does not need to be set directly. */ extern bool pcmk__is_daemon; // Number of elements in a statically defined array #define PCMK__NELEM(a) ((int) (sizeof(a)/sizeof(a[0])) ) #if PCMK__ENABLE_CIBSECRETS /* internal CIB utilities (from cib_secrets.c) */ int pcmk__substitute_secrets(const char *rsc_id, GHashTable *params); #endif /* internal main loop utilities (from mainloop.c) */ int pcmk__add_mainloop_ipc(crm_ipc_t *ipc, int priority, void *userdata, const struct ipc_client_callbacks *callbacks, mainloop_io_t **source); guint pcmk__mainloop_timer_get_period(const mainloop_timer_t *timer); /* internal name/value utilities (from nvpair.c) */ int pcmk__scan_nvpair(const gchar *input, gchar **name, gchar **value); char *pcmk__format_nvpair(const char *name, const char *value, const char *units); /* internal procfs utilities (from procfs.c) */ pid_t pcmk__procfs_pid_of(const char *name); unsigned int pcmk__procfs_num_cores(void); int pcmk__procfs_pid2path(pid_t pid, char **path); bool pcmk__procfs_has_pids(void); DIR *pcmk__procfs_fd_dir(void); void pcmk__sysrq_trigger(char t); bool pcmk__throttle_cib_load(const char *server, float *load); bool pcmk__throttle_load_avg(float *load); /* internal functions related to process IDs (from pid.c) */ /*! * \internal * \brief Check whether process exists (by PID and optionally executable path) * * \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 pcmk__pid_active(pid_t pid, const char *daemon); -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); - // bitwise arithmetic utilities /*! * \internal * \brief Set specified flags in a flag group * * \param[in] function Function name of caller * \param[in] line Line number of caller * \param[in] log_level Log a message at this level * \param[in] flag_type Label describing this flag group (for logging) * \param[in] target Name of object whose flags these are (for logging) * \param[in] flag_group Flag group being manipulated * \param[in] flags Which flags in the group should be set * \param[in] flags_str Readable equivalent of \p flags (for logging) * * \return Possibly modified flag group */ static inline uint64_t pcmk__set_flags_as(const char *function, int line, uint8_t log_level, const char *flag_type, const char *target, uint64_t flag_group, uint64_t flags, const char *flags_str) { uint64_t result = flag_group | flags; if (result != flag_group) { do_crm_log_unlikely(log_level, "%s flags %#.8" PRIx64 " (%s) for %s set by %s:%d", pcmk__s(flag_type, "Group of"), flags, pcmk__s(flags_str, "flags"), pcmk__s(target, "target"), function, line); } return result; } /*! * \internal * \brief Clear specified flags in a flag group * * \param[in] function Function name of caller * \param[in] line Line number of caller * \param[in] log_level Log a message at this level * \param[in] flag_type Label describing this flag group (for logging) * \param[in] target Name of object whose flags these are (for logging) * \param[in] flag_group Flag group being manipulated * \param[in] flags Which flags in the group should be cleared * \param[in] flags_str Readable equivalent of \p flags (for logging) * * \return Possibly modified flag group */ static inline uint64_t pcmk__clear_flags_as(const char *function, int line, uint8_t log_level, const char *flag_type, const char *target, uint64_t flag_group, uint64_t flags, const char *flags_str) { uint64_t result = flag_group & ~flags; if (result != flag_group) { do_crm_log_unlikely(log_level, "%s flags %#.8" PRIx64 " (%s) for %s cleared by %s:%d", pcmk__s(flag_type, "Group of"), flags, pcmk__s(flags_str, "flags"), pcmk__s(target, "target"), function, line); } return result; } /*! * \internal * \brief Get readable string for whether specified flags are set * * \param[in] flag_group Group of flags to check * \param[in] flags Which flags in \p flag_group should be checked * * \return "true" if all \p flags are set in \p flag_group, otherwise "false" */ static inline const char * pcmk__flag_text(uint64_t flag_group, uint64_t flags) { return pcmk__btoa(pcmk_all_flags_set(flag_group, flags)); } // miscellaneous utilities (from utils.c) -void pcmk__daemonize(const char *name, const char *pidfile); void pcmk__panic(const char *reason); pid_t pcmk__locate_sbd(void); void pcmk__sleep_ms(unsigned int ms); guint pcmk__create_timer(guint interval_ms, GSourceFunc fn, gpointer data); guint pcmk__timeout_ms2s(guint timeout_ms); extern int pcmk__score_red; extern int pcmk__score_green; extern int pcmk__score_yellow; /*! * \internal * \brief Allocate new zero-initialized memory, asserting on failure * * \param[in] file File where \p function is located * \param[in] function Calling function * \param[in] line Line within \p file * \param[in] nmemb Number of elements to allocate memory for * \param[in] size Size of each element * * \return Newly allocated memory of of size nmemb * size (guaranteed * not to be \c NULL) * * \note The caller is responsible for freeing the return value using \c free(). */ static inline void * pcmk__assert_alloc_as(const char *file, const char *function, uint32_t line, size_t nmemb, size_t size) { void *ptr = calloc(nmemb, size); if (ptr == NULL) { crm_abort(file, function, line, "Out of memory", FALSE, TRUE); crm_exit(CRM_EX_OSERR); } return ptr; } /*! * \internal * \brief Allocate new zero-initialized memory, asserting on failure * * \param[in] nmemb Number of elements to allocate memory for * \param[in] size Size of each element * * \return Newly allocated memory of of size nmemb * size (guaranteed * not to be \c NULL) * * \note The caller is responsible for freeing the return value using \c free(). */ #define pcmk__assert_alloc(nmemb, size) \ pcmk__assert_alloc_as(__FILE__, __func__, __LINE__, nmemb, size) /*! * \internal * \brief Resize a dynamically allocated memory block * * \param[in] ptr Memory block to resize (or NULL to allocate new memory) * \param[in] size New size of memory block in bytes (must be > 0) * * \return Pointer to resized memory block * * \note This asserts on error, so the result is guaranteed to be non-NULL * (which is the main advantage of this over directly using realloc()). */ static inline void * pcmk__realloc(void *ptr, size_t size) { void *new_ptr; // realloc(p, 0) can replace free(p) but this wrapper can't pcmk__assert(size > 0); new_ptr = realloc(ptr, size); if (new_ptr == NULL) { free(ptr); abort(); } return new_ptr; } static inline char * pcmk__getpid_s(void) { 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 PCMK__FAIL_COUNT_PREFIX "fail-count" #define PCMK__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 * pcmk__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 * pcmk__failcount_name(const char *rsc_id, const char *op, guint interval_ms) { return pcmk__fail_attr_name(PCMK__FAIL_COUNT_PREFIX, rsc_id, op, interval_ms); } static inline char * pcmk__lastfailure_name(const char *rsc_id, const char *op, guint interval_ms) { return pcmk__fail_attr_name(PCMK__LAST_FAILURE_PREFIX, rsc_id, op, interval_ms); } // internal resource agent functions (from agents.c) int pcmk__effective_rc(int rc); #ifdef __cplusplus } #endif #endif // PCMK__CRM_COMMON_INTERNAL__H diff --git a/include/crm/common/io_internal.h b/include/crm/common/io_internal.h index e9c68f6eb8..05ed97d01d 100644 --- a/include/crm/common/io_internal.h +++ b/include/crm/common/io_internal.h @@ -1,63 +1,76 @@ /* - * Copyright 2022-2024 the Pacemaker project contributors + * Copyright 2022-2025 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 PCMK__CRM_COMMON_IO_INTERNAL__H #define PCMK__CRM_COMMON_IO_INTERNAL__H #include // open() #include // bool +#include // freopen() #include // uid_t, gid_t #ifdef __cplusplus extern "C" { #endif int pcmk__real_path(const char *path, char **resolved_path); char *pcmk__series_filename(const char *directory, const char *series, unsigned 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); int pcmk__build_path(const char *path_c, mode_t mode); char *pcmk__full_path(const char *filename, const char *dirname); 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); +void pcmk__close_fds_in_child(void); /*! * \internal - * \brief Open /dev/null to consume next available file descriptor + * \brief Reopen the standard streams using \c /dev/null * - * Open /dev/null, disregarding the result. This is intended when daemonizing to - * be able to null stdin, stdout, and stderr. - * - * \param[in] flags O_RDONLY (stdin) or O_WRONLY (stdout and stderr) + * This is intended for use when daemonizing, to null \c stdin, \c stdout, + * and \c stderr. Failures are ignored. */ static inline void -pcmk__open_devnull(int flags) +pcmk__null_std_streams(void) { - (void) open("/dev/null", flags); + /* The "(void) !" is to suppress an obnoxious gcc warning. At least on some + * systems, freopen() has the attribute __warn_unused_result__. "(void)" + * alone does not suppress the warning. This policy is controversial but + * intentional and longstanding. + * + * https://stackoverflow.com/questions/40576003/ignoring-warning-wunused-result + * https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66425#c34 + * + * @TODO Consider replacing our uses of fork() with g_subprocess_*() or + * g_spawn_*(). These interfaces have arguments for redirecting standard + * streams to /dev/null. + */ + (void) !freopen("/dev/null", "r", stdin); + (void) !freopen("/dev/null", "w", stdout); + (void) !freopen("/dev/null", "w", stderr); } #ifdef __cplusplus } #endif #endif // PCMK__CRM_COMMON_IO_INTERNAL__H diff --git a/include/crm/common/logging.h b/include/crm/common/logging.h index 56f42107e9..28f0ed8681 100644 --- a/include/crm/common/logging.h +++ b/include/crm/common/logging.h @@ -1,400 +1,364 @@ /* - * Copyright 2004-2024 the Pacemaker project contributors + * Copyright 2004-2025 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. */ #ifndef PCMK__CRM_COMMON_LOGGING__H #define PCMK__CRM_COMMON_LOGGING__H #include #include // uint8_t, uint32_t #include #include #include #ifdef __cplusplus extern "C" { #endif /** * \file * \brief Wrappers for and extensions to libqb logging * \ingroup core */ /* Define custom log priorities. * * syslog(3) uses int for priorities, but libqb's struct qb_log_callsite uses * uint8_t, so make sure they fit in the latter. */ // Define something even less desired than debug #ifndef LOG_TRACE #define LOG_TRACE (LOG_DEBUG+1) #endif // Print message to stdout instead of logging it #ifndef LOG_STDOUT #define LOG_STDOUT 254 #endif // Don't send message anywhere #ifndef LOG_NEVER #define LOG_NEVER 255 #endif // @COMPAT Make internal when we can break API backward compatibility //! \deprecated Do not use extern unsigned int crm_log_level; // @COMPAT Make internal when we can break API backward compatibility //! \deprecated Do not use extern unsigned int crm_trace_nonlog; void crm_enable_blackbox(int nsig); void crm_disable_blackbox(int nsig); void crm_write_blackbox(int nsig, const struct qb_log_callsite *callsite); void crm_update_callsites(void); void crm_log_deinit(void); /*! * \brief Initializes the logging system and defaults to the least verbose output level * * \param[in] entity If not NULL, will be used as the identity for logging purposes * \param[in] argc The number of command line parameters * \param[in] argv The command line parameter values */ void crm_log_preinit(const char *entity, int argc, char *const *argv); gboolean crm_log_init(const char *entity, uint8_t level, gboolean daemon, gboolean to_stderr, int argc, char **argv, gboolean quiet); void crm_log_args(int argc, char **argv); void crm_log_output_fn(const char *file, const char *function, int line, int level, const char *prefix, const char *output); // Log a block of text line by line #define crm_log_output(level, prefix, output) \ crm_log_output_fn(__FILE__, __func__, __LINE__, level, prefix, output) void crm_bump_log_level(int argc, char **argv); void crm_enable_stderr(int enable); gboolean crm_is_callsite_active(struct qb_log_callsite *cs, uint8_t level, uint32_t tags); // NOTE: sbd (as of at least 1.5.2) uses this /* returns the old value */ unsigned int set_crm_log_level(unsigned int level); unsigned int get_crm_log_level(void); void pcmk_log_xml_as(const char *file, const char *function, uint32_t line, uint32_t tags, uint8_t level, const char *text, const xmlNode *xml); /* * Throughout the macros below, note the leading, pre-comma, space in the * various ' , ##args' occurrences to aid portability across versions of 'gcc'. * https://gcc.gnu.org/onlinedocs/cpp/Variadic-Macros.html#Variadic-Macros */ #if defined(__clang__) #define CRM_TRACE_INIT_DATA(name) #else #include // required by QB_LOG_INIT_DATA() macro #define CRM_TRACE_INIT_DATA(name) QB_LOG_INIT_DATA(name) #endif /*! * \internal * \brief Clip log_level to \p uint8_t range * * \param[in] level Log level to clip * * \return 0 if \p level is less than 0, \p UINT8_MAX if \p level is greater * than \p UINT8_MAX, or \p level otherwise */ /* @COMPAT: Make this function internal at a compatibility break. It's used in * public macros for now. */ static inline uint8_t pcmk__clip_log_level(int level) { if (level <= 0) { return 0; } if (level >= UINT8_MAX) { return UINT8_MAX; } return level; } /* Using "switch" instead of "if" in these macro definitions keeps * static analysis from complaining about constant evaluations */ /*! * \brief Log a message * * \param[in] level Priority at which to log the message * \param[in] fmt printf-style format string literal for message * \param[in] args Any arguments needed by format string */ #define do_crm_log(level, fmt, args...) do { \ uint8_t _level = pcmk__clip_log_level(level); \ \ switch (_level) { \ case LOG_STDOUT: \ printf(fmt "\n" , ##args); \ break; \ case LOG_NEVER: \ break; \ default: \ qb_log_from_external_source(__func__, __FILE__, fmt, \ _level, __LINE__, 0 , ##args); \ break; \ } \ } while (0) /*! * \brief Log a message that is likely to be filtered out * * \param[in] level Priority at which to log the message * \param[in] fmt printf-style format string for message * \param[in] args Any arguments needed by format string * * \note This does nothing when level is \p LOG_STDOUT. */ #define do_crm_log_unlikely(level, fmt, args...) do { \ uint8_t _level = pcmk__clip_log_level(level); \ \ switch (_level) { \ case LOG_STDOUT: case LOG_NEVER: \ break; \ default: { \ static struct qb_log_callsite *trace_cs = NULL; \ if (trace_cs == NULL) { \ trace_cs = qb_log_callsite_get(__func__, __FILE__, fmt, \ _level, __LINE__, 0); \ } \ if (crm_is_callsite_active(trace_cs, _level, 0)) { \ qb_log_from_external_source(__func__, __FILE__, fmt, \ _level, __LINE__, 0 , \ ##args); \ } \ } \ break; \ } \ } while (0) #define CRM_LOG_ASSERT(expr) do { \ if (!(expr)) { \ static struct qb_log_callsite *core_cs = NULL; \ if(core_cs == NULL) { \ core_cs = qb_log_callsite_get(__func__, __FILE__, \ "log-assert", LOG_TRACE, \ __LINE__, 0); \ } \ crm_abort(__FILE__, __func__, __LINE__, #expr, \ core_cs?core_cs->targets:FALSE, TRUE); \ } \ } while(0) // NOTE: sbd (as of at least 1.5.2) uses this /* 'failure_action' MUST NOT be 'continue' as it will apply to the * macro's do-while loop */ #define CRM_CHECK(expr, failure_action) do { \ if (!(expr)) { \ static struct qb_log_callsite *core_cs = NULL; \ if (core_cs == NULL) { \ core_cs = qb_log_callsite_get(__func__, __FILE__, \ "check-assert", \ LOG_TRACE, __LINE__, 0); \ } \ crm_abort(__FILE__, __func__, __LINE__, #expr, \ (core_cs? core_cs->targets: FALSE), TRUE); \ failure_action; \ } \ } while(0) /*! * \brief Log XML line-by-line in a formatted fashion * * \param[in] level Priority at which to log the messages * \param[in] text Prefix for each line * \param[in] xml XML to log * * \note This does nothing when \p level is \p LOG_STDOUT. */ #define do_crm_log_xml(level, text, xml) do { \ uint8_t _level = pcmk__clip_log_level(level); \ static struct qb_log_callsite *xml_cs = NULL; \ \ switch (_level) { \ case LOG_STDOUT: \ case LOG_NEVER: \ break; \ default: \ if (xml_cs == NULL) { \ xml_cs = qb_log_callsite_get(__func__, __FILE__, \ "xml-blob", _level, \ __LINE__, 0); \ } \ if (crm_is_callsite_active(xml_cs, _level, 0)) { \ pcmk_log_xml_as(__FILE__, __func__, __LINE__, 0, \ _level, text, (xml)); \ } \ break; \ } \ } while(0) /*! * \brief Log a message as if it came from a different code location * * \param[in] level Priority at which to log the message * \param[in] file Source file name to use instead of __FILE__ * \param[in] function Source function name to use instead of __func__ * \param[in] line Source line number to use instead of __line__ * \param[in] fmt printf-style format string literal for message * \param[in] args Any arguments needed by format string */ #define do_crm_log_alias(level, file, function, line, fmt, args...) do { \ uint8_t _level = pcmk__clip_log_level(level); \ \ switch (_level) { \ case LOG_STDOUT: \ printf(fmt "\n" , ##args); \ break; \ case LOG_NEVER: \ break; \ default: \ qb_log_from_external_source(function, file, fmt, _level, \ line, 0 , ##args); \ break; \ } \ } while (0) -// NOTE: sbd (as of at least 1.5.2) uses this -/*! - * \brief Send a system error message to both the log and stderr - * - * \param[in] level Priority at which to log the message - * \param[in] fmt printf-style format string for message - * \param[in] args Any arguments needed by format string - * - * \deprecated One of the other logging functions should be used with - * pcmk_strerror() instead. - * \note This is a macro, and \p level may be evaluated more than once. - * \note Because crm_perror() adds the system error message and error number - * onto the end of fmt, that information will become extended information - * if QB_XS is used inside fmt and will not show up in syslog. - */ -#define crm_perror(level, fmt, args...) do { \ - uint8_t _level = pcmk__clip_log_level(level); \ - \ - switch (_level) { \ - case LOG_NEVER: \ - break; \ - default: { \ - const char *err = strerror(errno); \ - if (_level <= crm_log_level) { \ - fprintf(stderr, fmt ": %s (%d)\n" , ##args, err, \ - errno); \ - } \ - /* Pass original level arg since do_crm_log() also declares \ - * _level \ - */ \ - do_crm_log((level), fmt ": %s (%d)" , ##args, err, errno); \ - } \ - break; \ - } \ - } while (0) - /*! * \brief Log a message with a tag (for use with PCMK_trace_tags) * * \param[in] level Priority at which to log the message * \param[in] tag String to tag message with * \param[in] fmt printf-style format string for message * \param[in] args Any arguments needed by format string * * \note This does nothing when level is LOG_STDOUT. */ #define crm_log_tag(level, tag, fmt, args...) do { \ uint8_t _level = pcmk__clip_log_level(level); \ \ switch (_level) { \ case LOG_STDOUT: case LOG_NEVER: \ break; \ default: { \ static struct qb_log_callsite *trace_tag_cs = NULL; \ int converted_tag = g_quark_try_string(tag); \ if (trace_tag_cs == NULL) { \ trace_tag_cs = qb_log_callsite_get(__func__, __FILE__, \ fmt, _level, \ __LINE__, \ converted_tag); \ } \ if (crm_is_callsite_active(trace_tag_cs, _level, \ converted_tag)) { \ qb_log_from_external_source(__func__, __FILE__, fmt, \ _level, __LINE__, \ converted_tag , ##args); \ } \ } \ } \ } while (0) #define crm_emerg(fmt, args...) qb_log(LOG_EMERG, fmt , ##args) #define crm_crit(fmt, args...) qb_logt(LOG_CRIT, 0, fmt , ##args) // NOTE: sbd (as of at least 1.5.2) uses this #define crm_err(fmt, args...) qb_logt(LOG_ERR, 0, fmt , ##args) // NOTE: sbd (as of at least 1.5.2) uses this #define crm_warn(fmt, args...) qb_logt(LOG_WARNING, 0, fmt , ##args) // NOTE: sbd (as of at least 1.5.2) uses this #define crm_notice(fmt, args...) qb_logt(LOG_NOTICE, 0, fmt , ##args) #define crm_info(fmt, args...) qb_logt(LOG_INFO, 0, fmt , ##args) // // NOTE: sbd (as of at least 1.5.2) uses this #define crm_debug(fmt, args...) do_crm_log_unlikely(LOG_DEBUG, fmt , ##args) #define crm_trace(fmt, args...) do_crm_log_unlikely(LOG_TRACE, fmt , ##args) #define crm_log_xml_crit(xml, text) do_crm_log_xml(LOG_CRIT, text, xml) #define crm_log_xml_err(xml, text) do_crm_log_xml(LOG_ERR, text, xml) #define crm_log_xml_warn(xml, text) do_crm_log_xml(LOG_WARNING, text, xml) #define crm_log_xml_notice(xml, text) do_crm_log_xml(LOG_NOTICE, text, xml) #define crm_log_xml_info(xml, text) do_crm_log_xml(LOG_INFO, text, xml) #define crm_log_xml_debug(xml, text) do_crm_log_xml(LOG_DEBUG, text, xml) #define crm_log_xml_trace(xml, text) do_crm_log_xml(LOG_TRACE, text, xml) #define crm_log_xml_explicit(xml, text) do { \ static struct qb_log_callsite *digest_cs = NULL; \ digest_cs = qb_log_callsite_get( \ __func__, __FILE__, text, LOG_TRACE, __LINE__, \ crm_trace_nonlog); \ if (digest_cs && digest_cs->targets) { \ do_crm_log_xml(LOG_TRACE, text, xml); \ } \ } while(0) #ifdef __cplusplus } #endif #if !defined(PCMK_ALLOW_DEPRECATED) || (PCMK_ALLOW_DEPRECATED == 1) #include #endif #endif diff --git a/include/crm/common/logging_compat.h b/include/crm/common/logging_compat.h index 4864c2b423..7818f3ece0 100644 --- a/include/crm/common/logging_compat.h +++ b/include/crm/common/logging_compat.h @@ -1,29 +1,61 @@ /* - * Copyright 2024 the Pacemaker project contributors + * Copyright 2024-2025 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. */ #ifndef PCMK__CRM_COMMON_LOGGING_COMPAT__H #define PCMK__CRM_COMMON_LOGGING_COMPAT__H #include // qb_log_ctl(), QB_LOG_CONF_EXTENDED #ifdef __cplusplus extern "C" { #endif +/*! + * \file + * \brief Deprecated Pacemaker logging API + * \ingroup core + * \deprecated Do not include this header directly. The logging APIs in this + * header, and the header itself, will be removed in a future + * release. + */ + //! \deprecated use QB_XS instead #define CRM_XS QB_XS //! \deprecated Use qb_log_ctl() directly instead #define crm_extended_logging(t, e) qb_log_ctl((t), QB_LOG_CONF_EXTENDED, (e)) +// NOTE: sbd (as of at least 1.5.2) uses this +//! \deprecated Do not use Pacemaker for general-purpose logging +#define crm_perror(level, fmt, args...) do { \ + uint8_t _level = pcmk__clip_log_level(level); \ + \ + switch (_level) { \ + case LOG_NEVER: \ + break; \ + default: { \ + const char *err = strerror(errno); \ + if (_level <= crm_log_level) { \ + fprintf(stderr, fmt ": %s (%d)\n" , ##args, err, \ + errno); \ + } \ + /* Pass original level arg since do_crm_log() also declares \ + * _level \ + */ \ + do_crm_log((level), fmt ": %s (%d)" , ##args, err, errno); \ + } \ + break; \ + } \ + } while (0) + #ifdef __cplusplus } #endif #endif // PCMK__CRM_COMMON_LOGGING_COMPAT__H diff --git a/include/crm/common/output_internal.h b/include/crm/common/output_internal.h index 64d22e62a0..0dab4075ca 100644 --- a/include/crm/common/output_internal.h +++ b/include/crm/common/output_internal.h @@ -1,1017 +1,1019 @@ /* * Copyright 2019-2025 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 PCMK__CRM_COMMON_OUTPUT_INTERNAL__H #define PCMK__CRM_COMMON_OUTPUT_INTERNAL__H #include #include #include #include #include #include #include #ifdef __cplusplus extern "C" { #endif /** * \file * \brief Formatted output for pacemaker tools */ #if defined(PCMK__WITH_ATTRIBUTE_OUTPUT_ARGS) #define PCMK__OUTPUT_ARGS(ARGS...) __attribute__((output_args(ARGS))) #else #define PCMK__OUTPUT_ARGS(ARGS...) #endif typedef struct pcmk__output_s pcmk__output_t; /*! * \internal * \brief The type of a function that creates a ::pcmk__output_t. * * Instances of this type are passed to pcmk__register_format(), stored in an * internal data structure, and later accessed by pcmk__output_new(). For * examples, see pcmk__mk_xml_output() and pcmk__mk_text_output(). * * \param[in] argv The list of command line arguments. */ typedef pcmk__output_t * (*pcmk__output_factory_t)(char **argv); /*! * \internal * \brief The type of a custom message formatting function. * * These functions are defined by various libraries to support formatting of * types aside from the basic types provided by a ::pcmk__output_t. * * The meaning of the return value will be different for each message. * In general, however, 0 should be returned on success and a positive value * on error. * * \param[in,out] out Output object to use to display message * \param[in,out] args Message-specific arguments needed * * \note These functions must not call va_start or va_end - that is done * automatically before the custom formatting function is called. */ typedef int (*pcmk__message_fn_t)(pcmk__output_t *out, va_list args); /*! * \internal * \brief Internal type for tracking custom messages. * * Each library can register functions that format custom message types. These * are commonly used to handle some library-specific type. Registration is * done by first defining a table of ::pcmk__message_entry_t structures and * then passing that table to pcmk__register_messages(). Separate handlers * can be defined for the same message, but for different formats (xml vs. * text). Unknown formats will be ignored. * * Additionally, a "default" value for fmt_table can be used. In this case, * fn will be registered for all supported formats. It is also possible to * register a default and then override that registration with a format-specific * function if necessary. * * \note The ::pcmk__message_entry_t table is processed in one pass, in order, * from top to bottom. This means later entries with the same message_id will * override previous ones. Thus, any default entry must come before any * format-specific entries for the same message_id. */ typedef struct pcmk__message_entry_s { /*! * \brief The message to be handled. * * This must be the same ID that is passed to the message function of * a ::pcmk__output_t. Unknown message IDs will be ignored. */ const char *message_id; /*! * \brief The format type this handler is for. * * This name must match the fmt_name of the currently active formatter in * order for the registered function to be called. It is valid to have * multiple entries for the same message_id but with different fmt_name * values. */ const char *fmt_name; /*! * \brief The function to be called for message_id given a match on * fmt_name. See comments on ::pcmk__message_fn_t. */ pcmk__message_fn_t fn; } pcmk__message_entry_t; /*! * \internal * \brief This structure contains everything needed to add support for a * single output formatter to a command line program. */ typedef struct pcmk__supported_format_s { /*! * \brief The name of this output formatter, which should match the * fmt_name parameter in some ::pcmk__output_t structure. */ const char *name; /*! * \brief A function that creates a ::pcmk__output_t. */ pcmk__output_factory_t create; /*! * \brief Format-specific command line options. This can be NULL if * no command line options should be supported. */ GOptionEntry *options; } pcmk__supported_format_t; /* The following three blocks need to be updated each time a new base formatter * is added. */ extern GOptionEntry pcmk__html_output_entries[]; pcmk__output_t *pcmk__mk_html_output(char **argv); pcmk__output_t *pcmk__mk_log_output(char **argv); pcmk__output_t *pcmk__mk_none_output(char **argv); pcmk__output_t *pcmk__mk_text_output(char **argv); pcmk__output_t *pcmk__mk_xml_output(char **argv); #define PCMK__SUPPORTED_FORMAT_HTML { "html", pcmk__mk_html_output, pcmk__html_output_entries } #define PCMK__SUPPORTED_FORMAT_LOG { "log", pcmk__mk_log_output, NULL } #define PCMK__SUPPORTED_FORMAT_NONE { PCMK_VALUE_NONE, pcmk__mk_none_output, NULL } #define PCMK__SUPPORTED_FORMAT_TEXT { "text", pcmk__mk_text_output, NULL } #define PCMK__SUPPORTED_FORMAT_XML { "xml", pcmk__mk_xml_output, NULL } /*! * \brief This structure contains everything that makes up a single output * formatter. * * Instances of this structure may be created by calling pcmk__output_new() * with the name of the desired formatter. They should later be freed with * pcmk__output_free(). */ struct pcmk__output_s { /*! * \brief The name of this output formatter. */ const char *fmt_name; /*! * \brief Should this formatter supress most output? * * \note This setting is not respected by all formatters. In general, * machine-readable output formats will not support this while * user-oriented formats will. Callers should use is_quiet() * to test whether to print or not. */ bool quiet; /*! * \brief A copy of the request that generated this output. * * In the case of command line usage, this would be the command line * arguments. For other use cases, it could be different. */ gchar *request; /*! * \brief Where output should be written. * * This could be a file handle, or stdout or stderr. This is really only * useful internally. */ FILE *dest; /*! * \brief Custom messages that are currently registered on this formatter. * * Keys are the string message IDs, values are ::pcmk__message_fn_t function * pointers. */ GHashTable *messages; /*! * \brief Implementation-specific private data. * * Each individual formatter may have some private data useful in its * implementation. This points to that data. Callers should not rely on * its contents or structure. */ void *priv; /*! * \internal * \brief Take whatever actions are necessary to prepare out for use. This is * called by pcmk__output_new(). End users should not need to call this. * * \note For formatted output implementers - This function should be written in * such a way that it can be called repeatedly on an already initialized * object without causing problems, or on a previously finished object * without crashing. * * \param[in,out] out The output functions structure. * * \return true on success, false on error. */ bool (*init) (pcmk__output_t *out); /*! * \internal * \brief Free the private formatter-specific data. * * This is called from pcmk__output_free() and does not typically need to be * called directly. * * \param[in,out] out The output functions structure. */ void (*free_priv) (pcmk__output_t *out); /*! * \internal * \brief Take whatever actions are necessary to end formatted output. * * This could include flushing output to a file, but does not include freeing * anything. The finish method can potentially be fairly complicated, adding * additional information to the internal data structures or doing whatever * else. It is therefore suggested that finish only be called once. * * \note The print parameter will only affect those formatters that do all * their output at the end. Console-oriented formatters typically print * a line at a time as they go, so this parameter will not affect them. * Structured formatters will honor it, however. * * \note The copy_dest parameter does not apply to all formatters. Console- * oriented formatters do not build up a structure as they go, and thus * do not have anything to return. Structured formatters will honor it, * however. Note that each type of formatter will return a different * type of value in this parameter. To use this parameter, call this * function like so: * * \code * xmlNode *dest = NULL; * out->finish(out, exit_code, false, (void **) &dest); * \endcode * * \param[in,out] out The output functions structure. * \param[in] exit_status The exit value of the whole program. * \param[in] print Whether this function should write any output. * \param[out] copy_dest A destination to store a copy of the internal * data structure for this output, or NULL if no * copy is required. The caller should free this * memory when done with it. */ void (*finish) (pcmk__output_t *out, crm_exit_t exit_status, bool print, void **copy_dest); /*! * \internal * \brief Finalize output and then immediately set back up to start a new set * of output. * * This is conceptually the same as calling finish and then init, though in * practice more be happening behind the scenes. * * \note This function differs from finish in that no exit_status is added. * The idea is that the program is not shutting down, so there is not * yet a final exit code. Call finish on the last time through if this * is needed. * * \param[in,out] out The output functions structure. */ void (*reset) (pcmk__output_t *out); /*! * \internal * \brief Register a custom message. * * \param[in,out] out The output functions structure. * \param[in] message_id The name of the message to register. This name * will be used as the message_id parameter to the * message function in order to call the custom * format function. * \param[in] fn The custom format function to call for message_id. */ void (*register_message) (pcmk__output_t *out, const char *message_id, pcmk__message_fn_t fn); /*! * \internal * \brief Call a previously registered custom message. * * \param[in,out] out The output functions structure. * \param[in] message_id The name of the message to call. This name must * be the same as the message_id parameter of some * previous call to register_message. * \param[in] ... Arguments to be passed to the registered function. * * \return A standard Pacemaker return code. Generally: 0 if a function was * registered for the message, that function was called, and returned * successfully; EINVAL if no function was registered; or pcmk_rc_no_output * if a function was called but produced no output. */ int (*message) (pcmk__output_t *out, const char *message_id, ...); /*! * \internal * \brief Format the output of a completed subprocess. * * \param[in,out] out The output functions structure. * \param[in] exit_status The exit value of the subprocess. * \param[in] proc_stdout stdout from the completed subprocess. * \param[in] proc_stderr stderr from the completed subprocess. */ void (*subprocess_output) (pcmk__output_t *out, int exit_status, const char *proc_stdout, const char *proc_stderr); /*! * \internal * \brief Format version information. This is useful for the --version * argument of command line tools. * * \param[in,out] out The output functions structure. */ void (*version)(pcmk__output_t *out); /*! * \internal * \brief Format an informational message that should be shown to * to an interactive user. Not all formatters will do this. * * \note A newline will automatically be added to the end of the format * string, so callers should not include a newline. * * \note It is possible for a formatter that supports this method to * still not print anything out if is_quiet returns true. * * \param[in,out] out The output functions structure. * \param[in] buf The message to be printed. * \param[in] ... Arguments to be formatted. * * \return A standard Pacemaker return code. Generally: pcmk_rc_ok * if output was produced and pcmk_rc_no_output if it was not. * As not all formatters implement this function, those that * do not will always just return pcmk_rc_no_output. */ int (*info) (pcmk__output_t *out, const char *format, ...) G_GNUC_PRINTF(2, 3); /*! * \internal * \brief Like \p info() but for messages that should appear only * transiently. Not all formatters will do this. * * The originally envisioned use case is for console output, where a * transient status-related message may be quickly overwritten by a refresh. * * \param[in,out] out The output functions structure. * \param[in] format The format string of the message to be printed. * \param[in] ... Arguments to be formatted. * * \return A standard Pacemaker return code. Generally: \p pcmk_rc_ok if * output was produced and \p pcmk_rc_no_output if it was not. As * not all formatters implement this function, those that do not * will always just return \p pcmk_rc_no_output. */ int (*transient) (pcmk__output_t *out, const char *format, ...) G_GNUC_PRINTF(2, 3); /*! * \internal * \brief Format an error message that should be shown to an interactive * user. Not all formatters will do this. * * \note A newline will automatically be added to the end of the format * string, so callers should not include a newline. * * \note Formatters that support this method should always generate output, * even if is_quiet returns true. * * \param[in,out] out The output functions structure. * \param[in] buf The message to be printed. * \param[in] ... Arguments to be formatted. */ void (*err) (pcmk__output_t *out, const char *format, ...) G_GNUC_PRINTF(2, 3); /*! * \internal * \brief Format already formatted XML. * * \param[in,out] out The output functions structure. * \param[in] name A name to associate with the XML. * \param[in] buf The XML in a string. */ void (*output_xml) (pcmk__output_t *out, const char *name, const char *buf); /*! * \internal * \brief Start a new list of items. * * \note For text output, this corresponds to another level of indentation. For * XML output, this corresponds to wrapping any following output in another * layer of tags. * * \note If singular_noun and plural_noun are non-NULL, calling end_list will * result in a summary being added. * * \param[in,out] out The output functions structure. * \param[in] singular_noun When outputting the summary for a list with * one item, the noun to use. * \param[in] plural_noun When outputting the summary for a list with * more than one item, the noun to use. * \param[in] format The format string. * \param[in] ... Arguments to be formatted. */ void (*begin_list) (pcmk__output_t *out, const char *singular_noun, const char *plural_noun, const char *format, ...) G_GNUC_PRINTF(4, 5); /*! * \internal * \brief Format a single item in a list. * * \param[in,out] out The output functions structure. * \param[in] name A name to associate with this item. * \param[in] format The format string. * \param[in] ... Arguments to be formatted. */ void (*list_item) (pcmk__output_t *out, const char *name, const char *format, ...) G_GNUC_PRINTF(3, 4); /*! * \internal * \brief Increment the internal counter of the current list's length. * * Typically, this counter is maintained behind the scenes as a side effect * of calling list_item(). However, custom functions that maintain lists * some other way will need to manage this counter manually. This is * useful for implementing custom message functions and should not be * needed otherwise. * * \param[in,out] out The output functions structure. */ void (*increment_list) (pcmk__output_t *out); /*! * \internal * \brief Conclude a list. * * \note If begin_list was called with non-NULL for both the singular_noun * and plural_noun arguments, this function will output a summary. * Otherwise, no summary will be added. * * \param[in,out] out The output functions structure. */ void (*end_list) (pcmk__output_t *out); /*! * \internal * \brief Should anything be printed to the user? * * \note This takes into account both the \p quiet value as well as the * current formatter. * * \param[in,out] out The output functions structure. * * \return true if output should be supressed, false otherwise. */ bool (*is_quiet) (pcmk__output_t *out); /*! * \internal * \brief Output a spacer. Not all formatters will do this. * * \param[in,out] out The output functions structure. */ void (*spacer) (pcmk__output_t *out); /*! * \internal * \brief Output a progress indicator. This is likely only useful for * plain text, console based formatters. * * \param[in,out] out The output functions structure * \param[in] end If true, output a newline afterwards (this should * only be used the last time this function is called) * */ void (*progress) (pcmk__output_t *out, bool end); /*! * \internal * \brief Prompt the user for input. Not all formatters will do this. * * \note This function is part of pcmk__output_t, but unlike all other * function it does not take that as an argument. In general, a * prompt will go directly to the screen and therefore bypass any * need to use the formatted output code to decide where and how * to display. * * \param[in] prompt The prompt to display. This is required. * \param[in] echo If true, echo the user's input to the screen. Set * to false for password entry. * \param[out] dest Where to store the user's response. This is * required. */ void (*prompt) (const char *prompt, bool echo, char **dest); }; /*! * \internal * \brief Call a formatting function for a previously registered message. * * \note This function is for implementing custom formatters. It should not * be called directly. Instead, call out->message. * * \param[in,out] out The output functions structure. * \param[in] message_id The message to be handled. Unknown messages * will be ignored. * \param[in] ... Arguments to be passed to the registered function. */ int pcmk__call_message(pcmk__output_t *out, const char *message_id, ...); /*! * \internal * \brief Free a ::pcmk__output_t structure that was previously created by * pcmk__output_new(). * * \note While the create and finish functions are designed in such a way that * they can be called repeatedly, this function will completely free the * memory of the object. Once this function has been called, producing * more output requires starting over from pcmk__output_new(). * * \param[in,out] out The output structure. */ void pcmk__output_free(pcmk__output_t *out); /*! * \internal * \brief Create a new ::pcmk__output_t structure. * * This also registers message functions from libcrmcommon. * * \param[in,out] out The destination of the new ::pcmk__output_t. * \param[in] fmt_name How should output be formatted? * \param[in] filename Where should formatted output be written to? This * can be a filename (which will be overwritten if it * already exists), or NULL or "-" for stdout. For no * output, pass a filename of "/dev/null". * \param[in] argv The list of command line arguments. * * \return Standard Pacemaker return code */ int pcmk__output_new(pcmk__output_t **out, const char *fmt_name, const char *filename, char **argv); /*! * \internal * \brief Register a new output formatter, making it available for use * the same as a base formatter. * * \param[in,out] group A ::GOptionGroup that formatted output related command * line arguments should be added to. This can be NULL * for use outside of command line programs. * \param[in] name The name of the format. This will be used to select a * format from command line options and for displaying help. * \param[in] create A function that creates a ::pcmk__output_t. * \param[in] options Format-specific command line options. These will be * added to the context. This argument can also be NULL. * * \return Standard Pacemaker return code */ int pcmk__register_format(GOptionGroup *group, const char *name, pcmk__output_factory_t create, const GOptionEntry *options); /*! * \internal * \brief Register an entire table of output formatters at once. * * \param[in,out] group A ::GOptionGroup that formatted output related command * line arguments should be added to. This can be NULL * for use outside of command line programs. * \param[in] table An array of ::pcmk__supported_format_t which should * all be registered. This array must be NULL-terminated. * */ void pcmk__register_formats(GOptionGroup *group, const pcmk__supported_format_t *table); /*! * \internal * \brief Unregister a previously registered table of custom formatting * functions and destroy the internal data structures associated with them. */ void pcmk__unregister_formats(void); /*! * \internal * \brief Register a function to handle a custom message. * * \note This function is for implementing custom formatters. It should not * be called directly. Instead, call out->register_message. * * \param[in,out] out The output functions structure. * \param[in] message_id The message to be handled. * \param[in] fn The custom format function to call for message_id. */ void pcmk__register_message(pcmk__output_t *out, const char *message_id, pcmk__message_fn_t fn); /*! * \internal * \brief Register an entire table of custom formatting functions at once. * * This table can contain multiple formatting functions for the same message ID * if they are for different format types. * * \param[in,out] out The output functions structure. * \param[in] table An array of ::pcmk__message_entry_t values which should * all be registered. This array must be NULL-terminated. */ void pcmk__register_messages(pcmk__output_t *out, const pcmk__message_entry_t *table); /* Functions that are useful for implementing custom message formatters */ void pcmk__output_text_set_fancy(pcmk__output_t *out, bool enabled); /*! * \internal * \brief A printf-like function. * * This function writes to out->dest and indents the text to the current level * of the text formatter's nesting. This function should be used when implementing * custom message functions for the text output format. It should not be used * for any other purpose. * * Typically, this function should be used instead of printf. * * \param[in,out] out The output functions structure. * \param[in] format The format string. * \param[in] ... Arguments to be passed to the format string. */ void pcmk__indented_printf(pcmk__output_t *out, const char *format, ...) G_GNUC_PRINTF(2, 3); /*! * \internal * \brief A vprintf-like function. * * This function is like pcmk__indented_printf(), except it takes a va_list instead * of a list of arguments. This function should be used when implementing custom * functions for the text output format. It should not be used for any other purpose. * * Typically, this function should be used instead of vprintf. * * \param[in,out] out The output functions structure. * \param[in] format The format string. * \param[in] args A list of arguments to apply to the format string. */ void pcmk__indented_vprintf(pcmk__output_t *out, const char *format, va_list args) G_GNUC_PRINTF(2, 0); /*! * \internal * \brief A printf-like function. * * This function writes to out->dest without indenting the text. This function * should be used when implementing custom message functions for the text output * format. It should not be used for any other purpose. * * \param[in,out] out The output functions structure. * \param[in] format The format string. * \param[in] ... Arguments to be passed to the format string. */ void pcmk__formatted_printf(pcmk__output_t *out, const char *format, ...) G_GNUC_PRINTF(2, 3); /*! * \internal * \brief A vprintf-like function. * * This function is like pcmk__formatted_printf(), except it takes a va_list instead * of a list of arguments. This function should be used when implementing custom * message functions for the text output format. It should not be used for any * other purpose. * * \param[in,out] out The output functions structure. * \param[in] format The format string. * \param[in] args A list of arguments to apply to the format string. */ void pcmk__formatted_vprintf(pcmk__output_t *out, const char *format, va_list args) G_GNUC_PRINTF(2, 0); /*! * \internal * \brief Prompt the user for input. * * \param[in] prompt The prompt to display * \param[in] echo If true, echo the user's input to the screen. Set * to false for password entry. * \param[out] dest Where to store the user's response. */ void pcmk__text_prompt(const char *prompt, bool echo, char **dest); uint8_t pcmk__output_get_log_level(const pcmk__output_t *out); void pcmk__output_set_log_level(pcmk__output_t *out, uint8_t log_level); void pcmk__output_set_log_filter(pcmk__output_t *out, const char *file, const char *function, uint32_t line, uint32_t tags); /*! * \internal * \brief Create and return a new XML node with the given name, as a child of the * current list parent. The new node is then added as the new list parent, * meaning all subsequent nodes will be its children. This is used when * implementing custom functions. * * \param[in,out] out The output functions structure. * \param[in] name The name of the node to be created. * \param[in] ... Name/value pairs to set as XML properties. */ xmlNodePtr pcmk__output_xml_create_parent(pcmk__output_t *out, const char *name, ...) G_GNUC_NULL_TERMINATED; /*! * \internal * \brief Add a copy of the given node as a child of the current list parent. * This is used when implementing custom message functions. * * \param[in,out] out The output functions structure. * \param[in] node An XML node to copy as a child. */ void pcmk__output_xml_add_node_copy(pcmk__output_t *out, xmlNodePtr node); /*! * \internal * \brief Create and return a new XML node with the given name, as a child of the * current list parent. This is used when implementing custom functions. * * \param[in,out] out The output functions structure. * \param[in] name The name of the node to be created. * \param[in] ... Name/value pairs to set as XML properties. */ xmlNodePtr pcmk__output_create_xml_node(pcmk__output_t *out, const char *name, ...) G_GNUC_NULL_TERMINATED; /*! * \internal * \brief Like pcmk__output_create_xml_node(), but add the given text content to the * new node. * * \param[in,out] out The output functions structure. * \param[in] name The name of the node to be created. * \param[in] content The text content of the node. */ xmlNodePtr pcmk__output_create_xml_text_node(pcmk__output_t *out, const char *name, const char *content); /*! * \internal * \brief Push a parent XML node onto the stack. This is used when implementing * custom message functions. * * The XML output formatter maintains an internal stack to keep track of which nodes * are parents in order to build up the tree structure. This function can be used * to temporarily push a new node onto the stack. After calling this function, any * other formatting functions will have their nodes added as children of this new * parent. * * \param[in,out] out The output functions structure * \param[in] parent XML node to add */ void pcmk__output_xml_push_parent(pcmk__output_t *out, xmlNodePtr parent); /*! * \internal * \brief Pop a parent XML node onto the stack. This is used when implementing * custom message functions. * * This function removes a parent node from the stack. See pcmk__xml_push_parent() * for more details. * * \note Little checking is done with this function. Be sure you only pop parents * that were previously pushed. In general, it is best to keep the code between * push and pop simple. * * \param[in,out] out The output functions structure. */ void pcmk__output_xml_pop_parent(pcmk__output_t *out); /*! * \internal * \brief Peek a parent XML node onto the stack. This is used when implementing * custom message functions. * * This function peeks a parent node on stack. See pcmk__xml_push_parent() * for more details. It has no side-effect and can be called for an empty stack. * * \note Little checking is done with this function. * * \param[in,out] out The output functions structure. * * \return NULL if stack is empty, otherwise the parent of the stack. */ xmlNodePtr pcmk__output_xml_peek_parent(pcmk__output_t *out); /*! * \internal * \brief Create a new XML node consisting of the provided text inside an HTML * element node of the given name. * * \param[in,out] out The output functions structure. * \param[in] element_name The name of the new HTML element. * \param[in] id The CSS ID selector to apply to this element. * If NULL, no ID is added. * \param[in] class_name The CSS class selector to apply to this element. * If NULL, no class is added. * \param[in] text The text content of the node. */ xmlNodePtr pcmk__output_create_html_node(pcmk__output_t *out, const char *element_name, const char *id, const char *class_name, const char *text); xmlNode *pcmk__html_create(xmlNode *parent, const char *name, const char *id, const char *class_name); +void pcmk__html_set_title(const char *name); + /*! * \internal * \brief Add an HTML tag to the section. * * The arguments after name are a NULL-terminated list of keys and values, * all of which will be added as attributes to the given tag. For instance, * the following code would generate the tag * "": * * \code * pcmk__html_add_header(PCMK__XE_META, * PCMK__XA_HTTP_EQUIV, PCMK__VALUE_REFRESH, * PCMK__XA_CONTENT, "19", * NULL); * \endcode * * \param[in] name The HTML tag for the new node. * \param[in] ... A NULL-terminated key/value list of attributes. */ void pcmk__html_add_header(const char *name, ...) G_GNUC_NULL_TERMINATED; /*! * \internal * \brief Handle end-of-program error reporting * * \param[in,out] error A GError object potentially containing some error. * If NULL, do nothing. * \param[in,out] out The output functions structure. If NULL, any errors * will simply be printed to stderr. */ void pcmk__output_and_clear_error(GError **error, pcmk__output_t *out); int pcmk__xml_output_new(pcmk__output_t **out, xmlNodePtr *xml); void pcmk__xml_output_finish(pcmk__output_t *out, crm_exit_t exit_status, xmlNodePtr *xml); int pcmk__log_output_new(pcmk__output_t **out); int pcmk__text_output_new(pcmk__output_t **out, const char *filename); /*! * \internal * \brief Check whether older style XML output is enabled * * The legacy flag should be used sparingly. Its meaning depends on the context * in which it's used. * * \param[in] out Output object * * \return \c true if the \c legacy_xml flag is enabled for \p out, or \c false * otherwise */ // @COMPAT This can be removed when `crm_mon -X` and daemon metadata are removed bool pcmk__output_get_legacy_xml(pcmk__output_t *out); /*! * \internal * \brief Enable older style XML output * * The legacy flag should be used sparingly. Its meaning depends on the context * in which it's used. * * \param[in,out] out Output object */ // @COMPAT This can be removed when `crm_mon -X` and daemon metadata are removed void pcmk__output_set_legacy_xml(pcmk__output_t *out); /*! * \internal * \brief Enable using the element for lists * * \note This function is only used in limited places and should not be * used anywhere new. We are trying to discourage and ultimately remove * uses of this style of list. * * @COMPAT This can be removed when the stonith_admin and crm_resource * schemas can be changed */ void pcmk__output_enable_list_element(pcmk__output_t *out); /*! * \internal * \brief Select an updated return code for an operation on a \p pcmk__output_t * * This function helps to keep an up-to-date record of the most relevant return * code from a series of operations on a \p pcmk__output_t object. For example, * suppose the object has already produced some output, and we've saved a * \p pcmk_rc_ok return code. A new operation did not produce any output and * returned \p pcmk_rc_no_output. We can ignore the new \p pcmk_rc_no_output * return code and keep the previous \p pcmk_rc_ok return code. * * It prioritizes return codes as follows (from highest to lowest priority): * 1. Other return codes (unexpected errors) * 2. \p pcmk_rc_ok * 3. \p pcmk_rc_no_output * * \param[in] old_rc Saved return code from \p pcmk__output_t operations * \param[in] new_rc New return code from a \p pcmk__output_t operation * * \retval \p old_rc \p new_rc is \p pcmk_rc_no_output, or \p new_rc is * \p pcmk_rc_ok and \p old_rc is not \p pcmk_rc_no_output * \retval \p new_rc Otherwise */ static inline int pcmk__output_select_rc(int old_rc, int new_rc) { switch (new_rc) { case pcmk_rc_no_output: return old_rc; case pcmk_rc_ok: switch (old_rc) { case pcmk_rc_no_output: return new_rc; default: return old_rc; } default: return new_rc; } } #if defined(PCMK__UNIT_TESTING) /* If we are building libcrmcommon_test.a, add this accessor function so we can * inspect the internal formatters hash table. */ GHashTable *pcmk__output_formatters(void); #endif #define PCMK__OUTPUT_SPACER_IF(out_obj, cond) \ if (cond) { \ out->spacer(out); \ } #define PCMK__OUTPUT_LIST_HEADER(out_obj, cond, retcode, title...) \ if (retcode == pcmk_rc_no_output) { \ PCMK__OUTPUT_SPACER_IF(out_obj, cond); \ retcode = pcmk_rc_ok; \ out_obj->begin_list(out_obj, NULL, NULL, title); \ } #define PCMK__OUTPUT_LIST_FOOTER(out_obj, retcode) \ if (retcode == pcmk_rc_ok) { \ out_obj->end_list(out_obj); \ } #ifdef __cplusplus } #endif #endif // PCMK__CRM_COMMON_OUTPUT_INTERNAL__H diff --git a/lib/cib/cib_client.c b/lib/cib/cib_client.c index 220410dc2c..ad298e46b6 100644 --- a/lib/cib/cib_client.c +++ b/lib/cib/cib_client.c @@ -1,767 +1,779 @@ /* * Copyright 2004-2025 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 #include #include #include static GHashTable *cib_op_callback_table = NULL; static gint ciblib_GCompareFunc(gconstpointer a, gconstpointer b) { int rc = 0; const cib_notify_client_t *a_client = a; const cib_notify_client_t *b_client = b; CRM_CHECK(a_client->event != NULL && b_client->event != NULL, return 0); rc = strcmp(a_client->event, b_client->event); if (rc == 0) { if (a_client->callback == b_client->callback) { return 0; } else if (((long)a_client->callback) < ((long)b_client->callback)) { crm_trace("callbacks for %s are not equal: %p < %p", a_client->event, a_client->callback, b_client->callback); return -1; } crm_trace("callbacks for %s are not equal: %p > %p", a_client->event, a_client->callback, b_client->callback); return 1; } return rc; } static int cib_client_add_notify_callback(cib_t * cib, const char *event, void (*callback) (const char *event, xmlNode * msg)) { GList *list_item = NULL; cib_notify_client_t *new_client = NULL; if ((cib->variant != cib_native) && (cib->variant != cib_remote)) { return -EPROTONOSUPPORT; } crm_trace("Adding callback for %s events (%d)", event, g_list_length(cib->notify_list)); new_client = pcmk__assert_alloc(1, sizeof(cib_notify_client_t)); new_client->event = event; new_client->callback = callback; list_item = g_list_find_custom(cib->notify_list, new_client, ciblib_GCompareFunc); if (list_item != NULL) { crm_warn("Callback already present"); free(new_client); return -EINVAL; } else { cib->notify_list = g_list_append(cib->notify_list, new_client); cib->cmds->register_notification(cib, event, 1); crm_trace("Callback added (%d)", g_list_length(cib->notify_list)); } return pcmk_ok; } static int get_notify_list_event_count(cib_t *cib, const char *event) { int count = 0; for (GList *iter = g_list_first(cib->notify_list); iter != NULL; iter = iter->next) { cib_notify_client_t *client = (cib_notify_client_t *) iter->data; if (strcmp(client->event, event) == 0) { count++; } } crm_trace("event(%s) count : %d", event, count); return count; } static int cib_client_del_notify_callback(cib_t *cib, const char *event, void (*callback) (const char *event, xmlNode *msg)) { GList *list_item = NULL; cib_notify_client_t *new_client = NULL; if (cib->variant != cib_native && cib->variant != cib_remote) { return -EPROTONOSUPPORT; } if (get_notify_list_event_count(cib, event) == 0) { crm_debug("The callback of the event does not exist(%s)", event); return pcmk_ok; } crm_debug("Removing callback for %s events", event); new_client = pcmk__assert_alloc(1, sizeof(cib_notify_client_t)); new_client->event = event; new_client->callback = callback; list_item = g_list_find_custom(cib->notify_list, new_client, ciblib_GCompareFunc); if (list_item != NULL) { cib_notify_client_t *list_client = list_item->data; cib->notify_list = g_list_remove(cib->notify_list, list_client); free(list_client); crm_trace("Removed callback"); } else { crm_trace("Callback not present"); } if (get_notify_list_event_count(cib, event) == 0) { /* When there is not the registration of the event, the processing turns off a notice. */ cib->cmds->register_notification(cib, event, 0); } free(new_client); return pcmk_ok; } static gboolean cib_async_timeout_handler(gpointer data) { struct timer_rec_s *timer = data; crm_debug("Async call %d timed out after %ds", timer->call_id, timer->timeout); cib_native_callback(timer->cib, NULL, timer->call_id, -ETIME); // We remove the handler in remove_cib_op_callback() return G_SOURCE_CONTINUE; } static gboolean cib_client_register_callback_full(cib_t *cib, int call_id, int timeout, gboolean only_success, void *user_data, const char *callback_name, void (*callback)(xmlNode *, int, int, xmlNode *, void *), void (*free_func)(void *)) { cib_callback_client_t *blob = NULL; if (call_id < 0) { if (only_success == FALSE) { callback(NULL, call_id, call_id, NULL, user_data); } else { crm_warn("CIB call failed: %s", pcmk_strerror(call_id)); } if (user_data && free_func) { free_func(user_data); } return FALSE; } blob = pcmk__assert_alloc(1, sizeof(cib_callback_client_t)); blob->id = callback_name; blob->only_success = only_success; blob->user_data = user_data; blob->callback = callback; blob->free_func = free_func; if (timeout > 0) { struct timer_rec_s *async_timer = pcmk__assert_alloc(1, sizeof(struct timer_rec_s)); blob->timer = async_timer; async_timer->cib = cib; async_timer->call_id = call_id; async_timer->timeout = timeout * 1000; async_timer->ref = pcmk__create_timer(async_timer->timeout, cib_async_timeout_handler, async_timer); } crm_trace("Adding callback %s for call %d", callback_name, call_id); pcmk__intkey_table_insert(cib_op_callback_table, call_id, blob); return TRUE; } static gboolean cib_client_register_callback(cib_t *cib, int call_id, int timeout, gboolean only_success, void *user_data, const char *callback_name, void (*callback) (xmlNode *, int, int, xmlNode *, void *)) { return cib_client_register_callback_full(cib, call_id, timeout, only_success, user_data, callback_name, callback, NULL); } static int cib_client_noop(cib_t * cib, int call_options) { return cib_internal_op(cib, PCMK__CIB_REQUEST_NOOP, NULL, NULL, NULL, NULL, call_options, cib->user); } static int cib_client_ping(cib_t * cib, xmlNode ** output_data, int call_options) { return cib_internal_op(cib, CRM_OP_PING, NULL, NULL, NULL, output_data, call_options, cib->user); } static int cib_client_query(cib_t * cib, const char *section, xmlNode ** output_data, int call_options) { return cib->cmds->query_from(cib, NULL, section, output_data, call_options); } static int cib_client_query_from(cib_t * cib, const char *host, const char *section, xmlNode ** output_data, int call_options) { return cib_internal_op(cib, PCMK__CIB_REQUEST_QUERY, host, section, NULL, output_data, call_options, cib->user); } static int set_secondary(cib_t *cib, int call_options) { return cib_internal_op(cib, PCMK__CIB_REQUEST_SECONDARY, NULL, NULL, NULL, NULL, call_options, cib->user); } static int set_primary(cib_t *cib, int call_options) { return cib_internal_op(cib, PCMK__CIB_REQUEST_PRIMARY, NULL, NULL, NULL, NULL, call_options, cib->user); } static int cib_client_bump_epoch(cib_t * cib, int call_options) { return cib_internal_op(cib, PCMK__CIB_REQUEST_BUMP, NULL, NULL, NULL, NULL, call_options, cib->user); } static int cib_client_upgrade(cib_t * cib, int call_options) { return cib_internal_op(cib, PCMK__CIB_REQUEST_UPGRADE, NULL, NULL, NULL, NULL, call_options, cib->user); } static int cib_client_sync(cib_t * cib, const char *section, int call_options) { return cib->cmds->sync_from(cib, NULL, section, call_options); } static int cib_client_sync_from(cib_t * cib, const char *host, const char *section, int call_options) { return cib_internal_op(cib, PCMK__CIB_REQUEST_SYNC_TO_ALL, host, section, NULL, NULL, call_options, cib->user); } static int cib_client_create(cib_t * cib, const char *section, xmlNode * data, int call_options) { return cib_internal_op(cib, PCMK__CIB_REQUEST_CREATE, NULL, section, data, NULL, call_options, cib->user); } static int cib_client_modify(cib_t * cib, const char *section, xmlNode * data, int call_options) { return cib_internal_op(cib, PCMK__CIB_REQUEST_MODIFY, NULL, section, data, NULL, call_options, cib->user); } static int cib_client_replace(cib_t * cib, const char *section, xmlNode * data, int call_options) { return cib_internal_op(cib, PCMK__CIB_REQUEST_REPLACE, NULL, section, data, NULL, call_options, cib->user); } static int cib_client_delete(cib_t * cib, const char *section, xmlNode * data, int call_options) { return cib_internal_op(cib, PCMK__CIB_REQUEST_DELETE, NULL, section, data, NULL, call_options, cib->user); } static int cib_client_erase(cib_t * cib, xmlNode ** output_data, int call_options) { return cib_internal_op(cib, PCMK__CIB_REQUEST_ERASE, NULL, NULL, NULL, output_data, call_options, cib->user); } static int cib_client_init_transaction(cib_t *cib) { int rc = pcmk_rc_ok; if (cib == NULL) { return -EINVAL; } if (cib->transaction != NULL) { // A client can have at most one transaction at a time rc = pcmk_rc_already; } if (rc == pcmk_rc_ok) { cib->transaction = pcmk__xe_create(NULL, PCMK__XE_CIB_TRANSACTION); } if (rc != pcmk_rc_ok) { const char *client_id = NULL; cib->cmds->client_id(cib, NULL, &client_id); crm_err("Failed to initialize CIB transaction for client %s: %s", client_id, pcmk_rc_str(rc)); } return pcmk_rc2legacy(rc); } static int cib_client_end_transaction(cib_t *cib, bool commit, int call_options) { const char *client_id = NULL; int rc = pcmk_ok; if (cib == NULL) { return -EINVAL; } cib->cmds->client_id(cib, NULL, &client_id); client_id = pcmk__s(client_id, "(unidentified)"); if (commit) { if (cib->transaction == NULL) { rc = pcmk_rc_no_transaction; crm_err("Failed to commit transaction for CIB client %s: %s", client_id, pcmk_rc_str(rc)); return pcmk_rc2legacy(rc); } rc = cib_internal_op(cib, PCMK__CIB_REQUEST_COMMIT_TRANSACT, NULL, NULL, cib->transaction, NULL, call_options, cib->user); } else { // Discard always succeeds if (cib->transaction != NULL) { crm_trace("Discarded transaction for CIB client %s", client_id); } else { crm_trace("No transaction found for CIB client %s", client_id); } } pcmk__xml_free(cib->transaction); cib->transaction = NULL; return rc; } static int cib_client_fetch_schemas(cib_t *cib, xmlNode **output_data, const char *after_ver, int call_options) { xmlNode *data = pcmk__xe_create(NULL, PCMK__XA_SCHEMA); int rc = pcmk_ok; crm_xml_add(data, PCMK_XA_VERSION, after_ver); rc = cib_internal_op(cib, PCMK__CIB_REQUEST_SCHEMAS, NULL, NULL, data, output_data, call_options, NULL); pcmk__xml_free(data); return rc; } static void cib_client_set_user(cib_t *cib, const char *user) { pcmk__str_update(&(cib->user), user); } static void cib_destroy_op_callback(gpointer data) { cib_callback_client_t *blob = data; if (blob->timer && blob->timer->ref > 0) { g_source_remove(blob->timer->ref); } free(blob->timer); if (blob->user_data && blob->free_func) { blob->free_func(blob->user_data); } free(blob); } static void destroy_op_callback_table(void) { if (cib_op_callback_table != NULL) { g_hash_table_destroy(cib_op_callback_table); cib_op_callback_table = NULL; } } char * get_shadow_file(const char *suffix) { char *cib_home = NULL; char *fullname = NULL; char *name = crm_strdup_printf("shadow.%s", suffix); const char *dir = getenv("CIB_shadow_dir"); if (dir == NULL) { - uid_t uid = geteuid(); - struct passwd *pwent = getpwuid(uid); + /* @TODO This basically duplicates pcmk__uid2username(), but we need the + * password database entry, not just the user name from it. We should + * reduce the duplication. + */ + struct passwd *pwent = NULL; const char *user = NULL; + errno = 0; + pwent = getpwuid(geteuid()); + if (pwent) { user = pwent->pw_name; + } else { + // Save errno before getenv() + int rc = errno; + user = getenv("USER"); - crm_perror(LOG_ERR, - "Assuming %s because cannot get user details for user ID %d", - (user? user : "unprivileged user"), uid); + crm_warn("Could not get password database entry for effective user " + "ID %lld: %s. Assuming user is %s.", + (long long) geteuid(), + ((rc != 0)? strerror(rc) : "No matching entry found"), + pcmk__s(user, "unprivileged user")); } if (pcmk__strcase_any_of(user, "root", CRM_DAEMON_USER, NULL)) { dir = CRM_CONFIG_DIR; } else { const char *home = NULL; if ((home = getenv("HOME")) == NULL) { if (pwent) { home = pwent->pw_dir; } } dir = pcmk__get_tmpdir(); if (home && home[0] == '/') { int rc = 0; cib_home = crm_strdup_printf("%s/.cib", home); rc = mkdir(cib_home, 0700); if (rc < 0 && errno != EEXIST) { - crm_perror(LOG_ERR, "Couldn't create user-specific shadow directory: %s", - cib_home); - errno = 0; + crm_err("Couldn't create user-specific shadow directory " + "%s: %s", + cib_home, strerror(errno)); } else { dir = cib_home; } } } } fullname = crm_strdup_printf("%s/%s", dir, name); free(cib_home); free(name); return fullname; } cib_t * cib_shadow_new(const char *shadow) { cib_t *new_cib = NULL; char *shadow_file = NULL; CRM_CHECK(shadow != NULL, return NULL); shadow_file = get_shadow_file(shadow); new_cib = cib_file_new(shadow_file); free(shadow_file); return new_cib; } /*! * \brief Create a new CIB connection object * * Create a new live, remote, file, or shadow file CIB connection object based * on the values of CIB-related environment variables (CIB_shadow, CIB_file, * CIB_port, CIB_server, CIB_user, and CIB_passwd). The object will not be * connected. * * \return Newly allocated CIB connection object * \note The CIB API does not fully support opening multiple CIB connection * objects simultaneously, so the returned object should be treated as a * singleton. */ /* @TODO Ensure all APIs support multiple simultaneous CIB connection objects * (at least cib_free_callbacks() currently does not). */ cib_t * cib_new(void) { const char *value = getenv("CIB_shadow"); const char *server = NULL; const char *user = NULL; const char *pass = NULL; gboolean encrypted = TRUE; int port; if (!pcmk__str_empty(value)) { return cib_shadow_new(value); } value = getenv("CIB_file"); if (!pcmk__str_empty(value)) { return cib_file_new(value); } value = getenv("CIB_port"); if (pcmk__str_empty(value)) { return cib_native_new(); } /* We don't ensure port is valid (>= 0) because cib_new() currently can't * return NULL in practice, and introducing a NULL return here could cause * core dumps that would previously just cause signon() failures. */ pcmk__scan_port(value, &port); if (!crm_is_true(getenv("CIB_encrypted"))) { encrypted = FALSE; } server = getenv("CIB_server"); user = getenv("CIB_user"); pass = getenv("CIB_passwd"); if (pcmk__str_empty(user)) { user = CRM_DAEMON_USER; } if (pcmk__str_empty(server)) { server = "localhost"; } crm_debug("Initializing %s remote CIB access to %s:%d as user %s", (encrypted? "encrypted" : "plain-text"), server, port, user); return cib_remote_new(server, user, pass, port, encrypted); } /*! * \internal * \brief Create a generic CIB connection instance * * \return Newly allocated and initialized cib_t instance * * \note This is called by each variant's cib_*_new() function before setting * variant-specific values. */ cib_t * cib_new_variant(void) { cib_t *new_cib = NULL; new_cib = calloc(1, sizeof(cib_t)); if (new_cib == NULL) { return NULL; } remove_cib_op_callback(0, TRUE); /* remove all */ new_cib->call_id = 1; new_cib->variant = cib_undefined; new_cib->type = cib_no_connection; new_cib->state = cib_disconnected; new_cib->variant_opaque = NULL; new_cib->notify_list = NULL; /* the rest will get filled in by the variant constructor */ new_cib->cmds = calloc(1, sizeof(cib_api_operations_t)); if (new_cib->cmds == NULL) { free(new_cib); return NULL; } new_cib->cmds->add_notify_callback = cib_client_add_notify_callback; new_cib->cmds->del_notify_callback = cib_client_del_notify_callback; new_cib->cmds->register_callback = cib_client_register_callback; new_cib->cmds->register_callback_full = cib_client_register_callback_full; new_cib->cmds->noop = cib_client_noop; // Deprecated method new_cib->cmds->ping = cib_client_ping; new_cib->cmds->query = cib_client_query; new_cib->cmds->sync = cib_client_sync; new_cib->cmds->query_from = cib_client_query_from; new_cib->cmds->sync_from = cib_client_sync_from; new_cib->cmds->set_primary = set_primary; new_cib->cmds->set_secondary = set_secondary; new_cib->cmds->upgrade = cib_client_upgrade; new_cib->cmds->bump_epoch = cib_client_bump_epoch; new_cib->cmds->create = cib_client_create; new_cib->cmds->modify = cib_client_modify; new_cib->cmds->replace = cib_client_replace; new_cib->cmds->remove = cib_client_delete; new_cib->cmds->erase = cib_client_erase; new_cib->cmds->init_transaction = cib_client_init_transaction; new_cib->cmds->end_transaction = cib_client_end_transaction; new_cib->cmds->set_user = cib_client_set_user; new_cib->cmds->fetch_schemas = cib_client_fetch_schemas; return new_cib; } void cib_free_notify(cib_t *cib) { if (cib) { GList *list = cib->notify_list; while (list != NULL) { cib_notify_client_t *client = g_list_nth_data(list, 0); list = g_list_remove(list, client); free(client); } cib->notify_list = NULL; } } /*! * \brief Free all callbacks for a CIB connection * * \param[in,out] cib CIB connection to clean up */ void cib_free_callbacks(cib_t *cib) { cib_free_notify(cib); destroy_op_callback_table(); } /*! * \brief Free all memory used by CIB connection * * \param[in,out] cib CIB connection to delete */ void cib_delete(cib_t *cib) { cib_free_callbacks(cib); if (cib) { cib->cmds->free(cib); } } void remove_cib_op_callback(int call_id, gboolean all_callbacks) { if (all_callbacks) { destroy_op_callback_table(); cib_op_callback_table = pcmk__intkey_table(cib_destroy_op_callback); } else { pcmk__intkey_table_remove(cib_op_callback_table, call_id); } } int num_cib_op_callbacks(void) { if (cib_op_callback_table == NULL) { return 0; } return g_hash_table_size(cib_op_callback_table); } static void cib_dump_pending_op(gpointer key, gpointer value, gpointer user_data) { int call = GPOINTER_TO_INT(key); cib_callback_client_t *blob = value; crm_debug("Call %d (%s): pending", call, pcmk__s(blob->id, "without ID")); } void cib_dump_pending_callbacks(void) { if (cib_op_callback_table == NULL) { return; } return g_hash_table_foreach(cib_op_callback_table, cib_dump_pending_op, NULL); } cib_callback_client_t* cib__lookup_id (int call_id) { return pcmk__intkey_table_lookup(cib_op_callback_table, call_id); } // Deprecated functions kept only for backward API compatibility // LCOV_EXCL_START #include cib_t * cib_new_no_shadow(void) { const char *shadow = getenv("CIB_shadow"); cib_t *cib = NULL; unsetenv("CIB_shadow"); cib = cib_new(); if (shadow != NULL) { setenv("CIB_shadow", shadow, 1); } return cib; } // LCOV_EXCL_STOP // End deprecated API diff --git a/lib/cib/cib_file.c b/lib/cib/cib_file.c index 7348ce5f06..bcf089cb9e 100644 --- a/lib/cib/cib_file.c +++ b/lib/cib/cib_file.c @@ -1,1178 +1,1185 @@ /* * Original copyright 2004 International Business Machines * Later changes copyright 2008-2025 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 #include #include #include #include #include #include #include #define CIB_SERIES "cib" #define CIB_SERIES_MAX 100 #define CIB_SERIES_BZIP FALSE /* Must be false because archived copies are created with hard links */ #define CIB_LIVE_NAME CIB_SERIES ".xml" // key: client ID (const char *) -> value: client (cib_t *) static GHashTable *client_table = NULL; enum cib_file_flags { cib_file_flag_dirty = (1 << 0), cib_file_flag_live = (1 << 1), }; typedef struct cib_file_opaque_s { char *id; char *filename; uint32_t flags; // Group of enum cib_file_flags xmlNode *cib_xml; } cib_file_opaque_t; static int cib_file_process_commit_transaction(const char *op, int options, const char *section, xmlNode *req, xmlNode *input, xmlNode *existing_cib, xmlNode **result_cib, xmlNode **answer); /*! * \internal * \brief Add a CIB file client to client table * * \param[in] cib CIB client */ static void register_client(const cib_t *cib) { cib_file_opaque_t *private = cib->variant_opaque; if (client_table == NULL) { client_table = pcmk__strkey_table(NULL, NULL); } g_hash_table_insert(client_table, private->id, (gpointer) cib); } /*! * \internal * \brief Remove a CIB file client from client table * * \param[in] cib CIB client */ static void unregister_client(const cib_t *cib) { cib_file_opaque_t *private = cib->variant_opaque; if (client_table == NULL) { return; } g_hash_table_remove(client_table, private->id); /* @COMPAT: Add to crm_exit() when libcib and libcrmcommon are merged, * instead of destroying the client table when there are no more clients. */ if (g_hash_table_size(client_table) == 0) { g_hash_table_destroy(client_table); client_table = NULL; } } /*! * \internal * \brief Look up a CIB file client by its ID * * \param[in] client_id CIB client ID * * \return CIB client with matching ID if found, or \p NULL otherwise */ static cib_t * get_client(const char *client_id) { if (client_table == NULL) { return NULL; } return g_hash_table_lookup(client_table, (gpointer) client_id); } static const cib__op_fn_t cib_op_functions[] = { [cib__op_apply_patch] = cib_process_diff, [cib__op_bump] = cib_process_bump, [cib__op_commit_transact] = cib_file_process_commit_transaction, [cib__op_create] = cib_process_create, [cib__op_delete] = cib_process_delete, [cib__op_erase] = cib_process_erase, [cib__op_modify] = cib_process_modify, [cib__op_query] = cib_process_query, [cib__op_replace] = cib_process_replace, [cib__op_upgrade] = cib_process_upgrade, }; /* cib_file_backup() and cib_file_write_with_digest() need to chown the * written files only in limited circumstances, so these variables allow * that to be indicated without affecting external callers */ static uid_t cib_file_owner = 0; static uid_t cib_file_group = 0; static gboolean cib_do_chown = FALSE; #define cib_set_file_flags(cibfile, flags_to_set) do { \ (cibfile)->flags = pcmk__set_flags_as(__func__, __LINE__, \ LOG_TRACE, "CIB file", \ cibfile->filename, \ (cibfile)->flags, \ (flags_to_set), \ #flags_to_set); \ } while (0) #define cib_clear_file_flags(cibfile, flags_to_clear) do { \ (cibfile)->flags = pcmk__clear_flags_as(__func__, __LINE__, \ LOG_TRACE, "CIB file", \ cibfile->filename, \ (cibfile)->flags, \ (flags_to_clear), \ #flags_to_clear); \ } while (0) /*! * \internal * \brief Get the function that performs a given CIB file operation * * \param[in] operation Operation whose function to look up * * \return Function that performs \p operation for a CIB file client */ static cib__op_fn_t file_get_op_function(const cib__operation_t *operation) { enum cib__op_type type = operation->type; pcmk__assert(type >= 0); if (type >= PCMK__NELEM(cib_op_functions)) { return NULL; } return cib_op_functions[type]; } /*! * \internal * \brief Check whether a file is the live CIB * * \param[in] filename Name of file to check * * \return TRUE if file exists and its real path is same as live CIB's */ static gboolean cib_file_is_live(const char *filename) { gboolean same = FALSE; if (filename != NULL) { // Canonicalize file names for true comparison char *real_filename = NULL; if (pcmk__real_path(filename, &real_filename) == pcmk_rc_ok) { char *real_livename = NULL; if (pcmk__real_path(CRM_CONFIG_DIR "/" CIB_LIVE_NAME, &real_livename) == pcmk_rc_ok) { same = !strcmp(real_filename, real_livename); free(real_livename); } free(real_filename); } } return same; } static int cib_file_process_request(cib_t *cib, xmlNode *request, xmlNode **output) { int rc = pcmk_ok; const cib__operation_t *operation = NULL; cib__op_fn_t op_function = NULL; int call_id = 0; uint32_t call_options = cib_none; const char *op = pcmk__xe_get(request, PCMK__XA_CIB_OP); const char *section = pcmk__xe_get(request, PCMK__XA_CIB_SECTION); xmlNode *wrapper = pcmk__xe_first_child(request, PCMK__XE_CIB_CALLDATA, NULL, NULL); xmlNode *data = pcmk__xe_first_child(wrapper, NULL, NULL, NULL); bool changed = false; bool read_only = false; xmlNode *result_cib = NULL; xmlNode *cib_diff = NULL; cib_file_opaque_t *private = cib->variant_opaque; // We error checked these in callers cib__get_operation(op, &operation); op_function = file_get_op_function(operation); pcmk__xe_get_int(request, PCMK__XA_CIB_CALLID, &call_id); rc = pcmk__xe_get_flags(request, PCMK__XA_CIB_CALLOPT, &call_options, cib_none); if (rc != pcmk_rc_ok) { crm_warn("Couldn't parse options from request: %s", pcmk_rc_str(rc)); } read_only = !pcmk_is_set(operation->flags, cib__op_attr_modifies); // Mirror the logic in prepare_input() in the CIB manager if ((section != NULL) && pcmk__xe_is(data, PCMK_XE_CIB)) { data = pcmk_find_cib_element(data, section); } rc = cib_perform_op(cib, op, call_options, op_function, read_only, section, request, data, true, &changed, &private->cib_xml, &result_cib, &cib_diff, output); if (pcmk_is_set(call_options, cib_transaction)) { /* The rest of the logic applies only to the transaction as a whole, not * to individual requests. */ goto done; } if (rc == -pcmk_err_schema_validation) { // Show validation errors to stderr pcmk__validate_xml(result_cib, NULL, NULL, NULL); } else if ((rc == pcmk_ok) && !read_only) { pcmk__log_xml_patchset(LOG_DEBUG, cib_diff); if (result_cib != private->cib_xml) { pcmk__xml_free(private->cib_xml); private->cib_xml = result_cib; } cib_set_file_flags(private, cib_file_flag_dirty); } done: if ((result_cib != private->cib_xml) && (result_cib != *output)) { pcmk__xml_free(result_cib); } pcmk__xml_free(cib_diff); return rc; } static int cib_file_perform_op_delegate(cib_t *cib, const char *op, const char *host, const char *section, xmlNode *data, xmlNode **output_data, int call_options, const char *user_name) { int rc = pcmk_ok; xmlNode *request = NULL; xmlNode *output = NULL; cib_file_opaque_t *private = cib->variant_opaque; const cib__operation_t *operation = NULL; crm_info("Handling %s operation for %s as %s", pcmk__s(op, "invalid"), pcmk__s(section, "entire CIB"), pcmk__s(user_name, "default user")); if (output_data != NULL) { *output_data = NULL; } if (cib->state == cib_disconnected) { return -ENOTCONN; } rc = cib__get_operation(op, &operation); rc = pcmk_rc2legacy(rc); if (rc != pcmk_ok) { // @COMPAT: At compatibility break, use rc directly return -EPROTONOSUPPORT; } if (file_get_op_function(operation) == NULL) { // @COMPAT: At compatibility break, use EOPNOTSUPP crm_err("Operation %s is not supported by CIB file clients", op); return -EPROTONOSUPPORT; } cib__set_call_options(call_options, "file operation", cib_no_mtime); rc = cib__create_op(cib, op, host, section, data, call_options, user_name, NULL, &request); if (rc != pcmk_ok) { return rc; } crm_xml_add(request, PCMK__XA_ACL_TARGET, user_name); crm_xml_add(request, PCMK__XA_CIB_CLIENTID, private->id); if (pcmk_is_set(call_options, cib_transaction)) { rc = cib__extend_transaction(cib, request); goto done; } rc = cib_file_process_request(cib, request, &output); if ((output_data != NULL) && (output != NULL)) { if (output->doc == private->cib_xml->doc) { *output_data = pcmk__xml_copy(NULL, output); } else { *output_data = output; } } done: if ((output != NULL) && (output->doc != private->cib_xml->doc) && ((output_data == NULL) || (output != *output_data))) { pcmk__xml_free(output); } pcmk__xml_free(request); return rc; } /*! * \internal * \brief Read CIB from disk and validate it against XML schema * * \param[in] filename Name of file to read CIB from * \param[out] output Where to store the read CIB XML * * \return pcmk_ok on success, * -ENXIO if file does not exist (or stat() otherwise fails), or * -pcmk_err_schema_validation if XML doesn't parse or validate * \note If filename is the live CIB, this will *not* verify its digest, * though that functionality would be trivial to add here. * Also, this will *not* verify that the file is writable, * because some callers might not need to write. */ static int load_file_cib(const char *filename, xmlNode **output) { struct stat buf; xmlNode *root = NULL; /* Ensure file is readable */ if (strcmp(filename, "-") && (stat(filename, &buf) < 0)) { return -ENXIO; } /* Parse XML from file */ root = pcmk__xml_read(filename); if (root == NULL) { return -pcmk_err_schema_validation; } /* Add a status section if not already present */ if (pcmk__xe_first_child(root, PCMK_XE_STATUS, NULL, NULL) == NULL) { pcmk__xe_create(root, PCMK_XE_STATUS); } /* Validate XML against its specified schema */ if (!pcmk__configured_schema_validates(root)) { pcmk__xml_free(root); return -pcmk_err_schema_validation; } /* Remember the parsed XML for later use */ *output = root; return pcmk_ok; } static int cib_file_signon(cib_t *cib, const char *name, enum cib_conn_type type) { int rc = pcmk_ok; cib_file_opaque_t *private = cib->variant_opaque; if (private->filename == NULL) { rc = -EINVAL; } else { rc = load_file_cib(private->filename, &private->cib_xml); } if (rc == pcmk_ok) { crm_debug("Opened connection to local file '%s' for %s", private->filename, pcmk__s(name, "client")); cib->state = cib_connected_command; cib->type = cib_command; register_client(cib); } else { crm_info("Connection to local file '%s' for %s (client %s) failed: %s", private->filename, pcmk__s(name, "client"), private->id, pcmk_strerror(rc)); } return rc; } /*! * \internal * \brief Write out the in-memory CIB to a live CIB file * * \param[in] cib_root Root of XML tree to write * \param[in,out] path Full path to file to write * * \return 0 on success, -1 on failure */ static int cib_file_write_live(xmlNode *cib_root, char *path) { uid_t uid = geteuid(); struct passwd *daemon_pwent; char *sep = strrchr(path, '/'); const char *cib_dirname, *cib_filename; int rc = 0; /* Get the desired uid/gid */ errno = 0; daemon_pwent = getpwnam(CRM_DAEMON_USER); if (daemon_pwent == NULL) { - crm_perror(LOG_ERR, "Could not find %s user", CRM_DAEMON_USER); + crm_err("Could not find " CRM_DAEMON_USER " user: %s", strerror(errno)); return -1; } /* If we're root, we can change the ownership; * if we're daemon, anything we create will be OK; * otherwise, block access so we don't create wrong owner */ if ((uid != 0) && (uid != daemon_pwent->pw_uid)) { - crm_perror(LOG_ERR, "Must be root or %s to modify live CIB", - CRM_DAEMON_USER); + crm_err("Must be root or " CRM_DAEMON_USER " to modify live CIB"); return 0; } /* fancy footwork to separate dirname from filename * (we know the canonical name maps to the live CIB, * but the given name might be relative, or symlinked) */ if (sep == NULL) { /* no directory component specified */ cib_dirname = "./"; cib_filename = path; } else if (sep == path) { /* given name is in / */ cib_dirname = "/"; cib_filename = path + 1; } else { /* typical case; split given name into parts */ *sep = '\0'; cib_dirname = path; cib_filename = sep + 1; } /* if we're root, we want to update the file ownership */ if (uid == 0) { cib_file_owner = daemon_pwent->pw_uid; cib_file_group = daemon_pwent->pw_gid; cib_do_chown = TRUE; } /* write the file */ if (cib_file_write_with_digest(cib_root, cib_dirname, cib_filename) != pcmk_ok) { rc = -1; } /* turn off file ownership changes, for other callers */ if (uid == 0) { cib_do_chown = FALSE; } /* undo fancy stuff */ if ((sep != NULL) && (*sep == '\0')) { *sep = '/'; } return rc; } /*! * \internal * \brief Sign-off method for CIB file variants * * This will write the file to disk if needed, and free the in-memory CIB. If * the file is the live CIB, it will compute and write a signature as well. * * \param[in,out] cib CIB object to sign off * * \return pcmk_ok on success, pcmk_err_generic on failure * \todo This method should refuse to write the live CIB if the CIB manager is * running. */ static int cib_file_signoff(cib_t *cib) { int rc = pcmk_ok; cib_file_opaque_t *private = cib->variant_opaque; crm_debug("Disconnecting from the CIB manager"); cib->state = cib_disconnected; cib->type = cib_no_connection; unregister_client(cib); cib->cmds->end_transaction(cib, false, cib_none); /* If the in-memory CIB has been changed, write it to disk */ if (pcmk_is_set(private->flags, cib_file_flag_dirty)) { /* If this is the live CIB, write it out with a digest */ if (pcmk_is_set(private->flags, cib_file_flag_live)) { if (cib_file_write_live(private->cib_xml, private->filename) < 0) { rc = pcmk_err_generic; } /* Otherwise, it's a simple write */ } else { bool compress = g_str_has_suffix(private->filename, ".bz2"); if (pcmk__xml_write_file(private->cib_xml, private->filename, compress) != pcmk_rc_ok) { rc = pcmk_err_generic; } } if (rc == pcmk_ok) { crm_info("Wrote CIB to %s", private->filename); cib_clear_file_flags(private, cib_file_flag_dirty); } else { crm_err("Could not write CIB to %s", private->filename); } } /* Free the in-memory CIB */ pcmk__xml_free(private->cib_xml); private->cib_xml = NULL; return rc; } static int cib_file_free(cib_t *cib) { int rc = pcmk_ok; if (cib->state != cib_disconnected) { rc = cib_file_signoff(cib); } if (rc == pcmk_ok) { cib_file_opaque_t *private = cib->variant_opaque; free(private->id); free(private->filename); free(private); free(cib->cmds); free(cib->user); free(cib); } else { fprintf(stderr, "Couldn't sign off: %d\n", rc); } return rc; } static int cib_file_register_notification(cib_t *cib, const char *callback, int enabled) { return -EPROTONOSUPPORT; } static int cib_file_set_connection_dnotify(cib_t *cib, void (*dnotify) (gpointer user_data)) { return -EPROTONOSUPPORT; } /*! * \internal * \brief Get the given CIB connection's unique client identifier * * \param[in] cib CIB connection * \param[out] async_id If not \p NULL, where to store asynchronous client ID * \param[out] sync_id If not \p NULL, where to store synchronous client ID * * \return Legacy Pacemaker return code * * \note This is the \p cib_file variant implementation of * \p cib_api_operations_t:client_id(). */ static int cib_file_client_id(const cib_t *cib, const char **async_id, const char **sync_id) { cib_file_opaque_t *private = cib->variant_opaque; if (async_id != NULL) { *async_id = private->id; } if (sync_id != NULL) { *sync_id = private->id; } return pcmk_ok; } cib_t * cib_file_new(const char *cib_location) { cib_t *cib = NULL; cib_file_opaque_t *private = NULL; char *filename = NULL; if (cib_location == NULL) { cib_location = getenv("CIB_file"); if (cib_location == NULL) { return NULL; // Shouldn't be possible if we were called internally } } cib = cib_new_variant(); if (cib == NULL) { return NULL; } filename = strdup(cib_location); if (filename == NULL) { free(cib); return NULL; } private = calloc(1, sizeof(cib_file_opaque_t)); if (private == NULL) { free(cib); free(filename); return NULL; } private->id = crm_generate_uuid(); private->filename = filename; cib->variant = cib_file; cib->variant_opaque = private; private->flags = 0; if (cib_file_is_live(cib_location)) { cib_set_file_flags(private, cib_file_flag_live); crm_trace("File %s detected as live CIB", cib_location); } /* assign variant specific ops */ cib->delegate_fn = cib_file_perform_op_delegate; cib->cmds->signon = cib_file_signon; cib->cmds->signoff = cib_file_signoff; cib->cmds->free = cib_file_free; cib->cmds->register_notification = cib_file_register_notification; cib->cmds->set_connection_dnotify = cib_file_set_connection_dnotify; cib->cmds->client_id = cib_file_client_id; return cib; } /*! * \internal * \brief Compare the calculated digest of an XML tree against a signature file * * \param[in] root Root of XML tree to compare * \param[in] sigfile Name of signature file containing digest to compare * * \return TRUE if digests match or signature file does not exist, else FALSE */ static gboolean cib_file_verify_digest(xmlNode *root, const char *sigfile) { gboolean passed = FALSE; char *expected; int rc = pcmk__file_contents(sigfile, &expected); switch (rc) { case pcmk_rc_ok: if (expected == NULL) { crm_err("On-disk digest at %s is empty", sigfile); return FALSE; } break; case ENOENT: crm_warn("No on-disk digest present at %s", sigfile); return TRUE; default: crm_err("Could not read on-disk digest from %s: %s", sigfile, pcmk_rc_str(rc)); return FALSE; } passed = pcmk__verify_digest(root, expected); free(expected); return passed; } /*! * \internal * \brief Read an XML tree from a file and verify its digest * * \param[in] filename Name of XML file to read * \param[in] sigfile Name of signature file containing digest to compare * \param[out] root If non-NULL, will be set to pointer to parsed XML tree * * \return 0 if file was successfully read, parsed and verified, otherwise: * -errno on stat() failure, * -pcmk_err_cib_corrupt if file size is 0 or XML is not parseable, or * -pcmk_err_cib_modified if digests do not match * \note If root is non-NULL, it is the caller's responsibility to free *root on * successful return. */ int cib_file_read_and_verify(const char *filename, const char *sigfile, xmlNode **root) { int s_res; struct stat buf; char *local_sigfile = NULL; xmlNode *local_root = NULL; pcmk__assert(filename != NULL); if (root) { *root = NULL; } /* Verify that file exists and its size is nonzero */ s_res = stat(filename, &buf); if (s_res < 0) { - crm_perror(LOG_WARNING, "Could not verify cluster configuration file %s", filename); + crm_warn("Could not verify cluster configuration file %s: " + "stat() failed: %s", + filename, strerror(errno)); return -errno; } else if (buf.st_size == 0) { crm_warn("Cluster configuration file %s is corrupt (size is zero)", filename); return -pcmk_err_cib_corrupt; } /* Parse XML */ local_root = pcmk__xml_read(filename); if (local_root == NULL) { crm_warn("Cluster configuration file %s is corrupt (unparseable as XML)", filename); return -pcmk_err_cib_corrupt; } /* If sigfile is not specified, use original file name plus .sig */ if (sigfile == NULL) { sigfile = local_sigfile = crm_strdup_printf("%s.sig", filename); } /* Verify that digests match */ if (cib_file_verify_digest(local_root, sigfile) == FALSE) { free(local_sigfile); pcmk__xml_free(local_root); return -pcmk_err_cib_modified; } free(local_sigfile); if (root) { *root = local_root; } else { pcmk__xml_free(local_root); } return pcmk_ok; } /*! * \internal * \brief Back up a CIB * * \param[in] cib_dirname Directory containing CIB file and backups * \param[in] cib_filename Name (relative to cib_dirname) of CIB file to back up * * \return 0 on success, -1 on error */ static int cib_file_backup(const char *cib_dirname, const char *cib_filename) { int rc = 0; unsigned int seq = 0U; char *cib_path = crm_strdup_printf("%s/%s", cib_dirname, cib_filename); char *cib_digest = crm_strdup_printf("%s.sig", cib_path); char *backup_path; char *backup_digest; // Determine backup and digest file names if (pcmk__read_series_sequence(cib_dirname, CIB_SERIES, &seq) != pcmk_rc_ok) { // @TODO maybe handle errors better ... seq = 0U; } backup_path = pcmk__series_filename(cib_dirname, CIB_SERIES, seq, CIB_SERIES_BZIP); backup_digest = crm_strdup_printf("%s.sig", backup_path); /* Remove the old backups if they exist */ unlink(backup_path); unlink(backup_digest); /* Back up the CIB, by hard-linking it to the backup name */ if ((link(cib_path, backup_path) < 0) && (errno != ENOENT)) { - crm_perror(LOG_ERR, "Could not archive %s by linking to %s", - cib_path, backup_path); + crm_err("Could not archive %s by linking to %s: %s", cib_path, + backup_path, strerror(errno)); rc = -1; /* Back up the CIB signature similarly */ } else if ((link(cib_digest, backup_digest) < 0) && (errno != ENOENT)) { - crm_perror(LOG_ERR, "Could not archive %s by linking to %s", - cib_digest, backup_digest); + crm_err("Could not archive %s by linking to %s: %s", cib_digest, + backup_digest, strerror(errno)); rc = -1; /* Update the last counter and ensure everything is sync'd to media */ } else { pcmk__write_series_sequence(cib_dirname, CIB_SERIES, ++seq, CIB_SERIES_MAX); if (cib_do_chown) { int rc2; if ((chown(backup_path, cib_file_owner, cib_file_group) < 0) - && (errno != ENOENT)) { - crm_perror(LOG_ERR, "Could not set owner of %s", backup_path); + && (errno != ENOENT)) { + + crm_err("Could not set owner of %s: %s", backup_path, + strerror(errno)); rc = -1; } if ((chown(backup_digest, cib_file_owner, cib_file_group) < 0) - && (errno != ENOENT)) { - crm_perror(LOG_ERR, "Could not set owner of %s", backup_digest); + && (errno != ENOENT)) { + + crm_err("Could not set owner of %s: %s", backup_digest, + strerror(errno)); rc = -1; } rc2 = pcmk__chown_series_sequence(cib_dirname, CIB_SERIES, cib_file_owner, cib_file_group); if (rc2 != pcmk_rc_ok) { crm_err("Could not set owner of sequence file in %s: %s", cib_dirname, pcmk_rc_str(rc2)); rc = -1; } } pcmk__sync_directory(cib_dirname); crm_info("Archived previous version as %s", backup_path); } free(cib_path); free(cib_digest); free(backup_path); free(backup_digest); return rc; } /*! * \internal * \brief Prepare CIB XML to be written to disk * * Set \c PCMK_XA_NUM_UPDATES to 0, set \c PCMK_XA_CIB_LAST_WRITTEN to the * current timestamp, and strip out the status section. * * \param[in,out] root Root of CIB XML tree * * \return void */ static void cib_file_prepare_xml(xmlNode *root) { xmlNode *cib_status_root = NULL; /* Always write out with num_updates=0 and current last-written timestamp */ crm_xml_add(root, PCMK_XA_NUM_UPDATES, "0"); pcmk__xe_add_last_written(root); /* Delete status section before writing to file, because * we discard it on startup anyway, and users get confused by it */ cib_status_root = pcmk__xe_first_child(root, PCMK_XE_STATUS, NULL, NULL); CRM_CHECK(cib_status_root != NULL, return); pcmk__xml_free(cib_status_root); } /*! * \internal * \brief Write CIB to disk, along with a signature file containing its digest * * \param[in,out] cib_root Root of XML tree to write * \param[in] cib_dirname Directory containing CIB and signature files * \param[in] cib_filename Name (relative to cib_dirname) of file to write * * \return pcmk_ok on success, * pcmk_err_cib_modified if existing cib_filename doesn't match digest, * pcmk_err_cib_backup if existing cib_filename couldn't be backed up, * or pcmk_err_cib_save if new cib_filename couldn't be saved */ int cib_file_write_with_digest(xmlNode *cib_root, const char *cib_dirname, const char *cib_filename) { int exit_rc = pcmk_ok; int rc, fd; char *digest = NULL; /* Detect CIB version for diagnostic purposes */ const char *epoch = pcmk__xe_get(cib_root, PCMK_XA_EPOCH); const char *admin_epoch = pcmk__xe_get(cib_root, PCMK_XA_ADMIN_EPOCH); /* Determine full CIB and signature pathnames */ char *cib_path = crm_strdup_printf("%s/%s", cib_dirname, cib_filename); char *digest_path = crm_strdup_printf("%s.sig", cib_path); /* Create temporary file name patterns for writing out CIB and signature */ char *tmp_cib = crm_strdup_printf("%s/cib.XXXXXX", cib_dirname); char *tmp_digest = crm_strdup_printf("%s/cib.XXXXXX", cib_dirname); /* Ensure the admin didn't modify the existing CIB underneath us */ crm_trace("Reading cluster configuration file %s", cib_path); rc = cib_file_read_and_verify(cib_path, NULL, NULL); if ((rc != pcmk_ok) && (rc != -ENOENT)) { crm_err("%s was manually modified while the cluster was active!", cib_path); exit_rc = pcmk_err_cib_modified; goto cleanup; } /* Back up the existing CIB */ if (cib_file_backup(cib_dirname, cib_filename) < 0) { exit_rc = pcmk_err_cib_backup; goto cleanup; } crm_debug("Writing CIB to disk"); umask(S_IWGRP | S_IWOTH | S_IROTH); cib_file_prepare_xml(cib_root); /* Write the CIB to a temporary file, so we can deploy (near) atomically */ fd = mkstemp(tmp_cib); if (fd < 0) { - crm_perror(LOG_ERR, "Couldn't open temporary file %s for writing CIB", - tmp_cib); + crm_err("Couldn't open temporary file %s for writing CIB: %s", tmp_cib, + strerror(errno)); exit_rc = pcmk_err_cib_save; goto cleanup; } /* Protect the temporary file */ if (fchmod(fd, S_IRUSR | S_IWUSR) < 0) { - crm_perror(LOG_ERR, "Couldn't protect temporary file %s for writing CIB", - tmp_cib); + crm_err("Couldn't protect temporary file %s for writing CIB: %s", + tmp_cib, strerror(errno)); exit_rc = pcmk_err_cib_save; goto cleanup; } if (cib_do_chown && (fchown(fd, cib_file_owner, cib_file_group) < 0)) { - crm_perror(LOG_ERR, "Couldn't protect temporary file %s for writing CIB", - tmp_cib); + crm_err("Couldn't protect temporary file %s for writing CIB: %s", + tmp_cib, strerror(errno)); exit_rc = pcmk_err_cib_save; goto cleanup; } /* Write out the CIB */ if (pcmk__xml_write_fd(cib_root, tmp_cib, fd) != pcmk_rc_ok) { crm_err("Changes couldn't be written to %s", tmp_cib); exit_rc = pcmk_err_cib_save; goto cleanup; } /* Calculate CIB digest */ digest = pcmk__digest_on_disk_cib(cib_root); pcmk__assert(digest != NULL); crm_info("Wrote version %s.%s.0 of the CIB to disk (digest: %s)", (admin_epoch ? admin_epoch : "0"), (epoch ? epoch : "0"), digest); /* Write the CIB digest to a temporary file */ fd = mkstemp(tmp_digest); if (fd < 0) { - crm_perror(LOG_ERR, "Could not create temporary file for CIB digest"); + crm_err("Could not create temporary file %s for CIB digest: %s", + tmp_digest, strerror(errno)); exit_rc = pcmk_err_cib_save; goto cleanup; } if (cib_do_chown && (fchown(fd, cib_file_owner, cib_file_group) < 0)) { - crm_perror(LOG_ERR, "Couldn't protect temporary file %s for writing CIB", - tmp_cib); + crm_err("Couldn't protect temporary file %s for writing CIB: %s", + tmp_cib, strerror(errno)); exit_rc = pcmk_err_cib_save; close(fd); goto cleanup; } rc = pcmk__write_sync(fd, digest); if (rc != pcmk_rc_ok) { crm_err("Could not write digest to %s: %s", tmp_digest, pcmk_rc_str(rc)); exit_rc = pcmk_err_cib_save; close(fd); goto cleanup; } close(fd); crm_debug("Wrote digest %s to disk", digest); /* Verify that what we wrote is sane */ crm_info("Reading cluster configuration file %s (digest: %s)", tmp_cib, tmp_digest); rc = cib_file_read_and_verify(tmp_cib, tmp_digest, NULL); pcmk__assert(rc == 0); /* Rename temporary files to live, and sync directory changes to media */ crm_debug("Activating %s", tmp_cib); if (rename(tmp_cib, cib_path) < 0) { - crm_perror(LOG_ERR, "Couldn't rename %s as %s", tmp_cib, cib_path); + crm_err("Couldn't rename %s as %s: %s", tmp_cib, cib_path, + strerror(errno)); exit_rc = pcmk_err_cib_save; } if (rename(tmp_digest, digest_path) < 0) { - crm_perror(LOG_ERR, "Couldn't rename %s as %s", tmp_digest, - digest_path); + crm_err("Couldn't rename %s as %s: %s", tmp_digest, digest_path, + strerror(errno)); exit_rc = pcmk_err_cib_save; } pcmk__sync_directory(cib_dirname); cleanup: free(cib_path); free(digest_path); free(digest); free(tmp_digest); free(tmp_cib); return exit_rc; } /*! * \internal * \brief Process requests in a CIB transaction * * Stop when a request fails or when all requests have been processed. * * \param[in,out] cib CIB client * \param[in,out] transaction CIB transaction * * \return Standard Pacemaker return code */ static int cib_file_process_transaction_requests(cib_t *cib, xmlNode *transaction) { cib_file_opaque_t *private = cib->variant_opaque; for (xmlNode *request = pcmk__xe_first_child(transaction, PCMK__XE_CIB_COMMAND, NULL, NULL); request != NULL; request = pcmk__xe_next(request, PCMK__XE_CIB_COMMAND)) { xmlNode *output = NULL; const char *op = pcmk__xe_get(request, PCMK__XA_CIB_OP); int rc = cib_file_process_request(cib, request, &output); rc = pcmk_legacy2rc(rc); if (rc != pcmk_rc_ok) { crm_err("Aborting transaction for CIB file client (%s) on file " "'%s' due to failed %s request: %s", private->id, private->filename, op, pcmk_rc_str(rc)); crm_log_xml_info(request, "Failed request"); return rc; } crm_trace("Applied %s request to transaction working CIB for CIB file " "client (%s) on file '%s'", op, private->id, private->filename); crm_log_xml_trace(request, "Successful request"); } return pcmk_rc_ok; } /*! * \internal * \brief Commit a given CIB file client's transaction to a working CIB copy * * \param[in,out] cib CIB file client * \param[in] transaction CIB transaction * \param[in,out] result_cib Where to store result CIB * * \return Standard Pacemaker return code * * \note The caller is responsible for replacing the \p cib argument's * \p private->cib_xml with \p result_cib on success, and for freeing * \p result_cib using \p pcmk__xml_free() on failure. */ static int cib_file_commit_transaction(cib_t *cib, xmlNode *transaction, xmlNode **result_cib) { int rc = pcmk_rc_ok; cib_file_opaque_t *private = cib->variant_opaque; xmlNode *saved_cib = private->cib_xml; CRM_CHECK(pcmk__xe_is(transaction, PCMK__XE_CIB_TRANSACTION), return pcmk_rc_no_transaction); /* *result_cib should be a copy of private->cib_xml (created by * cib_perform_op()). If not, make a copy now. Change tracking isn't * strictly required here because: * * Each request in the transaction will have changes tracked and ACLs * checked if appropriate. * * cib_perform_op() will infer changes for the commit request at the end. */ CRM_CHECK((*result_cib != NULL) && (*result_cib != private->cib_xml), *result_cib = pcmk__xml_copy(NULL, private->cib_xml)); crm_trace("Committing transaction for CIB file client (%s) on file '%s' to " "working CIB", private->id, private->filename); // Apply all changes to a working copy of the CIB private->cib_xml = *result_cib; rc = cib_file_process_transaction_requests(cib, transaction); crm_trace("Transaction commit %s for CIB file client (%s) on file '%s'", ((rc == pcmk_rc_ok)? "succeeded" : "failed"), private->id, private->filename); /* Some request types (for example, erase) may have freed private->cib_xml * (the working copy) and pointed it at a new XML object. In that case, it * follows that *result_cib (the working copy) was freed. * * Point *result_cib at the updated working copy stored in private->cib_xml. */ *result_cib = private->cib_xml; // Point private->cib_xml back to the unchanged original copy private->cib_xml = saved_cib; return rc; } static int cib_file_process_commit_transaction(const char *op, int options, const char *section, xmlNode *req, xmlNode *input, xmlNode *existing_cib, xmlNode **result_cib, xmlNode **answer) { int rc = pcmk_rc_ok; const char *client_id = pcmk__xe_get(req, PCMK__XA_CIB_CLIENTID); cib_t *cib = NULL; CRM_CHECK(client_id != NULL, return -EINVAL); cib = get_client(client_id); CRM_CHECK(cib != NULL, return -EINVAL); rc = cib_file_commit_transaction(cib, input, result_cib); if (rc != pcmk_rc_ok) { cib_file_opaque_t *private = cib->variant_opaque; crm_err("Could not commit transaction for CIB file client (%s) on " "file '%s': %s", private->id, private->filename, pcmk_rc_str(rc)); } return pcmk_rc2legacy(rc); } diff --git a/lib/cluster/election.c b/lib/cluster/election.c index 0c2bf8d683..1e40a2bd24 100644 --- a/lib/cluster/election.c +++ b/lib/cluster/election.c @@ -1,742 +1,743 @@ /* * Copyright 2004-2025 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 "crmcluster_private.h" #define STORM_INTERVAL 2 /* in seconds */ struct pcmk__election { enum election_result state; // Current state of election guint count; // How many times local node has voted void (*cb)(pcmk_cluster_t *); // Function to call if election is won GHashTable *voted; // Key = node name, value = how node voted mainloop_timer_t *timeout; // When to abort if all votes not received int election_wins; // Track wins, for storm detection bool wrote_blackbox; // Write a storm blackbox at most once time_t expires; // When storm detection period ends time_t last_election_loss; // When dampening period ends }; static void election_complete(pcmk_cluster_t *cluster) { pcmk__assert((cluster != NULL) && (cluster->priv->election != NULL)); cluster->priv->election->state = election_won; if (cluster->priv->election->cb != NULL) { cluster->priv->election->cb(cluster); } election_reset(cluster); } static gboolean election_timer_cb(gpointer user_data) { pcmk_cluster_t *cluster = user_data; crm_info("Declaring local node as winner after election timed out"); election_complete(cluster); return FALSE; } /*! * \internal * \brief Get current state of an election * * \param[in] cluster Cluster with election * * \return Current state of \e */ enum election_result election_state(const pcmk_cluster_t *cluster) { if ((cluster == NULL) || (cluster->priv->election == NULL)) { return election_error; } return cluster->priv->election->state; } /* The local node will be declared the winner if missing votes are not received * within this time. The value is chosen to be the same as the default for the * election-timeout cluster option. */ #define ELECTION_TIMEOUT_MS 120000 /*! * \internal * \brief Track election state in a cluster * * Every node that wishes to participate in an election must initialize the * election once, typically at start-up. * * \param[in] cluster Cluster that election is for * \param[in] cb Function to call if local node wins election */ void election_init(pcmk_cluster_t *cluster, void (*cb)(pcmk_cluster_t *)) { const char *name = pcmk__s(crm_system_name, "election"); CRM_CHECK(cluster->priv->election == NULL, return); cluster->priv->election = pcmk__assert_alloc(1, sizeof(pcmk__election_t)); cluster->priv->election->cb = cb; cluster->priv->election->timeout = mainloop_timer_add(name, ELECTION_TIMEOUT_MS, FALSE, election_timer_cb, cluster); } /*! * \internal * \brief Disregard any previous vote by specified peer * * This discards any recorded vote from a specified peer. Election users should * call this whenever a voting peer becomes inactive. * * \param[in,out] cluster Cluster with election * \param[in] uname Name of peer to disregard */ void election_remove(pcmk_cluster_t *cluster, const char *uname) { if ((cluster != NULL) && (cluster->priv->election != NULL) && (uname != NULL) && (cluster->priv->election->voted != NULL)) { crm_trace("Discarding (no-)vote from lost peer %s", uname); g_hash_table_remove(cluster->priv->election->voted, uname); } } /*! * \internal * \brief Stop election timer and disregard all votes * * \param[in,out] cluster Cluster with election */ void election_reset(pcmk_cluster_t *cluster) { if ((cluster != NULL) && (cluster->priv->election != NULL)) { crm_trace("Resetting election"); mainloop_timer_stop(cluster->priv->election->timeout); if (cluster->priv->election->voted != NULL) { g_hash_table_destroy(cluster->priv->election->voted); cluster->priv->election->voted = NULL; } } } /*! * \internal * \brief Free an election object * * Free all memory associated with an election object, stopping its * election timer (if running). * * \param[in,out] cluster Cluster with election */ void election_fini(pcmk_cluster_t *cluster) { if ((cluster != NULL) && (cluster->priv->election != NULL)) { election_reset(cluster); crm_trace("Destroying election"); mainloop_timer_del(cluster->priv->election->timeout); free(cluster->priv->election); cluster->priv->election = NULL; } } static void election_timeout_start(pcmk_cluster_t *cluster) { mainloop_timer_start(cluster->priv->election->timeout); } /*! * \internal * \brief Stop an election's timer, if running * * \param[in,out] cluster Cluster with election */ void election_timeout_stop(pcmk_cluster_t *cluster) { if ((cluster != NULL) && (cluster->priv->election != NULL)) { mainloop_timer_stop(cluster->priv->election->timeout); } } /*! * \internal * \brief Change an election's timeout (restarting timer if running) * * \param[in,out] cluster Cluster with election * \param[in] period New timeout */ void election_timeout_set_period(pcmk_cluster_t *cluster, guint period) { CRM_CHECK((cluster != NULL) && (cluster->priv->election != NULL), return); mainloop_timer_set_period(cluster->priv->election->timeout, period); } static int get_uptime(struct timeval *output) { static time_t expires = 0; static struct rusage info; time_t tm_now = time(NULL); if (expires < tm_now) { int rc = 0; - info.ru_utime.tv_sec = 0; - info.ru_utime.tv_usec = 0; - rc = getrusage(RUSAGE_SELF, &info); - output->tv_sec = 0; output->tv_usec = 0; + info.ru_utime.tv_sec = 0; + info.ru_utime.tv_usec = 0; + + rc = getrusage(RUSAGE_SELF, &info); if (rc < 0) { - crm_perror(LOG_ERR, "Could not calculate the current uptime"); + crm_err("Could not calculate the current uptime: %s", + strerror(errno)); expires = 0; return -1; } crm_debug("Current CPU usage is: %lds, %ldus", (long)info.ru_utime.tv_sec, (long)info.ru_utime.tv_usec); } expires = tm_now + STORM_INTERVAL; /* N seconds after the last _access_ */ output->tv_sec = info.ru_utime.tv_sec; output->tv_usec = info.ru_utime.tv_usec; return 1; } static int compare_age(struct timeval your_age) { struct timeval our_age; get_uptime(&our_age); /* If an error occurred, our_age will be compared as {0,0} */ if (our_age.tv_sec > your_age.tv_sec) { crm_debug("Win: %ld vs %ld (seconds)", (long)our_age.tv_sec, (long)your_age.tv_sec); return 1; } else if (our_age.tv_sec < your_age.tv_sec) { crm_debug("Lose: %ld vs %ld (seconds)", (long)our_age.tv_sec, (long)your_age.tv_sec); return -1; } else if (our_age.tv_usec > your_age.tv_usec) { crm_debug("Win: %ld.%06ld vs %ld.%06ld (usec)", (long)our_age.tv_sec, (long)our_age.tv_usec, (long)your_age.tv_sec, (long)your_age.tv_usec); return 1; } else if (our_age.tv_usec < your_age.tv_usec) { crm_debug("Lose: %ld.%06ld vs %ld.%06ld (usec)", (long)our_age.tv_sec, (long)our_age.tv_usec, (long)your_age.tv_sec, (long)your_age.tv_usec); return -1; } return 0; } /*! * \internal * \brief Start a new election by offering local node's candidacy * * Broadcast a "vote" election message containing the local node's ID, * (incremented) election counter, and uptime, and start the election timer. * * \param[in,out] cluster Cluster with election * * \note Any nodes agreeing to the candidacy will send a "no-vote" reply, and if * all active peers do so, or if the election times out, the local node * wins the election. (If we lose to any peer vote, we will stop the * timer, so a timeout means we did not lose -- either some peer did not * vote, or we did not call election_check() in time.) */ void election_vote(pcmk_cluster_t *cluster) { struct timeval age; xmlNode *vote = NULL; pcmk__node_status_t *our_node = NULL; const char *message_type = NULL; CRM_CHECK((cluster != NULL) && (cluster->priv->election != NULL), return); if (cluster->priv->node_name == NULL) { crm_err("Cannot start an election: Local node name unknown"); return; } our_node = pcmk__get_node(0, cluster->priv->node_name, NULL, pcmk__node_search_cluster_member); if (!pcmk__cluster_is_node_active(our_node)) { crm_trace("Cannot vote yet: local node not connected to cluster"); return; } election_reset(cluster); cluster->priv->election->state = election_in_progress; message_type = pcmk__server_message_type(cluster->priv->server); /* @COMPAT We use message_type as the sender and recipient system for * backward compatibility (see T566). */ vote = pcmk__new_request(cluster->priv->server, message_type, NULL, message_type, CRM_OP_VOTE, NULL); cluster->priv->election->count++; crm_xml_add(vote, PCMK__XA_ELECTION_OWNER, pcmk__cluster_get_xml_id(our_node)); crm_xml_add_int(vote, PCMK__XA_ELECTION_ID, cluster->priv->election->count); // Warning: PCMK__XA_ELECTION_AGE_NANO_SEC value is actually microseconds get_uptime(&age); crm_xml_add_timeval(vote, PCMK__XA_ELECTION_AGE_SEC, PCMK__XA_ELECTION_AGE_NANO_SEC, &age); pcmk__cluster_send_message(NULL, cluster->priv->server, vote); pcmk__xml_free(vote); crm_debug("Started election round %u", cluster->priv->election->count); election_timeout_start(cluster); return; } /*! * \internal * \brief Check whether local node has won an election * * If all known peers have sent no-vote messages, stop the election timer, set * the election state to won, and call any registered win callback. * * \param[in,out] cluster Cluster with election * * \return TRUE if local node has won, FALSE otherwise * \note If all known peers have sent no-vote messages, but the election owner * does not call this function, the election will not be won (and the * callback will not be called) until the election times out. * \note This should be called when election_count_vote() returns * \c election_in_progress. */ bool election_check(pcmk_cluster_t *cluster) { int voted_size = 0; int num_members = 0; CRM_CHECK((cluster != NULL) && (cluster->priv->election != NULL), return false); if (cluster->priv->election->voted == NULL) { crm_trace("Election check requested, but no votes received yet"); return FALSE; } voted_size = g_hash_table_size(cluster->priv->election->voted); num_members = pcmk__cluster_num_active_nodes(); /* in the case of #voted > #members, it is better to * wait for the timeout and give the cluster time to * stabilize */ if (voted_size >= num_members) { /* we won and everyone has voted */ election_timeout_stop(cluster); if (voted_size > num_members) { GHashTableIter gIter; const pcmk__node_status_t *node = NULL; char *key = NULL; crm_warn("Received too many votes in election"); g_hash_table_iter_init(&gIter, pcmk__peer_cache); while (g_hash_table_iter_next(&gIter, NULL, (gpointer *) & node)) { if (pcmk__cluster_is_node_active(node)) { crm_warn("* expected vote: %s", node->name); } } g_hash_table_iter_init(&gIter, cluster->priv->election->voted); while (g_hash_table_iter_next(&gIter, (gpointer *) & key, NULL)) { crm_warn("* actual vote: %s", key); } } crm_info("Election won by local node"); election_complete(cluster); return TRUE; } else { crm_debug("Election still waiting on %d of %d vote%s", num_members - voted_size, num_members, pcmk__plural_s(num_members)); } return FALSE; } #define LOSS_DAMPEN 2 /* in seconds */ struct vote { const char *op; const char *from; const char *version; const char *election_owner; int election_id; struct timeval age; }; /*! * \internal * \brief Unpack an election message * * \param[in] message Election message XML * \param[out] vote Parsed fields from message * * \return TRUE if election message and election are valid, FALSE otherwise * \note The parsed struct's pointer members are valid only for the lifetime of * the message argument. */ static bool parse_election_message(const xmlNode *message, struct vote *vote) { CRM_CHECK(message && vote, return FALSE); vote->election_id = -1; vote->age.tv_sec = -1; vote->age.tv_usec = -1; vote->op = pcmk__xe_get(message, PCMK__XA_CRM_TASK); vote->from = pcmk__xe_get(message, PCMK__XA_SRC); vote->version = pcmk__xe_get(message, PCMK_XA_VERSION); vote->election_owner = pcmk__xe_get(message, PCMK__XA_ELECTION_OWNER); pcmk__xe_get_int(message, PCMK__XA_ELECTION_ID, &(vote->election_id)); if ((vote->op == NULL) || (vote->from == NULL) || (vote->version == NULL) || (vote->election_owner == NULL) || (vote->election_id < 0)) { crm_warn("Invalid %s message from %s", pcmk__s(vote->op, "election"), pcmk__s(vote->from, "unspecified node")); crm_log_xml_trace(message, "bad-vote"); return FALSE; } // Op-specific validation if (pcmk__str_eq(vote->op, CRM_OP_VOTE, pcmk__str_none)) { /* Only vote ops have uptime. Warning: PCMK__XA_ELECTION_AGE_NANO_SEC value is in microseconds. */ if ((pcmk__xe_get_timeval(message, PCMK__XA_ELECTION_AGE_SEC, PCMK__XA_ELECTION_AGE_NANO_SEC, &(vote->age)) != pcmk_rc_ok) || (vote->age.tv_sec < 0) || (vote->age.tv_usec < 0)) { crm_warn("Cannot count election %s from %s because uptime is " "missing or invalid", vote->op, vote->from); return FALSE; } } else if (!pcmk__str_eq(vote->op, CRM_OP_NOVOTE, pcmk__str_none)) { crm_info("Cannot process election message from %s " "because %s is not a known election op", vote->from, vote->op); return FALSE; } /* If the membership cache is NULL, we REALLY shouldn't be voting -- * the question is how we managed to get here. */ if (pcmk__peer_cache == NULL) { crm_info("Cannot count election %s from %s " "because no peer information available", vote->op, vote->from); return FALSE; } return TRUE; } static void record_vote(pcmk_cluster_t *cluster, struct vote *vote) { pcmk__assert((vote->from != NULL) && (vote->op != NULL)); if (cluster->priv->election->voted == NULL) { cluster->priv->election->voted = pcmk__strkey_table(free, free); } pcmk__insert_dup(cluster->priv->election->voted, vote->from, vote->op); } static void send_no_vote(pcmk_cluster_t *cluster, pcmk__node_status_t *peer, struct vote *vote) { const char *message_type = NULL; xmlNode *novote = NULL; message_type = pcmk__server_message_type(cluster->priv->server); novote = pcmk__new_request(cluster->priv->server, message_type, vote->from, message_type, CRM_OP_NOVOTE, NULL); crm_xml_add(novote, PCMK__XA_ELECTION_OWNER, vote->election_owner); crm_xml_add_int(novote, PCMK__XA_ELECTION_ID, vote->election_id); pcmk__cluster_send_message(peer, cluster->priv->server, novote); pcmk__xml_free(novote); } /*! * \internal * \brief Process an election message (vote or no-vote) from a peer * * \param[in,out] cluster Cluster with election * \param[in] message Election message XML from peer * \param[in] can_win Whether local node is eligible to win * * \return Election state after new vote is considered * \note If the peer message is a vote, and we prefer the peer to win, this will * send a no-vote reply to the peer. * \note The situations "we lost to this vote" from "this is a late no-vote * after we've already lost" both return election_lost. If a caller needs * to distinguish them, it should save the current state before calling * this function, and then compare the result. */ enum election_result election_count_vote(pcmk_cluster_t *cluster, const xmlNode *message, bool can_win) { int log_level = LOG_INFO; gboolean done = FALSE; gboolean we_lose = FALSE; const char *reason = NULL; bool we_are_owner = FALSE; pcmk__node_status_t *our_node = NULL; pcmk__node_status_t *your_node = NULL; time_t tm_now = time(NULL); struct vote vote; CRM_CHECK((cluster != NULL) && (cluster->priv->election != NULL) && (message != NULL) && (cluster->priv->node_name != NULL), return election_error); if (!parse_election_message(message, &vote)) { return election_error; } your_node = pcmk__get_node(0, vote.from, NULL, pcmk__node_search_cluster_member); our_node = pcmk__get_node(0, cluster->priv->node_name, NULL, pcmk__node_search_cluster_member); we_are_owner = (our_node != NULL) && pcmk__str_eq(pcmk__cluster_get_xml_id(our_node), vote.election_owner, pcmk__str_none); if (!can_win) { reason = "Not eligible"; we_lose = TRUE; } else if (!pcmk__cluster_is_node_active(our_node)) { reason = "We are not part of the cluster"; log_level = LOG_ERR; we_lose = TRUE; } else if (we_are_owner && (vote.election_id != cluster->priv->election->count)) { log_level = LOG_TRACE; reason = "Superseded"; done = TRUE; } else if (!pcmk__cluster_is_node_active(your_node)) { /* Possibly we cached the message in the FSA queue at a point that it wasn't */ reason = "Peer is not part of our cluster"; log_level = LOG_WARNING; done = TRUE; } else if (pcmk__str_eq(vote.op, CRM_OP_NOVOTE, pcmk__str_none) || pcmk__str_eq(vote.from, cluster->priv->node_name, pcmk__str_casei)) { /* Receiving our own broadcast vote, or a no-vote from peer, is a vote * for us to win */ if (!we_are_owner) { crm_warn("Cannot count election round %d %s from %s " "because we did not start election (node ID %s did)", vote.election_id, vote.op, vote.from, vote.election_owner); return election_error; } if (cluster->priv->election->state != election_in_progress) { // Should only happen if we already lost crm_debug("Not counting election round %d %s from %s " "because no election in progress", vote.election_id, vote.op, vote.from); return cluster->priv->election->state; } record_vote(cluster, &vote); reason = "Recorded"; done = TRUE; } else { // A peer vote requires a comparison to determine which node is better int age_result = compare_age(vote.age); int version_result = compare_version(vote.version, CRM_FEATURE_SET); if (version_result < 0) { reason = "Version"; we_lose = TRUE; } else if (version_result > 0) { reason = "Version"; } else if (age_result < 0) { reason = "Uptime"; we_lose = TRUE; } else if (age_result > 0) { reason = "Uptime"; } else if (strcasecmp(cluster->priv->node_name, vote.from) > 0) { reason = "Host name"; we_lose = TRUE; } else { reason = "Host name"; } } if (cluster->priv->election->expires < tm_now) { cluster->priv->election->election_wins = 0; cluster->priv->election->expires = tm_now + STORM_INTERVAL; } else if (done == FALSE && we_lose == FALSE) { int peers = 1 + g_hash_table_size(pcmk__peer_cache); /* If every node has to vote down every other node, thats N*(N-1) total elections * Allow some leeway before _really_ complaining */ cluster->priv->election->election_wins++; if (cluster->priv->election->election_wins > (peers * peers)) { crm_warn("Election storm detected: %d wins in %d seconds", cluster->priv->election->election_wins, STORM_INTERVAL); cluster->priv->election->election_wins = 0; cluster->priv->election->expires = tm_now + STORM_INTERVAL; if (!(cluster->priv->election->wrote_blackbox)) { /* It's questionable whether a black box (from every node in the * cluster) would be truly helpful in diagnosing an election * storm. It's also highly doubtful a production environment * would get multiple election storms from distinct causes, so * saving one blackbox per process lifetime should be * sufficient. Alternatives would be to save a timestamp of the * last blackbox write instead of a boolean, and write a new one * if some amount of time has passed; or to save a storm count, * write a blackbox on every Nth occurrence. */ crm_write_blackbox(0, NULL); cluster->priv->election->wrote_blackbox = true; } } } if (done) { do_crm_log(log_level + 1, "Processed election round %u %s (current round %d) " "from %s (%s)", vote.election_id, vote.op, cluster->priv->election->count, vote.from, reason); return cluster->priv->election->state; } else if (we_lose == FALSE) { /* We track the time of the last election loss to implement an election * dampening period, reducing the likelihood of an election storm. If * this node has lost within the dampening period, don't start a new * election, even if we win against a peer's vote -- the peer we lost to * should win again. * * @TODO This has a problem case: if an election winner immediately * leaves the cluster, and a new election is immediately called, all * nodes could lose, with no new winner elected. The ideal solution * would be to tie the election structure with the peer caches, which * would allow us to clear the dampening when the previous winner * leaves (and would allow other improvements as well). */ if ((cluster->priv->election->last_election_loss == 0) || ((tm_now - cluster->priv->election->last_election_loss) > (time_t) LOSS_DAMPEN)) { do_crm_log(log_level, "Election round %d (started by node ID %s) pass: " "%s from %s (%s)", vote.election_id, vote.election_owner, vote.op, vote.from, reason); cluster->priv->election->last_election_loss = 0; election_timeout_stop(cluster); /* Start a new election by voting down this, and other, peers */ cluster->priv->election->state = election_start; return cluster->priv->election->state; } else { char *loss_time = NULL; loss_time = ctime(&(cluster->priv->election->last_election_loss)); if (loss_time) { // Show only HH:MM:SS loss_time += 11; loss_time[8] = '\0'; } crm_info("Ignoring election round %d (started by node ID %s) pass " "vs %s because we lost less than %ds ago at %s", vote.election_id, vote.election_owner, vote.from, LOSS_DAMPEN, (loss_time? loss_time : "unknown")); } } cluster->priv->election->last_election_loss = tm_now; do_crm_log(log_level, "Election round %d (started by node ID %s) lost: " "%s from %s (%s)", vote.election_id, vote.election_owner, vote.op, vote.from, reason); election_reset(cluster); send_no_vote(cluster, your_node, &vote); cluster->priv->election->state = election_lost; return cluster->priv->election->state; } /*! * \internal * \brief Reset any election dampening currently in effect * * \param[in,out] cluster Cluster with election */ void election_clear_dampening(pcmk_cluster_t *cluster) { if ((cluster != NULL) && (cluster->priv->election != NULL)) { cluster->priv->election->last_election_loss = 0; } } diff --git a/lib/common/acl.c b/lib/common/acl.c index 9e2205980c..4d430127e3 100644 --- a/lib/common/acl.c +++ b/lib/common/acl.c @@ -1,930 +1,936 @@ /* * Copyright 2004-2025 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 // xmlNode, etc. #include // xmlChar #include // xmlXPathObject, etc. #include #include #include #include "crmcommon_private.h" typedef struct xml_acl_s { enum pcmk__xml_flags mode; gchar *xpath; } xml_acl_t; static void free_acl(void *data) { if (data) { xml_acl_t *acl = data; g_free(acl->xpath); free(acl); } } void pcmk__free_acls(GList *acls) { g_list_free_full(acls, free_acl); } static GList * create_acl(const xmlNode *xml, GList *acls, enum pcmk__xml_flags mode) { xml_acl_t *acl = NULL; const char *tag = pcmk__xe_get(xml, PCMK_XA_OBJECT_TYPE); const char *ref = pcmk__xe_get(xml, PCMK_XA_REFERENCE); const char *xpath = pcmk__xe_get(xml, PCMK_XA_XPATH); const char *attr = pcmk__xe_get(xml, PCMK_XA_ATTRIBUTE); if ((tag == NULL) && (ref == NULL) && (xpath == NULL)) { // Schema should prevent this, but to be safe ... crm_trace("Ignoring ACL <%s> element without selection criteria", xml->name); return NULL; } acl = pcmk__assert_alloc(1, sizeof (xml_acl_t)); acl->mode = mode; if (xpath) { acl->xpath = g_strdup(xpath); crm_trace("Unpacked ACL <%s> element using xpath: %s", xml->name, acl->xpath); } else { GString *buf = g_string_sized_new(128); if ((ref != NULL) && (attr != NULL)) { // NOTE: schema currently does not allow this pcmk__g_strcat(buf, "//", pcmk__s(tag, "*"), "[@" PCMK_XA_ID "='", ref, "' and @", attr, "]", NULL); } else if (ref != NULL) { pcmk__g_strcat(buf, "//", pcmk__s(tag, "*"), "[@" PCMK_XA_ID "='", ref, "']", NULL); } else if (attr != NULL) { pcmk__g_strcat(buf, "//", pcmk__s(tag, "*"), "[@", attr, "]", NULL); } else { pcmk__g_strcat(buf, "//", pcmk__s(tag, "*"), NULL); } acl->xpath = buf->str; g_string_free(buf, FALSE); crm_trace("Unpacked ACL <%s> element as xpath: %s", xml->name, acl->xpath); } return g_list_append(acls, acl); } /*! * \internal * \brief Unpack a user, group, or role subtree of the ACLs section * * \param[in] acl_top XML of entire ACLs section * \param[in] acl_entry XML of ACL element being unpacked * \param[in,out] acls List of ACLs unpacked so far * * \return New head of (possibly modified) acls * * \note This function is recursive */ static GList * parse_acl_entry(const xmlNode *acl_top, const xmlNode *acl_entry, GList *acls) { for (const xmlNode *child = pcmk__xe_first_child(acl_entry, NULL, NULL, NULL); child != NULL; child = pcmk__xe_next(child, NULL)) { if (pcmk__xe_is(child, PCMK_XE_ACL_PERMISSION)) { const char *kind = pcmk__xe_get(child, PCMK_XA_KIND); pcmk__assert(kind != NULL); crm_trace("Unpacking <" PCMK_XE_ACL_PERMISSION "> element of " "kind '%s'", kind); if (pcmk__str_eq(kind, PCMK_VALUE_READ, pcmk__str_none)) { acls = create_acl(child, acls, pcmk__xf_acl_read); } else if (pcmk__str_eq(kind, PCMK_VALUE_WRITE, pcmk__str_none)) { acls = create_acl(child, acls, pcmk__xf_acl_write); } else if (pcmk__str_eq(kind, PCMK_VALUE_DENY, pcmk__str_none)) { acls = create_acl(child, acls, pcmk__xf_acl_deny); } else { crm_warn("Ignoring unknown ACL kind '%s'", kind); } } else if (pcmk__xe_is(child, PCMK_XE_ROLE)) { const char *ref_role = pcmk__xe_get(child, PCMK_XA_ID); crm_trace("Unpacking <" PCMK_XE_ROLE "> element"); if (ref_role == NULL) { continue; } for (xmlNode *role = pcmk__xe_first_child(acl_top, NULL, NULL, NULL); role != NULL; role = pcmk__xe_next(role, NULL)) { const char *role_id = NULL; if (!pcmk__xe_is(role, PCMK_XE_ACL_ROLE)) { continue; } role_id = pcmk__xe_get(role, PCMK_XA_ID); if (pcmk__str_eq(ref_role, role_id, pcmk__str_none)) { crm_trace("Unpacking referenced role '%s' in <%s> element", role_id, acl_entry->name); acls = parse_acl_entry(acl_top, role, acls); break; } } } } return acls; } /* */ static const char * acl_to_text(enum pcmk__xml_flags flags) { if (pcmk_is_set(flags, pcmk__xf_acl_deny)) { return "deny"; } else if (pcmk_any_flags_set(flags, pcmk__xf_acl_write|pcmk__xf_acl_create)) { return "read/write"; } else if (pcmk_is_set(flags, pcmk__xf_acl_read)) { return "read"; } return "none"; } void pcmk__apply_acl(xmlNode *xml) { GList *aIter = NULL; xml_doc_private_t *docpriv = NULL; xml_node_private_t *nodepriv = NULL; xmlXPathObject *xpathObj = NULL; pcmk__assert(xml != NULL); docpriv = xml->doc->_private; if (!pcmk__xml_doc_all_flags_set(xml->doc, pcmk__xf_acl_enabled)) { crm_trace("Skipping ACLs for user '%s' because not enabled for this XML", pcmk__s(docpriv->acl_user, "(unknown)")); return; } for (aIter = docpriv->acls; aIter != NULL; aIter = aIter->next) { int max = 0, lpc = 0; xml_acl_t *acl = aIter->data; xpathObj = pcmk__xpath_search(xml->doc, acl->xpath); max = pcmk__xpath_num_results(xpathObj); for (lpc = 0; lpc < max; lpc++) { xmlNode *match = pcmk__xpath_result(xpathObj, lpc); if (match == NULL) { continue; } /* @COMPAT If the ACL's XPath matches a node that is neither an * element nor a document, we apply the ACL to the parent element * rather than to the matched node. For example, if the XPath * matches a "score" attribute, then it applies to every element * that contains a "score" attribute. That is, the XPath expression * "//@score" matches all attributes named "score", but we apply the * ACL to all elements containing such an attribute. * * This behavior is incorrect from an XPath standpoint and is thus * confusing and counterintuitive. The correct way to match all * elements containing a "score" attribute is to use an XPath * predicate: "// *[@score]". (Space inserted after slashes so that * GCC doesn't throw an error about nested comments.) * * Additionally, if an XPath expression matches the entire document * (for example, "/"), then the ACL applies to the document's root * element if it exists. * * These behaviors should be changed so that the ACL applies to the * nodes matched by the XPath expression, or so that it doesn't * apply at all if applying an ACL to an attribute doesn't make * sense. * * Unfortunately, we document in Pacemaker Explained that matching * attributes is a valid way to match elements: "Attributes may be * specified in the XPath to select particular elements, but the * permissions apply to the entire element." * * So we have to keep this behavior at least until a compatibility * break. Even then, it's not feasible in the general case to * transform such XPath expressions using XSLT. */ match = pcmk__xpath_match_element(match); if (match == NULL) { continue; } nodepriv = match->_private; pcmk__set_xml_flags(nodepriv, acl->mode); // Build a GString only if tracing is enabled pcmk__if_tracing( { GString *path = pcmk__element_xpath(match); crm_trace("Applying %s ACL to %s matched by %s", acl_to_text(acl->mode), path->str, acl->xpath); g_string_free(path, TRUE); }, {} ); } crm_trace("Applied %s ACL %s (%d match%s)", acl_to_text(acl->mode), acl->xpath, max, ((max == 1)? "" : "es")); xmlXPathFreeObject(xpathObj); } } /*! * \internal * \brief Unpack ACLs for a given user into the * metadata of the target XML tree * * Taking the description of ACLs from the source XML tree and * marking up the target XML tree with access information for the * given user by tacking it onto the relevant nodes * * \param[in] source XML with ACL definitions * \param[in,out] target XML that ACLs will be applied to * \param[in] user Username whose ACLs need to be unpacked */ void pcmk__unpack_acl(xmlNode *source, xmlNode *target, const char *user) { xml_doc_private_t *docpriv = NULL; if ((target == NULL) || (target->doc == NULL) || (target->doc->_private == NULL)) { return; } docpriv = target->doc->_private; if (!pcmk_acl_required(user)) { crm_trace("Not unpacking ACLs because not required for user '%s'", user); } else if (docpriv->acls == NULL) { xmlNode *acls = pcmk__xpath_find_one(source->doc, "//" PCMK_XE_ACLS, LOG_NEVER); pcmk__str_update(&(docpriv->acl_user), user); if (acls) { xmlNode *child = NULL; for (child = pcmk__xe_first_child(acls, NULL, NULL, NULL); child != NULL; child = pcmk__xe_next(child, NULL)) { if (pcmk__xe_is(child, PCMK_XE_ACL_TARGET)) { const char *id = pcmk__xe_get(child, PCMK_XA_NAME); if (id == NULL) { id = pcmk__xe_get(child, PCMK_XA_ID); } if (id && strcmp(id, user) == 0) { crm_debug("Unpacking ACLs for user '%s'", id); docpriv->acls = parse_acl_entry(acls, child, docpriv->acls); } } else if (pcmk__xe_is(child, PCMK_XE_ACL_GROUP)) { const char *id = pcmk__xe_get(child, PCMK_XA_NAME); if (id == NULL) { id = pcmk__xe_get(child, PCMK_XA_ID); } if (id && pcmk__is_user_in_group(user,id)) { crm_debug("Unpacking ACLs for group '%s'", id); docpriv->acls = parse_acl_entry(acls, child, docpriv->acls); } } } } } } /*! * \internal * \brief Copy source to target and set xf_acl_enabled flag in target * * \param[in] acl_source XML with ACL definitions * \param[in,out] target XML that ACLs will be applied to * \param[in] user Username whose ACLs need to be set */ void pcmk__enable_acl(xmlNode *acl_source, xmlNode *target, const char *user) { if (target == NULL) { return; } pcmk__unpack_acl(acl_source, target, user); pcmk__xml_doc_set_flags(target->doc, pcmk__xf_acl_enabled); pcmk__apply_acl(target); } static inline bool test_acl_mode(enum pcmk__xml_flags allowed, enum pcmk__xml_flags requested) { if (pcmk_is_set(allowed, pcmk__xf_acl_deny)) { return false; } else if (pcmk_all_flags_set(allowed, requested)) { return true; } else if (pcmk_is_set(requested, pcmk__xf_acl_read) && pcmk_is_set(allowed, pcmk__xf_acl_write)) { return true; } else if (pcmk_is_set(requested, pcmk__xf_acl_create) && pcmk_any_flags_set(allowed, pcmk__xf_acl_write|pcmk__xf_created)) { return true; } return false; } /*! * \internal * \brief Rid XML tree of all unreadable nodes and node properties * * \param[in,out] xml Root XML node to be purged of attributes * * \return true if this node or any of its children are readable * if false is returned, xml will be freed * * \note This function is recursive */ static bool purge_xml_attributes(xmlNode *xml) { xmlNode *child = NULL; xmlAttr *xIter = NULL; bool readable_children = false; xml_node_private_t *nodepriv = xml->_private; if (test_acl_mode(nodepriv->flags, pcmk__xf_acl_read)) { crm_trace("%s[@" PCMK_XA_ID "=%s] is readable", xml->name, pcmk__xe_id(xml)); return true; } xIter = xml->properties; while (xIter != NULL) { xmlAttr *tmp = xIter; const char *prop_name = (const char *)xIter->name; xIter = xIter->next; if (strcmp(prop_name, PCMK_XA_ID) == 0) { continue; } pcmk__xa_remove(tmp, true); } child = pcmk__xml_first_child(xml); while ( child != NULL ) { xmlNode *tmp = child; child = pcmk__xml_next(child); readable_children |= purge_xml_attributes(tmp); } if (!readable_children) { // Nothing readable under here, so purge completely pcmk__xml_free(xml); } return readable_children; } /*! * \brief Copy ACL-allowed portions of specified XML * * \param[in] user Username whose ACLs should be used * \param[in] acl_source XML containing ACLs * \param[in] xml XML to be copied * \param[out] result Copy of XML portions readable via ACLs * * \return true if xml exists and ACLs are required for user, false otherwise * \note If this returns true, caller should use \p result rather than \p xml */ bool xml_acl_filtered_copy(const char *user, xmlNode *acl_source, xmlNode *xml, xmlNode **result) { GList *aIter = NULL; xmlNode *target = NULL; xml_doc_private_t *docpriv = NULL; *result = NULL; if ((xml == NULL) || !pcmk_acl_required(user)) { crm_trace("Not filtering XML because ACLs not required for user '%s'", user); return false; } crm_trace("Filtering XML copy using user '%s' ACLs", user); target = pcmk__xml_copy(NULL, xml); if (target == NULL) { return true; } pcmk__enable_acl(acl_source, target, user); docpriv = target->doc->_private; for(aIter = docpriv->acls; aIter != NULL && target; aIter = aIter->next) { int max = 0; xml_acl_t *acl = aIter->data; if (acl->mode != pcmk__xf_acl_deny) { /* Nothing to do */ } else if (acl->xpath) { int lpc = 0; xmlXPathObject *xpathObj = pcmk__xpath_search(target->doc, acl->xpath); max = pcmk__xpath_num_results(xpathObj); for(lpc = 0; lpc < max; lpc++) { xmlNode *match = pcmk__xpath_result(xpathObj, lpc); if (match == NULL) { continue; } // @COMPAT See COMPAT comment in pcmk__apply_acl() match = pcmk__xpath_match_element(match); if (match == NULL) { continue; } if (!purge_xml_attributes(match) && (match == target)) { crm_trace("ACLs deny user '%s' access to entire XML document", user); xmlXPathFreeObject(xpathObj); return true; } } crm_trace("ACLs deny user '%s' access to %s (%d %s)", user, acl->xpath, max, pcmk__plural_alt(max, "match", "matches")); xmlXPathFreeObject(xpathObj); } } if (!purge_xml_attributes(target)) { crm_trace("ACLs deny user '%s' access to entire XML document", user); return true; } if (docpriv->acls) { g_list_free_full(docpriv->acls, free_acl); docpriv->acls = NULL; } else { crm_trace("User '%s' without ACLs denied access to entire XML document", user); pcmk__xml_free(target); target = NULL; } if (target) { *result = target; } return true; } /*! * \internal * \brief Check whether creation of an XML element is implicitly allowed * * Check whether XML is a "scaffolding" element whose creation is implicitly * allowed regardless of ACLs (that is, it is not in the ACL section and has * no attributes other than \c PCMK_XA_ID). * * \param[in] xml XML element to check * * \return true if XML element is implicitly allowed, false otherwise */ static bool implicitly_allowed(const xmlNode *xml) { GString *path = NULL; for (xmlAttr *prop = xml->properties; prop != NULL; prop = prop->next) { if (strcmp((const char *) prop->name, PCMK_XA_ID) != 0) { return false; } } path = pcmk__element_xpath(xml); pcmk__assert(path != NULL); if (strstr((const char *) path->str, "/" PCMK_XE_ACLS "/") != NULL) { g_string_free(path, TRUE); return false; } g_string_free(path, TRUE); return true; } #define display_id(xml) pcmk__s(pcmk__xe_id(xml), "") /*! * \internal * \brief Drop XML nodes created in violation of ACLs * * Given an XML element, free all of its descendant nodes created in violation * of ACLs, with the exception of allowing "scaffolding" elements (i.e. those * that aren't in the ACL section and don't have any attributes other than * \c PCMK_XA_ID). * * \param[in,out] xml XML to check * \param[in] check_top Whether to apply checks to argument itself * (if true, xml might get freed) * * \note This function is recursive */ void pcmk__apply_creation_acl(xmlNode *xml, bool check_top) { xml_node_private_t *nodepriv = xml->_private; if (pcmk_is_set(nodepriv->flags, pcmk__xf_created)) { if (implicitly_allowed(xml)) { crm_trace("Creation of <%s> scaffolding with " PCMK_XA_ID "=\"%s\"" " is implicitly allowed", xml->name, display_id(xml)); } else if (pcmk__check_acl(xml, NULL, pcmk__xf_acl_write)) { crm_trace("ACLs allow creation of <%s> with " PCMK_XA_ID "=\"%s\"", xml->name, display_id(xml)); } else if (check_top) { /* is_root=true should be impossible with check_top=true, but check * for sanity */ bool is_root = (xmlDocGetRootElement(xml->doc) == xml); xml_doc_private_t *docpriv = xml->doc->_private; crm_trace("ACLs disallow creation of %s<%s> with " PCMK_XA_ID "=\"%s\"", (is_root? "root element " : ""), xml->name, display_id(xml)); // pcmk__xml_free() checks ACLs if enabled, which would fail pcmk__clear_xml_flags(docpriv, pcmk__xf_acl_enabled); pcmk__xml_free(xml); if (!is_root) { // If root, the document was freed. Otherwise re-enable ACLs. pcmk__set_xml_flags(docpriv, pcmk__xf_acl_enabled); } return; } else { crm_notice("ACLs would disallow creation of %s<%s> with " PCMK_XA_ID "=\"%s\"", ((xml == xmlDocGetRootElement(xml->doc))? "root element " : ""), xml->name, display_id(xml)); } } for (xmlNode *cIter = pcmk__xml_first_child(xml); cIter != NULL; ) { xmlNode *child = cIter; cIter = pcmk__xml_next(cIter); /* In case it is free'd */ pcmk__apply_creation_acl(child, true); } } /*! * \brief Check whether or not an XML node is ACL-denied * * \param[in] xml node to check * * \return true if XML node exists and is ACL-denied, false otherwise */ bool xml_acl_denied(const xmlNode *xml) { if (xml && xml->doc && xml->doc->_private){ xml_doc_private_t *docpriv = xml->doc->_private; return pcmk_is_set(docpriv->flags, pcmk__xf_acl_denied); } return false; } void xml_acl_disable(xmlNode *xml) { if ((xml != NULL) && pcmk__xml_doc_all_flags_set(xml->doc, pcmk__xf_acl_enabled)) { xml_doc_private_t *docpriv = xml->doc->_private; /* Catch anything that was created but shouldn't have been */ pcmk__apply_acl(xml); pcmk__apply_creation_acl(xml, false); pcmk__clear_xml_flags(docpriv, pcmk__xf_acl_enabled); } } /*! * \internal * \brief Deny access to an XML tree's document based on ACLs * * \param[in,out] xml XML tree * \param[in] attr_name Name of attribute being accessed in \p xml (for * logging only) * \param[in] prefix Prefix describing ACL that denied access (for * logging only) * \param[in] user User accessing \p xml (for logging only) * \param[in] mode Access mode (for logging only) */ #define check_acl_deny(xml, attr_name, prefix, user, mode) do { \ xmlNode *tree = xml; \ \ pcmk__xml_doc_set_flags(tree->doc, pcmk__xf_acl_denied); \ pcmk__if_tracing( \ { \ GString *xpath = pcmk__element_xpath(tree); \ \ if ((attr_name) != NULL) { \ pcmk__g_strcat(xpath, "[@", attr_name, "]", NULL); \ } \ qb_log_from_external_source(__func__, __FILE__, \ "%sACL denies user '%s' %s " \ "access to %s", \ LOG_TRACE, __LINE__, 0 , \ prefix, user, \ acl_to_text(mode), xpath->str); \ g_string_free(xpath, TRUE); \ }, \ {} \ ); \ } while (false); bool pcmk__check_acl(xmlNode *xml, const char *attr_name, enum pcmk__xml_flags mode) { xml_doc_private_t *docpriv = NULL; pcmk__assert((xml != NULL) && (xml->doc->_private != NULL)); if (!pcmk__xml_doc_all_flags_set(xml->doc, pcmk__xf_tracking|pcmk__xf_acl_enabled)) { return true; } docpriv = xml->doc->_private; if (docpriv->acls == NULL) { check_acl_deny(xml, attr_name, "Lack of ", docpriv->acl_user, mode); return false; } /* Walk the tree upwards looking for xml_acl_* flags * - Creating an attribute requires write permissions for the node * - Creating a child requires write permissions for the parent */ if (attr_name != NULL) { xmlAttr *attr = xmlHasProp(xml, (const xmlChar *) attr_name); if ((attr != NULL) && (mode == pcmk__xf_acl_create)) { mode = pcmk__xf_acl_write; } } for (const xmlNode *parent = xml; (parent != NULL) && (parent->_private != NULL); parent = parent->parent) { const xml_node_private_t *nodepriv = parent->_private; if (test_acl_mode(nodepriv->flags, mode)) { return true; } if (pcmk_is_set(nodepriv->flags, pcmk__xf_acl_deny)) { const char *pfx = (parent != xml)? "Parent " : ""; check_acl_deny(xml, attr_name, pfx, docpriv->acl_user, mode); return false; } } check_acl_deny(xml, attr_name, "Default ", docpriv->acl_user, mode); return false; } /*! * \brief Check whether ACLs are required for a given user * * \param[in] User name to check * * \return true if the user requires ACLs, false otherwise */ bool pcmk_acl_required(const char *user) { if (pcmk__str_empty(user)) { crm_trace("ACLs not required because no user set"); return false; } else if (!strcmp(user, CRM_DAEMON_USER) || !strcmp(user, "root")) { crm_trace("ACLs not required for privileged user %s", user); return false; } crm_trace("ACLs required for %s", user); return true; } char * pcmk__uid2username(uid_t uid) { - struct passwd *pwent = getpwuid(uid); + struct passwd *pwent = NULL; + + errno = 0; + pwent = getpwuid(uid); if (pwent == NULL) { - crm_perror(LOG_INFO, "Cannot get user details for user ID %d", uid); + crm_err("Cannot get name from password database for user ID %lld: %s", + (long long) uid, + ((errno != 0)? strerror(errno) : "No matching entry found")); return NULL; } + return pcmk__str_copy(pwent->pw_name); } /*! * \internal * \brief Set the ACL user field properly on an XML request * * Multiple user names are potentially involved in an XML request: the effective * user of the current process; the user name known from an IPC client * connection; and the user name obtained from the request itself, whether by * the current standard XML attribute name or an older legacy attribute name. * This function chooses the appropriate one that should be used for ACLs, sets * it in the request (using the standard attribute name, and the legacy name if * given), and returns it. * * \param[in,out] request XML request to update * \param[in] field Alternate name for ACL user name XML attribute * \param[in] peer_user User name as known from IPC connection * * \return ACL user name actually used */ const char * pcmk__update_acl_user(xmlNode *request, const char *field, const char *peer_user) { static const char *effective_user = NULL; const char *requested_user = NULL; const char *user = NULL; if (effective_user == NULL) { effective_user = pcmk__uid2username(geteuid()); if (effective_user == NULL) { effective_user = pcmk__str_copy("#unprivileged"); crm_err("Unable to determine effective user, assuming unprivileged for ACLs"); } } requested_user = pcmk__xe_get(request, PCMK__XA_ACL_TARGET); if (requested_user == NULL) { /* Currently, different XML attribute names are used for the ACL user in * different contexts (PCMK__XA_ATTR_USER, PCMK__XA_CIB_USER, etc.). * The caller may specify that name as the field argument. * * @TODO Standardize on PCMK__XA_ACL_TARGET and eventually drop the * others once rolling upgrades from versions older than that are no * longer supported. */ requested_user = pcmk__xe_get(request, field); } if (!pcmk__is_privileged(effective_user)) { /* We're not running as a privileged user, set or overwrite any existing * value for PCMK__XA_ACL_TARGET */ user = effective_user; } else if (peer_user == NULL && requested_user == NULL) { /* No user known or requested, use 'effective_user' and make sure one is * set for the request */ user = effective_user; } else if (peer_user == NULL) { /* No user known, trusting 'requested_user' */ user = requested_user; } else if (!pcmk__is_privileged(peer_user)) { /* The peer is not a privileged user, set or overwrite any existing * value for PCMK__XA_ACL_TARGET */ user = peer_user; } else if (requested_user == NULL) { /* Even if we're privileged, make sure there is always a value set */ user = peer_user; } else { /* Legal delegation to 'requested_user' */ user = requested_user; } // This requires pointer comparison, not string comparison if (user != pcmk__xe_get(request, PCMK__XA_ACL_TARGET)) { crm_xml_add(request, PCMK__XA_ACL_TARGET, user); } if ((field != NULL) && (user != pcmk__xe_get(request, field))) { crm_xml_add(request, field, user); } return requested_user; } // Deprecated functions kept only for backward API compatibility // LCOV_EXCL_START #include #include bool xml_acl_enabled(const xmlNode *xml) { if (xml && xml->doc && xml->doc->_private){ xml_doc_private_t *docpriv = xml->doc->_private; return pcmk_is_set(docpriv->flags, pcmk__xf_acl_enabled); } return false; } // LCOV_EXCL_STOP // End deprecated API diff --git a/lib/common/cmdline.c b/lib/common/cmdline.c index 5b6b1654ed..6ecc9cb1e4 100644 --- a/lib/common/cmdline.c +++ b/lib/common/cmdline.c @@ -1,344 +1,317 @@ /* * Copyright 2019-2025 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 static gboolean bump_verbosity(const gchar *option_name, const gchar *optarg, gpointer data, GError **error) { pcmk__common_args_t *common_args = (pcmk__common_args_t *) data; common_args->verbosity++; return TRUE; } pcmk__common_args_t * pcmk__new_common_args(const char *summary) { pcmk__common_args_t *args = NULL; args = calloc(1, sizeof(pcmk__common_args_t)); if (args == NULL) { crm_exit(CRM_EX_OSERR); } // cppcheck-suppress nullPointerOutOfMemory args->summary = strdup(summary); // cppcheck-suppress nullPointerOutOfMemory if (args->summary == NULL) { free(args); args = NULL; crm_exit(CRM_EX_OSERR); } return args; } static void free_common_args(gpointer data) { pcmk__common_args_t *common_args = (pcmk__common_args_t *) data; free(common_args->summary); free(common_args->output_ty); free(common_args->output_dest); if (common_args->output_as_descr != NULL) { free(common_args->output_as_descr); } free(common_args); } GOptionContext * pcmk__build_arg_context(pcmk__common_args_t *common_args, const char *fmts, GOptionGroup **output_group, const char *param_string) { GOptionContext *context; GOptionGroup *main_group; GOptionEntry main_entries[3] = { { "version", '$', 0, G_OPTION_ARG_NONE, &(common_args->version), N_("Display software version and exit"), NULL }, { "verbose", 'V', G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, bump_verbosity, N_("Increase debug output (may be specified multiple times)"), NULL }, { NULL } }; main_group = g_option_group_new(NULL, "Application Options:", NULL, common_args, free_common_args); g_option_group_add_entries(main_group, main_entries); context = g_option_context_new(param_string); g_option_context_set_summary(context, common_args->summary); g_option_context_set_description(context, "Report bugs to " PCMK__BUG_URL "\n"); g_option_context_set_main_group(context, main_group); if (fmts != NULL) { GOptionEntry output_entries[3] = { { "output-as", 0, 0, G_OPTION_ARG_STRING, &(common_args->output_ty), NULL, N_("FORMAT") }, { "output-to", 0, 0, G_OPTION_ARG_STRING, &(common_args->output_dest), N_( "Specify file name for output (or \"-\" for stdout)"), N_("DEST") }, { NULL } }; if (*output_group == NULL) { *output_group = g_option_group_new("output", N_("Output Options:"), N_("Show output help"), NULL, NULL); } common_args->output_as_descr = crm_strdup_printf("Specify output format as one of: %s", fmts); output_entries[0].description = common_args->output_as_descr; g_option_group_add_entries(*output_group, output_entries); g_option_context_add_group(context, *output_group); } // main_group is now owned by context, we don't free it here return context; } void pcmk__free_arg_context(GOptionContext *context) { if (context == NULL) { return; } g_option_context_free(context); } void pcmk__add_main_args(GOptionContext *context, const GOptionEntry entries[]) { GOptionGroup *main_group = g_option_context_get_main_group(context); g_option_group_add_entries(main_group, entries); } void pcmk__add_arg_group(GOptionContext *context, const char *name, const char *header, const char *desc, const GOptionEntry entries[]) { GOptionGroup *group = NULL; group = g_option_group_new(name, header, desc, NULL, NULL); g_option_group_add_entries(group, entries); g_option_context_add_group(context, group); // group is now owned by context, we don't free it here } gchar * pcmk__quote_cmdline(gchar **argv) { GString *cmdline = NULL; if (argv == NULL) { return NULL; } for (int i = 0; argv[i] != NULL; i++) { gint argc = 0; /* Quote the argument if it's unparsable as-is (empty, all whitespace, * or having mismatched quotes), or if it contains more than one token */ if (!g_shell_parse_argv(argv[i], &argc, NULL, NULL) || (argc > 1)) { gchar *quoted = g_shell_quote(argv[i]); pcmk__add_word(&cmdline, 128, quoted); g_free(quoted); } else { pcmk__add_word(&cmdline, 128, argv[i]); } } if (cmdline == NULL) { return NULL; } return g_string_free(cmdline, FALSE); } gchar ** pcmk__cmdline_preproc(char *const *argv, const char *special) { GPtrArray *arr = NULL; bool saw_dash_dash = false; bool copy_option = false; if (argv == NULL) { return NULL; } if (g_get_prgname() == NULL && argv && *argv) { gchar *basename = g_path_get_basename(*argv); g_set_prgname(basename); g_free(basename); } arr = g_ptr_array_new(); for (int i = 0; argv[i] != NULL; i++) { /* If this is the first time we saw "--" in the command line, set * a flag so we know to just copy everything after it over. We also * want to copy the "--" over so whatever actually parses the command * line when we're done knows where arguments end. */ if (saw_dash_dash == false && strcmp(argv[i], "--") == 0) { saw_dash_dash = true; } if (saw_dash_dash == true) { g_ptr_array_add(arr, g_strdup(argv[i])); continue; } if (copy_option == true) { g_ptr_array_add(arr, g_strdup(argv[i])); copy_option = false; continue; } /* This is just a dash by itself. That could indicate stdin/stdout, or * it could be user error. Copy it over and let glib figure it out. */ if (pcmk__str_eq(argv[i], "-", pcmk__str_casei)) { g_ptr_array_add(arr, g_strdup(argv[i])); continue; } /* "-INFINITY" is almost certainly meant as a string, not as an option * list */ if (strcmp(argv[i], "-INFINITY") == 0) { g_ptr_array_add(arr, g_strdup(argv[i])); continue; } /* This is a short argument, or perhaps several. Iterate over it * and explode them out into individual arguments. */ if (g_str_has_prefix(argv[i], "-") && !g_str_has_prefix(argv[i], "--")) { /* Skip over leading dash */ const char *ch = argv[i]+1; /* This looks like the start of a number, which means it is a negative * number. It's probably the argument to the preceeding option, but * we can't know that here. Copy it over and let whatever handles * arguments next figure it out. */ if (*ch != '\0' && *ch >= '1' && *ch <= '9') { bool is_numeric = true; while (*ch != '\0') { if (!isdigit(*ch)) { is_numeric = false; break; } ch++; } if (is_numeric) { g_ptr_array_add(arr, g_strdup_printf("%s", argv[i])); continue; } else { /* This argument wasn't entirely numeric. Reset ch to the * beginning so we can process it one character at a time. */ ch = argv[i]+1; } } while (*ch != '\0') { /* This is a special short argument that takes an option. getopt * allows values to be interspersed with a list of arguments, but * glib does not. Grab both the argument and its value and * separate them into a new argument. */ if (special != NULL && strchr(special, *ch) != NULL) { /* The argument does not occur at the end of this string of * arguments. Take everything through the end as its value. */ if (*(ch+1) != '\0') { fprintf(stderr, "Deprecated argument format '-%c%s' used.\n", *ch, ch+1); fprintf(stderr, "Please use '-%c %s' instead. " "Support will be removed in a future release.\n", *ch, ch+1); g_ptr_array_add(arr, g_strdup_printf("-%c", *ch)); g_ptr_array_add(arr, g_strdup(ch+1)); break; /* The argument occurs at the end of this string. Hopefully * whatever comes next in argv is its value. It may not be, * but that is not for us to decide. */ } else { g_ptr_array_add(arr, g_strdup_printf("-%c", *ch)); copy_option = true; ch++; } /* This is a regular short argument. Just copy it over. */ } else { g_ptr_array_add(arr, g_strdup_printf("-%c", *ch)); ch++; } } /* This is a long argument, or an option, or something else. * Copy it over - everything else is copied, so this keeps it easy for * the caller to know what to do with the memory when it's done. */ } else { g_ptr_array_add(arr, g_strdup(argv[i])); } } g_ptr_array_add(arr, NULL); return (char **) g_ptr_array_free(arr, FALSE); } - -G_GNUC_PRINTF(3, 4) -gboolean -pcmk__force_args(GOptionContext *context, GError **error, const char *format, ...) { - int len = 0; - char *buf = NULL; - gchar **extra_args = NULL; - va_list ap; - gboolean retval = TRUE; - - va_start(ap, format); - len = vasprintf(&buf, format, ap); - pcmk__assert(len > 0); - va_end(ap); - - if (!g_shell_parse_argv(buf, NULL, &extra_args, error)) { - g_strfreev(extra_args); - free(buf); - return FALSE; - } - - retval = g_option_context_parse_strv(context, &extra_args, error); - - g_strfreev(extra_args); - free(buf); - return retval; -} diff --git a/lib/common/digest.c b/lib/common/digest.c index cdc4a61879..e7138490af 100644 --- a/lib/common/digest.c +++ b/lib/common/digest.c @@ -1,402 +1,402 @@ /* * Copyright 2015-2025 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 // GString, etc. #include // gnutls_hash_fast(), gnutls_hash_get_len() #include // gnutls_strerror() #include #include #include "crmcommon_private.h" #define BEST_EFFORT_STATUS 0 /* * Pacemaker uses digests (MD5 hashes) of stringified XML to detect changes in * the CIB as a whole, a particular resource's agent parameters, and the device * parameters last used to unfence a particular node. * * "v2" digests hash pcmk__xml_string() directly, while less efficient "v1" * digests do the same with a prefixed space, suffixed newline, and optional * pre-sorting. * * On-disk CIB digests use v1 without sorting. * * Operation digests use v1 with sorting, and are stored in a resource's * operation history in the CIB status section. They come in three flavors: * - a digest of (nearly) all resource parameters and options, used to detect * any resource configuration change; * - a digest of resource parameters marked as nonreloadable, used to decide * whether a reload or full restart is needed after a configuration change; * - and a digest of resource parameters not marked as private, used in * simulations where private parameters have been removed from the input. * * Unfencing digests are set as node attributes, and are used to require * that nodes be unfenced again after a device's configuration changes. */ /*! * \internal * \brief Dump XML in a format used with v1 digests * * \param[in] xml Root of XML to dump * * \return Newly allocated buffer containing dumped XML */ static GString * dump_xml_for_digest(const xmlNode *xml) { GString *buffer = g_string_sized_new(1024); /* for compatibility with the old result which is used for v1 digests */ g_string_append_c(buffer, ' '); pcmk__xml_string(xml, 0, buffer, 0); g_string_append_c(buffer, '\n'); return buffer; } /*! * \internal * \brief Calculate and return v1 digest of XML tree * * \param[in] input Root of XML to digest * * \return Newly allocated string containing digest * * \note Example return value: "c048eae664dba840e1d2060f00299e9d" */ static char * calculate_xml_digest_v1(const xmlNode *input) { GString *buffer = dump_xml_for_digest(input); char *digest = NULL; // buffer->len > 2 for initial space and trailing newline CRM_CHECK(buffer->len > 2, g_string_free(buffer, TRUE); return NULL); digest = crm_md5sum((const char *) buffer->str); crm_log_xml_trace(input, "digest:source"); g_string_free(buffer, TRUE); return digest; } /*! * \internal * \brief Calculate and return the digest of a CIB, suitable for storing on disk * * \param[in] input Root of XML to digest * * \return Newly allocated string containing digest */ char * pcmk__digest_on_disk_cib(const xmlNode *input) { /* Always use the v1 format for on-disk digests. * * Switching to v2 affects even full-restart upgrades, so it would be a * compatibility nightmare. * * We only use this once at startup. All other invocations are in a * separate child process. */ return calculate_xml_digest_v1(input); } /*! * \internal * \brief Calculate and return digest of a \c PCMK_XE_PARAMETERS element * * This is intended for parameters of a resource operation (also known as * resource action). A \c PCMK_XE_PARAMETERS element from a different source * (for example, resource agent metadata) may have child elements, which are not * allowed here. * * The digest is invariant to changes in the order of XML attributes. * * \param[in] input XML element to digest (must have no children) * * \return Newly allocated string containing digest */ char * pcmk__digest_op_params(const xmlNode *input) { /* Switching to v2 digests would likely cause restarts during rolling * upgrades. * * @TODO Confirm this. Switch to v2 if safe, or drop this TODO otherwise. */ char *digest = NULL; xmlNode *sorted = NULL; pcmk__assert(input->children == NULL); sorted = pcmk__xe_create(NULL, (const char *) input->name); pcmk__xe_copy_attrs(sorted, input, pcmk__xaf_none); pcmk__xe_sort_attrs(sorted); digest = calculate_xml_digest_v1(sorted); pcmk__xml_free(sorted); return digest; } /*! * \internal * \brief Calculate and return the digest of an XML tree * * \param[in] xml XML tree to digest * \param[in] filter Whether to filter certain XML attributes * * \return Newly allocated string containing digest */ char * pcmk__digest_xml(const xmlNode *xml, bool filter) { /* @TODO Filtering accounts for significant CPU usage. Consider removing if * possible. */ char *digest = NULL; GString *buf = g_string_sized_new(1024); pcmk__xml_string(xml, (filter? pcmk__xml_fmt_filtered : 0), buf, 0); digest = crm_md5sum(buf->str); if (digest == NULL) { goto done; } pcmk__if_tracing( { char *trace_file = crm_strdup_printf("%s/digest-%s", pcmk__get_tmpdir(), digest); crm_trace("Saving %s.%s.%s to %s", pcmk__xe_get(xml, PCMK_XA_ADMIN_EPOCH), pcmk__xe_get(xml, PCMK_XA_EPOCH), pcmk__xe_get(xml, PCMK_XA_NUM_UPDATES), trace_file); save_xml_to_file(xml, "digest input", trace_file); free(trace_file); }, {} ); done: g_string_free(buf, TRUE); return digest; } /*! * \internal * \brief Check whether calculated digest of given XML matches expected digest * * \param[in] input Root of XML tree to digest * \param[in] expected Expected digest in on-disk format * * \return true if digests match, false on mismatch or error */ bool pcmk__verify_digest(const xmlNode *input, const char *expected) { char *calculated = NULL; bool passed; if (input != NULL) { calculated = pcmk__digest_on_disk_cib(input); if (calculated == NULL) { - crm_perror(LOG_ERR, "Could not calculate digest for comparison"); + crm_err("Could not calculate digest for comparison"); return false; } } passed = pcmk__str_eq(expected, calculated, pcmk__str_casei); if (passed) { crm_trace("Digest comparison passed: %s", calculated); } else { crm_err("Digest comparison failed: expected %s, calculated %s", expected, calculated); } free(calculated); return passed; } /*! * \internal * \brief Check whether an XML attribute should be excluded from CIB digests * * \param[in] name XML attribute name * * \return true if XML attribute should be excluded from CIB digest calculation */ bool pcmk__xa_filterable(const char *name) { static const char *filter[] = { PCMK_XA_CRM_DEBUG_ORIGIN, PCMK_XA_CIB_LAST_WRITTEN, PCMK_XA_UPDATE_ORIGIN, PCMK_XA_UPDATE_CLIENT, PCMK_XA_UPDATE_USER, }; for (int i = 0; i < PCMK__NELEM(filter); i++) { if (strcmp(name, filter[i]) == 0) { return true; } } return false; } char * crm_md5sum(const char *buffer) { char *digest = NULL; gchar *raw_digest = NULL; /* g_compute_checksum_for_string returns NULL if the input string is empty. * There are instances where we may want to hash an empty, but non-NULL, * string so here we just hardcode the result. */ if (buffer == NULL) { return NULL; } else if (pcmk__str_empty(buffer)) { return pcmk__str_copy("d41d8cd98f00b204e9800998ecf8427e"); } raw_digest = g_compute_checksum_for_string(G_CHECKSUM_MD5, buffer, -1); if (raw_digest == NULL) { crm_err("Failed to calculate hash"); return NULL; } digest = pcmk__str_copy(raw_digest); g_free(raw_digest); crm_trace("Digest %s.", digest); return digest; } // Return true if a is an attribute that should be filtered static bool should_filter_for_digest(xmlAttrPtr a, void *user_data) { if (strncmp((const char *) a->name, CRM_META "_", sizeof(CRM_META " ") - 1) == 0) { return true; } return pcmk__str_any_of((const char *) a->name, PCMK_XA_ID, PCMK_XA_CRM_FEATURE_SET, PCMK__XA_OP_DIGEST, PCMK__META_ON_NODE, PCMK__META_ON_NODE_UUID, "pcmk_external_ip", NULL); } /*! * \internal * \brief Remove XML attributes not needed for operation digest * * \param[in,out] param_set XML with operation parameters */ void pcmk__filter_op_for_digest(xmlNode *param_set) { char *key = NULL; char *timeout = NULL; guint interval_ms = 0; if (param_set == NULL) { return; } /* Timeout is useful for recurring operation digests, so grab it before * removing meta-attributes */ key = crm_meta_name(PCMK_META_INTERVAL); pcmk__xe_get_guint(param_set, key, &interval_ms); free(key); key = NULL; if (interval_ms != 0) { key = crm_meta_name(PCMK_META_TIMEOUT); timeout = pcmk__xe_get_copy(param_set, key); } // Remove all CRM_meta_* attributes and certain other attributes pcmk__xe_remove_matching_attrs(param_set, false, should_filter_for_digest, NULL); // Add timeout back for recurring operation digests if (timeout != NULL) { crm_xml_add(param_set, key, timeout); } free(timeout); free(key); } // Deprecated functions kept only for backward API compatibility // LCOV_EXCL_START #include #include char * calculate_on_disk_digest(xmlNode *input) { return calculate_xml_digest_v1(input); } char * calculate_operation_digest(xmlNode *input, const char *version) { xmlNode *sorted = sorted_xml(input, NULL, true); char *digest = calculate_xml_digest_v1(sorted); pcmk__xml_free(sorted); return digest; } char * calculate_xml_versioned_digest(xmlNode *input, gboolean sort, gboolean do_filter, const char *version) { if ((version == NULL) || (compare_version("3.0.5", version) > 0)) { xmlNode *sorted = NULL; char *digest = NULL; if (sort) { xmlNode *sorted = sorted_xml(input, NULL, true); input = sorted; } crm_trace("Using v1 digest algorithm for %s", pcmk__s(version, "unknown feature set")); digest = calculate_xml_digest_v1(input); pcmk__xml_free(sorted); return digest; } crm_trace("Using v2 digest algorithm for %s", version); return pcmk__digest_xml(input, do_filter); } // LCOV_EXCL_STOP // End deprecated API diff --git a/lib/common/io.c b/lib/common/io.c index 93535ea539..ff5e58d61d 100644 --- a/lib/common/io.c +++ b/lib/common/io.c @@ -1,640 +1,635 @@ /* * Copyright 2004-2025 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 #include #include #include #include #include #include #include /*! * \internal * \brief Create a directory, including any parent directories needed * * \param[in] path_c Pathname of the directory to create * \param[in] mode Permissions to be used (with current umask) when creating * * \return Standard Pacemaker return code */ int pcmk__build_path(const char *path_c, mode_t mode) { int offset = 1, len = 0; int rc = pcmk_rc_ok; char *path = strdup(path_c); CRM_CHECK(path != NULL, return -ENOMEM); for (len = strlen(path); offset < len; offset++) { if (path[offset] == '/') { path[offset] = 0; if ((mkdir(path, mode) < 0) && (errno != EEXIST)) { rc = errno; goto done; } path[offset] = '/'; } } if ((mkdir(path, mode) < 0) && (errno != EEXIST)) { rc = errno; } done: free(path); return rc; } /*! * \internal * \brief Return canonicalized form of a path name * * \param[in] path Pathname to canonicalize * \param[out] resolved_path Where to store canonicalized pathname * * \return Standard Pacemaker return code * \note The caller is responsible for freeing \p resolved_path on success. * \note This function exists because not all C library versions of * realpath(path, resolved_path) support a NULL resolved_path. */ int pcmk__real_path(const char *path, char **resolved_path) { CRM_CHECK((path != NULL) && (resolved_path != NULL), return EINVAL); #if _POSIX_VERSION >= 200809L /* Recent C libraries can dynamically allocate memory as needed */ *resolved_path = realpath(path, NULL); return (*resolved_path == NULL)? errno : pcmk_rc_ok; #elif defined(PATH_MAX) /* Older implementations require pre-allocated memory */ /* (this is less desirable because PATH_MAX may be huge or not defined) */ *resolved_path = malloc(PATH_MAX); if ((*resolved_path == NULL) || (realpath(path, *resolved_path) == NULL)) { return errno; } return pcmk_rc_ok; #else *resolved_path = NULL; return ENOTSUP; #endif } /*! * \internal * \brief Create a file name using a sequence number * * \param[in] directory Directory that contains the file series * \param[in] series Start of file name * \param[in] sequence Sequence number * \param[in] bzip Whether to use ".bz2" instead of ".raw" as extension * * \return Newly allocated file path (asserts on error, so always non-NULL) * \note The caller is responsible for freeing the return value. */ char * pcmk__series_filename(const char *directory, const char *series, unsigned int sequence, bool bzip) { pcmk__assert((directory != NULL) && (series != NULL)); return crm_strdup_printf("%s/%s-%u.%s", directory, series, sequence, (bzip? "bz2" : "raw")); } /*! * \internal * \brief Read sequence number stored in a file series' .last file * * \param[in] directory Directory that contains the file series * \param[in] series Start of file name * \param[out] seq Where to store the sequence number * * \return Standard Pacemaker return code */ int pcmk__read_series_sequence(const char *directory, const char *series, unsigned int *seq) { int rc; FILE *fp = NULL; char *series_file = NULL; if ((directory == NULL) || (series == NULL) || (seq == NULL)) { return EINVAL; } series_file = crm_strdup_printf("%s/%s.last", directory, series); fp = fopen(series_file, "r"); if (fp == NULL) { rc = errno; crm_debug("Could not open series file %s: %s", series_file, strerror(rc)); free(series_file); return rc; } errno = 0; if (fscanf(fp, "%u", seq) != 1) { rc = (errno == 0)? ENODATA : errno; crm_debug("Could not read sequence number from series file %s: %s", series_file, pcmk_rc_str(rc)); fclose(fp); return rc; } fclose(fp); crm_trace("Found last sequence number %u in series file %s", *seq, series_file); free(series_file); return pcmk_rc_ok; } /*! * \internal * \brief Write sequence number to a file series' .last file * * \param[in] directory Directory that contains the file series * \param[in] series Start of file name * \param[in] sequence Sequence number to write * \param[in] max Maximum sequence value, after which it is reset to 0 * * \note This function logs some errors but does not return any to the caller */ void pcmk__write_series_sequence(const char *directory, const char *series, unsigned int sequence, int max) { - int rc = 0; FILE *file_strm = NULL; char *series_file = NULL; CRM_CHECK(directory != NULL, return); CRM_CHECK(series != NULL, return); if (max == 0) { return; } if (max > 0 && sequence >= max) { sequence = 0; } series_file = crm_strdup_printf("%s/%s.last", directory, series); file_strm = fopen(series_file, "w"); if (file_strm != NULL) { - rc = fprintf(file_strm, "%u", sequence); + int rc = fprintf(file_strm, "%u", sequence); + if (rc < 0) { - crm_perror(LOG_ERR, "Cannot write to series file %s", series_file); + crm_err("Cannot write to series file %s", series_file); } + fflush(file_strm); + fclose(file_strm); } else { crm_err("Cannot open series file %s for writing", series_file); } - if (file_strm != NULL) { - fflush(file_strm); - fclose(file_strm); - } - - crm_trace("Wrote %d to %s", sequence, series_file); free(series_file); } /*! * \internal * \brief Change the owner and group of a file series' .last file * * \param[in] directory Directory that contains series * \param[in] series Series to change * \param[in] uid User ID of desired file owner * \param[in] gid Group ID of desired file group * * \return Standard Pacemaker return code * \note The caller must have the appropriate privileges. */ int pcmk__chown_series_sequence(const char *directory, const char *series, uid_t uid, gid_t gid) { char *series_file = NULL; int rc = pcmk_rc_ok; if ((directory == NULL) || (series == NULL)) { return EINVAL; } series_file = crm_strdup_printf("%s/%s.last", directory, series); if (chown(series_file, uid, gid) < 0) { rc = errno; } free(series_file); return rc; } static bool pcmk__daemon_user_can_write(const char *target_name, struct stat *target_stat) { struct passwd *sys_user = NULL; errno = 0; sys_user = getpwnam(CRM_DAEMON_USER); if (sys_user == NULL) { crm_notice("Could not find user %s: %s", CRM_DAEMON_USER, pcmk_rc_str(errno)); return FALSE; } if (target_stat->st_uid != sys_user->pw_uid) { crm_notice("%s is not owned by user %s " QB_XS " uid %d != %d", target_name, CRM_DAEMON_USER, sys_user->pw_uid, target_stat->st_uid); return FALSE; } if ((target_stat->st_mode & (S_IRUSR | S_IWUSR)) == 0) { crm_notice("%s is not readable and writable by user %s " QB_XS " st_mode=0%lo", target_name, CRM_DAEMON_USER, (unsigned long) target_stat->st_mode); return FALSE; } return TRUE; } static bool pcmk__daemon_group_can_write(const char *target_name, struct stat *target_stat) { struct group *sys_grp = NULL; errno = 0; sys_grp = getgrnam(CRM_DAEMON_GROUP); if (sys_grp == NULL) { crm_notice("Could not find group %s: %s", CRM_DAEMON_GROUP, pcmk_rc_str(errno)); return FALSE; } if (target_stat->st_gid != sys_grp->gr_gid) { crm_notice("%s is not owned by group %s " QB_XS " uid %d != %d", target_name, CRM_DAEMON_GROUP, sys_grp->gr_gid, target_stat->st_gid); return FALSE; } if ((target_stat->st_mode & (S_IRGRP | S_IWGRP)) == 0) { crm_notice("%s is not readable and writable by group %s " QB_XS " st_mode=0%lo", target_name, CRM_DAEMON_GROUP, (unsigned long) target_stat->st_mode); return FALSE; } return TRUE; } /*! * \internal * \brief Check whether a directory or file is writable by the cluster daemon * * Return true if either the cluster daemon user or cluster daemon group has * write permission on a specified file or directory. * * \param[in] dir Directory to check (this argument must be specified, and * the directory must exist) * \param[in] file File to check (only the directory will be checked if this * argument is not specified or the file does not exist) * * \return true if target is writable by cluster daemon, false otherwise */ bool pcmk__daemon_can_write(const char *dir, const char *file) { int s_res = 0; struct stat buf; char *full_file = NULL; const char *target = NULL; // Caller must supply directory pcmk__assert(dir != NULL); // If file is given, check whether it exists as a regular file if (file != NULL) { full_file = crm_strdup_printf("%s/%s", dir, file); target = full_file; s_res = stat(full_file, &buf); if (s_res < 0) { crm_notice("%s not found: %s", target, pcmk_rc_str(errno)); free(full_file); full_file = NULL; target = NULL; } else if (S_ISREG(buf.st_mode) == FALSE) { crm_err("%s must be a regular file " QB_XS " st_mode=0%lo", target, (unsigned long) buf.st_mode); free(full_file); return false; } } // If file is not given, ensure dir exists as directory if (target == NULL) { target = dir; s_res = stat(dir, &buf); if (s_res < 0) { crm_err("%s not found: %s", dir, pcmk_rc_str(errno)); return false; } else if (S_ISDIR(buf.st_mode) == FALSE) { crm_err("%s must be a directory " QB_XS " st_mode=0%lo", dir, (unsigned long) buf.st_mode); return false; } } if (!pcmk__daemon_user_can_write(target, &buf) && !pcmk__daemon_group_can_write(target, &buf)) { crm_err("%s must be owned and writable by either user %s or group %s " QB_XS " st_mode=0%lo", target, CRM_DAEMON_USER, CRM_DAEMON_GROUP, (unsigned long) buf.st_mode); free(full_file); return false; } free(full_file); return true; } /*! * \internal * \brief Flush and sync a directory to disk * * \param[in] name Directory to flush and sync * \note This function logs errors but does not return them to the caller */ void pcmk__sync_directory(const char *name) { int fd; DIR *directory; directory = opendir(name); if (directory == NULL) { - crm_perror(LOG_ERR, "Could not open %s for syncing", name); + crm_err("Could not open %s for syncing: %s", name, strerror(errno)); return; } fd = dirfd(directory); if (fd < 0) { - crm_perror(LOG_ERR, "Could not obtain file descriptor for %s", name); + crm_err("Could not obtain file descriptor for %s: %s", name, + strerror(errno)); return; } if (fsync(fd) < 0) { - crm_perror(LOG_ERR, "Could not sync %s", name); + crm_err("Could not sync %s: %s", name, strerror(errno)); } if (closedir(directory) < 0) { - crm_perror(LOG_ERR, "Could not close %s after fsync", name); + crm_err("Could not close %s after fsync: %s", name, strerror(errno)); } } /*! * \internal * \brief Read the contents of a file * * \param[in] filename Name of file to read * \param[out] contents Where to store file contents * * \return Standard Pacemaker return code * \note On success, the caller is responsible for freeing contents. */ int pcmk__file_contents(const char *filename, char **contents) { FILE *fp; int length, read_len; int rc = pcmk_rc_ok; if ((filename == NULL) || (contents == NULL)) { return EINVAL; } fp = fopen(filename, "r"); if ((fp == NULL) || (fseek(fp, 0L, SEEK_END) < 0)) { rc = errno; goto bail; } length = ftell(fp); if (length < 0) { rc = errno; goto bail; } if (length == 0) { *contents = NULL; } else { *contents = calloc(length + 1, sizeof(char)); if (*contents == NULL) { rc = errno; goto bail; } errno = 0; rewind(fp); if (errno != 0) { rc = errno; goto bail; } read_len = fread(*contents, 1, length, fp); if (read_len != length) { free(*contents); *contents = NULL; rc = EIO; } else { /* Coverity thinks *contents isn't null-terminated. It doesn't * understand calloc(). */ (*contents)[length] = '\0'; } } bail: if (fp != NULL) { fclose(fp); } return rc; } /*! * \internal * \brief Write text to a file, flush and sync it to disk, then close the file * * \param[in] fd File descriptor opened for writing * \param[in] contents String to write to file * * \return Standard Pacemaker return code */ int pcmk__write_sync(int fd, const char *contents) { int rc = 0; FILE *fp = fdopen(fd, "w"); if (fp == NULL) { return errno; } if ((contents != NULL) && (fprintf(fp, "%s", contents) < 0)) { rc = EIO; } if (fflush(fp) != 0) { rc = errno; } if (fsync(fileno(fp)) < 0) { rc = errno; } fclose(fp); return rc; } /*! * \internal * \brief Set a file descriptor to non-blocking * * \param[in] fd File descriptor to use * * \return Standard Pacemaker return code */ int pcmk__set_nonblocking(int fd) { int flag = fcntl(fd, F_GETFL); if (flag < 0) { return errno; } if (fcntl(fd, F_SETFL, flag | O_NONBLOCK) < 0) { return errno; } return pcmk_rc_ok; } /*! * \internal * \brief Get directory name for temporary files * * Return the value of the TMPDIR environment variable if it is set to a * full path, otherwise return "/tmp". * * \return Name of directory to be used for temporary files */ const char * pcmk__get_tmpdir(void) { const char *dir = getenv("TMPDIR"); return (dir && (*dir == '/'))? dir : "/tmp"; } /*! * \internal - * \brief Close open file descriptors - * - * Close all file descriptors (except optionally stdin, stdout, and stderr), - * which is a best practice for a new child process forked for the purpose of - * executing an external program. + * \brief Close open file descriptors except standard streams * - * \param[in] bool If true, close stdin, stdout, and stderr as well + * Close all file descriptors (except stdin, stdout, and stderr), which is a + * best practice for a new child process forked for the purpose of executing an + * external program. */ void -pcmk__close_fds_in_child(bool all) +pcmk__close_fds_in_child(void) { DIR *dir; struct rlimit rlim; rlim_t max_fd; - int min_fd = (all? 0 : (STDERR_FILENO + 1)); + const int min_fd = STDERR_FILENO + 1; /* Find the current process's (soft) limit for open files. getrlimit() * should always work, but have a fallback just in case. */ if (getrlimit(RLIMIT_NOFILE, &rlim) == 0) { max_fd = rlim.rlim_cur - 1; } else { long conf_max = sysconf(_SC_OPEN_MAX); max_fd = (conf_max > 0)? conf_max : 1024; } /* First try /proc. If that returns NULL (either because opening the * directory failed, or because procfs isn't supported on this platform), * fall back to /dev/fd. */ dir = pcmk__procfs_fd_dir(); if (dir == NULL) { dir = opendir("/dev/fd"); } if (dir != NULL) { struct dirent *entry; int dir_fd = dirfd(dir); while ((entry = readdir(dir)) != NULL) { int lpc = atoi(entry->d_name); /* How could one of these entries be higher than max_fd, you ask? * It isn't possible in normal operation, but when run under * valgrind, valgrind can open high-numbered file descriptors for * its own use that are higher than the process's soft limit. * These will show up in the fd directory but aren't closable. */ if ((lpc >= min_fd) && (lpc <= max_fd) && (lpc != dir_fd)) { close(lpc); } } closedir(dir); return; } /* If no fd directory is available, iterate over all possible descriptors. * This is less efficient due to the overhead of many system calls. */ for (int lpc = max_fd; lpc >= min_fd; lpc--) { close(lpc); } } /*! * \brief Duplicate a file path, inserting a prefix if not absolute * * \param[in] filename File path to duplicate * \param[in] dirname If filename is not absolute, prefix to add * * \return Newly allocated memory with full path (guaranteed non-NULL) */ char * pcmk__full_path(const char *filename, const char *dirname) { pcmk__assert(filename != NULL); if (filename[0] == '/') { return pcmk__str_copy(filename); } pcmk__assert(dirname != NULL); return crm_strdup_printf("%s/%s", dirname, filename); } diff --git a/lib/common/logging.c b/lib/common/logging.c index 5fa8cd20c6..a879a32c27 100644 --- a/lib/common/logging.c +++ b/lib/common/logging.c @@ -1,1310 +1,1313 @@ /* * Copyright 2004-2025 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 #include #include #include #include #include #include #include #include #include #include #include // Use high-resolution (millisecond) timestamps if libqb supports them #ifdef QB_FEATURE_LOG_HIRES_TIMESTAMPS #define TIMESTAMP_FORMAT_SPEC "%%T" typedef struct timespec *log_time_t; #else #define TIMESTAMP_FORMAT_SPEC "%%t" typedef time_t log_time_t; #endif unsigned int crm_log_level = LOG_INFO; unsigned int crm_trace_nonlog = 0; bool pcmk__is_daemon = false; static unsigned int crm_log_priority = LOG_NOTICE; static guint pcmk__log_id = 0; static guint pcmk__glib_log_id = 0; static guint pcmk__gio_log_id = 0; static guint pcmk__gmodule_log_id = 0; static guint pcmk__gthread_log_id = 0; static pcmk__output_t *logger_out = NULL; pcmk__config_error_func pcmk__config_error_handler = NULL; pcmk__config_warning_func pcmk__config_warning_handler = NULL; void *pcmk__config_error_context = NULL; void *pcmk__config_warning_context = NULL; static gboolean crm_tracing_enabled(void); static void crm_glib_handler(const gchar * log_domain, GLogLevelFlags flags, const gchar * message, gpointer user_data) { int log_level = LOG_WARNING; GLogLevelFlags msg_level = (flags & G_LOG_LEVEL_MASK); static struct qb_log_callsite *glib_cs = NULL; if (glib_cs == NULL) { glib_cs = qb_log_callsite_get(__func__, __FILE__, "glib-handler", LOG_DEBUG, __LINE__, crm_trace_nonlog); } switch (msg_level) { case G_LOG_LEVEL_CRITICAL: log_level = LOG_CRIT; if (!crm_is_callsite_active(glib_cs, LOG_DEBUG, crm_trace_nonlog)) { /* log and record how we got here */ crm_abort(__FILE__, __func__, __LINE__, message, TRUE, TRUE); } break; case G_LOG_LEVEL_ERROR: log_level = LOG_ERR; break; case G_LOG_LEVEL_MESSAGE: log_level = LOG_NOTICE; break; case G_LOG_LEVEL_INFO: log_level = LOG_INFO; break; case G_LOG_LEVEL_DEBUG: log_level = LOG_DEBUG; break; case G_LOG_LEVEL_WARNING: case G_LOG_FLAG_RECURSION: case G_LOG_FLAG_FATAL: case G_LOG_LEVEL_MASK: log_level = LOG_WARNING; break; } do_crm_log(log_level, "%s: %s", log_domain, message); } #ifndef NAME_MAX # define NAME_MAX 256 #endif /*! * \internal * \brief Write out a blackbox (enabling blackboxes if needed) * * \param[in] nsig Signal number that was received * * \note This is a true signal handler, and so must be async-safe. */ static void crm_trigger_blackbox(int nsig) { if(nsig == SIGTRAP) { /* Turn it on if it wasn't already */ crm_enable_blackbox(nsig); } crm_write_blackbox(nsig, NULL); } void crm_log_deinit(void) { if (pcmk__log_id == 0) { return; } g_log_remove_handler(G_LOG_DOMAIN, pcmk__log_id); pcmk__log_id = 0; g_log_remove_handler("GLib", pcmk__glib_log_id); pcmk__glib_log_id = 0; g_log_remove_handler("GLib-GIO", pcmk__gio_log_id); pcmk__gio_log_id = 0; g_log_remove_handler("GModule", pcmk__gmodule_log_id); pcmk__gmodule_log_id = 0; g_log_remove_handler("GThread", pcmk__gthread_log_id); pcmk__gthread_log_id = 0; } /*! * \internal * \brief Set the log format string based on the passed-in method * * \param[in] method The detail level of the log output * \param[in] daemon The daemon ID included in error messages * \param[in] use_pid Cached result of getpid() call, for efficiency * \param[in] use_nodename Cached result of uname() call, for efficiency * */ /* XXX __attribute__((nonnull)) for use_nodename parameter */ static void set_format_string(int method, const char *daemon, pid_t use_pid, const char *use_nodename) { if (method == QB_LOG_SYSLOG) { // The system log gets a simplified, user-friendly format qb_log_ctl(method, QB_LOG_CONF_EXTENDED, QB_FALSE); qb_log_format_set(method, "%g %p: %b"); } else { // Everything else gets more detail, for advanced troubleshooting GString *fmt = g_string_sized_new(256); if (method > QB_LOG_STDERR) { // If logging to file, prefix with timestamp, node name, daemon ID g_string_append_printf(fmt, TIMESTAMP_FORMAT_SPEC " %s %-20s[%lld] ", use_nodename, daemon, (long long) use_pid); } // Add function name (in parentheses) g_string_append(fmt, "(%n"); if (crm_tracing_enabled()) { // When tracing, add file and line number g_string_append(fmt, "@%f:%l"); } g_string_append_c(fmt, ')'); // Add tag (if any), severity, and actual message g_string_append(fmt, " %g\t%p: %b"); CRM_LOG_ASSERT(fmt->len > 0); qb_log_format_set(method, fmt->str); g_string_free(fmt, TRUE); } } #define DEFAULT_LOG_FILE CRM_LOG_DIR "/pacemaker.log" static bool logfile_disabled(const char *filename) { return pcmk__str_eq(filename, PCMK_VALUE_NONE, pcmk__str_casei) || pcmk__str_eq(filename, "/dev/null", pcmk__str_none); } /*! * \internal * \brief Fix log file ownership if group is wrong or doesn't have access * * \param[in] filename Log file name (for logging only) * \param[in] logfd Log file descriptor * * \return Standard Pacemaker return code */ static int chown_logfile(const char *filename, int logfd) { uid_t pcmk_uid = 0; gid_t pcmk_gid = 0; struct stat st; int rc; // Get the log file's current ownership and permissions if (fstat(logfd, &st) < 0) { return errno; } // Any other errors don't prevent file from being used as log rc = pcmk_daemon_user(&pcmk_uid, &pcmk_gid); if (rc != pcmk_ok) { rc = pcmk_legacy2rc(rc); crm_warn("Not changing '%s' ownership because user information " "unavailable: %s", filename, pcmk_rc_str(rc)); return pcmk_rc_ok; } if ((st.st_gid == pcmk_gid) && ((st.st_mode & S_IRWXG) == (S_IRGRP|S_IWGRP))) { return pcmk_rc_ok; } if (fchown(logfd, pcmk_uid, pcmk_gid) < 0) { crm_warn("Couldn't change '%s' ownership to user %s gid %d: %s", filename, CRM_DAEMON_USER, pcmk_gid, strerror(errno)); } return pcmk_rc_ok; } // Reset log file permissions (using environment variable if set) static void chmod_logfile(const char *filename, int logfd) { const char *modestr = pcmk__env_option(PCMK__ENV_LOGFILE_MODE); mode_t filemode = S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP; if (modestr != NULL) { long filemode_l = strtol(modestr, NULL, 8); if ((filemode_l != LONG_MIN) && (filemode_l != LONG_MAX)) { filemode = (mode_t) filemode_l; } } if ((filemode != 0) && (fchmod(logfd, filemode) < 0)) { crm_warn("Couldn't change '%s' mode to %04o: %s", filename, filemode, strerror(errno)); } } // If we're root, correct a log file's permissions if needed static int set_logfile_permissions(const char *filename, FILE *logfile) { if (geteuid() == 0) { int logfd = fileno(logfile); int rc = chown_logfile(filename, logfd); if (rc != pcmk_rc_ok) { return rc; } chmod_logfile(filename, logfd); } return pcmk_rc_ok; } // Enable libqb logging to a new log file static void enable_logfile(int fd) { qb_log_ctl(fd, QB_LOG_CONF_ENABLED, QB_TRUE); #if 0 qb_log_ctl(fd, QB_LOG_CONF_FILE_SYNC, 1); // Turn on synchronous writes #endif #ifdef HAVE_qb_log_conf_QB_LOG_CONF_MAX_LINE_LEN // Longer than default, for logging long XML lines qb_log_ctl(fd, QB_LOG_CONF_MAX_LINE_LEN, 800); #endif crm_update_callsites(); } static inline void disable_logfile(int fd) { qb_log_ctl(fd, QB_LOG_CONF_ENABLED, QB_FALSE); } static void setenv_logfile(const char *filename) { // Some resource agents will log only if environment variable is set if (pcmk__env_option(PCMK__ENV_LOGFILE) == NULL) { pcmk__set_env_option(PCMK__ENV_LOGFILE, filename, true); } } /*! * \brief Add a file to be used as a Pacemaker detail log * * \param[in] filename Name of log file to use * * \return Standard Pacemaker return code */ int pcmk__add_logfile(const char *filename) { /* No log messages from this function will be logged to the new log! * If another target such as syslog has already been added, the messages * should show up there. */ int fd = 0; int rc = pcmk_rc_ok; FILE *logfile = NULL; bool is_default = false; static int default_fd = -1; static bool have_logfile = false; // Use default if caller didn't specify (and we don't already have one) if (filename == NULL) { if (have_logfile) { return pcmk_rc_ok; } filename = DEFAULT_LOG_FILE; } // If the user doesn't want logging, we're done if (logfile_disabled(filename)) { return pcmk_rc_ok; } // If the caller wants the default and we already have it, we're done is_default = pcmk__str_eq(filename, DEFAULT_LOG_FILE, pcmk__str_none); if (is_default && (default_fd >= 0)) { return pcmk_rc_ok; } // Check whether we have write access to the file logfile = fopen(filename, "a"); if (logfile == NULL) { rc = errno; crm_warn("Logging to '%s' is disabled: %s " QB_XS " uid=%u gid=%u", filename, strerror(rc), geteuid(), getegid()); return rc; } rc = set_logfile_permissions(filename, logfile); if (rc != pcmk_rc_ok) { crm_warn("Logging to '%s' is disabled: %s " QB_XS " permissions", filename, strerror(rc)); fclose(logfile); return rc; } // Close and reopen as libqb logging target fclose(logfile); fd = qb_log_file_open(filename); if (fd < 0) { crm_warn("Logging to '%s' is disabled: %s " QB_XS " qb_log_file_open", filename, strerror(-fd)); return -fd; // == +errno } if (is_default) { default_fd = fd; setenv_logfile(filename); } else if (default_fd >= 0) { crm_notice("Switching logging to %s", filename); disable_logfile(default_fd); } crm_notice("Additional logging available in %s", filename); enable_logfile(fd); have_logfile = true; return pcmk_rc_ok; } /*! * \brief Add multiple additional log files * * \param[in] log_files Array of log files to add * \param[in] out Output object to use for error reporting * * \return Standard Pacemaker return code */ void pcmk__add_logfiles(gchar **log_files, pcmk__output_t *out) { if (log_files == NULL) { return; } for (gchar **fname = log_files; *fname != NULL; fname++) { int rc = pcmk__add_logfile(*fname); if (rc != pcmk_rc_ok) { out->err(out, "Logging to %s is disabled: %s", *fname, pcmk_rc_str(rc)); } } } static int blackbox_trigger = 0; static volatile char *blackbox_file_prefix = NULL; static void blackbox_logger(int32_t t, struct qb_log_callsite *cs, log_time_t timestamp, const char *msg) { if(cs && cs->priority < LOG_ERR) { crm_write_blackbox(SIGTRAP, cs); /* Bypass the over-dumping logic */ } else { crm_write_blackbox(0, cs); } } static void crm_control_blackbox(int nsig, bool enable) { int lpc = 0; if (blackbox_file_prefix == NULL) { pid_t pid = getpid(); blackbox_file_prefix = crm_strdup_printf("%s/%s-%lu", CRM_BLACKBOX_DIR, crm_system_name, (unsigned long) pid); } if (enable && qb_log_ctl(QB_LOG_BLACKBOX, QB_LOG_CONF_STATE_GET, 0) != QB_LOG_STATE_ENABLED) { qb_log_ctl(QB_LOG_BLACKBOX, QB_LOG_CONF_SIZE, 5 * 1024 * 1024); /* Any size change drops existing entries */ qb_log_ctl(QB_LOG_BLACKBOX, QB_LOG_CONF_ENABLED, QB_TRUE); /* Setting the size seems to disable it */ /* Enable synchronous logging */ for (lpc = QB_LOG_BLACKBOX; lpc < QB_LOG_TARGET_MAX; lpc++) { qb_log_ctl(lpc, QB_LOG_CONF_FILE_SYNC, QB_TRUE); } crm_notice("Initiated blackbox recorder: %s", blackbox_file_prefix); /* Save to disk on abnormal termination */ crm_signal_handler(SIGSEGV, crm_trigger_blackbox); crm_signal_handler(SIGABRT, crm_trigger_blackbox); crm_signal_handler(SIGILL, crm_trigger_blackbox); crm_signal_handler(SIGBUS, crm_trigger_blackbox); crm_signal_handler(SIGFPE, crm_trigger_blackbox); crm_update_callsites(); blackbox_trigger = qb_log_custom_open(blackbox_logger, NULL, NULL, NULL); qb_log_ctl(blackbox_trigger, QB_LOG_CONF_ENABLED, QB_TRUE); crm_trace("Trigger: %d is %d %d", blackbox_trigger, qb_log_ctl(blackbox_trigger, QB_LOG_CONF_STATE_GET, 0), QB_LOG_STATE_ENABLED); crm_update_callsites(); } else if (!enable && qb_log_ctl(QB_LOG_BLACKBOX, QB_LOG_CONF_STATE_GET, 0) == QB_LOG_STATE_ENABLED) { qb_log_ctl(QB_LOG_BLACKBOX, QB_LOG_CONF_ENABLED, QB_FALSE); /* Disable synchronous logging again when the blackbox is disabled */ for (lpc = QB_LOG_BLACKBOX; lpc < QB_LOG_TARGET_MAX; lpc++) { qb_log_ctl(lpc, QB_LOG_CONF_FILE_SYNC, QB_FALSE); } } } void crm_enable_blackbox(int nsig) { crm_control_blackbox(nsig, TRUE); } void crm_disable_blackbox(int nsig) { crm_control_blackbox(nsig, FALSE); } /*! * \internal * \brief Write out a blackbox, if blackboxes are enabled * * \param[in] nsig Signal that was received * \param[in] cs libqb callsite * * \note This may be called via a true signal handler and so must be async-safe. * @TODO actually make this async-safe */ void crm_write_blackbox(int nsig, const struct qb_log_callsite *cs) { static volatile int counter = 1; static volatile time_t last = 0; char *buffer = NULL; int rc = 0; time_t now = time(NULL); if (blackbox_file_prefix == NULL) { return; } switch (nsig) { case 0: case SIGTRAP: /* The graceful case - such as assertion failure or user request */ if (nsig == 0 && now == last) { /* Prevent over-dumping */ return; } buffer = crm_strdup_printf("%s.%d", blackbox_file_prefix, counter++); if (nsig == SIGTRAP) { crm_notice("Blackbox dump requested, please see %s for contents", buffer); } else if (cs) { syslog(LOG_NOTICE, "Problem detected at %s:%d (%s), please see %s for additional details", cs->function, cs->lineno, cs->filename, buffer); } else { crm_notice("Problem detected, please see %s for additional details", buffer); } last = now; rc = qb_log_blackbox_write_to_file(buffer); if (rc < 0) { // System errno crm_err("Failed to write blackbox file %s: %s", buffer, strerror(-rc)); } /* Flush the existing contents * A size change would also work */ qb_log_ctl(QB_LOG_BLACKBOX, QB_LOG_CONF_ENABLED, QB_FALSE); qb_log_ctl(QB_LOG_BLACKBOX, QB_LOG_CONF_ENABLED, QB_TRUE); break; default: /* Do as little as possible, just try to get what we have out * We logged the filename when the blackbox was enabled */ crm_signal_handler(nsig, SIG_DFL); qb_log_blackbox_write_to_file((const char *)blackbox_file_prefix); qb_log_ctl(QB_LOG_BLACKBOX, QB_LOG_CONF_ENABLED, QB_FALSE); raise(nsig); break; } free(buffer); } static const char * crm_quark_to_string(uint32_t tag) { const char *text = g_quark_to_string(tag); if (text) { return text; } return ""; } static void crm_log_filter_source(int source, const char *trace_files, const char *trace_fns, const char *trace_fmts, const char *trace_tags, const char *trace_blackbox, struct qb_log_callsite *cs) { if (qb_log_ctl(source, QB_LOG_CONF_STATE_GET, 0) != QB_LOG_STATE_ENABLED) { return; } else if (cs->tags != crm_trace_nonlog && source == QB_LOG_BLACKBOX) { /* Blackbox gets everything if enabled */ qb_bit_set(cs->targets, source); } else if (source == blackbox_trigger && blackbox_trigger > 0) { /* Should this log message result in the blackbox being dumped */ if (cs->priority <= LOG_ERR) { qb_bit_set(cs->targets, source); } else if (trace_blackbox) { char *key = crm_strdup_printf("%s:%d", cs->function, cs->lineno); if (strstr(trace_blackbox, key) != NULL) { qb_bit_set(cs->targets, source); } free(key); } } else if (source == QB_LOG_SYSLOG) { /* No tracing to syslog */ if (cs->priority <= crm_log_priority && cs->priority <= crm_log_level) { qb_bit_set(cs->targets, source); } /* Log file tracing options... */ } else if (cs->priority <= crm_log_level) { qb_bit_set(cs->targets, source); } else if (trace_files && strstr(trace_files, cs->filename) != NULL) { qb_bit_set(cs->targets, source); } else if (trace_fns && strstr(trace_fns, cs->function) != NULL) { qb_bit_set(cs->targets, source); } else if (trace_fmts && strstr(trace_fmts, cs->format) != NULL) { qb_bit_set(cs->targets, source); } else if (trace_tags && cs->tags != 0 && cs->tags != crm_trace_nonlog && g_quark_to_string(cs->tags) != NULL) { qb_bit_set(cs->targets, source); } } #ifndef HAVE_STRCHRNUL /* strchrnul() is a GNU extension. If not present, use our own definition. * The GNU version returns char*, but we only need it to be const char*. */ static const char * strchrnul(const char *s, int c) { while ((*s != c) && (*s != '\0')) { ++s; } return s; } #endif static void crm_log_filter(struct qb_log_callsite *cs) { int lpc = 0; static int need_init = 1; static const char *trace_fns = NULL; static const char *trace_tags = NULL; static const char *trace_fmts = NULL; static const char *trace_files = NULL; static const char *trace_blackbox = NULL; if (need_init) { need_init = 0; trace_fns = pcmk__env_option(PCMK__ENV_TRACE_FUNCTIONS); trace_fmts = pcmk__env_option(PCMK__ENV_TRACE_FORMATS); trace_tags = pcmk__env_option(PCMK__ENV_TRACE_TAGS); trace_files = pcmk__env_option(PCMK__ENV_TRACE_FILES); trace_blackbox = pcmk__env_option(PCMK__ENV_TRACE_BLACKBOX); if (trace_tags != NULL) { uint32_t tag; const char *offset = NULL; const char *next = trace_tags; // @TODO Use g_strsplit() to simplify do { char *token = NULL; offset = next; next = strchrnul(offset, ','); token = crm_strdup_printf("%.*s", (int) (next - offset), offset); tag = g_quark_from_string(token); crm_info("Created GQuark %u from token '%s' in '%s'", tag, token, trace_tags); free(token); if (next[0] != 0) { next++; } } while (next != NULL && next[0] != 0); } } cs->targets = 0; /* Reset then find targets to enable */ for (lpc = QB_LOG_SYSLOG; lpc < QB_LOG_TARGET_MAX; lpc++) { crm_log_filter_source(lpc, trace_files, trace_fns, trace_fmts, trace_tags, trace_blackbox, cs); } } gboolean crm_is_callsite_active(struct qb_log_callsite *cs, uint8_t level, uint32_t tags) { gboolean refilter = FALSE; if (cs == NULL) { return FALSE; } if (cs->priority != level) { cs->priority = level; refilter = TRUE; } if (cs->tags != tags) { cs->tags = tags; refilter = TRUE; } if (refilter) { crm_log_filter(cs); } if (cs->targets == 0) { return FALSE; } return TRUE; } void crm_update_callsites(void) { static gboolean log = TRUE; if (log) { log = FALSE; crm_debug ("Enabling callsites based on priority=%d, files=%s, functions=%s, formats=%s, tags=%s", crm_log_level, pcmk__env_option(PCMK__ENV_TRACE_FILES), pcmk__env_option(PCMK__ENV_TRACE_FUNCTIONS), pcmk__env_option(PCMK__ENV_TRACE_FORMATS), pcmk__env_option(PCMK__ENV_TRACE_TAGS)); } qb_log_filter_fn_set(crm_log_filter); } static gboolean crm_tracing_enabled(void) { return (crm_log_level == LOG_TRACE) || (pcmk__env_option(PCMK__ENV_TRACE_FILES) != NULL) || (pcmk__env_option(PCMK__ENV_TRACE_FUNCTIONS) != NULL) || (pcmk__env_option(PCMK__ENV_TRACE_FORMATS) != NULL) || (pcmk__env_option(PCMK__ENV_TRACE_TAGS) != NULL); } static int crm_priority2int(const char *name) { struct syslog_names { const char *name; int priority; }; static struct syslog_names p_names[] = { {"emerg", LOG_EMERG}, {"alert", LOG_ALERT}, {"crit", LOG_CRIT}, {"error", LOG_ERR}, {"warning", LOG_WARNING}, {"notice", LOG_NOTICE}, {"info", LOG_INFO}, {"debug", LOG_DEBUG}, {NULL, -1} }; int lpc; for (lpc = 0; name != NULL && p_names[lpc].name != NULL; lpc++) { if (pcmk__str_eq(p_names[lpc].name, name, pcmk__str_none)) { return p_names[lpc].priority; } } return crm_log_priority; } /*! * \internal * \brief Set the identifier for the current process * * If the identifier crm_system_name is not already set, then it is set as follows: * - it is passed to the function via the "entity" parameter, or * - it is derived from the executable name * * The identifier can be used in logs, IPC, and more. * * This method also sets the PCMK_service environment variable. * * \param[in] entity If not NULL, will be assigned to the identifier * \param[in] argc The number of command line parameters * \param[in] argv The command line parameter values */ static void set_identity(const char *entity, int argc, char *const *argv) { if (crm_system_name != NULL) { return; // Already set, don't overwrite } if (entity != NULL) { crm_system_name = pcmk__str_copy(entity); } else if ((argc > 0) && (argv != NULL)) { char *mutable = strdup(argv[0]); char *modified = basename(mutable); if (strstr(modified, "lt-") == modified) { modified += 3; } crm_system_name = pcmk__str_copy(modified); free(mutable); } else { crm_system_name = pcmk__str_copy("Unknown"); } // Used by fencing.py.py (in fence-agents) pcmk__set_env_option(PCMK__ENV_SERVICE, crm_system_name, false); } void crm_log_preinit(const char *entity, int argc, char *const *argv) { /* Configure libqb logging with nothing turned on */ struct utsname res; int lpc = 0; int32_t qb_facility = 0; pid_t pid = getpid(); const char *nodename = "localhost"; static bool have_logging = false; GLogLevelFlags log_levels; if (have_logging) { return; } have_logging = true; /* @TODO Try to create a more obvious "global Pacemaker initializer" * function than crm_log_preinit(), and call pcmk__schema_init() there. * See also https://projects.clusterlabs.org/T840. */ pcmk__schema_init(); if (crm_trace_nonlog == 0) { crm_trace_nonlog = g_quark_from_static_string("Pacemaker non-logging tracepoint"); } umask(S_IWGRP | S_IWOTH | S_IROTH); /* Add a log handler for messages from our log domain at any log level. */ log_levels = G_LOG_LEVEL_MASK | G_LOG_FLAG_FATAL | G_LOG_FLAG_RECURSION; pcmk__log_id = g_log_set_handler(G_LOG_DOMAIN, log_levels, crm_glib_handler, NULL); /* Add a log handler for messages from the GLib domains at any log level. */ pcmk__glib_log_id = g_log_set_handler("GLib", log_levels, crm_glib_handler, NULL); pcmk__gio_log_id = g_log_set_handler("GLib-GIO", log_levels, crm_glib_handler, NULL); pcmk__gmodule_log_id = g_log_set_handler("GModule", log_levels, crm_glib_handler, NULL); pcmk__gthread_log_id = g_log_set_handler("GThread", log_levels, crm_glib_handler, NULL); /* glib should not abort for any messages from the Pacemaker domain, but * other domains are still free to specify their own behavior. However, * note that G_LOG_LEVEL_ERROR is always fatal regardless of what we do * here. */ g_log_set_fatal_mask(G_LOG_DOMAIN, 0); /* Set crm_system_name, which is used as the logging name. It may also * be used for other purposes such as an IPC client name. */ set_identity(entity, argc, argv); qb_facility = qb_log_facility2int("local0"); qb_log_init(crm_system_name, qb_facility, LOG_ERR); crm_log_level = LOG_CRIT; /* Nuke any syslog activity until it's asked for */ qb_log_ctl(QB_LOG_SYSLOG, QB_LOG_CONF_ENABLED, QB_FALSE); #ifdef HAVE_qb_log_conf_QB_LOG_CONF_MAX_LINE_LEN // Shorter than default, generous for what we *should* send to syslog qb_log_ctl(QB_LOG_SYSLOG, QB_LOG_CONF_MAX_LINE_LEN, 256); #endif if (uname(memset(&res, 0, sizeof(res))) == 0 && *res.nodename != '\0') { nodename = res.nodename; } /* Set format strings and disable threading * Pacemaker and threads do not mix well (due to the amount of forking) */ qb_log_tags_stringify_fn_set(crm_quark_to_string); for (lpc = QB_LOG_SYSLOG; lpc < QB_LOG_TARGET_MAX; lpc++) { qb_log_ctl(lpc, QB_LOG_CONF_THREADED, QB_FALSE); #ifdef HAVE_qb_log_conf_QB_LOG_CONF_ELLIPSIS // End truncated lines with '...' qb_log_ctl(lpc, QB_LOG_CONF_ELLIPSIS, QB_TRUE); #endif set_format_string(lpc, crm_system_name, pid, nodename); } #ifdef ENABLE_NLS /* Enable translations (experimental). Currently we only have a few * proof-of-concept translations for some option help. The goal would be to * offer translations for option help and man pages rather than logs or * documentation, to reduce the burden of maintaining them. */ // Load locale information for the local host from the environment setlocale(LC_ALL, ""); // Tell gettext where to find Pacemaker message catalogs pcmk__assert(bindtextdomain(PACKAGE, PCMK__LOCALE_DIR) != NULL); // Tell gettext to use the Pacemaker message catalogs pcmk__assert(textdomain(PACKAGE) != NULL); // Tell gettext that the translated strings are stored in UTF-8 bind_textdomain_codeset(PACKAGE, "UTF-8"); #endif } gboolean crm_log_init(const char *entity, uint8_t level, gboolean daemon, gboolean to_stderr, int argc, char **argv, gboolean quiet) { const char *syslog_priority = NULL; const char *facility = pcmk__env_option(PCMK__ENV_LOGFACILITY); const char *f_copy = facility; pcmk__is_daemon = daemon; crm_log_preinit(entity, argc, argv); if (level > LOG_TRACE) { level = LOG_TRACE; } if(level > crm_log_level) { crm_log_level = level; } /* Should we log to syslog */ if (facility == NULL) { if (pcmk__is_daemon) { facility = "daemon"; } else { facility = PCMK_VALUE_NONE; } pcmk__set_env_option(PCMK__ENV_LOGFACILITY, facility, true); } if (pcmk__str_eq(facility, PCMK_VALUE_NONE, pcmk__str_casei)) { quiet = TRUE; } else { qb_log_ctl(QB_LOG_SYSLOG, QB_LOG_CONF_FACILITY, qb_log_facility2int(facility)); } if (pcmk__env_option_enabled(crm_system_name, PCMK__ENV_DEBUG)) { /* Override the default setting */ crm_log_level = LOG_DEBUG; } /* What lower threshold do we have for sending to syslog */ syslog_priority = pcmk__env_option(PCMK__ENV_LOGPRIORITY); if (syslog_priority) { crm_log_priority = crm_priority2int(syslog_priority); } qb_log_filter_ctl(QB_LOG_SYSLOG, QB_LOG_FILTER_ADD, QB_LOG_FILTER_FILE, "*", crm_log_priority); // Log to syslog unless requested to be quiet if (!quiet) { qb_log_ctl(QB_LOG_SYSLOG, QB_LOG_CONF_ENABLED, QB_TRUE); } /* Should we log to stderr */ if (pcmk__env_option_enabled(crm_system_name, PCMK__ENV_STDERR)) { /* Override the default setting */ to_stderr = TRUE; } crm_enable_stderr(to_stderr); // Log to a file if we're a daemon or user asked for one { const char *logfile = pcmk__env_option(PCMK__ENV_LOGFILE); if (!pcmk__str_eq(PCMK_VALUE_NONE, logfile, pcmk__str_casei) && (pcmk__is_daemon || (logfile != NULL))) { // Daemons always get a log file, unless explicitly set to "none" pcmk__add_logfile(logfile); } } if (pcmk__is_daemon && pcmk__env_option_enabled(crm_system_name, PCMK__ENV_BLACKBOX)) { crm_enable_blackbox(0); } /* Summary */ crm_trace("Quiet: %d, facility %s", quiet, f_copy); pcmk__env_option(PCMK__ENV_LOGFILE); pcmk__env_option(PCMK__ENV_LOGFACILITY); crm_update_callsites(); /* Ok, now we can start logging... */ // Disable daemon request if user isn't root or Pacemaker daemon user if (pcmk__is_daemon) { const char *user = getenv("USER"); if (user != NULL && !pcmk__strcase_any_of(user, "root", CRM_DAEMON_USER, NULL)) { crm_trace("Not switching to corefile directory for %s", user); pcmk__is_daemon = false; } } if (pcmk__is_daemon) { - int user = getuid(); - struct passwd *pwent = getpwuid(user); + char *user = pcmk__uid2username(getuid()); - if (pwent == NULL) { - crm_perror(LOG_ERR, "Cannot get name for uid: %d", user); + if (user == NULL) { + // Error already logged - } else if (!pcmk__strcase_any_of(pwent->pw_name, "root", CRM_DAEMON_USER, NULL)) { - crm_trace("Don't change active directory for regular user: %s", pwent->pw_name); + } else if (!pcmk__str_any_of(user, "root", CRM_DAEMON_USER, NULL)) { + crm_trace("Don't change active directory for regular user %s", + user); } else if (chdir(CRM_CORE_DIR) < 0) { - crm_perror(LOG_INFO, "Cannot change active directory to " CRM_CORE_DIR); + crm_info("Cannot change active directory to " CRM_CORE_DIR ": %s", + strerror(errno)); } else { crm_info("Changed active directory to " CRM_CORE_DIR); } /* Original meanings from signal(7) * * Signal Value Action Comment * SIGTRAP 5 Core Trace/breakpoint trap * SIGUSR1 30,10,16 Term User-defined signal 1 * SIGUSR2 31,12,17 Term User-defined signal 2 * * Our usage is as similar as possible */ mainloop_add_signal(SIGUSR1, crm_enable_blackbox); mainloop_add_signal(SIGUSR2, crm_disable_blackbox); mainloop_add_signal(SIGTRAP, crm_trigger_blackbox); + free(user); + } else if (!quiet) { crm_log_args(argc, argv); } return TRUE; } /* returns the old value */ unsigned int set_crm_log_level(unsigned int level) { unsigned int old = crm_log_level; if (level > LOG_TRACE) { level = LOG_TRACE; } crm_log_level = level; crm_update_callsites(); crm_trace("New log level: %d", level); return old; } void crm_enable_stderr(int enable) { if (enable && qb_log_ctl(QB_LOG_STDERR, QB_LOG_CONF_STATE_GET, 0) != QB_LOG_STATE_ENABLED) { qb_log_ctl(QB_LOG_STDERR, QB_LOG_CONF_ENABLED, QB_TRUE); crm_update_callsites(); } else if (enable == FALSE) { qb_log_ctl(QB_LOG_STDERR, QB_LOG_CONF_ENABLED, QB_FALSE); } } /*! * \brief Make logging more verbose * * If logging to stderr is not already enabled when this function is called, * enable it. Otherwise, increase the log level by 1. * * \param[in] argc Ignored * \param[in] argv Ignored */ void crm_bump_log_level(int argc, char **argv) { if (qb_log_ctl(QB_LOG_STDERR, QB_LOG_CONF_STATE_GET, 0) != QB_LOG_STATE_ENABLED) { crm_enable_stderr(TRUE); } else { set_crm_log_level(crm_log_level + 1); } } unsigned int get_crm_log_level(void) { return crm_log_level; } /*! * \brief Log the command line (once) * * \param[in] Number of values in \p argv * \param[in] Command-line arguments (including command name) * * \note This function will only log once, even if called with different * arguments. */ void crm_log_args(int argc, char **argv) { static bool logged = false; gchar *arg_string = NULL; if ((argc == 0) || (argv == NULL) || logged) { return; } logged = true; arg_string = g_strjoinv(" ", argv); crm_notice("Invoked: %s", arg_string); g_free(arg_string); } void crm_log_output_fn(const char *file, const char *function, int line, int level, const char *prefix, const char *output) { const char *next = NULL; const char *offset = NULL; if (level == LOG_NEVER) { return; } if (output == NULL) { if (level != LOG_STDOUT) { level = LOG_TRACE; } output = "-- empty --"; } next = output; do { offset = next; next = strchrnul(offset, '\n'); do_crm_log_alias(level, file, function, line, "%s [ %.*s ]", prefix, (int)(next - offset), offset); if (next[0] != 0) { next++; } } while (next != NULL && next[0] != 0); } void pcmk__cli_init_logging(const char *name, unsigned int verbosity) { crm_log_init(name, LOG_ERR, FALSE, FALSE, 0, NULL, TRUE); for (int i = 0; i < verbosity; i++) { /* These arguments are ignored, so pass placeholders. */ crm_bump_log_level(0, NULL); } } /*! * \brief Log XML line-by-line in a formatted fashion * * \param[in] file File name to use for log filtering * \param[in] function Function name to use for log filtering * \param[in] line Line number to use for log filtering * \param[in] tags Logging tags to use for log filtering * \param[in] level Priority at which to log the messages * \param[in] text Prefix for each line * \param[in] xml XML to log * * \note This does nothing when \p level is \p LOG_STDOUT. * \note Do not call this function directly. It should be called only from the * \p do_crm_log_xml() macro. */ void pcmk_log_xml_as(const char *file, const char *function, uint32_t line, uint32_t tags, uint8_t level, const char *text, const xmlNode *xml) { if (xml == NULL) { do_crm_log(level, "%s%sNo data to dump as XML", pcmk__s(text, ""), pcmk__str_empty(text)? "" : " "); } else { if (logger_out == NULL) { CRM_CHECK(pcmk__log_output_new(&logger_out) == pcmk_rc_ok, return); } pcmk__output_set_log_level(logger_out, level); pcmk__output_set_log_filter(logger_out, file, function, line, tags); pcmk__xml_show(logger_out, text, xml, 1, pcmk__xml_fmt_pretty |pcmk__xml_fmt_open |pcmk__xml_fmt_children |pcmk__xml_fmt_close); pcmk__output_set_log_filter(logger_out, NULL, NULL, 0U, 0U); } } /*! * \internal * \brief Log XML changes line-by-line in a formatted fashion * * \param[in] file File name to use for log filtering * \param[in] function Function name to use for log filtering * \param[in] line Line number to use for log filtering * \param[in] tags Logging tags to use for log filtering * \param[in] level Priority at which to log the messages * \param[in] xml XML whose changes to log * * \note This does nothing when \p level is \c LOG_STDOUT. */ void pcmk__log_xml_changes_as(const char *file, const char *function, uint32_t line, uint32_t tags, uint8_t level, const xmlNode *xml) { if (xml == NULL) { do_crm_log(level, "No XML to dump"); return; } if (logger_out == NULL) { CRM_CHECK(pcmk__log_output_new(&logger_out) == pcmk_rc_ok, return); } pcmk__output_set_log_level(logger_out, level); pcmk__output_set_log_filter(logger_out, file, function, line, tags); pcmk__xml_show_changes(logger_out, xml); pcmk__output_set_log_filter(logger_out, NULL, NULL, 0U, 0U); } /*! * \internal * \brief Log an XML patchset line-by-line in a formatted fashion * * \param[in] file File name to use for log filtering * \param[in] function Function name to use for log filtering * \param[in] line Line number to use for log filtering * \param[in] tags Logging tags to use for log filtering * \param[in] level Priority at which to log the messages * \param[in] patchset XML patchset to log * * \note This does nothing when \p level is \c LOG_STDOUT. */ void pcmk__log_xml_patchset_as(const char *file, const char *function, uint32_t line, uint32_t tags, uint8_t level, const xmlNode *patchset) { if (patchset == NULL) { do_crm_log(level, "No patchset to dump"); return; } if (logger_out == NULL) { CRM_CHECK(pcmk__log_output_new(&logger_out) == pcmk_rc_ok, return); } pcmk__output_set_log_level(logger_out, level); pcmk__output_set_log_filter(logger_out, file, function, line, tags); logger_out->message(logger_out, "xml-patchset", patchset); pcmk__output_set_log_filter(logger_out, NULL, NULL, 0U, 0U); } /*! * \internal * \brief Free the logging library's internal log output object */ void pcmk__free_common_logger(void) { if (logger_out != NULL) { logger_out->finish(logger_out, CRM_EX_OK, true, NULL); pcmk__output_free(logger_out); logger_out = NULL; } } void pcmk__set_config_error_handler(pcmk__config_error_func error_handler, void *error_context) { pcmk__config_error_handler = error_handler; pcmk__config_error_context = error_context; } void pcmk__set_config_warning_handler(pcmk__config_warning_func warning_handler, void *warning_context) { pcmk__config_warning_handler = warning_handler; pcmk__config_warning_context = warning_context; } diff --git a/lib/common/mainloop.c b/lib/common/mainloop.c index fc115e7dff..91322ee754 100644 --- a/lib/common/mainloop.c +++ b/lib/common/mainloop.c @@ -1,1471 +1,1471 @@ /* * Copyright 2004-2025 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 #include struct mainloop_child_s { pid_t pid; char *desc; unsigned timerid; gboolean timeout; void *privatedata; enum mainloop_child_flags flags; /* Called when a process dies */ void (*callback) (mainloop_child_t * p, pid_t pid, int core, int signo, int exitcode); }; struct trigger_s { GSource source; gboolean running; gboolean trigger; void *user_data; guint id; }; struct mainloop_timer_s { guint id; guint period_ms; bool repeat; char *name; GSourceFunc cb; void *userdata; }; static gboolean crm_trigger_prepare(GSource * source, gint * timeout) { crm_trigger_t *trig = (crm_trigger_t *) source; /* cluster-glue's FD and IPC related sources make use of * g_source_add_poll() but do not set a timeout in their prepare * functions * * This means mainloop's poll() will block until an event for one * of these sources occurs - any /other/ type of source, such as * this one or g_idle_*, that doesn't use g_source_add_poll() is * S-O-L and won't be processed until there is something fd-based * happens. * * Luckily the timeout we can set here affects all sources and * puts an upper limit on how long poll() can take. * * So unconditionally set a small-ish timeout, not too small that * we're in constant motion, which will act as an upper bound on * how long the signal handling might be delayed for. */ *timeout = 500; /* Timeout in ms */ return trig->trigger; } static gboolean crm_trigger_check(GSource * source) { crm_trigger_t *trig = (crm_trigger_t *) source; return trig->trigger; } /*! * \internal * \brief GSource dispatch function for crm_trigger_t * * \param[in] source crm_trigger_t being dispatched * \param[in] callback Callback passed at source creation * \param[in,out] userdata User data passed at source creation * * \return G_SOURCE_REMOVE to remove source, G_SOURCE_CONTINUE to keep it */ static gboolean crm_trigger_dispatch(GSource *source, GSourceFunc callback, gpointer userdata) { gboolean rc = G_SOURCE_CONTINUE; crm_trigger_t *trig = (crm_trigger_t *) source; if (trig->running) { /* Wait until the existing job is complete before starting the next one */ return G_SOURCE_CONTINUE; } trig->trigger = FALSE; if (callback) { int callback_rc = callback(trig->user_data); if (callback_rc < 0) { crm_trace("Trigger handler %p not yet complete", trig); trig->running = TRUE; } else if (callback_rc == 0) { rc = G_SOURCE_REMOVE; } } return rc; } static void crm_trigger_finalize(GSource * source) { crm_trace("Trigger %p destroyed", source); } static GSourceFuncs crm_trigger_funcs = { crm_trigger_prepare, crm_trigger_check, crm_trigger_dispatch, crm_trigger_finalize, }; static crm_trigger_t * mainloop_setup_trigger(GSource * source, int priority, int (*dispatch) (gpointer user_data), gpointer userdata) { crm_trigger_t *trigger = NULL; trigger = (crm_trigger_t *) source; trigger->id = 0; trigger->trigger = FALSE; trigger->user_data = userdata; if (dispatch) { g_source_set_callback(source, dispatch, trigger, NULL); } g_source_set_priority(source, priority); g_source_set_can_recurse(source, FALSE); trigger->id = g_source_attach(source, NULL); return trigger; } void mainloop_trigger_complete(crm_trigger_t * trig) { crm_trace("Trigger handler %p complete", trig); trig->running = FALSE; } /*! * \brief Create a trigger to be used as a mainloop source * * \param[in] priority Relative priority of source (lower number is higher priority) * \param[in] dispatch Trigger dispatch function (should return 0 to remove the * trigger from the mainloop, -1 if the trigger should be * kept but the job is still running and not complete, and * 1 if the trigger should be kept and the job is complete) * \param[in] userdata Pointer to pass to \p dispatch * * \return Newly allocated mainloop source for trigger */ crm_trigger_t * mainloop_add_trigger(int priority, int (*dispatch) (gpointer user_data), gpointer userdata) { GSource *source = NULL; pcmk__assert(sizeof(crm_trigger_t) > sizeof(GSource)); source = g_source_new(&crm_trigger_funcs, sizeof(crm_trigger_t)); return mainloop_setup_trigger(source, priority, dispatch, userdata); } void mainloop_set_trigger(crm_trigger_t * source) { if(source) { source->trigger = TRUE; } } gboolean mainloop_destroy_trigger(crm_trigger_t * source) { GSource *gs = NULL; if(source == NULL) { return TRUE; } gs = (GSource *)source; g_source_destroy(gs); /* Remove from mainloop, ref_count-- */ g_source_unref(gs); /* The caller no longer carries a reference to source * * At this point the source should be free'd, * unless we're currently processing said * source, in which case mainloop holds an * additional reference and it will be free'd * once our processing completes */ return TRUE; } // Define a custom glib source for signal handling // Data structure for custom glib source typedef struct signal_s { crm_trigger_t trigger; // trigger that invoked source (must be first) void (*handler) (int sig); // signal handler int signal; // signal that was received } crm_signal_t; // Table to associate signal handlers with signal numbers static crm_signal_t *crm_signals[NSIG]; /*! * \internal * \brief Dispatch an event from custom glib source for signals * * Given an signal event, clear the event trigger and call any registered * signal handler. * * \param[in] source glib source that triggered this dispatch * \param[in] callback (ignored) * \param[in] userdata (ignored) */ static gboolean crm_signal_dispatch(GSource *source, GSourceFunc callback, gpointer userdata) { crm_signal_t *sig = (crm_signal_t *) source; if(sig->signal != SIGCHLD) { crm_notice("Caught '%s' signal " QB_XS " %d (%s handler)", strsignal(sig->signal), sig->signal, (sig->handler? "invoking" : "no")); } sig->trigger.trigger = FALSE; if (sig->handler) { sig->handler(sig->signal); } return TRUE; } /*! * \internal * \brief Handle a signal by setting a trigger for signal source * * \param[in] sig Signal number that was received * * \note This is the true signal handler for the mainloop signal source, and * must be async-safe. */ static void mainloop_signal_handler(int sig) { if (sig > 0 && sig < NSIG && crm_signals[sig] != NULL) { mainloop_set_trigger((crm_trigger_t *) crm_signals[sig]); } } // Functions implementing our custom glib source for signal handling static GSourceFuncs crm_signal_funcs = { crm_trigger_prepare, crm_trigger_check, crm_signal_dispatch, crm_trigger_finalize, }; /*! * \internal * \brief Set a true signal handler * * signal()-like interface to sigaction() * * \param[in] sig Signal number to register handler for * \param[in] dispatch Signal handler * * \return The previous value of the signal handler, or SIG_ERR on error * \note The dispatch function must be async-safe. */ sighandler_t crm_signal_handler(int sig, sighandler_t dispatch) { sigset_t mask; struct sigaction sa; struct sigaction old; if (sigemptyset(&mask) < 0) { - crm_err("Could not set handler for signal %d: %s", - sig, pcmk_rc_str(errno)); + crm_err("Could not %sset handler for signal %d: %s", + ((dispatch == NULL)? "un" : ""), sig, strerror(errno)); return SIG_ERR; } memset(&sa, 0, sizeof(struct sigaction)); sa.sa_handler = dispatch; sa.sa_flags = SA_RESTART; sa.sa_mask = mask; if (sigaction(sig, &sa, &old) < 0) { - crm_err("Could not set handler for signal %d: %s", - sig, pcmk_rc_str(errno)); + crm_err("Could not %sset handler for signal %d: %s", + ((dispatch == NULL)? "un" : ""), sig, strerror(errno)); return SIG_ERR; } return old.sa_handler; } static void mainloop_destroy_signal_entry(int sig) { crm_signal_t *tmp = crm_signals[sig]; if (tmp != NULL) { crm_signals[sig] = NULL; crm_trace("Unregistering mainloop handler for signal %d", sig); mainloop_destroy_trigger((crm_trigger_t *) tmp); } } /*! * \internal * \brief Add a signal handler to a mainloop * * \param[in] sig Signal number to handle * \param[in] dispatch Signal handler function * * \note The true signal handler merely sets a mainloop trigger to call this * dispatch function via the mainloop. Therefore, the dispatch function * does not need to be async-safe. */ gboolean mainloop_add_signal(int sig, void (*dispatch) (int sig)) { GSource *source = NULL; int priority = G_PRIORITY_HIGH - 1; if (sig == SIGTERM) { /* TERM is higher priority than other signals, * signals are higher priority than other ipc. * Yes, minus: smaller is "higher" */ priority--; } if (sig >= NSIG || sig < 0) { crm_err("Signal %d is out of range", sig); return FALSE; } else if (crm_signals[sig] != NULL && crm_signals[sig]->handler == dispatch) { crm_trace("Signal handler for %d is already installed", sig); return TRUE; } else if (crm_signals[sig] != NULL) { crm_err("Different signal handler for %d is already installed", sig); return FALSE; } pcmk__assert(sizeof(crm_signal_t) > sizeof(GSource)); source = g_source_new(&crm_signal_funcs, sizeof(crm_signal_t)); crm_signals[sig] = (crm_signal_t *) mainloop_setup_trigger(source, priority, NULL, NULL); pcmk__assert(crm_signals[sig] != NULL); crm_signals[sig]->handler = dispatch; crm_signals[sig]->signal = sig; if (crm_signal_handler(sig, mainloop_signal_handler) == SIG_ERR) { mainloop_destroy_signal_entry(sig); return FALSE; } return TRUE; } gboolean mainloop_destroy_signal(int sig) { if (sig >= NSIG || sig < 0) { crm_err("Signal %d is out of range", sig); return FALSE; } else if (crm_signal_handler(sig, NULL) == SIG_ERR) { - crm_perror(LOG_ERR, "Could not uninstall signal handler for signal %d", sig); + // Error already logged return FALSE; } else if (crm_signals[sig] == NULL) { return TRUE; } mainloop_destroy_signal_entry(sig); return TRUE; } static qb_array_t *gio_map = NULL; void mainloop_cleanup(void) { if (gio_map != NULL) { qb_array_free(gio_map); gio_map = NULL; } for (int sig = 0; sig < NSIG; ++sig) { mainloop_destroy_signal_entry(sig); } } /* * libqb... */ struct gio_to_qb_poll { int32_t is_used; guint source; int32_t events; void *data; qb_ipcs_dispatch_fn_t fn; enum qb_loop_priority p; }; static gboolean gio_read_socket(GIOChannel * gio, GIOCondition condition, gpointer data) { struct gio_to_qb_poll *adaptor = (struct gio_to_qb_poll *)data; gint fd = g_io_channel_unix_get_fd(gio); crm_trace("%p.%d %d", data, fd, condition); /* if this assert get's hit, then there is a race condition between * when we destroy a fd and when mainloop actually gives it up */ pcmk__assert(adaptor->is_used > 0); return (adaptor->fn(fd, condition, adaptor->data) == 0); } static void gio_poll_destroy(gpointer data) { struct gio_to_qb_poll *adaptor = (struct gio_to_qb_poll *)data; adaptor->is_used--; pcmk__assert(adaptor->is_used >= 0); if (adaptor->is_used == 0) { crm_trace("Marking adaptor %p unused", adaptor); adaptor->source = 0; } } /*! * \internal * \brief Convert libqb's poll priority into GLib's one * * \param[in] prio libqb's poll priority (#QB_LOOP_MED assumed as fallback) * * \return best matching GLib's priority */ static gint conv_prio_libqb2glib(enum qb_loop_priority prio) { switch (prio) { case QB_LOOP_LOW: return G_PRIORITY_LOW; case QB_LOOP_HIGH: return G_PRIORITY_HIGH; default: return G_PRIORITY_DEFAULT; // QB_LOOP_MED } } /*! * \internal * \brief Convert libqb's poll priority to rate limiting spec * * \param[in] prio libqb's poll priority (#QB_LOOP_MED assumed as fallback) * * \return best matching rate limiting spec * \note This is the inverse of libqb's qb_ipcs_request_rate_limit(). */ static enum qb_ipcs_rate_limit conv_libqb_prio2ratelimit(enum qb_loop_priority prio) { switch (prio) { case QB_LOOP_LOW: return QB_IPCS_RATE_SLOW; case QB_LOOP_HIGH: return QB_IPCS_RATE_FAST; default: return QB_IPCS_RATE_NORMAL; // QB_LOOP_MED } } static int32_t gio_poll_dispatch_update(enum qb_loop_priority p, int32_t fd, int32_t evts, void *data, qb_ipcs_dispatch_fn_t fn, int32_t add) { struct gio_to_qb_poll *adaptor; GIOChannel *channel; int32_t res = 0; res = qb_array_index(gio_map, fd, (void **)&adaptor); if (res < 0) { crm_err("Array lookup failed for fd=%d: %d", fd, res); return res; } crm_trace("Adding fd=%d to mainloop as adaptor %p", fd, adaptor); if (add && adaptor->source) { crm_err("Adaptor for descriptor %d is still in-use", fd); return -EEXIST; } if (!add && !adaptor->is_used) { crm_err("Adaptor for descriptor %d is not in-use", fd); return -ENOENT; } /* channel is created with ref_count = 1 */ channel = g_io_channel_unix_new(fd); if (!channel) { crm_err("No memory left to add fd=%d", fd); return -ENOMEM; } if (adaptor->source) { g_source_remove(adaptor->source); adaptor->source = 0; } /* Because unlike the poll() API, glib doesn't tell us about HUPs by default */ evts |= (G_IO_HUP | G_IO_NVAL | G_IO_ERR); adaptor->fn = fn; adaptor->events = evts; adaptor->data = data; adaptor->p = p; adaptor->is_used++; adaptor->source = g_io_add_watch_full(channel, conv_prio_libqb2glib(p), evts, gio_read_socket, adaptor, gio_poll_destroy); /* Now that mainloop now holds a reference to channel, * thanks to g_io_add_watch_full(), drop ours from g_io_channel_unix_new(). * * This means that channel will be free'd by: * g_main_context_dispatch() * -> g_source_destroy_internal() * -> g_source_callback_unref() * shortly after gio_poll_destroy() completes */ g_io_channel_unref(channel); crm_trace("Added to mainloop with gsource id=%d", adaptor->source); if (adaptor->source > 0) { return 0; } return -EINVAL; } static int32_t gio_poll_dispatch_add(enum qb_loop_priority p, int32_t fd, int32_t evts, void *data, qb_ipcs_dispatch_fn_t fn) { return gio_poll_dispatch_update(p, fd, evts, data, fn, QB_TRUE); } static int32_t gio_poll_dispatch_mod(enum qb_loop_priority p, int32_t fd, int32_t evts, void *data, qb_ipcs_dispatch_fn_t fn) { return gio_poll_dispatch_update(p, fd, evts, data, fn, QB_FALSE); } static int32_t gio_poll_dispatch_del(int32_t fd) { struct gio_to_qb_poll *adaptor; crm_trace("Looking for fd=%d", fd); if (qb_array_index(gio_map, fd, (void **)&adaptor) == 0) { if (adaptor->source) { g_source_remove(adaptor->source); adaptor->source = 0; } } return 0; } struct qb_ipcs_poll_handlers gio_poll_funcs = { .job_add = NULL, .dispatch_add = gio_poll_dispatch_add, .dispatch_mod = gio_poll_dispatch_mod, .dispatch_del = gio_poll_dispatch_del, }; static enum qb_ipc_type pick_ipc_type(enum qb_ipc_type requested) { const char *env = pcmk__env_option(PCMK__ENV_IPC_TYPE); if (env && strcmp("shared-mem", env) == 0) { return QB_IPC_SHM; } else if (env && strcmp("socket", env) == 0) { return QB_IPC_SOCKET; } else if (env && strcmp("posix", env) == 0) { return QB_IPC_POSIX_MQ; } else if (env && strcmp("sysv", env) == 0) { return QB_IPC_SYSV_MQ; } else if (requested == QB_IPC_NATIVE) { /* We prefer shared memory because the server never blocks on * send. If part of a message fits into the socket, libqb * needs to block until the remainder can be sent also. * Otherwise the client will wait forever for the remaining * bytes. */ return QB_IPC_SHM; } return requested; } qb_ipcs_service_t * mainloop_add_ipc_server(const char *name, enum qb_ipc_type type, struct qb_ipcs_service_handlers *callbacks) { return mainloop_add_ipc_server_with_prio(name, type, callbacks, QB_LOOP_MED); } qb_ipcs_service_t * mainloop_add_ipc_server_with_prio(const char *name, enum qb_ipc_type type, struct qb_ipcs_service_handlers *callbacks, enum qb_loop_priority prio) { int rc = 0; qb_ipcs_service_t *server = NULL; if (gio_map == NULL) { gio_map = qb_array_create_2(64, sizeof(struct gio_to_qb_poll), 1); } server = qb_ipcs_create(name, 0, pick_ipc_type(type), callbacks); if (server == NULL) { crm_err("Could not create %s IPC server: %s (%d)", name, pcmk_rc_str(errno), errno); return NULL; } if (prio != QB_LOOP_MED) { qb_ipcs_request_rate_limit(server, conv_libqb_prio2ratelimit(prio)); } // Enforce a minimum IPC buffer size on all clients qb_ipcs_enforce_buffer_size(server, crm_ipc_default_buffer_size()); qb_ipcs_poll_handlers_set(server, &gio_poll_funcs); rc = qb_ipcs_run(server); if (rc < 0) { crm_err("Could not start %s IPC server: %s (%d)", name, pcmk_strerror(rc), rc); return NULL; // qb_ipcs_run() destroys server on failure } return server; } void mainloop_del_ipc_server(qb_ipcs_service_t * server) { if (server) { qb_ipcs_destroy(server); } } struct mainloop_io_s { char *name; void *userdata; int fd; guint source; crm_ipc_t *ipc; GIOChannel *channel; int (*dispatch_fn_ipc) (const char *buffer, ssize_t length, gpointer userdata); int (*dispatch_fn_io) (gpointer userdata); void (*destroy_fn) (gpointer userdata); }; /*! * \internal * \brief I/O watch callback function (GIOFunc) * * \param[in] gio I/O channel being watched * \param[in] condition I/O condition satisfied * \param[in] data User data passed when source was created * * \return G_SOURCE_REMOVE to remove source, G_SOURCE_CONTINUE to keep it */ static gboolean mainloop_gio_callback(GIOChannel *gio, GIOCondition condition, gpointer data) { gboolean rc = G_SOURCE_CONTINUE; mainloop_io_t *client = data; pcmk__assert(client->fd == g_io_channel_unix_get_fd(gio)); if (condition & G_IO_IN) { if (client->ipc) { long read_rc = 0L; int max = 10; do { read_rc = crm_ipc_read(client->ipc); if (read_rc <= 0) { crm_trace("Could not read IPC message from %s: %s (%ld)", client->name, pcmk_strerror(read_rc), read_rc); if (read_rc == -EAGAIN) { continue; } } else if (client->dispatch_fn_ipc) { const char *buffer = crm_ipc_buffer(client->ipc); crm_trace("New %ld-byte IPC message from %s " "after I/O condition %d", read_rc, client->name, (int) condition); if (client->dispatch_fn_ipc(buffer, read_rc, client->userdata) < 0) { crm_trace("Connection to %s no longer required", client->name); rc = G_SOURCE_REMOVE; } } pcmk__ipc_free_client_buffer(client->ipc); } while ((rc == G_SOURCE_CONTINUE) && (read_rc > 0) && --max > 0); } else { crm_trace("New I/O event for %s after I/O condition %d", client->name, (int) condition); if (client->dispatch_fn_io) { if (client->dispatch_fn_io(client->userdata) < 0) { crm_trace("Connection to %s no longer required", client->name); rc = G_SOURCE_REMOVE; } } } } if (client->ipc && !crm_ipc_connected(client->ipc)) { crm_err("Connection to %s closed " QB_XS " client=%p condition=%d", client->name, client, condition); rc = G_SOURCE_REMOVE; } else if (condition & (G_IO_HUP | G_IO_NVAL | G_IO_ERR)) { crm_trace("The connection %s[%p] has been closed (I/O condition=%d)", client->name, client, condition); rc = G_SOURCE_REMOVE; } else if ((condition & G_IO_IN) == 0) { /* #define GLIB_SYSDEF_POLLIN =1 #define GLIB_SYSDEF_POLLPRI =2 #define GLIB_SYSDEF_POLLOUT =4 #define GLIB_SYSDEF_POLLERR =8 #define GLIB_SYSDEF_POLLHUP =16 #define GLIB_SYSDEF_POLLNVAL =32 typedef enum { G_IO_IN GLIB_SYSDEF_POLLIN, G_IO_OUT GLIB_SYSDEF_POLLOUT, G_IO_PRI GLIB_SYSDEF_POLLPRI, G_IO_ERR GLIB_SYSDEF_POLLERR, G_IO_HUP GLIB_SYSDEF_POLLHUP, G_IO_NVAL GLIB_SYSDEF_POLLNVAL } GIOCondition; A bitwise combination representing a condition to watch for on an event source. G_IO_IN There is data to read. G_IO_OUT Data can be written (without blocking). G_IO_PRI There is urgent data to read. G_IO_ERR Error condition. G_IO_HUP Hung up (the connection has been broken, usually for pipes and sockets). G_IO_NVAL Invalid request. The file descriptor is not open. */ crm_err("Strange condition: %d", condition); } /* G_SOURCE_REMOVE results in mainloop_gio_destroy() being called * just before the source is removed from mainloop */ return rc; } static void mainloop_gio_destroy(gpointer c) { mainloop_io_t *client = c; char *c_name = strdup(client->name); /* client->source is valid but about to be destroyed (ref_count == 0) in gmain.c * client->channel will still have ref_count > 0... should be == 1 */ crm_trace("Destroying client %s[%p]", c_name, c); if (client->ipc) { crm_ipc_close(client->ipc); } if (client->destroy_fn) { void (*destroy_fn) (gpointer userdata) = client->destroy_fn; client->destroy_fn = NULL; destroy_fn(client->userdata); } if (client->ipc) { crm_ipc_t *ipc = client->ipc; client->ipc = NULL; crm_ipc_destroy(ipc); } crm_trace("Destroyed client %s[%p]", c_name, c); free(client->name); client->name = NULL; free(client); free(c_name); } /*! * \brief Connect to IPC and add it as a main loop source * * \param[in,out] ipc IPC connection to add * \param[in] priority Event source priority to use for connection * \param[in] userdata Data to register with callbacks * \param[in] callbacks Dispatch and destroy callbacks for connection * \param[out] source Newly allocated event source * * \return Standard Pacemaker return code * * \note On failure, the caller is still responsible for ipc. On success, the * caller should call mainloop_del_ipc_client() when source is no longer * needed, which will lead to the disconnection of the IPC later in the * main loop if it is connected. However the IPC disconnects, * mainloop_gio_destroy() will free ipc and source after calling the * destroy callback. */ int pcmk__add_mainloop_ipc(crm_ipc_t *ipc, int priority, void *userdata, const struct ipc_client_callbacks *callbacks, mainloop_io_t **source) { int rc = pcmk_rc_ok; int fd = -1; const char *ipc_name = NULL; CRM_CHECK((ipc != NULL) && (callbacks != NULL), return EINVAL); ipc_name = pcmk__s(crm_ipc_name(ipc), "Pacemaker"); rc = pcmk__connect_generic_ipc(ipc); if (rc != pcmk_rc_ok) { crm_debug("Connection to %s failed: %s", ipc_name, pcmk_rc_str(rc)); return rc; } rc = pcmk__ipc_fd(ipc, &fd); if (rc != pcmk_rc_ok) { crm_debug("Could not obtain file descriptor for %s IPC: %s", ipc_name, pcmk_rc_str(rc)); crm_ipc_close(ipc); return rc; } *source = mainloop_add_fd(ipc_name, priority, fd, userdata, NULL); if (*source == NULL) { rc = errno; crm_ipc_close(ipc); return rc; } (*source)->ipc = ipc; (*source)->destroy_fn = callbacks->destroy; (*source)->dispatch_fn_ipc = callbacks->dispatch; return pcmk_rc_ok; } /*! * \brief Get period for mainloop timer * * \param[in] timer Timer * * \return Period in ms */ guint pcmk__mainloop_timer_get_period(const mainloop_timer_t *timer) { if (timer) { return timer->period_ms; } return 0; } mainloop_io_t * mainloop_add_ipc_client(const char *name, int priority, size_t max_size, void *userdata, struct ipc_client_callbacks *callbacks) { crm_ipc_t *ipc = crm_ipc_new(name, 0); mainloop_io_t *source = NULL; int rc = pcmk__add_mainloop_ipc(ipc, priority, userdata, callbacks, &source); if (rc != pcmk_rc_ok) { if (crm_log_level == LOG_STDOUT) { fprintf(stderr, "Connection to %s failed: %s", name, pcmk_rc_str(rc)); } crm_ipc_destroy(ipc); if (rc > 0) { errno = rc; } else { errno = ENOTCONN; } return NULL; } return source; } void mainloop_del_ipc_client(mainloop_io_t * client) { mainloop_del_fd(client); } crm_ipc_t * mainloop_get_ipc_client(mainloop_io_t * client) { if (client) { return client->ipc; } return NULL; } mainloop_io_t * mainloop_add_fd(const char *name, int priority, int fd, void *userdata, struct mainloop_fd_callbacks * callbacks) { mainloop_io_t *client = NULL; if (fd >= 0) { client = calloc(1, sizeof(mainloop_io_t)); if (client == NULL) { return NULL; } client->name = strdup(name); client->userdata = userdata; if (callbacks) { client->destroy_fn = callbacks->destroy; client->dispatch_fn_io = callbacks->dispatch; } client->fd = fd; client->channel = g_io_channel_unix_new(fd); client->source = g_io_add_watch_full(client->channel, priority, (G_IO_IN | G_IO_HUP | G_IO_NVAL | G_IO_ERR), mainloop_gio_callback, client, mainloop_gio_destroy); /* Now that mainloop now holds a reference to channel, * thanks to g_io_add_watch_full(), drop ours from g_io_channel_unix_new(). * * This means that channel will be free'd by: * g_main_context_dispatch() or g_source_remove() * -> g_source_destroy_internal() * -> g_source_callback_unref() * shortly after mainloop_gio_destroy() completes */ g_io_channel_unref(client->channel); crm_trace("Added connection %d for %s[%p].%d", client->source, client->name, client, fd); } else { errno = EINVAL; } return client; } void mainloop_del_fd(mainloop_io_t * client) { if (client != NULL) { crm_trace("Removing client %s[%p]", client->name, client); if (client->source) { /* Results in mainloop_gio_destroy() being called just * before the source is removed from mainloop */ g_source_remove(client->source); } } } static GList *child_list = NULL; pid_t mainloop_child_pid(mainloop_child_t * child) { return child->pid; } const char * mainloop_child_name(mainloop_child_t * child) { return child->desc; } int mainloop_child_timeout(mainloop_child_t * child) { return child->timeout; } void * mainloop_child_userdata(mainloop_child_t * child) { return child->privatedata; } void mainloop_clear_child_userdata(mainloop_child_t * child) { child->privatedata = NULL; } /* good function name */ static void child_free(mainloop_child_t *child) { if (child->timerid != 0) { crm_trace("Removing timer %d", child->timerid); g_source_remove(child->timerid); child->timerid = 0; } free(child->desc); free(child); } /* terrible function name */ static int child_kill_helper(mainloop_child_t *child) { int rc; if (child->flags & mainloop_leave_pid_group) { crm_debug("Kill pid %d only. leave group intact.", child->pid); rc = kill(child->pid, SIGKILL); } else { crm_debug("Kill pid %d's group", child->pid); rc = kill(-child->pid, SIGKILL); } if (rc < 0) { if (errno != ESRCH) { - crm_perror(LOG_ERR, "kill(%d, KILL) failed", child->pid); + crm_err("kill(%d, KILL) failed: %s", child->pid, strerror(errno)); } return -errno; } return 0; } static gboolean child_timeout_callback(gpointer p) { mainloop_child_t *child = p; int rc = 0; child->timerid = 0; if (child->timeout) { crm_warn("%s process (PID %d) will not die!", child->desc, (int)child->pid); return FALSE; } rc = child_kill_helper(child); if (rc == -ESRCH) { /* Nothing left to do. pid doesn't exist */ return FALSE; } child->timeout = TRUE; crm_debug("%s process (PID %d) timed out", child->desc, (int)child->pid); child->timerid = pcmk__create_timer(5000, child_timeout_callback, child); return FALSE; } static bool child_waitpid(mainloop_child_t *child, int flags) { int rc = 0; int core = 0; int signo = 0; int status = 0; int exitcode = 0; bool callback_needed = true; rc = waitpid(child->pid, &status, flags); if (rc == 0) { // WNOHANG in flags, and child status is not available crm_trace("Child process %d (%s) still active", child->pid, child->desc); callback_needed = false; } else if (rc != child->pid) { /* According to POSIX, possible conditions: * - child->pid was non-positive (process group or any child), * and rc is specific child * - errno ECHILD (pid does not exist or is not child) * - errno EINVAL (invalid flags) * - errno EINTR (caller interrupted by signal) * * @TODO Handle these cases more specifically. */ signo = SIGCHLD; exitcode = 1; crm_notice("Wait for child process %d (%s) interrupted: %s", child->pid, child->desc, pcmk_rc_str(errno)); } else if (WIFEXITED(status)) { exitcode = WEXITSTATUS(status); crm_trace("Child process %d (%s) exited with status %d", child->pid, child->desc, exitcode); } else if (WIFSIGNALED(status)) { signo = WTERMSIG(status); crm_trace("Child process %d (%s) exited with signal %d (%s)", child->pid, child->desc, signo, strsignal(signo)); #ifdef WCOREDUMP // AIX, SunOS, maybe others } else if (WCOREDUMP(status)) { core = 1; crm_err("Child process %d (%s) dumped core", child->pid, child->desc); #endif } else { // flags must contain WUNTRACED and/or WCONTINUED to reach this crm_trace("Child process %d (%s) stopped or continued", child->pid, child->desc); callback_needed = false; } if (callback_needed && child->callback) { child->callback(child, child->pid, core, signo, exitcode); } return callback_needed; } static void child_death_dispatch(int signal) { for (GList *iter = child_list; iter; ) { GList *saved = iter; mainloop_child_t *child = iter->data; iter = iter->next; if (child_waitpid(child, WNOHANG)) { crm_trace("Removing completed process %d from child list", child->pid); child_list = g_list_remove_link(child_list, saved); g_list_free(saved); child_free(child); } } } static gboolean child_signal_init(gpointer p) { crm_trace("Installed SIGCHLD handler"); /* Do NOT use g_child_watch_add() and friends, they rely on pthreads */ mainloop_add_signal(SIGCHLD, child_death_dispatch); /* In case they terminated before the signal handler was installed */ child_death_dispatch(SIGCHLD); return FALSE; } gboolean mainloop_child_kill(pid_t pid) { GList *iter; mainloop_child_t *child = NULL; mainloop_child_t *match = NULL; /* It is impossible to block SIGKILL, this allows us to * call waitpid without WNOHANG flag.*/ int waitflags = 0, rc = 0; for (iter = child_list; iter != NULL && match == NULL; iter = iter->next) { child = iter->data; if (pid == child->pid) { match = child; } } if (match == NULL) { return FALSE; } rc = child_kill_helper(match); if(rc == -ESRCH) { /* It's gone, but hasn't shown up in waitpid() yet. Wait until we get * SIGCHLD and let handler clean it up as normal (so we get the correct * return code/status). The blocking alternative would be to call * child_waitpid(match, 0). */ crm_trace("Waiting for signal that child process %d completed", match->pid); return TRUE; } else if(rc != 0) { /* If KILL for some other reason set the WNOHANG flag since we * can't be certain what happened. */ waitflags = WNOHANG; } if (!child_waitpid(match, waitflags)) { /* not much we can do if this occurs */ return FALSE; } child_list = g_list_remove(child_list, match); child_free(match); return TRUE; } /* Create/Log a new tracked process * To track a process group, use -pid * * @TODO Using a non-positive pid (i.e. any child, or process group) would * likely not be useful since we will free the child after the first * completed process. */ void mainloop_child_add_with_flags(pid_t pid, int timeout, const char *desc, void *privatedata, enum mainloop_child_flags flags, void (*callback) (mainloop_child_t * p, pid_t pid, int core, int signo, int exitcode)) { static bool need_init = TRUE; mainloop_child_t *child = pcmk__assert_alloc(1, sizeof(mainloop_child_t)); child->pid = pid; child->timerid = 0; child->timeout = FALSE; child->privatedata = privatedata; child->callback = callback; child->flags = flags; child->desc = pcmk__str_copy(desc); if (timeout) { child->timerid = pcmk__create_timer(timeout, child_timeout_callback, child); } child_list = g_list_append(child_list, child); if(need_init) { need_init = FALSE; /* SIGCHLD processing has to be invoked from mainloop. * We do not want it to be possible to both add a child pid * to mainloop, and have the pid's exit callback invoked within * the same callstack. */ pcmk__create_timer(1, child_signal_init, NULL); } } void mainloop_child_add(pid_t pid, int timeout, const char *desc, void *privatedata, void (*callback) (mainloop_child_t * p, pid_t pid, int core, int signo, int exitcode)) { mainloop_child_add_with_flags(pid, timeout, desc, privatedata, 0, callback); } static gboolean mainloop_timer_cb(gpointer user_data) { int id = 0; bool repeat = FALSE; struct mainloop_timer_s *t = user_data; pcmk__assert(t != NULL); id = t->id; t->id = 0; /* Ensure it's unset during callbacks so that * mainloop_timer_running() works as expected */ if(t->cb) { crm_trace("Invoking callbacks for timer %s", t->name); repeat = t->repeat; if(t->cb(t->userdata) == FALSE) { crm_trace("Timer %s complete", t->name); repeat = FALSE; } } if(repeat) { /* Restore if repeating */ t->id = id; } return repeat; } bool mainloop_timer_running(mainloop_timer_t *t) { if(t && t->id != 0) { return TRUE; } return FALSE; } void mainloop_timer_start(mainloop_timer_t *t) { mainloop_timer_stop(t); if(t && t->period_ms > 0) { crm_trace("Starting timer %s", t->name); t->id = pcmk__create_timer(t->period_ms, mainloop_timer_cb, t); } } void mainloop_timer_stop(mainloop_timer_t *t) { if(t && t->id != 0) { crm_trace("Stopping timer %s", t->name); g_source_remove(t->id); t->id = 0; } } guint mainloop_timer_set_period(mainloop_timer_t *t, guint period_ms) { guint last = 0; if(t) { last = t->period_ms; t->period_ms = period_ms; } if(t && t->id != 0 && last != t->period_ms) { mainloop_timer_start(t); } return last; } mainloop_timer_t * mainloop_timer_add(const char *name, guint period_ms, bool repeat, GSourceFunc cb, void *userdata) { mainloop_timer_t *t = pcmk__assert_alloc(1, sizeof(mainloop_timer_t)); if (name != NULL) { t->name = crm_strdup_printf("%s-%u-%d", name, period_ms, repeat); } else { t->name = crm_strdup_printf("%p-%u-%d", t, period_ms, repeat); } t->id = 0; t->period_ms = period_ms; t->repeat = repeat; t->cb = cb; t->userdata = userdata; crm_trace("Created timer %s with %p %p", t->name, userdata, t->userdata); return t; } void mainloop_timer_del(mainloop_timer_t *t) { if(t) { crm_trace("Destroying timer %s", t->name); mainloop_timer_stop(t); free(t->name); free(t); } } /* * Helpers to make sure certain events aren't lost at shutdown */ static gboolean drain_timeout_cb(gpointer user_data) { bool *timeout_popped = (bool*) user_data; *timeout_popped = TRUE; return FALSE; } /*! * \brief Drain some remaining main loop events then quit it * * \param[in,out] mloop Main loop to drain and quit * \param[in] n Drain up to this many pending events */ void pcmk_quit_main_loop(GMainLoop *mloop, unsigned int n) { if ((mloop != NULL) && g_main_loop_is_running(mloop)) { GMainContext *ctx = g_main_loop_get_context(mloop); /* Drain up to n events in case some memory clean-up is pending * (helpful to reduce noise in valgrind output). */ for (int i = 0; (i < n) && g_main_context_pending(ctx); ++i) { g_main_context_dispatch(ctx); } g_main_loop_quit(mloop); } } /*! * \brief Process main loop events while a certain condition is met * * \param[in,out] mloop Main loop to process * \param[in] timer_ms Don't process longer than this amount of time * \param[in] check Function that returns true if events should be * processed * * \note This function is intended to be called at shutdown if certain important * events should not be missed. The caller would likely quit the main loop * or exit after calling this function. The check() function will be * passed the remaining timeout in milliseconds. */ void pcmk_drain_main_loop(GMainLoop *mloop, guint timer_ms, bool (*check)(guint)) { bool timeout_popped = FALSE; guint timer = 0; GMainContext *ctx = NULL; CRM_CHECK(mloop && check, return); ctx = g_main_loop_get_context(mloop); if (ctx) { time_t start_time = time(NULL); timer = pcmk__create_timer(timer_ms, drain_timeout_cb, &timeout_popped); while (!timeout_popped && check(timer_ms - (time(NULL) - start_time) * 1000)) { g_main_context_iteration(ctx, TRUE); } } if (!timeout_popped && (timer > 0)) { g_source_remove(timer); } } diff --git a/lib/common/output_html.c b/lib/common/output_html.c index 65383d8e81..b2b3d95ee2 100644 --- a/lib/common/output_html.c +++ b/lib/common/output_html.c @@ -1,522 +1,532 @@ /* * Copyright 2019-2025 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 // xmlNode #include // xmlChar #include #include static const char *stylesheet_default = "." PCMK__VALUE_BOLD " { font-weight: bold }\n" "." PCMK_VALUE_ONLINE " { color: green }\n" "." PCMK_VALUE_OFFLINE " { color: red }\n" "." PCMK__VALUE_MAINT " { color: blue }\n" "." PCMK_VALUE_STANDBY " { color: blue }\n" "." PCMK__VALUE_HEALTH_RED " { color: red }\n" "." PCMK__VALUE_HEALTH_YELLOW " { color: GoldenRod }\n" "." PCMK__VALUE_RSC_FAILED " { color: red }\n" "." PCMK__VALUE_RSC_FAILURE_IGNORED " { color: DarkGreen }\n" "." PCMK__VALUE_RSC_MANAGED " { color: blue }\n" "." PCMK__VALUE_RSC_MULTIPLE " { color: orange }\n" "." PCMK__VALUE_RSC_OK " { color: green }\n" "." PCMK__VALUE_WARNING " { color: red; font-weight: bold }"; +/* @TODO stylesheet_link, title, and extra_headers should be set + * per-output-object and should be freed before exit + */ static gboolean cgi_output = FALSE; -static char *stylesheet_link = NULL; -static char *title = NULL; +static gchar *stylesheet_link = NULL; +static gchar *title = NULL; static GSList *extra_headers = NULL; GOptionEntry pcmk__html_output_entries[] = { { "html-cgi", 0, 0, G_OPTION_ARG_NONE, &cgi_output, "Add CGI headers (requires --output-as=html)", NULL }, { "html-stylesheet", 0, 0, G_OPTION_ARG_STRING, &stylesheet_link, "Link to an external stylesheet (requires --output-as=html)", "URI" }, { "html-title", 0, 0, G_OPTION_ARG_STRING, &title, "Specify a page title (requires --output-as=html)", "TITLE" }, { NULL } }; /* The first several elements of this struct must be the same as the first * several elements of private_data_s in lib/common/output_xml.c. This * struct gets passed to a bunch of the pcmk__output_xml_* functions which * assume an XML private_data_s. Keeping them laid out the same means this * still works. */ typedef struct private_data_s { /* Begin members that must match the XML version */ xmlNode *root; GQueue *parent_q; GSList *errors; /* End members that must match the XML version */ } private_data_t; static void html_free_priv(pcmk__output_t *out) { private_data_t *priv = NULL; if (out == NULL || out->priv == NULL) { return; } priv = out->priv; pcmk__xml_free(priv->root); /* The elements of parent_q are xmlNodes that are a part of the * priv->root document, so the above line already frees them. Don't * call g_queue_free_full here. */ g_queue_free(priv->parent_q); g_slist_free_full(priv->errors, free); free(priv); out->priv = NULL; } static bool html_init(pcmk__output_t *out) { private_data_t *priv = NULL; pcmk__assert(out != NULL); /* If html_init was previously called on this output struct, just return. */ if (out->priv != NULL) { return true; } else { out->priv = calloc(1, sizeof(private_data_t)); if (out->priv == NULL) { return false; } priv = out->priv; } priv->parent_q = g_queue_new(); priv->root = pcmk__xe_create(NULL, "html"); xmlCreateIntSubset(priv->root->doc, (const xmlChar *) "html", NULL, NULL); crm_xml_add(priv->root, PCMK_XA_LANG, PCMK__VALUE_EN); g_queue_push_tail(priv->parent_q, priv->root); priv->errors = NULL; pcmk__output_xml_create_parent(out, "body", NULL); return true; } static void add_error_node(gpointer data, gpointer user_data) { char *str = (char *) data; pcmk__output_t *out = (pcmk__output_t *) user_data; out->list_item(out, NULL, "%s", str); } static void html_finish(pcmk__output_t *out, crm_exit_t exit_status, bool print, void **copy_dest) { private_data_t *priv = NULL; htmlNodePtr head_node = NULL; htmlNodePtr charset_node = NULL; xmlNode *child_node = NULL; pcmk__assert(out != NULL); priv = out->priv; /* If root is NULL, html_init failed and we are being called from pcmk__output_free * in the pcmk__output_new path. */ if (priv == NULL || priv->root == NULL) { return; } if (cgi_output && print) { fprintf(out->dest, "Content-Type: text/html\n\n"); } /* Add the head node last - it's not needed earlier because it doesn't contain * anything else that the user could add, and we want it done last to pick up * any options that may have been given. */ head_node = pcmk__xe_create(priv->root, "head"); xmlAddPrevSibling(priv->root->children, head_node); if (title != NULL ) { child_node = pcmk__xe_create(head_node, "title"); pcmk__xe_set_content(child_node, "%s", title); } else if (out->request != NULL) { child_node = pcmk__xe_create(head_node, "title"); pcmk__xe_set_content(child_node, "%s", out->request); } charset_node = pcmk__xe_create(head_node, PCMK__XE_META); crm_xml_add(charset_node, "charset", "utf-8"); /* Add any extra header nodes the caller might have created. */ for (GSList *iter = extra_headers; iter != NULL; iter = iter->next) { pcmk__xml_copy(head_node, (xmlNode *) iter->data); } /* Stylesheets are included two different ways. The first is via a built-in * default (see the stylesheet_default const above). The second is via the * html-stylesheet option, and this should obviously be a link to a * stylesheet. The second can override the first. At least one should be * given. */ child_node = pcmk__xe_create(head_node, "style"); pcmk__xe_set_content(child_node, "%s", stylesheet_default); if (stylesheet_link != NULL) { htmlNodePtr link_node = pcmk__xe_create(head_node, "link"); pcmk__xe_set_props(link_node, "rel", "stylesheet", "href", stylesheet_link, NULL); } if (g_slist_length(priv->errors) > 0) { out->begin_list(out, "Errors", NULL, NULL); g_slist_foreach(priv->errors, add_error_node, (gpointer) out); out->end_list(out); } if (print) { htmlDocDump(out->dest, priv->root->doc); } if (copy_dest != NULL) { *copy_dest = pcmk__xml_copy(NULL, priv->root); } g_slist_free_full(extra_headers, (GDestroyNotify) pcmk__xml_free); extra_headers = NULL; } static void html_reset(pcmk__output_t *out) { pcmk__assert(out != NULL); out->dest = freopen(NULL, "w", out->dest); pcmk__assert(out->dest != NULL); html_free_priv(out); html_init(out); } static void html_subprocess_output(pcmk__output_t *out, int exit_status, const char *proc_stdout, const char *proc_stderr) { char *rc_buf = NULL; pcmk__assert(out != NULL); rc_buf = crm_strdup_printf("Return code: %d", exit_status); pcmk__output_create_xml_text_node(out, "h2", "Command Output"); pcmk__output_create_html_node(out, PCMK__XE_DIV, NULL, NULL, rc_buf); if (proc_stdout != NULL) { pcmk__output_create_html_node(out, PCMK__XE_DIV, NULL, NULL, "Stdout"); pcmk__output_create_html_node(out, PCMK__XE_DIV, NULL, PCMK__VALUE_OUTPUT, proc_stdout); } if (proc_stderr != NULL) { pcmk__output_create_html_node(out, PCMK__XE_DIV, NULL, NULL, "Stderr"); pcmk__output_create_html_node(out, PCMK__XE_DIV, NULL, PCMK__VALUE_OUTPUT, proc_stderr); } free(rc_buf); } static void html_version(pcmk__output_t *out) { pcmk__assert(out != NULL); pcmk__output_create_xml_text_node(out, "h2", "Version Information"); pcmk__output_create_html_node(out, PCMK__XE_DIV, NULL, NULL, "Program: Pacemaker"); pcmk__output_create_html_node(out, PCMK__XE_DIV, NULL, NULL, "Version: " PACEMAKER_VERSION); pcmk__output_create_html_node(out, PCMK__XE_DIV, NULL, NULL, "Author: Andrew Beekhof and " "the Pacemaker project contributors"); pcmk__output_create_html_node(out, PCMK__XE_DIV, NULL, NULL, "Build: " BUILD_VERSION); pcmk__output_create_html_node(out, PCMK__XE_DIV, NULL, NULL, "Features: " CRM_FEATURES); } G_GNUC_PRINTF(2, 3) static void html_err(pcmk__output_t *out, const char *format, ...) { private_data_t *priv = NULL; int len = 0; char *buf = NULL; va_list ap; pcmk__assert((out != NULL) && (out->priv != NULL)); priv = out->priv; va_start(ap, format); len = vasprintf(&buf, format, ap); pcmk__assert(len >= 0); va_end(ap); priv->errors = g_slist_append(priv->errors, buf); } G_GNUC_PRINTF(2, 3) static int html_info(pcmk__output_t *out, const char *format, ...) { return pcmk_rc_no_output; } static void html_output_xml(pcmk__output_t *out, const char *name, const char *buf) { htmlNodePtr node = NULL; pcmk__assert(out != NULL); node = pcmk__output_create_html_node(out, "pre", NULL, NULL, buf); crm_xml_add(node, PCMK_XA_LANG, "xml"); } G_GNUC_PRINTF(4, 5) static void html_begin_list(pcmk__output_t *out, const char *singular_noun, const char *plural_noun, const char *format, ...) { int q_len = 0; private_data_t *priv = NULL; xmlNodePtr node = NULL; pcmk__assert((out != NULL) && (out->priv != NULL)); priv = out->priv; /* If we are already in a list (the queue depth is always at least * one because of the element), first create a
  • element * to hold the

    and the new list. */ q_len = g_queue_get_length(priv->parent_q); if (q_len > 2) { pcmk__output_xml_create_parent(out, "li", NULL); } if (format != NULL) { va_list ap; char *buf = NULL; int len; va_start(ap, format); len = vasprintf(&buf, format, ap); va_end(ap); pcmk__assert(len >= 0); if (q_len > 2) { pcmk__output_create_xml_text_node(out, "h3", buf); } else { pcmk__output_create_xml_text_node(out, "h2", buf); } free(buf); } node = pcmk__output_xml_create_parent(out, "ul", NULL); g_queue_push_tail(priv->parent_q, node); } G_GNUC_PRINTF(3, 4) static void html_list_item(pcmk__output_t *out, const char *name, const char *format, ...) { htmlNodePtr item_node = NULL; va_list ap; char *buf = NULL; int len; pcmk__assert(out != NULL); va_start(ap, format); len = vasprintf(&buf, format, ap); pcmk__assert(len >= 0); va_end(ap); item_node = pcmk__output_create_xml_text_node(out, "li", buf); free(buf); if (name != NULL) { crm_xml_add(item_node, PCMK_XA_CLASS, name); } } static void html_increment_list(pcmk__output_t *out) { /* This function intentially left blank */ } static void html_end_list(pcmk__output_t *out) { private_data_t *priv = NULL; pcmk__assert((out != NULL) && (out->priv != NULL)); priv = out->priv; /* Remove the
      tag, but do not free this result - it's still * part of the document. */ g_queue_pop_tail(priv->parent_q); pcmk__output_xml_pop_parent(out); /* Remove the
    • created for nested lists. */ if (g_queue_get_length(priv->parent_q) > 2) { pcmk__output_xml_pop_parent(out); } } static bool html_is_quiet(pcmk__output_t *out) { return false; } static void html_spacer(pcmk__output_t *out) { pcmk__assert(out != NULL); pcmk__output_create_xml_node(out, "br", NULL); } static void html_progress(pcmk__output_t *out, bool end) { /* This function intentially left blank */ } pcmk__output_t * pcmk__mk_html_output(char **argv) { pcmk__output_t *retval = calloc(1, sizeof(pcmk__output_t)); if (retval == NULL) { return NULL; } retval->fmt_name = "html"; retval->request = pcmk__quote_cmdline(argv); retval->init = html_init; retval->free_priv = html_free_priv; retval->finish = html_finish; retval->reset = html_reset; retval->register_message = pcmk__register_message; retval->message = pcmk__call_message; retval->subprocess_output = html_subprocess_output; retval->version = html_version; retval->info = html_info; retval->transient = html_info; retval->err = html_err; retval->output_xml = html_output_xml; retval->begin_list = html_begin_list; retval->list_item = html_list_item; retval->increment_list = html_increment_list; retval->end_list = html_end_list; retval->is_quiet = html_is_quiet; retval->spacer = html_spacer; retval->progress = html_progress; retval->prompt = pcmk__text_prompt; return retval; } xmlNodePtr pcmk__output_create_html_node(pcmk__output_t *out, const char *element_name, const char *id, const char *class_name, const char *text) { htmlNodePtr node = NULL; pcmk__assert(out != NULL); CRM_CHECK(pcmk__str_eq(out->fmt_name, "html", pcmk__str_none), return NULL); node = pcmk__output_create_xml_text_node(out, element_name, text); if (class_name != NULL) { crm_xml_add(node, PCMK_XA_CLASS, class_name); } if (id != NULL) { crm_xml_add(node, PCMK_XA_ID, id); } return node; } /*! * \internal * \brief Create a new HTML element under a given parent with ID and class * * \param[in,out] parent XML element that will be the new element's parent * (\c NULL to create a new XML document with the new * node as root) * \param[in] name Name of new element * \param[in] id CSS ID of new element (can be \c NULL) * \param[in] class_name CSS class of new element (can be \c NULL) * * \return Newly created XML element (guaranteed not to be \c NULL) */ xmlNode * pcmk__html_create(xmlNode *parent, const char *name, const char *id, const char *class_name) { xmlNode *node = pcmk__xe_create(parent, name); pcmk__xe_set_props(node, PCMK_XA_CLASS, class_name, PCMK_XA_ID, id, NULL); return node; } +void +pcmk__html_set_title(const char *name) +{ + g_free(title); + title = g_strdup(name); +} + void pcmk__html_add_header(const char *name, ...) { htmlNodePtr header_node; va_list ap; va_start(ap, name); header_node = pcmk__xe_create(NULL, name); while (1) { char *key = va_arg(ap, char *); char *value; if (key == NULL) { break; } value = va_arg(ap, char *); crm_xml_add(header_node, key, value); } extra_headers = g_slist_append(extra_headers, header_node); va_end(ap); } diff --git a/lib/common/pid.c b/lib/common/pid.c index 3011f8d470..47f278b4a1 100644 --- a/lib/common/pid.c +++ b/lib/common/pid.c @@ -1,256 +1,95 @@ /* * Copyright 2004-2025 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 int pcmk__pid_active(pid_t pid, const char *daemon) { static pid_t last_asked_pid = 0; /* log spam prevention */ int rc = 0; if (pid <= 0) { return EINVAL; } rc = kill(pid, 0); if ((rc < 0) && (errno == ESRCH)) { return ESRCH; /* no such PID detected */ } else if ((daemon == NULL) || !pcmk__procfs_has_pids()) { // The kill result is all we have, we can't check the name if (rc == 0) { return pcmk_rc_ok; } rc = errno; if (last_asked_pid != pid) { crm_info("Cannot examine PID %lld: %s", (long long) pid, pcmk_rc_str(rc)); last_asked_pid = pid; } return rc; /* errno != ESRCH */ } else { /* 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); bool paths_equal = false; char *exe_path = NULL; char *myexe_path = NULL; rc = pcmk__procfs_pid2path(pid, &exe_path); if (rc != pcmk_rc_ok) { if (rc != EACCES) { // Check again to filter out races if ((kill(pid, 0) < 0) && (errno == ESRCH)) { return ESRCH; } } if (last_asked_pid != pid) { if (rc == EACCES) { crm_info("Could not get executable for PID %lld: %s " QB_XS " rc=%d", (long long) pid, pcmk_rc_str(rc), rc); } else { crm_err("Could not get executable for PID %lld: %s " QB_XS " rc=%d", (long long) pid, pcmk_rc_str(rc), rc); } last_asked_pid = pid; } if (rc == EACCES) { // Trust kill if it was OK (we can't double-check via path) return checked_through_kill? pcmk_rc_ok : EACCES; } else { return ESRCH; /* most likely errno == ENOENT */ } } if (daemon[0] != '/') { myexe_path = crm_strdup_printf(CRM_DAEMON_DIR "/%s", daemon); } else { myexe_path = pcmk__str_copy(daemon); } paths_equal = pcmk__str_eq(exe_path, myexe_path, pcmk__str_none); free(exe_path); free(myexe_path); if (paths_equal) { return pcmk_rc_ok; } } return ESRCH; } - -#define LOCKSTRLEN 11 - -/*! - * \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 - */ -static int -read_pidfile(const char *filename, pid_t *pid) -{ - int fd; - struct stat sbuf; - int rc = pcmk_rc_ok; - 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) { - 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; - } - - errno = 0; - rc = sscanf(buf, "%lld", &pid_read); - - if (rc > 0) { - if (pid_read <= 0) { - rc = ESRCH; - } else { - rc = pcmk_rc_ok; - *pid = (pid_t) pid_read; - crm_trace("Read pid %lld from %s", pid_read, filename); - } - } else if (rc == 0) { - rc = ENODATA; - } else { - rc = errno; - } - - bail: - close(fd); - return rc; -} - -/*! - * \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) -{ - pid_t pidfile_pid = 0; - int rc = read_pidfile(filename, &pidfile_pid); - - if (pid) { - *pid = pidfile_pid; - } - - if (rc != pcmk_rc_ok) { - // Error reading PID file or invalid contents - unlink(filename); - rc = ENOENT; - - } else if ((expected_pid > 0) && (pidfile_pid == expected_pid)) { - // PID in file matches what was expected - rc = pcmk_rc_ok; - - } else if (pcmk__pid_active(pidfile_pid, expected_name) == ESRCH) { - // Contains a stale value - unlink(filename); - rc = ENOENT; - - } else if ((expected_pid > 0) && (pidfile_pid != expected_pid)) { - // Locked by existing process - rc = EEXIST; - } - - 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 -pcmk__lock_pidfile(const char *filename, const char *name) -{ - pid_t mypid = getpid(); - int fd = 0; - int rc = 0; - char buf[LOCKSTRLEN + 2]; - - 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; - } - - /* @FIXME If the string representation of the max PID is longer than - * sizeof(buf), then the PID will be truncated. The read and write - * buffers should be large enough to hold the max PID on all systems. - * - * Also, if the max PID's string representation is longer than - * (LOCKSTRLEN - 1), then snprintf() will return a value greater than - * (LOCKSTRLEN - 1). - */ - pcmk__assert(snprintf(buf, sizeof(buf), "%*lld\n", LOCKSTRLEN - 1, - (long long) mypid) >= (LOCKSTRLEN - 1)); - rc = write(fd, buf, LOCKSTRLEN); - close(fd); - - if (rc != LOCKSTRLEN) { - crm_perror(LOG_ERR, "Incomplete write to %s", filename); - return errno; - } - - 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 35ffd5c6de..71c4deb2e1 100644 --- a/lib/common/procfs.c +++ b/lib/common/procfs.c @@ -1,478 +1,478 @@ /* * Copyright 2015-2025 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 // g_str_has_prefix() #if HAVE_LINUX_PROCFS /*! * \internal * \brief Return name of /proc file containing the CIB daemon's load statistics * * \return Newly allocated memory with file name on success, NULL otherwise * * \note It is the caller's responsibility to free the return value. * This will return NULL if the daemon is being run via valgrind. * This should be called only on Linux systems. */ static char * find_cib_loadfile(const char *server) { pid_t pid = pcmk__procfs_pid_of(server); return pid? crm_strdup_printf("/proc/%lld/stat", (long long) pid) : NULL; } /*! * \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 Standard Pacemaker return code * \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. */ static int pcmk__procfs_process_info(const struct dirent *entry, char *name, pid_t *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 = (pid_t) 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; } #endif // HAVE_LINUX_PROCFS /*! * \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. */ pid_t pcmk__procfs_pid_of(const char *name) { #if HAVE_LINUX_PROCFS DIR *dp; struct dirent *entry; pid_t 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 ((pcmk__procfs_process_info(entry, entry_name, &pid) == pcmk_rc_ok) && pcmk__str_eq(entry_name, name, pcmk__str_casei) && (pcmk__pid_active(pid, NULL) == pcmk_rc_ok)) { crm_info("Found %s active as process %lld", name, (long long) pid); break; } pid = 0; } closedir(dp); return pid; #else return 0; #endif // HAVE_LINUX_PROCFS } /*! * \internal * \brief Calculate number of logical CPU cores from procfs * * \return Number of cores (or 1 if unable to determine) */ unsigned int pcmk__procfs_num_cores(void) { #if HAVE_LINUX_PROCFS 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"); + crm_info("Could not open /proc/stat: %s", strerror(errno)); } else { char buffer[2048]; while (fgets(buffer, sizeof(buffer), stream)) { if (g_str_has_prefix(buffer, "cpu") && isdigit(buffer[3])) { ++cores; } } fclose(stream); } return cores? cores : 1; #else return 1; #endif // HAVE_LINUX_PROCFS } /*! * \internal * \brief Get the executable path corresponding to a process ID * * \param[in] pid Process ID to check * \param[out] path Where to store executable path (can be \c NULL) * * \return Standard Pacemaker error code (as possible errno values from * readlink()) */ int pcmk__procfs_pid2path(pid_t pid, char **path) { #if HAVE_LINUX_PROCFS char *procfs_path = NULL; ssize_t link_rc; /* The readlink(2) man page recommends calling lstat() to get the required * buffer size, and then dynamically allocate the buffer. However, * st_size == 0 for symlinks under /proc. So we use PATH_MAX. */ char real_path[PATH_MAX] = { '\0', }; pcmk__assert((path == NULL) || (*path == NULL)); procfs_path = crm_strdup_printf("/proc/%lld/exe", (long long) pid); link_rc = readlink(procfs_path, real_path, sizeof(real_path)); free(procfs_path); if (link_rc < 0) { return errno; } else if (link_rc >= sizeof(real_path)) { return ENAMETOOLONG; } if (path != NULL) { /* Make Coverity happy; we already zero-initialized real_path, and we * returned ENAMETOOLONG if it's no longer null-terminated */ real_path[link_rc] = '\0'; *path = pcmk__str_copy(real_path); } return pcmk_rc_ok; #else return EOPNOTSUPP; #endif // HAVE_LINUX_PROCFS } /*! * \internal * \brief Check whether process ID information is available from procfs * * \return true if process ID information is available, otherwise false */ bool pcmk__procfs_has_pids(void) { #if HAVE_LINUX_PROCFS static bool have_pids = false; static bool checked = false; if (!checked) { have_pids = pcmk__procfs_pid2path(getpid(), NULL) == pcmk_rc_ok; checked = true; } return have_pids; #else return false; #endif // HAVE_LINUX_PROCFS } /*! * \internal * \brief Return an open handle on the directory containing links to open file * descriptors, or NULL on error */ DIR * pcmk__procfs_fd_dir(void) { DIR *dir = NULL; /* /proc/self/fd (on Linux) or /dev/fd (on most OSes) contains symlinks to * all open files for the current process, named as the file descriptor. * Use this if available, because it's more efficient than a shotgun * approach to closing descriptors. */ #if HAVE_LINUX_PROCFS dir = opendir("/proc/self/fd"); #endif // HAVE_LINUX_PROCFS return dir; } /*! * \internal * \brief Trigger a sysrq command if supported on current platform * * \param[in] t Sysrq command to trigger */ void pcmk__sysrq_trigger(char t) { #if HAVE_LINUX_PROCFS // Root can always write here, regardless of kernel.sysrq value FILE *procf = fopen("/proc/sysrq-trigger", "a"); if (procf == NULL) { crm_warn("Could not open sysrq-trigger: %s", strerror(errno)); } else { fprintf(procf, "%c\n", t); fclose(procf); } #endif // HAVE_LINUX_PROCFS } bool pcmk__throttle_cib_load(const char *server, float *load) { /* /proc/[pid]/stat * * Status information about the process. This is used by ps(1). It is defined * in /usr/src/linux/fs/proc/array.c. * * The fields, in order, with their proper scanf(3) format specifiers, are: * * pid %d (1) The process ID. * comm %s (2) The filename of the executable, in parentheses. This is * visible whether or not the executable is swapped out. * state %c (3) One character from the string "RSDZTW" where R is running, * S is sleeping in an interruptible wait, D is waiting in * uninterruptible disk sleep, Z is zombie, T is traced or * stopped (on a signal), and W is paging. * ppid %d (4) The PID of the parent. * pgrp %d (5) The process group ID of the process. * session %d (6) The session ID of the process. * tty_nr %d (7) The controlling terminal of the process. (The minor device * number is contained in the combination of bits 31 to 20 and * 7 to 0; the major device number is in bits 15 to 8.) * tpgid %d (8) The ID of the foreground process group of the controlling * terminal of the process. * flags %u (9) The kernel flags word of the process. For bit meanings, see * the PF_* defines in the Linux kernel source file include/linux/sched.h. * Details depend on the kernel version. * minflt %lu (10) The number of minor faults the process has made which have * not required loading a memory page from disk. * cminflt %lu (11) The number of minor faults that the process's waited-for * children have made. * majflt %lu (12) The number of major faults the process has made which have * required loading a memory page from disk. * cmajflt %lu (13) The number of major faults that the process's waited-for * children have made. * utime %lu (14) Amount of time that this process has been scheduled in user * mode, measured in clock ticks (divide by sysconf(_SC_CLK_TCK)). * This includes guest time, guest_time (time spent running a * virtual CPU, see below), so that applications that are not * aware of the guest time field do not lose that time from * their calculations. * stime %lu (15) Amount of time that this process has been scheduled in * kernel mode, measured in clock ticks (divide by sysconf(_SC_CLK_TCK)). */ #if HAVE_LINUX_PROCFS static char *loadfile = NULL; static time_t last_call = 0; static long ticks_per_s = 0; static unsigned long last_utime, last_stime; char buffer[64*1024]; FILE *stream = NULL; time_t now = time(NULL); if (load == NULL) { return false; } else { *load = 0.0; } if (loadfile == NULL) { last_call = 0; last_utime = 0; last_stime = 0; loadfile = find_cib_loadfile(server); if (loadfile == NULL) { crm_warn("Couldn't find CIB load file"); return false; } ticks_per_s = sysconf(_SC_CLK_TCK); crm_trace("Found %s", loadfile); } stream = fopen(loadfile, "r"); if (stream == NULL) { int rc = errno; crm_warn("Couldn't read %s: %s (%d)", loadfile, pcmk_rc_str(rc), rc); free(loadfile); loadfile = NULL; return false; } if (fgets(buffer, sizeof(buffer), stream) != NULL) { char *comm = pcmk__assert_alloc(1, 256); char state = 0; int rc = 0, pid = 0, ppid = 0, pgrp = 0, session = 0, tty_nr = 0, tpgid = 0; unsigned long flags = 0, minflt = 0, cminflt = 0, majflt = 0, cmajflt = 0, utime = 0, stime = 0; rc = sscanf(buffer, "%d %[^ ] %c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu", &pid, comm, &state, &ppid, &pgrp, &session, &tty_nr, &tpgid, &flags, &minflt, &cminflt, &majflt, &cmajflt, &utime, &stime); free(comm); if (rc != 15) { crm_err("Only %d of 15 fields found in %s", rc, loadfile); fclose(stream); return false; } else if ((last_call > 0) && (last_call < now) && (last_utime <= utime) && (last_stime <= stime)) { time_t elapsed = now - last_call; unsigned long delta_utime = utime - last_utime; unsigned long delta_stime = stime - last_stime; *load = delta_utime + delta_stime; /* Cast to a float before division */ *load /= ticks_per_s; *load /= elapsed; crm_debug("cib load: %f (%lu ticks in %lds)", *load, delta_utime + delta_stime, (long) elapsed); } else { crm_debug("Init %lu + %lu ticks at %ld (%lu tps)", utime, stime, (long) now, ticks_per_s); } last_call = now; last_utime = utime; last_stime = stime; fclose(stream); return true; } fclose(stream); #endif // HAVE_LINUX_PROCFS return false; } bool pcmk__throttle_load_avg(float *load) { #if HAVE_LINUX_PROCFS char buffer[256]; FILE *stream = NULL; const char *loadfile = "/proc/loadavg"; if (load == NULL) { return false; } stream = fopen(loadfile, "r"); if (stream == NULL) { int rc = errno; crm_warn("Couldn't read %s: %s (%d)", loadfile, pcmk_rc_str(rc), rc); return false; } if (fgets(buffer, sizeof(buffer), stream) != NULL) { char *nl = strstr(buffer, "\n"); /* Grab the 1-minute average, ignore the rest */ *load = strtof(buffer, NULL); if (nl != NULL) { nl[0] = 0; } fclose(stream); return true; } fclose(stream); #endif // HAVE_LINUX_PROCFS return false; } diff --git a/lib/common/utils.c b/lib/common/utils.c index fe1595dbad..1f66c6dced 100644 --- a/lib/common/utils.c +++ b/lib/common/utils.c @@ -1,506 +1,448 @@ /* * Copyright 2004-2025 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 #include #include #include #include #include #include #include #include #include #include #include #include #include // xmlCleanupParser() #include #include "crmcommon_private.h" CRM_TRACE_INIT_DATA(common); bool pcmk__config_has_error = false; bool pcmk__config_has_warning = false; char *crm_system_name = NULL; /*! * \brief Free all memory used by libcrmcommon * * Free all global memory allocated by the libcrmcommon library. This should be * called before exiting a process that uses the library, and the process should * not call any libcrmcommon or libxml2 APIs after calling this one. */ void pcmk_common_cleanup(void) { // @TODO This isn't really everything, move all cleanup here mainloop_cleanup(); pcmk__schema_cleanup(); pcmk__free_common_logger(); free(crm_system_name); crm_system_name = NULL; // Clean up external library global state qb_log_fini(); // Don't log anything after this point xmlCleanupParser(); } bool pcmk__is_user_in_group(const char *user, const char *group) { struct group *grent; char **gr_mem; if (user == NULL || group == NULL) { return false; } setgrent(); while ((grent = getgrent()) != NULL) { if (grent->gr_mem == NULL) { continue; } if(strcmp(group, grent->gr_name) != 0) { continue; } gr_mem = grent->gr_mem; while (*gr_mem != NULL) { if (!strcmp(user, *gr_mem++)) { endgrent(); return true; } } } endgrent(); return false; } 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, PCMK__PW_BUFFER_LEN); if (buffer == NULL) { return -ENOMEM; } rc = getpwnam_r(name, &pwd, buffer, PCMK__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_ok; 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; } /*! * \internal * \brief Return the integer equivalent of a portion of a string * * \param[in] text Pointer to beginning of string portion * \param[out] end_text This will point to next character after integer */ static int version_helper(const char *text, const char **end_text) { int atoi_result = -1; pcmk__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 = version_helper(ver1_iter, &ver1_iter); } if (ver2_iter != NULL) { digit2 = 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; } -/*! - * \internal - * \brief Convert the current process to a daemon process - * - * Fork a child process, exit the parent, create a PID file with the current - * process ID, and close the standard input/output/error file descriptors. - * Exit instead if a daemon is already running and using the PID file. - * - * \param[in] name Daemon executable name - * \param[in] pidfile File name to use as PID file - */ -void -pcmk__daemonize(const char *name, const char *pidfile) -{ - int rc; - pid_t pid; - - /* Check before we even try... */ - 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 = pcmk__lock_pidfile(pidfile, name); - if (rc != pcmk_rc_ok) { - crm_err("Could not lock '%s' for %s: %s " QB_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); - pcmk__open_devnull(O_RDONLY); // stdin (fd 0) - - close(STDOUT_FILENO); - pcmk__open_devnull(O_WRONLY); // stdout (fd 1) - - close(STDERR_FILENO); - pcmk__open_devnull(O_WRONLY); // stderr (fd 2) -} - #ifdef HAVE_UUID_UUID_H # include #endif char * crm_generate_uuid(void) { unsigned char uuid[16]; char *buffer = malloc(37); /* Including NUL byte */ pcmk__mem_assert(buffer); uuid_generate(uuid); uuid_unparse(uuid, buffer); return buffer; } /*! * \internal * \brief Sleep for given milliseconds * * \param[in] ms Time to sleep * * \note The full time might not be slept if a signal is received. */ void pcmk__sleep_ms(unsigned int ms) { // @TODO Impose a sane maximum sleep to avoid hanging a process for long //CRM_CHECK(ms <= MAX_SLEEP, ms = MAX_SLEEP); // Use sleep() for any whole seconds if (ms >= 1000) { sleep(ms / 1000); ms -= ms / 1000; } if (ms == 0) { return; } #if defined(HAVE_NANOSLEEP) // nanosleep() is POSIX-2008, so prefer that { struct timespec req = { .tv_sec = 0, .tv_nsec = (long) (ms * 1000000) }; nanosleep(&req, NULL); } #elif defined(HAVE_USLEEP) // usleep() is widely available, though considered obsolete usleep((useconds_t) ms); #else // Otherwise use a trick with select() timeout { struct timeval tv = { .tv_sec = 0, .tv_usec = (suseconds_t) ms }; select(0, NULL, NULL, NULL, &tv); } #endif } /*! * \internal * \brief Add a timer * * \param[in] interval_ms The interval for the function to be called, in ms * \param[in] fn The function to be called * \param[in] data Data to be passed to fn (can be NULL) * * \return The ID of the event source */ guint pcmk__create_timer(guint interval_ms, GSourceFunc fn, gpointer data) { pcmk__assert(interval_ms != 0 && fn != NULL); if (interval_ms % 1000 == 0) { /* In case interval_ms is 0, the call to pcmk__timeout_ms2s ensures * an interval of one second. */ return g_timeout_add_seconds(pcmk__timeout_ms2s(interval_ms), fn, data); } else { return g_timeout_add(interval_ms, fn, data); } } /*! * \internal * \brief Convert milliseconds to seconds * * \param[in] timeout_ms The interval, in ms * * \return If \p timeout_ms is 0, return 0. Otherwise, return the number of * seconds, rounded to the nearest integer, with a minimum of 1. */ guint pcmk__timeout_ms2s(guint timeout_ms) { guint quot, rem; if (timeout_ms == 0) { return 0; } else if (timeout_ms < 1000) { return 1; } quot = timeout_ms / 1000; rem = timeout_ms % 1000; if (rem >= 500) { quot += 1; } return quot; } // Deprecated functions kept only for backward API compatibility // LCOV_EXCL_START #include // gnutls_global_init(), etc. #include static void _gnutls_log_func(int level, const char *msg) { crm_trace("%s", msg); } void crm_gnutls_global_init(void) { signal(SIGPIPE, SIG_IGN); gnutls_global_init(); gnutls_global_set_log_level(8); gnutls_global_set_log_function(_gnutls_log_func); } /*! * \brief Check whether string represents a client name used by cluster daemons * * \param[in] name String to check * * \return true if name is standard client name used by daemons, false otherwise * * \note This is provided by the client, and so cannot be used by itself as a * secure means of authentication. */ bool crm_is_daemon_name(const char *name) { return pcmk__str_any_of(name, "attrd", CRM_SYSTEM_CIB, CRM_SYSTEM_CRMD, CRM_SYSTEM_DC, CRM_SYSTEM_LRMD, CRM_SYSTEM_MCP, CRM_SYSTEM_PENGINE, CRM_SYSTEM_TENGINE, "pacemaker-attrd", "pacemaker-based", "pacemaker-controld", "pacemaker-execd", "pacemaker-fenced", "pacemaker-remoted", "pacemaker-schedulerd", "stonith-ng", "stonithd", NULL); } // LCOV_EXCL_STOP // End deprecated API diff --git a/lib/common/watchdog.c b/lib/common/watchdog.c index 7342df8d20..35bc21a0fa 100644 --- a/lib/common/watchdog.c +++ b/lib/common/watchdog.c @@ -1,297 +1,310 @@ /* * Copyright 2013-2025 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 #include // g_str_has_prefix() static pid_t sbd_pid = 0; /*! * \internal * \brief Tell pacemakerd to panic the local host * * \param[in] ppid Process ID of parent process */ static void panic_local_nonroot(pid_t ppid) { if (ppid > 1) { // pacemakerd is still our parent crm_emerg("Escalating panic to " PCMK__SERVER_PACEMAKERD "[%lld]", (long long) ppid); } else { // Signal (non-parent) pacemakerd if possible ppid = pcmk__procfs_pid_of(PCMK__SERVER_PACEMAKERD); if (ppid > 0) { union sigval signal_value; crm_emerg("Signaling " PCMK__SERVER_PACEMAKERD "[%lld] to panic", (long long) ppid); memset(&signal_value, 0, sizeof(signal_value)); if (sigqueue(ppid, SIGQUIT, signal_value) < 0) { crm_emerg("Exiting after signal failure: %s", strerror(errno)); } } else { crm_emerg("Exiting with no known " PCMK__SERVER_PACEMAKERD "process"); } } crm_exit(CRM_EX_PANIC); } /*! * \internal * \brief Panic the local host (if root) or tell pacemakerd to do so */ static void panic_local(void) { const char *full_panic_action = pcmk__env_option(PCMK__ENV_PANIC_ACTION); const char *panic_action = full_panic_action; int reboot_cmd = RB_AUTOBOOT; // Default panic action is reboot if (geteuid() != 0) { // Non-root caller such as the controller panic_local_nonroot(getppid()); return; } if ((full_panic_action != NULL) && g_str_has_prefix(full_panic_action, "sync-")) { panic_action += sizeof("sync-") - 1; sync(); } if (pcmk__str_empty(full_panic_action) || pcmk__str_eq(panic_action, PCMK_VALUE_REBOOT, pcmk__str_none)) { pcmk__sysrq_trigger('b'); } else if (pcmk__str_eq(panic_action, PCMK_VALUE_CRASH, pcmk__str_none)) { pcmk__sysrq_trigger('c'); } else if (pcmk__str_eq(panic_action, PCMK_VALUE_OFF, pcmk__str_none)) { pcmk__sysrq_trigger('o'); #ifdef RB_POWER_OFF reboot_cmd = RB_POWER_OFF; #elif defined(RB_POWEROFF) reboot_cmd = RB_POWEROFF; #endif } else { crm_warn("Using default '" PCMK_VALUE_REBOOT "' for local option PCMK_" PCMK__ENV_PANIC_ACTION " because '%s' is not a valid value", full_panic_action); pcmk__sysrq_trigger('b'); } // sysrq failed or is not supported on this platform, so fall back to reboot reboot(reboot_cmd); // Even reboot failed, nothing left to do but exit crm_emerg("Exiting after reboot failed: %s", strerror(errno)); if (getppid() > 1) { // pacemakerd is parent process crm_exit(CRM_EX_PANIC); } else { // This is pacemakerd, or an orphaned subdaemon crm_exit(CRM_EX_FATAL); } } /*! * \internal * \brief Tell sbd to kill the local host, then exit */ static void panic_sbd(void) { union sigval signal_value; pid_t ppid = getppid(); memset(&signal_value, 0, sizeof(signal_value)); /* TODO: Arrange for a slightly less brutal option? */ if(sigqueue(sbd_pid, SIGKILL, signal_value) < 0) { crm_emerg("Panicking directly because couldn't signal sbd"); panic_local(); } if(ppid > 1) { /* child daemon */ crm_exit(CRM_EX_PANIC); } else { /* pacemakerd or orphan child */ crm_exit(CRM_EX_FATAL); } } /*! * \internal * \brief Panic the local host * * Panic the local host either by sbd (if running), directly, or by asking * pacemakerd. If trace logging this function, exit instead. * * \param[in] reason Why panic is needed (for logging only) */ void pcmk__panic(const char *reason) { if (pcmk__locate_sbd() > 1) { crm_emerg("Signaling sbd[%lld] to panic the system: %s", (long long) sbd_pid, reason); panic_sbd(); } else { crm_emerg("Panicking the system directly: %s", reason); panic_local(); } } +#define PIDFILE PCMK__RUN_DIR "/sbd.pid" + /*! * \internal * \brief Return the process ID of sbd (or 0 if it is not running) */ pid_t pcmk__locate_sbd(void) { - const char *pidfile = PCMK__RUN_DIR "/sbd.pid"; - int rc; + gchar *contents = NULL; + long long pid_read = 0; - if(sbd_pid > 1) { + if (sbd_pid > 1) { return sbd_pid; } - /* Read the pid file */ - rc = pcmk__pidfile_matches(pidfile, 0, SBIN_DIR "/sbd", &sbd_pid); - if (rc == pcmk_rc_ok) { - crm_trace("SBD detected at pid %lld (via PID file %s)", - (long long) sbd_pid, pidfile); + if (g_file_get_contents(PIDFILE, &contents, NULL, NULL) + && (pcmk__scan_ll(contents, &pid_read, 0) == pcmk_rc_ok) + && (pcmk__pid_active((pid_t) pid_read, + SBIN_DIR "/sbd") != ESRCH)) { + + /* If the pcmk__pid_active() return code is neither pcmk__rc_ok nor + * ESRCH, then we couldn't determine whether the PID belongs to + * SBIN_DIR "/sbd". In that case, we assume that it does. + * + * @TODO Make sure that's what we want to do. + */ + crm_trace("SBD detected at pid %lld (via PID file " PIDFILE ")", + pid_read); + sbd_pid = (pid_t) pid_read; + } else { - /* Fall back to /proc for systems that support it */ - sbd_pid = pcmk__procfs_pid_of("sbd"); + unlink(PIDFILE); + // Fall back to /proc for systems that support it + sbd_pid = pcmk__procfs_pid_of("sbd"); if (sbd_pid != 0) { crm_trace("SBD detected at pid %lld (via procfs)", (long long) sbd_pid); } } - if(sbd_pid < 0) { + if (sbd_pid <= 0) { sbd_pid = 0; crm_trace("SBD not detected"); } - + g_free(contents); return sbd_pid; } long pcmk__get_sbd_watchdog_timeout(void) { static long sbd_timeout = -2; if (sbd_timeout == -2) { sbd_timeout = crm_get_msec(getenv("SBD_WATCHDOG_TIMEOUT")); } return sbd_timeout; } bool pcmk__get_sbd_sync_resource_startup(void) { static int sync_resource_startup = PCMK__SBD_SYNC_DEFAULT; static bool checked_sync_resource_startup = false; if (!checked_sync_resource_startup) { const char *sync_env = getenv("SBD_SYNC_RESOURCE_STARTUP"); if (sync_env == NULL) { crm_trace("Defaulting to %sstart-up synchronization with sbd", (PCMK__SBD_SYNC_DEFAULT? "" : "no ")); } else if (crm_str_to_boolean(sync_env, &sync_resource_startup) < 0) { crm_warn("Defaulting to %sstart-up synchronization with sbd " "because environment value '%s' is invalid", (PCMK__SBD_SYNC_DEFAULT? "" : "no "), sync_env); } checked_sync_resource_startup = true; } return sync_resource_startup != 0; } long pcmk__auto_stonith_watchdog_timeout(void) { long sbd_timeout = pcmk__get_sbd_watchdog_timeout(); return (sbd_timeout <= 0)? 0 : (2 * sbd_timeout); } bool pcmk__valid_stonith_watchdog_timeout(const char *value) { /* @COMPAT At a compatibility break, accept either negative values or a * specific string like "auto" (but not both) to mean "auto-calculate the * timeout." Reject other values that aren't parsable as timeouts. */ long long st_timeout = 0; if (value != NULL) { /* @COMPAT So far it has been documented that a negative value is * valid. Parse it as an integer first to avoid the warning from * crm_get_msec(). */ int rc = pcmk__scan_ll(value, &st_timeout, PCMK__PARSE_INT_DEFAULT); if (rc != pcmk_rc_ok || st_timeout >= 0) { st_timeout = crm_get_msec(value); } } if (st_timeout < 0) { st_timeout = pcmk__auto_stonith_watchdog_timeout(); crm_debug("Using calculated value %lld for " PCMK_OPT_STONITH_WATCHDOG_TIMEOUT " (%s)", st_timeout, value); } if (st_timeout == 0) { crm_debug("Watchdog may be enabled but " PCMK_OPT_STONITH_WATCHDOG_TIMEOUT " is disabled (%s)", value? value : "default"); } else if (pcmk__locate_sbd() == 0) { crm_emerg("Shutting down: " PCMK_OPT_STONITH_WATCHDOG_TIMEOUT " configured (%s) but SBD not active", pcmk__s(value, "auto")); crm_exit(CRM_EX_FATAL); return false; } else { long sbd_timeout = pcmk__get_sbd_watchdog_timeout(); if (st_timeout < sbd_timeout) { crm_emerg("Shutting down: " PCMK_OPT_STONITH_WATCHDOG_TIMEOUT " (%s) too short (must be >%ldms)", value, sbd_timeout); crm_exit(CRM_EX_FATAL); return false; } crm_info("Watchdog configured with " PCMK_OPT_STONITH_WATCHDOG_TIMEOUT " %s and SBD timeout %ldms", value, sbd_timeout); } return true; } diff --git a/lib/common/xml_io.c b/lib/common/xml_io.c index acb0fc3eb9..c092aedd69 100644 --- a/lib/common/xml_io.c +++ b/lib/common/xml_io.c @@ -1,619 +1,619 @@ /* * Copyright 2004-2025 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 // xmlOutputBuffer* #include // xmlChar #include #include #include #include "crmcommon_private.h" /*! * \internal * \brief Decompress a bzip2-compressed file into a string buffer * * \param[in] filename Name of file to decompress * * \return Newly allocated string with the decompressed contents of \p filename, * or \c NULL on error. * * \note The caller is responsible for freeing the return value using \c free(). */ static char * decompress_file(const char *filename) { char *buffer = NULL; int rc = pcmk_rc_ok; size_t length = 0; BZFILE *bz_file = NULL; FILE *input = fopen(filename, "r"); if (input == NULL) { - crm_perror(LOG_ERR, "Could not open %s for reading", filename); + crm_err("Could not open %s for reading: %s", filename, strerror(errno)); return NULL; } bz_file = BZ2_bzReadOpen(&rc, input, 0, 0, NULL, 0); rc = pcmk__bzlib2rc(rc); if (rc != pcmk_rc_ok) { crm_err("Could not prepare to read compressed %s: %s " QB_XS " rc=%d", filename, pcmk_rc_str(rc), rc); goto done; } do { int read_len = 0; buffer = pcmk__realloc(buffer, length + PCMK__BUFFER_SIZE + 1); read_len = BZ2_bzRead(&rc, bz_file, buffer + length, PCMK__BUFFER_SIZE); if ((rc == BZ_OK) || (rc == BZ_STREAM_END)) { crm_trace("Read %ld bytes from file: %d", (long) read_len, rc); length += read_len; } } while (rc == BZ_OK); rc = pcmk__bzlib2rc(rc); if (rc != pcmk_rc_ok) { rc = pcmk__bzlib2rc(rc); crm_err("Could not read compressed %s: %s " QB_XS " rc=%d", filename, pcmk_rc_str(rc), rc); free(buffer); buffer = NULL; } else { buffer[length] = '\0'; } done: BZ2_bzReadClose(&rc, bz_file); fclose(input); return buffer; } /*! * \internal * \brief Parse XML from a file * * \param[in] filename Name of file containing XML (\c NULL or \c "-" for * \c stdin); if \p filename ends in \c ".bz2", the file * will be decompressed using \c bzip2 * * \return XML tree parsed from the given file on success, otherwise \c NULL */ xmlNode * pcmk__xml_read(const char *filename) { bool use_stdin = pcmk__str_eq(filename, "-", pcmk__str_null_matches); xmlNode *xml = NULL; xmlDoc *output = NULL; xmlParserCtxt *ctxt = NULL; const xmlError *last_error = NULL; // Create a parser context ctxt = xmlNewParserCtxt(); CRM_CHECK(ctxt != NULL, return NULL); xmlCtxtResetLastError(ctxt); xmlSetGenericErrorFunc(ctxt, pcmk__log_xmllib_err); if (use_stdin) { output = xmlCtxtReadFd(ctxt, STDIN_FILENO, NULL, NULL, XML_PARSE_NOBLANKS); } else if (g_str_has_suffix(filename, ".bz2")) { char *input = decompress_file(filename); if (input != NULL) { output = xmlCtxtReadDoc(ctxt, (const xmlChar *) input, NULL, NULL, XML_PARSE_NOBLANKS); free(input); } } else { output = xmlCtxtReadFile(ctxt, filename, NULL, XML_PARSE_NOBLANKS); } if (output != NULL) { pcmk__xml_new_private_data((xmlNode *) output); xml = xmlDocGetRootElement(output); if (xml != NULL) { /* @TODO Should we really be stripping out text? This seems like an * overly broad way to get rid of whitespace, if that's the goal. * Text nodes may be invalid in most or all Pacemaker inputs, but * stripping them in a generic "parse XML from file" function may * not be the best way to ignore them. */ pcmk__strip_xml_text(xml); } } last_error = xmlCtxtGetLastError(ctxt); if ((last_error != NULL) && (xml != NULL)) { crm_log_xml_debug(xml, "partial"); pcmk__xml_free(xml); xml = NULL; } xmlFreeParserCtxt(ctxt); return xml; } /*! * \internal * \brief Parse XML from a string * * \param[in] input String to parse * * \return XML tree parsed from the given string on success, otherwise \c NULL */ xmlNode * pcmk__xml_parse(const char *input) { xmlNode *xml = NULL; xmlDoc *output = NULL; xmlParserCtxt *ctxt = NULL; const xmlError *last_error = NULL; if (input == NULL) { return NULL; } ctxt = xmlNewParserCtxt(); if (ctxt == NULL) { return NULL; } xmlCtxtResetLastError(ctxt); xmlSetGenericErrorFunc(ctxt, pcmk__log_xmllib_err); output = xmlCtxtReadDoc(ctxt, (const xmlChar *) input, NULL, NULL, XML_PARSE_NOBLANKS); if (output != NULL) { pcmk__xml_new_private_data((xmlNode *) output); xml = xmlDocGetRootElement(output); } last_error = xmlCtxtGetLastError(ctxt); if ((last_error != NULL) && (xml != NULL)) { crm_log_xml_debug(xml, "partial"); pcmk__xml_free(xml); xml = NULL; } xmlFreeParserCtxt(ctxt); return xml; } /*! * \internal * \brief Append a string representation of an XML element to a buffer * * \param[in] data XML whose representation to append * \param[in] options Group of \p pcmk__xml_fmt_options flags * \param[in,out] buffer Where to append the content (must not be \p NULL) * \param[in] depth Current indentation level */ static void dump_xml_element(const xmlNode *data, uint32_t options, GString *buffer, int depth) { bool pretty = pcmk_is_set(options, pcmk__xml_fmt_pretty); bool filtered = pcmk_is_set(options, pcmk__xml_fmt_filtered); int spaces = pretty? (2 * depth) : 0; for (int lpc = 0; lpc < spaces; lpc++) { g_string_append_c(buffer, ' '); } pcmk__g_strcat(buffer, "<", data->name, NULL); for (const xmlAttr *attr = pcmk__xe_first_attr(data); attr != NULL; attr = attr->next) { if (!filtered || !pcmk__xa_filterable((const char *) (attr->name))) { pcmk__dump_xml_attr(attr, buffer); } } if (data->children == NULL) { g_string_append(buffer, "/>"); } else { g_string_append_c(buffer, '>'); } if (pretty) { g_string_append_c(buffer, '\n'); } if (data->children) { for (const xmlNode *child = data->children; child != NULL; child = child->next) { pcmk__xml_string(child, options, buffer, depth + 1); } for (int lpc = 0; lpc < spaces; lpc++) { g_string_append_c(buffer, ' '); } pcmk__g_strcat(buffer, "name, ">", NULL); if (pretty) { g_string_append_c(buffer, '\n'); } } } /*! * \internal * \brief Append XML text content to a buffer * * \param[in] data XML whose content to append * \param[in] options Group of enum pcmk__xml_fmt_options * \param[in,out] buffer Where to append the content (must not be \p NULL) * \param[in] depth Current indentation level */ static void dump_xml_text(const xmlNode *data, uint32_t options, GString *buffer, int depth) { bool pretty = pcmk_is_set(options, pcmk__xml_fmt_pretty); int spaces = pretty? (2 * depth) : 0; const char *content = (const char *) data->content; gchar *content_esc = NULL; if (pcmk__xml_needs_escape(content, pcmk__xml_escape_text)) { content_esc = pcmk__xml_escape(content, pcmk__xml_escape_text); content = content_esc; } for (int lpc = 0; lpc < spaces; lpc++) { g_string_append_c(buffer, ' '); } g_string_append(buffer, content); if (pretty) { g_string_append_c(buffer, '\n'); } g_free(content_esc); } /*! * \internal * \brief Append XML CDATA content to a buffer * * \param[in] data XML whose content to append * \param[in] options Group of \p pcmk__xml_fmt_options flags * \param[in,out] buffer Where to append the content (must not be \p NULL) * \param[in] depth Current indentation level */ static void dump_xml_cdata(const xmlNode *data, uint32_t options, GString *buffer, int depth) { bool pretty = pcmk_is_set(options, pcmk__xml_fmt_pretty); int spaces = pretty? (2 * depth) : 0; for (int lpc = 0; lpc < spaces; lpc++) { g_string_append_c(buffer, ' '); } pcmk__g_strcat(buffer, "content, "]]>", NULL); if (pretty) { g_string_append_c(buffer, '\n'); } } /*! * \internal * \brief Append an XML comment to a buffer * * \param[in] data XML whose content to append * \param[in] options Group of \p pcmk__xml_fmt_options flags * \param[in,out] buffer Where to append the content (must not be \p NULL) * \param[in] depth Current indentation level */ static void dump_xml_comment(const xmlNode *data, uint32_t options, GString *buffer, int depth) { bool pretty = pcmk_is_set(options, pcmk__xml_fmt_pretty); int spaces = pretty? (2 * depth) : 0; for (int lpc = 0; lpc < spaces; lpc++) { g_string_append_c(buffer, ' '); } pcmk__g_strcat(buffer, "", NULL); if (pretty) { g_string_append_c(buffer, '\n'); } } /*! * \internal * \brief Create a string representation of an XML object * * libxml2's \c xmlNodeDumpOutput() doesn't allow filtering, doesn't escape * special characters thoroughly, and doesn't allow a const argument. * * \param[in] data XML to convert * \param[in] options Group of \p pcmk__xml_fmt_options flags * \param[in,out] buffer Where to store the text (must not be \p NULL) * \param[in] depth Current indentation level * * \todo Create a wrapper that doesn't require \p depth. Only used with * recursive calls currently. */ void pcmk__xml_string(const xmlNode *data, uint32_t options, GString *buffer, int depth) { if (data == NULL) { crm_trace("Nothing to dump"); return; } pcmk__assert(buffer != NULL); CRM_CHECK(depth >= 0, depth = 0); switch(data->type) { case XML_ELEMENT_NODE: /* Handle below */ dump_xml_element(data, options, buffer, depth); break; case XML_TEXT_NODE: if (pcmk_is_set(options, pcmk__xml_fmt_text)) { dump_xml_text(data, options, buffer, depth); } break; case XML_COMMENT_NODE: dump_xml_comment(data, options, buffer, depth); break; case XML_CDATA_SECTION_NODE: dump_xml_cdata(data, options, buffer, depth); break; default: crm_warn("Cannot convert XML %s node to text " QB_XS " type=%d", pcmk__xml_element_type_text(data->type), data->type); break; } } /*! * \internal * \brief Write a string to a file stream, compressed using \c bzip2 * * \param[in] text String to write * \param[in] filename Name of file being written (for logging only) * \param[in,out] stream Open file stream to write to * \param[out] bytes_out Number of bytes written (valid only on success) * * \return Standard Pacemaker return code */ static int write_compressed_stream(char *text, const char *filename, FILE *stream, unsigned int *bytes_out) { unsigned int bytes_in = 0; int rc = pcmk_rc_ok; // (5, 0, 0): (intermediate block size, silent, default workFactor) BZFILE *bz_file = BZ2_bzWriteOpen(&rc, stream, 5, 0, 0); rc = pcmk__bzlib2rc(rc); if (rc != pcmk_rc_ok) { crm_warn("Not compressing %s: could not prepare file stream: %s " QB_XS " rc=%d", filename, pcmk_rc_str(rc), rc); goto done; } BZ2_bzWrite(&rc, bz_file, text, strlen(text)); rc = pcmk__bzlib2rc(rc); if (rc != pcmk_rc_ok) { crm_warn("Not compressing %s: could not compress data: %s " QB_XS " rc=%d errno=%d", filename, pcmk_rc_str(rc), rc, errno); goto done; } BZ2_bzWriteClose(&rc, bz_file, 0, &bytes_in, bytes_out); bz_file = NULL; rc = pcmk__bzlib2rc(rc); if (rc != pcmk_rc_ok) { crm_warn("Not compressing %s: could not write compressed data: %s " QB_XS " rc=%d errno=%d", filename, pcmk_rc_str(rc), rc, errno); goto done; } crm_trace("Compressed XML for %s from %u bytes to %u", filename, bytes_in, *bytes_out); done: if (bz_file != NULL) { BZ2_bzWriteClose(&rc, bz_file, 0, NULL, NULL); } return rc; } /*! * \internal * \brief Write XML to a file stream * * \param[in] xml XML to write * \param[in] filename Name of file being written (for logging only) * \param[in,out] stream Open file stream corresponding to filename (closed * when this function returns) * \param[in] compress Whether to compress XML before writing * * \return Standard Pacemaker return code */ static int write_xml_stream(const xmlNode *xml, const char *filename, FILE *stream, bool compress) { GString *buffer = g_string_sized_new(1024); unsigned int bytes_out = 0; int rc = pcmk_rc_ok; pcmk__xml_string(xml, pcmk__xml_fmt_pretty, buffer, 0); CRM_CHECK(!pcmk__str_empty(buffer->str), crm_log_xml_info(xml, "dump-failed"); rc = pcmk_rc_error; goto done); crm_log_xml_trace(xml, "writing"); if (compress && (write_compressed_stream(buffer->str, filename, stream, &bytes_out) == pcmk_rc_ok)) { goto done; } rc = fprintf(stream, "%s", buffer->str); if (rc < 0) { rc = EIO; - crm_perror(LOG_ERR, "writing %s", filename); + crm_err("Error writing %s", filename); goto done; } bytes_out = (unsigned int) rc; rc = pcmk_rc_ok; done: if (fflush(stream) != 0) { rc = errno; - crm_perror(LOG_ERR, "flushing %s", filename); + crm_err("Error flushing %s: %s", filename, strerror(errno)); } // Don't report error if the file does not support synchronization if ((fsync(fileno(stream)) < 0) && (errno != EROFS) && (errno != EINVAL)) { rc = errno; - crm_perror(LOG_ERR, "synchronizing %s", filename); + crm_err("Error synchronizing %s: %s", filename, strerror(errno)); } fclose(stream); crm_trace("Saved %u bytes to %s as XML", bytes_out, filename); g_string_free(buffer, TRUE); return rc; } /*! * \internal * \brief Write XML to a file descriptor * * \param[in] xml XML to write * \param[in] filename Name of file being written (for logging only) * \param[in] fd Open file descriptor corresponding to \p filename * * \return Standard Pacemaker return code */ int pcmk__xml_write_fd(const xmlNode *xml, const char *filename, int fd) { FILE *stream = NULL; CRM_CHECK((xml != NULL) && (fd > 0), return EINVAL); stream = fdopen(fd, "w"); if (stream == NULL) { return errno; } return write_xml_stream(xml, pcmk__s(filename, "unnamed file"), stream, false); } /*! * \internal * \brief Write XML to a file * * \param[in] xml XML to write * \param[in] filename Name of file to write * \param[in] compress If \c true, compress XML before writing * * \return Standard Pacemaker return code */ int pcmk__xml_write_file(const xmlNode *xml, const char *filename, bool compress) { FILE *stream = NULL; CRM_CHECK((xml != NULL) && (filename != NULL), return EINVAL); stream = fopen(filename, "w"); if (stream == NULL) { return errno; } return write_xml_stream(xml, filename, stream, compress); } /*! * \internal * \brief Serialize XML (using libxml) into provided descriptor * * \param[in] fd File descriptor to (piece-wise) write to * \param[in] cur XML subtree to proceed * * \return a standard Pacemaker return code */ int pcmk__xml2fd(int fd, xmlNode *cur) { bool success; xmlOutputBuffer *fd_out = xmlOutputBufferCreateFd(fd, NULL); pcmk__mem_assert(fd_out); xmlNodeDumpOutput(fd_out, cur->doc, cur, 0, pcmk__xml_fmt_pretty, NULL); success = xmlOutputBufferWrite(fd_out, sizeof("\n") - 1, "\n") != -1; success = xmlOutputBufferClose(fd_out) != -1 && success; if (!success) { return EIO; } fsync(fd); return pcmk_rc_ok; } void save_xml_to_file(const xmlNode *xml, const char *desc, const char *filename) { char *f = NULL; if (filename == NULL) { char *uuid = crm_generate_uuid(); f = crm_strdup_printf("%s/%s", pcmk__get_tmpdir(), uuid); filename = f; free(uuid); } crm_info("Saving %s to %s", desc, filename); pcmk__xml_write_file(xml, filename, false); free(f); } diff --git a/lib/fencing/st_client.c b/lib/fencing/st_client.c index 6278579b8a..ebaa74e5a3 100644 --- a/lib/fencing/st_client.c +++ b/lib/fencing/st_client.c @@ -1,2954 +1,2956 @@ /* * Copyright 2004-2025 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 // xmlNode #include // xmlXPathObject, etc. #include #include #include #include #include #include "fencing_private.h" CRM_TRACE_INIT_DATA(stonith); // Used as stonith_t:st_private typedef struct stonith_private_s { char *token; crm_ipc_t *ipc; mainloop_io_t *source; GHashTable *stonith_op_callback_table; GList *notify_list; int notify_refcnt; bool notify_deletes; void (*op_callback) (stonith_t * st, stonith_callback_data_t * data); } stonith_private_t; // Used as stonith_event_t:opaque struct event_private { pcmk__action_result_t result; }; typedef struct stonith_notify_client_s { const char *event; const char *obj_id; /* implement one day */ const char *obj_type; /* implement one day */ void (*notify) (stonith_t * st, stonith_event_t * e); bool delete; } stonith_notify_client_t; typedef struct stonith_callback_client_s { void (*callback) (stonith_t * st, stonith_callback_data_t * data); const char *id; void *user_data; gboolean only_success; gboolean allow_timeout_updates; struct timer_rec_s *timer; } stonith_callback_client_t; struct notify_blob_s { stonith_t *stonith; xmlNode *xml; }; struct timer_rec_s { int call_id; int timeout; guint ref; stonith_t *stonith; }; typedef int (*stonith_op_t) (const char *, int, const char *, xmlNode *, xmlNode *, xmlNode *, xmlNode **, xmlNode **); xmlNode *stonith_create_op(int call_id, const char *token, const char *op, xmlNode * data, int call_options); static int stonith_send_command(stonith_t *stonith, const char *op, xmlNode *data, xmlNode **output_data, int call_options, int timeout); static void stonith_connection_destroy(gpointer user_data); static void stonith_send_notification(gpointer data, gpointer user_data); static int stonith_api_del_notification(stonith_t *stonith, const char *event); /*! * \internal * \brief Parse fence agent namespace from a string * * \param[in] namespace_s Name of namespace as string * * \return enum value parsed from \p namespace_s */ static enum stonith_namespace parse_namespace(const char *namespace_s) { if (pcmk__str_eq(namespace_s, "any", pcmk__str_null_matches)) { return st_namespace_any; } /* @TODO Is "redhat" still necessary except for stonith_text2namespace() * backward compatibility? */ if (pcmk__str_any_of(namespace_s, "redhat", "stonith-ng", NULL)) { return st_namespace_rhcs; } if (pcmk__str_eq(namespace_s, "internal", pcmk__str_none)) { return st_namespace_internal; } if (pcmk__str_eq(namespace_s, "heartbeat", pcmk__str_none)) { return st_namespace_lha; } return st_namespace_invalid; } /*! * \internal * \brief Get name of a fence agent namespace as a string * * \param[in] st_namespace Namespace as enum value * * \return Name of \p st_namespace as a string */ static const char * namespace_text(enum stonith_namespace st_namespace) { switch (st_namespace) { case st_namespace_any: return "any"; case st_namespace_rhcs: return "stonith-ng"; case st_namespace_internal: return "internal"; case st_namespace_lha: return "heartbeat"; default: return "unsupported"; } } /*! * \internal * \brief Determine fence agent namespace from agent name * * This involves external checks (for example, checking the existence of a file * or calling an external library function). * * \param[in] agent Fence agent name * * \return Namespace to which \p agent belongs, or \c st_namespace_invalid if * not found */ static enum stonith_namespace get_namespace_from_agent(const char *agent) { if (stonith__agent_is_rhcs(agent)) { return st_namespace_rhcs; } #if HAVE_STONITH_STONITH_H if (stonith__agent_is_lha(agent)) { return st_namespace_lha; } #endif // HAVE_STONITH_STONITH_H return st_namespace_invalid; } gboolean stonith__watchdog_fencing_enabled_for_node_api(stonith_t *st, const char *node) { gboolean rv = FALSE; stonith_t *stonith_api = (st != NULL)? st : stonith__api_new(); char *list = NULL; if(stonith_api) { if (stonith_api->state == stonith_disconnected) { int rc = stonith_api->cmds->connect(stonith_api, "stonith-api", NULL); if (rc != pcmk_ok) { crm_err("Failed connecting to Stonith-API for watchdog-fencing-query."); } } if (stonith_api->state != stonith_disconnected) { /* caveat!!! * this might fail when when stonithd is just updating the device-list * probably something we should fix as well for other api-calls */ int rc = stonith_api->cmds->list(stonith_api, st_opt_sync_call, STONITH_WATCHDOG_ID, &list, 0); if ((rc != pcmk_ok) || (list == NULL)) { /* due to the race described above it can happen that * we drop in here - so as not to make remote nodes * panic on that answer */ if (rc == -ENODEV) { crm_notice("Cluster does not have watchdog fencing device"); } else { crm_warn("Could not check for watchdog fencing device: %s", pcmk_strerror(rc)); } } else if (list[0] == '\0') { rv = TRUE; } else { GList *targets = stonith__parse_targets(list); rv = pcmk__str_in_list(node, targets, pcmk__str_casei); g_list_free_full(targets, free); } free(list); if (!st) { /* if we're provided the api we still might have done the * connection - but let's assume the caller won't bother */ stonith_api->cmds->disconnect(stonith_api); } } if (!st) { stonith__api_free(stonith_api); } } else { crm_err("Stonith-API for watchdog-fencing-query couldn't be created."); } crm_trace("Pacemaker assumes node %s %sto do watchdog-fencing.", node, rv?"":"not "); return rv; } gboolean stonith__watchdog_fencing_enabled_for_node(const char *node) { return stonith__watchdog_fencing_enabled_for_node_api(NULL, node); } /* when cycling through the list we don't want to delete items so just mark them and when we know nobody is using the list loop over it to remove the marked items */ static void foreach_notify_entry (stonith_private_t *private, GFunc func, gpointer user_data) { private->notify_refcnt++; g_list_foreach(private->notify_list, func, user_data); private->notify_refcnt--; if ((private->notify_refcnt == 0) && private->notify_deletes) { GList *list_item = private->notify_list; private->notify_deletes = FALSE; while (list_item != NULL) { stonith_notify_client_t *list_client = list_item->data; GList *next = g_list_next(list_item); if (list_client->delete) { free(list_client); private->notify_list = g_list_delete_link(private->notify_list, list_item); } list_item = next; } } } static void stonith_connection_destroy(gpointer user_data) { stonith_t *stonith = user_data; stonith_private_t *native = NULL; struct notify_blob_s blob; crm_trace("Sending destroyed notification"); blob.stonith = stonith; blob.xml = pcmk__xe_create(NULL, PCMK__XE_NOTIFY); native = stonith->st_private; native->ipc = NULL; native->source = NULL; free(native->token); native->token = NULL; stonith->state = stonith_disconnected; crm_xml_add(blob.xml, PCMK__XA_T, PCMK__VALUE_ST_NOTIFY); crm_xml_add(blob.xml, PCMK__XA_SUBT, PCMK__VALUE_ST_NOTIFY_DISCONNECT); foreach_notify_entry(native, stonith_send_notification, &blob); pcmk__xml_free(blob.xml); } xmlNode * create_device_registration_xml(const char *id, enum stonith_namespace standard, const char *agent, const stonith_key_value_t *params, const char *rsc_provides) { xmlNode *data = pcmk__xe_create(NULL, PCMK__XE_ST_DEVICE_ID); xmlNode *args = pcmk__xe_create(data, PCMK__XE_ATTRIBUTES); #if HAVE_STONITH_STONITH_H if (standard == st_namespace_any) { standard = get_namespace_from_agent(agent); } if (standard == st_namespace_lha) { hash2field((gpointer) "plugin", (gpointer) agent, args); agent = "fence_legacy"; } #endif crm_xml_add(data, PCMK_XA_ID, id); crm_xml_add(data, PCMK__XA_ST_ORIGIN, __func__); crm_xml_add(data, PCMK_XA_AGENT, agent); if ((standard != st_namespace_any) && (standard != st_namespace_invalid)) { crm_xml_add(data, PCMK__XA_NAMESPACE, namespace_text(standard)); } if (rsc_provides) { crm_xml_add(data, PCMK__XA_RSC_PROVIDES, rsc_provides); } for (; params; params = params->next) { hash2field((gpointer) params->key, (gpointer) params->value, args); } return data; } static int stonith_api_register_device(stonith_t *st, int call_options, const char *id, const char *namespace_s, const char *agent, const stonith_key_value_t *params) { int rc = 0; xmlNode *data = NULL; data = create_device_registration_xml(id, parse_namespace(namespace_s), agent, params, NULL); rc = stonith_send_command(st, STONITH_OP_DEVICE_ADD, data, NULL, call_options, 0); pcmk__xml_free(data); return rc; } static int stonith_api_remove_device(stonith_t * st, int call_options, const char *name) { int rc = 0; xmlNode *data = NULL; data = pcmk__xe_create(NULL, PCMK__XE_ST_DEVICE_ID); crm_xml_add(data, PCMK__XA_ST_ORIGIN, __func__); crm_xml_add(data, PCMK_XA_ID, name); rc = stonith_send_command(st, STONITH_OP_DEVICE_DEL, data, NULL, call_options, 0); pcmk__xml_free(data); return rc; } static int stonith_api_remove_level_full(stonith_t *st, int options, const char *node, const char *pattern, const char *attr, const char *value, int level) { int rc = 0; xmlNode *data = NULL; CRM_CHECK(node || pattern || (attr && value), return -EINVAL); data = pcmk__xe_create(NULL, PCMK_XE_FENCING_LEVEL); crm_xml_add(data, PCMK__XA_ST_ORIGIN, __func__); if (node) { crm_xml_add(data, PCMK_XA_TARGET, node); } else if (pattern) { crm_xml_add(data, PCMK_XA_TARGET_PATTERN, pattern); } else { crm_xml_add(data, PCMK_XA_TARGET_ATTRIBUTE, attr); crm_xml_add(data, PCMK_XA_TARGET_VALUE, value); } crm_xml_add_int(data, PCMK_XA_INDEX, level); rc = stonith_send_command(st, STONITH_OP_LEVEL_DEL, data, NULL, options, 0); pcmk__xml_free(data); return rc; } static int stonith_api_remove_level(stonith_t * st, int options, const char *node, int level) { return stonith_api_remove_level_full(st, options, node, NULL, NULL, NULL, level); } /*! * \internal * \brief Create XML for fence topology level registration request * * \param[in] node If not NULL, target level by this node name * \param[in] pattern If not NULL, target by node name using this regex * \param[in] attr If not NULL, target by this node attribute * \param[in] value If not NULL, target by this node attribute value * \param[in] level Index number of level to register * \param[in] device_list List of devices in level * * \return Newly allocated XML tree on success, NULL otherwise * * \note The caller should set only one of node, pattern or attr/value. */ xmlNode * create_level_registration_xml(const char *node, const char *pattern, const char *attr, const char *value, int level, const stonith_key_value_t *device_list) { GString *list = NULL; xmlNode *data; CRM_CHECK(node || pattern || (attr && value), return NULL); data = pcmk__xe_create(NULL, PCMK_XE_FENCING_LEVEL); crm_xml_add(data, PCMK__XA_ST_ORIGIN, __func__); crm_xml_add_int(data, PCMK_XA_ID, level); crm_xml_add_int(data, PCMK_XA_INDEX, level); if (node) { crm_xml_add(data, PCMK_XA_TARGET, node); } else if (pattern) { crm_xml_add(data, PCMK_XA_TARGET_PATTERN, pattern); } else { crm_xml_add(data, PCMK_XA_TARGET_ATTRIBUTE, attr); crm_xml_add(data, PCMK_XA_TARGET_VALUE, value); } for (; device_list; device_list = device_list->next) { pcmk__add_separated_word(&list, 1024, device_list->value, ","); } if (list != NULL) { crm_xml_add(data, PCMK_XA_DEVICES, (const char *) list->str); g_string_free(list, TRUE); } return data; } static int stonith_api_register_level_full(stonith_t *st, int options, const char *node, const char *pattern, const char *attr, const char *value, int level, const stonith_key_value_t *device_list) { int rc = 0; xmlNode *data = create_level_registration_xml(node, pattern, attr, value, level, device_list); CRM_CHECK(data != NULL, return -EINVAL); rc = stonith_send_command(st, STONITH_OP_LEVEL_ADD, data, NULL, options, 0); pcmk__xml_free(data); return rc; } static int stonith_api_register_level(stonith_t * st, int options, const char *node, int level, const stonith_key_value_t * device_list) { return stonith_api_register_level_full(st, options, node, NULL, NULL, NULL, level, device_list); } static int stonith_api_device_list(stonith_t *stonith, int call_options, const char *namespace_s, stonith_key_value_t **devices, int timeout) { int count = 0; enum stonith_namespace ns = parse_namespace(namespace_s); if (devices == NULL) { crm_err("Parameter error: stonith_api_device_list"); return -EFAULT; } #if HAVE_STONITH_STONITH_H // Include Linux-HA agents if requested if ((ns == st_namespace_any) || (ns == st_namespace_lha)) { count += stonith__list_lha_agents(devices); } #endif // Include Red Hat agents if requested if ((ns == st_namespace_any) || (ns == st_namespace_rhcs)) { count += stonith__list_rhcs_agents(devices); } return count; } // See stonith_api_operations_t:metadata() documentation static int stonith_api_device_metadata(stonith_t *stonith, int call_options, const char *agent, const char *namespace_s, char **output, int timeout_sec) { /* By executing meta-data directly, we can get it from stonith_admin when * the cluster is not running, which is important for higher-level tools. */ enum stonith_namespace ns = get_namespace_from_agent(agent); if (timeout_sec <= 0) { timeout_sec = PCMK_DEFAULT_ACTION_TIMEOUT_MS; } crm_trace("Looking up metadata for %s agent %s", namespace_text(ns), agent); switch (ns) { case st_namespace_rhcs: return stonith__rhcs_metadata(agent, timeout_sec, output); #if HAVE_STONITH_STONITH_H case st_namespace_lha: return stonith__lha_metadata(agent, timeout_sec, output); #endif default: crm_err("Can't get fence agent '%s' meta-data: No such agent", agent); break; } return -ENODEV; } static int stonith_api_query(stonith_t * stonith, int call_options, const char *target, stonith_key_value_t ** devices, int timeout) { int rc = 0, lpc = 0, max = 0; xmlNode *data = NULL; xmlNode *output = NULL; xmlXPathObject *xpathObj = NULL; CRM_CHECK(devices != NULL, return -EINVAL); data = pcmk__xe_create(NULL, PCMK__XE_ST_DEVICE_ID); crm_xml_add(data, PCMK__XA_ST_ORIGIN, __func__); crm_xml_add(data, PCMK__XA_ST_TARGET, target); crm_xml_add(data, PCMK__XA_ST_DEVICE_ACTION, PCMK_ACTION_OFF); rc = stonith_send_command(stonith, STONITH_OP_QUERY, data, &output, call_options, timeout); if (rc < 0) { return rc; } xpathObj = pcmk__xpath_search(output->doc, "//*[@" PCMK_XA_AGENT "]"); if (xpathObj) { max = pcmk__xpath_num_results(xpathObj); for (lpc = 0; lpc < max; lpc++) { xmlNode *match = pcmk__xpath_result(xpathObj, lpc); CRM_LOG_ASSERT(match != NULL); if(match != NULL) { const char *match_id = pcmk__xe_get(match, PCMK_XA_ID); xmlChar *match_path = xmlGetNodePath(match); crm_info("//*[@" PCMK_XA_AGENT "][%d] = %s", lpc, match_path); free(match_path); *devices = stonith__key_value_add(*devices, NULL, match_id); } } xmlXPathFreeObject(xpathObj); } pcmk__xml_free(output); pcmk__xml_free(data); return max; } /*! * \internal * \brief Make a STONITH_OP_EXEC request * * \param[in,out] stonith Fencer connection * \param[in] call_options Bitmask of \c stonith_call_options * \param[in] id Fence device ID that request is for * \param[in] action Agent action to request (list, status, monitor) * \param[in] target Name of target node for requested action * \param[in] timeout_sec Error if not completed within this many seconds * \param[out] output Where to set agent output */ static int stonith_api_call(stonith_t *stonith, int call_options, const char *id, const char *action, const char *target, int timeout_sec, xmlNode **output) { int rc = 0; xmlNode *data = NULL; data = pcmk__xe_create(NULL, PCMK__XE_ST_DEVICE_ID); crm_xml_add(data, PCMK__XA_ST_ORIGIN, __func__); crm_xml_add(data, PCMK__XA_ST_DEVICE_ID, id); crm_xml_add(data, PCMK__XA_ST_DEVICE_ACTION, action); crm_xml_add(data, PCMK__XA_ST_TARGET, target); rc = stonith_send_command(stonith, STONITH_OP_EXEC, data, output, call_options, timeout_sec); pcmk__xml_free(data); return rc; } static int stonith_api_list(stonith_t * stonith, int call_options, const char *id, char **list_info, int timeout) { int rc; xmlNode *output = NULL; rc = stonith_api_call(stonith, call_options, id, PCMK_ACTION_LIST, NULL, timeout, &output); if (output && list_info) { const char *list_str; list_str = pcmk__xe_get(output, PCMK__XA_ST_OUTPUT); if (list_str) { *list_info = strdup(list_str); } } if (output) { pcmk__xml_free(output); } return rc; } static int stonith_api_monitor(stonith_t * stonith, int call_options, const char *id, int timeout) { return stonith_api_call(stonith, call_options, id, PCMK_ACTION_MONITOR, NULL, timeout, NULL); } static int stonith_api_status(stonith_t * stonith, int call_options, const char *id, const char *port, int timeout) { return stonith_api_call(stonith, call_options, id, PCMK_ACTION_STATUS, port, timeout, NULL); } static int stonith_api_fence_with_delay(stonith_t * stonith, int call_options, const char *node, const char *action, int timeout, int tolerance, int delay) { int rc = 0; xmlNode *data = NULL; data = pcmk__xe_create(NULL, __func__); crm_xml_add(data, PCMK__XA_ST_TARGET, node); crm_xml_add(data, PCMK__XA_ST_DEVICE_ACTION, action); crm_xml_add_int(data, PCMK__XA_ST_TIMEOUT, timeout); crm_xml_add_int(data, PCMK__XA_ST_TOLERANCE, tolerance); crm_xml_add_int(data, PCMK__XA_ST_DELAY, delay); rc = stonith_send_command(stonith, STONITH_OP_FENCE, data, NULL, call_options, timeout); pcmk__xml_free(data); return rc; } static int stonith_api_fence(stonith_t * stonith, int call_options, const char *node, const char *action, int timeout, int tolerance) { return stonith_api_fence_with_delay(stonith, call_options, node, action, timeout, tolerance, 0); } static int stonith_api_confirm(stonith_t * stonith, int call_options, const char *target) { stonith__set_call_options(call_options, target, st_opt_manual_ack); return stonith_api_fence(stonith, call_options, target, PCMK_ACTION_OFF, 0, 0); } static int stonith_api_history(stonith_t * stonith, int call_options, const char *node, stonith_history_t ** history, int timeout) { int rc = 0; xmlNode *data = NULL; xmlNode *output = NULL; stonith_history_t *last = NULL; *history = NULL; if (node) { data = pcmk__xe_create(NULL, __func__); crm_xml_add(data, PCMK__XA_ST_TARGET, node); } stonith__set_call_options(call_options, node, st_opt_sync_call); rc = stonith_send_command(stonith, STONITH_OP_FENCE_HISTORY, data, &output, call_options, timeout); pcmk__xml_free(data); if (rc == 0) { xmlNode *op = NULL; xmlNode *reply = pcmk__xpath_find_one(output->doc, "//" PCMK__XE_ST_HISTORY, LOG_NEVER); for (op = pcmk__xe_first_child(reply, NULL, NULL, NULL); op != NULL; op = pcmk__xe_next(op, NULL)) { stonith_history_t *kvp = pcmk__assert_alloc(1, sizeof(stonith_history_t)); long long completed_nsec = 0LL; kvp->target = pcmk__xe_get_copy(op, PCMK__XA_ST_TARGET); kvp->action = pcmk__xe_get_copy(op, PCMK__XA_ST_DEVICE_ACTION); kvp->origin = pcmk__xe_get_copy(op, PCMK__XA_ST_ORIGIN); kvp->delegate = pcmk__xe_get_copy(op, PCMK__XA_ST_DELEGATE); kvp->client = pcmk__xe_get_copy(op, PCMK__XA_ST_CLIENTNAME); pcmk__xe_get_time(op, PCMK__XA_ST_DATE, &kvp->completed); pcmk__xe_get_ll(op, PCMK__XA_ST_DATE_NSEC, &completed_nsec); // Coverity complains here if long is the same size as long long // coverity[result_independent_of_operands:FALSE] if ((completed_nsec >= LONG_MIN) && (completed_nsec <= LONG_MAX)) { kvp->completed_nsec = (long) completed_nsec; } pcmk__xe_get_int(op, PCMK__XA_ST_STATE, &kvp->state); kvp->exit_reason = pcmk__xe_get_copy(op, PCMK_XA_EXIT_REASON); if (last) { last->next = kvp; } else { *history = kvp; } last = kvp; } } pcmk__xml_free(output); return rc; } /*! * \internal * \brief Free a list of fencing history objects and all members of each object * * \param[in,out] head Head of fencing history object list */ void stonith__history_free(stonith_history_t *head) { /* @COMPAT Drop "next" member of stonith_history_t, use a GList or GSList, * and use the appropriate free function (while ensuring the members get * freed) */ while (head != NULL) { stonith_history_t *next = head->next; free(head->target); free(head->action); free(head->origin); free(head->delegate); free(head->client); free(head->exit_reason); free(head); head = next; } } static gint stonithlib_GCompareFunc(gconstpointer a, gconstpointer b) { int rc = 0; const stonith_notify_client_t *a_client = a; const stonith_notify_client_t *b_client = b; if (a_client->delete || b_client->delete) { /* make entries marked for deletion not findable */ return -1; } CRM_CHECK(a_client->event != NULL && b_client->event != NULL, return 0); rc = strcmp(a_client->event, b_client->event); if (rc == 0) { if (a_client->notify == NULL || b_client->notify == NULL) { return 0; } else if (a_client->notify == b_client->notify) { return 0; } else if (((long)a_client->notify) < ((long)b_client->notify)) { crm_err("callbacks for %s are not equal: %p vs. %p", a_client->event, a_client->notify, b_client->notify); return -1; } crm_err("callbacks for %s are not equal: %p vs. %p", a_client->event, a_client->notify, b_client->notify); return 1; } return rc; } xmlNode * stonith_create_op(int call_id, const char *token, const char *op, xmlNode * data, int call_options) { xmlNode *op_msg = NULL; CRM_CHECK(token != NULL, return NULL); op_msg = pcmk__xe_create(NULL, PCMK__XE_STONITH_COMMAND); crm_xml_add(op_msg, PCMK__XA_T, PCMK__VALUE_STONITH_NG); crm_xml_add(op_msg, PCMK__XA_ST_OP, op); crm_xml_add_int(op_msg, PCMK__XA_ST_CALLID, call_id); crm_trace("Sending call options: %.8lx, %d", (long)call_options, call_options); crm_xml_add_int(op_msg, PCMK__XA_ST_CALLOPT, call_options); if (data != NULL) { xmlNode *wrapper = pcmk__xe_create(op_msg, PCMK__XE_ST_CALLDATA); pcmk__xml_copy(wrapper, data); } return op_msg; } static void stonith_destroy_op_callback(gpointer data) { stonith_callback_client_t *blob = data; if (blob->timer && blob->timer->ref > 0) { g_source_remove(blob->timer->ref); } free(blob->timer); free(blob); } static int stonith_api_signoff(stonith_t * stonith) { stonith_private_t *native = stonith->st_private; crm_debug("Disconnecting from the fencer"); if (native->source != NULL) { /* Attached to mainloop */ mainloop_del_ipc_client(native->source); native->source = NULL; native->ipc = NULL; } else if (native->ipc) { /* Not attached to mainloop */ crm_ipc_t *ipc = native->ipc; native->ipc = NULL; crm_ipc_close(ipc); crm_ipc_destroy(ipc); } free(native->token); native->token = NULL; stonith->state = stonith_disconnected; return pcmk_ok; } static int stonith_api_del_callback(stonith_t * stonith, int call_id, bool all_callbacks) { stonith_private_t *private = stonith->st_private; if (all_callbacks) { private->op_callback = NULL; g_hash_table_destroy(private->stonith_op_callback_table); private->stonith_op_callback_table = pcmk__intkey_table(stonith_destroy_op_callback); } else if (call_id == 0) { private->op_callback = NULL; } else { pcmk__intkey_table_remove(private->stonith_op_callback_table, call_id); } return pcmk_ok; } /*! * \internal * \brief Invoke a (single) specified fence action callback * * \param[in,out] st Fencer API connection * \param[in] call_id If positive, call ID of completed fence action, * otherwise legacy return code for early failure * \param[in,out] result Full result for action * \param[in,out] userdata User data to pass to callback * \param[in] callback Fence action callback to invoke */ static void invoke_fence_action_callback(stonith_t *st, int call_id, pcmk__action_result_t *result, void *userdata, void (*callback) (stonith_t *st, stonith_callback_data_t *data)) { stonith_callback_data_t data = { 0, }; data.call_id = call_id; data.rc = pcmk_rc2legacy(stonith__result2rc(result)); data.userdata = userdata; data.opaque = (void *) result; callback(st, &data); } /*! * \internal * \brief Invoke any callbacks registered for a specified fence action result * * Given a fence action result from the fencer, invoke any callback registered * for that action, as well as any global callback registered. * * \param[in,out] stonith Fencer API connection * \param[in] msg If non-NULL, fencer reply * \param[in] call_id If \p msg is NULL, call ID of action that timed out */ static void invoke_registered_callbacks(stonith_t *stonith, const xmlNode *msg, int call_id) { stonith_private_t *private = NULL; stonith_callback_client_t *cb_info = NULL; pcmk__action_result_t result = PCMK__UNKNOWN_RESULT; CRM_CHECK(stonith != NULL, return); CRM_CHECK(stonith->st_private != NULL, return); private = stonith->st_private; if (msg == NULL) { // Fencer didn't reply in time pcmk__set_result(&result, CRM_EX_ERROR, PCMK_EXEC_TIMEOUT, "Fencer accepted request but did not reply in time"); CRM_LOG_ASSERT(call_id > 0); } else { // We have the fencer reply if ((pcmk__xe_get_int(msg, PCMK__XA_ST_CALLID, &call_id) != pcmk_rc_ok) || (call_id <= 0)) { crm_log_xml_warn(msg, "Bad fencer reply"); } stonith__xe_get_result(msg, &result); } if (call_id > 0) { cb_info = pcmk__intkey_table_lookup(private->stonith_op_callback_table, call_id); } if ((cb_info != NULL) && (cb_info->callback != NULL) && (pcmk__result_ok(&result) || !(cb_info->only_success))) { crm_trace("Invoking callback %s for call %d", pcmk__s(cb_info->id, "without ID"), call_id); invoke_fence_action_callback(stonith, call_id, &result, cb_info->user_data, cb_info->callback); } else if ((private->op_callback == NULL) && !pcmk__result_ok(&result)) { crm_warn("Fencing action without registered callback failed: %d (%s%s%s)", result.exit_status, pcmk_exec_status_str(result.execution_status), ((result.exit_reason == NULL)? "" : ": "), ((result.exit_reason == NULL)? "" : result.exit_reason)); crm_log_xml_debug(msg, "Failed fence update"); } if (private->op_callback != NULL) { crm_trace("Invoking global callback for call %d", call_id); invoke_fence_action_callback(stonith, call_id, &result, NULL, private->op_callback); } if (cb_info != NULL) { stonith_api_del_callback(stonith, call_id, FALSE); } pcmk__reset_result(&result); } static gboolean stonith_async_timeout_handler(gpointer data) { struct timer_rec_s *timer = data; crm_err("Async call %d timed out after %dms", timer->call_id, timer->timeout); invoke_registered_callbacks(timer->stonith, NULL, timer->call_id); /* Always return TRUE, never remove the handler * We do that in stonith_del_callback() */ return TRUE; } static void set_callback_timeout(stonith_callback_client_t * callback, stonith_t * stonith, int call_id, int timeout) { struct timer_rec_s *async_timer = callback->timer; if (timeout <= 0) { return; } if (!async_timer) { async_timer = pcmk__assert_alloc(1, sizeof(struct timer_rec_s)); callback->timer = async_timer; } async_timer->stonith = stonith; async_timer->call_id = call_id; /* Allow a fair bit of grace to allow the server to tell us of a timeout * This is only a fallback */ async_timer->timeout = (timeout + 60) * 1000; if (async_timer->ref) { g_source_remove(async_timer->ref); } async_timer->ref = pcmk__create_timer(async_timer->timeout, stonith_async_timeout_handler, async_timer); } static void update_callback_timeout(int call_id, int timeout, stonith_t * st) { stonith_callback_client_t *callback = NULL; stonith_private_t *private = st->st_private; callback = pcmk__intkey_table_lookup(private->stonith_op_callback_table, call_id); if (!callback || !callback->allow_timeout_updates) { return; } set_callback_timeout(callback, st, call_id, timeout); } static int stonith_dispatch_internal(const char *buffer, ssize_t length, gpointer userdata) { const char *type = NULL; struct notify_blob_s blob; stonith_t *st = userdata; stonith_private_t *private = NULL; pcmk__assert(st != NULL); private = st->st_private; blob.stonith = st; blob.xml = pcmk__xml_parse(buffer); if (blob.xml == NULL) { crm_warn("Received malformed message from fencer: %s", buffer); return 0; } /* do callbacks */ type = pcmk__xe_get(blob.xml, PCMK__XA_T); crm_trace("Activating %s callbacks...", type); if (pcmk__str_eq(type, PCMK__VALUE_STONITH_NG, pcmk__str_none)) { invoke_registered_callbacks(st, blob.xml, 0); } else if (pcmk__str_eq(type, PCMK__VALUE_ST_NOTIFY, pcmk__str_none)) { foreach_notify_entry(private, stonith_send_notification, &blob); } else if (pcmk__str_eq(type, PCMK__VALUE_ST_ASYNC_TIMEOUT_VALUE, pcmk__str_none)) { int call_id = 0; int timeout = 0; pcmk__xe_get_int(blob.xml, PCMK__XA_ST_TIMEOUT, &timeout); pcmk__xe_get_int(blob.xml, PCMK__XA_ST_CALLID, &call_id); update_callback_timeout(call_id, timeout, st); } else { crm_err("Unknown message type: %s", type); crm_log_xml_warn(blob.xml, "BadReply"); } pcmk__xml_free(blob.xml); return 1; } static int stonith_api_signon(stonith_t * stonith, const char *name, int *stonith_fd) { int rc = pcmk_ok; stonith_private_t *native = NULL; const char *display_name = name? name : "client"; struct ipc_client_callbacks st_callbacks = { .dispatch = stonith_dispatch_internal, .destroy = stonith_connection_destroy }; CRM_CHECK(stonith != NULL, return -EINVAL); native = stonith->st_private; pcmk__assert(native != NULL); crm_debug("Attempting fencer connection by %s with%s mainloop", display_name, (stonith_fd? "out" : "")); stonith->state = stonith_connected_command; if (stonith_fd) { /* No mainloop */ native->ipc = crm_ipc_new("stonith-ng", 0); if (native->ipc != NULL) { rc = pcmk__connect_generic_ipc(native->ipc); if (rc == pcmk_rc_ok) { rc = pcmk__ipc_fd(native->ipc, stonith_fd); if (rc != pcmk_rc_ok) { crm_debug("Couldn't get file descriptor for IPC: %s", pcmk_rc_str(rc)); } } if (rc != pcmk_rc_ok) { crm_ipc_close(native->ipc); crm_ipc_destroy(native->ipc); native->ipc = NULL; } } } else { /* With mainloop */ native->source = mainloop_add_ipc_client("stonith-ng", G_PRIORITY_MEDIUM, 0, stonith, &st_callbacks); native->ipc = mainloop_get_ipc_client(native->source); } if (native->ipc == NULL) { rc = -ENOTCONN; } else { xmlNode *reply = NULL; xmlNode *hello = pcmk__xe_create(NULL, PCMK__XE_STONITH_COMMAND); crm_xml_add(hello, PCMK__XA_T, PCMK__VALUE_STONITH_NG); crm_xml_add(hello, PCMK__XA_ST_OP, CRM_OP_REGISTER); crm_xml_add(hello, PCMK__XA_ST_CLIENTNAME, name); rc = crm_ipc_send(native->ipc, hello, crm_ipc_client_response, -1, &reply); if (rc < 0) { crm_debug("Couldn't register with the fencer: %s " QB_XS " rc=%d", pcmk_strerror(rc), rc); rc = -ECOMM; } else if (reply == NULL) { crm_debug("Couldn't register with the fencer: no reply"); rc = -EPROTO; } else { const char *msg_type = pcmk__xe_get(reply, PCMK__XA_ST_OP); native->token = pcmk__xe_get_copy(reply, PCMK__XA_ST_CLIENTID); if (!pcmk__str_eq(msg_type, CRM_OP_REGISTER, pcmk__str_none)) { crm_debug("Couldn't register with the fencer: invalid reply type '%s'", (msg_type? msg_type : "(missing)")); crm_log_xml_debug(reply, "Invalid fencer reply"); rc = -EPROTO; } else if (native->token == NULL) { crm_debug("Couldn't register with the fencer: no token in reply"); crm_log_xml_debug(reply, "Invalid fencer reply"); rc = -EPROTO; } else { crm_debug("Connection to fencer by %s succeeded (registration token: %s)", display_name, native->token); rc = pcmk_ok; } } pcmk__xml_free(reply); pcmk__xml_free(hello); } if (rc != pcmk_ok) { crm_debug("Connection attempt to fencer by %s failed: %s " QB_XS " rc=%d", display_name, pcmk_strerror(rc), rc); stonith->cmds->disconnect(stonith); } return rc; } static int stonith_set_notification(stonith_t * stonith, const char *callback, int enabled) { int rc = pcmk_ok; xmlNode *notify_msg = pcmk__xe_create(NULL, __func__); stonith_private_t *native = stonith->st_private; if (stonith->state != stonith_disconnected) { crm_xml_add(notify_msg, PCMK__XA_ST_OP, STONITH_OP_NOTIFY); if (enabled) { crm_xml_add(notify_msg, PCMK__XA_ST_NOTIFY_ACTIVATE, callback); } else { crm_xml_add(notify_msg, PCMK__XA_ST_NOTIFY_DEACTIVATE, callback); } rc = crm_ipc_send(native->ipc, notify_msg, crm_ipc_client_response, -1, NULL); if (rc < 0) { - crm_perror(LOG_DEBUG, "Couldn't register for fencing notifications: %d", rc); + crm_debug("Couldn't register for fencing notifications: %s", + pcmk_strerror(rc)); rc = -ECOMM; } else { rc = pcmk_ok; } } pcmk__xml_free(notify_msg); return rc; } static int stonith_api_add_notification(stonith_t * stonith, const char *event, void (*callback) (stonith_t * stonith, stonith_event_t * e)) { GList *list_item = NULL; stonith_notify_client_t *new_client = NULL; stonith_private_t *private = NULL; private = stonith->st_private; crm_trace("Adding callback for %s events (%d)", event, g_list_length(private->notify_list)); new_client = pcmk__assert_alloc(1, sizeof(stonith_notify_client_t)); new_client->event = event; new_client->notify = callback; list_item = g_list_find_custom(private->notify_list, new_client, stonithlib_GCompareFunc); if (list_item != NULL) { crm_warn("Callback already present"); free(new_client); return -ENOTUNIQ; } else { private->notify_list = g_list_append(private->notify_list, new_client); stonith_set_notification(stonith, event, 1); crm_trace("Callback added (%d)", g_list_length(private->notify_list)); } return pcmk_ok; } static void del_notify_entry(gpointer data, gpointer user_data) { stonith_notify_client_t *entry = data; stonith_t * stonith = user_data; if (!entry->delete) { crm_debug("Removing callback for %s events", entry->event); stonith_api_del_notification(stonith, entry->event); } } static int stonith_api_del_notification(stonith_t * stonith, const char *event) { GList *list_item = NULL; stonith_notify_client_t *new_client = NULL; stonith_private_t *private = stonith->st_private; if (event == NULL) { foreach_notify_entry(private, del_notify_entry, stonith); crm_trace("Removed callback"); return pcmk_ok; } crm_debug("Removing callback for %s events", event); new_client = pcmk__assert_alloc(1, sizeof(stonith_notify_client_t)); new_client->event = event; new_client->notify = NULL; list_item = g_list_find_custom(private->notify_list, new_client, stonithlib_GCompareFunc); stonith_set_notification(stonith, event, 0); if (list_item != NULL) { stonith_notify_client_t *list_client = list_item->data; if (private->notify_refcnt) { list_client->delete = TRUE; private->notify_deletes = TRUE; } else { private->notify_list = g_list_remove(private->notify_list, list_client); free(list_client); } crm_trace("Removed callback"); } else { crm_trace("Callback not present"); } free(new_client); return pcmk_ok; } static int stonith_api_add_callback(stonith_t * stonith, int call_id, int timeout, int options, void *user_data, const char *callback_name, void (*callback) (stonith_t * st, stonith_callback_data_t * data)) { stonith_callback_client_t *blob = NULL; stonith_private_t *private = NULL; CRM_CHECK(stonith != NULL, return -EINVAL); CRM_CHECK(stonith->st_private != NULL, return -EINVAL); private = stonith->st_private; if (call_id == 0) { // Add global callback private->op_callback = callback; } else if (call_id < 0) { // Call failed immediately, so call callback now if (!(options & st_opt_report_only_success)) { pcmk__action_result_t result = PCMK__UNKNOWN_RESULT; crm_trace("Call failed, calling %s: %s", callback_name, pcmk_strerror(call_id)); pcmk__set_result(&result, CRM_EX_ERROR, stonith__legacy2status(call_id), NULL); invoke_fence_action_callback(stonith, call_id, &result, user_data, callback); } else { crm_warn("Fencer call failed: %s", pcmk_strerror(call_id)); } return FALSE; } blob = pcmk__assert_alloc(1, sizeof(stonith_callback_client_t)); blob->id = callback_name; blob->only_success = (options & st_opt_report_only_success) ? TRUE : FALSE; blob->user_data = user_data; blob->callback = callback; blob->allow_timeout_updates = (options & st_opt_timeout_updates) ? TRUE : FALSE; if (timeout > 0) { set_callback_timeout(blob, stonith, call_id, timeout); } pcmk__intkey_table_insert(private->stonith_op_callback_table, call_id, blob); crm_trace("Added callback to %s for call %d", callback_name, call_id); return TRUE; } /*! * \internal * \brief Get the data section of a fencer notification * * \param[in] msg Notification XML * \param[in] ntype Notification type */ static xmlNode * get_event_data_xml(xmlNode *msg, const char *ntype) { char *data_addr = crm_strdup_printf("//%s", ntype); xmlNode *data = pcmk__xpath_find_one(msg->doc, data_addr, LOG_DEBUG); free(data_addr); return data; } /* */ static stonith_event_t * xml_to_event(xmlNode *msg) { stonith_event_t *event = pcmk__assert_alloc(1, sizeof(stonith_event_t)); struct event_private *event_private = NULL; event->opaque = pcmk__assert_alloc(1, sizeof(struct event_private)); event_private = (struct event_private *) event->opaque; crm_log_xml_trace(msg, "stonith_notify"); // All notification types have the operation result and notification subtype stonith__xe_get_result(msg, &event_private->result); event->operation = pcmk__xe_get_copy(msg, PCMK__XA_ST_OP); // @COMPAT The API originally provided the result as a legacy return code event->result = pcmk_rc2legacy(stonith__result2rc(&event_private->result)); // Some notification subtypes have additional information if (pcmk__str_eq(event->operation, PCMK__VALUE_ST_NOTIFY_FENCE, pcmk__str_none)) { xmlNode *data = get_event_data_xml(msg, event->operation); if (data == NULL) { crm_err("No data for %s event", event->operation); crm_log_xml_notice(msg, "BadEvent"); } else { event->origin = pcmk__xe_get_copy(data, PCMK__XA_ST_ORIGIN); event->action = pcmk__xe_get_copy(data, PCMK__XA_ST_DEVICE_ACTION); event->target = pcmk__xe_get_copy(data, PCMK__XA_ST_TARGET); event->executioner = pcmk__xe_get_copy(data, PCMK__XA_ST_DELEGATE); event->id = pcmk__xe_get_copy(data, PCMK__XA_ST_REMOTE_OP); event->client_origin = pcmk__xe_get_copy(data, PCMK__XA_ST_CLIENTNAME); event->device = pcmk__xe_get_copy(data, PCMK__XA_ST_DEVICE_ID); } } else if (pcmk__str_any_of(event->operation, STONITH_OP_DEVICE_ADD, STONITH_OP_DEVICE_DEL, STONITH_OP_LEVEL_ADD, STONITH_OP_LEVEL_DEL, NULL)) { xmlNode *data = get_event_data_xml(msg, event->operation); if (data == NULL) { crm_err("No data for %s event", event->operation); crm_log_xml_notice(msg, "BadEvent"); } else { event->device = pcmk__xe_get_copy(data, PCMK__XA_ST_DEVICE_ID); } } return event; } static void event_free(stonith_event_t * event) { struct event_private *event_private = event->opaque; free(event->id); free(event->operation); free(event->origin); free(event->action); free(event->target); free(event->executioner); free(event->device); free(event->client_origin); pcmk__reset_result(&event_private->result); free(event->opaque); free(event); } static void stonith_send_notification(gpointer data, gpointer user_data) { struct notify_blob_s *blob = user_data; stonith_notify_client_t *entry = data; stonith_event_t *st_event = NULL; const char *event = NULL; if (blob->xml == NULL) { crm_warn("Skipping callback - NULL message"); return; } event = pcmk__xe_get(blob->xml, PCMK__XA_SUBT); if (entry == NULL) { crm_warn("Skipping callback - NULL callback client"); return; } else if (entry->delete) { crm_trace("Skipping callback - marked for deletion"); return; } else if (entry->notify == NULL) { crm_warn("Skipping callback - NULL callback"); return; } else if (!pcmk__str_eq(entry->event, event, pcmk__str_none)) { crm_trace("Skipping callback - event mismatch %p/%s vs. %s", entry, entry->event, event); return; } st_event = xml_to_event(blob->xml); crm_trace("Invoking callback for %p/%s event...", entry, event); // coverity[null_field] entry->notify(blob->stonith, st_event); crm_trace("Callback invoked..."); event_free(st_event); } /*! * \internal * \brief Create and send an API request * * \param[in,out] stonith Stonith connection * \param[in] op API operation to request * \param[in] data Data to attach to request * \param[out] output_data If not NULL, will be set to reply if synchronous * \param[in] call_options Bitmask of stonith_call_options to use * \param[in] timeout Error if not completed within this many seconds * * \return pcmk_ok (for synchronous requests) or positive call ID * (for asynchronous requests) on success, -errno otherwise */ static int stonith_send_command(stonith_t * stonith, const char *op, xmlNode * data, xmlNode ** output_data, int call_options, int timeout) { int rc = 0; int reply_id = -1; xmlNode *op_msg = NULL; xmlNode *op_reply = NULL; stonith_private_t *native = NULL; pcmk__assert((stonith != NULL) && (stonith->st_private != NULL) && (op != NULL)); native = stonith->st_private; if (output_data != NULL) { *output_data = NULL; } if ((stonith->state == stonith_disconnected) || (native->token == NULL)) { return -ENOTCONN; } /* Increment the call ID, which must be positive to avoid conflicting with * error codes. This shouldn't be a problem unless the client mucked with * it or the counter wrapped around. */ stonith->call_id++; if (stonith->call_id < 1) { stonith->call_id = 1; } op_msg = stonith_create_op(stonith->call_id, native->token, op, data, call_options); if (op_msg == NULL) { return -EINVAL; } crm_xml_add_int(op_msg, PCMK__XA_ST_TIMEOUT, timeout); crm_trace("Sending %s message to fencer with timeout %ds", op, timeout); if (data) { const char *delay_s = pcmk__xe_get(data, PCMK__XA_ST_DELAY); if (delay_s) { crm_xml_add(op_msg, PCMK__XA_ST_DELAY, delay_s); } } { enum crm_ipc_flags ipc_flags = crm_ipc_flags_none; if (call_options & st_opt_sync_call) { pcmk__set_ipc_flags(ipc_flags, "stonith command", crm_ipc_client_response); } rc = crm_ipc_send(native->ipc, op_msg, ipc_flags, 1000 * (timeout + 60), &op_reply); } pcmk__xml_free(op_msg); if (rc < 0) { - crm_perror(LOG_ERR, "Couldn't perform %s operation (timeout=%ds): %d", op, timeout, rc); + crm_err("Couldn't perform %s operation (timeout=%ds): %s", op, timeout, + pcmk_strerror(rc)); rc = -ECOMM; goto done; } crm_log_xml_trace(op_reply, "Reply"); if (!(call_options & st_opt_sync_call)) { crm_trace("Async call %d, returning", stonith->call_id); pcmk__xml_free(op_reply); return stonith->call_id; } pcmk__xe_get_int(op_reply, PCMK__XA_ST_CALLID, &reply_id); if (reply_id == stonith->call_id) { pcmk__action_result_t result = PCMK__UNKNOWN_RESULT; crm_trace("Synchronous reply %d received", reply_id); stonith__xe_get_result(op_reply, &result); rc = pcmk_rc2legacy(stonith__result2rc(&result)); pcmk__reset_result(&result); if ((call_options & st_opt_discard_reply) || output_data == NULL) { crm_trace("Discarding reply"); } else { *output_data = op_reply; op_reply = NULL; /* Prevent subsequent free */ } } else if (reply_id <= 0) { crm_err("Received bad reply: No id set"); crm_log_xml_err(op_reply, "Bad reply"); pcmk__xml_free(op_reply); op_reply = NULL; rc = -ENOMSG; } else { crm_err("Received bad reply: %d (wanted %d)", reply_id, stonith->call_id); crm_log_xml_err(op_reply, "Old reply"); pcmk__xml_free(op_reply); op_reply = NULL; rc = -ENOMSG; } done: if (!crm_ipc_connected(native->ipc)) { crm_err("Fencer disconnected"); free(native->token); native->token = NULL; stonith->state = stonith_disconnected; } pcmk__xml_free(op_reply); return rc; } /*! * \internal * \brief Process IPC messages for a fencer API connection * * This is used for testing purposes in scenarios that don't use a mainloop to * dispatch messages automatically. * * \param[in,out] stonith_api Fencer API connetion object * * \return Standard Pacemaker return code */ int stonith__api_dispatch(stonith_t *stonith_api) { stonith_private_t *private = NULL; pcmk__assert(stonith_api != NULL); private = stonith_api->st_private; while (crm_ipc_ready(private->ipc)) { if (crm_ipc_read(private->ipc) > 0) { const char *msg = crm_ipc_buffer(private->ipc); stonith_dispatch_internal(msg, strlen(msg), stonith_api); pcmk__ipc_free_client_buffer(private->ipc); } if (!crm_ipc_connected(private->ipc)) { crm_err("Connection closed"); return ENOTCONN; } } return pcmk_rc_ok; } static int free_stonith_api(stonith_t *stonith) { int rc = pcmk_ok; crm_trace("Destroying %p", stonith); if (stonith->state != stonith_disconnected) { crm_trace("Unregistering notifications and disconnecting %p first", stonith); stonith->cmds->remove_notification(stonith, NULL); rc = stonith->cmds->disconnect(stonith); } if (stonith->state == stonith_disconnected) { stonith_private_t *private = stonith->st_private; crm_trace("Removing %d callbacks", g_hash_table_size(private->stonith_op_callback_table)); g_hash_table_destroy(private->stonith_op_callback_table); crm_trace("Destroying %d notification clients", g_list_length(private->notify_list)); g_list_free_full(private->notify_list, free); free(stonith->st_private); free(stonith->cmds); free(stonith); } else { crm_err("Not free'ing active connection: %s (%d)", pcmk_strerror(rc), rc); } return rc; } static gboolean is_stonith_param(gpointer key, gpointer value, gpointer user_data) { return pcmk_stonith_param(key); } int stonith__validate(stonith_t *st, int call_options, const char *rsc_id, const char *agent, GHashTable *params, int timeout_sec, char **output, char **error_output) { int rc = pcmk_rc_ok; /* Use a dummy node name in case the agent requires a target. We assume the * actual target doesn't matter for validation purposes (if in practice, * that is incorrect, we will need to allow the caller to pass the target). */ const char *target = "node1"; char *host_arg = NULL; if (params != NULL) { host_arg = pcmk__str_copy(g_hash_table_lookup(params, PCMK_STONITH_HOST_ARGUMENT)); /* Remove special stonith params from the table before doing anything else */ g_hash_table_foreach_remove(params, is_stonith_param, NULL); } #if PCMK__ENABLE_CIBSECRETS rc = pcmk__substitute_secrets(rsc_id, params); if (rc != pcmk_rc_ok) { crm_warn("Could not replace secret parameters for validation of %s: %s", agent, pcmk_rc_str(rc)); // rc is standard return value, don't return it in this function } #endif if (output) { *output = NULL; } if (error_output) { *error_output = NULL; } if (timeout_sec <= 0) { timeout_sec = PCMK_DEFAULT_ACTION_TIMEOUT_MS; } switch (get_namespace_from_agent(agent)) { case st_namespace_rhcs: rc = stonith__rhcs_validate(st, call_options, target, agent, params, host_arg, timeout_sec, output, error_output); rc = pcmk_legacy2rc(rc); break; #if HAVE_STONITH_STONITH_H case st_namespace_lha: rc = stonith__lha_validate(st, call_options, target, agent, params, timeout_sec, output, error_output); rc = pcmk_legacy2rc(rc); break; #endif case st_namespace_invalid: errno = ENOENT; rc = errno; if (error_output) { *error_output = crm_strdup_printf("Agent %s not found", agent); } else { crm_err("Agent %s not found", agent); } break; default: errno = EOPNOTSUPP; rc = errno; if (error_output) { *error_output = crm_strdup_printf("Agent %s does not support validation", agent); } else { crm_err("Agent %s does not support validation", agent); } break; } free(host_arg); return rc; } static int stonith_api_validate(stonith_t *st, int call_options, const char *rsc_id, const char *namespace_s, const char *agent, const stonith_key_value_t *params, int timeout_sec, char **output, char **error_output) { /* Validation should be done directly via the agent, so we can get it from * stonith_admin when the cluster is not running, which is important for * higher-level tools. */ int rc = pcmk_ok; GHashTable *params_table = pcmk__strkey_table(free, free); // Convert parameter list to a hash table for (; params; params = params->next) { if (!pcmk_stonith_param(params->key)) { pcmk__insert_dup(params_table, params->key, params->value); } } rc = stonith__validate(st, call_options, rsc_id, agent, params_table, timeout_sec, output, error_output); g_hash_table_destroy(params_table); return rc; } /*! * \internal * \brief Create a new fencer API connection object * * \return Newly allocated fencer API connection object, or \c NULL on * allocation failure */ stonith_t * stonith__api_new(void) { stonith_t *new_stonith = NULL; stonith_private_t *private = NULL; new_stonith = calloc(1, sizeof(stonith_t)); if (new_stonith == NULL) { return NULL; } private = calloc(1, sizeof(stonith_private_t)); if (private == NULL) { free(new_stonith); return NULL; } new_stonith->st_private = private; private->stonith_op_callback_table = pcmk__intkey_table(stonith_destroy_op_callback); private->notify_list = NULL; private->notify_refcnt = 0; private->notify_deletes = FALSE; new_stonith->call_id = 1; new_stonith->state = stonith_disconnected; new_stonith->cmds = calloc(1, sizeof(stonith_api_operations_t)); if (new_stonith->cmds == NULL) { free(new_stonith->st_private); free(new_stonith); return NULL; } new_stonith->cmds->free = free_stonith_api; new_stonith->cmds->connect = stonith_api_signon; new_stonith->cmds->disconnect = stonith_api_signoff; new_stonith->cmds->list = stonith_api_list; new_stonith->cmds->monitor = stonith_api_monitor; new_stonith->cmds->status = stonith_api_status; new_stonith->cmds->fence = stonith_api_fence; new_stonith->cmds->fence_with_delay = stonith_api_fence_with_delay; new_stonith->cmds->confirm = stonith_api_confirm; new_stonith->cmds->history = stonith_api_history; new_stonith->cmds->list_agents = stonith_api_device_list; new_stonith->cmds->metadata = stonith_api_device_metadata; new_stonith->cmds->query = stonith_api_query; new_stonith->cmds->remove_device = stonith_api_remove_device; new_stonith->cmds->register_device = stonith_api_register_device; new_stonith->cmds->remove_level = stonith_api_remove_level; new_stonith->cmds->remove_level_full = stonith_api_remove_level_full; new_stonith->cmds->register_level = stonith_api_register_level; new_stonith->cmds->register_level_full = stonith_api_register_level_full; new_stonith->cmds->remove_callback = stonith_api_del_callback; new_stonith->cmds->register_callback = stonith_api_add_callback; new_stonith->cmds->remove_notification = stonith_api_del_notification; new_stonith->cmds->register_notification = stonith_api_add_notification; new_stonith->cmds->validate = stonith_api_validate; return new_stonith; } /*! * \internal * \brief Free a fencer API connection object * * \param[in,out] stonith_api Fencer API connection object */ void stonith__api_free(stonith_t *stonith_api) { crm_trace("Destroying %p", stonith_api); if (stonith_api != NULL) { stonith_api->cmds->free(stonith_api); } } /*! * \internal * \brief Connect to the fencer, retrying on failure * * \param[in,out] stonith Fencer API connection object * \param[in] name Client name to use with fencer * \param[in] max_attempts Maximum number of attempts * * \return \c pcmk_rc_ok on success, or result of last attempt otherwise */ int stonith__api_connect_retry(stonith_t *stonith_api, const char *name, int max_attempts) { int rc = EINVAL; // if max_attempts is not positive for (int attempt = 1; attempt <= max_attempts; attempt++) { rc = stonith_api->cmds->connect(stonith_api, name, NULL); rc = pcmk_legacy2rc(rc); if (rc == pcmk_rc_ok) { return rc; } if (attempt < max_attempts) { crm_notice("Fencer connection attempt %d of %d failed " "(retrying in 2s): %s " QB_XS " rc=%d", attempt, max_attempts, pcmk_rc_str(rc), rc); sleep(2); } } crm_notice("Could not connect to fencer: %s " QB_XS " rc=%d", pcmk_rc_str(rc), rc); return rc; } /*! * \internal * \brief Append a newly allocated STONITH key-value pair to a list * * \param[in,out] head Head of key-value pair list (\c NULL for new list) * \param[in] key Key to add * \param[in] value Value to add * * \return Head of appended-to list (equal to \p head if \p head is not \c NULL) * \note The caller is responsible for freeing the return value using * \c stonith__key_value_freeall(). */ stonith_key_value_t * stonith__key_value_add(stonith_key_value_t *head, const char *key, const char *value) { /* @COMPAT Replace this function with pcmk_prepend_nvpair(), and reverse the * list when finished adding to it; or with a hash table where order does * not matter */ stonith_key_value_t *pair = pcmk__assert_alloc(1, sizeof(stonith_key_value_t)); pair->key = pcmk__str_copy(key); pair->value = pcmk__str_copy(value); if (head != NULL) { stonith_key_value_t *end = head; for (; end->next != NULL; end = end->next); end->next = pair; } else { head = pair; } return head; } /*! * \internal * \brief Free all items in a \c stonith_key_value_t list * * This means freeing the list itself with all of its nodes. Keys and values may * be freed depending on arguments. * * \param[in,out] head Head of list * \param[in] keys If \c true, free all keys * \param[in] values If \c true, free all values */ void stonith__key_value_freeall(stonith_key_value_t *head, bool keys, bool values) { while (head != NULL) { stonith_key_value_t *next = head->next; if (keys) { free(head->key); } if (values) { free(head->value); } free(head); head = next; } } #define api_log_open() openlog("stonith-api", LOG_CONS | LOG_NDELAY | LOG_PID, LOG_DAEMON) #define api_log(level, fmt, args...) syslog(level, "%s: "fmt, __func__, args) int stonith_api_kick(uint32_t nodeid, const char *uname, int timeout, bool off) { int rc = pcmk_ok; stonith_t *st = stonith__api_new(); const char *action = off? PCMK_ACTION_OFF : PCMK_ACTION_REBOOT; api_log_open(); if (st == NULL) { api_log(LOG_ERR, "API initialization failed, could not kick (%s) node %u/%s", action, nodeid, uname); return -EPROTO; } rc = st->cmds->connect(st, "stonith-api", NULL); if (rc != pcmk_ok) { api_log(LOG_ERR, "Connection failed, could not kick (%s) node %u/%s : %s (%d)", action, nodeid, uname, pcmk_strerror(rc), rc); } else { char *name = (uname == NULL)? pcmk__itoa(nodeid) : strdup(uname); int opts = 0; stonith__set_call_options(opts, name, st_opt_sync_call|st_opt_allow_self_fencing); if ((uname == NULL) && (nodeid > 0)) { stonith__set_call_options(opts, name, st_opt_cs_nodeid); } rc = st->cmds->fence(st, opts, name, action, timeout, 0); free(name); if (rc != pcmk_ok) { api_log(LOG_ERR, "Could not kick (%s) node %u/%s : %s (%d)", action, nodeid, uname, pcmk_strerror(rc), rc); } else { api_log(LOG_NOTICE, "Node %u/%s kicked: %s", nodeid, uname, action); } } stonith__api_free(st); return rc; } time_t stonith_api_time(uint32_t nodeid, const char *uname, bool in_progress) { int rc = pcmk_ok; time_t when = 0; stonith_t *st = stonith__api_new(); stonith_history_t *history = NULL, *hp = NULL; if (st == NULL) { api_log(LOG_ERR, "Could not retrieve fence history for %u/%s: " "API initialization failed", nodeid, uname); return when; } rc = st->cmds->connect(st, "stonith-api", NULL); if (rc != pcmk_ok) { api_log(LOG_NOTICE, "Connection failed: %s (%d)", pcmk_strerror(rc), rc); } else { int entries = 0; int progress = 0; int completed = 0; int opts = 0; char *name = (uname == NULL)? pcmk__itoa(nodeid) : strdup(uname); stonith__set_call_options(opts, name, st_opt_sync_call); if ((uname == NULL) && (nodeid > 0)) { stonith__set_call_options(opts, name, st_opt_cs_nodeid); } rc = st->cmds->history(st, opts, name, &history, 120); free(name); for (hp = history; hp; hp = hp->next) { entries++; if (in_progress) { progress++; if (hp->state != st_done && hp->state != st_failed) { when = time(NULL); } } else if (hp->state == st_done) { completed++; if (hp->completed > when) { when = hp->completed; } } } stonith__history_free(history); if(rc == pcmk_ok) { api_log(LOG_INFO, "Found %d entries for %u/%s: %d in progress, %d completed", entries, nodeid, uname, progress, completed); } else { api_log(LOG_ERR, "Could not retrieve fence history for %u/%s: %s (%d)", nodeid, uname, pcmk_strerror(rc), rc); } } stonith__api_free(st); if(when) { api_log(LOG_INFO, "Node %u/%s last kicked at: %ld", nodeid, uname, (long int)when); } return when; } /*! * \internal * \brief Check whether a fence agent with a given name exists * * \param[in] name Agent name * * \retval \c true If a fence agent named \p name exists * \retval \c false Otherwise */ bool stonith__agent_exists(const char *name) { stonith_t *stonith_api = NULL; stonith_key_value_t *agents = NULL; bool rc = false; if (name == NULL) { return false; } stonith_api = stonith__api_new(); if (stonith_api == NULL) { crm_err("Could not list fence agents: API memory allocation failed"); return false; } // The list_agents method ignores its timeout argument stonith_api->cmds->list_agents(stonith_api, st_opt_sync_call, NULL, &agents, 0); for (const stonith_key_value_t *iter = agents; iter != NULL; iter = iter->next) { if (pcmk__str_eq(iter->value, name, pcmk__str_none)) { rc = true; break; } } stonith__key_value_freeall(agents, true, true); stonith__api_free(stonith_api); return rc; } /*! * \internal * \brief Parse a target name from one line of a target list string * * \param[in] line One line of a target list string * \param[in] len String length of line * \param[in,out] output List to add newly allocated target name to */ static void parse_list_line(const char *line, int len, GList **output) { size_t i = 0; size_t entry_start = 0; if (line == NULL) { return; } /* Skip complaints about additional parameters device doesn't understand * * @TODO Document or eliminate the implied restriction of target names */ if (strstr(line, "invalid") || strstr(line, "variable")) { crm_debug("Skipping list output line: %s", line); return; } // Process line content, character by character for (i = 0; i <= len; i++) { if (isspace(line[i]) || (line[i] == ',') || (line[i] == ';') || (line[i] == '\0')) { // We've found a separator (i.e. the end of an entry) int rc = 0; char *entry = NULL; if (i == entry_start) { // Skip leading and sequential separators entry_start = i + 1; continue; } entry = pcmk__assert_alloc(i - entry_start + 1, sizeof(char)); /* Read entry, stopping at first separator * * @TODO Document or eliminate these character restrictions */ rc = sscanf(line + entry_start, "%[a-zA-Z0-9_-.]", entry); if (rc != 1) { crm_warn("Could not parse list output entry: %s " QB_XS " entry_start=%d position=%d", line + entry_start, entry_start, i); free(entry); } else if (pcmk__strcase_any_of(entry, PCMK_ACTION_ON, PCMK_ACTION_OFF, NULL)) { /* Some agents print the target status in the list output, * though none are known now (the separate list-status command * is used for this, but it can also print "UNKNOWN"). To handle * this possibility, skip such entries. * * @TODO Document or eliminate the implied restriction of target * names. */ free(entry); } else { // We have a valid entry *output = g_list_append(*output, entry); } entry_start = i + 1; } } } /*! * \internal * \brief Parse a list of targets from a string * * \param[in] list_output Target list as a string * * \return List of target names * \note The target list string format is flexible, to allow for user-specified * lists such pcmk_host_list and the output of an agent's list action * (whether direct or via the API, which escapes newlines). There may be * multiple lines, separated by either a newline or an escaped newline * (backslash n). Each line may have one or more target names, separated * by any combination of whitespace, commas, and semi-colons. Lines * containing "invalid" or "variable" will be ignored entirely. Target * names "on" or "off" (case-insensitive) will be ignored. Target names * may contain only alphanumeric characters, underbars (_), dashes (-), * and dots (.) (if any other character occurs in the name, it and all * subsequent characters in the name will be ignored). * \note The caller is responsible for freeing the result with * g_list_free_full(result, free). */ GList * stonith__parse_targets(const char *target_spec) { GList *targets = NULL; if (target_spec != NULL) { size_t out_len = strlen(target_spec); size_t line_start = 0; // Starting index of line being processed for (size_t i = 0; i <= out_len; ++i) { if ((target_spec[i] == '\n') || (target_spec[i] == '\0') || ((target_spec[i] == '\\') && (target_spec[i + 1] == 'n'))) { // We've reached the end of one line of output int len = i - line_start; if (len > 0) { char *line = strndup(target_spec + line_start, len); pcmk__assert(line != NULL); // cppcheck-suppress nullPointerOutOfMemory line[len] = '\0'; // Because it might be a newline parse_list_line(line, len, &targets); free(line); } if (target_spec[i] == '\\') { ++i; // backslash-n takes up two positions } line_start = i + 1; } } } return targets; } /*! * \internal * \brief Check whether a fencing failure was followed by an equivalent success * * \param[in] event Fencing failure * \param[in] top_history Complete fencing history (must be sorted by * stonith__sort_history() beforehand) * * \return The name of the node that executed the fencing if a later successful * event exists, or NULL if no such event exists */ const char * stonith__later_succeeded(const stonith_history_t *event, const stonith_history_t *top_history) { const char *other = NULL; for (const stonith_history_t *prev_hp = top_history; prev_hp != NULL; prev_hp = prev_hp->next) { if (prev_hp == event) { break; } if ((prev_hp->state == st_done) && pcmk__str_eq(event->target, prev_hp->target, pcmk__str_casei) && pcmk__str_eq(event->action, prev_hp->action, pcmk__str_none) && ((event->completed < prev_hp->completed) || ((event->completed == prev_hp->completed) && (event->completed_nsec < prev_hp->completed_nsec)))) { if ((event->delegate == NULL) || pcmk__str_eq(event->delegate, prev_hp->delegate, pcmk__str_casei)) { // Prefer equivalent fencing by same executioner return prev_hp->delegate; } else if (other == NULL) { // Otherwise remember first successful executioner other = (prev_hp->delegate == NULL)? "some node" : prev_hp->delegate; } } } return other; } /*! * \internal * \brief Sort fencing history, pending first then by most recently completed * * \param[in,out] history List of stonith actions * * \return New head of sorted \p history */ stonith_history_t * stonith__sort_history(stonith_history_t *history) { stonith_history_t *new = NULL, *pending = NULL, *hp, *np, *tmp; for (hp = history; hp; ) { tmp = hp->next; if ((hp->state == st_done) || (hp->state == st_failed)) { /* sort into new */ if ((!new) || (hp->completed > new->completed) || ((hp->completed == new->completed) && (hp->completed_nsec > new->completed_nsec))) { hp->next = new; new = hp; } else { np = new; do { if ((!np->next) || (hp->completed > np->next->completed) || ((hp->completed == np->next->completed) && (hp->completed_nsec > np->next->completed_nsec))) { hp->next = np->next; np->next = hp; break; } np = np->next; } while (1); } } else { /* put into pending */ hp->next = pending; pending = hp; } hp = tmp; } /* pending actions don't have a completed-stamp so make them go front */ if (pending) { stonith_history_t *last_pending = pending; while (last_pending->next) { last_pending = last_pending->next; } last_pending->next = new; new = pending; } return new; } /*! * \internal * \brief Return string equivalent of a fencing operation state value * * \param[in] state Fencing operation state value * * \return Human-friendly string equivalent of \p state */ const char * stonith__op_state_text(enum op_state state) { // @COMPAT Move this to the fencer after dropping stonith_op_state_str() switch (state) { case st_query: return "querying"; case st_exec: return "executing"; case st_done: return "completed"; case st_duplicate: return "duplicate"; case st_failed: return "failed"; default: return "unknown"; } } stonith_history_t * stonith__first_matching_event(stonith_history_t *history, bool (*matching_fn)(stonith_history_t *, void *), void *user_data) { for (stonith_history_t *hp = history; hp; hp = hp->next) { if (matching_fn(hp, user_data)) { return hp; } } return NULL; } bool stonith__event_state_pending(stonith_history_t *history, void *user_data) { return history->state != st_failed && history->state != st_done; } bool stonith__event_state_eq(stonith_history_t *history, void *user_data) { return history->state == GPOINTER_TO_INT(user_data); } bool stonith__event_state_neq(stonith_history_t *history, void *user_data) { return history->state != GPOINTER_TO_INT(user_data); } /*! * \internal * \brief Check whether a given parameter exists in a fence agent's metadata * * \param[in] metadata Agent metadata * \param[in] name Parameter name * * \retval \c true If \p name exists as a parameter in \p metadata * \retval \c false Otherwise */ static bool param_is_supported(xmlNode *metadata, const char *name) { char *xpath_s = crm_strdup_printf("//" PCMK_XE_PARAMETER "[@" PCMK_XA_NAME "='%s']", name); xmlXPathObject *xpath = pcmk__xpath_search(metadata->doc, xpath_s); bool supported = (pcmk__xpath_num_results(xpath) > 0); free(xpath_s); xmlXPathFreeObject(xpath); return supported; } /*! * \internal * \brief Get the default host argument based on a device's agent metadata * * If an agent supports the "plug" parameter, default to that. Otherwise default * to the "port" parameter if supported. Otherwise return \c NULL. * * \param[in] metadata Agent metadata * * \return Parameter name for default host argument */ const char * stonith__default_host_arg(xmlNode *metadata) { CRM_CHECK(metadata != NULL, return NULL); if (param_is_supported(metadata, "plug")) { return "plug"; } if (param_is_supported(metadata, "port")) { return "port"; } return NULL; } /*! * \internal * \brief Retrieve fence agent meta-data asynchronously * * \param[in] agent Agent to execute * \param[in] timeout_sec Error if not complete within this time * \param[in] callback Function to call with result (this will always be * called, whether by this function directly or * later via the main loop, and on success the * metadata will be in its result argument's * action_stdout) * \param[in,out] user_data User data to pass to callback * * \return Standard Pacemaker return code * \note The caller must use a main loop. This function is not a * stonith_api_operations_t method because it does not need a stonith_t * object and does not go through the fencer, but executes the agent * directly. */ int stonith__metadata_async(const char *agent, int timeout_sec, void (*callback)(int pid, const pcmk__action_result_t *result, void *user_data), void *user_data) { switch (get_namespace_from_agent(agent)) { case st_namespace_rhcs: { stonith_action_t *action = NULL; int rc = pcmk_ok; action = stonith__action_create(agent, PCMK_ACTION_METADATA, NULL, timeout_sec, NULL, NULL, NULL); rc = stonith__execute_async(action, user_data, callback, NULL); if (rc != pcmk_ok) { callback(0, stonith__action_result(action), user_data); stonith__destroy_action(action); } return pcmk_legacy2rc(rc); } #if HAVE_STONITH_STONITH_H case st_namespace_lha: // LHA metadata is simply synthesized, so simulate async { pcmk__action_result_t result = { .exit_status = CRM_EX_OK, .execution_status = PCMK_EXEC_DONE, .exit_reason = NULL, .action_stdout = NULL, .action_stderr = NULL, }; stonith__lha_metadata(agent, timeout_sec, &result.action_stdout); callback(0, &result, user_data); pcmk__reset_result(&result); return pcmk_rc_ok; } #endif default: { pcmk__action_result_t result = { .exit_status = CRM_EX_NOSUCH, .execution_status = PCMK_EXEC_ERROR_HARD, .exit_reason = crm_strdup_printf("No such agent '%s'", agent), .action_stdout = NULL, .action_stderr = NULL, }; callback(0, &result, user_data); pcmk__reset_result(&result); return ENOENT; } } } /*! * \internal * \brief Return the exit status from an async action callback * * \param[in] data Callback data * * \return Exit status from callback data */ int stonith__exit_status(const stonith_callback_data_t *data) { if ((data == NULL) || (data->opaque == NULL)) { return CRM_EX_ERROR; } return ((pcmk__action_result_t *) data->opaque)->exit_status; } /*! * \internal * \brief Return the execution status from an async action callback * * \param[in] data Callback data * * \return Execution status from callback data */ int stonith__execution_status(const stonith_callback_data_t *data) { if ((data == NULL) || (data->opaque == NULL)) { return PCMK_EXEC_UNKNOWN; } return ((pcmk__action_result_t *) data->opaque)->execution_status; } /*! * \internal * \brief Return the exit reason from an async action callback * * \param[in] data Callback data * * \return Exit reason from callback data */ const char * stonith__exit_reason(const stonith_callback_data_t *data) { if ((data == NULL) || (data->opaque == NULL)) { return NULL; } return ((pcmk__action_result_t *) data->opaque)->exit_reason; } /*! * \internal * \brief Return the exit status from an event notification * * \param[in] event Event * * \return Exit status from event */ int stonith__event_exit_status(const stonith_event_t *event) { if ((event == NULL) || (event->opaque == NULL)) { return CRM_EX_ERROR; } else { struct event_private *event_private = event->opaque; return event_private->result.exit_status; } } /*! * \internal * \brief Return the execution status from an event notification * * \param[in] event Event * * \return Execution status from event */ int stonith__event_execution_status(const stonith_event_t *event) { if ((event == NULL) || (event->opaque == NULL)) { return PCMK_EXEC_UNKNOWN; } else { struct event_private *event_private = event->opaque; return event_private->result.execution_status; } } /*! * \internal * \brief Return the exit reason from an event notification * * \param[in] event Event * * \return Exit reason from event */ const char * stonith__event_exit_reason(const stonith_event_t *event) { if ((event == NULL) || (event->opaque == NULL)) { return NULL; } else { struct event_private *event_private = event->opaque; return event_private->result.exit_reason; } } /*! * \internal * \brief Return a human-friendly description of a fencing event * * \param[in] event Event to describe * * \return Newly allocated string with description of \p event * \note The caller is responsible for freeing the return value. * This function asserts on memory errors and never returns NULL. */ char * stonith__event_description(const stonith_event_t *event) { // Use somewhat readable defaults const char *origin = pcmk__s(event->client_origin, "a client"); const char *origin_node = pcmk__s(event->origin, "a node"); const char *executioner = pcmk__s(event->executioner, "the cluster"); const char *device = pcmk__s(event->device, "unknown"); const char *action = pcmk__s(event->action, event->operation); const char *target = pcmk__s(event->target, "no node"); const char *reason = stonith__event_exit_reason(event); const char *status; if (action == NULL) { action = "(unknown)"; } if (stonith__event_execution_status(event) != PCMK_EXEC_DONE) { status = pcmk_exec_status_str(stonith__event_execution_status(event)); } else if (stonith__event_exit_status(event) != CRM_EX_OK) { status = pcmk_exec_status_str(PCMK_EXEC_ERROR); } else { status = crm_exit_str(CRM_EX_OK); } if (pcmk__str_eq(event->operation, PCMK__VALUE_ST_NOTIFY_HISTORY, pcmk__str_none)) { return crm_strdup_printf("Fencing history may have changed"); } else if (pcmk__str_eq(event->operation, STONITH_OP_DEVICE_ADD, pcmk__str_none)) { return crm_strdup_printf("A fencing device (%s) was added", device); } else if (pcmk__str_eq(event->operation, STONITH_OP_DEVICE_DEL, pcmk__str_none)) { return crm_strdup_printf("A fencing device (%s) was removed", device); } else if (pcmk__str_eq(event->operation, STONITH_OP_LEVEL_ADD, pcmk__str_none)) { return crm_strdup_printf("A fencing topology level (%s) was added", device); } else if (pcmk__str_eq(event->operation, STONITH_OP_LEVEL_DEL, pcmk__str_none)) { return crm_strdup_printf("A fencing topology level (%s) was removed", device); } // event->operation should be PCMK__VALUE_ST_NOTIFY_FENCE at this point return crm_strdup_printf("Operation %s of %s by %s for %s@%s: %s%s%s%s (ref=%s)", action, target, executioner, origin, origin_node, status, ((reason == NULL)? "" : " ("), pcmk__s(reason, ""), ((reason == NULL)? "" : ")"), pcmk__s(event->id, "(none)")); } // Deprecated functions kept only for backward API compatibility // LCOV_EXCL_START // See comments in stonith-ng.h for why we re-declare before defining stonith_t *stonith_api_new(void); stonith_t * stonith_api_new(void) { return stonith__api_new(); } void stonith_api_delete(stonith_t *stonith); void stonith_api_delete(stonith_t *stonith) { stonith__api_free(stonith); } static void stonith_dump_pending_op(gpointer key, gpointer value, gpointer user_data) { int call = GPOINTER_TO_INT(key); stonith_callback_client_t *blob = value; crm_debug("Call %d (%s): pending", call, pcmk__s(blob->id, "no ID")); } void stonith_dump_pending_callbacks(stonith_t *stonith); void stonith_dump_pending_callbacks(stonith_t *stonith) { stonith_private_t *private = stonith->st_private; if (private->stonith_op_callback_table == NULL) { return; } return g_hash_table_foreach(private->stonith_op_callback_table, stonith_dump_pending_op, NULL); } bool stonith_dispatch(stonith_t *stonith_api); bool stonith_dispatch(stonith_t *stonith_api) { return (stonith__api_dispatch(stonith_api) == pcmk_rc_ok); } stonith_key_value_t *stonith_key_value_add(stonith_key_value_t *head, const char *key, const char *value); stonith_key_value_t * stonith_key_value_add(stonith_key_value_t *head, const char *key, const char *value) { return stonith__key_value_add(head, key, value); } void stonith_key_value_freeall(stonith_key_value_t *head, int keys, int values); void stonith_key_value_freeall(stonith_key_value_t *head, int keys, int values) { stonith__key_value_freeall(head, (keys != 0), (values != 0)); } void stonith_history_free(stonith_history_t *head); void stonith_history_free(stonith_history_t *head) { stonith__history_free(head); } int stonith_api_connect_retry(stonith_t *st, const char *name, int max_attempts); int stonith_api_connect_retry(stonith_t *st, const char *name, int max_attempts) { return pcmk_rc2legacy(stonith__api_connect_retry(st, name, max_attempts)); } const char *stonith_op_state_str(enum op_state state); const char * stonith_op_state_str(enum op_state state) { return stonith__op_state_text(state); } bool stonith_agent_exists(const char *agent, int timeout); bool stonith_agent_exists(const char *agent, int timeout) { return stonith__agent_exists(agent); } const char *stonith_action_str(const char *action); const char * stonith_action_str(const char *action) { if (action == NULL) { return "fencing"; } else if (strcmp(action, PCMK_ACTION_ON) == 0) { return "unfencing"; } else if (strcmp(action, PCMK_ACTION_OFF) == 0) { return "turning off"; } else { return action; } } enum stonith_namespace stonith_text2namespace(const char *namespace_s); enum stonith_namespace stonith_text2namespace(const char *namespace_s) { return parse_namespace(namespace_s); } const char *stonith_namespace2text(enum stonith_namespace st_namespace); const char * stonith_namespace2text(enum stonith_namespace st_namespace) { return namespace_text(st_namespace); } enum stonith_namespace stonith_get_namespace(const char *agent, const char *namespace_s); enum stonith_namespace stonith_get_namespace(const char *agent, const char *namespace_s) { if (pcmk__str_eq(namespace_s, "internal", pcmk__str_none)) { return st_namespace_internal; } return get_namespace_from_agent(agent); } // LCOV_EXCL_STOP // End deprecated API diff --git a/lib/fencing/st_lha.c b/lib/fencing/st_lha.c index 90bc88015c..20a060dacc 100644 --- a/lib/fencing/st_lha.c +++ b/lib/fencing/st_lha.c @@ -1,309 +1,309 @@ /* * Copyright 2004-2025 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 #include #include #include "fencing_private.h" #define LHA_STONITH_LIBRARY "libstonith.so.1" static void *lha_agents_lib = NULL; // @TODO Use XML string constants and maybe a real XML object static const char META_TEMPLATE[] = "\n" "<" PCMK_XE_RESOURCE_AGENT " " PCMK_XA_NAME "=\"%s\">\n" " <" PCMK_XE_VERSION ">1.1\n" " <" PCMK_XE_LONGDESC " " PCMK_XA_LANG "=\"" PCMK__VALUE_EN "\">\n" "%s\n" " \n" " <" PCMK_XE_SHORTDESC " " PCMK_XA_LANG "=\"" PCMK__VALUE_EN "\">" "%s" "\n" "%s\n" " <" PCMK_XE_ACTIONS ">\n" " <" PCMK_XE_ACTION " " PCMK_XA_NAME "=\"" PCMK_ACTION_START "\"" " " PCMK_META_TIMEOUT "=\"%s\" />\n" " <" PCMK_XE_ACTION " " PCMK_XA_NAME "=\"" PCMK_ACTION_STOP "\"" " " PCMK_META_TIMEOUT "=\"15s\" />\n" " <" PCMK_XE_ACTION " " PCMK_XA_NAME "=\"" PCMK_ACTION_STATUS "\"" " " PCMK_META_TIMEOUT "=\"%s\" />\n" " <" PCMK_XE_ACTION " " PCMK_XA_NAME "=\"" PCMK_ACTION_MONITOR "\"" " " PCMK_META_TIMEOUT "=\"%s\"" " " PCMK_META_INTERVAL "=\"3600s\" />\n" " <" PCMK_XE_ACTION " " PCMK_XA_NAME "=\"" PCMK_ACTION_META_DATA "\"" " " PCMK_META_TIMEOUT "=\"15s\" />\n" " \n" " <" PCMK_XE_SPECIAL " " PCMK_XA_TAG "=\"heartbeat\">\n" " <" PCMK_XE_VERSION ">2.0\n" " \n" "\n"; static void * find_library_function(void **handle, const char *lib, const char *fn) { void *a_function; if (*handle == NULL) { *handle = dlopen(lib, RTLD_LAZY); if ((*handle) == NULL) { crm_err("Could not open %s: %s", lib, dlerror()); return NULL; } } a_function = dlsym(*handle, fn); if (a_function == NULL) { crm_err("Could not find %s in %s: %s", fn, lib, dlerror()); } return a_function; } /*! * \internal * \brief Check whether a given fence agent is an LHA agent * * \param[in] agent Fence agent type * * \return true if \p agent is an LHA agent, otherwise false */ bool stonith__agent_is_lha(const char *agent) { Stonith *stonith_obj = NULL; static bool need_init = true; static Stonith *(*st_new_fn) (const char *) = NULL; static void (*st_del_fn) (Stonith *) = NULL; if (need_init) { need_init = false; st_new_fn = find_library_function(&lha_agents_lib, LHA_STONITH_LIBRARY, "stonith_new"); st_del_fn = find_library_function(&lha_agents_lib, LHA_STONITH_LIBRARY, "stonith_delete"); } if (lha_agents_lib && st_new_fn && st_del_fn) { stonith_obj = (*st_new_fn) (agent); if (stonith_obj) { (*st_del_fn) (stonith_obj); return true; } } return false; } int stonith__list_lha_agents(stonith_key_value_t **devices) { static gboolean need_init = TRUE; int count = 0; char **entry = NULL; char **type_list = NULL; static char **(*type_list_fn) (void) = NULL; static void (*type_free_fn) (char **) = NULL; if (need_init) { need_init = FALSE; type_list_fn = find_library_function(&lha_agents_lib, LHA_STONITH_LIBRARY, "stonith_types"); type_free_fn = find_library_function(&lha_agents_lib, LHA_STONITH_LIBRARY, "stonith_free_hostlist"); } if (type_list_fn) { type_list = (*type_list_fn) (); } for (entry = type_list; entry != NULL && *entry; ++entry) { crm_trace("Added: %s", *entry); *devices = stonith__key_value_add(*devices, NULL, *entry); count++; } if (type_list && type_free_fn) { (*type_free_fn) (type_list); } return count; } static void stonith_plugin(int priority, const char *fmt, ...) G_GNUC_PRINTF(2, 3); static void stonith_plugin(int priority, const char *format, ...) { int err = errno; va_list ap; int len = 0; char *string = NULL; va_start(ap, format); len = vasprintf (&string, format, ap); va_end(ap); pcmk__assert(len > 0); do_crm_log_alias(priority, __FILE__, __func__, __LINE__, "%s", string); free(string); errno = err; } int stonith__lha_metadata(const char *agent, int timeout, char **output) { int rc = 0; char *buffer = NULL; static const char *no_parameter_info = ""; Stonith *stonith_obj = NULL; static gboolean need_init = TRUE; static Stonith *(*st_new_fn) (const char *) = NULL; static const char *(*st_info_fn) (Stonith *, int) = NULL; static void (*st_del_fn) (Stonith *) = NULL; static void (*st_log_fn) (Stonith *, PILLogFun) = NULL; if (need_init) { need_init = FALSE; st_new_fn = find_library_function(&lha_agents_lib, LHA_STONITH_LIBRARY, "stonith_new"); st_del_fn = find_library_function(&lha_agents_lib, LHA_STONITH_LIBRARY, "stonith_delete"); st_log_fn = find_library_function(&lha_agents_lib, LHA_STONITH_LIBRARY, "stonith_set_log"); st_info_fn = find_library_function(&lha_agents_lib, LHA_STONITH_LIBRARY, "stonith_get_info"); } if (lha_agents_lib && st_new_fn && st_del_fn && st_info_fn && st_log_fn) { char *meta_longdesc = NULL; char *meta_shortdesc = NULL; char *meta_param = NULL; const char *timeout_str = NULL; gchar *meta_longdesc_esc = NULL; gchar *meta_shortdesc_esc = NULL; stonith_obj = st_new_fn(agent); if (stonith_obj != NULL) { st_log_fn(stonith_obj, (PILLogFun) &stonith_plugin); /* A st_info_fn() may free any existing output buffer every time * when it's called. Copy the output every time. */ meta_longdesc = pcmk__str_copy(st_info_fn(stonith_obj, ST_DEVICEDESCR)); if (meta_longdesc == NULL) { crm_warn("no long description in %s's metadata.", agent); meta_longdesc = pcmk__str_copy(no_parameter_info); } meta_shortdesc = pcmk__str_copy(st_info_fn(stonith_obj, ST_DEVICEID)); if (meta_shortdesc == NULL) { crm_warn("no short description in %s's metadata.", agent); meta_shortdesc = pcmk__str_copy(no_parameter_info); } meta_param = pcmk__str_copy(st_info_fn(stonith_obj, ST_CONF_XML)); if (meta_param == NULL) { crm_warn("no list of parameters in %s's metadata.", agent); meta_param = pcmk__str_copy(no_parameter_info); } st_del_fn(stonith_obj); } else { errno = EINVAL; - crm_perror(LOG_ERR, "Agent %s not found", agent); + crm_err("Agent %s not found", agent); return -EINVAL; } if (pcmk__xml_needs_escape(meta_longdesc, pcmk__xml_escape_text)) { meta_longdesc_esc = pcmk__xml_escape(meta_longdesc, pcmk__xml_escape_text); } if (pcmk__xml_needs_escape(meta_shortdesc, pcmk__xml_escape_text)) { meta_shortdesc_esc = pcmk__xml_escape(meta_shortdesc, pcmk__xml_escape_text); } /* @TODO This needs a string that's parsable by crm_get_msec(). In * general, pcmk__readable_interval() doesn't provide that. It works * here because PCMK_DEFAULT_ACTION_TIMEOUT_MS is 20000 -> "20s". */ timeout_str = pcmk__readable_interval(PCMK_DEFAULT_ACTION_TIMEOUT_MS); buffer = crm_strdup_printf(META_TEMPLATE, agent, ((meta_longdesc_esc != NULL) ? meta_longdesc_esc : meta_longdesc), ((meta_shortdesc_esc != NULL) ? meta_shortdesc_esc : meta_shortdesc), meta_param, timeout_str, timeout_str, timeout_str); g_free(meta_longdesc_esc); g_free(meta_shortdesc_esc); free(meta_longdesc); free(meta_shortdesc); free(meta_param); } if (output) { *output = buffer; } else { free(buffer); } return rc; } /* Implement a dummy function that uses -lpils so that linkers don't drop the * reference. */ #include const char *i_hate_pils(int rc); const char * i_hate_pils(int rc) { return PIL_strerror(rc); } int stonith__lha_validate(stonith_t *st, int call_options, const char *target, const char *agent, GHashTable *params, int timeout, char **output, char **error_output) { errno = EOPNOTSUPP; - crm_perror(LOG_ERR, "Cannot validate Linux-HA fence agents"); + crm_err("Cannot validate Linux-HA fence agents"); return -EOPNOTSUPP; } diff --git a/lib/lrmd/lrmd_client.c b/lib/lrmd/lrmd_client.c index a9c50a3474..3c51a4bd05 100644 --- a/lib/lrmd/lrmd_client.c +++ b/lib/lrmd/lrmd_client.c @@ -1,2675 +1,2677 @@ /* * Copyright 2012-2025 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 // uint32_t, uint64_t #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include // stonith__* #include #include #include #include #include #include #define MAX_TLS_RECV_WAIT 10000 CRM_TRACE_INIT_DATA(lrmd); static int lrmd_api_disconnect(lrmd_t * lrmd); static int lrmd_api_is_connected(lrmd_t * lrmd); /* IPC proxy functions */ int lrmd_internal_proxy_send(lrmd_t * lrmd, xmlNode *msg); static void lrmd_internal_proxy_dispatch(lrmd_t *lrmd, xmlNode *msg); void lrmd_internal_set_proxy_callback(lrmd_t * lrmd, void *userdata, void (*callback)(lrmd_t *lrmd, void *userdata, xmlNode *msg)); // GnuTLS client handshake timeout in seconds #define TLS_HANDSHAKE_TIMEOUT 5 static void lrmd_tls_disconnect(lrmd_t * lrmd); static int global_remote_msg_id = 0; static void lrmd_tls_connection_destroy(gpointer userdata); static int add_tls_to_mainloop(lrmd_t *lrmd, bool do_api_handshake); typedef struct lrmd_private_s { uint64_t type; char *token; mainloop_io_t *source; /* IPC parameters */ crm_ipc_t *ipc; pcmk__remote_t *remote; /* Extra TLS parameters */ char *remote_nodename; char *server; int port; pcmk__tls_t *tls; /* while the async connection is occurring, this is the id * of the connection timeout timer. */ int async_timer; int sock; /* since tls requires a round trip across the network for a * request/reply, there are times where we just want to be able * to send a request from the client and not wait around (or even care * about) what the reply is. */ int expected_late_replies; GList *pending_notify; crm_trigger_t *process_notify; crm_trigger_t *handshake_trigger; lrmd_event_callback callback; /* Internal IPC proxy msg passing for remote guests */ void (*proxy_callback)(lrmd_t *lrmd, void *userdata, xmlNode *msg); void *proxy_callback_userdata; char *peer_version; } lrmd_private_t; static int process_lrmd_handshake_reply(xmlNode *reply, lrmd_private_t *native); static void report_async_connection_result(lrmd_t * lrmd, int rc); static lrmd_list_t * lrmd_list_add(lrmd_list_t * head, const char *value) { lrmd_list_t *p, *end; p = pcmk__assert_alloc(1, sizeof(lrmd_list_t)); p->val = strdup(value); end = head; while (end && end->next) { end = end->next; } if (end) { end->next = p; } else { head = p; } return head; } void lrmd_list_freeall(lrmd_list_t * head) { lrmd_list_t *p; while (head) { char *val = (char *)head->val; p = head->next; free(val); free(head); head = p; } } lrmd_key_value_t * lrmd_key_value_add(lrmd_key_value_t * head, const char *key, const char *value) { lrmd_key_value_t *p, *end; p = pcmk__assert_alloc(1, sizeof(lrmd_key_value_t)); p->key = strdup(key); p->value = strdup(value); end = head; while (end && end->next) { end = end->next; } if (end) { end->next = p; } else { head = p; } return head; } void lrmd_key_value_freeall(lrmd_key_value_t * head) { lrmd_key_value_t *p; while (head) { p = head->next; free(head->key); free(head->value); free(head); head = p; } } /*! * \brief Create a new lrmd_event_data_t object * * \param[in] rsc_id ID of resource involved in event * \param[in] task Action name * \param[in] interval_ms Action interval * * \return Newly allocated and initialized lrmd_event_data_t * \note This functions asserts on memory errors, so the return value is * guaranteed to be non-NULL. The caller is responsible for freeing the * result with lrmd_free_event(). */ lrmd_event_data_t * lrmd_new_event(const char *rsc_id, const char *task, guint interval_ms) { lrmd_event_data_t *event = pcmk__assert_alloc(1, sizeof(lrmd_event_data_t)); // lrmd_event_data_t has (const char *) members that lrmd_free_event() frees event->rsc_id = pcmk__str_copy(rsc_id); event->op_type = pcmk__str_copy(task); event->interval_ms = interval_ms; return event; } lrmd_event_data_t * lrmd_copy_event(lrmd_event_data_t * event) { lrmd_event_data_t *copy = NULL; copy = pcmk__assert_alloc(1, sizeof(lrmd_event_data_t)); copy->type = event->type; // lrmd_event_data_t has (const char *) members that lrmd_free_event() frees copy->rsc_id = pcmk__str_copy(event->rsc_id); copy->op_type = pcmk__str_copy(event->op_type); copy->user_data = pcmk__str_copy(event->user_data); copy->output = pcmk__str_copy(event->output); copy->remote_nodename = pcmk__str_copy(event->remote_nodename); copy->exit_reason = pcmk__str_copy(event->exit_reason); copy->call_id = event->call_id; copy->timeout = event->timeout; copy->interval_ms = event->interval_ms; copy->start_delay = event->start_delay; copy->rsc_deleted = event->rsc_deleted; copy->rc = event->rc; copy->op_status = event->op_status; copy->t_run = event->t_run; copy->t_rcchange = event->t_rcchange; copy->exec_time = event->exec_time; copy->queue_time = event->queue_time; copy->connection_rc = event->connection_rc; copy->params = pcmk__str_table_dup(event->params); return copy; } /*! * \brief Free an executor event * * \param[in,out] Executor event object to free */ void lrmd_free_event(lrmd_event_data_t *event) { if (event == NULL) { return; } // @TODO Why are these const char *? free((void *) event->rsc_id); free((void *) event->op_type); free((void *) event->user_data); free((void *) event->remote_nodename); lrmd__reset_result(event); if (event->params != NULL) { g_hash_table_destroy(event->params); } free(event); } static void lrmd_dispatch_internal(gpointer data, gpointer user_data) { xmlNode *msg = data; lrmd_t *lrmd = user_data; const char *type; const char *proxy_session = pcmk__xe_get(msg, PCMK__XA_LRMD_IPC_SESSION); lrmd_private_t *native = lrmd->lrmd_private; lrmd_event_data_t event = { 0, }; if (proxy_session != NULL) { /* this is proxy business */ lrmd_internal_proxy_dispatch(lrmd, msg); return; } else if (!native->callback) { /* no callback set */ crm_trace("notify event received but client has not set callback"); return; } event.remote_nodename = native->remote_nodename; type = pcmk__xe_get(msg, PCMK__XA_LRMD_OP); pcmk__xe_get_int(msg, PCMK__XA_LRMD_CALLID, &event.call_id); event.rsc_id = pcmk__xe_get(msg, PCMK__XA_LRMD_RSC_ID); if (pcmk__str_eq(type, LRMD_OP_RSC_REG, pcmk__str_none)) { event.type = lrmd_event_register; } else if (pcmk__str_eq(type, LRMD_OP_RSC_UNREG, pcmk__str_none)) { event.type = lrmd_event_unregister; } else if (pcmk__str_eq(type, LRMD_OP_RSC_EXEC, pcmk__str_none)) { int rc = 0; int exec_time = 0; int queue_time = 0; pcmk__xe_get_int(msg, PCMK__XA_LRMD_TIMEOUT, &event.timeout); pcmk__xe_get_guint(msg, PCMK__XA_LRMD_RSC_INTERVAL, &event.interval_ms); pcmk__xe_get_int(msg, PCMK__XA_LRMD_RSC_START_DELAY, &event.start_delay); pcmk__xe_get_int(msg, PCMK__XA_LRMD_EXEC_RC, &rc); event.rc = (enum ocf_exitcode) rc; pcmk__xe_get_int(msg, PCMK__XA_LRMD_EXEC_OP_STATUS, &event.op_status); pcmk__xe_get_int(msg, PCMK__XA_LRMD_RSC_DELETED, &event.rsc_deleted); pcmk__xe_get_time(msg, PCMK__XA_LRMD_RUN_TIME, &event.t_run); pcmk__xe_get_time(msg, PCMK__XA_LRMD_RCCHANGE_TIME, &event.t_rcchange); pcmk__xe_get_int(msg, PCMK__XA_LRMD_EXEC_TIME, &exec_time); CRM_LOG_ASSERT(exec_time >= 0); event.exec_time = QB_MAX(0, exec_time); pcmk__xe_get_int(msg, PCMK__XA_LRMD_QUEUE_TIME, &queue_time); CRM_LOG_ASSERT(queue_time >= 0); event.queue_time = QB_MAX(0, queue_time); event.op_type = pcmk__xe_get(msg, PCMK__XA_LRMD_RSC_ACTION); event.user_data = pcmk__xe_get(msg, PCMK__XA_LRMD_RSC_USERDATA_STR); event.type = lrmd_event_exec_complete; /* output and exit_reason may be freed by a callback */ event.output = pcmk__xe_get_copy(msg, PCMK__XA_LRMD_RSC_OUTPUT); lrmd__set_result(&event, event.rc, event.op_status, pcmk__xe_get(msg, PCMK__XA_LRMD_RSC_EXIT_REASON)); event.params = xml2list(msg); } else if (pcmk__str_eq(type, LRMD_OP_NEW_CLIENT, pcmk__str_none)) { event.type = lrmd_event_new_client; } else if (pcmk__str_eq(type, LRMD_OP_POKE, pcmk__str_none)) { event.type = lrmd_event_poke; } else { return; } crm_trace("op %s notify event received", type); native->callback(&event); if (event.params) { g_hash_table_destroy(event.params); } lrmd__reset_result(&event); } // \return Always 0, to indicate that IPC mainloop source should be kept static int lrmd_ipc_dispatch(const char *buffer, ssize_t length, gpointer userdata) { lrmd_t *lrmd = userdata; lrmd_private_t *native = lrmd->lrmd_private; if (native->callback != NULL) { xmlNode *msg = pcmk__xml_parse(buffer); lrmd_dispatch_internal(msg, lrmd); pcmk__xml_free(msg); } return 0; } static void lrmd_free_xml(gpointer userdata) { pcmk__xml_free((xmlNode *) userdata); } static bool remote_executor_connected(lrmd_t * lrmd) { lrmd_private_t *native = lrmd->lrmd_private; return (native->remote->tls_session != NULL); } static void handle_remote_msg(xmlNode *xml, lrmd_t *lrmd) { lrmd_private_t *native = lrmd->lrmd_private; const char *msg_type = NULL; msg_type = pcmk__xe_get(xml, PCMK__XA_LRMD_REMOTE_MSG_TYPE); if (pcmk__str_eq(msg_type, "notify", pcmk__str_casei)) { lrmd_dispatch_internal(xml, lrmd); } else if (pcmk__str_eq(msg_type, "reply", pcmk__str_casei)) { const char *op = pcmk__xe_get(xml, PCMK__XA_LRMD_OP); if (native->expected_late_replies > 0) { native->expected_late_replies--; /* The register op message we get as a response to lrmd_handshake_async * is a reply, so we have to handle that here. */ if (pcmk__str_eq(op, "register", pcmk__str_casei)) { int rc = process_lrmd_handshake_reply(xml, native); report_async_connection_result(lrmd, pcmk_rc2legacy(rc)); } } else { int reply_id = 0; pcmk__xe_get_int(xml, PCMK__XA_LRMD_CALLID, &reply_id); /* if this happens, we want to know about it */ crm_err("Got outdated Pacemaker Remote reply %d", reply_id); } } } /*! * \internal * \brief Notify trigger handler * * \param[in,out] userdata API connection * * \return Always return G_SOURCE_CONTINUE to leave this trigger handler in the * mainloop */ static int process_pending_notifies(gpointer userdata) { lrmd_t *lrmd = userdata; lrmd_private_t *native = lrmd->lrmd_private; if (native->pending_notify == NULL) { return G_SOURCE_CONTINUE; } crm_trace("Processing pending notifies"); g_list_foreach(native->pending_notify, lrmd_dispatch_internal, lrmd); g_list_free_full(native->pending_notify, lrmd_free_xml); native->pending_notify = NULL; return G_SOURCE_CONTINUE; } /*! * \internal * \brief TLS dispatch function for file descriptor sources * * \param[in,out] userdata API connection * * \return -1 on error to remove the source from the mainloop, or 0 otherwise * to leave it in the mainloop */ static int lrmd_tls_dispatch(gpointer userdata) { lrmd_t *lrmd = userdata; lrmd_private_t *native = lrmd->lrmd_private; xmlNode *xml = NULL; int rc = pcmk_rc_ok; if (!remote_executor_connected(lrmd)) { crm_trace("TLS dispatch triggered after disconnect"); return -1; } crm_trace("TLS dispatch triggered"); rc = pcmk__remote_ready(native->remote, 0); if (rc == pcmk_rc_ok) { rc = pcmk__read_remote_message(native->remote, -1); } if (rc != pcmk_rc_ok && rc != ETIME) { crm_info("Lost %s executor connection while reading data", (native->remote_nodename? native->remote_nodename : "local")); lrmd_tls_disconnect(lrmd); return -1; } /* If rc is ETIME, there was nothing to read but we may already have a * full message in the buffer */ xml = pcmk__remote_message_xml(native->remote); if (xml == NULL) { return 0; } handle_remote_msg(xml, lrmd); pcmk__xml_free(xml); return 0; } /* Not used with mainloop */ int lrmd_poll(lrmd_t * lrmd, int timeout) { lrmd_private_t *native = lrmd->lrmd_private; switch (native->type) { case pcmk__client_ipc: return crm_ipc_ready(native->ipc); case pcmk__client_tls: if (native->pending_notify) { return 1; } else { int rc = pcmk__remote_ready(native->remote, 0); switch (rc) { case pcmk_rc_ok: return 1; case ETIME: return 0; default: return pcmk_rc2legacy(rc); } } default: crm_err("Unsupported executor connection type (bug?): %d", native->type); return -EPROTONOSUPPORT; } } /* Not used with mainloop */ bool lrmd_dispatch(lrmd_t * lrmd) { lrmd_private_t *private = NULL; pcmk__assert(lrmd != NULL); private = lrmd->lrmd_private; switch (private->type) { case pcmk__client_ipc: while (crm_ipc_ready(private->ipc)) { if (crm_ipc_read(private->ipc) > 0) { const char *msg = crm_ipc_buffer(private->ipc); lrmd_ipc_dispatch(msg, strlen(msg), lrmd); pcmk__ipc_free_client_buffer(private->ipc); } } break; case pcmk__client_tls: lrmd_tls_dispatch(lrmd); break; default: crm_err("Unsupported executor connection type (bug?): %d", private->type); } if (lrmd_api_is_connected(lrmd) == FALSE) { crm_err("Connection closed"); return FALSE; } return TRUE; } static xmlNode * lrmd_create_op(const char *token, const char *op, xmlNode *data, int timeout, enum lrmd_call_options options) { xmlNode *op_msg = NULL; CRM_CHECK(token != NULL, return NULL); op_msg = pcmk__xe_create(NULL, PCMK__XE_LRMD_COMMAND); crm_xml_add(op_msg, PCMK__XA_T, PCMK__VALUE_LRMD); crm_xml_add(op_msg, PCMK__XA_LRMD_OP, op); crm_xml_add_int(op_msg, PCMK__XA_LRMD_TIMEOUT, timeout); crm_xml_add_int(op_msg, PCMK__XA_LRMD_CALLOPT, options); if (data != NULL) { xmlNode *wrapper = pcmk__xe_create(op_msg, PCMK__XE_LRMD_CALLDATA); pcmk__xml_copy(wrapper, data); } crm_trace("Created executor %s command with call options %.8lx (%d)", op, (long)options, options); return op_msg; } static void lrmd_ipc_connection_destroy(gpointer userdata) { lrmd_t *lrmd = userdata; lrmd_private_t *native = lrmd->lrmd_private; switch (native->type) { case pcmk__client_ipc: crm_info("Disconnected from local executor"); break; case pcmk__client_tls: crm_info("Disconnected from remote executor on %s", native->remote_nodename); break; default: crm_err("Unsupported executor connection type %d (bug?)", native->type); } /* Prevent these from being cleaned up in lrmd_api_disconnect() */ native->ipc = NULL; native->source = NULL; if (native->callback) { lrmd_event_data_t event = { 0, }; event.type = lrmd_event_disconnect; event.remote_nodename = native->remote_nodename; native->callback(&event); } } static void lrmd_tls_connection_destroy(gpointer userdata) { lrmd_t *lrmd = userdata; lrmd_private_t *native = lrmd->lrmd_private; crm_info("TLS connection destroyed"); if (native->remote->tls_session) { gnutls_bye(native->remote->tls_session, GNUTLS_SHUT_RDWR); gnutls_deinit(native->remote->tls_session); native->remote->tls_session = NULL; } if (native->tls) { pcmk__free_tls(native->tls); native->tls = NULL; } if (native->sock >= 0) { close(native->sock); } if (native->process_notify) { mainloop_destroy_trigger(native->process_notify); native->process_notify = NULL; } if (native->pending_notify) { g_list_free_full(native->pending_notify, lrmd_free_xml); native->pending_notify = NULL; } if (native->handshake_trigger != NULL) { mainloop_destroy_trigger(native->handshake_trigger); native->handshake_trigger = NULL; } free(native->remote->buffer); free(native->remote->start_state); native->remote->buffer = NULL; native->remote->start_state = NULL; native->source = 0; native->sock = -1; if (native->callback) { lrmd_event_data_t event = { 0, }; event.remote_nodename = native->remote_nodename; event.type = lrmd_event_disconnect; native->callback(&event); } return; } // \return Standard Pacemaker return code int lrmd__remote_send_xml(pcmk__remote_t *session, xmlNode *msg, uint32_t id, const char *msg_type) { crm_xml_add_int(msg, PCMK__XA_LRMD_REMOTE_MSG_ID, id); crm_xml_add(msg, PCMK__XA_LRMD_REMOTE_MSG_TYPE, msg_type); return pcmk__remote_send_xml(session, msg); } // \return Standard Pacemaker return code static int read_remote_reply(lrmd_t *lrmd, int total_timeout, int expected_reply_id, xmlNode **reply) { lrmd_private_t *native = lrmd->lrmd_private; time_t start = time(NULL); const char *msg_type = NULL; int reply_id = 0; int remaining_timeout = 0; int rc = pcmk_rc_ok; /* A timeout of 0 here makes no sense. We have to wait a period of time * for the response to come back. If -1 or 0, default to 10 seconds. */ if (total_timeout <= 0 || total_timeout > MAX_TLS_RECV_WAIT) { total_timeout = MAX_TLS_RECV_WAIT; } for (*reply = NULL; *reply == NULL; ) { *reply = pcmk__remote_message_xml(native->remote); if (*reply == NULL) { /* read some more off the tls buffer if we still have time left. */ if (remaining_timeout) { remaining_timeout = total_timeout - ((time(NULL) - start) * 1000); } else { remaining_timeout = total_timeout; } if (remaining_timeout <= 0) { return ETIME; } rc = pcmk__read_remote_message(native->remote, remaining_timeout); if (rc != pcmk_rc_ok) { return rc; } *reply = pcmk__remote_message_xml(native->remote); if (*reply == NULL) { return ENOMSG; } } pcmk__xe_get_int(*reply, PCMK__XA_LRMD_REMOTE_MSG_ID, &reply_id); msg_type = pcmk__xe_get(*reply, PCMK__XA_LRMD_REMOTE_MSG_TYPE); if (!msg_type) { crm_err("Empty msg type received while waiting for reply"); pcmk__xml_free(*reply); *reply = NULL; } else if (pcmk__str_eq(msg_type, "notify", pcmk__str_casei)) { /* got a notify while waiting for reply, trigger the notify to be processed later */ crm_info("queueing notify"); native->pending_notify = g_list_append(native->pending_notify, *reply); if (native->process_notify) { crm_info("notify trigger set."); mainloop_set_trigger(native->process_notify); } *reply = NULL; } else if (!pcmk__str_eq(msg_type, "reply", pcmk__str_casei)) { /* msg isn't a reply, make some noise */ crm_err("Expected a reply, got %s", msg_type); pcmk__xml_free(*reply); *reply = NULL; } else if (reply_id != expected_reply_id) { if (native->expected_late_replies > 0) { native->expected_late_replies--; } else { crm_err("Got outdated reply, expected id %d got id %d", expected_reply_id, reply_id); } pcmk__xml_free(*reply); *reply = NULL; } } if (native->remote->buffer && native->process_notify) { mainloop_set_trigger(native->process_notify); } return rc; } // \return Standard Pacemaker return code static int send_remote_message(lrmd_t *lrmd, xmlNode *msg) { int rc = pcmk_rc_ok; lrmd_private_t *native = lrmd->lrmd_private; global_remote_msg_id++; if (global_remote_msg_id <= 0) { global_remote_msg_id = 1; } rc = lrmd__remote_send_xml(native->remote, msg, global_remote_msg_id, "request"); if (rc != pcmk_rc_ok) { crm_err("Disconnecting because TLS message could not be sent to " "Pacemaker Remote: %s", pcmk_rc_str(rc)); lrmd_tls_disconnect(lrmd); } return rc; } static int lrmd_tls_send_recv(lrmd_t * lrmd, xmlNode * msg, int timeout, xmlNode ** reply) { int rc = 0; xmlNode *xml = NULL; if (!remote_executor_connected(lrmd)) { return -ENOTCONN; } rc = send_remote_message(lrmd, msg); if (rc != pcmk_rc_ok) { return pcmk_rc2legacy(rc); } rc = read_remote_reply(lrmd, timeout, global_remote_msg_id, &xml); if (rc != pcmk_rc_ok) { crm_err("Disconnecting remote after request %d reply not received: %s " QB_XS " rc=%d timeout=%dms", global_remote_msg_id, pcmk_rc_str(rc), rc, timeout); lrmd_tls_disconnect(lrmd); } if (reply) { *reply = xml; } else { pcmk__xml_free(xml); } return pcmk_rc2legacy(rc); } static int lrmd_send_xml(lrmd_t * lrmd, xmlNode * msg, int timeout, xmlNode ** reply) { int rc = pcmk_ok; lrmd_private_t *native = lrmd->lrmd_private; switch (native->type) { case pcmk__client_ipc: rc = crm_ipc_send(native->ipc, msg, crm_ipc_client_response, timeout, reply); break; case pcmk__client_tls: rc = lrmd_tls_send_recv(lrmd, msg, timeout, reply); break; default: crm_err("Unsupported executor connection type (bug?): %d", native->type); rc = -EPROTONOSUPPORT; } return rc; } static int lrmd_send_xml_no_reply(lrmd_t * lrmd, xmlNode * msg) { int rc = pcmk_ok; lrmd_private_t *native = lrmd->lrmd_private; switch (native->type) { case pcmk__client_ipc: rc = crm_ipc_send(native->ipc, msg, crm_ipc_flags_none, 0, NULL); break; case pcmk__client_tls: rc = send_remote_message(lrmd, msg); if (rc == pcmk_rc_ok) { /* we don't want to wait around for the reply, but * since the request/reply protocol needs to behave the same * as libqb, a reply will eventually come later anyway. */ native->expected_late_replies++; } rc = pcmk_rc2legacy(rc); break; default: crm_err("Unsupported executor connection type (bug?): %d", native->type); rc = -EPROTONOSUPPORT; } return rc; } static int lrmd_api_is_connected(lrmd_t * lrmd) { lrmd_private_t *native = lrmd->lrmd_private; switch (native->type) { case pcmk__client_ipc: return crm_ipc_connected(native->ipc); case pcmk__client_tls: return remote_executor_connected(lrmd); default: crm_err("Unsupported executor connection type (bug?): %d", native->type); return 0; } } /*! * \internal * \brief Send a prepared API command to the executor * * \param[in,out] lrmd Existing connection to the executor * \param[in] op Name of API command to send * \param[in] data Command data XML to add to the sent command * \param[out] output_data If expecting a reply, it will be stored here * \param[in] timeout Timeout in milliseconds (if 0, defaults to * a sensible value per the type of connection, * standard vs. pacemaker remote); * also propagated to the command XML * \param[in] call_options Call options to pass to server when sending * \param[in] expect_reply If true, wait for a reply from the server; * must be true for IPC (as opposed to TLS) clients * * \return pcmk_ok on success, -errno on error */ static int lrmd_send_command(lrmd_t *lrmd, const char *op, xmlNode *data, xmlNode **output_data, int timeout, enum lrmd_call_options options, bool expect_reply) { int rc = pcmk_ok; lrmd_private_t *native = lrmd->lrmd_private; xmlNode *op_msg = NULL; xmlNode *op_reply = NULL; if (!lrmd_api_is_connected(lrmd)) { return -ENOTCONN; } if (op == NULL) { crm_err("No operation specified"); return -EINVAL; } CRM_LOG_ASSERT(native->token != NULL); crm_trace("Sending %s op to executor", op); op_msg = lrmd_create_op(native->token, op, data, timeout, options); if (op_msg == NULL) { return -EINVAL; } if (expect_reply) { rc = lrmd_send_xml(lrmd, op_msg, timeout, &op_reply); } else { rc = lrmd_send_xml_no_reply(lrmd, op_msg); goto done; } if (rc < 0) { - crm_perror(LOG_ERR, "Couldn't perform %s operation (timeout=%d): %d", op, timeout, rc); + crm_err("Couldn't perform %s operation (timeout=%d): %d", op, timeout, + pcmk_strerror(rc)); goto done; } else if (op_reply == NULL) { rc = -ENOMSG; goto done; } rc = pcmk_ok; crm_trace("%s op reply received", op); if (pcmk__xe_get_int(op_reply, PCMK__XA_LRMD_RC, &rc) != pcmk_rc_ok) { rc = -ENOMSG; goto done; } crm_log_xml_trace(op_reply, "Reply"); if (output_data) { *output_data = op_reply; op_reply = NULL; /* Prevent subsequent free */ } done: if (lrmd_api_is_connected(lrmd) == FALSE) { crm_err("Executor disconnected"); } pcmk__xml_free(op_msg); pcmk__xml_free(op_reply); return rc; } static int lrmd_api_poke_connection(lrmd_t * lrmd) { int rc; lrmd_private_t *native = lrmd->lrmd_private; xmlNode *data = pcmk__xe_create(NULL, PCMK__XE_LRMD_RSC); crm_xml_add(data, PCMK__XA_LRMD_ORIGIN, __func__); rc = lrmd_send_command(lrmd, LRMD_OP_POKE, data, NULL, 0, 0, (native->type == pcmk__client_ipc)); pcmk__xml_free(data); return rc < 0 ? rc : pcmk_ok; } // \return Standard Pacemaker return code int lrmd__validate_remote_settings(lrmd_t *lrmd, GHashTable *hash) { int rc = pcmk_rc_ok; const char *value; lrmd_private_t *native = lrmd->lrmd_private; xmlNode *data = pcmk__xe_create(NULL, PCMK__XA_LRMD_OP); crm_xml_add(data, PCMK__XA_LRMD_ORIGIN, __func__); value = g_hash_table_lookup(hash, PCMK_OPT_STONITH_WATCHDOG_TIMEOUT); if ((value) && (stonith__watchdog_fencing_enabled_for_node(native->remote_nodename))) { crm_xml_add(data, PCMK__XA_LRMD_WATCHDOG, value); } rc = lrmd_send_command(lrmd, LRMD_OP_CHECK, data, NULL, 0, 0, (native->type == pcmk__client_ipc)); pcmk__xml_free(data); return (rc < 0)? pcmk_legacy2rc(rc) : pcmk_rc_ok; } static xmlNode * lrmd_handshake_hello_msg(const char *name, bool is_proxy) { xmlNode *hello = pcmk__xe_create(NULL, PCMK__XE_LRMD_COMMAND); crm_xml_add(hello, PCMK__XA_T, PCMK__VALUE_LRMD); crm_xml_add(hello, PCMK__XA_LRMD_OP, CRM_OP_REGISTER); crm_xml_add(hello, PCMK__XA_LRMD_CLIENTNAME, name); crm_xml_add(hello, PCMK__XA_LRMD_PROTOCOL_VERSION, LRMD_PROTOCOL_VERSION); /* advertise that we are a proxy provider */ if (is_proxy) { pcmk__xe_set_bool_attr(hello, PCMK__XA_LRMD_IS_IPC_PROVIDER, true); } return hello; } static int process_lrmd_handshake_reply(xmlNode *reply, lrmd_private_t *native) { int rc = pcmk_rc_ok; const char *version = pcmk__xe_get(reply, PCMK__XA_LRMD_PROTOCOL_VERSION); const char *msg_type = pcmk__xe_get(reply, PCMK__XA_LRMD_OP); const char *tmp_ticket = pcmk__xe_get(reply, PCMK__XA_LRMD_CLIENTID); const char *start_state = pcmk__xe_get(reply, PCMK__XA_NODE_START_STATE); pcmk__xe_get_int(reply, PCMK__XA_LRMD_RC, &rc); rc = pcmk_legacy2rc(rc); /* The remote executor may add its uptime to the XML reply, which is useful * in handling transient attributes when the connection to the remote node * unexpectedly drops. If no parameter is given, just default to -1. */ native->remote->uptime = -1; pcmk__xe_get_time(reply, PCMK__XA_UPTIME, &native->remote->uptime); if (start_state) { native->remote->start_state = strdup(start_state); } if (rc == EPROTO) { crm_err("Executor protocol version mismatch between client (%s) and server (%s)", LRMD_PROTOCOL_VERSION, version); crm_log_xml_err(reply, "Protocol Error"); } else if (!pcmk__str_eq(msg_type, CRM_OP_REGISTER, pcmk__str_casei)) { crm_err("Invalid registration message: %s", msg_type); crm_log_xml_err(reply, "Bad reply"); rc = EPROTO; } else if (tmp_ticket == NULL) { crm_err("No registration token provided"); crm_log_xml_err(reply, "Bad reply"); rc = EPROTO; } else { crm_trace("Obtained registration token: %s", tmp_ticket); native->token = strdup(tmp_ticket); native->peer_version = strdup(version?version:"1.0"); /* Included since 1.1 */ rc = pcmk_rc_ok; } return rc; } static int lrmd_handshake(lrmd_t * lrmd, const char *name) { int rc = pcmk_rc_ok; lrmd_private_t *native = lrmd->lrmd_private; xmlNode *reply = NULL; xmlNode *hello = lrmd_handshake_hello_msg(name, native->proxy_callback != NULL); rc = lrmd_send_xml(lrmd, hello, -1, &reply); if (rc < 0) { - crm_perror(LOG_DEBUG, "Couldn't complete registration with the executor API: %d", rc); + crm_debug("Couldn't complete registration with the executor API: %s", + pcmk_strerror(rc)); rc = ECOMM; } else if (reply == NULL) { crm_err("Did not receive registration reply"); rc = EPROTO; } else { rc = process_lrmd_handshake_reply(reply, native); } pcmk__xml_free(reply); pcmk__xml_free(hello); if (rc != pcmk_rc_ok) { lrmd_api_disconnect(lrmd); } return rc; } static int lrmd_handshake_async(lrmd_t * lrmd, const char *name) { int rc = pcmk_rc_ok; lrmd_private_t *native = lrmd->lrmd_private; xmlNode *hello = lrmd_handshake_hello_msg(name, native->proxy_callback != NULL); rc = send_remote_message(lrmd, hello); if (rc == pcmk_rc_ok) { native->expected_late_replies++; } else { lrmd_api_disconnect(lrmd); } pcmk__xml_free(hello); return rc; } static int lrmd_ipc_connect(lrmd_t * lrmd, int *fd) { int rc = pcmk_ok; lrmd_private_t *native = lrmd->lrmd_private; struct ipc_client_callbacks lrmd_callbacks = { .dispatch = lrmd_ipc_dispatch, .destroy = lrmd_ipc_connection_destroy }; crm_info("Connecting to executor"); if (fd) { /* No mainloop */ native->ipc = crm_ipc_new(CRM_SYSTEM_LRMD, 0); if (native->ipc != NULL) { rc = pcmk__connect_generic_ipc(native->ipc); if (rc == pcmk_rc_ok) { rc = pcmk__ipc_fd(native->ipc, fd); } if (rc != pcmk_rc_ok) { crm_err("Connection to executor failed: %s", pcmk_rc_str(rc)); rc = -ENOTCONN; } } } else { native->source = mainloop_add_ipc_client(CRM_SYSTEM_LRMD, G_PRIORITY_HIGH, 0, lrmd, &lrmd_callbacks); native->ipc = mainloop_get_ipc_client(native->source); } if (native->ipc == NULL) { crm_debug("Could not connect to the executor API"); rc = -ENOTCONN; } return rc; } static void copy_gnutls_datum(gnutls_datum_t *dest, gnutls_datum_t *source) { pcmk__assert((dest != NULL) && (source != NULL) && (source->data != NULL)); dest->data = gnutls_malloc(source->size); pcmk__mem_assert(dest->data); memcpy(dest->data, source->data, source->size); dest->size = source->size; } static void clear_gnutls_datum(gnutls_datum_t *datum) { gnutls_free(datum->data); datum->data = NULL; datum->size = 0; } #define KEY_READ_LEN 256 // Chunk size for reading key from file // \return Standard Pacemaker return code static int read_gnutls_key(const char *location, gnutls_datum_t *key) { FILE *stream = NULL; size_t buf_len = KEY_READ_LEN; if ((location == NULL) || (key == NULL)) { return EINVAL; } stream = fopen(location, "r"); if (stream == NULL) { return errno; } key->data = gnutls_malloc(buf_len); key->size = 0; while (!feof(stream)) { int next = fgetc(stream); if (next == EOF) { if (!feof(stream)) { crm_warn("Pacemaker Remote key read was partially successful " "(copy in memory may be corrupted)"); } break; } if (key->size == buf_len) { buf_len = key->size + KEY_READ_LEN; key->data = gnutls_realloc(key->data, buf_len); pcmk__assert(key->data); } key->data[key->size++] = (unsigned char) next; } fclose(stream); if (key->size == 0) { clear_gnutls_datum(key); return ENOKEY; } return pcmk_rc_ok; } // Cache the most recently used Pacemaker Remote authentication key struct key_cache_s { time_t updated; // When cached key was read (valid for 1 minute) const char *location; // Where cached key was read from gnutls_datum_t key; // Cached key }; static bool key_is_cached(struct key_cache_s *key_cache) { return key_cache->updated != 0; } static bool key_cache_expired(struct key_cache_s *key_cache) { return (time(NULL) - key_cache->updated) >= 60; } static void clear_key_cache(struct key_cache_s *key_cache) { clear_gnutls_datum(&(key_cache->key)); if ((key_cache->updated != 0) || (key_cache->location != NULL)) { key_cache->updated = 0; key_cache->location = NULL; crm_debug("Cleared Pacemaker Remote key cache"); } } static void get_cached_key(struct key_cache_s *key_cache, gnutls_datum_t *key) { copy_gnutls_datum(key, &(key_cache->key)); crm_debug("Using cached Pacemaker Remote key from %s", pcmk__s(key_cache->location, "unknown location")); } static void cache_key(struct key_cache_s *key_cache, gnutls_datum_t *key, const char *location) { key_cache->updated = time(NULL); key_cache->location = location; copy_gnutls_datum(&(key_cache->key), key); crm_debug("Using (and cacheing) Pacemaker Remote key from %s", pcmk__s(location, "unknown location")); } /*! * \internal * \brief Get Pacemaker Remote authentication key from file or cache * * \param[in] location Path to key file to try (this memory must * persist across all calls of this function) * \param[out] key Key from location or cache * * \return Standard Pacemaker return code */ static int get_remote_key(const char *location, gnutls_datum_t *key) { static struct key_cache_s key_cache = { 0, }; int rc = pcmk_rc_ok; if ((location == NULL) || (key == NULL)) { return EINVAL; } if (key_is_cached(&key_cache)) { if (key_cache_expired(&key_cache)) { clear_key_cache(&key_cache); } else { get_cached_key(&key_cache, key); return pcmk_rc_ok; } } rc = read_gnutls_key(location, key); if (rc != pcmk_rc_ok) { return rc; } cache_key(&key_cache, key, location); return pcmk_rc_ok; } /*! * \internal * \brief Initialize the Pacemaker Remote authentication key * * Try loading the Pacemaker Remote authentication key from cache if available, * otherwise from these locations, in order of preference: * * - The value of the PCMK_authkey_location environment variable, if set * - The Pacemaker default key file location * * \param[out] key Where to store key * * \return Standard Pacemaker return code */ int lrmd__init_remote_key(gnutls_datum_t *key) { static const char *env_location = NULL; static bool need_env = true; int rc = pcmk_rc_ok; if (need_env) { env_location = pcmk__env_option(PCMK__ENV_AUTHKEY_LOCATION); need_env = false; } // Try location in environment variable, if set if (env_location != NULL) { rc = get_remote_key(env_location, key); if (rc == pcmk_rc_ok) { return pcmk_rc_ok; } crm_warn("Could not read Pacemaker Remote key from %s: %s", env_location, pcmk_rc_str(rc)); return ENOKEY; } // Try default location, if environment wasn't explicitly set to it rc = get_remote_key(DEFAULT_REMOTE_KEY_LOCATION, key); if (rc == pcmk_rc_ok) { return pcmk_rc_ok; } crm_warn("Could not read Pacemaker Remote key from default location %s: %s", DEFAULT_REMOTE_KEY_LOCATION, pcmk_rc_str(rc)); return ENOKEY; } static void report_async_connection_result(lrmd_t * lrmd, int rc) { lrmd_private_t *native = lrmd->lrmd_private; if (native->callback) { lrmd_event_data_t event = { 0, }; event.type = lrmd_event_connect; event.remote_nodename = native->remote_nodename; event.connection_rc = rc; native->callback(&event); } } static void tls_handshake_failed(lrmd_t *lrmd, int tls_rc, int rc) { lrmd_private_t *native = lrmd->lrmd_private; crm_warn("Disconnecting after TLS handshake with " "Pacemaker Remote server %s:%d failed: %s", native->server, native->port, (rc == EPROTO)? gnutls_strerror(tls_rc) : pcmk_rc_str(rc)); report_async_connection_result(lrmd, pcmk_rc2legacy(rc)); gnutls_deinit(native->remote->tls_session); native->remote->tls_session = NULL; lrmd_tls_connection_destroy(lrmd); } static void tls_handshake_succeeded(lrmd_t *lrmd) { int rc = pcmk_rc_ok; lrmd_private_t *native = lrmd->lrmd_private; /* Now that the handshake is done, see if any client TLS certificate is * close to its expiration date and log if so. If a TLS certificate is not * in use, this function will just return so we don't need to check for the * session type here. */ pcmk__tls_check_cert_expiration(native->remote->tls_session); crm_info("TLS connection to Pacemaker Remote server %s:%d succeeded", native->server, native->port); rc = add_tls_to_mainloop(lrmd, true); /* If add_tls_to_mainloop failed, report that right now. Otherwise, we have * to wait until we read the async reply to report anything. */ if (rc != pcmk_rc_ok) { report_async_connection_result(lrmd, pcmk_rc2legacy(rc)); } } /*! * \internal * \brief Perform a TLS client handshake with a Pacemaker Remote server * * \param[in] lrmd Newly established Pacemaker Remote executor connection * * \return Standard Pacemaker return code */ static int tls_client_handshake(lrmd_t *lrmd) { lrmd_private_t *native = lrmd->lrmd_private; int tls_rc = GNUTLS_E_SUCCESS; int rc = pcmk__tls_client_handshake(native->remote, TLS_HANDSHAKE_TIMEOUT, &tls_rc); if (rc != pcmk_rc_ok) { tls_handshake_failed(lrmd, tls_rc, rc); } return rc; } /*! * \internal * \brief Add trigger and file descriptor mainloop sources for TLS * * \param[in,out] lrmd API connection with established TLS session * \param[in] do_api_handshake Whether to perform executor handshake * * \return Standard Pacemaker return code */ static int add_tls_to_mainloop(lrmd_t *lrmd, bool do_api_handshake) { lrmd_private_t *native = lrmd->lrmd_private; int rc = pcmk_rc_ok; char *name = crm_strdup_printf("pacemaker-remote-%s:%d", native->server, native->port); struct mainloop_fd_callbacks tls_fd_callbacks = { .dispatch = lrmd_tls_dispatch, .destroy = lrmd_tls_connection_destroy, }; native->process_notify = mainloop_add_trigger(G_PRIORITY_HIGH, process_pending_notifies, lrmd); native->source = mainloop_add_fd(name, G_PRIORITY_HIGH, native->sock, lrmd, &tls_fd_callbacks); /* Async connections lose the client name provided by the API caller, so we * have to use our generated name here to perform the executor handshake. * * @TODO Keep track of the caller-provided name. Perhaps we should be using * that name in this function instead of generating one anyway. */ if (do_api_handshake) { rc = lrmd_handshake_async(lrmd, name); } free(name); return rc; } struct handshake_data_s { lrmd_t *lrmd; time_t start_time; int timeout_sec; }; static gboolean try_handshake_cb(gpointer user_data) { struct handshake_data_s *hs = user_data; lrmd_t *lrmd = hs->lrmd; lrmd_private_t *native = lrmd->lrmd_private; pcmk__remote_t *remote = native->remote; int rc = pcmk_rc_ok; int tls_rc = GNUTLS_E_SUCCESS; if (time(NULL) >= hs->start_time + hs->timeout_sec) { rc = ETIME; tls_handshake_failed(lrmd, GNUTLS_E_TIMEDOUT, rc); free(hs); return 0; } rc = pcmk__tls_client_try_handshake(remote, &tls_rc); if (rc == pcmk_rc_ok) { tls_handshake_succeeded(lrmd); free(hs); return 0; } else if (rc == EAGAIN) { mainloop_set_trigger(native->handshake_trigger); return 1; } else { rc = EKEYREJECTED; tls_handshake_failed(lrmd, tls_rc, rc); free(hs); return 0; } } static void lrmd_tcp_connect_cb(void *userdata, int rc, int sock) { lrmd_t *lrmd = userdata; lrmd_private_t *native = lrmd->lrmd_private; int tls_rc = GNUTLS_E_SUCCESS; bool use_cert = pcmk__x509_enabled(); native->async_timer = 0; if (rc != pcmk_rc_ok) { lrmd_tls_connection_destroy(lrmd); crm_info("Could not connect to Pacemaker Remote at %s:%d: %s " QB_XS " rc=%d", native->server, native->port, pcmk_rc_str(rc), rc); report_async_connection_result(lrmd, pcmk_rc2legacy(rc)); return; } /* The TCP connection was successful, so establish the TLS connection. */ native->sock = sock; if (native->tls == NULL) { rc = pcmk__init_tls(&native->tls, false, use_cert ? GNUTLS_CRD_CERTIFICATE : GNUTLS_CRD_PSK); if ((rc != pcmk_rc_ok) || (native->tls == NULL)) { lrmd_tls_connection_destroy(lrmd); report_async_connection_result(lrmd, pcmk_rc2legacy(rc)); return; } } if (!use_cert) { gnutls_datum_t psk_key = { NULL, 0 }; rc = lrmd__init_remote_key(&psk_key); if (rc != pcmk_rc_ok) { crm_info("Could not connect to Pacemaker Remote at %s:%d: %s " QB_XS " rc=%d", native->server, native->port, pcmk_rc_str(rc), rc); lrmd_tls_connection_destroy(lrmd); report_async_connection_result(lrmd, pcmk_rc2legacy(rc)); return; } pcmk__tls_add_psk_key(native->tls, &psk_key); gnutls_free(psk_key.data); } native->remote->tls_session = pcmk__new_tls_session(native->tls, sock); if (native->remote->tls_session == NULL) { lrmd_tls_connection_destroy(lrmd); report_async_connection_result(lrmd, -EPROTO); return; } /* If the TLS handshake immediately succeeds or fails, we can handle that * now without having to deal with mainloops and retries. Otherwise, add a * trigger to keep trying until we get a result (or it times out). */ rc = pcmk__tls_client_try_handshake(native->remote, &tls_rc); if (rc == EAGAIN) { struct handshake_data_s *hs = NULL; if (native->handshake_trigger != NULL) { return; } hs = pcmk__assert_alloc(1, sizeof(struct handshake_data_s)); hs->lrmd = lrmd; hs->start_time = time(NULL); hs->timeout_sec = TLS_HANDSHAKE_TIMEOUT; native->handshake_trigger = mainloop_add_trigger(G_PRIORITY_LOW, try_handshake_cb, hs); mainloop_set_trigger(native->handshake_trigger); } else if (rc == pcmk_rc_ok) { tls_handshake_succeeded(lrmd); } else { tls_handshake_failed(lrmd, tls_rc, rc); } } static int lrmd_tls_connect_async(lrmd_t * lrmd, int timeout /*ms */ ) { int rc = pcmk_rc_ok; int timer_id = 0; lrmd_private_t *native = lrmd->lrmd_private; native->sock = -1; rc = pcmk__connect_remote(native->server, native->port, timeout, &timer_id, &(native->sock), lrmd, lrmd_tcp_connect_cb); if (rc != pcmk_rc_ok) { crm_warn("Pacemaker Remote connection to %s:%d failed: %s " QB_XS " rc=%d", native->server, native->port, pcmk_rc_str(rc), rc); return rc; } native->async_timer = timer_id; return rc; } static int lrmd_tls_connect(lrmd_t * lrmd, int *fd) { int rc = pcmk_rc_ok; bool use_cert = pcmk__x509_enabled(); lrmd_private_t *native = lrmd->lrmd_private; native->sock = -1; rc = pcmk__connect_remote(native->server, native->port, 0, NULL, &(native->sock), NULL, NULL); if (rc != pcmk_rc_ok) { crm_warn("Pacemaker Remote connection to %s:%d failed: %s " QB_XS " rc=%d", native->server, native->port, pcmk_rc_str(rc), rc); lrmd_tls_connection_destroy(lrmd); return ENOTCONN; } if (native->tls == NULL) { rc = pcmk__init_tls(&native->tls, false, use_cert ? GNUTLS_CRD_CERTIFICATE : GNUTLS_CRD_PSK); if (rc != pcmk_rc_ok) { lrmd_tls_connection_destroy(lrmd); return rc; } } if (!use_cert) { gnutls_datum_t psk_key = { NULL, 0 }; rc = lrmd__init_remote_key(&psk_key); if (rc != pcmk_rc_ok) { lrmd_tls_connection_destroy(lrmd); return rc; } pcmk__tls_add_psk_key(native->tls, &psk_key); gnutls_free(psk_key.data); } native->remote->tls_session = pcmk__new_tls_session(native->tls, native->sock); if (native->remote->tls_session == NULL) { lrmd_tls_connection_destroy(lrmd); return EPROTO; } if (tls_client_handshake(lrmd) != pcmk_rc_ok) { return EKEYREJECTED; } crm_info("Client TLS connection established with Pacemaker Remote server %s:%d", native->server, native->port); if (fd) { *fd = native->sock; } else { rc = add_tls_to_mainloop(lrmd, false); } return rc; } static int lrmd_api_connect(lrmd_t * lrmd, const char *name, int *fd) { int rc = -ENOTCONN; lrmd_private_t *native = lrmd->lrmd_private; switch (native->type) { case pcmk__client_ipc: rc = lrmd_ipc_connect(lrmd, fd); break; case pcmk__client_tls: rc = lrmd_tls_connect(lrmd, fd); rc = pcmk_rc2legacy(rc); break; default: crm_err("Unsupported executor connection type (bug?): %d", native->type); rc = -EPROTONOSUPPORT; } if (rc == pcmk_ok) { rc = lrmd_handshake(lrmd, name); rc = pcmk_rc2legacy(rc); } return rc; } static int lrmd_api_connect_async(lrmd_t * lrmd, const char *name, int timeout) { int rc = pcmk_ok; lrmd_private_t *native = lrmd->lrmd_private; CRM_CHECK(native && native->callback, return -EINVAL); switch (native->type) { case pcmk__client_ipc: /* fake async connection with ipc. it should be fast * enough that we gain very little from async */ rc = lrmd_api_connect(lrmd, name, NULL); if (!rc) { report_async_connection_result(lrmd, rc); } break; case pcmk__client_tls: rc = lrmd_tls_connect_async(lrmd, timeout); rc = pcmk_rc2legacy(rc); break; default: crm_err("Unsupported executor connection type (bug?): %d", native->type); rc = -EPROTONOSUPPORT; } return rc; } static void lrmd_ipc_disconnect(lrmd_t * lrmd) { lrmd_private_t *native = lrmd->lrmd_private; if (native->source != NULL) { /* Attached to mainloop */ mainloop_del_ipc_client(native->source); native->source = NULL; native->ipc = NULL; } else if (native->ipc) { /* Not attached to mainloop */ crm_ipc_t *ipc = native->ipc; native->ipc = NULL; crm_ipc_close(ipc); crm_ipc_destroy(ipc); } } static void lrmd_tls_disconnect(lrmd_t * lrmd) { lrmd_private_t *native = lrmd->lrmd_private; if (native->remote->tls_session) { gnutls_bye(native->remote->tls_session, GNUTLS_SHUT_RDWR); gnutls_deinit(native->remote->tls_session); native->remote->tls_session = NULL; } if (native->async_timer) { g_source_remove(native->async_timer); native->async_timer = 0; } if (native->source != NULL) { /* Attached to mainloop */ mainloop_del_ipc_client(native->source); native->source = NULL; } else if (native->sock >= 0) { close(native->sock); native->sock = -1; } if (native->pending_notify) { g_list_free_full(native->pending_notify, lrmd_free_xml); native->pending_notify = NULL; } } static int lrmd_api_disconnect(lrmd_t * lrmd) { lrmd_private_t *native = lrmd->lrmd_private; int rc = pcmk_ok; switch (native->type) { case pcmk__client_ipc: crm_debug("Disconnecting from local executor"); lrmd_ipc_disconnect(lrmd); break; case pcmk__client_tls: crm_debug("Disconnecting from remote executor on %s", native->remote_nodename); lrmd_tls_disconnect(lrmd); break; default: crm_err("Unsupported executor connection type (bug?): %d", native->type); rc = -EPROTONOSUPPORT; } free(native->token); native->token = NULL; free(native->peer_version); native->peer_version = NULL; return rc; } static int lrmd_api_register_rsc(lrmd_t * lrmd, const char *rsc_id, const char *class, const char *provider, const char *type, enum lrmd_call_options options) { int rc = pcmk_ok; xmlNode *data = NULL; if (!class || !type || !rsc_id) { return -EINVAL; } if (pcmk_is_set(pcmk_get_ra_caps(class), pcmk_ra_cap_provider) && (provider == NULL)) { return -EINVAL; } data = pcmk__xe_create(NULL, PCMK__XE_LRMD_RSC); crm_xml_add(data, PCMK__XA_LRMD_ORIGIN, __func__); crm_xml_add(data, PCMK__XA_LRMD_RSC_ID, rsc_id); crm_xml_add(data, PCMK__XA_LRMD_CLASS, class); crm_xml_add(data, PCMK__XA_LRMD_PROVIDER, provider); crm_xml_add(data, PCMK__XA_LRMD_TYPE, type); rc = lrmd_send_command(lrmd, LRMD_OP_RSC_REG, data, NULL, 0, options, true); pcmk__xml_free(data); return rc; } static int lrmd_api_unregister_rsc(lrmd_t * lrmd, const char *rsc_id, enum lrmd_call_options options) { int rc = pcmk_ok; xmlNode *data = pcmk__xe_create(NULL, PCMK__XE_LRMD_RSC); crm_xml_add(data, PCMK__XA_LRMD_ORIGIN, __func__); crm_xml_add(data, PCMK__XA_LRMD_RSC_ID, rsc_id); rc = lrmd_send_command(lrmd, LRMD_OP_RSC_UNREG, data, NULL, 0, options, true); pcmk__xml_free(data); return rc; } lrmd_rsc_info_t * lrmd_new_rsc_info(const char *rsc_id, const char *standard, const char *provider, const char *type) { lrmd_rsc_info_t *rsc_info = pcmk__assert_alloc(1, sizeof(lrmd_rsc_info_t)); rsc_info->id = pcmk__str_copy(rsc_id); rsc_info->standard = pcmk__str_copy(standard); rsc_info->provider = pcmk__str_copy(provider); rsc_info->type = pcmk__str_copy(type); return rsc_info; } lrmd_rsc_info_t * lrmd_copy_rsc_info(lrmd_rsc_info_t * rsc_info) { return lrmd_new_rsc_info(rsc_info->id, rsc_info->standard, rsc_info->provider, rsc_info->type); } void lrmd_free_rsc_info(lrmd_rsc_info_t * rsc_info) { if (!rsc_info) { return; } free(rsc_info->id); free(rsc_info->type); free(rsc_info->standard); free(rsc_info->provider); free(rsc_info); } static lrmd_rsc_info_t * lrmd_api_get_rsc_info(lrmd_t * lrmd, const char *rsc_id, enum lrmd_call_options options) { lrmd_rsc_info_t *rsc_info = NULL; xmlNode *data = pcmk__xe_create(NULL, PCMK__XE_LRMD_RSC); xmlNode *output = NULL; const char *class = NULL; const char *provider = NULL; const char *type = NULL; crm_xml_add(data, PCMK__XA_LRMD_ORIGIN, __func__); crm_xml_add(data, PCMK__XA_LRMD_RSC_ID, rsc_id); lrmd_send_command(lrmd, LRMD_OP_RSC_INFO, data, &output, 0, options, true); pcmk__xml_free(data); if (!output) { return NULL; } class = pcmk__xe_get(output, PCMK__XA_LRMD_CLASS); provider = pcmk__xe_get(output, PCMK__XA_LRMD_PROVIDER); type = pcmk__xe_get(output, PCMK__XA_LRMD_TYPE); if (!class || !type) { pcmk__xml_free(output); return NULL; } else if (pcmk_is_set(pcmk_get_ra_caps(class), pcmk_ra_cap_provider) && !provider) { pcmk__xml_free(output); return NULL; } rsc_info = lrmd_new_rsc_info(rsc_id, class, provider, type); pcmk__xml_free(output); return rsc_info; } void lrmd_free_op_info(lrmd_op_info_t *op_info) { if (op_info) { free(op_info->rsc_id); free(op_info->action); free(op_info->interval_ms_s); free(op_info->timeout_ms_s); free(op_info); } } static int lrmd_api_get_recurring_ops(lrmd_t *lrmd, const char *rsc_id, int timeout_ms, enum lrmd_call_options options, GList **output) { xmlNode *data = NULL; xmlNode *output_xml = NULL; int rc = pcmk_ok; if (output == NULL) { return -EINVAL; } *output = NULL; // Send request if (rsc_id) { data = pcmk__xe_create(NULL, PCMK__XE_LRMD_RSC); crm_xml_add(data, PCMK__XA_LRMD_ORIGIN, __func__); crm_xml_add(data, PCMK__XA_LRMD_RSC_ID, rsc_id); } rc = lrmd_send_command(lrmd, LRMD_OP_GET_RECURRING, data, &output_xml, timeout_ms, options, true); if (data) { pcmk__xml_free(data); } // Process reply if ((rc != pcmk_ok) || (output_xml == NULL)) { return rc; } for (const xmlNode *rsc_xml = pcmk__xe_first_child(output_xml, PCMK__XE_LRMD_RSC, NULL, NULL); (rsc_xml != NULL) && (rc == pcmk_ok); rsc_xml = pcmk__xe_next(rsc_xml, PCMK__XE_LRMD_RSC)) { rsc_id = pcmk__xe_get(rsc_xml, PCMK__XA_LRMD_RSC_ID); if (rsc_id == NULL) { crm_err("Could not parse recurring operation information from executor"); continue; } for (const xmlNode *op_xml = pcmk__xe_first_child(rsc_xml, PCMK__XE_LRMD_RSC_OP, NULL, NULL); op_xml != NULL; op_xml = pcmk__xe_next(op_xml, PCMK__XE_LRMD_RSC_OP)) { lrmd_op_info_t *op_info = calloc(1, sizeof(lrmd_op_info_t)); if (op_info == NULL) { rc = -ENOMEM; break; } op_info->rsc_id = strdup(rsc_id); op_info->action = pcmk__xe_get_copy(op_xml, PCMK__XA_LRMD_RSC_ACTION); op_info->interval_ms_s = pcmk__xe_get_copy(op_xml, PCMK__XA_LRMD_RSC_INTERVAL); op_info->timeout_ms_s = pcmk__xe_get_copy(op_xml, PCMK__XA_LRMD_TIMEOUT); *output = g_list_prepend(*output, op_info); } } pcmk__xml_free(output_xml); return rc; } static void lrmd_api_set_callback(lrmd_t * lrmd, lrmd_event_callback callback) { lrmd_private_t *native = lrmd->lrmd_private; native->callback = callback; } void lrmd_internal_set_proxy_callback(lrmd_t * lrmd, void *userdata, void (*callback)(lrmd_t *lrmd, void *userdata, xmlNode *msg)) { lrmd_private_t *native = lrmd->lrmd_private; native->proxy_callback = callback; native->proxy_callback_userdata = userdata; } void lrmd_internal_proxy_dispatch(lrmd_t *lrmd, xmlNode *msg) { lrmd_private_t *native = lrmd->lrmd_private; if (native->proxy_callback) { crm_log_xml_trace(msg, "PROXY_INBOUND"); native->proxy_callback(lrmd, native->proxy_callback_userdata, msg); } } int lrmd_internal_proxy_send(lrmd_t * lrmd, xmlNode *msg) { if (lrmd == NULL) { return -ENOTCONN; } crm_xml_add(msg, PCMK__XA_LRMD_OP, CRM_OP_IPC_FWD); crm_log_xml_trace(msg, "PROXY_OUTBOUND"); return lrmd_send_xml_no_reply(lrmd, msg); } static int stonith_get_metadata(const char *type, char **output) { int rc = pcmk_ok; stonith_t *stonith_api = stonith__api_new(); if (stonith_api == NULL) { crm_err("Could not get fence agent meta-data: API memory allocation failed"); return -ENOMEM; } rc = stonith_api->cmds->metadata(stonith_api, st_opt_sync_call, type, NULL, output, 0); if ((rc == pcmk_ok) && (*output == NULL)) { rc = -EIO; } stonith_api->cmds->free(stonith_api); return rc; } static int lrmd_api_get_metadata(lrmd_t *lrmd, const char *standard, const char *provider, const char *type, char **output, enum lrmd_call_options options) { return lrmd->cmds->get_metadata_params(lrmd, standard, provider, type, output, options, NULL); } static int lrmd_api_get_metadata_params(lrmd_t *lrmd, const char *standard, const char *provider, const char *type, char **output, enum lrmd_call_options options, lrmd_key_value_t *params) { svc_action_t *action = NULL; GHashTable *params_table = NULL; if (!standard || !type) { lrmd_key_value_freeall(params); return -EINVAL; } if (pcmk__str_eq(standard, PCMK_RESOURCE_CLASS_STONITH, pcmk__str_casei)) { lrmd_key_value_freeall(params); // stonith-class resources don't support a provider return stonith_get_metadata(type, output); } params_table = pcmk__strkey_table(free, free); for (const lrmd_key_value_t *param = params; param; param = param->next) { pcmk__insert_dup(params_table, param->key, param->value); } action = services__create_resource_action(type, standard, provider, type, PCMK_ACTION_META_DATA, 0, PCMK_DEFAULT_ACTION_TIMEOUT_MS, params_table, 0); lrmd_key_value_freeall(params); if (action == NULL) { return -ENOMEM; } if (action->rc != PCMK_OCF_UNKNOWN) { services_action_free(action); return -EINVAL; } if (!services_action_sync(action)) { crm_err("Failed to retrieve meta-data for %s:%s:%s", standard, provider, type); services_action_free(action); return -EIO; } if (!action->stdout_data) { crm_err("Failed to receive meta-data for %s:%s:%s", standard, provider, type); services_action_free(action); return -EIO; } *output = strdup(action->stdout_data); services_action_free(action); return pcmk_ok; } static int lrmd_api_exec(lrmd_t *lrmd, const char *rsc_id, const char *action, const char *userdata, guint interval_ms, int timeout, /* ms */ int start_delay, /* ms */ enum lrmd_call_options options, lrmd_key_value_t * params) { int rc = pcmk_ok; xmlNode *data = pcmk__xe_create(NULL, PCMK__XE_LRMD_RSC); xmlNode *args = pcmk__xe_create(data, PCMK__XE_ATTRIBUTES); lrmd_key_value_t *tmp = NULL; crm_xml_add(data, PCMK__XA_LRMD_ORIGIN, __func__); crm_xml_add(data, PCMK__XA_LRMD_RSC_ID, rsc_id); crm_xml_add(data, PCMK__XA_LRMD_RSC_ACTION, action); crm_xml_add(data, PCMK__XA_LRMD_RSC_USERDATA_STR, userdata); crm_xml_add_ms(data, PCMK__XA_LRMD_RSC_INTERVAL, interval_ms); crm_xml_add_int(data, PCMK__XA_LRMD_TIMEOUT, timeout); crm_xml_add_int(data, PCMK__XA_LRMD_RSC_START_DELAY, start_delay); for (tmp = params; tmp; tmp = tmp->next) { hash2smartfield((gpointer) tmp->key, (gpointer) tmp->value, args); } rc = lrmd_send_command(lrmd, LRMD_OP_RSC_EXEC, data, NULL, timeout, options, true); pcmk__xml_free(data); lrmd_key_value_freeall(params); return rc; } /* timeout is in ms */ static int lrmd_api_exec_alert(lrmd_t *lrmd, const char *alert_id, const char *alert_path, int timeout, lrmd_key_value_t *params) { int rc = pcmk_ok; xmlNode *data = pcmk__xe_create(NULL, PCMK__XE_LRMD_ALERT); xmlNode *args = pcmk__xe_create(data, PCMK__XE_ATTRIBUTES); lrmd_key_value_t *tmp = NULL; crm_xml_add(data, PCMK__XA_LRMD_ORIGIN, __func__); crm_xml_add(data, PCMK__XA_LRMD_ALERT_ID, alert_id); crm_xml_add(data, PCMK__XA_LRMD_ALERT_PATH, alert_path); crm_xml_add_int(data, PCMK__XA_LRMD_TIMEOUT, timeout); for (tmp = params; tmp; tmp = tmp->next) { hash2smartfield((gpointer) tmp->key, (gpointer) tmp->value, args); } rc = lrmd_send_command(lrmd, LRMD_OP_ALERT_EXEC, data, NULL, timeout, lrmd_opt_notify_orig_only, true); pcmk__xml_free(data); lrmd_key_value_freeall(params); return rc; } static int lrmd_api_cancel(lrmd_t *lrmd, const char *rsc_id, const char *action, guint interval_ms) { int rc = pcmk_ok; xmlNode *data = pcmk__xe_create(NULL, PCMK__XE_LRMD_RSC); crm_xml_add(data, PCMK__XA_LRMD_ORIGIN, __func__); crm_xml_add(data, PCMK__XA_LRMD_RSC_ACTION, action); crm_xml_add(data, PCMK__XA_LRMD_RSC_ID, rsc_id); crm_xml_add_ms(data, PCMK__XA_LRMD_RSC_INTERVAL, interval_ms); rc = lrmd_send_command(lrmd, LRMD_OP_RSC_CANCEL, data, NULL, 0, 0, true); pcmk__xml_free(data); return rc; } static int list_stonith_agents(lrmd_list_t ** resources) { int rc = 0; stonith_t *stonith_api = stonith__api_new(); stonith_key_value_t *stonith_resources = NULL; stonith_key_value_t *dIter = NULL; if (stonith_api == NULL) { crm_err("Could not list fence agents: API memory allocation failed"); return -ENOMEM; } stonith_api->cmds->list_agents(stonith_api, st_opt_sync_call, NULL, &stonith_resources, 0); stonith_api->cmds->free(stonith_api); for (dIter = stonith_resources; dIter; dIter = dIter->next) { rc++; if (resources) { *resources = lrmd_list_add(*resources, dIter->value); } } stonith__key_value_freeall(stonith_resources, true, false); return rc; } static int lrmd_api_list_agents(lrmd_t * lrmd, lrmd_list_t ** resources, const char *class, const char *provider) { int rc = 0; int stonith_count = 0; // Initially, whether to include stonith devices if (pcmk__str_eq(class, PCMK_RESOURCE_CLASS_STONITH, pcmk__str_casei)) { stonith_count = 1; } else { GList *gIter = NULL; GList *agents = resources_list_agents(class, provider); for (gIter = agents; gIter != NULL; gIter = gIter->next) { *resources = lrmd_list_add(*resources, (const char *)gIter->data); rc++; } g_list_free_full(agents, free); if (!class) { stonith_count = 1; } } if (stonith_count) { // Now, if stonith devices are included, how many there are stonith_count = list_stonith_agents(resources); if (stonith_count > 0) { rc += stonith_count; } } if (rc == 0) { crm_notice("No agents found for class %s", class); rc = -EPROTONOSUPPORT; } return rc; } static bool does_provider_have_agent(const char *agent, const char *provider, const char *class) { bool found = false; GList *agents = NULL; GList *gIter2 = NULL; agents = resources_list_agents(class, provider); for (gIter2 = agents; gIter2 != NULL; gIter2 = gIter2->next) { if (pcmk__str_eq(agent, gIter2->data, pcmk__str_casei)) { found = true; } } g_list_free_full(agents, free); return found; } static int lrmd_api_list_ocf_providers(lrmd_t * lrmd, const char *agent, lrmd_list_t ** providers) { int rc = pcmk_ok; char *provider = NULL; GList *ocf_providers = NULL; GList *gIter = NULL; ocf_providers = resources_list_providers(PCMK_RESOURCE_CLASS_OCF); for (gIter = ocf_providers; gIter != NULL; gIter = gIter->next) { provider = gIter->data; if (!agent || does_provider_have_agent(agent, provider, PCMK_RESOURCE_CLASS_OCF)) { *providers = lrmd_list_add(*providers, (const char *)gIter->data); rc++; } } g_list_free_full(ocf_providers, free); return rc; } static int lrmd_api_list_standards(lrmd_t * lrmd, lrmd_list_t ** supported) { int rc = 0; GList *standards = NULL; GList *gIter = NULL; standards = resources_list_standards(); for (gIter = standards; gIter != NULL; gIter = gIter->next) { *supported = lrmd_list_add(*supported, (const char *)gIter->data); rc++; } if (list_stonith_agents(NULL) > 0) { *supported = lrmd_list_add(*supported, PCMK_RESOURCE_CLASS_STONITH); rc++; } g_list_free_full(standards, free); return rc; } /*! * \internal * \brief Create an executor API object * * \param[out] api Will be set to newly created API object (it is the * caller's responsibility to free this value with * lrmd_api_delete() if this function succeeds) * \param[in] nodename If the object will be used for a remote connection, * the node name to use in cluster for remote executor * \param[in] server If the object will be used for a remote connection, * the resolvable host name to connect to * \param[in] port If the object will be used for a remote connection, * port number on \p server to connect to * * \return Standard Pacemaker return code * \note If the caller leaves one of \p nodename or \p server NULL, the other's * value will be used for both. If the caller leaves both NULL, an API * object will be created for a local executor connection. */ int lrmd__new(lrmd_t **api, const char *nodename, const char *server, int port) { lrmd_private_t *pvt = NULL; if (api == NULL) { return EINVAL; } *api = NULL; // Allocate all memory needed *api = calloc(1, sizeof(lrmd_t)); if (*api == NULL) { return ENOMEM; } pvt = calloc(1, sizeof(lrmd_private_t)); if (pvt == NULL) { lrmd_api_delete(*api); *api = NULL; return ENOMEM; } (*api)->lrmd_private = pvt; // @TODO Do we need to do this for local connections? pvt->remote = calloc(1, sizeof(pcmk__remote_t)); (*api)->cmds = calloc(1, sizeof(lrmd_api_operations_t)); if ((pvt->remote == NULL) || ((*api)->cmds == NULL)) { lrmd_api_delete(*api); *api = NULL; return ENOMEM; } // Set methods (*api)->cmds->connect = lrmd_api_connect; (*api)->cmds->connect_async = lrmd_api_connect_async; (*api)->cmds->is_connected = lrmd_api_is_connected; (*api)->cmds->poke_connection = lrmd_api_poke_connection; (*api)->cmds->disconnect = lrmd_api_disconnect; (*api)->cmds->register_rsc = lrmd_api_register_rsc; (*api)->cmds->unregister_rsc = lrmd_api_unregister_rsc; (*api)->cmds->get_rsc_info = lrmd_api_get_rsc_info; (*api)->cmds->get_recurring_ops = lrmd_api_get_recurring_ops; (*api)->cmds->set_callback = lrmd_api_set_callback; (*api)->cmds->get_metadata = lrmd_api_get_metadata; (*api)->cmds->exec = lrmd_api_exec; (*api)->cmds->cancel = lrmd_api_cancel; (*api)->cmds->list_agents = lrmd_api_list_agents; (*api)->cmds->list_ocf_providers = lrmd_api_list_ocf_providers; (*api)->cmds->list_standards = lrmd_api_list_standards; (*api)->cmds->exec_alert = lrmd_api_exec_alert; (*api)->cmds->get_metadata_params = lrmd_api_get_metadata_params; if ((nodename == NULL) && (server == NULL)) { pvt->type = pcmk__client_ipc; } else { if (nodename == NULL) { nodename = server; } else if (server == NULL) { server = nodename; } pvt->type = pcmk__client_tls; pvt->remote_nodename = strdup(nodename); pvt->server = strdup(server); if ((pvt->remote_nodename == NULL) || (pvt->server == NULL)) { lrmd_api_delete(*api); *api = NULL; return ENOMEM; } pvt->port = port; if (pvt->port == 0) { pvt->port = crm_default_remote_port(); } } return pcmk_rc_ok; } lrmd_t * lrmd_api_new(void) { lrmd_t *api = NULL; pcmk__assert(lrmd__new(&api, NULL, NULL, 0) == pcmk_rc_ok); return api; } lrmd_t * lrmd_remote_api_new(const char *nodename, const char *server, int port) { lrmd_t *api = NULL; pcmk__assert(lrmd__new(&api, nodename, server, port) == pcmk_rc_ok); return api; } void lrmd_api_delete(lrmd_t * lrmd) { if (lrmd == NULL) { return; } if (lrmd->cmds != NULL) { // Never NULL, but make static analysis happy if (lrmd->cmds->disconnect != NULL) { // Also never really NULL lrmd->cmds->disconnect(lrmd); // No-op if already disconnected } free(lrmd->cmds); } if (lrmd->lrmd_private != NULL) { lrmd_private_t *native = lrmd->lrmd_private; free(native->server); free(native->remote_nodename); free(native->remote); free(native->token); free(native->peer_version); free(lrmd->lrmd_private); } free(lrmd); } struct metadata_cb { void (*callback)(int pid, const pcmk__action_result_t *result, void *user_data); void *user_data; }; /*! * \internal * \brief Process asynchronous metadata completion * * \param[in,out] action Metadata action that completed */ static void metadata_complete(svc_action_t *action) { struct metadata_cb *metadata_cb = (struct metadata_cb *) action->cb_data; pcmk__action_result_t result = PCMK__UNKNOWN_RESULT; services__copy_result(action, &result); pcmk__set_result_output(&result, action->stdout_data, action->stderr_data); metadata_cb->callback(0, &result, metadata_cb->user_data); result.action_stdout = NULL; // Prevent free, because action owns it result.action_stderr = NULL; // Prevent free, because action owns it pcmk__reset_result(&result); free(metadata_cb); } /*! * \internal * \brief Retrieve agent metadata asynchronously * * \param[in] rsc Resource agent specification * \param[in] callback Function to call with result (this will always be * called, whether by this function directly or later * via the main loop, and on success the metadata will * be in its result argument's action_stdout) * \param[in,out] user_data User data to pass to callback * * \return Standard Pacemaker return code * \note This function is not a lrmd_api_operations_t method because it does not * need an lrmd_t object and does not go through the executor, but * executes the agent directly. */ int lrmd__metadata_async(const lrmd_rsc_info_t *rsc, void (*callback)(int pid, const pcmk__action_result_t *result, void *user_data), void *user_data) { svc_action_t *action = NULL; struct metadata_cb *metadata_cb = NULL; pcmk__action_result_t result = PCMK__UNKNOWN_RESULT; CRM_CHECK(callback != NULL, return EINVAL); if ((rsc == NULL) || (rsc->standard == NULL) || (rsc->type == NULL)) { pcmk__set_result(&result, PCMK_OCF_NOT_CONFIGURED, PCMK_EXEC_ERROR_FATAL, "Invalid resource specification"); callback(0, &result, user_data); pcmk__reset_result(&result); return EINVAL; } if (strcmp(rsc->standard, PCMK_RESOURCE_CLASS_STONITH) == 0) { return stonith__metadata_async(rsc->type, pcmk__timeout_ms2s(PCMK_DEFAULT_ACTION_TIMEOUT_MS), callback, user_data); } action = services__create_resource_action(pcmk__s(rsc->id, rsc->type), rsc->standard, rsc->provider, rsc->type, PCMK_ACTION_META_DATA, 0, PCMK_DEFAULT_ACTION_TIMEOUT_MS, NULL, 0); if (action == NULL) { pcmk__set_result(&result, PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_ERROR, "Out of memory"); callback(0, &result, user_data); pcmk__reset_result(&result); return ENOMEM; } if (action->rc != PCMK_OCF_UNKNOWN) { services__copy_result(action, &result); callback(0, &result, user_data); pcmk__reset_result(&result); services_action_free(action); return EINVAL; } action->cb_data = calloc(1, sizeof(struct metadata_cb)); if (action->cb_data == NULL) { services_action_free(action); pcmk__set_result(&result, PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_ERROR, "Out of memory"); callback(0, &result, user_data); pcmk__reset_result(&result); return ENOMEM; } metadata_cb = (struct metadata_cb *) action->cb_data; metadata_cb->callback = callback; metadata_cb->user_data = user_data; if (!services_action_async(action, metadata_complete)) { services_action_free(action); return pcmk_rc_error; // @TODO Derive from action->rc and ->status } // The services library has taken responsibility for action return pcmk_rc_ok; } /*! * \internal * \brief Set the result of an executor event * * \param[in,out] event Executor event to set * \param[in] rc OCF exit status of event * \param[in] op_status Executor status of event * \param[in] exit_reason Human-friendly description of event */ void lrmd__set_result(lrmd_event_data_t *event, enum ocf_exitcode rc, int op_status, const char *exit_reason) { if (event == NULL) { return; } event->rc = rc; event->op_status = op_status; // lrmd_event_data_t has (const char *) members that lrmd_free_event() frees pcmk__str_update((char **) &event->exit_reason, exit_reason); } /*! * \internal * \brief Clear an executor event's exit reason, output, and error output * * \param[in,out] event Executor event to reset */ void lrmd__reset_result(lrmd_event_data_t *event) { if (event == NULL) { return; } free((void *) event->exit_reason); event->exit_reason = NULL; free((void *) event->output); event->output = NULL; } /*! * \internal * \brief Get the uptime of a remote resource connection * * When the cluster connects to a remote resource, part of that resource's * handshake includes the uptime of the remote resource's connection. This * uptime is stored in the lrmd_t object. * * \return The connection's uptime, or -1 if unknown */ time_t lrmd__uptime(lrmd_t *lrmd) { lrmd_private_t *native = lrmd->lrmd_private; if (native->remote == NULL) { return -1; } else { return native->remote->uptime; } } const char * lrmd__node_start_state(lrmd_t *lrmd) { lrmd_private_t *native = lrmd->lrmd_private; if (native->remote == NULL) { return NULL; } else { return native->remote->start_state; } } diff --git a/lib/pacemaker/pcmk_graph_consumer.c b/lib/pacemaker/pcmk_graph_consumer.c index d69dd2ba95..a688177916 100644 --- a/lib/pacemaker/pcmk_graph_consumer.c +++ b/lib/pacemaker/pcmk_graph_consumer.c @@ -1,865 +1,869 @@ /* * Copyright 2004-2025 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 /* * Functions for freeing transition graph objects */ /*! * \internal * \brief Free a transition graph action object * * \param[in,out] user_data Action to free */ static void free_graph_action(gpointer user_data) { pcmk__graph_action_t *action = user_data; if (action->timer != 0) { crm_warn("Cancelling timer for graph action %d", action->id); g_source_remove(action->timer); } if (action->params != NULL) { g_hash_table_destroy(action->params); } pcmk__xml_free(action->xml); free(action); } /*! * \internal * \brief Free a transition graph synapse object * * \param[in,out] user_data Synapse to free */ static void free_graph_synapse(gpointer user_data) { pcmk__graph_synapse_t *synapse = user_data; g_list_free_full(synapse->actions, free_graph_action); g_list_free_full(synapse->inputs, free_graph_action); free(synapse); } /*! * \internal * \brief Free a transition graph object * * \param[in,out] graph Transition graph to free */ void pcmk__free_graph(pcmk__graph_t *graph) { if (graph != NULL) { g_list_free_full(graph->synapses, free_graph_synapse); free(graph->source); free(graph->failed_stop_offset); free(graph->failed_start_offset); free(graph); } } /* * Functions for updating graph */ /*! * \internal * \brief Update synapse after completed prerequisite * * A synapse is ready to be executed once all its prerequisite actions (inputs) * complete. Given a completed action, check whether it is an input for a given * synapse, and if so, mark the input as confirmed, and mark the synapse as * ready if appropriate. * * \param[in,out] synapse Transition graph synapse to update * \param[in] action_id ID of an action that completed * * \note The only substantial effect here is confirming synapse inputs. * should_fire_synapse() will recalculate pcmk__synapse_ready, so the only * thing that uses the pcmk__synapse_ready from here is * synapse_state_str(). */ static void update_synapse_ready(pcmk__graph_synapse_t *synapse, int action_id) { if (pcmk_is_set(synapse->flags, pcmk__synapse_ready)) { return; // All inputs have already been confirmed } // Presume ready until proven otherwise pcmk__set_synapse_flags(synapse, pcmk__synapse_ready); for (GList *lpc = synapse->inputs; lpc != NULL; lpc = lpc->next) { pcmk__graph_action_t *prereq = (pcmk__graph_action_t *) lpc->data; if (prereq->id == action_id) { crm_trace("Confirming input %d of synapse %d", action_id, synapse->id); pcmk__set_graph_action_flags(prereq, pcmk__graph_action_confirmed); } else if (!pcmk_is_set(prereq->flags, pcmk__graph_action_confirmed)) { pcmk__clear_synapse_flags(synapse, pcmk__synapse_ready); crm_trace("Synapse %d still not ready after action %d", synapse->id, action_id); } } if (pcmk_is_set(synapse->flags, pcmk__synapse_ready)) { crm_trace("Synapse %d is now ready to execute", synapse->id); } } /*! * \internal * \brief Update action and synapse confirmation after action completion * * \param[in,out] synapse Transition graph synapse that action belongs to * \param[in] action_id ID of action that completed */ static void update_synapse_confirmed(pcmk__graph_synapse_t *synapse, int action_id) { bool all_confirmed = true; for (GList *lpc = synapse->actions; lpc != NULL; lpc = lpc->next) { pcmk__graph_action_t *action = (pcmk__graph_action_t *) lpc->data; if (action->id == action_id) { crm_trace("Confirmed action %d of synapse %d", action_id, synapse->id); pcmk__set_graph_action_flags(action, pcmk__graph_action_confirmed); } else if (all_confirmed && !pcmk_is_set(action->flags, pcmk__graph_action_confirmed)) { all_confirmed = false; crm_trace("Synapse %d still not confirmed after action %d", synapse->id, action_id); } } if (all_confirmed && !pcmk_is_set(synapse->flags, pcmk__synapse_confirmed)) { crm_trace("Confirmed synapse %d", synapse->id); pcmk__set_synapse_flags(synapse, pcmk__synapse_confirmed); } } /*! * \internal * \brief Update the transition graph with a completed action result * * \param[in,out] graph Transition graph to update * \param[in] action Action that completed */ void pcmk__update_graph(pcmk__graph_t *graph, const pcmk__graph_action_t *action) { for (GList *lpc = graph->synapses; lpc != NULL; lpc = lpc->next) { pcmk__graph_synapse_t *synapse = (pcmk__graph_synapse_t *) lpc->data; if (pcmk_any_flags_set(synapse->flags, pcmk__synapse_confirmed|pcmk__synapse_failed)) { continue; // This synapse already completed } else if (pcmk_is_set(synapse->flags, pcmk__synapse_executed)) { update_synapse_confirmed(synapse, action->id); } else if (!pcmk_is_set(action->flags, pcmk__graph_action_failed) || (synapse->priority == PCMK_SCORE_INFINITY)) { update_synapse_ready(synapse, action->id); } } } /* * Functions for executing graph */ /* A transition graph consists of various types of actions. The library caller * registers execution functions for each action type, which will be stored * here. */ static pcmk__graph_functions_t *graph_fns = NULL; /*! * \internal * \brief Set transition graph execution functions * * \param[in] Execution functions to use */ void pcmk__set_graph_functions(pcmk__graph_functions_t *fns) { pcmk__assert((fns != NULL) && (fns->rsc != NULL) && (fns->cluster != NULL) && (fns->pseudo != NULL) && (fns->fence != NULL)); crm_debug("Setting custom functions for executing transition graphs"); graph_fns = fns; } /*! * \internal * \brief Check whether a graph synapse is ready to be executed * * \param[in,out] graph Transition graph that synapse is part of * \param[in,out] synapse Synapse to check * * \return true if synapse is ready, false otherwise */ static bool should_fire_synapse(pcmk__graph_t *graph, pcmk__graph_synapse_t *synapse) { GList *lpc = NULL; pcmk__set_synapse_flags(synapse, pcmk__synapse_ready); for (lpc = synapse->inputs; lpc != NULL; lpc = lpc->next) { pcmk__graph_action_t *prereq = (pcmk__graph_action_t *) lpc->data; if (!(pcmk_is_set(prereq->flags, pcmk__graph_action_confirmed))) { crm_trace("Input %d for synapse %d not yet confirmed", prereq->id, synapse->id); pcmk__clear_synapse_flags(synapse, pcmk__synapse_ready); break; } else if (pcmk_is_set(prereq->flags, pcmk__graph_action_failed)) { crm_trace("Input %d for synapse %d confirmed but failed", prereq->id, synapse->id); pcmk__clear_synapse_flags(synapse, pcmk__synapse_ready); break; } } if (pcmk_is_set(synapse->flags, pcmk__synapse_ready)) { crm_trace("Synapse %d is ready to execute", synapse->id); } else { return false; } for (lpc = synapse->actions; lpc != NULL; lpc = lpc->next) { pcmk__graph_action_t *a = (pcmk__graph_action_t *) lpc->data; if (a->type == pcmk__pseudo_graph_action) { /* None of the below applies to pseudo ops */ } else if (synapse->priority < graph->abort_priority) { crm_trace("Skipping synapse %d: priority %d is less than " "abort priority %d", synapse->id, synapse->priority, graph->abort_priority); graph->skipped++; return false; } else if (graph_fns->allowed && !(graph_fns->allowed(graph, a))) { crm_trace("Deferring synapse %d: not allowed", synapse->id); return false; } } return true; } /*! * \internal * \brief Initiate an action from a transition graph * * \param[in,out] graph Transition graph containing action * \param[in,out] action Action to execute * * \return Standard Pacemaker return code */ static int initiate_action(pcmk__graph_t *graph, pcmk__graph_action_t *action) { const char *id = pcmk__xe_id(action->xml); CRM_CHECK(id != NULL, return EINVAL); CRM_CHECK(!pcmk_is_set(action->flags, pcmk__graph_action_executed), return pcmk_rc_already); pcmk__set_graph_action_flags(action, pcmk__graph_action_executed); switch (action->type) { case pcmk__pseudo_graph_action: crm_trace("Executing pseudo-action %d (%s)", action->id, id); return graph_fns->pseudo(graph, action); case pcmk__rsc_graph_action: crm_trace("Executing resource action %d (%s)", action->id, id); return graph_fns->rsc(graph, action); case pcmk__cluster_graph_action: if (pcmk__str_eq(pcmk__xe_get(action->xml, PCMK_XA_OPERATION), PCMK_ACTION_STONITH, pcmk__str_none)) { crm_trace("Executing fencing action %d (%s)", action->id, id); return graph_fns->fence(graph, action); } crm_trace("Executing cluster action %d (%s)", action->id, id); return graph_fns->cluster(graph, action); default: crm_err("Unsupported graph action type <%s " PCMK_XA_ID "='%s'> " "(bug?)", action->xml->name, id); return EINVAL; } } /*! * \internal * \brief Execute a graph synapse * * \param[in,out] graph Transition graph with synapse to execute * \param[in,out] synapse Synapse to execute * * \return Standard Pacemaker return value */ static int fire_synapse(pcmk__graph_t *graph, pcmk__graph_synapse_t *synapse) { pcmk__set_synapse_flags(synapse, pcmk__synapse_executed); for (GList *lpc = synapse->actions; lpc != NULL; lpc = lpc->next) { pcmk__graph_action_t *action = (pcmk__graph_action_t *) lpc->data; int rc = initiate_action(graph, action); if (rc != pcmk_rc_ok) { crm_err("Failed initiating <%s " PCMK_XA_ID "=%d> in synapse %d: " "%s", action->xml->name, action->id, synapse->id, pcmk_rc_str(rc)); pcmk__set_synapse_flags(synapse, pcmk__synapse_confirmed); pcmk__set_graph_action_flags(action, pcmk__graph_action_confirmed |pcmk__graph_action_failed); return pcmk_rc_error; } } return pcmk_rc_ok; } /*! * \internal * \brief Dummy graph method that can be used with simulations * * \param[in,out] graph Transition graph containing action * \param[in,out] action Graph action to be initiated * * \return Standard Pacemaker return code * \note If the PE_fail environment variable is set to the action ID, * then the graph action will be marked as failed. */ static int pseudo_action_dummy(pcmk__graph_t *graph, pcmk__graph_action_t *action) { static int fail = -1; if (fail < 0) { long long fail_ll; if ((pcmk__scan_ll(getenv("PE_fail"), &fail_ll, 0LL) == pcmk_rc_ok) && (fail_ll > 0LL) && (fail_ll <= INT_MAX)) { fail = (int) fail_ll; } else { fail = 0; } } if (action->id == fail) { crm_err("Dummy event handler: pretending action %d failed", action->id); pcmk__set_graph_action_flags(action, pcmk__graph_action_failed); graph->abort_priority = PCMK_SCORE_INFINITY; } else { crm_trace("Dummy event handler: action %d initiated", action->id); } pcmk__set_graph_action_flags(action, pcmk__graph_action_confirmed); pcmk__update_graph(graph, action); return pcmk_rc_ok; } static pcmk__graph_functions_t default_fns = { pseudo_action_dummy, pseudo_action_dummy, pseudo_action_dummy, pseudo_action_dummy }; /*! * \internal * \brief Execute all actions in a transition graph * * \param[in,out] graph Transition graph to execute * * \return Status of transition after execution */ enum pcmk__graph_status pcmk__execute_graph(pcmk__graph_t *graph) { GList *lpc = NULL; int log_level = LOG_DEBUG; enum pcmk__graph_status pass_result = pcmk__graph_active; const char *status = "In progress"; if (graph_fns == NULL) { graph_fns = &default_fns; } if (graph == NULL) { return pcmk__graph_complete; } graph->fired = 0; graph->pending = 0; graph->skipped = 0; graph->completed = 0; graph->incomplete = 0; // Count completed and in-flight synapses for (lpc = graph->synapses; lpc != NULL; lpc = lpc->next) { pcmk__graph_synapse_t *synapse = (pcmk__graph_synapse_t *) lpc->data; if (pcmk_is_set(synapse->flags, pcmk__synapse_confirmed)) { graph->completed++; } else if (!pcmk_is_set(synapse->flags, pcmk__synapse_failed) && pcmk_is_set(synapse->flags, pcmk__synapse_executed)) { graph->pending++; } } crm_trace("Executing graph %d (%d synapses already completed, %d pending)", graph->id, graph->completed, graph->pending); // Execute any synapses that are ready for (lpc = graph->synapses; lpc != NULL; lpc = lpc->next) { pcmk__graph_synapse_t *synapse = (pcmk__graph_synapse_t *) lpc->data; if ((graph->batch_limit > 0) && (graph->pending >= graph->batch_limit)) { crm_debug("Throttling graph execution: batch limit (%d) reached", graph->batch_limit); break; } else if (pcmk_is_set(synapse->flags, pcmk__synapse_failed)) { graph->skipped++; continue; } else if (pcmk_any_flags_set(synapse->flags, pcmk__synapse_confirmed |pcmk__synapse_executed)) { continue; // Already handled } else if (should_fire_synapse(graph, synapse)) { graph->fired++; if (fire_synapse(graph, synapse) != pcmk_rc_ok) { crm_err("Synapse %d failed to fire", synapse->id); log_level = LOG_ERR; graph->abort_priority = PCMK_SCORE_INFINITY; graph->incomplete++; graph->fired--; } if (!(pcmk_is_set(synapse->flags, pcmk__synapse_confirmed))) { graph->pending++; } } else { crm_trace("Synapse %d cannot fire", synapse->id); graph->incomplete++; } } if ((graph->pending == 0) && (graph->fired == 0)) { graph->complete = true; if ((graph->incomplete != 0) && (graph->abort_priority <= 0)) { log_level = LOG_WARNING; pass_result = pcmk__graph_terminated; status = "Terminated"; } else if (graph->skipped != 0) { log_level = LOG_NOTICE; pass_result = pcmk__graph_complete; status = "Stopped"; } else { log_level = LOG_NOTICE; pass_result = pcmk__graph_complete; status = "Complete"; } } else if (graph->fired == 0) { pass_result = pcmk__graph_pending; } do_crm_log(log_level, "Transition %d (Complete=%d, Pending=%d," " Fired=%d, Skipped=%d, Incomplete=%d, Source=%s): %s", graph->id, graph->completed, graph->pending, graph->fired, graph->skipped, graph->incomplete, graph->source, status); return pass_result; } /* * Functions for unpacking transition graph XML into structs */ /*! * \internal * \brief Unpack a transition graph action from XML * * \param[in] parent Synapse that action is part of * \param[in] xml_action Action XML to unparse * * \return Newly allocated action on success, or NULL otherwise */ static pcmk__graph_action_t * unpack_action(pcmk__graph_synapse_t *parent, xmlNode *xml_action) { enum pcmk__graph_action_type action_type; pcmk__graph_action_t *action = NULL; const char *value = pcmk__xe_id(xml_action); if (value == NULL) { crm_err("Ignoring transition graph action without " PCMK_XA_ID " (bug?)"); crm_log_xml_trace(xml_action, "invalid"); return NULL; } if (pcmk__xe_is(xml_action, PCMK__XE_RSC_OP)) { action_type = pcmk__rsc_graph_action; } else if (pcmk__xe_is(xml_action, PCMK__XE_PSEUDO_EVENT)) { action_type = pcmk__pseudo_graph_action; } else if (pcmk__xe_is(xml_action, PCMK__XE_CRM_EVENT)) { action_type = pcmk__cluster_graph_action; } else { crm_err("Ignoring transition graph action of unknown type '%s' (bug?)", xml_action->name); crm_log_xml_trace(xml_action, "invalid"); return NULL; } + /* @TODO Should we use pcmk__assert_alloc() here? A crash seems preferable + * to returning a graph with missing actions. Besides, there will likely be + * more allocation failures after this. + */ action = calloc(1, sizeof(pcmk__graph_action_t)); if (action == NULL) { - crm_perror(LOG_CRIT, "Cannot unpack transition graph action"); + crm_crit("Cannot unpack transition graph action: %s", strerror(errno)); crm_log_xml_trace(xml_action, "lost"); return NULL; } pcmk__scan_min_int(value, &(action->id), -1); action->type = pcmk__rsc_graph_action; action->xml = pcmk__xml_copy(NULL, xml_action); action->synapse = parent; action->type = action_type; action->params = xml2list(action->xml); value = crm_meta_value(action->params, PCMK_META_TIMEOUT); pcmk__scan_min_int(value, &(action->timeout), 0); /* Take PCMK_META_START_DELAY into account for the timeout of the action * timer */ value = crm_meta_value(action->params, PCMK_META_START_DELAY); { int start_delay; pcmk__scan_min_int(value, &start_delay, 0); action->timeout += start_delay; } if (pcmk__guint_from_hash(action->params, CRM_META "_" PCMK_META_INTERVAL, 0, &(action->interval_ms)) != pcmk_rc_ok) { action->interval_ms = 0; } crm_trace("Action %d has timer set to %dms", action->id, action->timeout); return action; } /*! * \internal * \brief Unpack transition graph synapse from XML * * \param[in,out] new_graph Transition graph that synapse is part of * \param[in] xml_synapse Synapse XML * * \return Newly allocated synapse on success, or NULL otherwise */ static pcmk__graph_synapse_t * unpack_synapse(pcmk__graph_t *new_graph, const xmlNode *xml_synapse) { const char *value = NULL; xmlNode *action_set = NULL; pcmk__graph_synapse_t *new_synapse = NULL; crm_trace("Unpacking synapse %s", pcmk__xe_id(xml_synapse)); new_synapse = calloc(1, sizeof(pcmk__graph_synapse_t)); if (new_synapse == NULL) { return NULL; } pcmk__scan_min_int(pcmk__xe_id(xml_synapse), &(new_synapse->id), 0); value = pcmk__xe_get(xml_synapse, PCMK__XA_PRIORITY); pcmk__scan_min_int(value, &(new_synapse->priority), 0); CRM_CHECK(new_synapse->id >= 0, free_graph_synapse((gpointer) new_synapse); return NULL); new_graph->num_synapses++; crm_trace("Unpacking synapse %s action sets", pcmk__xe_get(xml_synapse, PCMK_XA_ID)); for (action_set = pcmk__xe_first_child(xml_synapse, PCMK__XE_ACTION_SET, NULL, NULL); action_set != NULL; action_set = pcmk__xe_next(action_set, PCMK__XE_ACTION_SET)) { for (xmlNode *action = pcmk__xe_first_child(action_set, NULL, NULL, NULL); action != NULL; action = pcmk__xe_next(action, NULL)) { pcmk__graph_action_t *new_action = unpack_action(new_synapse, action); if (new_action == NULL) { continue; } crm_trace("Adding action %d to synapse %d", new_action->id, new_synapse->id); new_graph->num_actions++; new_synapse->actions = g_list_append(new_synapse->actions, new_action); } } crm_trace("Unpacking synapse %s inputs", pcmk__xe_id(xml_synapse)); for (xmlNode *inputs = pcmk__xe_first_child(xml_synapse, PCMK__XE_INPUTS, NULL, NULL); inputs != NULL; inputs = pcmk__xe_next(inputs, PCMK__XE_INPUTS)) { for (xmlNode *trigger = pcmk__xe_first_child(inputs, PCMK__XE_TRIGGER, NULL, NULL); trigger != NULL; trigger = pcmk__xe_next(trigger, PCMK__XE_TRIGGER)) { for (xmlNode *input = pcmk__xe_first_child(trigger, NULL, NULL, NULL); input != NULL; input = pcmk__xe_next(input, NULL)) { pcmk__graph_action_t *new_input = unpack_action(new_synapse, input); if (new_input == NULL) { continue; } crm_trace("Adding input %d to synapse %d", new_input->id, new_synapse->id); new_synapse->inputs = g_list_append(new_synapse->inputs, new_input); } } } return new_synapse; } /*! * \internal * \brief Unpack transition graph XML * * \param[in] xml_graph Transition graph XML to unpack * \param[in] reference Where the XML came from (for logging) * * \return Newly allocated transition graph on success, NULL otherwise * \note The caller is responsible for freeing the return value using * pcmk__free_graph(). * \note The XML is expected to be structured like: ... ... */ pcmk__graph_t * pcmk__unpack_graph(const xmlNode *xml_graph, const char *reference) { pcmk__graph_t *new_graph = NULL; new_graph = calloc(1, sizeof(pcmk__graph_t)); if (new_graph == NULL) { return NULL; } new_graph->source = strdup(pcmk__s(reference, "unknown")); if (new_graph->source == NULL) { pcmk__free_graph(new_graph); return NULL; } new_graph->completion_action = pcmk__graph_done; // Parse top-level attributes from PCMK__XE_TRANSITION_GRAPH if (xml_graph != NULL) { const char *buf = pcmk__xe_get(xml_graph, "transition_id"); CRM_CHECK(buf != NULL, pcmk__free_graph(new_graph); return NULL); pcmk__scan_min_int(buf, &(new_graph->id), 1); buf = pcmk__xe_get(xml_graph, PCMK_OPT_CLUSTER_DELAY); CRM_CHECK(buf != NULL, pcmk__free_graph(new_graph); return NULL); pcmk_parse_interval_spec(buf, &(new_graph->network_delay)); buf = pcmk__xe_get(xml_graph, PCMK_OPT_STONITH_TIMEOUT); if (buf == NULL) { new_graph->stonith_timeout = new_graph->network_delay; } else { pcmk_parse_interval_spec(buf, &(new_graph->stonith_timeout)); } // Use 0 (dynamic limit) as default/invalid, -1 (no limit) as minimum buf = pcmk__xe_get(xml_graph, PCMK_OPT_BATCH_LIMIT); if ((buf == NULL) || (pcmk__scan_min_int(buf, &(new_graph->batch_limit), -1) != pcmk_rc_ok)) { new_graph->batch_limit = 0; } buf = pcmk__xe_get(xml_graph, PCMK_OPT_MIGRATION_LIMIT); pcmk__scan_min_int(buf, &(new_graph->migration_limit), -1); new_graph->failed_stop_offset = pcmk__xe_get_copy(xml_graph, PCMK__XA_FAILED_STOP_OFFSET); new_graph->failed_start_offset = pcmk__xe_get_copy(xml_graph, PCMK__XA_FAILED_START_OFFSET); pcmk__xe_get_time(xml_graph, "recheck-by", &(new_graph->recheck_by)); } // Unpack each child element for (const xmlNode *synapse_xml = pcmk__xe_first_child(xml_graph, PCMK__XE_SYNAPSE, NULL, NULL); synapse_xml != NULL; synapse_xml = pcmk__xe_next(synapse_xml, PCMK__XE_SYNAPSE)) { pcmk__graph_synapse_t *new_synapse = unpack_synapse(new_graph, synapse_xml); if (new_synapse != NULL) { new_graph->synapses = g_list_append(new_graph->synapses, new_synapse); } } crm_debug("Unpacked transition %d from %s: %d actions in %d synapses", new_graph->id, new_graph->source, new_graph->num_actions, new_graph->num_synapses); return new_graph; } /* * Other transition graph utilities */ /*! * \internal * \brief Synthesize an executor event from a graph action * * \param[in] resource If not NULL, use greater call ID than in this XML * \param[in] action Graph action * \param[in] status What to use as event execution status * \param[in] rc What to use as event exit status * \param[in] exit_reason What to use as event exit reason * * \return Newly allocated executor event on success, or NULL otherwise */ lrmd_event_data_t * pcmk__event_from_graph_action(const xmlNode *resource, const pcmk__graph_action_t *action, int status, int rc, const char *exit_reason) { lrmd_event_data_t *op = NULL; GHashTableIter iter; const char *name = NULL; const char *value = NULL; xmlNode *action_resource = NULL; CRM_CHECK(action != NULL, return NULL); CRM_CHECK(action->type == pcmk__rsc_graph_action, return NULL); action_resource = pcmk__xe_first_child(action->xml, PCMK_XE_PRIMITIVE, NULL, NULL); CRM_CHECK(action_resource != NULL, crm_log_xml_warn(action->xml, "invalid"); return NULL); op = lrmd_new_event(pcmk__xe_id(action_resource), pcmk__xe_get(action->xml, PCMK_XA_OPERATION), action->interval_ms); lrmd__set_result(op, rc, status, exit_reason); op->t_run = time(NULL); op->t_rcchange = op->t_run; op->params = pcmk__strkey_table(free, free); g_hash_table_iter_init(&iter, action->params); while (g_hash_table_iter_next(&iter, (void **)&name, (void **)&value)) { pcmk__insert_dup(op->params, name, value); } for (xmlNode *xop = pcmk__xe_first_child(resource, NULL, NULL, NULL); xop != NULL; xop = pcmk__xe_next(xop, NULL)) { int tmp = 0; pcmk__xe_get_int(xop, PCMK__XA_CALL_ID, &tmp); crm_debug("Got call_id=%d for %s", tmp, pcmk__xe_id(resource)); if (tmp > op->call_id) { op->call_id = tmp; } } op->call_id++; return op; } diff --git a/lib/pengine/pe_digest.c b/lib/pengine/pe_digest.c index 68f88ecf6c..edbc9a14a9 100644 --- a/lib/pengine/pe_digest.c +++ b/lib/pengine/pe_digest.c @@ -1,611 +1,609 @@ /* * Copyright 2004-2025 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 "pe_status_private.h" -extern bool pcmk__is_daemon; - /*! * \internal * \brief Free an operation digest cache entry * * \param[in,out] ptr Pointer to cache entry to free * * \note The argument is a gpointer so this can be used as a hash table * free function. */ void pe__free_digests(gpointer ptr) { pcmk__op_digest_t *data = ptr; if (data != NULL) { pcmk__xml_free(data->params_all); pcmk__xml_free(data->params_secure); pcmk__xml_free(data->params_restart); free(data->digest_all_calc); free(data->digest_restart_calc); free(data->digest_secure_calc); free(data); } } // Return true if XML attribute name is not substring of a given string static bool attr_not_in_string(xmlAttrPtr a, void *user_data) { bool filter = false; char *name = crm_strdup_printf(" %s ", (const char *) a->name); if (strstr((const char *) user_data, name) == NULL) { crm_trace("Filtering %s (not found in '%s')", (const char *) a->name, (const char *) user_data); filter = true; } free(name); return filter; } // Return true if XML attribute name is substring of a given string static bool attr_in_string(xmlAttrPtr a, void *user_data) { bool filter = false; char *name = crm_strdup_printf(" %s ", (const char *) a->name); if (strstr((const char *) user_data, name) != NULL) { crm_trace("Filtering %s (found in '%s')", (const char *) a->name, (const char *) user_data); filter = true; } free(name); return filter; } /*! * \internal * \brief Add digest of all parameters to a digest cache entry * * \param[out] data Digest cache entry to modify * \param[in,out] rsc Resource that action was for * \param[in] node Node action was performed on * \param[in] params Resource parameters evaluated for node * \param[in] task Name of action performed * \param[in,out] interval_ms Action's interval (will be reset if in overrides) * \param[in] xml_op Unused * \param[in] op_version CRM feature set to use for digest calculation * \param[in] overrides Key/value table to override resource parameters * \param[in,out] scheduler Scheduler data */ static void calculate_main_digest(pcmk__op_digest_t *data, pcmk_resource_t *rsc, const pcmk_node_t *node, GHashTable *params, const char *task, guint *interval_ms, const xmlNode *xml_op, const char *op_version, GHashTable *overrides, pcmk_scheduler_t *scheduler) { xmlNode *action_config = NULL; data->params_all = pcmk__xe_create(NULL, PCMK_XE_PARAMETERS); /* REMOTE_CONTAINER_HACK: Allow Pacemaker Remote nodes to run containers * that themselves are Pacemaker Remote nodes */ (void) pe__add_bundle_remote_name(rsc, data->params_all, PCMK_REMOTE_RA_ADDR); if (overrides != NULL) { // If interval was overridden, reset it const char *meta_name = CRM_META "_" PCMK_META_INTERVAL; const char *interval_s = g_hash_table_lookup(overrides, meta_name); if (interval_s != NULL) { long long value_ll; if ((pcmk__scan_ll(interval_s, &value_ll, 0LL) == pcmk_rc_ok) && (value_ll >= 0) && (value_ll <= G_MAXUINT)) { *interval_ms = (guint) value_ll; } } // Add overrides to list of all parameters g_hash_table_foreach(overrides, hash2field, data->params_all); } // Add provided instance parameters g_hash_table_foreach(params, hash2field, data->params_all); // Find action configuration XML in CIB action_config = pcmk__find_action_config(rsc, task, *interval_ms, true); /* Add action-specific resource instance attributes to the digest list. * * If this is a one-time action with action-specific instance attributes, * enforce a restart instead of reload-agent in case the main digest doesn't * match, even if the restart digest does. This ensures any changes of the * action-specific parameters get applied for this specific action, and * digests calculated for the resulting history will be correct. Default the * result to RSC_DIGEST_RESTART for the case where the main digest doesn't * match. */ params = pcmk__unpack_action_rsc_params(action_config, node->priv->attrs, scheduler); if ((*interval_ms == 0) && (g_hash_table_size(params) > 0)) { data->rc = pcmk__digest_restart; } g_hash_table_foreach(params, hash2field, data->params_all); g_hash_table_destroy(params); // Add action meta-attributes params = pcmk__unpack_action_meta(rsc, node, task, *interval_ms, action_config); g_hash_table_foreach(params, hash2metafield, data->params_all); g_hash_table_destroy(params); pcmk__filter_op_for_digest(data->params_all); data->digest_all_calc = pcmk__digest_op_params(data->params_all); } // Return true if XML attribute name is a Pacemaker-defined fencing parameter static bool is_fence_param(xmlAttrPtr attr, void *user_data) { return pcmk_stonith_param((const char *) attr->name); } /*! * \internal * \brief Add secure digest to a digest cache entry * * \param[out] data Digest cache entry to modify * \param[in] rsc Resource that action was for * \param[in] params Resource parameters evaluated for node * \param[in] xml_op XML of operation in CIB status (if available) * \param[in] op_version CRM feature set to use for digest calculation * \param[in] overrides Key/value hash table to override resource parameters */ static void calculate_secure_digest(pcmk__op_digest_t *data, const pcmk_resource_t *rsc, GHashTable *params, const xmlNode *xml_op, const char *op_version, GHashTable *overrides) { const char *class = pcmk__xe_get(rsc->priv->xml, PCMK_XA_CLASS); const char *secure_list = NULL; bool old_version = (compare_version(op_version, "3.16.0") < 0); if (xml_op == NULL) { secure_list = " passwd password user "; } else { secure_list = pcmk__xe_get(xml_op, PCMK__XA_OP_SECURE_PARAMS); } if (old_version) { data->params_secure = pcmk__xe_create(NULL, PCMK_XE_PARAMETERS); if (overrides != NULL) { g_hash_table_foreach(overrides, hash2field, data->params_secure); } g_hash_table_foreach(params, hash2field, data->params_secure); } else { // Start with a copy of all parameters data->params_secure = pcmk__xml_copy(NULL, data->params_all); } if (secure_list != NULL) { pcmk__xe_remove_matching_attrs(data->params_secure, false, attr_in_string, (void *) secure_list); } if (old_version && pcmk_is_set(pcmk_get_ra_caps(class), pcmk_ra_cap_fence_params)) { /* For stonith resources, Pacemaker adds special parameters, * but these are not listed in fence agent meta-data, so with older * versions of DC, the controller will not hash them. That means we have * to filter them out before calculating our hash for comparison. */ pcmk__xe_remove_matching_attrs(data->params_secure, false, is_fence_param, NULL); } pcmk__filter_op_for_digest(data->params_secure); /* CRM_meta_timeout *should* be part of a digest for recurring operations. * However, with older versions of DC, the controller does not add timeout * to secure digests, because it only includes parameters declared by the * resource agent. * Remove any timeout that made it this far, to match. */ if (old_version) { pcmk__xe_remove_attr(data->params_secure, CRM_META "_" PCMK_META_TIMEOUT); } data->digest_secure_calc = pcmk__digest_op_params(data->params_secure); } /*! * \internal * \brief Add restart digest to a digest cache entry * * \param[out] data Digest cache entry to modify * \param[in] xml_op XML of operation in CIB status (if available) * \param[in] op_version CRM feature set to use for digest calculation * * \note This function doesn't need to handle overrides because it starts with * data->params_all, which already has overrides applied. */ static void calculate_restart_digest(pcmk__op_digest_t *data, const xmlNode *xml_op, const char *op_version) { const char *value = NULL; // We must have XML of resource operation history if (xml_op == NULL) { return; } // And the history must have a restart digest to compare against if (pcmk__xe_get(xml_op, PCMK__XA_OP_RESTART_DIGEST) == NULL) { return; } // Start with a copy of all parameters data->params_restart = pcmk__xml_copy(NULL, data->params_all); // Then filter out reloadable parameters, if any value = pcmk__xe_get(xml_op, PCMK__XA_OP_FORCE_RESTART); if (value != NULL) { pcmk__xe_remove_matching_attrs(data->params_restart, false, attr_not_in_string, (void *) value); } data->digest_restart_calc = pcmk__digest_op_params(data->params_restart); } /*! * \internal * \brief Create a new digest cache entry with calculated digests * * \param[in,out] rsc Resource that action was for * \param[in] task Name of action performed * \param[in,out] interval_ms Action's interval (will be reset if in overrides) * \param[in] node Node action was performed on * \param[in] xml_op XML of operation in CIB status (if available) * \param[in] overrides Key/value table to override resource parameters * \param[in] calc_secure Whether to calculate secure digest * \param[in,out] scheduler Scheduler data * * \return Pointer to new digest cache entry (or NULL on memory error) * \note It is the caller's responsibility to free the result using * pe__free_digests(). */ pcmk__op_digest_t * pe__calculate_digests(pcmk_resource_t *rsc, const char *task, guint *interval_ms, const pcmk_node_t *node, const xmlNode *xml_op, GHashTable *overrides, bool calc_secure, pcmk_scheduler_t *scheduler) { pcmk__op_digest_t *data = NULL; const char *op_version = NULL; GHashTable *params = NULL; CRM_CHECK(scheduler != NULL, return NULL); data = calloc(1, sizeof(pcmk__op_digest_t)); if (data == NULL) { pcmk__sched_err(scheduler, "Could not allocate memory for operation digest"); return NULL; } data->rc = pcmk__digest_match; if (xml_op != NULL) { op_version = pcmk__xe_get(xml_op, PCMK_XA_CRM_FEATURE_SET); } if ((op_version == NULL) && (scheduler->input != NULL)) { op_version = pcmk__xe_get(scheduler->input, PCMK_XA_CRM_FEATURE_SET); } if (op_version == NULL) { op_version = CRM_FEATURE_SET; } params = pe_rsc_params(rsc, node, scheduler); calculate_main_digest(data, rsc, node, params, task, interval_ms, xml_op, op_version, overrides, scheduler); if (calc_secure) { calculate_secure_digest(data, rsc, params, xml_op, op_version, overrides); } calculate_restart_digest(data, xml_op, op_version); return data; } /*! * \internal * \brief Calculate action digests and store in node's digest cache * * \param[in,out] rsc Resource that action was for * \param[in] task Name of action performed * \param[in] interval_ms Action's interval * \param[in,out] node Node action was performed on * \param[in] xml_op XML of operation in CIB status (if available) * \param[in] calc_secure Whether to calculate secure digest * \param[in,out] scheduler Scheduler data * * \return Pointer to node's digest cache entry */ static pcmk__op_digest_t * rsc_action_digest(pcmk_resource_t *rsc, const char *task, guint interval_ms, pcmk_node_t *node, const xmlNode *xml_op, bool calc_secure, pcmk_scheduler_t *scheduler) { pcmk__op_digest_t *data = NULL; char *key = pcmk__op_key(rsc->id, task, interval_ms); data = g_hash_table_lookup(node->priv->digest_cache, key); if (data == NULL) { data = pe__calculate_digests(rsc, task, &interval_ms, node, xml_op, NULL, calc_secure, scheduler); pcmk__assert(data != NULL); g_hash_table_insert(node->priv->digest_cache, strdup(key), data); } free(key); return data; } /*! * \internal * \brief Calculate operation digests and compare against an XML history entry * * \param[in,out] rsc Resource to check * \param[in] xml_op Resource history XML * \param[in,out] node Node to use for digest calculation * \param[in,out] scheduler Scheduler data * * \return Pointer to node's digest cache entry, with comparison result set */ pcmk__op_digest_t * rsc_action_digest_cmp(pcmk_resource_t *rsc, const xmlNode *xml_op, pcmk_node_t *node, pcmk_scheduler_t *scheduler) { pcmk__op_digest_t *data = NULL; guint interval_ms = 0; const char *op_version; const char *task = pcmk__xe_get(xml_op, PCMK_XA_OPERATION); const char *digest_all; const char *digest_restart; pcmk__assert(node != NULL); op_version = pcmk__xe_get(xml_op, PCMK_XA_CRM_FEATURE_SET); digest_all = pcmk__xe_get(xml_op, PCMK__XA_OP_DIGEST); digest_restart = pcmk__xe_get(xml_op, PCMK__XA_OP_RESTART_DIGEST); pcmk__xe_get_guint(xml_op, PCMK_META_INTERVAL, &interval_ms); data = rsc_action_digest(rsc, task, interval_ms, node, xml_op, pcmk_is_set(scheduler->flags, pcmk__sched_sanitized), scheduler); if (!pcmk__str_eq(data->digest_restart_calc, digest_restart, pcmk__str_null_matches)) { pcmk__rsc_info(rsc, "Parameter change for %s-interval %s of %s on %s " "requires restart (hash now %s vs. %s " "with op feature set %s for transition %s)", pcmk__readable_interval(interval_ms), task, rsc->id, pcmk__node_name(node), data->digest_restart_calc, pcmk__s(digest_restart, "missing"), op_version, pcmk__xe_get(xml_op, PCMK__XA_TRANSITION_MAGIC)); data->rc = pcmk__digest_restart; } else if (digest_all == NULL) { /* it is unknown what the previous op digest was */ data->rc = pcmk__digest_unknown; } else if (strcmp(digest_all, data->digest_all_calc) != 0) { /* Given a non-recurring operation with extra parameters configured, * in case that the main digest doesn't match, even if the restart * digest matches, enforce a restart rather than a reload-agent anyway. * So that it ensures any changes of the extra parameters get applied * for this specific operation, and the digests calculated for the * resulting PCMK__XE_LRM_RSC_OP will be correct. * Preserve the implied rc pcmk__digest_restart for the case that the * main digest doesn't match. */ if ((interval_ms == 0) && (data->rc == pcmk__digest_restart)) { pcmk__rsc_info(rsc, "Parameters containing extra ones to %ums-interval" " %s action for %s on %s " "changed: hash was %s vs. now %s (restart:%s) %s", interval_ms, task, rsc->id, pcmk__node_name(node), pcmk__s(digest_all, "missing"), data->digest_all_calc, op_version, pcmk__xe_get(xml_op, PCMK__XA_TRANSITION_MAGIC)); } else { pcmk__rsc_info(rsc, "Parameters to %ums-interval %s action for %s on %s " "changed: hash was %s vs. now %s (%s:%s) %s", interval_ms, task, rsc->id, pcmk__node_name(node), pcmk__s(digest_all, "missing"), data->digest_all_calc, (interval_ms > 0)? "reschedule" : "reload", op_version, pcmk__xe_get(xml_op, PCMK__XA_TRANSITION_MAGIC)); data->rc = pcmk__digest_mismatch; } } else { data->rc = pcmk__digest_match; } return data; } /*! * \internal * \brief Create an unfencing summary for use in special node attribute * * Create a string combining a fence device's resource ID, agent type, and * parameter digest (whether for all parameters or just non-private parameters). * This can be stored in a special node attribute, allowing us to detect changes * in either the agent type or parameters, to know whether unfencing must be * redone or can be safely skipped when the device's history is cleaned. * * \param[in] rsc_id Fence device resource ID * \param[in] agent_type Fence device agent * \param[in] param_digest Fence device parameter digest * * \return Newly allocated string with unfencing digest * \note The caller is responsible for freeing the result. */ static inline char * create_unfencing_summary(const char *rsc_id, const char *agent_type, const char *param_digest) { return crm_strdup_printf("%s:%s:%s", rsc_id, agent_type, param_digest); } /*! * \internal * \brief Check whether a node can skip unfencing * * Check whether a fence device's current definition matches a node's * stored summary of when it was last unfenced by the device. * * \param[in] rsc_id Fence device's resource ID * \param[in] agent Fence device's agent type * \param[in] digest_calc Fence device's current parameter digest * \param[in] node_summary Value of node's special unfencing node attribute * (a comma-separated list of unfencing summaries for * all devices that have unfenced this node) * * \return TRUE if digest matches, FALSE otherwise */ static bool unfencing_digest_matches(const char *rsc_id, const char *agent, const char *digest_calc, const char *node_summary) { bool matches = FALSE; if (rsc_id && agent && digest_calc && node_summary) { char *search_secure = create_unfencing_summary(rsc_id, agent, digest_calc); /* The digest was calculated including the device ID and agent, * so there is no risk of collision using strstr(). */ matches = (strstr(node_summary, search_secure) != NULL); crm_trace("Calculated unfencing digest '%s' %sfound in '%s'", search_secure, matches? "" : "not ", node_summary); free(search_secure); } return matches; } /* Magic string to use as action name for digest cache entries used for * unfencing checks. This is not a real action name (i.e. "on"), so * pcmk__check_action_config() won't confuse these entries with real actions. */ #define STONITH_DIGEST_TASK "stonith-on" /*! * \internal * \brief Calculate fence device digests and digest comparison result * * \param[in,out] rsc Fence device resource * \param[in] agent Fence device's agent type * \param[in,out] node Node with digest cache to use * \param[in,out] scheduler Scheduler data * * \return Node's digest cache entry */ pcmk__op_digest_t * pe__compare_fencing_digest(pcmk_resource_t *rsc, const char *agent, pcmk_node_t *node, pcmk_scheduler_t *scheduler) { const char *node_summary = NULL; // Calculate device's current parameter digests pcmk__op_digest_t *data = rsc_action_digest(rsc, STONITH_DIGEST_TASK, 0U, node, NULL, TRUE, scheduler); // Check whether node has special unfencing summary node attribute node_summary = pcmk__node_attr(node, CRM_ATTR_DIGESTS_ALL, NULL, pcmk__rsc_node_current); if (node_summary == NULL) { data->rc = pcmk__digest_unknown; return data; } // Check whether full parameter digest matches if (unfencing_digest_matches(rsc->id, agent, data->digest_all_calc, node_summary)) { data->rc = pcmk__digest_match; return data; } // Check whether secure parameter digest matches node_summary = pcmk__node_attr(node, CRM_ATTR_DIGESTS_SECURE, NULL, pcmk__rsc_node_current); if (unfencing_digest_matches(rsc->id, agent, data->digest_secure_calc, node_summary)) { data->rc = pcmk__digest_match; if (!pcmk__is_daemon && (scheduler->priv->out != NULL)) { pcmk__output_t *out = scheduler->priv->out; out->info(out, "Only 'private' parameters to %s " "for unfencing %s changed", rsc->id, pcmk__node_name(node)); } return data; } // Parameters don't match data->rc = pcmk__digest_mismatch; if (pcmk_is_set(scheduler->flags, pcmk__sched_sanitized) && (data->digest_secure_calc != NULL)) { if (scheduler->priv->out != NULL) { pcmk__output_t *out = scheduler->priv->out; char *digest = create_unfencing_summary(rsc->id, agent, data->digest_secure_calc); out->info(out, "Parameters to %s for unfencing " "%s changed, try '%s'", rsc->id, pcmk__node_name(node), digest); free(digest); } else if (!pcmk__is_daemon) { char *digest = create_unfencing_summary(rsc->id, agent, data->digest_secure_calc); printf("Parameters to %s for unfencing %s changed, try '%s'\n", rsc->id, pcmk__node_name(node), digest); free(digest); } } return data; } diff --git a/lib/pengine/utils.c b/lib/pengine/utils.c index ea3efa6f77..c65672b3e6 100644 --- a/lib/pengine/utils.c +++ b/lib/pengine/utils.c @@ -1,924 +1,922 @@ /* * Copyright 2004-2025 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 "pe_status_private.h" -extern bool pcmk__is_daemon; - gboolean ghash_free_str_str(gpointer key, gpointer value, gpointer user_data); /*! * \internal * \brief Check whether we can fence a particular node * * \param[in] scheduler Scheduler data * \param[in] node Name of node to check * * \return true if node can be fenced, false otherwise */ bool pe_can_fence(const pcmk_scheduler_t *scheduler, const pcmk_node_t *node) { if (pcmk__is_guest_or_bundle_node(node)) { /* A guest or bundle node is fenced by stopping its launcher, which is * possible if the launcher's host is either online or fenceable. */ pcmk_resource_t *rsc = node->priv->remote->priv->launcher; for (GList *n = rsc->priv->active_nodes; n != NULL; n = n->next) { pcmk_node_t *launcher_node = n->data; if (!launcher_node->details->online && !pe_can_fence(scheduler, launcher_node)) { return false; } } return true; } else if (!pcmk_is_set(scheduler->flags, pcmk__sched_fencing_enabled)) { return false; /* Turned off */ } else if (!pcmk_is_set(scheduler->flags, pcmk__sched_have_fencing)) { return false; /* No devices */ } else if (pcmk_is_set(scheduler->flags, pcmk__sched_quorate)) { return true; } else if (scheduler->no_quorum_policy == pcmk_no_quorum_ignore) { return true; } else if (node == NULL) { return false; } else if (node->details->online) { /* Remote nodes are marked online when we assign their resource to a * node, not when they are actually started (see remote_connection_assigned) * so the above test by itself isn't good enough. */ if (pcmk__is_pacemaker_remote_node(node) && !pcmk_is_set(scheduler->flags, pcmk__sched_fence_remote_no_quorum)) { /* If we're on a system without quorum, it's entirely possible that * the remote resource was automatically moved to a node on the * partition with quorum. We can't tell that from this node - the * best we can do is check if it's possible for the resource to run * on another node in the partition with quorum. If so, it has * likely been moved and we shouldn't fence it. * * NOTE: This condition appears to only come up in very limited * circumstances. It at least requires some very lengthy fencing * timeouts set, some way for fencing to still take place (a second * NIC is how I've reproduced it in testing, but fence_scsi or * sbd could work too), and a resource that runs on the remote node. */ pcmk_resource_t *rsc = node->priv->remote; pcmk_node_t *n = NULL; GHashTableIter iter; g_hash_table_iter_init(&iter, rsc->priv->allowed_nodes); while (g_hash_table_iter_next(&iter, NULL, (void **) &n)) { /* A node that's not online according to this non-quorum node * is a node that's in another partition. */ if (!n->details->online) { return false; } } } crm_notice("We can fence %s without quorum because they're in our membership", pcmk__node_name(node)); return true; } crm_trace("Cannot fence %s", pcmk__node_name(node)); return false; } /*! * \internal * \brief Copy a node object * * \param[in] this_node Node object to copy * * \return Newly allocated shallow copy of this_node * \note This function asserts on errors and is guaranteed to return non-NULL. * The caller is responsible for freeing the result using * pcmk__free_node_copy(). */ pcmk_node_t * pe__copy_node(const pcmk_node_t *this_node) { pcmk_node_t *new_node = NULL; pcmk__assert(this_node != NULL); new_node = pcmk__assert_alloc(1, sizeof(pcmk_node_t)); new_node->assign = pcmk__assert_alloc(1, sizeof(struct pcmk__node_assignment)); new_node->assign->probe_mode = this_node->assign->probe_mode; new_node->assign->score = this_node->assign->score; new_node->assign->count = this_node->assign->count; new_node->details = this_node->details; new_node->priv = this_node->priv; return new_node; } /*! * \internal * \brief Create a hash table of node copies from a list of nodes * * \param[in] list Node list * * \return Hash table equivalent of node list */ GHashTable * pe__node_list2table(const GList *list) { GHashTable *result = NULL; result = pcmk__strkey_table(NULL, pcmk__free_node_copy); for (const GList *gIter = list; gIter != NULL; gIter = gIter->next) { pcmk_node_t *new_node = NULL; new_node = pe__copy_node((const pcmk_node_t *) gIter->data); g_hash_table_insert(result, (gpointer) new_node->priv->id, new_node); } return result; } /*! * \internal * \brief Compare two nodes by name, with numeric portions sorted numerically * * Sort two node names case-insensitively like strcasecmp(), but with any * numeric portions of the name sorted numerically. For example, "node10" will * sort higher than "node9" but lower than "remotenode9". * * \param[in] a First node to compare (can be \c NULL) * \param[in] b Second node to compare (can be \c NULL) * * \retval -1 \c a comes before \c b (or \c a is \c NULL and \c b is not) * \retval 0 \c a and \c b are equal (or both are \c NULL) * \retval 1 \c a comes after \c b (or \c b is \c NULL and \c a is not) */ gint pe__cmp_node_name(gconstpointer a, gconstpointer b) { const pcmk_node_t *node1 = (const pcmk_node_t *) a; const pcmk_node_t *node2 = (const pcmk_node_t *) b; if ((node1 == NULL) && (node2 == NULL)) { return 0; } if (node1 == NULL) { return -1; } if (node2 == NULL) { return 1; } return pcmk__numeric_strcasecmp(node1->priv->name, node2->priv->name); } /*! * \internal * \brief Output node weights to stdout * * \param[in] rsc Use allowed nodes for this resource * \param[in] comment Text description to prefix lines with * \param[in] nodes If rsc is not specified, use these nodes * \param[in,out] scheduler Scheduler data */ static void pe__output_node_weights(const pcmk_resource_t *rsc, const char *comment, GHashTable *nodes, pcmk_scheduler_t *scheduler) { pcmk__output_t *out = scheduler->priv->out; // Sort the nodes so the output is consistent for regression tests GList *list = g_list_sort(g_hash_table_get_values(nodes), pe__cmp_node_name); for (const GList *gIter = list; gIter != NULL; gIter = gIter->next) { const pcmk_node_t *node = (const pcmk_node_t *) gIter->data; out->message(out, "node-weight", rsc, comment, node->priv->name, pcmk_readable_score(node->assign->score)); } g_list_free(list); } /*! * \internal * \brief Log node weights at trace level * * \param[in] file Caller's filename * \param[in] function Caller's function name * \param[in] line Caller's line number * \param[in] rsc If not NULL, include this resource's ID in logs * \param[in] comment Text description to prefix lines with * \param[in] nodes Nodes whose scores should be logged */ static void pe__log_node_weights(const char *file, const char *function, int line, const pcmk_resource_t *rsc, const char *comment, GHashTable *nodes) { GHashTableIter iter; pcmk_node_t *node = NULL; // Don't waste time if we're not tracing at this point pcmk__if_tracing({}, return); g_hash_table_iter_init(&iter, nodes); while (g_hash_table_iter_next(&iter, NULL, (void **) &node)) { if (rsc) { qb_log_from_external_source(function, file, "%s: %s allocation score on %s: %s", LOG_TRACE, line, 0, comment, rsc->id, pcmk__node_name(node), pcmk_readable_score(node->assign->score)); } else { qb_log_from_external_source(function, file, "%s: %s = %s", LOG_TRACE, line, 0, comment, pcmk__node_name(node), pcmk_readable_score(node->assign->score)); } } } /*! * \internal * \brief Log or output node weights * * \param[in] file Caller's filename * \param[in] function Caller's function name * \param[in] line Caller's line number * \param[in] to_log Log if true, otherwise output * \param[in] rsc If not NULL, use this resource's ID in logs, * and show scores recursively for any children * \param[in] comment Text description to prefix lines with * \param[in] nodes Nodes whose scores should be shown * \param[in,out] scheduler Scheduler data */ void pe__show_node_scores_as(const char *file, const char *function, int line, bool to_log, const pcmk_resource_t *rsc, const char *comment, GHashTable *nodes, pcmk_scheduler_t *scheduler) { if ((rsc != NULL) && pcmk_is_set(rsc->flags, pcmk__rsc_removed)) { // Don't show allocation scores for orphans return; } if (nodes == NULL) { // Nothing to show return; } if (to_log) { pe__log_node_weights(file, function, line, rsc, comment, nodes); } else { pe__output_node_weights(rsc, comment, nodes, scheduler); } if (rsc == NULL) { return; } // If this resource has children, repeat recursively for each for (GList *gIter = rsc->priv->children; gIter != NULL; gIter = gIter->next) { pcmk_resource_t *child = (pcmk_resource_t *) gIter->data; pe__show_node_scores_as(file, function, line, to_log, child, comment, child->priv->allowed_nodes, scheduler); } } /*! * \internal * \brief Compare two resources by priority * * \param[in] a First resource to compare (can be \c NULL) * \param[in] b Second resource to compare (can be \c NULL) * * \retval -1 a's priority > b's priority (or \c b is \c NULL and \c a is not) * \retval 0 a's priority == b's priority (or both \c a and \c b are \c NULL) * \retval 1 a's priority < b's priority (or \c a is \c NULL and \c b is not) */ gint pe__cmp_rsc_priority(gconstpointer a, gconstpointer b) { const pcmk_resource_t *resource1 = (const pcmk_resource_t *)a; const pcmk_resource_t *resource2 = (const pcmk_resource_t *)b; if (a == NULL && b == NULL) { return 0; } if (a == NULL) { return 1; } if (b == NULL) { return -1; } if (resource1->priv->priority > resource2->priv->priority) { return -1; } if (resource1->priv->priority < resource2->priv->priority) { return 1; } return 0; } static void resource_node_score(pcmk_resource_t *rsc, const pcmk_node_t *node, int score, const char *tag) { pcmk_node_t *match = NULL; if ((pcmk_is_set(rsc->flags, pcmk__rsc_exclusive_probes) || (node->assign->probe_mode == pcmk__probe_never)) && pcmk__str_eq(tag, "symmetric_default", pcmk__str_casei)) { /* This string comparision may be fragile, but exclusive resources and * exclusive nodes should not have the symmetric_default constraint * applied to them. */ return; } else { for (GList *gIter = rsc->priv->children; gIter != NULL; gIter = gIter->next) { pcmk_resource_t *child_rsc = (pcmk_resource_t *) gIter->data; resource_node_score(child_rsc, node, score, tag); } } match = g_hash_table_lookup(rsc->priv->allowed_nodes, node->priv->id); if (match == NULL) { match = pe__copy_node(node); g_hash_table_insert(rsc->priv->allowed_nodes, (gpointer) match->priv->id, match); } match->assign->score = pcmk__add_scores(match->assign->score, score); pcmk__rsc_trace(rsc, "Enabling %s preference (%s) for %s on %s (now %s)", tag, pcmk_readable_score(score), rsc->id, pcmk__node_name(node), pcmk_readable_score(match->assign->score)); } void resource_location(pcmk_resource_t *rsc, const pcmk_node_t *node, int score, const char *tag, pcmk_scheduler_t *scheduler) { if (node != NULL) { resource_node_score(rsc, node, score, tag); } else if (scheduler != NULL) { GList *gIter = scheduler->nodes; for (; gIter != NULL; gIter = gIter->next) { pcmk_node_t *node_iter = (pcmk_node_t *) gIter->data; resource_node_score(rsc, node_iter, score, tag); } } else { GHashTableIter iter; pcmk_node_t *node_iter = NULL; g_hash_table_iter_init(&iter, rsc->priv->allowed_nodes); while (g_hash_table_iter_next(&iter, NULL, (void **)&node_iter)) { resource_node_score(rsc, node_iter, score, tag); } } if ((node == NULL) && (score == -PCMK_SCORE_INFINITY) && (rsc->priv->assigned_node != NULL)) { // @TODO Should this be more like pcmk__unassign_resource()? crm_info("Unassigning %s from %s", rsc->id, pcmk__node_name(rsc->priv->assigned_node)); pcmk__free_node_copy(rsc->priv->assigned_node); rsc->priv->assigned_node = NULL; } } gboolean get_target_role(const pcmk_resource_t *rsc, enum rsc_role_e *role) { enum rsc_role_e local_role = pcmk_role_unknown; const char *value = g_hash_table_lookup(rsc->priv->meta, PCMK_META_TARGET_ROLE); CRM_CHECK(role != NULL, return FALSE); if (pcmk__str_eq(value, PCMK_ROLE_STARTED, pcmk__str_null_matches|pcmk__str_casei)) { return FALSE; } if (pcmk__str_eq(PCMK_VALUE_DEFAULT, value, pcmk__str_casei)) { // @COMPAT Deprecated since 2.1.8 pcmk__config_warn("Support for setting " PCMK_META_TARGET_ROLE " to the explicit value '" PCMK_VALUE_DEFAULT "' is deprecated and will be removed in a " "future release (just leave it unset)"); return FALSE; } local_role = pcmk_parse_role(value); if (local_role == pcmk_role_unknown) { pcmk__config_err("Ignoring '" PCMK_META_TARGET_ROLE "' for %s " "because '%s' is not valid", rsc->id, value); return FALSE; } else if (local_role > pcmk_role_started) { if (pcmk_is_set(pe__const_top_resource(rsc, false)->flags, pcmk__rsc_promotable)) { if (local_role > pcmk_role_unpromoted) { /* This is what we'd do anyway, just leave the default to avoid messing up the placement algorithm */ return FALSE; } } else { pcmk__config_err("Ignoring '" PCMK_META_TARGET_ROLE "' for %s " "because '%s' only makes sense for promotable " "clones", rsc->id, value); return FALSE; } } *role = local_role; return TRUE; } gboolean order_actions(pcmk_action_t *first, pcmk_action_t *then, uint32_t flags) { GList *gIter = NULL; pcmk__related_action_t *wrapper = NULL; GList *list = NULL; if (flags == pcmk__ar_none) { return FALSE; } if ((first == NULL) || (then == NULL)) { return FALSE; } crm_trace("Creating action wrappers for ordering: %s then %s", first->uuid, then->uuid); /* Ensure we never create a dependency on ourselves... it's happened */ pcmk__assert(first != then); /* Filter dups, otherwise update_action_states() has too much work to do */ gIter = first->actions_after; for (; gIter != NULL; gIter = gIter->next) { pcmk__related_action_t *after = gIter->data; if ((after->action == then) && pcmk_any_flags_set(after->flags, flags)) { return FALSE; } } wrapper = pcmk__assert_alloc(1, sizeof(pcmk__related_action_t)); wrapper->action = then; wrapper->flags = flags; list = first->actions_after; list = g_list_prepend(list, wrapper); first->actions_after = list; wrapper = pcmk__assert_alloc(1, sizeof(pcmk__related_action_t)); wrapper->action = first; wrapper->flags = flags; list = then->actions_before; list = g_list_prepend(list, wrapper); then->actions_before = list; return TRUE; } void destroy_ticket(gpointer data) { pcmk__ticket_t *ticket = data; if (ticket->state) { g_hash_table_destroy(ticket->state); } free(ticket->id); free(ticket); } pcmk__ticket_t * ticket_new(const char *ticket_id, pcmk_scheduler_t *scheduler) { pcmk__ticket_t *ticket = NULL; if (pcmk__str_empty(ticket_id)) { return NULL; } if (scheduler->priv->ticket_constraints == NULL) { scheduler->priv->ticket_constraints = pcmk__strkey_table(free, destroy_ticket); } ticket = g_hash_table_lookup(scheduler->priv->ticket_constraints, ticket_id); if (ticket == NULL) { ticket = calloc(1, sizeof(pcmk__ticket_t)); if (ticket == NULL) { pcmk__sched_err(scheduler, "Cannot allocate ticket '%s'", ticket_id); return NULL; } crm_trace("Creating ticket entry for %s", ticket_id); ticket->id = strdup(ticket_id); ticket->last_granted = -1; ticket->state = pcmk__strkey_table(free, free); g_hash_table_insert(scheduler->priv->ticket_constraints, pcmk__str_copy(ticket->id), ticket); } return ticket; } const char * rsc_printable_id(const pcmk_resource_t *rsc) { if (pcmk_is_set(rsc->flags, pcmk__rsc_unique)) { return rsc->id; } return pcmk__xe_id(rsc->priv->xml); } void pe__clear_resource_flags_recursive(pcmk_resource_t *rsc, uint64_t flags) { pcmk__clear_rsc_flags(rsc, flags); for (GList *gIter = rsc->priv->children; gIter != NULL; gIter = gIter->next) { pe__clear_resource_flags_recursive((pcmk_resource_t *) gIter->data, flags); } } void pe__clear_resource_flags_on_all(pcmk_scheduler_t *scheduler, uint64_t flag) { for (GList *lpc = scheduler->priv->resources; lpc != NULL; lpc = lpc->next) { pcmk_resource_t *r = (pcmk_resource_t *) lpc->data; pe__clear_resource_flags_recursive(r, flag); } } void pe__set_resource_flags_recursive(pcmk_resource_t *rsc, uint64_t flags) { pcmk__set_rsc_flags(rsc, flags); for (GList *gIter = rsc->priv->children; gIter != NULL; gIter = gIter->next) { pe__set_resource_flags_recursive((pcmk_resource_t *) gIter->data, flags); } } void trigger_unfencing(pcmk_resource_t *rsc, pcmk_node_t *node, const char *reason, pcmk_action_t *dependency, pcmk_scheduler_t *scheduler) { if (!pcmk_is_set(scheduler->flags, pcmk__sched_enable_unfencing)) { /* No resources require it */ return; } else if ((rsc != NULL) && !pcmk_is_set(rsc->flags, pcmk__rsc_fence_device)) { /* Wasn't a stonith device */ return; } else if(node && node->details->online && node->details->unclean == FALSE && node->details->shutdown == FALSE) { pcmk_action_t *unfence = pe_fence_op(node, PCMK_ACTION_ON, FALSE, reason, FALSE, scheduler); if(dependency) { order_actions(unfence, dependency, pcmk__ar_ordered); } } else if(rsc) { GHashTableIter iter; g_hash_table_iter_init(&iter, rsc->priv->allowed_nodes); while (g_hash_table_iter_next(&iter, NULL, (void **)&node)) { if(node->details->online && node->details->unclean == FALSE && node->details->shutdown == FALSE) { trigger_unfencing(rsc, node, reason, dependency, scheduler); } } } } /*! * \internal * \brief Check whether shutdown has been requested for a node * * \param[in] node Node to check * * \return TRUE if node has shutdown attribute set and nonzero, FALSE otherwise * \note This differs from simply using node->details->shutdown in that it can * be used before that has been determined (and in fact to determine it), * and it can also be used to distinguish requested shutdown from implicit * shutdown of remote nodes by virtue of their connection stopping. */ bool pe__shutdown_requested(const pcmk_node_t *node) { const char *shutdown = pcmk__node_attr(node, PCMK__NODE_ATTR_SHUTDOWN, NULL, pcmk__rsc_node_current); return !pcmk__str_eq(shutdown, "0", pcmk__str_null_matches); } /*! * \internal * \brief Extract nvpair blocks contained by a CIB XML element into a hash table * * \param[in] xml_obj XML element containing blocks of nvpair elements * \param[in] set_name If not NULL, only use blocks of this element * \param[in] rule_input Values used to evaluate rule criteria * (node_attrs member must be NULL if \p set_name * is PCMK_XE_META_ATTRIBUTES) * \param[out] hash Where to store extracted name/value pairs * \param[in] always_first If not NULL, process block with this ID first * \param[in,out] scheduler Scheduler data containing \p xml_obj */ void pe__unpack_dataset_nvpairs(const xmlNode *xml_obj, const char *set_name, const pcmk_rule_input_t *rule_input, GHashTable *hash, const char *always_first, pcmk_scheduler_t *scheduler) { crm_time_t *next_change = NULL; CRM_CHECK((set_name != NULL) && (rule_input != NULL) && (hash != NULL) && (scheduler != NULL), return); // Node attribute expressions are not allowed for meta-attributes CRM_CHECK((rule_input->node_attrs == NULL) || (strcmp(set_name, PCMK_XE_META_ATTRIBUTES) != 0), return); if (xml_obj == NULL) { return; } next_change = crm_time_new_undefined(); pcmk_unpack_nvpair_blocks(xml_obj, set_name, always_first, rule_input, hash, next_change); if (crm_time_is_defined(next_change)) { time_t recheck = (time_t) crm_time_get_seconds_since_epoch(next_change); pcmk__update_recheck_time(recheck, scheduler, "rule evaluation"); } crm_time_free(next_change); } bool pe__resource_is_disabled(const pcmk_resource_t *rsc) { const char *target_role = NULL; CRM_CHECK(rsc != NULL, return false); target_role = g_hash_table_lookup(rsc->priv->meta, PCMK_META_TARGET_ROLE); if (target_role) { // If invalid, we've already logged an error when unpacking enum rsc_role_e target_role_e = pcmk_parse_role(target_role); if ((target_role_e == pcmk_role_stopped) || ((target_role_e == pcmk_role_unpromoted) && pcmk_is_set(pe__const_top_resource(rsc, false)->flags, pcmk__rsc_promotable))) { return true; } } return false; } /*! * \internal * \brief Check whether a resource is running only on given node * * \param[in] rsc Resource to check * \param[in] node Node to check * * \return true if \p rsc is running only on \p node, otherwise false */ bool pe__rsc_running_on_only(const pcmk_resource_t *rsc, const pcmk_node_t *node) { return (rsc != NULL) && pcmk__list_of_1(rsc->priv->active_nodes) && pcmk__same_node((const pcmk_node_t *) rsc->priv->active_nodes->data, node); } bool pe__rsc_running_on_any(pcmk_resource_t *rsc, GList *node_list) { if (rsc != NULL) { for (GList *ele = rsc->priv->active_nodes; ele; ele = ele->next) { pcmk_node_t *node = (pcmk_node_t *) ele->data; if (pcmk__str_in_list(node->priv->name, node_list, pcmk__str_star_matches|pcmk__str_casei)) { return true; } } } return false; } bool pcmk__rsc_filtered_by_node(pcmk_resource_t *rsc, GList *only_node) { return rsc->priv->fns->active(rsc, false) && !pe__rsc_running_on_any(rsc, only_node); } GList * pe__filter_rsc_list(GList *rscs, GList *filter) { GList *retval = NULL; for (GList *gIter = rscs; gIter; gIter = gIter->next) { pcmk_resource_t *rsc = (pcmk_resource_t *) gIter->data; /* I think the second condition is safe here for all callers of this * function. If not, it needs to move into pe__node_text. */ if (pcmk__str_in_list(rsc_printable_id(rsc), filter, pcmk__str_star_matches) || ((rsc->priv->parent != NULL) && pcmk__str_in_list(rsc_printable_id(rsc->priv->parent), filter, pcmk__str_star_matches))) { retval = g_list_prepend(retval, rsc); } } return retval; } GList * pe__build_node_name_list(pcmk_scheduler_t *scheduler, const char *s) { GList *nodes = NULL; if (pcmk__str_eq(s, "*", pcmk__str_null_matches)) { /* Nothing was given so return a list of all node names. Or, '*' was * given. This would normally fall into the pe__unames_with_tag branch * where it will return an empty list. Catch it here instead. */ nodes = g_list_prepend(nodes, strdup("*")); } else { pcmk_node_t *node = pcmk_find_node(scheduler, s); if (node) { /* The given string was a valid uname for a node. Return a * singleton list containing just that uname. */ nodes = g_list_prepend(nodes, strdup(s)); } else { /* The given string was not a valid uname. It's either a tag or * it's a typo or something. In the first case, we'll return a * list of all the unames of the nodes with the given tag. In the * second case, we'll return a NULL pointer and nothing will * get displayed. */ nodes = pe__unames_with_tag(scheduler, s); } } return nodes; } GList * pe__build_rsc_list(pcmk_scheduler_t *scheduler, const char *s) { GList *resources = NULL; if (pcmk__str_eq(s, "*", pcmk__str_null_matches)) { resources = g_list_prepend(resources, strdup("*")); } else { const uint32_t flags = pcmk_rsc_match_history|pcmk_rsc_match_basename; pcmk_resource_t *rsc = pe_find_resource_with_flags(scheduler->priv->resources, s, flags); if (rsc) { /* A colon in the name we were given means we're being asked to filter * on a specific instance of a cloned resource. Put that exact string * into the filter list. Otherwise, use the printable ID of whatever * resource was found that matches what was asked for. */ if (strstr(s, ":") != NULL) { resources = g_list_prepend(resources, strdup(rsc->id)); } else { resources = g_list_prepend(resources, strdup(rsc_printable_id(rsc))); } } else { /* The given string was not a valid resource name. It's a tag or a * typo or something. See pe__build_node_name_list() for more * detail. */ resources = pe__rscs_with_tag(scheduler, s); } } return resources; } xmlNode * pe__failed_probe_for_rsc(const pcmk_resource_t *rsc, const char *name) { const pcmk_resource_t *parent = pe__const_top_resource(rsc, false); const char *rsc_id = rsc->id; const pcmk_scheduler_t *scheduler = rsc->priv->scheduler; if (pcmk__is_clone(parent)) { rsc_id = pe__clone_child_id(parent); } for (xmlNode *xml_op = pcmk__xe_first_child(scheduler->priv->failed, NULL, NULL, NULL); xml_op != NULL; xml_op = pcmk__xe_next(xml_op, NULL)) { const char *value = NULL; char *op_id = NULL; /* This resource operation is not a failed probe. */ if (!pcmk_xe_mask_probe_failure(xml_op)) { continue; } /* This resource operation was not run on the given node. Note that if name is * NULL, this will always succeed. */ value = pcmk__xe_get(xml_op, PCMK__META_ON_NODE); if (value == NULL || !pcmk__str_eq(value, name, pcmk__str_casei|pcmk__str_null_matches)) { continue; } if (!parse_op_key(pcmk__xe_history_key(xml_op), &op_id, NULL, NULL)) { continue; // This history entry is missing an operation key } /* This resource operation's ID does not match the rsc_id we are looking for. */ if (!pcmk__str_eq(op_id, rsc_id, pcmk__str_none)) { free(op_id); continue; } free(op_id); return xml_op; } return NULL; } diff --git a/lib/services/services_linux.c b/lib/services/services_linux.c index 2666e84a18..fcd979734b 100644 --- a/lib/services/services_linux.c +++ b/lib/services/services_linux.c @@ -1,1489 +1,1505 @@ /* * Copyright 2010-2025 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 #include #include "crm/crm.h" #include "crm/common/mainloop.h" #include "crm/services.h" #include "crm/services_internal.h" #include "services_private.h" static void close_pipe(int fildes[]); /* We have two alternative ways of handling SIGCHLD when synchronously waiting * for spawned processes to complete. Both rely on polling a file descriptor to * discover SIGCHLD events. * * If sys/signalfd.h is available (e.g. on Linux), we call signalfd() to * generate the file descriptor. Otherwise, we use the "self-pipe trick" * (opening a pipe and writing a byte to it when SIGCHLD is received). */ #ifdef HAVE_SYS_SIGNALFD_H // signalfd() implementation #include // Everything needed to manage SIGCHLD handling struct sigchld_data_s { sigset_t mask; // Signals to block now (including SIGCHLD) sigset_t old_mask; // Previous set of blocked signals bool ignored; // If SIGCHLD for another child has been ignored }; // Initialize SIGCHLD data and prepare for use static bool sigchld_setup(struct sigchld_data_s *data) { sigemptyset(&(data->mask)); sigaddset(&(data->mask), SIGCHLD); sigemptyset(&(data->old_mask)); // Block SIGCHLD (saving previous set of blocked signals to restore later) if (sigprocmask(SIG_BLOCK, &(data->mask), &(data->old_mask)) < 0) { crm_info("Wait for child process completion failed: %s " QB_XS " source=sigprocmask", pcmk_rc_str(errno)); return false; } data->ignored = false; return true; } // Get a file descriptor suitable for polling for SIGCHLD events static int sigchld_open(struct sigchld_data_s *data) { int fd; CRM_CHECK(data != NULL, return -1); fd = signalfd(-1, &(data->mask), SFD_NONBLOCK); if (fd < 0) { crm_info("Wait for child process completion failed: %s " QB_XS " source=signalfd", pcmk_rc_str(errno)); } return fd; } // Close a file descriptor returned by sigchld_open() static void sigchld_close(int fd) { if (fd > 0) { close(fd); } } // Return true if SIGCHLD was received from polled fd static bool sigchld_received(int fd, int pid, struct sigchld_data_s *data) { struct signalfd_siginfo fdsi; ssize_t s; if (fd < 0) { return false; } s = read(fd, &fdsi, sizeof(struct signalfd_siginfo)); if (s != sizeof(struct signalfd_siginfo)) { crm_info("Wait for child process completion failed: %s " QB_XS " source=read", pcmk_rc_str(errno)); } else if (fdsi.ssi_signo == SIGCHLD) { if (fdsi.ssi_pid == pid) { return true; } else { /* This SIGCHLD is for another child. We have to ignore it here but * will still need to resend it after this synchronous action has * completed and SIGCHLD has been restored to be handled by the * previous SIGCHLD handler, so that it will be handled. */ data->ignored = true; return false; } } return false; } // Do anything needed after done waiting for SIGCHLD static void sigchld_cleanup(struct sigchld_data_s *data) { // Restore the original set of blocked signals if ((sigismember(&(data->old_mask), SIGCHLD) == 0) && (sigprocmask(SIG_UNBLOCK, &(data->mask), NULL) < 0)) { crm_warn("Could not clean up after child process completion: %s", pcmk_rc_str(errno)); } // Resend any ignored SIGCHLD for other children so that they'll be handled. if (data->ignored && kill(getpid(), SIGCHLD) != 0) { crm_warn("Could not resend ignored SIGCHLD to ourselves: %s", pcmk_rc_str(errno)); } } #else // HAVE_SYS_SIGNALFD_H not defined // Self-pipe implementation (see above for function descriptions) struct sigchld_data_s { int pipe_fd[2]; // Pipe file descriptors struct sigaction sa; // Signal handling info (with SIGCHLD) struct sigaction old_sa; // Previous signal handling info bool ignored; // If SIGCHLD for another child has been ignored }; // We need a global to use in the signal handler volatile struct sigchld_data_s *last_sigchld_data = NULL; static void sigchld_handler(void) { // We received a SIGCHLD, so trigger pipe polling if ((last_sigchld_data != NULL) && (last_sigchld_data->pipe_fd[1] >= 0) && (write(last_sigchld_data->pipe_fd[1], "", 1) == -1)) { crm_info("Wait for child process completion failed: %s " QB_XS " source=write", pcmk_rc_str(errno)); } } static bool sigchld_setup(struct sigchld_data_s *data) { int rc; data->pipe_fd[0] = data->pipe_fd[1] = -1; if (pipe(data->pipe_fd) == -1) { crm_info("Wait for child process completion failed: %s " QB_XS " source=pipe", pcmk_rc_str(errno)); return false; } rc = pcmk__set_nonblocking(data->pipe_fd[0]); if (rc != pcmk_rc_ok) { crm_info("Could not set pipe input non-blocking: %s " QB_XS " rc=%d", pcmk_rc_str(rc), rc); } rc = pcmk__set_nonblocking(data->pipe_fd[1]); if (rc != pcmk_rc_ok) { crm_info("Could not set pipe output non-blocking: %s " QB_XS " rc=%d", pcmk_rc_str(rc), rc); } // Set SIGCHLD handler data->sa.sa_handler = (sighandler_t) sigchld_handler; data->sa.sa_flags = 0; sigemptyset(&(data->sa.sa_mask)); if (sigaction(SIGCHLD, &(data->sa), &(data->old_sa)) < 0) { crm_info("Wait for child process completion failed: %s " QB_XS " source=sigaction", pcmk_rc_str(errno)); } data->ignored = false; // Remember data for use in signal handler last_sigchld_data = data; return true; } static int sigchld_open(struct sigchld_data_s *data) { CRM_CHECK(data != NULL, return -1); return data->pipe_fd[0]; } static void sigchld_close(int fd) { // Pipe will be closed in sigchld_cleanup() return; } static bool sigchld_received(int fd, int pid, struct sigchld_data_s *data) { char ch; if (fd < 0) { return false; } // Clear out the self-pipe while (read(fd, &ch, 1) == 1) /*omit*/; return true; } static void sigchld_cleanup(struct sigchld_data_s *data) { // Restore the previous SIGCHLD handler if (sigaction(SIGCHLD, &(data->old_sa), NULL) < 0) { crm_warn("Could not clean up after child process completion: %s", pcmk_rc_str(errno)); } close_pipe(data->pipe_fd); // Resend any ignored SIGCHLD for other children so that they'll be handled. if (data->ignored && kill(getpid(), SIGCHLD) != 0) { crm_warn("Could not resend ignored SIGCHLD to ourselves: %s", pcmk_rc_str(errno)); } } #endif /*! * \internal * \brief Close the two file descriptors of a pipe * * \param[in,out] fildes Array of file descriptors opened by pipe() */ static void close_pipe(int fildes[]) { if (fildes[0] >= 0) { close(fildes[0]); fildes[0] = -1; } if (fildes[1] >= 0) { close(fildes[1]); fildes[1] = -1; } } #define out_type(is_stderr) ((is_stderr)? "stderr" : "stdout") // Maximum number of bytes of stdout or stderr we'll accept #define MAX_OUTPUT (10 * 1024 * 1024) static gboolean svc_read_output(int fd, svc_action_t * op, bool is_stderr) { char *data = NULL; ssize_t rc = 0; size_t len = 0; size_t discarded = 0; char buf[500]; static const size_t buf_read_len = sizeof(buf) - 1; if (fd < 0) { crm_trace("No fd for %s", op->id); return FALSE; } if (is_stderr && op->stderr_data) { len = strlen(op->stderr_data); data = op->stderr_data; crm_trace("Reading %s stderr into offset %zu", op->id, len); } else if (is_stderr == FALSE && op->stdout_data) { len = strlen(op->stdout_data); data = op->stdout_data; crm_trace("Reading %s stdout into offset %zu", op->id, len); } else { crm_trace("Reading %s %s", op->id, out_type(is_stderr)); } do { errno = 0; rc = read(fd, buf, buf_read_len); if (rc > 0) { if (len < MAX_OUTPUT) { buf[rc] = 0; crm_trace("Received %zd bytes of %s %s: %.80s", rc, op->id, out_type(is_stderr), buf); data = pcmk__realloc(data, len + rc + 1); strcpy(data + len, buf); len += rc; } else { discarded += rc; } } else if (errno != EINTR) { // Fatal error or EOF rc = 0; break; } } while ((rc == buf_read_len) || (rc < 0)); if (discarded > 0) { crm_warn("Truncated %s %s to %zu bytes (discarded %zu)", op->id, out_type(is_stderr), len, discarded); } if (is_stderr) { op->stderr_data = data; } else { op->stdout_data = data; } return rc != 0; } static int dispatch_stdout(gpointer userdata) { svc_action_t *op = (svc_action_t *) userdata; return svc_read_output(op->opaque->stdout_fd, op, FALSE); } static int dispatch_stderr(gpointer userdata) { svc_action_t *op = (svc_action_t *) userdata; return svc_read_output(op->opaque->stderr_fd, op, TRUE); } static void pipe_out_done(gpointer user_data) { svc_action_t *op = (svc_action_t *) user_data; crm_trace("%p", op); op->opaque->stdout_gsource = NULL; if (op->opaque->stdout_fd > STDOUT_FILENO) { close(op->opaque->stdout_fd); } op->opaque->stdout_fd = -1; } static void pipe_err_done(gpointer user_data) { svc_action_t *op = (svc_action_t *) user_data; op->opaque->stderr_gsource = NULL; if (op->opaque->stderr_fd > STDERR_FILENO) { close(op->opaque->stderr_fd); } op->opaque->stderr_fd = -1; } static struct mainloop_fd_callbacks stdout_callbacks = { .dispatch = dispatch_stdout, .destroy = pipe_out_done, }; static struct mainloop_fd_callbacks stderr_callbacks = { .dispatch = dispatch_stderr, .destroy = pipe_err_done, }; static void set_ocf_env(const char *key, const char *value, gpointer user_data) { + // @FIXME @COMPAT This seems like it should be a fatal error if (setenv(key, value, 1) != 0) { - crm_perror(LOG_ERR, "setenv failed for key:%s and value:%s", key, value); + int rc = errno; + + crm_err("setenv failed for key='%s' and value='%s': %s", + pcmk__s(key, ""), pcmk__s(value, ""), strerror(rc)); } } static void set_ocf_env_with_prefix(gpointer key, gpointer value, gpointer user_data) { const char *ckey = key; if (pcmk__str_eq(ckey, "OCF_CHECK_LEVEL", pcmk__str_none)) { set_ocf_env(ckey, value, user_data); } else { char *buffer = crm_strdup_printf("OCF_RESKEY_%s", ckey); set_ocf_env(buffer, value, user_data); free(buffer); } } static void set_alert_env(gpointer key, gpointer value, gpointer user_data) { int rc; if (value != NULL) { rc = setenv(key, value, 1); } else { rc = unsetenv(key); } if (rc < 0) { - crm_perror(LOG_ERR, "setenv %s=%s", - (char*)key, (value? (char*)value : "")); + // @FIXME @COMPAT This seems like it should be a fatal error + rc = errno; + + if (value != NULL) { + crm_err("setenv %s='%s' failed: %s", (const char *) key, + (const char *) value, strerror(rc)); + } else { + crm_err("unsetenv %s failed: %s", (const char *) key, strerror(rc)); + } + } else { - crm_trace("setenv %s=%s", (char*)key, (value? (char*)value : "")); + if (value != NULL) { + crm_trace("setenv %s='%s'", (const char *) key, (const char *) value); + } else { + crm_trace("unsetenv %s", (const char *) key); + } } } /*! * \internal * \brief Add environment variables suitable for an action * * \param[in] op Action to use */ static void add_action_env_vars(const svc_action_t *op) { void (*env_setter)(gpointer, gpointer, gpointer) = NULL; if (op->agent == NULL) { env_setter = set_alert_env; /* we deal with alert handler */ } else if (pcmk__str_eq(op->standard, PCMK_RESOURCE_CLASS_OCF, pcmk__str_casei)) { env_setter = set_ocf_env_with_prefix; } if (env_setter != NULL && op->params != NULL) { g_hash_table_foreach(op->params, env_setter, NULL); } if (env_setter == NULL || env_setter == set_alert_env) { return; } set_ocf_env("OCF_RA_VERSION_MAJOR", PCMK_OCF_MAJOR_VERSION, NULL); set_ocf_env("OCF_RA_VERSION_MINOR", PCMK_OCF_MINOR_VERSION, NULL); set_ocf_env("OCF_ROOT", PCMK_OCF_ROOT, NULL); set_ocf_env("OCF_EXIT_REASON_PREFIX", PCMK_OCF_REASON_PREFIX, NULL); if (op->rsc) { set_ocf_env("OCF_RESOURCE_INSTANCE", op->rsc, NULL); } if (op->agent != NULL) { set_ocf_env("OCF_RESOURCE_TYPE", op->agent, NULL); } /* Notes: this is not added to specification yet. Sept 10,2004 */ if (op->provider != NULL) { set_ocf_env("OCF_RESOURCE_PROVIDER", op->provider, NULL); } } static void pipe_in_single_parameter(gpointer key, gpointer value, gpointer user_data) { svc_action_t *op = user_data; char *buffer = crm_strdup_printf("%s=%s\n", (char *)key, (char *) value); size_t len = strlen(buffer); size_t total = 0; ssize_t ret = 0; do { errno = 0; ret = write(op->opaque->stdin_fd, buffer + total, len - total); if (ret > 0) { total += ret; } } while ((errno == EINTR) && (total < len)); free(buffer); } /*! * \internal * \brief Pipe parameters in via stdin for action * * \param[in] op Action to use */ static void pipe_in_action_stdin_parameters(const svc_action_t *op) { if (op->params) { g_hash_table_foreach(op->params, pipe_in_single_parameter, (gpointer) op); } } gboolean recurring_action_timer(gpointer data) { svc_action_t *op = data; crm_debug("Scheduling another invocation of %s", op->id); /* Clean out the old result */ free(op->stdout_data); op->stdout_data = NULL; free(op->stderr_data); op->stderr_data = NULL; op->opaque->repeat_timer = 0; services_action_async(op, NULL); return FALSE; } /*! * \internal * \brief Finalize handling of an asynchronous operation * * Given a completed asynchronous operation, cancel or reschedule it as * appropriate if recurring, call its callback if registered, stop tracking it, * and clean it up. * * \param[in,out] op Operation to finalize * * \return Standard Pacemaker return code * \retval EINVAL Caller supplied NULL or invalid \p op * \retval EBUSY Uncanceled recurring action has only been cleaned up * \retval pcmk_rc_ok Action has been freed * * \note If the return value is not pcmk_rc_ok, the caller is responsible for * freeing the action. */ int services__finalize_async_op(svc_action_t *op) { CRM_CHECK((op != NULL) && !(op->synchronous), return EINVAL); if (op->interval_ms != 0) { // Recurring operations must be either cancelled or rescheduled if (op->cancel) { services__set_cancelled(op); cancel_recurring_action(op); } else { op->opaque->repeat_timer = pcmk__create_timer(op->interval_ms, recurring_action_timer, op); } } if (op->opaque->callback != NULL) { op->opaque->callback(op); } // Stop tracking the operation (as in-flight or blocked) op->pid = 0; services_untrack_op(op); if ((op->interval_ms != 0) && !(op->cancel)) { // Do not free recurring actions (they will get freed when cancelled) services_action_cleanup(op); return EBUSY; } services_action_free(op); return pcmk_rc_ok; } static void close_op_input(svc_action_t *op) { if (op->opaque->stdin_fd >= 0) { close(op->opaque->stdin_fd); } } static void finish_op_output(svc_action_t *op, bool is_stderr) { mainloop_io_t **source; int fd; if (is_stderr) { source = &(op->opaque->stderr_gsource); fd = op->opaque->stderr_fd; } else { source = &(op->opaque->stdout_gsource); fd = op->opaque->stdout_fd; } if (op->synchronous || *source) { crm_trace("Finish reading %s[%d] %s", op->id, op->pid, (is_stderr? "stderr" : "stdout")); svc_read_output(fd, op, is_stderr); if (op->synchronous) { close(fd); } else { mainloop_del_fd(*source); *source = NULL; } } } // Log an operation's stdout and stderr static void log_op_output(svc_action_t *op) { char *prefix = crm_strdup_printf("%s[%d] error output", op->id, op->pid); /* The library caller has better context to know how important the output * is, so log it at info and debug severity here. They can log it again at * higher severity if appropriate. */ crm_log_output(LOG_INFO, prefix, op->stderr_data); strcpy(prefix + strlen(prefix) - strlen("error output"), "output"); crm_log_output(LOG_DEBUG, prefix, op->stdout_data); free(prefix); } // Truncate exit reasons at this many characters #define EXIT_REASON_MAX_LEN 128 static void parse_exit_reason_from_stderr(svc_action_t *op) { const char *reason_start = NULL; const char *reason_end = NULL; const int prefix_len = strlen(PCMK_OCF_REASON_PREFIX); if ((op->stderr_data == NULL) || // Only OCF agents have exit reasons in stderr !pcmk__str_eq(op->standard, PCMK_RESOURCE_CLASS_OCF, pcmk__str_none)) { return; } // Find the last occurrence of the magic string indicating an exit reason for (const char *cur = strstr(op->stderr_data, PCMK_OCF_REASON_PREFIX); cur != NULL; cur = strstr(cur, PCMK_OCF_REASON_PREFIX)) { cur += prefix_len; // Skip over magic string reason_start = cur; } if ((reason_start == NULL) || (reason_start[0] == '\n') || (reason_start[0] == '\0')) { return; // No or empty exit reason } // Exit reason goes to end of line (or end of output) reason_end = strchr(reason_start, '\n'); if (reason_end == NULL) { reason_end = reason_start + strlen(reason_start); } // Limit size of exit reason to something reasonable if (reason_end > (reason_start + EXIT_REASON_MAX_LEN)) { reason_end = reason_start + EXIT_REASON_MAX_LEN; } free(op->opaque->exit_reason); op->opaque->exit_reason = strndup(reason_start, reason_end - reason_start); } /*! * \internal * \brief Process the completion of an asynchronous child process * * \param[in,out] p Child process that completed * \param[in] pid Process ID of child * \param[in] core (Unused) * \param[in] signo Signal that interrupted child, if any * \param[in] exitcode Exit status of child process */ static void async_action_complete(mainloop_child_t *p, pid_t pid, int core, int signo, int exitcode) { svc_action_t *op = mainloop_child_userdata(p); mainloop_clear_child_userdata(p); CRM_CHECK(op->pid == pid, services__set_result(op, services__generic_error(op), PCMK_EXEC_ERROR, "Bug in mainloop handling"); return); /* Depending on the priority the mainloop gives the stdout and stderr * file descriptors, this function could be called before everything has * been read from them, so force a final read now. */ finish_op_output(op, true); finish_op_output(op, false); close_op_input(op); if (signo == 0) { crm_debug("%s[%d] exited with status %d", op->id, op->pid, exitcode); services__set_result(op, exitcode, PCMK_EXEC_DONE, NULL); log_op_output(op); parse_exit_reason_from_stderr(op); } else if (mainloop_child_timeout(p)) { const char *kind = services__action_kind(op); crm_info("%s %s[%d] timed out after %s", kind, op->id, op->pid, pcmk__readable_interval(op->timeout)); services__format_result(op, services__generic_error(op), PCMK_EXEC_TIMEOUT, "%s did not complete within %s", kind, pcmk__readable_interval(op->timeout)); } else if (op->cancel) { /* If an in-flight recurring operation was killed because it was * cancelled, don't treat that as a failure. */ crm_info("%s[%d] terminated with signal %d (%s)", op->id, op->pid, signo, strsignal(signo)); services__set_result(op, PCMK_OCF_OK, PCMK_EXEC_CANCELLED, NULL); } else { crm_info("%s[%d] terminated with signal %d (%s)", op->id, op->pid, signo, strsignal(signo)); services__format_result(op, PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_ERROR, "%s interrupted by %s signal", services__action_kind(op), strsignal(signo)); } services__finalize_async_op(op); } /*! * \internal * \brief Return agent standard's exit status for "generic error" * * When returning an internal error for an action, a value that is appropriate * to the action's agent standard must be used. This function returns a value * appropriate for errors in general. * * \param[in] op Action that error is for * * \return Exit status appropriate to agent standard * \note Actions without a standard will get PCMK_OCF_UNKNOWN_ERROR. */ int services__generic_error(const svc_action_t *op) { if ((op == NULL) || (op->standard == NULL)) { return PCMK_OCF_UNKNOWN_ERROR; } #if PCMK__ENABLE_LSB if (pcmk__str_eq(op->standard, PCMK_RESOURCE_CLASS_LSB, pcmk__str_casei) && pcmk__str_eq(op->action, PCMK_ACTION_STATUS, pcmk__str_casei)) { return PCMK_LSB_STATUS_UNKNOWN; } #endif return PCMK_OCF_UNKNOWN_ERROR; } /*! * \internal * \brief Return agent standard's exit status for "not installed" * * When returning an internal error for an action, a value that is appropriate * to the action's agent standard must be used. This function returns a value * appropriate for "not installed" errors. * * \param[in] op Action that error is for * * \return Exit status appropriate to agent standard * \note Actions without a standard will get PCMK_OCF_UNKNOWN_ERROR. */ int services__not_installed_error(const svc_action_t *op) { if ((op == NULL) || (op->standard == NULL)) { return PCMK_OCF_UNKNOWN_ERROR; } #if PCMK__ENABLE_LSB if (pcmk__str_eq(op->standard, PCMK_RESOURCE_CLASS_LSB, pcmk__str_casei) && pcmk__str_eq(op->action, PCMK_ACTION_STATUS, pcmk__str_casei)) { return PCMK_LSB_STATUS_NOT_INSTALLED; } #endif return PCMK_OCF_NOT_INSTALLED; } /*! * \internal * \brief Return agent standard's exit status for "insufficient privileges" * * When returning an internal error for an action, a value that is appropriate * to the action's agent standard must be used. This function returns a value * appropriate for "insufficient privileges" errors. * * \param[in] op Action that error is for * * \return Exit status appropriate to agent standard * \note Actions without a standard will get PCMK_OCF_UNKNOWN_ERROR. */ int services__authorization_error(const svc_action_t *op) { if ((op == NULL) || (op->standard == NULL)) { return PCMK_OCF_UNKNOWN_ERROR; } #if PCMK__ENABLE_LSB if (pcmk__str_eq(op->standard, PCMK_RESOURCE_CLASS_LSB, pcmk__str_casei) && pcmk__str_eq(op->action, PCMK_ACTION_STATUS, pcmk__str_casei)) { return PCMK_LSB_STATUS_INSUFFICIENT_PRIV; } #endif return PCMK_OCF_INSUFFICIENT_PRIV; } /*! * \internal * \brief Return agent standard's exit status for "not configured" * * When returning an internal error for an action, a value that is appropriate * to the action's agent standard must be used. This function returns a value * appropriate for "not configured" errors. * * \param[in] op Action that error is for * \param[in] is_fatal Whether problem is cluster-wide instead of only local * * \return Exit status appropriate to agent standard * \note Actions without a standard will get PCMK_OCF_UNKNOWN_ERROR. */ int services__configuration_error(const svc_action_t *op, bool is_fatal) { if ((op == NULL) || (op->standard == NULL)) { return PCMK_OCF_UNKNOWN_ERROR; } #if PCMK__ENABLE_LSB if (pcmk__str_eq(op->standard, PCMK_RESOURCE_CLASS_LSB, pcmk__str_casei) && pcmk__str_eq(op->action, PCMK_ACTION_STATUS, pcmk__str_casei)) { return PCMK_LSB_NOT_CONFIGURED; } #endif return is_fatal? PCMK_OCF_NOT_CONFIGURED : PCMK_OCF_INVALID_PARAM; } /*! * \internal * \brief Set operation rc and status per errno from stat(), fork() or execvp() * * \param[in,out] op Operation to set rc and status for * \param[in] error Value of errno after system call * * \return void */ void services__handle_exec_error(svc_action_t * op, int error) { const char *name = op->opaque->exec; if (name == NULL) { name = op->agent; if (name == NULL) { name = op->id; } } switch (error) { /* see execve(2), stat(2) and fork(2) */ case ENOENT: /* No such file or directory */ case EISDIR: /* Is a directory */ case ENOTDIR: /* Path component is not a directory */ case EINVAL: /* Invalid executable format */ case ENOEXEC: /* Invalid executable format */ services__format_result(op, services__not_installed_error(op), PCMK_EXEC_NOT_INSTALLED, "%s: %s", name, pcmk_rc_str(error)); break; case EACCES: /* permission denied (various errors) */ case EPERM: /* permission denied (various errors) */ services__format_result(op, services__authorization_error(op), PCMK_EXEC_ERROR, "%s: %s", name, pcmk_rc_str(error)); break; default: services__set_result(op, services__generic_error(op), PCMK_EXEC_ERROR, pcmk_rc_str(error)); } } /*! * \internal * \brief Exit a child process that failed before executing agent * * \param[in] op Action that failed * \param[in] exit_status Exit status code to use * \param[in] exit_reason Exit reason to output if for OCF agent */ static void exit_child(const svc_action_t *op, int exit_status, const char *exit_reason) { if ((op != NULL) && (exit_reason != NULL) && pcmk__str_eq(op->standard, PCMK_RESOURCE_CLASS_OCF, pcmk__str_none)) { fprintf(stderr, PCMK_OCF_REASON_PREFIX "%s\n", exit_reason); } pcmk_common_cleanup(); _exit(exit_status); } static void action_launch_child(svc_action_t *op) { int rc; /* SIGPIPE is ignored (which is different from signal blocking) by the gnutls library. * Depending on the libqb version in use, libqb may set SIGPIPE to be ignored as well. * We do not want this to be inherited by the child process. By resetting this the signal * to the default behavior, we avoid some potential odd problems that occur during OCF * scripts when SIGPIPE is ignored by the environment. */ signal(SIGPIPE, SIG_DFL); if (sched_getscheduler(0) != SCHED_OTHER) { struct sched_param sp; memset(&sp, 0, sizeof(sp)); sp.sched_priority = 0; if (sched_setscheduler(0, SCHED_OTHER, &sp) == -1) { crm_info("Could not reset scheduling policy for %s", op->id); } } if (setpriority(PRIO_PROCESS, 0, 0) == -1) { crm_info("Could not reset process priority for %s", op->id); } /* Man: The call setpgrp() is equivalent to setpgid(0,0) * _and_ compiles on BSD variants too * need to investigate if it works the same too. */ setpgid(0, 0); - pcmk__close_fds_in_child(false); + pcmk__close_fds_in_child(); /* It would be nice if errors in this function could be reported as * execution status (for example, PCMK_EXEC_NO_SECRETS for the secrets error * below) instead of exit status. However, we've already forked, so * exit status is all we have. At least for OCF actions, we can output an * exit reason for the parent to parse. * * @TODO It might be better to substitute secrets in the parent before * forking, so that if it fails, we can give a better message and result, * and avoid the fork. */ #if PCMK__ENABLE_CIBSECRETS rc = pcmk__substitute_secrets(op->rsc, op->params); if (rc != pcmk_rc_ok) { if (pcmk__str_eq(op->action, PCMK_ACTION_STOP, pcmk__str_casei)) { crm_info("Proceeding with stop operation for %s " "despite being unable to load CIB secrets (%s)", op->rsc, pcmk_rc_str(rc)); } else { crm_err("Considering %s unconfigured " "because unable to load CIB secrets: %s", op->rsc, pcmk_rc_str(rc)); exit_child(op, services__configuration_error(op, false), "Unable to load CIB secrets"); } } #endif add_action_env_vars(op); /* Become the desired user */ if (op->opaque->uid && (geteuid() == 0)) { // If requested, set effective group if (op->opaque->gid && (setgid(op->opaque->gid) < 0)) { crm_err("Considering %s unauthorized because could not set " "child group to %d: %s", op->id, op->opaque->gid, strerror(errno)); exit_child(op, services__authorization_error(op), "Could not set group for child process"); } // Erase supplementary group list // (We could do initgroups() if we kept a copy of the username) if (setgroups(0, NULL) < 0) { crm_err("Considering %s unauthorized because could not " "clear supplementary groups: %s", op->id, strerror(errno)); exit_child(op, services__authorization_error(op), "Could not clear supplementary groups for child process"); } // Set effective user if (setuid(op->opaque->uid) < 0) { crm_err("Considering %s unauthorized because could not set user " "to %d: %s", op->id, op->opaque->uid, strerror(errno)); exit_child(op, services__authorization_error(op), "Could not set user for child process"); } } // Execute the agent (doesn't return if successful) execvp(op->opaque->exec, op->opaque->args); // An earlier stat() should have avoided most possible errors rc = errno; services__handle_exec_error(op, rc); crm_err("Unable to execute %s: %s", op->id, strerror(rc)); exit_child(op, op->rc, "Child process was unable to execute file"); } /*! * \internal * \brief Wait for synchronous action to complete, and set its result * * \param[in,out] op Action to wait for * \param[in,out] data Child signal data */ static void wait_for_sync_result(svc_action_t *op, struct sigchld_data_s *data) { int status = 0; int timeout = op->timeout; time_t start = time(NULL); struct pollfd fds[3]; int wait_rc = 0; const char *wait_reason = NULL; fds[0].fd = op->opaque->stdout_fd; fds[0].events = POLLIN; fds[0].revents = 0; fds[1].fd = op->opaque->stderr_fd; fds[1].events = POLLIN; fds[1].revents = 0; fds[2].fd = sigchld_open(data); fds[2].events = POLLIN; fds[2].revents = 0; crm_trace("Waiting for %s[%d]", op->id, op->pid); do { int poll_rc = poll(fds, 3, timeout); wait_reason = NULL; if (poll_rc > 0) { if (fds[0].revents & POLLIN) { svc_read_output(op->opaque->stdout_fd, op, FALSE); } if (fds[1].revents & POLLIN) { svc_read_output(op->opaque->stderr_fd, op, TRUE); } if ((fds[2].revents & POLLIN) && sigchld_received(fds[2].fd, op->pid, data)) { wait_rc = waitpid(op->pid, &status, WNOHANG); if ((wait_rc > 0) || ((wait_rc < 0) && (errno == ECHILD))) { // Child process exited or doesn't exist break; } else if (wait_rc < 0) { wait_reason = pcmk_rc_str(errno); crm_info("Wait for completion of %s[%d] failed: %s " QB_XS " source=waitpid", op->id, op->pid, wait_reason); wait_rc = 0; // Act as if process is still running #ifndef HAVE_SYS_SIGNALFD_H } else { /* The child hasn't exited, so this SIGCHLD could be for * another child. We have to ignore it here but will still * need to resend it after this synchronous action has * completed and SIGCHLD has been restored to be handled by * the previous handler, so that it will be handled. */ data->ignored = true; #endif } } } else if (poll_rc == 0) { // Poll timed out with no descriptors ready timeout = 0; break; } else if ((poll_rc < 0) && (errno != EINTR)) { wait_reason = pcmk_rc_str(errno); crm_info("Wait for completion of %s[%d] failed: %s " QB_XS " source=poll", op->id, op->pid, wait_reason); break; } timeout = op->timeout - (time(NULL) - start) * 1000; } while ((op->timeout < 0 || timeout > 0)); crm_trace("Stopped waiting for %s[%d]", op->id, op->pid); finish_op_output(op, true); finish_op_output(op, false); close_op_input(op); sigchld_close(fds[2].fd); if (wait_rc <= 0) { if ((op->timeout > 0) && (timeout <= 0)) { services__format_result(op, services__generic_error(op), PCMK_EXEC_TIMEOUT, "%s did not exit within specified timeout", services__action_kind(op)); crm_info("%s[%d] timed out after %dms", op->id, op->pid, op->timeout); } else { services__set_result(op, services__generic_error(op), PCMK_EXEC_ERROR, wait_reason); } /* If only child hasn't been successfully waited for, yet. This is to limit killing wrong target a bit more. */ if ((wait_rc == 0) && (waitpid(op->pid, &status, WNOHANG) == 0)) { if (kill(op->pid, SIGKILL)) { crm_warn("Could not kill rogue child %s[%d]: %s", op->id, op->pid, pcmk_rc_str(errno)); } /* Safe to skip WNOHANG here as we sent non-ignorable signal. */ while ((waitpid(op->pid, &status, 0) == (pid_t) -1) && (errno == EINTR)) { /* keep waiting */; } } } else if (WIFEXITED(status)) { services__set_result(op, WEXITSTATUS(status), PCMK_EXEC_DONE, NULL); parse_exit_reason_from_stderr(op); crm_info("%s[%d] exited with status %d", op->id, op->pid, op->rc); } else if (WIFSIGNALED(status)) { int signo = WTERMSIG(status); services__format_result(op, services__generic_error(op), PCMK_EXEC_ERROR, "%s interrupted by %s signal", services__action_kind(op), strsignal(signo)); crm_info("%s[%d] terminated with signal %d (%s)", op->id, op->pid, signo, strsignal(signo)); #ifdef WCOREDUMP if (WCOREDUMP(status)) { crm_warn("%s[%d] dumped core", op->id, op->pid); } #endif } else { // Shouldn't be possible to get here services__set_result(op, services__generic_error(op), PCMK_EXEC_ERROR, "Unable to wait for child to complete"); } } /*! * \internal * \brief Execute an action whose standard uses executable files * * \param[in,out] op Action to execute * * \return Standard Pacemaker return value * \retval EBUSY Recurring operation could not be initiated * \retval pcmk_rc_error Synchronous action failed * \retval pcmk_rc_ok Synchronous action succeeded, or asynchronous action * should not be freed (because it's pending or because * it failed to execute and was already freed) * * \note If the return value for an asynchronous action is not pcmk_rc_ok, the * caller is responsible for freeing the action. */ int services__execute_file(svc_action_t *op) { int stdout_fd[2]; int stderr_fd[2]; int stdin_fd[2] = {-1, -1}; int rc; struct stat st; struct sigchld_data_s data = { .ignored = false }; // Catch common failure conditions early if (stat(op->opaque->exec, &st) != 0) { rc = errno; crm_info("Cannot execute '%s': %s " QB_XS " stat rc=%d", op->opaque->exec, pcmk_rc_str(rc), rc); services__handle_exec_error(op, rc); goto done; } if (pipe(stdout_fd) < 0) { rc = errno; crm_info("Cannot execute '%s': %s " QB_XS " pipe(stdout) rc=%d", op->opaque->exec, pcmk_rc_str(rc), rc); services__handle_exec_error(op, rc); goto done; } if (pipe(stderr_fd) < 0) { rc = errno; close_pipe(stdout_fd); crm_info("Cannot execute '%s': %s " QB_XS " pipe(stderr) rc=%d", op->opaque->exec, pcmk_rc_str(rc), rc); services__handle_exec_error(op, rc); goto done; } if (pcmk_is_set(pcmk_get_ra_caps(op->standard), pcmk_ra_cap_stdin)) { if (pipe(stdin_fd) < 0) { rc = errno; close_pipe(stdout_fd); close_pipe(stderr_fd); crm_info("Cannot execute '%s': %s " QB_XS " pipe(stdin) rc=%d", op->opaque->exec, pcmk_rc_str(rc), rc); services__handle_exec_error(op, rc); goto done; } } if (op->synchronous && !sigchld_setup(&data)) { close_pipe(stdin_fd); close_pipe(stdout_fd); close_pipe(stderr_fd); sigchld_cleanup(&data); services__set_result(op, services__generic_error(op), PCMK_EXEC_ERROR, "Could not manage signals for child process"); goto done; } op->pid = fork(); switch (op->pid) { case -1: rc = errno; close_pipe(stdin_fd); close_pipe(stdout_fd); close_pipe(stderr_fd); crm_info("Cannot execute '%s': %s " QB_XS " fork rc=%d", op->opaque->exec, pcmk_rc_str(rc), rc); services__handle_exec_error(op, rc); if (op->synchronous) { sigchld_cleanup(&data); } goto done; break; case 0: /* Child */ close(stdout_fd[0]); close(stderr_fd[0]); if (stdin_fd[1] >= 0) { close(stdin_fd[1]); } if (STDOUT_FILENO != stdout_fd[1]) { if (dup2(stdout_fd[1], STDOUT_FILENO) != STDOUT_FILENO) { crm_warn("Can't redirect output from '%s': %s " QB_XS " errno=%d", op->opaque->exec, pcmk_rc_str(errno), errno); } close(stdout_fd[1]); } if (STDERR_FILENO != stderr_fd[1]) { if (dup2(stderr_fd[1], STDERR_FILENO) != STDERR_FILENO) { crm_warn("Can't redirect error output from '%s': %s " QB_XS " errno=%d", op->opaque->exec, pcmk_rc_str(errno), errno); } close(stderr_fd[1]); } if ((stdin_fd[0] >= 0) && (STDIN_FILENO != stdin_fd[0])) { if (dup2(stdin_fd[0], STDIN_FILENO) != STDIN_FILENO) { crm_warn("Can't redirect input to '%s': %s " QB_XS " errno=%d", op->opaque->exec, pcmk_rc_str(errno), errno); } close(stdin_fd[0]); } if (op->synchronous) { sigchld_cleanup(&data); } action_launch_child(op); pcmk__assert(false); // action_launch_child() should not return } /* Only the parent reaches here */ close(stdout_fd[1]); close(stderr_fd[1]); if (stdin_fd[0] >= 0) { close(stdin_fd[0]); } op->opaque->stdout_fd = stdout_fd[0]; rc = pcmk__set_nonblocking(op->opaque->stdout_fd); if (rc != pcmk_rc_ok) { crm_info("Could not set '%s' output non-blocking: %s " QB_XS " rc=%d", op->opaque->exec, pcmk_rc_str(rc), rc); } op->opaque->stderr_fd = stderr_fd[0]; rc = pcmk__set_nonblocking(op->opaque->stderr_fd); if (rc != pcmk_rc_ok) { crm_info("Could not set '%s' error output non-blocking: %s " QB_XS " rc=%d", op->opaque->exec, pcmk_rc_str(rc), rc); } op->opaque->stdin_fd = stdin_fd[1]; if (op->opaque->stdin_fd >= 0) { // using buffer behind non-blocking-fd here - that could be improved // as long as no other standard uses stdin_fd assume stonith rc = pcmk__set_nonblocking(op->opaque->stdin_fd); if (rc != pcmk_rc_ok) { crm_info("Could not set '%s' input non-blocking: %s " QB_XS " fd=%d,rc=%d", op->opaque->exec, pcmk_rc_str(rc), op->opaque->stdin_fd, rc); } pipe_in_action_stdin_parameters(op); // as long as we are handling parameters directly in here just close close(op->opaque->stdin_fd); op->opaque->stdin_fd = -1; } // after fds are setup properly and before we plug anything into mainloop if (op->opaque->fork_callback) { op->opaque->fork_callback(op); } if (op->synchronous) { wait_for_sync_result(op, &data); sigchld_cleanup(&data); goto done; } crm_trace("Waiting async for '%s'[%d]", op->opaque->exec, op->pid); mainloop_child_add_with_flags(op->pid, op->timeout, op->id, op, pcmk_is_set(op->flags, SVC_ACTION_LEAVE_GROUP)? mainloop_leave_pid_group : 0, async_action_complete); op->opaque->stdout_gsource = mainloop_add_fd(op->id, G_PRIORITY_LOW, op->opaque->stdout_fd, op, &stdout_callbacks); op->opaque->stderr_gsource = mainloop_add_fd(op->id, G_PRIORITY_LOW, op->opaque->stderr_fd, op, &stderr_callbacks); services_add_inflight_op(op); return pcmk_rc_ok; done: if (op->synchronous) { return (op->rc == PCMK_OCF_OK)? pcmk_rc_ok : pcmk_rc_error; } else { return services__finalize_async_op(op); } } GList * services_os_get_single_directory_list(const char *root, gboolean files, gboolean executable) { GList *list = NULL; struct dirent **namelist = NULL; int entries = 0, lpc = 0; entries = scandir(root, &namelist, NULL, alphasort); if (entries <= 0) { return list; } for (lpc = 0; lpc < entries; lpc++) { char *buffer = NULL; struct stat sb; int rc = 0; if ('.' == namelist[lpc]->d_name[0]) { free(namelist[lpc]); continue; } buffer = crm_strdup_printf("%s/%s", root, namelist[lpc]->d_name); rc = stat(buffer, &sb); free(buffer); if (rc != 0) { continue; } if (S_ISDIR(sb.st_mode)) { if (files) { free(namelist[lpc]); continue; } } else if (S_ISREG(sb.st_mode)) { if (files == FALSE) { free(namelist[lpc]); continue; } else if (executable && (sb.st_mode & S_IXUSR) == 0 && (sb.st_mode & S_IXGRP) == 0 && (sb.st_mode & S_IXOTH) == 0) { free(namelist[lpc]); continue; } } list = g_list_append(list, strdup(namelist[lpc]->d_name)); free(namelist[lpc]); } free(namelist); return list; } GList * services_os_get_directory_list(const char *root, gboolean files, gboolean executable) { GList *result = NULL; char *dirs = strdup(root); char *dir = NULL; if (pcmk__str_empty(dirs)) { free(dirs); return result; } for (dir = strtok(dirs, ":"); dir != NULL; dir = strtok(NULL, ":")) { GList *tmp = services_os_get_single_directory_list(dir, files, executable); if (tmp) { result = g_list_concat(result, tmp); } } free(dirs); return result; } diff --git a/tools/cibadmin.c b/tools/cibadmin.c index 953ba1d8aa..83e67f4367 100644 --- a/tools/cibadmin.c +++ b/tools/cibadmin.c @@ -1,1250 +1,1258 @@ /* * Copyright 2004-2025 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 // uint32_t, etc. #include // NULL #include #include #include #include #include #include #define SUMMARY "query and edit the Pacemaker configuration" #define DEFAULT_TIMEOUT 30 #define INDENT " " /*! * \internal * \brief How to interpret \c options.cib_section */ enum cibadmin_section_type { //! No section specified: the command applies to the entire CIB cibadmin_section_all = 0, //! Section is the name of the CIB element to which the command applies cibadmin_section_scope, //! Section is an XPath expression, and the command applies to matches cibadmin_section_xpath, }; /*! * \internal * \brief Commands for \c cibadmin */ enum cibadmin_cmd { cibadmin_cmd_bump, cibadmin_cmd_create, cibadmin_cmd_delete, cibadmin_cmd_delete_all, cibadmin_cmd_empty, cibadmin_cmd_erase, cibadmin_cmd_md5_sum, cibadmin_cmd_md5_sum_versioned, cibadmin_cmd_modify, cibadmin_cmd_patch, cibadmin_cmd_query, cibadmin_cmd_replace, cibadmin_cmd_upgrade, // Update this when adding new commands cibadmin_cmd_max = cibadmin_cmd_upgrade, }; /*! * \internal * \brief Flags to define attributes of a given \c cibadmin command */ enum cibadmin_command_flags { //! This flag has no effect cibadmin_cf_none = UINT32_C(0), /*! * \brief Command requires input * * There is no optional input. Either a command requires input, or it * ignores any input that was provided. */ cibadmin_cf_requires_input = (UINT32_C(1) << 0), /*! * \brief Command is especially unsafe * * Any command that modifies the CIB is unsafe. This flag is for commands * that are likely to be destructive to larger portions of the CIB and to be * used by mistake. */ cibadmin_cf_unsafe = (UINT32_C(1) << 1), /*! * \brief Command can use an XPath expression instead of input XML * * If \c options.section_type is \c cibadmin_section_xpath, then the command * uses \c options.cib_section rather than reading input XML. */ cibadmin_cf_xpath_input = (UINT32_C(1) << 2), }; /*! * \internal * \brief Setup function for a \c cibadmin command (before any CIB API call) */ typedef crm_exit_t (*cibadmin_pre_fn_t)(pcmk__output_t *, int *, xmlNode *, GError **); /*! * \internal * \brief Return/output handler for a \c cibadmin command (after CIB API call) */ typedef crm_exit_t (*cibadmin_post_fn_t)(pcmk__output_t *, cib_t *, int, xmlNode *, int, GError **); /*! * \internal * \brief Information about a \c cibadmin command type */ typedef struct { const char *cib_request; //!< Name of request to send to the CIB API cibadmin_pre_fn_t pre_fn; //!< Function to call before CIB API call cibadmin_post_fn_t post_fn; //!< Function to call after CIB API call //! Group of enum cibadmin_command_flags uint32_t flags; } cibadmin_cmd_info_t; static struct { enum cibadmin_cmd cmd; enum cibadmin_section_type section_type; char *cib_section; char *validate_with; gint timeout_sec; enum pcmk__acl_render_how acl_render_mode; gchar *cib_user; gchar *input_file; gchar *input_string; gboolean input_stdin; gboolean allow_create; gboolean force; gboolean get_node_path; gboolean no_children; gboolean score_update; // @COMPAT Deprecated since 3.0.2 gchar *dest_node; // @COMPAT Deprecated since 3.0.0 gboolean local; // @COMPAT Deprecated since 3.0.1 gboolean sync_call; } options = { .cmd = cibadmin_cmd_query, .timeout_sec = DEFAULT_TIMEOUT, }; /*! * \internal * \brief Determine whether the given CIB scope is valid for \p cibadmin * * \param[in] scope Scope to validate * * \return true if \p scope is valid, or false otherwise * \note An invalid scope applies the operation to the entire CIB. */ static inline bool scope_is_valid(const char *scope) { return pcmk__str_any_of(scope, PCMK_XE_CONFIGURATION, PCMK_XE_NODES, PCMK_XE_RESOURCES, PCMK_XE_CONSTRAINTS, PCMK_XE_CRM_CONFIG, PCMK_XE_RSC_DEFAULTS, PCMK_XE_OP_DEFAULTS, PCMK_XE_ACLS, PCMK_XE_FENCING_TOPOLOGY, PCMK_XE_TAGS, PCMK_XE_ALERTS, PCMK_XE_STATUS, NULL); } static void cibadmin_output_basic_xml(pcmk__output_t *out, const xmlNode *xml) { GString *buf = g_string_sized_new(1024); pcmk__xml_string(xml, pcmk__xml_fmt_pretty, buf, 0); out->output_xml(out, PCMK_XE_OUTPUT, buf->str); g_string_free(buf, TRUE); } static crm_exit_t cibadmin_pre_delete_all(pcmk__output_t *out, int *call_options, xmlNode *input, GError **error) { // Remove all matching objects. Meaningful only with cibadmin_section_xpath. cib__set_call_options(*call_options, crm_system_name, cib_multiple); return CRM_EX_OK; } static crm_exit_t cibadmin_pre_empty(pcmk__output_t *out, int *call_options, xmlNode *input, GError **error) { /* Output an empty CIB. * Handles entirety of empty command; there is no CIB request. */ xmlNode *output = createEmptyCib(1); crm_xml_add(output, PCMK_XA_VALIDATE_WITH, options.validate_with); cibadmin_output_basic_xml(out, output); pcmk__xml_free(output); return CRM_EX_OK; } static crm_exit_t cibadmin_pre_md5_sum(pcmk__output_t *out, int *call_options, xmlNode *input, GError **error) { // Handles entirety of md5_sum command; there is no CIB request int rc = pcmk_rc_ok; char *digest = pcmk__digest_on_disk_cib(input); if (digest == NULL) { /* On-disk digest should be non-NULL even if input is NULL or empty, * since whitespace gets added before and after dumping the XML */ g_set_error(error, PCMK__EXITC_ERROR, CRM_EX_SOFTWARE, "Bug: Null digest"); return CRM_EX_SOFTWARE; } rc = out->message(out, "cibadmin-md5-sum", digest); free(digest); return pcmk_rc2exitc(rc); } static crm_exit_t cibadmin_pre_md5_sum_versioned(pcmk__output_t *out, int *call_options, xmlNode *input, GError **error) { // Handles entirety of md5_sum_versioned command; there is no CIB request int rc = pcmk_rc_ok; char *digest = pcmk__digest_xml(input, true); if (digest == NULL) { int rc = pcmk_rc_bad_input; g_set_error(error, PCMK__RC_ERROR, rc, "Couldn't compute digest: %s", pcmk_rc_str(rc)); return pcmk_rc2exitc(rc); } rc = out->message(out, "cibadmin-md5-sum", digest); free(digest); return pcmk_rc2exitc(rc); } static crm_exit_t cibadmin_pre_modify(pcmk__output_t *out, int *call_options, xmlNode *input, GError **error) { /* @COMPAT When we drop default support for expansion in cibadmin, guard * with `if (options.score_update)` */ cib__set_call_options(*call_options, crm_system_name, cib_score_update); if (options.allow_create) { // Allow target to be created if it does not exist cib__set_call_options(*call_options, crm_system_name, cib_can_create); } return CRM_EX_OK; } static crm_exit_t cibadmin_pre_query(pcmk__output_t *out, int *call_options, xmlNode *input, GError **error) { if (options.get_node_path) { /* Enable getting node path of XPath query matches. Meaningful only with * cibadmin_section_xpath. */ cib__set_call_options(*call_options, crm_system_name, cib_xpath_address); } if (options.no_children) { // Don't include a match's children in the query result cib__set_call_options(*call_options, crm_system_name, cib_no_children); } return CRM_EX_OK; } static crm_exit_t cibadmin_pre_replace(pcmk__output_t *out, int *call_options, xmlNode *input, GError **error) { if (pcmk__xe_is(input, PCMK_XE_CIB)) { xmlNode *status = pcmk_find_cib_element(input, PCMK_XE_STATUS); if (status == NULL) { pcmk__xe_create(input, PCMK_XE_STATUS); } } return CRM_EX_OK; } static crm_exit_t cibadmin_post_upgrade(pcmk__output_t *out, cib_t *cib_conn, int call_options, xmlNode *output, int cib_rc, GError **error) { if (cib_rc == pcmk_rc_ok) { return CRM_EX_OK; } if (cib_rc == pcmk_rc_schema_unchanged) { out->info(out, "Upgrade unnecessary: %s", pcmk_rc_str(cib_rc)); return CRM_EX_OK; } g_set_error(error, PCMK__RC_ERROR, cib_rc, "CIB API call failed: %s", pcmk_rc_str(cib_rc)); if (cib_rc == pcmk_rc_schema_validation) { xmlNode *obj = NULL; if (cib_conn->cmds->query(cib_conn, NULL, &obj, call_options) == pcmk_ok) { pcmk__update_schema(&obj, NULL, true, false); } pcmk__xml_free(obj); } return pcmk_rc2exitc(cib_rc); } static crm_exit_t cibadmin_post_default(pcmk__output_t *out, cib_t *cib_conn, int call_options, xmlNode *output, int cib_rc, GError **error) { if (cib_rc != pcmk_rc_ok) { g_set_error(error, PCMK__RC_ERROR, cib_rc, "CIB API call failed: %s", pcmk_rc_str(cib_rc)); if ((cib_rc == pcmk_rc_schema_validation) && pcmk__xe_is(output, PCMK_XE_CIB)) { // Show validation errors to stderr pcmk__validate_xml(output, NULL, NULL, NULL); } return pcmk_rc2exitc(cib_rc); } return CRM_EX_OK; } static void cibadmin_output_xml(pcmk__output_t *out, xmlNode *xml, int call_options, const gchar *acl_user, crm_exit_t *exit_code, GError **error) { if ((options.acl_render_mode != pcmk__acl_render_none) && (*exit_code == CRM_EX_OK) && pcmk__xe_is(xml, PCMK_XE_CIB)) { xmlDoc *acl_evaled_doc = NULL; xmlChar *rendered = NULL; int rc = pcmk__acl_annotate_permissions(acl_user, xml->doc, &acl_evaled_doc); if (rc != pcmk_rc_ok) { *exit_code = CRM_EX_CONFIG; g_set_error(error, PCMK__EXITC_ERROR, *exit_code, "Could not evaluate ACLs for %s: %s", acl_user, pcmk_rc_str(rc)); return; } rc = pcmk__acl_evaled_render(acl_evaled_doc, options.acl_render_mode, &rendered); if (rc != pcmk_rc_ok) { *exit_code = CRM_EX_CONFIG; g_set_error(error, PCMK__EXITC_ERROR, *exit_code, "Could not render ACLs for %s: %s", acl_user, pcmk_rc_str(rc)); return; } out->message(out, "cibadmin-rendered-acls", (const char *) rendered); xmlFree(rendered); } else if (pcmk_is_set(call_options, cib_xpath_address) && pcmk__xe_is(xml, PCMK__XE_XPATH_QUERY)) { // @COMPAT Remove when -e/--node-path is removed out->message(out, "cibadmin-node-path", xml); } else { cibadmin_output_basic_xml(out, xml); } } static crm_exit_t cibadmin_handle_command(pcmk__output_t *out, const cibadmin_cmd_info_t *cmd_info, int call_options, const gchar *acl_user, xmlNode *input, GError **error) { int rc = pcmk_rc_ok; crm_exit_t exit_code = CRM_EX_OK; cib_t *cib_conn = NULL; xmlNode *output = NULL; if (cmd_info->pre_fn != NULL) { exit_code = cmd_info->pre_fn(out, &call_options, input, error); } if ((exit_code != CRM_EX_OK) || (cmd_info->cib_request == NULL)) { goto done; } if (options.section_type == cibadmin_section_xpath) { // Enable getting section by XPath cib__set_call_options(call_options, crm_system_name, cib_xpath); } else if ((options.section_type == cibadmin_section_scope) && !scope_is_valid(options.cib_section)) { // @COMPAT: Consider requiring --force to proceed out->err(out, "Invalid value '%s' for '--scope'. Operation will apply to the " "entire CIB", options.cib_section); } rc = cib__create_signon(&cib_conn); if (rc != pcmk_rc_ok) { exit_code = pcmk_rc2exitc(rc); g_set_error(error, PCMK__EXITC_ERROR, exit_code, "Could not connect to the CIB API: %s", pcmk_rc_str(rc)); goto done; } cib_conn->call_timeout = options.timeout_sec; if (cib_conn->call_timeout < 1) { out->err(out, "Timeout must be positive, defaulting to %d", DEFAULT_TIMEOUT); cib_conn->call_timeout = DEFAULT_TIMEOUT; } rc = cib_internal_op(cib_conn, cmd_info->cib_request, options.dest_node, options.cib_section, input, &output, call_options, options.cib_user); rc = pcmk_legacy2rc(rc); if (cmd_info->post_fn != NULL) { exit_code = cmd_info->post_fn(out, cib_conn, call_options, output, rc, error); } else { exit_code = cibadmin_post_default(out, cib_conn, call_options, output, rc, error); } if (output != NULL) { cibadmin_output_xml(out, output, call_options, acl_user, &exit_code, error); } done: pcmk__xml_free(output); rc = cib__clean_up_connection(&cib_conn); if (exit_code == CRM_EX_OK) { exit_code = pcmk_rc2exitc(rc); } return exit_code; } static const cibadmin_cmd_info_t cibadmin_command_info[] = { [cibadmin_cmd_bump] = { PCMK__CIB_REQUEST_BUMP, NULL, NULL, cibadmin_cf_none, }, [cibadmin_cmd_create] = { PCMK__CIB_REQUEST_CREATE, NULL, NULL, cibadmin_cf_requires_input, }, [cibadmin_cmd_delete] = { PCMK__CIB_REQUEST_DELETE, cibadmin_pre_delete_all, NULL, cibadmin_cf_requires_input|cibadmin_cf_xpath_input, }, [cibadmin_cmd_delete_all] = { PCMK__CIB_REQUEST_DELETE, NULL, NULL, cibadmin_cf_requires_input|cibadmin_cf_unsafe|cibadmin_cf_xpath_input, }, [cibadmin_cmd_empty] = { NULL, cibadmin_pre_empty, NULL, cibadmin_cf_none, }, [cibadmin_cmd_erase] = { PCMK__CIB_REQUEST_ERASE, NULL, NULL, cibadmin_cf_unsafe, }, [cibadmin_cmd_md5_sum] = { NULL, cibadmin_pre_md5_sum, NULL, cibadmin_cf_requires_input, }, [cibadmin_cmd_md5_sum_versioned] = { NULL, cibadmin_pre_md5_sum_versioned, NULL, cibadmin_cf_requires_input, }, [cibadmin_cmd_modify] = { PCMK__CIB_REQUEST_MODIFY, cibadmin_pre_modify, NULL, cibadmin_cf_requires_input, }, [cibadmin_cmd_patch] = { PCMK__CIB_REQUEST_APPLY_PATCH, NULL, NULL, cibadmin_cf_requires_input, }, [cibadmin_cmd_query] = { PCMK__CIB_REQUEST_QUERY, cibadmin_pre_query, NULL, cibadmin_cf_none, }, [cibadmin_cmd_replace] = { PCMK__CIB_REQUEST_REPLACE, cibadmin_pre_replace, NULL, cibadmin_cf_requires_input, }, /* @TODO Ideally, --upgrade wouldn't be considered unsafe if the CIB already * uses the latest schema. */ [cibadmin_cmd_upgrade] = { PCMK__CIB_REQUEST_UPGRADE, NULL, cibadmin_post_upgrade, cibadmin_cf_unsafe, }, }; static gboolean command_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error) { if (pcmk__str_any_of(option_name, "-u", "--upgrade", NULL)) { options.cmd = cibadmin_cmd_upgrade; } else if (pcmk__str_any_of(option_name, "-Q", "--query", NULL)) { options.cmd = cibadmin_cmd_query; } else if (pcmk__str_any_of(option_name, "-E", "--erase", NULL)) { options.cmd = cibadmin_cmd_erase; } else if (pcmk__str_any_of(option_name, "-B", "--bump", NULL)) { options.cmd = cibadmin_cmd_bump; } else if (pcmk__str_any_of(option_name, "-C", "--create", NULL)) { options.cmd = cibadmin_cmd_create; } else if (pcmk__str_any_of(option_name, "-M", "--modify", NULL)) { options.cmd = cibadmin_cmd_modify; } else if (pcmk__str_any_of(option_name, "-P", "--patch", NULL)) { options.cmd = cibadmin_cmd_patch; } else if (pcmk__str_any_of(option_name, "-R", "--replace", NULL)) { options.cmd = cibadmin_cmd_replace; } else if (pcmk__str_any_of(option_name, "-D", "--delete", NULL)) { options.cmd = cibadmin_cmd_delete; } else if (pcmk__str_any_of(option_name, "-d", "--delete-all", NULL)) { options.cmd = cibadmin_cmd_delete_all; } else if (pcmk__str_any_of(option_name, "-a", "--empty", NULL)) { options.cmd = cibadmin_cmd_empty; pcmk__str_update(&options.validate_with, optarg); } else if (pcmk__str_any_of(option_name, "-5", "--md5-sum", NULL)) { options.cmd = cibadmin_cmd_md5_sum; } else if (pcmk__str_any_of(option_name, "-6", "--md5-sum-versioned", NULL)) { options.cmd = cibadmin_cmd_md5_sum_versioned; } else { // Should be impossible return FALSE; } return TRUE; } static gboolean show_access_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error) { if (pcmk__str_eq(optarg, "auto", pcmk__str_null_matches)) { options.acl_render_mode = pcmk__acl_render_default; } else if (g_strcmp0(optarg, "namespace") == 0) { options.acl_render_mode = pcmk__acl_render_namespace; } else if (g_strcmp0(optarg, "text") == 0) { options.acl_render_mode = pcmk__acl_render_text; } else if (g_strcmp0(optarg, "color") == 0) { options.acl_render_mode = pcmk__acl_render_color; } else { g_set_error(error, PCMK__EXITC_ERROR, CRM_EX_USAGE, "Invalid value '%s' for option '%s'", optarg, option_name); return FALSE; } return TRUE; } static gboolean section_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **error) { if (pcmk__str_any_of(option_name, "-o", "--scope", NULL)) { options.section_type = cibadmin_section_scope; } else if (pcmk__str_any_of(option_name, "-A", "--xpath", NULL)) { options.section_type = cibadmin_section_xpath; } else { // Should be impossible return FALSE; } pcmk__str_update(&options.cib_section, optarg); return TRUE; } static GOptionEntry command_entries[] = { { "upgrade", 'u', G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, command_cb, "Upgrade the configuration to the latest syntax", NULL }, { "query", 'Q', G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, command_cb, "Query the contents of the CIB", NULL }, { "erase", 'E', G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, command_cb, "Erase the contents of the whole CIB", NULL }, { "bump", 'B', G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, command_cb, "Increase the CIB's epoch value by 1", NULL }, { "create", 'C', G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, command_cb, "Create an object in the CIB (will fail if object already exists)", NULL }, { "modify", 'M', G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, command_cb, "Find object somewhere in CIB's XML tree and update it (fails if object " "does not exist unless -c is also specified)", NULL }, { "patch", 'P', G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, command_cb, "Supply an update in the form of an XML diff (see crm_diff(8))", NULL }, { "replace", 'R', G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, command_cb, "Recursively replace an object in the CIB", NULL }, { "delete", 'D', G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, command_cb, "Delete first object matching supplied criteria (for example, " "<" PCMK_XE_OP " " PCMK_XA_ID "=\"rsc1_op1\" " PCMK_XA_NAME "=\"monitor\"/>).\n" INDENT "The XML element name and all attributes must match in order for " "the element to be deleted.", NULL }, { "delete-all", 'd', G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, command_cb, "When used with --xpath, remove all matching objects in the " "configuration instead of just the first one", NULL }, { "empty", 'a', G_OPTION_FLAG_OPTIONAL_ARG, G_OPTION_ARG_CALLBACK, command_cb, "Output an empty CIB. Accepts an optional schema name argument to use as " "the " PCMK_XA_VALIDATE_WITH " value.\n" INDENT "If no schema is given, the latest will be used.", "[schema]" }, { "md5-sum", '5', G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, command_cb, "Calculate the on-disk CIB digest", NULL }, { "md5-sum-versioned", '6', G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, command_cb, "Calculate an on-the-wire versioned CIB digest", NULL }, { NULL } }; static GOptionEntry data_entries[] = { // @COMPAT These arguments should be last-one-wins { "xml-file", 'x', G_OPTION_FLAG_NONE, G_OPTION_ARG_FILENAME, &options.input_file, "Retrieve XML from the named file. Currently this takes precedence\n" INDENT "over --xml-text and --xml-pipe. In a future release, the last\n" INDENT "one specified will be used.", "FILE" }, { "xml-text", 'X', G_OPTION_FLAG_NONE, G_OPTION_ARG_STRING, &options.input_string, "Retrieve XML from the supplied string. Currently this takes precedence\n" INDENT "over --xml-pipe, but --xml-file overrides this. In a future\n" INDENT "release, the last one specified will be used.", "STRING" }, { "xml-pipe", 'p', G_OPTION_FLAG_NONE, G_OPTION_ARG_NONE, &options.input_stdin, "Retrieve XML from stdin. Currently --xml-file and --xml-text override\n" INDENT "this. In a future release, the last one specified will be used.", NULL }, { NULL } }; static GOptionEntry addl_entries[] = { { "force", 'f', G_OPTION_FLAG_NONE, G_OPTION_ARG_NONE, &options.force, "Force the action to be performed", NULL }, { "timeout", 't', G_OPTION_FLAG_NONE, G_OPTION_ARG_INT, &options.timeout_sec, "Time (in seconds) to wait before declaring the operation failed", "value" }, { "user", 'U', G_OPTION_FLAG_NONE, G_OPTION_ARG_STRING, &options.cib_user, "Run the command with permissions of the named user (valid only for the " "root and " CRM_DAEMON_USER " accounts)", "value" }, { "scope", 'o', G_OPTION_FLAG_NONE, G_OPTION_ARG_CALLBACK, section_cb, "Limit scope of operation to specific section of CIB\n" INDENT "Valid values: " PCMK_XE_CONFIGURATION ", " PCMK_XE_NODES ", " PCMK_XE_RESOURCES ", " PCMK_XE_CONSTRAINTS ", " PCMK_XE_CRM_CONFIG ", " PCMK_XE_RSC_DEFAULTS ",\n" INDENT " " PCMK_XE_OP_DEFAULTS ", " PCMK_XE_ACLS ", " PCMK_XE_FENCING_TOPOLOGY ", " PCMK_XE_TAGS ", " PCMK_XE_ALERTS ", " PCMK_XE_STATUS "\n" INDENT "If both --scope/-o and --xpath/-a are specified, the last one to " "appear takes effect", "value" }, { "xpath", 'A', G_OPTION_FLAG_NONE, G_OPTION_ARG_CALLBACK, section_cb, "A valid XPath to use instead of --scope/-o\n" INDENT "If both --scope/-o and --xpath/-a are specified, the last one to " "appear takes effect", "value" }, { "show-access", 'S', G_OPTION_FLAG_OPTIONAL_ARG, G_OPTION_ARG_CALLBACK, show_access_cb, "Whether to use syntax highlighting for ACLs (with -Q/--query and " "-U/--user)\n" INDENT "Allowed values: 'color' (default for terminal), 'text' (plain text, " "default for non-terminal),\n" INDENT " 'namespace', or 'auto' (use default value)\n" INDENT "Default value: 'auto'", "[value]" }, { "score", 0, G_OPTION_FLAG_NONE, G_OPTION_ARG_NONE, &options.score_update, "Treat new attribute values as atomic score updates where possible " "(with --modify/-M).\n" INDENT "This currently happens by default and cannot be disabled, but\n" INDENT "this default behavior is deprecated and will be removed in a\n" INDENT "future release. Set this flag if this behavior is desired.\n" INDENT "This option takes effect when updating XML attributes. For an\n" INDENT "attribute named \"name\", if the new value is \"name++\" or\n" INDENT "\"name+=X\" for some score X, the new value is set as follows:\n" INDENT "If attribute \"name\" is not already set to some value in\n" INDENT "the element being updated, the new value is set as a literal\n" INDENT "string.\n" INDENT "If the new value is \"name++\", then the attribute is set to \n" INDENT "its existing value (parsed as a score) plus 1.\n" INDENT "If the new value is \"name+=X\" for some score X, then the\n" INDENT "attribute is set to its existing value plus X, where the\n" INDENT "existing value and X are parsed and added as scores.\n" INDENT "Scores are integer values capped at INFINITY and -INFINITY.\n" INDENT "Refer to Pacemaker Explained for more details on scores,\n" INDENT "including how they are parsed and added.", NULL }, { "allow-create", 'c', G_OPTION_FLAG_NONE, G_OPTION_ARG_NONE, &options.allow_create, "(Advanced) Allow target of --modify/-M to be created if it does not " "exist", NULL }, { "no-children", 'n', G_OPTION_FLAG_NONE, G_OPTION_ARG_NONE, &options.no_children, "(Advanced) When querying an object, do not include its children in the " "result", NULL }, // @COMPAT Deprecated since 3.0.0 { "local", 'l', G_OPTION_FLAG_HIDDEN, G_OPTION_ARG_NONE, &options.local, "(deprecated)", NULL }, // @COMPAT Deprecated since 3.0.2 { "node", 'N', G_OPTION_FLAG_HIDDEN, G_OPTION_ARG_STRING, &options.dest_node, "(deprecated)", "value" }, // @COMPAT Deprecated since 3.0.2 { "node-path", 'e', G_OPTION_FLAG_HIDDEN, G_OPTION_ARG_NONE, &options.get_node_path, "(deprecated)", NULL }, // @COMPAT Deprecated since 3.0.1 { "sync-call", 's', G_OPTION_FLAG_HIDDEN, G_OPTION_ARG_NONE, &options.sync_call, "(deprecated)", NULL }, { NULL } }; PCMK__OUTPUT_ARGS("cibadmin-md5-sum", "const char *") static int md5_sum_default(pcmk__output_t *out, va_list args) { const char *digest = va_arg(args, const char *); if (digest == NULL) { return pcmk_rc_no_output; } return out->info(out, "%s", digest); } PCMK__OUTPUT_ARGS("cibadmin-md5-sum", "const char *") static int md5_sum_xml(pcmk__output_t *out, va_list args) { const char *digest = va_arg(args, const char *); if (digest == NULL) { return pcmk_rc_no_output; } pcmk__output_create_xml_node(out, PCMK_XE_MD5_SUM, PCMK_XA_DIGEST, digest, NULL); return pcmk_rc_ok; } // @COMPAT Drop "cibadmin-node-path" and helper when dropping --node-path static int output_xml_id(xmlNode *xml, void *user_data) { pcmk__output_t *out = user_data; const char *id = pcmk__xe_id(xml); pcmk__assert(id != NULL); return out->info(out, "%s", id); } PCMK__OUTPUT_ARGS("cibadmin-node-path", "xmlNode *") static int node_path_default(pcmk__output_t *out, va_list args) { xmlNode *query_result = va_arg(args, xmlNode *); if (query_result == NULL) { return pcmk_rc_no_output; } return pcmk__xe_foreach_child(query_result, PCMK__XE_XPATH_QUERY_PATH, output_xml_id, out); } PCMK__OUTPUT_ARGS("cibadmin-node-path", "xmlNode *") static int node_path_xml(pcmk__output_t *out, va_list args) { xmlNode *query_result = va_arg(args, xmlNode *); if (query_result == NULL) { return pcmk_rc_no_output; } cibadmin_output_basic_xml(out, query_result); return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("cibadmin-rendered-acls", "const char *") static int rendered_acls_default(pcmk__output_t *out, va_list args) { const char *rendered = va_arg(args, const char *); if (rendered == NULL) { return pcmk_rc_no_output; } return out->info(out, "%s", rendered); } PCMK__OUTPUT_ARGS("cibadmin-rendered-acls", "const char *") static int rendered_acls_xml(pcmk__output_t *out, va_list args) { /* We want to create a CData block in a PCMK_XE_OUTPUT element. At the time * of writing, that's exactly what this call to xml_output_xml() does. * Note, however, that the "rendered" string is not XML if the ACL render * mode is color or text. * * @TODO Create a pcmk__output_xml_create_cdata() or similar, and share it * between xml_output_xml() and this function? */ const char *rendered = va_arg(args, const char *); if (rendered == NULL) { return pcmk_rc_no_output; } out->output_xml(out, PCMK_XE_OUTPUT, rendered); return pcmk_rc_ok; } static const pcmk__supported_format_t formats[] = { PCMK__SUPPORTED_FORMAT_NONE, PCMK__SUPPORTED_FORMAT_TEXT, PCMK__SUPPORTED_FORMAT_XML, { NULL, NULL, NULL } }; static const pcmk__message_entry_t fmt_functions[] = { { "cibadmin-md5-sum", "default", md5_sum_default }, { "cibadmin-md5-sum", "xml", md5_sum_xml }, { "cibadmin-node-path", "default", node_path_default }, { "cibadmin-node-path", "xml", node_path_xml }, { "cibadmin-rendered-acls", "default", rendered_acls_default }, { "cibadmin-rendered-acls", "xml", rendered_acls_xml }, { NULL, NULL, NULL } }; static GOptionContext * build_arg_context(pcmk__common_args_t *args, GOptionGroup **group) { const char *desc = NULL; GOptionContext *context = NULL; desc = "Examples:\n\n" "Query the configuration:\n\n" "\t# cibadmin --query\n\n" "or just:\n\n" "\t# cibadmin\n\n" "Query just the cluster options configuration:\n\n" "\t# cibadmin --query --scope " PCMK_XE_CRM_CONFIG "\n\n" "Query all '" PCMK_META_TARGET_ROLE "' settings:\n\n" "\t# cibadmin --query --xpath " "\"//" PCMK_XE_NVPAIR "[@" PCMK_XA_NAME "='" PCMK_META_TARGET_ROLE"']\"\n\n" "Remove all '" PCMK_META_IS_MANAGED "' settings:\n\n" "\t# cibadmin --delete-all --xpath " "\"//" PCMK_XE_NVPAIR "[@" PCMK_XA_NAME "='" PCMK_META_IS_MANAGED "']\"\n\n" "Remove the resource named 'old':\n\n" "\t# cibadmin --delete --xml-text " "'<" PCMK_XE_PRIMITIVE " " PCMK_XA_ID "=\"old\"/>'\n\n" "Remove all resources from the configuration:\n\n" "\t# cibadmin --replace --scope " PCMK_XE_RESOURCES " --xml-text '<" PCMK_XE_RESOURCES "/>'\n\n" "Replace complete configuration with contents of " "$HOME/pacemaker.xml:\n\n" "\t# cibadmin --replace --xml-file $HOME/pacemaker.xml\n\n" "Replace " PCMK_XE_CONSTRAINTS " section of configuration with " "contents of $HOME/constraints.xml:\n\n" "\t# cibadmin --replace --scope " PCMK_XE_CONSTRAINTS " --xml-file $HOME/constraints.xml\n\n" "Increase configuration version to prevent old configurations from " "being loaded accidentally:\n\n" "\t# cibadmin --modify --score --xml-text " "'<" PCMK_XE_CIB " " PCMK_XA_ADMIN_EPOCH "=\"" PCMK_XA_ADMIN_EPOCH "++\"/>'\n\n" "Edit the configuration with your favorite $EDITOR:\n\n" "\t# cibadmin --query > $HOME/local.xml\n\n" "\t# $EDITOR $HOME/local.xml\n\n" "\t# cibadmin --replace --xml-file $HOME/local.xml\n\n" "Assuming terminal, render configuration in color (green for " "writable, blue for readable, red for\n" "denied) to visualize permissions for user tony:\n\n" "\t# cibadmin --show-access=color --query --user tony | less -r\n\n" "SEE ALSO:\n" " crm(8), pcs(8), crm_shadow(8), crm_diff(8)\n"; context = pcmk__build_arg_context(args, "text (default), xml", group, "[]"); g_option_context_set_description(context, desc); pcmk__add_arg_group(context, "commands", "Commands:", "Show command help", command_entries); pcmk__add_arg_group(context, "data", "Data:", "Show data help", data_entries); pcmk__add_arg_group(context, "additional", "Additional Options:", "Show additional options", addl_entries); return context; } /*! * \internal * \brief Read input XML as specified on the command line * * Precedence is as follows: * 1. Input file * 2. Input string * 3. stdin * * If multiple input sources are given, only the last occurrence of the one with * the highest precedence is tried. * * If no input source is specified, this function does nothing. * * \param[out] input Where to store parsed input * \param[out] source Where to store string describing input source * * \return Standard Pacemaker return code */ static int read_input(xmlNode **input, const char **source) { if (options.input_file != NULL) { *source = options.input_file; *input = pcmk__xml_read(options.input_file); } else if (options.input_string != NULL) { *source = "input string"; *input = pcmk__xml_parse(options.input_string); } else if (options.input_stdin) { *source = "stdin"; *input = pcmk__xml_read(NULL); } else { *source = NULL; *input = NULL; return EINVAL; } if (*input == NULL) { return pcmk_rc_bad_input; } return pcmk_rc_ok; } int main(int argc, char **argv) { int rc = pcmk_rc_ok; crm_exit_t exit_code = CRM_EX_OK; const cibadmin_cmd_info_t *cmd_info = NULL; int call_options = cib_sync_call; xmlNode *input = NULL; gchar *acl_cred = NULL; pcmk__output_t *out = NULL; GError *error = NULL; GOptionGroup *output_group = NULL; pcmk__common_args_t *args = pcmk__new_common_args(SUMMARY); gchar **processed_args = pcmk__cmdline_preproc(argv, "ANSUXhotx"); GOptionContext *context = build_arg_context(args, &output_group); pcmk__register_formats(output_group, formats); if (!g_option_context_parse_strv(context, &processed_args, &error)) { exit_code = CRM_EX_USAGE; goto done; } /* At LOG_ERR, stderr for CIB calls is rather verbose. Several lines like * * (func@file:line) error: CIB failures * * In cibadmin we explicitly output the XML portion without the prefixes. So * we default to LOG_CRIT. */ pcmk__cli_init_logging("cibadmin", 0); set_crm_log_level(LOG_CRIT); if (args->verbosity > 0) { cib__set_call_options(call_options, crm_system_name, cib_verbose); for (int i = 0; i < args->verbosity; i++) { crm_bump_log_level(argc, argv); } } rc = pcmk__output_new(&out, args->output_ty, args->output_dest, argv); if (rc != pcmk_rc_ok) { exit_code = CRM_EX_ERROR; g_set_error(&error, PCMK__EXITC_ERROR, exit_code, "Error creating output format %s: %s", args->output_ty, pcmk_rc_str(rc)); goto done; } if (g_strv_length(processed_args) > 1) { gchar *extra = g_strjoinv(" ", processed_args + 1); gchar *help = g_option_context_get_help(context, TRUE, NULL); exit_code = CRM_EX_USAGE; g_set_error(&error, PCMK__EXITC_ERROR, exit_code, "non-option ARGV-elements: %s\n\n%s", extra, help); g_free(extra); g_free(help); goto done; } if (args->version) { out->version(out); goto done; } pcmk__register_messages(out, fmt_functions); // Ensure command is in valid range if ((options.cmd >= 0) && (options.cmd <= cibadmin_cmd_max)) { cmd_info = &cibadmin_command_info[options.cmd]; } if (cmd_info == NULL) { exit_code = CRM_EX_SOFTWARE; g_set_error(&error, PCMK__EXITC_ERROR, exit_code, "Bug: Unimplemented command: %d", (int) options.cmd); goto done; } if (pcmk_is_set(cmd_info->flags, cibadmin_cf_unsafe) && !options.force) { exit_code = CRM_EX_UNSAFE; g_set_error(&error, PCMK__EXITC_ERROR, exit_code, "The supplied command is considered dangerous. To prevent " "accidental destruction of the cluster, the --force flag " "is required in order to proceed."); goto done; } /* Query is the only command that produces output suitable for ACL * rendering. Ignore --show-access for other commands. */ if (options.acl_render_mode != pcmk__acl_render_none) { if (options.cmd == cibadmin_cmd_query) { char *username = NULL; if (options.cib_user == NULL) { exit_code = CRM_EX_USAGE; g_set_error(&error, PCMK__EXITC_ERROR, exit_code, "-U/--user is required with -S/--show-access"); goto done; } - // @COMPAT Fail if pcmk_acl_required(username) username = pcmk__uid2username(geteuid()); + if (username == NULL) { + exit_code = CRM_EX_NOSUCH; + g_set_error(&error, PCMK__EXITC_ERROR, exit_code, + "Failed to get username from password database for " + "effective user ID %lld", (long long) geteuid()); + goto done; + } + + // @COMPAT Fail if pcmk_acl_required(username) if (pcmk_acl_required(username)) { out->err(out, "Warning: cibadmin is being run as user %s, which is " "subject to ACLs. As a result, ACLs for user %s may " "be incorrect or incomplete in the output. In a " "future release, running as a privileged user (root " "or " CRM_DAEMON_USER ") will be required for " "-S/--show-access.", username, options.cib_user); } free(username); /* Note: acl_cred takes ownership of options.cib_user here. * options.cib_user is set to NULL so that the CIB is obtained as * the user running the cibadmin command. The CIB must be obtained * as a user with full permissions in order to show the CIB * correctly annotated for the options.cib_user's permissions. */ acl_cred = options.cib_user; options.cib_user = NULL; } else { options.acl_render_mode = pcmk__acl_render_none; } } if (pcmk_is_set(cmd_info->flags, cibadmin_cf_requires_input)) { bool accepts_xpath = pcmk_is_set(cmd_info->flags, cibadmin_cf_xpath_input); /* If true, use options.cib_section (an XPath expression) instead of * input XML */ bool as_xpath = accepts_xpath && (options.section_type == cibadmin_section_xpath); if (!as_xpath) { const char *source = NULL; rc = read_input(&input, &source); if (rc == EINVAL) { exit_code = CRM_EX_USAGE; g_set_error(&error, PCMK__EXITC_ERROR, exit_code, "The supplied command requires %sinput via " "--xml-file, --xml-text, or --xml-pipe", (accepts_xpath? "either --xpath or " : "")); goto done; } if (rc != pcmk_rc_ok) { exit_code = pcmk_rc2exitc(rc); g_set_error(&error, PCMK__EXITC_ERROR, exit_code, "Couldn't parse input from %s", pcmk__s(source, "(BUG: null source)")); goto done; } } } exit_code = cibadmin_handle_command(out, cmd_info, call_options, acl_cred, input, &error); done: g_strfreev(processed_args); pcmk__free_arg_context(context); g_free(options.cib_user); g_free(options.dest_node); g_free(options.input_file); g_free(options.input_string); free(options.cib_section); free(options.validate_with); g_free(acl_cred); pcmk__xml_free(input); pcmk__output_and_clear_error(&error, out); if (out != NULL) { out->finish(out, exit_code, true, NULL); pcmk__output_free(out); } pcmk__unregister_formats(); crm_exit(exit_code); } diff --git a/tools/crm_mon.c b/tools/crm_mon.c index 8fd69ec08a..f2c05e42d6 100644 --- a/tools/crm_mon.c +++ b/tools/crm_mon.c @@ -1,2193 +1,2180 @@ /* * Copyright 2004-2025 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 #include #include #include #include #include #include #include #include #include #include #include #include #include // g_str_has_prefix() #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include // stonith__* #include "crm_mon.h" #define SUMMARY "Provides a summary of cluster's current state.\n\n" \ "Outputs varying levels of detail in a number of different formats." /* * Definitions indicating which items to print */ static uint32_t show; static uint32_t show_opts = pcmk_show_pending; /* * Definitions indicating how to output */ static mon_output_format_t output_format = mon_output_unset; /* other globals */ static GIOChannel *io_channel = NULL; static GMainLoop *mainloop = NULL; static guint reconnect_timer = 0; static mainloop_timer_t *refresh_timer = NULL; static enum pcmk_pacemakerd_state pcmkd_state = pcmk_pacemakerd_state_invalid; static cib_t *cib = NULL; static stonith_t *st = NULL; static xmlNode *current_cib = NULL; static GError *error = NULL; static pcmk__common_args_t *args = NULL; static pcmk__output_t *out = NULL; static GOptionContext *context = NULL; static gchar **processed_args = NULL; static time_t last_refresh = 0; volatile crm_trigger_t *refresh_trigger = NULL; static pcmk_scheduler_t *scheduler = NULL; static enum pcmk__fence_history fence_history = pcmk__fence_history_none; int interactive_fence_level = 0; static pcmk__supported_format_t formats[] = { #if PCMK__ENABLE_CURSES CRM_MON_SUPPORTED_FORMAT_CURSES, #endif PCMK__SUPPORTED_FORMAT_HTML, PCMK__SUPPORTED_FORMAT_NONE, PCMK__SUPPORTED_FORMAT_TEXT, PCMK__SUPPORTED_FORMAT_XML, { NULL, NULL, NULL } }; PCMK__OUTPUT_ARGS("crm-mon-disconnected", "const char *", "enum pcmk_pacemakerd_state") static int crm_mon_disconnected_default(pcmk__output_t *out, va_list args) { return pcmk_rc_no_output; } PCMK__OUTPUT_ARGS("crm-mon-disconnected", "const char *", "enum pcmk_pacemakerd_state") static int crm_mon_disconnected_html(pcmk__output_t *out, va_list args) { const char *desc = va_arg(args, const char *); enum pcmk_pacemakerd_state state = (enum pcmk_pacemakerd_state) va_arg(args, int); if (out->dest != stdout) { out->reset(out); } pcmk__output_create_xml_text_node(out, PCMK__XE_SPAN, "Not connected to CIB"); if (desc != NULL) { pcmk__output_create_xml_text_node(out, PCMK__XE_SPAN, ": "); pcmk__output_create_xml_text_node(out, PCMK__XE_SPAN, desc); } if (state != pcmk_pacemakerd_state_invalid) { const char *state_s = pcmk__pcmkd_state_enum2friendly(state); pcmk__output_create_xml_text_node(out, PCMK__XE_SPAN, " ("); pcmk__output_create_xml_text_node(out, PCMK__XE_SPAN, state_s); pcmk__output_create_xml_text_node(out, PCMK__XE_SPAN, ")"); } out->finish(out, CRM_EX_DISCONNECT, true, NULL); return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("crm-mon-disconnected", "const char *", "enum pcmk_pacemakerd_state") static int crm_mon_disconnected_text(pcmk__output_t *out, va_list args) { const char *desc = va_arg(args, const char *); enum pcmk_pacemakerd_state state = (enum pcmk_pacemakerd_state) va_arg(args, int); int rc = pcmk_rc_ok; if (out->dest != stdout) { out->reset(out); } if (state != pcmk_pacemakerd_state_invalid) { rc = out->info(out, "Not connected to CIB%s%s (%s)", (desc != NULL)? ": " : "", pcmk__s(desc, ""), pcmk__pcmkd_state_enum2friendly(state)); } else { rc = out->info(out, "Not connected to CIB%s%s", (desc != NULL)? ": " : "", pcmk__s(desc, "")); } out->finish(out, CRM_EX_DISCONNECT, true, NULL); return rc; } PCMK__OUTPUT_ARGS("crm-mon-disconnected", "const char *", "enum pcmk_pacemakerd_state") static int crm_mon_disconnected_xml(pcmk__output_t *out, va_list args) { const char *desc = va_arg(args, const char *); enum pcmk_pacemakerd_state state = (enum pcmk_pacemakerd_state) va_arg(args, int); const char *state_s = NULL; if (out->dest != stdout) { out->reset(out); } if (state != pcmk_pacemakerd_state_invalid) { state_s = pcmk_pacemakerd_api_daemon_state_enum2text(state); } pcmk__output_create_xml_node(out, PCMK_XE_CRM_MON_DISCONNECTED, PCMK_XA_DESCRIPTION, desc, PCMK_XA_PACEMAKERD_STATE, state_s, NULL); out->finish(out, CRM_EX_DISCONNECT, true, NULL); return pcmk_rc_ok; } static pcmk__message_entry_t fmt_functions[] = { { "crm-mon-disconnected", "default", crm_mon_disconnected_default }, { "crm-mon-disconnected", "html", crm_mon_disconnected_html }, { "crm-mon-disconnected", "text", crm_mon_disconnected_text }, { "crm-mon-disconnected", "xml", crm_mon_disconnected_xml }, { NULL, NULL, NULL }, }; #define RECONNECT_MSECS 5000 struct { guint reconnect_ms; enum mon_exec_mode exec_mode; gboolean fence_connect; gboolean print_pending; gboolean show_bans; gboolean watch_fencing; - char *pid_file; - char *external_agent; - char *external_recipient; + gchar *external_agent; + gchar *external_recipient; char *neg_location_prefix; - char *only_node; - char *only_rsc; + gchar *only_node; + gchar *only_rsc; GSList *user_includes_excludes; GSList *includes_excludes; } options = { .reconnect_ms = RECONNECT_MSECS, .exec_mode = mon_exec_unset, .fence_connect = TRUE, }; static crm_exit_t clean_up(crm_exit_t exit_code); static void crm_diff_update(const char *event, xmlNode * msg); static void clean_up_on_connection_failure(int rc); static int mon_refresh_display(gpointer user_data); static int setup_cib_connection(void); static int setup_fencer_connection(void); static int setup_api_connections(void); static void mon_st_callback_event(stonith_t * st, stonith_event_t * e); static void mon_st_callback_display(stonith_t * st, stonith_event_t * e); static void refresh_after_event(gboolean data_updated, gboolean enforce); static uint32_t all_includes(mon_output_format_t fmt) { if ((fmt == mon_output_plain) || (fmt == mon_output_console)) { return ~pcmk_section_options; } else { return pcmk_section_all; } } static uint32_t default_includes(mon_output_format_t fmt) { switch (fmt) { case mon_output_plain: case mon_output_console: case mon_output_html: return pcmk_section_summary |pcmk_section_nodes |pcmk_section_resources |pcmk_section_failures; case mon_output_xml: return all_includes(fmt); default: return 0; } } struct { const char *name; uint32_t bit; } sections[] = { { "attributes", pcmk_section_attributes }, { "bans", pcmk_section_bans }, { "counts", pcmk_section_counts }, { "dc", pcmk_section_dc }, { "failcounts", pcmk_section_failcounts }, { "failures", pcmk_section_failures }, { PCMK_VALUE_FENCING, pcmk_section_fencing_all }, { "fencing-failed", pcmk_section_fence_failed }, { "fencing-pending", pcmk_section_fence_pending }, { "fencing-succeeded", pcmk_section_fence_worked }, { "maint-mode", pcmk_section_maint_mode }, { "nodes", pcmk_section_nodes }, { "operations", pcmk_section_operations }, { "options", pcmk_section_options }, { "resources", pcmk_section_resources }, { "stack", pcmk_section_stack }, { "summary", pcmk_section_summary }, { "tickets", pcmk_section_tickets }, { "times", pcmk_section_times }, { NULL } }; static uint32_t find_section_bit(const char *name) { for (int i = 0; sections[i].name != NULL; i++) { if (pcmk__str_eq(sections[i].name, name, pcmk__str_casei)) { return sections[i].bit; } } return 0; } static gboolean apply_exclude(const gchar *excludes, GError **error) { char **parts = NULL; gboolean result = TRUE; parts = g_strsplit(excludes, ",", 0); for (char **s = parts; *s != NULL; s++) { uint32_t bit = find_section_bit(*s); if (pcmk__str_eq(*s, "all", pcmk__str_none)) { show = 0; } else if (pcmk__str_eq(*s, PCMK_VALUE_NONE, pcmk__str_none)) { show = all_includes(output_format); } else if (bit != 0) { show &= ~bit; } else { g_set_error(error, PCMK__EXITC_ERROR, CRM_EX_USAGE, "--exclude options: all, attributes, bans, counts, dc, " "failcounts, failures, fencing, fencing-failed, " "fencing-pending, fencing-succeeded, maint-mode, nodes, " PCMK_VALUE_NONE ", operations, options, resources, " "stack, summary, tickets, times"); result = FALSE; break; } } g_strfreev(parts); return result; } static gboolean apply_include(const gchar *includes, GError **error) { char **parts = NULL; gboolean result = TRUE; parts = g_strsplit(includes, ",", 0); for (char **s = parts; *s != NULL; s++) { uint32_t bit = find_section_bit(*s); if (pcmk__str_eq(*s, "all", pcmk__str_none)) { show = all_includes(output_format); } else if (g_str_has_prefix(*s, "bans")) { show |= pcmk_section_bans; if (options.neg_location_prefix != NULL) { free(options.neg_location_prefix); options.neg_location_prefix = NULL; } if (strlen(*s) > 4 && (*s)[4] == ':') { options.neg_location_prefix = strdup(*s+5); } } else if (pcmk__str_any_of(*s, PCMK_VALUE_DEFAULT, "defaults", NULL)) { show |= default_includes(output_format); } else if (pcmk__str_eq(*s, PCMK_VALUE_NONE, pcmk__str_none)) { show = 0; } else if (bit != 0) { show |= bit; } else { g_set_error(error, PCMK__EXITC_ERROR, CRM_EX_USAGE, "--include options: all, attributes, bans[:PREFIX], counts, dc, " PCMK_VALUE_DEFAULT ", failcounts, failures, fencing, " "fencing-failed, fencing-pending, fencing-succeeded, " "maint-mode, nodes, " PCMK_VALUE_NONE ", operations, " "options, resources, stack, summary, tickets, times"); result = FALSE; break; } } g_strfreev(parts); return result; } static gboolean apply_include_exclude(GSList *lst, GError **error) { gboolean rc = TRUE; GSList *node = lst; while (node != NULL) { char *s = node->data; if (s == NULL) { } else if (g_str_has_prefix(s, "--include=")) { rc = apply_include(s+10, error); } else if (g_str_has_prefix(s, "-I=")) { rc = apply_include(s+3, error); } else if (g_str_has_prefix(s, "--exclude=")) { rc = apply_exclude(s+10, error); } else if (g_str_has_prefix(s, "-U=")) { rc = apply_exclude(s+3, error); } if (rc != TRUE) { break; } node = node->next; } return rc; } static gboolean user_include_exclude_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **err) { char *s = crm_strdup_printf("%s=%s", option_name, optarg); options.user_includes_excludes = g_slist_append(options.user_includes_excludes, s); return TRUE; } static gboolean include_exclude_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **err) { char *s = crm_strdup_printf("%s=%s", option_name, optarg); options.includes_excludes = g_slist_append(options.includes_excludes, s); return TRUE; } static gboolean as_xml_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **err) { pcmk__str_update(&args->output_ty, "xml"); output_format = mon_output_legacy_xml; return TRUE; } +static gboolean +pid_file_cb(const gchar *option_name, const gchar *optarg, gpointer data, + GError **err) +{ + return TRUE; +} + static gboolean fence_history_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **err) { if (optarg == NULL) { interactive_fence_level = 2; } else { pcmk__scan_min_int(optarg, &interactive_fence_level, 0); } switch (interactive_fence_level) { case 3: options.fence_connect = TRUE; fence_history = pcmk__fence_history_full; return include_exclude_cb("--include", PCMK_VALUE_FENCING, data, err); case 2: options.fence_connect = TRUE; fence_history = pcmk__fence_history_full; return include_exclude_cb("--include", PCMK_VALUE_FENCING, data, err); case 1: options.fence_connect = TRUE; fence_history = pcmk__fence_history_full; return include_exclude_cb("--include", "fencing-failed,fencing-pending", data, err); case 0: options.fence_connect = FALSE; fence_history = pcmk__fence_history_none; return include_exclude_cb("--exclude", PCMK_VALUE_FENCING, data, err); default: g_set_error(err, PCMK__EXITC_ERROR, CRM_EX_INVALID_PARAM, "Fence history must be 0-3"); return FALSE; } } static gboolean group_by_node_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **err) { show_opts |= pcmk_show_rscs_by_node; return TRUE; } static gboolean hide_headers_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **err) { return user_include_exclude_cb("--exclude", "summary", data, err); } static gboolean inactive_resources_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **err) { show_opts |= pcmk_show_inactive_rscs; return TRUE; } static gboolean print_brief_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **err) { show_opts |= pcmk_show_brief; return TRUE; } static gboolean print_detail_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **err) { show_opts |= pcmk_show_details; return TRUE; } static gboolean print_description_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **err) { show_opts |= pcmk_show_description; return TRUE; } static gboolean print_timing_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **err) { show_opts |= pcmk_show_timing; return user_include_exclude_cb("--include", "operations", data, err); } static gboolean reconnect_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **err) { int rc = crm_get_msec(optarg); if (rc == -1) { g_set_error(err, PCMK__EXITC_ERROR, CRM_EX_INVALID_PARAM, "Invalid value for -i: %s", optarg); return FALSE; } else { pcmk_parse_interval_spec(optarg, &options.reconnect_ms); if (options.exec_mode != mon_exec_daemonized) { // Reconnect interval applies to daemonized too, so don't override options.exec_mode = mon_exec_update; } } return TRUE; } /*! * \internal * \brief Enable one-shot mode * * \param[in] option_name Name of option being parsed (ignored) * \param[in] optarg Value to be parsed (ignored) * \param[in] data User data (ignored) * \param[out] err Where to store error (ignored) */ static gboolean one_shot_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **err) { options.exec_mode = mon_exec_one_shot; return TRUE; } /*! * \internal * \brief Enable daemonized mode * * \param[in] option_name Name of option being parsed (ignored) * \param[in] optarg Value to be parsed (ignored) * \param[in] data User data (ignored) * \param[out] err Where to store error (ignored) */ static gboolean daemonize_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **err) { options.exec_mode = mon_exec_daemonized; return TRUE; } static gboolean show_attributes_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **err) { return user_include_exclude_cb("--include", "attributes", data, err); } static gboolean show_bans_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **err) { if (optarg != NULL) { char *s = crm_strdup_printf("bans:%s", optarg); gboolean rc = user_include_exclude_cb("--include", s, data, err); free(s); return rc; } else { return user_include_exclude_cb("--include", "bans", data, err); } } static gboolean show_failcounts_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **err) { return user_include_exclude_cb("--include", "failcounts", data, err); } static gboolean show_operations_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **err) { return user_include_exclude_cb("--include", "failcounts,operations", data, err); } static gboolean show_tickets_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **err) { return user_include_exclude_cb("--include", "tickets", data, err); } static gboolean use_cib_file_cb(const gchar *option_name, const gchar *optarg, gpointer data, GError **err) { setenv("CIB_file", optarg, 1); options.exec_mode = mon_exec_one_shot; return TRUE; } #define INDENT " " /* *INDENT-OFF* */ static GOptionEntry addl_entries[] = { { "interval", 'i', 0, G_OPTION_ARG_CALLBACK, reconnect_cb, "Update frequency (default is 5 seconds). Note: When run interactively\n" INDENT "on a live cluster, the display will be updated automatically\n" INDENT "whenever the cluster configuration or status changes.", "TIMESPEC" }, { "one-shot", '1', G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, one_shot_cb, "Display the cluster status once and exit", NULL }, { "daemonize", 'd', G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, daemonize_cb, "Run in the background as a daemon.\n" INDENT "Requires at least one of --output-to and --external-agent.", NULL }, - { "pid-file", 'p', 0, G_OPTION_ARG_FILENAME, &options.pid_file, - "(Advanced) Daemon pid file location", - "FILE" }, - { "external-agent", 'E', 0, G_OPTION_ARG_FILENAME, &options.external_agent, "A program to run when resource operations take place", "FILE" }, { "external-recipient", 'e', 0, G_OPTION_ARG_STRING, &options.external_recipient, "A recipient for your program (assuming you want the program to send something to someone).", "RCPT" }, { "watch-fencing", 'W', 0, G_OPTION_ARG_NONE, &options.watch_fencing, "Listen for fencing events. For use with --external-agent.", NULL }, { "xml-file", 'x', G_OPTION_FLAG_HIDDEN, G_OPTION_ARG_CALLBACK, use_cib_file_cb, NULL, NULL }, + { "pid-file", 'p', G_OPTION_FLAG_HIDDEN|G_OPTION_FLAG_NO_ARG, + G_OPTION_ARG_CALLBACK, pid_file_cb, + "(deprecated)", "FILE" }, + { NULL } }; static GOptionEntry display_entries[] = { { "include", 'I', 0, G_OPTION_ARG_CALLBACK, user_include_exclude_cb, "A list of sections to include in the output.\n" INDENT "See `Output Control` help for more information.", "SECTION(s)" }, { "exclude", 'U', 0, G_OPTION_ARG_CALLBACK, user_include_exclude_cb, "A list of sections to exclude from the output.\n" INDENT "See `Output Control` help for more information.", "SECTION(s)" }, { "node", 0, 0, G_OPTION_ARG_STRING, &options.only_node, "When displaying information about nodes, show only what's related to the given\n" INDENT "node, or to all nodes tagged with the given tag", "NODE" }, { "resource", 0, 0, G_OPTION_ARG_STRING, &options.only_rsc, "When displaying information about resources, show only what's related to the given\n" INDENT "resource, or to all resources tagged with the given tag", "RSC" }, { "group-by-node", 'n', G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, group_by_node_cb, "Group resources by node", NULL }, { "inactive", 'r', G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, inactive_resources_cb, "Display inactive resources", NULL }, { "failcounts", 'f', G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, show_failcounts_cb, "Display resource fail counts", NULL }, { "operations", 'o', G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, show_operations_cb, "Display resource operation history", NULL }, { "timing-details", 't', G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, print_timing_cb, "Display resource operation history with timing details", NULL }, { "tickets", 'c', G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, show_tickets_cb, "Display cluster tickets", NULL }, { "fence-history", 'm', G_OPTION_FLAG_OPTIONAL_ARG, G_OPTION_ARG_CALLBACK, fence_history_cb, "Show fence history:\n" INDENT "0=off, 1=failures and pending (default without option),\n" INDENT "2=add successes (default without value for option),\n" INDENT "3=show full history without reduction to most recent of each flavor", "LEVEL" }, { "neg-locations", 'L', G_OPTION_FLAG_OPTIONAL_ARG, G_OPTION_ARG_CALLBACK, show_bans_cb, "Display negative location constraints [optionally filtered by id prefix]", NULL }, { "show-node-attributes", 'A', G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, show_attributes_cb, "Display node attributes", NULL }, { "hide-headers", 'D', G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, hide_headers_cb, "Hide all headers", NULL }, { "show-detail", 'R', G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, print_detail_cb, "Show more details (node IDs, individual clone instances)", NULL }, { "show-description", 0, G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, print_description_cb, "Show resource descriptions", NULL }, { "brief", 'b', G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, print_brief_cb, "Brief output", NULL }, { "pending", 'j', G_OPTION_FLAG_HIDDEN, G_OPTION_ARG_NONE, &options.print_pending, "Display pending state if '" PCMK_META_RECORD_PENDING "' is enabled", NULL }, - { NULL } -}; - -static GOptionEntry deprecated_entries[] = { /* @COMPAT resource-agents <4.15.0 uses --as-xml, so removing this option * must wait until we no longer support building on any platforms that ship * the older agents. + * + * Note: This enables one-shot mode. */ - { "as-xml", 'X', G_OPTION_FLAG_NO_ARG, G_OPTION_ARG_CALLBACK, as_xml_cb, - "Write cluster status as XML to stdout. This will enable one-shot mode.\n" - INDENT "Use --output-as=xml instead.", - NULL }, + { "as-xml", 'X', G_OPTION_FLAG_HIDDEN|G_OPTION_FLAG_NO_ARG, + G_OPTION_ARG_CALLBACK, as_xml_cb, + "(deprecated)" }, { NULL } }; /* *INDENT-ON* */ /* Reconnect to the CIB and fencing agent after reconnect_ms has passed. This sounds * like it would be more broadly useful, but only ever happens after a disconnect via * mon_cib_connection_destroy. */ static gboolean reconnect_after_timeout(gpointer data) { #if PCMK__ENABLE_CURSES if (output_format == mon_output_console) { clear(); refresh(); } #endif out->transient(out, "Reconnecting..."); if (setup_api_connections() == pcmk_rc_ok) { // Trigger redrawing the screen (needs reconnect_timer == 0) reconnect_timer = 0; refresh_after_event(FALSE, TRUE); return G_SOURCE_REMOVE; } out->message(out, "crm-mon-disconnected", "Latest connection attempt failed", pcmkd_state); reconnect_timer = pcmk__create_timer(options.reconnect_ms, reconnect_after_timeout, NULL); return G_SOURCE_REMOVE; } /* Called from various places when we are disconnected from the CIB or from the * fencing agent. If the CIB connection is still valid, this function will also * attempt to sign off and reconnect. */ static void mon_cib_connection_destroy(gpointer user_data) { const char *msg = "Connection to the cluster lost"; pcmkd_state = pcmk_pacemakerd_state_invalid; /* No crm-mon-disconnected message for console; a working implementation * is not currently worth the effort */ out->transient(out, "%s", msg); out->message(out, "crm-mon-disconnected", msg, pcmkd_state); if (refresh_timer != NULL) { /* we'll trigger a refresh after reconnect */ mainloop_timer_stop(refresh_timer); } if (reconnect_timer) { /* we'll trigger a new reconnect-timeout at the end */ g_source_remove(reconnect_timer); reconnect_timer = 0; } /* the client API won't properly reconnect notifications if they are still * in the table - so remove them */ if (st != NULL) { if (st->state != stonith_disconnected) { st->cmds->disconnect(st); } st->cmds->remove_notification(st, NULL); } if (cib) { cib->cmds->signoff(cib); reconnect_timer = pcmk__create_timer(options.reconnect_ms, reconnect_after_timeout, NULL); } } /* Signal handler installed into the mainloop for normal program shutdown */ static void mon_shutdown(int nsig) { clean_up(CRM_EX_OK); } #if PCMK__ENABLE_CURSES static volatile sighandler_t ncurses_winch_handler; /* Signal handler installed the regular way (not into the main loop) for when * the screen is resized. Commonly, this happens when running in an xterm and * the user changes its size. */ static void mon_winresize(int nsig) { static int not_done; int lines = 0, cols = 0; if (!not_done++) { if (ncurses_winch_handler) /* the original ncurses WINCH signal handler does the * magic of retrieving the new window size; * otherwise, we'd have to use ioctl or tgetent */ (*ncurses_winch_handler) (SIGWINCH); getmaxyx(stdscr, lines, cols); resizeterm(lines, cols); /* Alert the mainloop code we'd like the refresh_trigger to run next * time the mainloop gets around to checking. */ mainloop_set_trigger((crm_trigger_t *) refresh_trigger); } not_done--; } #endif static int setup_fencer_connection(void) { int rc = pcmk_ok; if (options.fence_connect && st == NULL) { st = stonith__api_new(); } if (!options.fence_connect || st == NULL || st->state != stonith_disconnected) { return rc; } rc = st->cmds->connect(st, crm_system_name, NULL); if (rc == pcmk_ok) { crm_trace("Setting up stonith callbacks"); if (options.watch_fencing) { st->cmds->register_notification(st, PCMK__VALUE_ST_NOTIFY_DISCONNECT, mon_st_callback_event); st->cmds->register_notification(st, PCMK__VALUE_ST_NOTIFY_FENCE, mon_st_callback_event); } else { st->cmds->register_notification(st, PCMK__VALUE_ST_NOTIFY_DISCONNECT, mon_st_callback_display); st->cmds->register_notification(st, PCMK__VALUE_ST_NOTIFY_HISTORY, mon_st_callback_display); } } else { stonith__api_free(st); st = NULL; } return rc; } static int setup_cib_connection(void) { int rc = pcmk_rc_ok; CRM_CHECK(cib != NULL, return EINVAL); if (cib->state != cib_disconnected) { // Already connected with notifications registered for CIB updates return rc; } rc = cib__signon_query(out, &cib, ¤t_cib); if (rc == pcmk_rc_ok) { rc = pcmk_legacy2rc(cib->cmds->set_connection_dnotify(cib, mon_cib_connection_destroy)); if (rc == EPROTONOSUPPORT) { out->err(out, "CIB client does not support connection loss " "notifications; crm_mon will be unable to reconnect after " "connection loss"); rc = pcmk_rc_ok; } if (rc == pcmk_rc_ok) { cib->cmds->del_notify_callback(cib, PCMK__VALUE_CIB_DIFF_NOTIFY, crm_diff_update); rc = cib->cmds->add_notify_callback(cib, PCMK__VALUE_CIB_DIFF_NOTIFY, crm_diff_update); rc = pcmk_legacy2rc(rc); } if (rc != pcmk_rc_ok) { if (rc == EPROTONOSUPPORT) { out->err(out, "CIB client does not support CIB diff " "notifications"); } else { out->err(out, "CIB diff notification setup failed"); } out->err(out, "Cannot monitor CIB changes; exiting"); cib__clean_up_connection(&cib); stonith__api_free(st); st = NULL; } } return rc; } /* This is used to set up the fencing options after the interactive UI has been stared. * fence_history_cb can't be used because it builds up a list of includes/excludes that * then have to be processed with apply_include_exclude and that could affect other * things. */ static void set_fencing_options(int level) { switch (level) { case 3: options.fence_connect = TRUE; fence_history = pcmk__fence_history_full; show |= pcmk_section_fencing_all; break; case 2: options.fence_connect = TRUE; fence_history = pcmk__fence_history_full; show |= pcmk_section_fencing_all; break; case 1: options.fence_connect = TRUE; fence_history = pcmk__fence_history_full; show |= pcmk_section_fence_failed | pcmk_section_fence_pending; break; default: interactive_fence_level = 0; options.fence_connect = FALSE; fence_history = pcmk__fence_history_none; show &= ~pcmk_section_fencing_all; break; } } static int setup_api_connections(void) { int rc = pcmk_rc_ok; CRM_CHECK(cib != NULL, return EINVAL); if (cib->state != cib_disconnected) { return rc; } if (cib->variant == cib_native) { rc = pcmk__pacemakerd_status(out, crm_system_name, options.reconnect_ms / 2, false, &pcmkd_state); if (rc != pcmk_rc_ok) { return rc; } switch (pcmkd_state) { case pcmk_pacemakerd_state_running: case pcmk_pacemakerd_state_remote: case pcmk_pacemakerd_state_shutting_down: /* Fencer and CIB may still be available while shutting down or * running on a Pacemaker Remote node */ break; default: // Fencer and CIB are definitely unavailable return ENOTCONN; } setup_fencer_connection(); } rc = setup_cib_connection(); return rc; } #if PCMK__ENABLE_CURSES static const char * get_option_desc(char c) { const char *desc = "No help available"; for (GOptionEntry *entry = display_entries; entry != NULL; entry++) { if (entry->short_name == c) { desc = entry->description; break; } } return desc; } #define print_option_help(out, option, condition) \ curses_formatted_printf(out, "%c %c: \t%s\n", ((condition)? '*': ' '), option, get_option_desc(option)); /* This function is called from the main loop when there is something to be read * on stdin, like an interactive user's keystroke. All it does is read the keystroke, * set flags (or show the page showing which keystrokes are valid), and redraw the * screen. It does not do anything with connections to the CIB or fencing agent * agent what would happen in mon_refresh_display. */ static gboolean detect_user_input(GIOChannel *channel, GIOCondition condition, gpointer user_data) { int c; gboolean config_mode = FALSE; gboolean rc = G_SOURCE_CONTINUE; /* If the attached pty device (pseudo-terminal) has been closed/deleted, * the condition (G_IO_IN | G_IO_ERR | G_IO_HUP) occurs. * Exit with an error, otherwise the process would persist in the * background and significantly raise the CPU usage. */ if ((condition & G_IO_ERR) && (condition & G_IO_HUP)) { rc = G_SOURCE_REMOVE; clean_up(CRM_EX_IOERR); } /* The connection/fd has been closed. Refresh the screen and remove this * event source hence ignore stdin. */ if (condition & (G_IO_HUP | G_IO_NVAL)) { rc = G_SOURCE_REMOVE; } if ((condition & G_IO_IN) == 0) { return rc; } while (1) { /* Get user input */ c = getchar(); switch (c) { case 'm': interactive_fence_level++; if (interactive_fence_level > 3) { interactive_fence_level = 0; } set_fencing_options(interactive_fence_level); break; case 'c': show ^= pcmk_section_tickets; break; case 'f': show ^= pcmk_section_failcounts; break; case 'n': show_opts ^= pcmk_show_rscs_by_node; break; case 'o': show ^= pcmk_section_operations; if (!pcmk_is_set(show, pcmk_section_operations)) { show_opts &= ~pcmk_show_timing; } break; case 'r': show_opts ^= pcmk_show_inactive_rscs; break; case 'R': show_opts ^= pcmk_show_details; break; case 't': show_opts ^= pcmk_show_timing; if (pcmk_is_set(show_opts, pcmk_show_timing)) { show |= pcmk_section_operations; } break; case 'A': show ^= pcmk_section_attributes; break; case 'L': show ^= pcmk_section_bans; break; case 'D': /* If any header is shown, clear them all, otherwise set them all */ if (pcmk_any_flags_set(show, pcmk_section_summary)) { show &= ~pcmk_section_summary; } else { show |= pcmk_section_summary; } /* Regardless, we don't show options in console mode. */ show &= ~pcmk_section_options; break; case 'b': show_opts ^= pcmk_show_brief; break; case 'j': show_opts ^= pcmk_show_pending; break; case '?': config_mode = TRUE; break; default: /* All other keys just redraw the screen. */ goto refresh; } if (!config_mode) goto refresh; clear(); refresh(); curses_formatted_printf(out, "%s", "Display option change mode\n"); print_option_help(out, 'c', pcmk_is_set(show, pcmk_section_tickets)); print_option_help(out, 'f', pcmk_is_set(show, pcmk_section_failcounts)); print_option_help(out, 'n', pcmk_is_set(show_opts, pcmk_show_rscs_by_node)); print_option_help(out, 'o', pcmk_is_set(show, pcmk_section_operations)); print_option_help(out, 'r', pcmk_is_set(show_opts, pcmk_show_inactive_rscs)); print_option_help(out, 't', pcmk_is_set(show_opts, pcmk_show_timing)); print_option_help(out, 'A', pcmk_is_set(show, pcmk_section_attributes)); print_option_help(out, 'L', pcmk_is_set(show, pcmk_section_bans)); print_option_help(out, 'D', !pcmk_is_set(show, pcmk_section_summary)); print_option_help(out, 'R', pcmk_any_flags_set(show_opts, pcmk_show_details)); print_option_help(out, 'b', pcmk_is_set(show_opts, pcmk_show_brief)); print_option_help(out, 'j', pcmk_is_set(show_opts, pcmk_show_pending)); curses_formatted_printf(out, "%d m: \t%s\n", interactive_fence_level, get_option_desc('m')); curses_formatted_printf(out, "%s", "\nToggle fields via field letter, type any other key to return\n"); } refresh: refresh_after_event(FALSE, TRUE); return rc; } #endif // PCMK__ENABLE_CURSES // Basically crm_signal_handler(SIGCHLD, SIG_IGN) plus the SA_NOCLDWAIT flag static void avoid_zombies(void) { struct sigaction sa; memset(&sa, 0, sizeof(struct sigaction)); if (sigemptyset(&sa.sa_mask) < 0) { crm_warn("Cannot avoid zombies: %s", pcmk_rc_str(errno)); return; } sa.sa_handler = SIG_IGN; sa.sa_flags = SA_RESTART|SA_NOCLDWAIT; if (sigaction(SIGCHLD, &sa, NULL) < 0) { crm_warn("Cannot avoid zombies: %s", pcmk_rc_str(errno)); } } static GOptionContext * build_arg_context(pcmk__common_args_t *args, GOptionGroup **group) { GOptionContext *context = NULL; GOptionEntry extra_prog_entries[] = { { "quiet", 'Q', 0, G_OPTION_ARG_NONE, &(args->quiet), "Be less descriptive in output.", NULL }, { NULL } }; #if PCMK__ENABLE_CURSES const char *fmts = "console (default), html, text, xml, none"; #else const char *fmts = "text (default), html, xml, none"; #endif // PCMK__ENABLE_CURSES const char *desc = NULL; desc = "Notes:\n\n" "Time Specification:\n\n" "The TIMESPEC in any command line option can be specified in many\n" "different formats. It can be an integer number of seconds, a\n" "number plus units (us/usec/ms/msec/s/sec/m/min/h/hr), or an ISO\n" "8601 period specification.\n\n" "Output Control:\n\n" "By default, a particular set of sections are written to the\n" "output destination. The default varies based on the output\n" "format: XML includes all sections by default, while other output\n" "formats include less. This set can be modified with the --include\n" "and --exclude command line options. Each option may be passed\n" "multiple times, and each can specify a comma-separated list of\n" "sections. The options are applied to the default set, in order\n" "from left to right as they are passed on the command line. For a\n" "list of valid sections, pass --include=list or --exclude=list.\n\n" "Interactive Use:\n\n" #if PCMK__ENABLE_CURSES "When run interactively, crm_mon can be told to hide and show\n" "various sections of output. To see a help screen explaining the\n" "options, press '?'. Any key stroke aside from those listed will\n" "cause the screen to refresh.\n\n" #else "The local installation of Pacemaker was built without support for\n" "interactive (console) mode. A curses library must be available at\n" "build time to support interactive mode.\n\n" #endif // PCMK__ENABLE_CURSES "Examples:\n\n" #if PCMK__ENABLE_CURSES "Display the cluster status on the console with updates as they\n" "occur:\n\n" "\tcrm_mon\n\n" #endif // PCMK__ENABLE_CURSES "Display the cluster status once and exit:\n\n" "\tcrm_mon -1\n\n" "Display the cluster status, group resources by node, and include\n" "inactive resources in the list:\n\n" "\tcrm_mon --group-by-node --inactive\n\n" "Start crm_mon as a background daemon and have it write the\n" "cluster status to an HTML file:\n\n" "\tcrm_mon --daemonize --output-as html " "--output-to /path/to/docroot/filename.html\n\n" "Display the cluster status as XML:\n\n" "\tcrm_mon --output-as xml\n\n"; context = pcmk__build_arg_context(args, fmts, group, NULL); pcmk__add_main_args(context, extra_prog_entries); g_option_context_set_description(context, desc); pcmk__add_arg_group(context, "display", "Display Options:", "Show display options", display_entries); pcmk__add_arg_group(context, "additional", "Additional Options:", "Show additional options", addl_entries); - pcmk__add_arg_group(context, "deprecated", "Deprecated Options:", - "Show deprecated options", deprecated_entries); return context; } /*! * \internal * \brief Set output format based on \c --output-as arguments and mode arguments * * When the deprecated \c --as-xml argument is parsed, a callback function sets * \c output_format. Otherwise, this function does the same based on the current * \c --output-as arguments and the \c --one-shot and \c --daemonize arguments. * * \param[in,out] args Command line arguments */ static void reconcile_output_format(pcmk__common_args_t *args) { if (output_format != mon_output_unset) { /* The deprecated --as-xml argument was used, and we're finished. Note * that this means the deprecated argument takes precedence. */ return; } if (pcmk__str_eq(args->output_ty, PCMK_VALUE_NONE, pcmk__str_none)) { output_format = mon_output_none; } else if (pcmk__str_eq(args->output_ty, "html", pcmk__str_none)) { output_format = mon_output_html; umask(S_IWGRP | S_IWOTH); // World-readable HTML } else if (pcmk__str_eq(args->output_ty, "xml", pcmk__str_none)) { output_format = mon_output_xml; #if PCMK__ENABLE_CURSES } else if (pcmk__str_eq(args->output_ty, "console", pcmk__str_null_matches)) { /* Console is the default format if no conflicting options are given. * * Use text output instead if one of the following conditions is met: * * We've requested daemonized or one-shot mode (console output is * incompatible with modes other than mon_exec_update) * * We requested the version, which is effectively one-shot + * * The CIB_file environment variable is set. We haven't created the + * cib object yet, so we can't simply check cib->variant, even though + * that abstraction feels cleaner than checking CIB_file. * * We specified a non-stdout output destination (console mode is * compatible only with stdout) */ if ((options.exec_mode == mon_exec_daemonized) || (options.exec_mode == mon_exec_one_shot) || args->version + || (getenv("CIB_file") != NULL) || !pcmk__str_eq(args->output_dest, "-", pcmk__str_null_matches)) { pcmk__str_update(&args->output_ty, "text"); output_format = mon_output_plain; } else { pcmk__str_update(&args->output_ty, "console"); output_format = mon_output_console; crm_enable_stderr(FALSE); } #endif // PCMK__ENABLE_CURSES } else if (pcmk__str_eq(args->output_ty, "text", pcmk__str_null_matches)) { /* Text output was explicitly requested, or it's the default because * curses is not enabled */ pcmk__str_update(&args->output_ty, "text"); output_format = mon_output_plain; } // Otherwise, invalid format. Let pcmk__output_new() throw an error. } /*! * \internal * \brief Set execution mode to the output format's default if appropriate * * \param[in,out] args Command line arguments */ static void set_default_exec_mode(const pcmk__common_args_t *args) { if (output_format == mon_output_console) { /* Update is the only valid mode for console, but set here instead of * reconcile_output_format() for isolation and consistency */ options.exec_mode = mon_exec_update; } else if (options.exec_mode == mon_exec_unset) { // Default to one-shot mode for all other formats options.exec_mode = mon_exec_one_shot; } else if ((options.exec_mode == mon_exec_update) && pcmk__str_eq(args->output_dest, "-", pcmk__str_null_matches)) { // If not using console format, update mode cannot be used with stdout options.exec_mode = mon_exec_one_shot; } } static void clean_up_on_connection_failure(int rc) { if (rc == ENOTCONN) { if (pcmkd_state == pcmk_pacemakerd_state_remote) { g_set_error(&error, PCMK__EXITC_ERROR, CRM_EX_ERROR, "Error: remote-node not connected to cluster"); } else { g_set_error(&error, PCMK__EXITC_ERROR, CRM_EX_ERROR, "Error: cluster is not available on this node"); } } else { g_set_error(&error, PCMK__EXITC_ERROR, CRM_EX_ERROR, "Connection to cluster failed: %s", pcmk_rc_str(rc)); } clean_up(pcmk_rc2exitc(rc)); } static void one_shot(void) { int rc = pcmk__status(out, cib, fence_history, show, show_opts, options.only_node, options.only_rsc, options.neg_location_prefix, 0); if (rc == pcmk_rc_ok) { clean_up(pcmk_rc2exitc(rc)); } else { clean_up_on_connection_failure(rc); } } -static void -exit_on_invalid_cib(void) -{ - if (cib != NULL) { - return; - } - - // Shouldn't really be possible - g_set_error(&error, PCMK__EXITC_ERROR, CRM_EX_ERROR, "Invalid CIB source"); - clean_up(CRM_EX_ERROR); -} - int main(int argc, char **argv) { int rc = pcmk_rc_ok; GOptionGroup *output_group = NULL; args = pcmk__new_common_args(SUMMARY); context = build_arg_context(args, &output_group); pcmk__register_formats(output_group, formats); - options.pid_file = strdup("/tmp/ClusterMon.pid"); pcmk__cli_init_logging("crm_mon", 0); // Avoid needing to wait for subprocesses forked for -E/--external-agent avoid_zombies(); - processed_args = pcmk__cmdline_preproc(argv, "eimpxEILU"); - fence_history_cb("--fence-history", "1", NULL, NULL); /* Set an HTML title regardless of what format we will eventually use. * Doing this here means the user can give their own title on the command * line. */ - if (!pcmk__force_args(context, &error, "%s --html-title \"Cluster Status\"", - g_get_prgname())) { - return clean_up(CRM_EX_USAGE); - } + pcmk__html_set_title("Cluster Status"); + + processed_args = pcmk__cmdline_preproc(argv, "eimpxEILU"); if (!g_option_context_parse_strv(context, &processed_args, &error)) { return clean_up(CRM_EX_USAGE); } for (int i = 0; i < args->verbosity; i++) { crm_bump_log_level(argc, argv); } - if (!args->version) { - if (args->quiet) { - include_exclude_cb("--exclude", "times", NULL, NULL); - } - - if (options.watch_fencing) { - fence_history_cb("--fence-history", "0", NULL, NULL); - options.fence_connect = TRUE; - } - - /* create the cib-object early to be able to do further - * decisions based on the cib-source - */ - cib = cib_new(); - - exit_on_invalid_cib(); - - switch (cib->variant) { - case cib_native: - // Everything (fencer, CIB, pcmkd status) should be available - break; - - case cib_file: - // Live fence history is not meaningful - fence_history_cb("--fence-history", "0", NULL, NULL); - - /* Notifications are unsupported; nothing to monitor - * @COMPAT: Let setup_cib_connection() handle this by exiting? - */ - options.exec_mode = mon_exec_one_shot; - break; - - case cib_remote: - // We won't receive any fencing updates - fence_history_cb("--fence-history", "0", NULL, NULL); - break; - - default: - /* something is odd */ - exit_on_invalid_cib(); - break; - } - - if ((options.exec_mode == mon_exec_daemonized) - && !options.external_agent - && pcmk__str_eq(args->output_dest, "-", pcmk__str_null_matches)) { - - g_set_error(&error, PCMK__EXITC_ERROR, CRM_EX_USAGE, - "--daemonize requires at least one of --output-to " - "(with value not set to '-') and --external-agent"); - return clean_up(CRM_EX_USAGE); - } - } - reconcile_output_format(args); set_default_exec_mode(args); rc = pcmk__output_new(&out, args->output_ty, args->output_dest, argv); if (rc != pcmk_rc_ok) { - g_set_error(&error, PCMK__EXITC_ERROR, CRM_EX_ERROR, "Error creating output format %s: %s", + g_set_error(&error, PCMK__EXITC_ERROR, CRM_EX_ERROR, + "Error creating output format %s: %s", args->output_ty, pcmk_rc_str(rc)); return clean_up(CRM_EX_ERROR); } if (output_format == mon_output_legacy_xml) { output_format = mon_output_xml; pcmk__output_set_legacy_xml(out); } /* output_format MUST NOT BE CHANGED AFTER THIS POINT. */ /* If we had a valid format for pcmk__output_new(), output_format should be * set by now. */ pcmk__assert(output_format != mon_output_unset); if (output_format == mon_output_plain) { pcmk__output_text_set_fancy(out, true); } - if (options.exec_mode == mon_exec_daemonized) { - if (!options.external_agent && (output_format == mon_output_none)) { - g_set_error(&error, PCMK__EXITC_ERROR, CRM_EX_USAGE, - "--daemonize requires --external-agent if used with " - "--output-as=none"); - return clean_up(CRM_EX_USAGE); - } - crm_enable_stderr(FALSE); - cib_delete(cib); - cib = NULL; - pcmk__daemonize(crm_system_name, options.pid_file); - cib = cib_new(); - exit_on_invalid_cib(); + pcmk__register_lib_messages(out); + crm_mon_register_messages(out); + pe__register_messages(out); + stonith__register_messages(out); + + // Messages internal to this file, nothing curses-specific + pcmk__register_messages(out, fmt_functions); + + if (args->version) { + out->version(out); + return clean_up(CRM_EX_OK); + } + + if (args->quiet) { + include_exclude_cb("--exclude", "times", NULL, NULL); + } + + if (options.watch_fencing) { + fence_history_cb("--fence-history", "0", NULL, NULL); + options.fence_connect = TRUE; } show = default_includes(output_format); /* Apply --include/--exclude flags we used internally. There's no error reporting * here because this would be a programming error. */ apply_include_exclude(options.includes_excludes, &error); /* And now apply any --include/--exclude flags the user gave on the command line. * These are done in a separate pass from the internal ones because we want to * make sure whatever the user specifies overrides whatever we do. */ if (!apply_include_exclude(options.user_includes_excludes, &error)) { return clean_up(CRM_EX_USAGE); } /* Sync up the initial value of interactive_fence_level with whatever was set with * --include/--exclude= options. */ if (pcmk_all_flags_set(show, pcmk_section_fencing_all)) { interactive_fence_level = 3; } else if (pcmk_is_set(show, pcmk_section_fence_worked)) { interactive_fence_level = 2; } else if (pcmk_any_flags_set(show, pcmk_section_fence_failed | pcmk_section_fence_pending)) { interactive_fence_level = 1; } else { interactive_fence_level = 0; } - pcmk__register_lib_messages(out); - crm_mon_register_messages(out); - pe__register_messages(out); - stonith__register_messages(out); - - // Messages internal to this file, nothing curses-specific - pcmk__register_messages(out, fmt_functions); - - if (args->version) { - out->version(out); - return clean_up(CRM_EX_OK); - } - if (output_format == mon_output_xml) { show_opts |= pcmk_show_inactive_rscs | pcmk_show_timing; } if ((output_format == mon_output_html) && (out->dest != stdout)) { char *content = pcmk__itoa(pcmk__timeout_ms2s(options.reconnect_ms)); pcmk__html_add_header(PCMK__XE_META, PCMK__XA_HTTP_EQUIV, PCMK__VALUE_REFRESH, PCMK__XA_CONTENT, content, NULL); free(content); } - crm_info("Starting %s", crm_system_name); + if (options.exec_mode == mon_exec_daemonized) { + pid_t pid = 0; + + if (options.external_agent == NULL) { + if (pcmk__str_eq(args->output_dest, "-", pcmk__str_null_matches)) { + g_set_error(&error, PCMK__EXITC_ERROR, CRM_EX_USAGE, + "--daemonize requires at least one of --output-to " + "(with value not set to '-') and --external-agent"); + return clean_up(CRM_EX_USAGE); + } + if (output_format == mon_output_none) { + g_set_error(&error, PCMK__EXITC_ERROR, CRM_EX_USAGE, + "--daemonize requires --external-agent if used " + "with --output-as=none"); + return clean_up(CRM_EX_USAGE); + } + } + + crm_enable_stderr(FALSE); + + pid = fork(); + if (pid < 0) { + g_set_error(&error, PCMK__EXITC_ERROR, CRM_EX_OSERR, + "Could not fork daemon: %s", strerror(errno)); + clean_up(CRM_EX_OSERR); + } + if (pid > 0) { + clean_up(CRM_EX_OK); + } + umask(S_IWGRP|S_IWOTH|S_IROTH); + pcmk__null_std_streams(); + } + + cib = cib_new(); + if (cib == NULL) { + /* For the foreseeable future, out-of-memory is the only possible + * reason for NULL return value + */ + rc = ENOMEM; + g_set_error(&error, PCMK__RC_ERROR, rc, + "Failed to create CIB API connection object: %s", + pcmk_rc_str(rc)); + clean_up(pcmk_rc2exitc(rc)); + } cib__set_output(cib, out); + switch (cib->variant) { + case cib_native: + // Everything (fencer, CIB, pcmkd status) should be available + break; + + case cib_file: + // Live fence history is not meaningful + fence_history_cb("--fence-history", "0", NULL, NULL); + + /* Notifications are unsupported; nothing to monitor + * @COMPAT: Let setup_cib_connection() handle this by exiting? + */ + options.exec_mode = mon_exec_one_shot; + break; + + case cib_remote: + // We won't receive any fencing updates + fence_history_cb("--fence-history", "0", NULL, NULL); + break; + + default: + // Should not be possible; would indicate a bug in CIB library + CRM_CHECK(false, clean_up(CRM_EX_SOFTWARE)); + break; + } + + crm_info("Starting %s", crm_system_name); + if (options.exec_mode == mon_exec_one_shot) { + // Needs cib but not scheduler one_shot(); } scheduler = pcmk_new_scheduler(); pcmk__mem_assert(scheduler); scheduler->priv->out = out; if ((cib->variant == cib_native) && pcmk_is_set(show, pcmk_section_times)) { // Currently used only in the times section pcmk__query_node_name(out, 0, &(scheduler->priv->local_node_name), 0); } out->message(out, "crm-mon-disconnected", "Waiting for initial connection", pcmkd_state); do { out->transient(out, "Connecting to cluster..."); rc = setup_api_connections(); if (rc != pcmk_rc_ok) { if ((rc == ENOTCONN) || (rc == ECONNREFUSED)) { out->transient(out, "Connection failed. Retrying in %s...", pcmk__readable_interval(options.reconnect_ms)); } // Give some time to view all output even if we won't retry pcmk__sleep_ms(options.reconnect_ms); #if PCMK__ENABLE_CURSES if (output_format == mon_output_console) { clear(); refresh(); } #endif } } while ((rc == ENOTCONN) || (rc == ECONNREFUSED)); if (rc != pcmk_rc_ok) { clean_up_on_connection_failure(rc); } set_fencing_options(interactive_fence_level); mon_refresh_display(NULL); mainloop = g_main_loop_new(NULL, FALSE); mainloop_add_signal(SIGTERM, mon_shutdown); mainloop_add_signal(SIGINT, mon_shutdown); #if PCMK__ENABLE_CURSES if (output_format == mon_output_console) { ncurses_winch_handler = crm_signal_handler(SIGWINCH, mon_winresize); if (ncurses_winch_handler == SIG_DFL || ncurses_winch_handler == SIG_IGN || ncurses_winch_handler == SIG_ERR) ncurses_winch_handler = NULL; io_channel = g_io_channel_unix_new(STDIN_FILENO); g_io_add_watch(io_channel, (G_IO_IN | G_IO_ERR | G_IO_HUP | G_IO_NVAL), detect_user_input, NULL); } #endif /* When refresh_trigger->trigger is set to TRUE, call mon_refresh_display. In * this file, that is anywhere mainloop_set_trigger is called. */ refresh_trigger = mainloop_add_trigger(G_PRIORITY_LOW, mon_refresh_display, NULL); g_main_loop_run(mainloop); g_main_loop_unref(mainloop); crm_info("Exiting %s", crm_system_name); return clean_up(CRM_EX_OK); } static int send_custom_trap(const char *node, const char *rsc, const char *task, int target_rc, int rc, int status, const char *desc) { pid_t pid; /*setenv needs chars, these are ints */ char *rc_s = pcmk__itoa(rc); char *status_s = pcmk__itoa(status); char *target_rc_s = pcmk__itoa(target_rc); crm_debug("Sending external notification to '%s' via '%s'", options.external_recipient, options.external_agent); if(rsc) { setenv("CRM_notify_rsc", rsc, 1); } - if (options.external_recipient) { + if (options.external_recipient != NULL) { setenv("CRM_notify_recipient", options.external_recipient, 1); } setenv("CRM_notify_node", node, 1); setenv("CRM_notify_task", task, 1); setenv("CRM_notify_desc", desc, 1); setenv("CRM_notify_rc", rc_s, 1); setenv("CRM_notify_target_rc", target_rc_s, 1); setenv("CRM_notify_status", status_s, 1); pid = fork(); if (pid == -1) { out->err(out, "notification fork() failed: %s", strerror(errno)); } if (pid == 0) { /* crm_debug("notification: I am the child. Executing the nofitication program."); */ execl(options.external_agent, options.external_agent, NULL); crm_exit(CRM_EX_ERROR); } crm_trace("Finished running custom notification program '%s'.", options.external_agent); free(target_rc_s); free(status_s); free(rc_s); return 0; } static int handle_rsc_op(xmlNode *xml, void *userdata) { const char *node_id = (const char *) userdata; int rc = -1; int status = -1; int target_rc = -1; gboolean notify = TRUE; char *rsc = NULL; char *task = NULL; const char *desc = NULL; const char *magic = NULL; const char *id = NULL; const char *node = NULL; xmlNode *n = xml; xmlNode * rsc_op = xml; if(strcmp((const char*)xml->name, PCMK__XE_LRM_RSC_OP) != 0) { pcmk__xe_foreach_child(xml, NULL, handle_rsc_op, (void *) node_id); return pcmk_rc_ok; } id = pcmk__xe_history_key(rsc_op); magic = pcmk__xe_get(rsc_op, PCMK__XA_TRANSITION_MAGIC); if (magic == NULL) { /* non-change */ return pcmk_rc_ok; } if (!decode_transition_magic(magic, NULL, NULL, NULL, &status, &rc, &target_rc)) { crm_err("Invalid event %s detected for %s", magic, id); return pcmk_rc_ok; } if (parse_op_key(id, &rsc, &task, NULL) == FALSE) { crm_err("Invalid event detected for %s", id); goto bail; } node = pcmk__xe_get(rsc_op, PCMK__META_ON_NODE); while ((n != NULL) && !pcmk__xe_is(n, PCMK__XE_NODE_STATE)) { n = n->parent; } if(node == NULL && n) { node = pcmk__xe_get(n, PCMK_XA_UNAME); } if (node == NULL && n) { node = pcmk__xe_id(n); } if (node == NULL) { node = node_id; } if (node == NULL) { crm_err("No node detected for event %s (%s)", magic, id); goto bail; } /* look up where we expected it to be? */ desc = pcmk_rc_str(pcmk_rc_ok); if ((status == PCMK_EXEC_DONE) && (target_rc == rc)) { crm_notice("%s of %s on %s completed: %s", task, rsc, node, desc); if (rc == PCMK_OCF_NOT_RUNNING) { notify = FALSE; } } else if (status == PCMK_EXEC_DONE) { desc = crm_exit_str(rc); crm_warn("%s of %s on %s failed: %s", task, rsc, node, desc); } else { desc = pcmk_exec_status_str(status); crm_warn("%s of %s on %s failed: %s", task, rsc, node, desc); } - if (notify && options.external_agent) { + if (notify && (options.external_agent != NULL)) { send_custom_trap(node, rsc, task, target_rc, rc, status, desc); } bail: free(rsc); free(task); return pcmk_rc_ok; } /* This function is just a wrapper around mainloop_set_trigger so that it can be * called from a mainloop directly. It's simply another way of ensuring the screen * gets redrawn. */ static gboolean mon_trigger_refresh(gpointer user_data) { mainloop_set_trigger((crm_trigger_t *) refresh_trigger); return FALSE; } static int handle_op_for_node(xmlNode *xml, void *userdata) { const char *node = pcmk__xe_get(xml, PCMK_XA_UNAME); if (node == NULL) { node = pcmk__xe_id(xml); } handle_rsc_op(xml, (void *) node); return pcmk_rc_ok; } static int crm_diff_update_element(xmlNode *change, void *userdata) { const char *name = NULL; const char *op = pcmk__xe_get(change, PCMK_XA_OPERATION); const char *xpath = pcmk__xe_get(change, PCMK_XA_PATH); xmlNode *match = NULL; const char *node = NULL; if (op == NULL) { return pcmk_rc_ok; } else if (strcmp(op, PCMK_VALUE_CREATE) == 0) { match = change->children; } else if (pcmk__str_any_of(op, PCMK_VALUE_MOVE, PCMK_VALUE_DELETE, NULL)) { return pcmk_rc_ok; } else if (strcmp(op, PCMK_VALUE_MODIFY) == 0) { match = pcmk__xe_first_child(change, PCMK_XE_CHANGE_RESULT, NULL, NULL); if(match) { match = match->children; } } if(match) { name = (const char *)match->name; } crm_trace("Handling %s operation for %s %p, %s", op, xpath, match, name); if(xpath == NULL) { /* Version field, ignore */ } else if(name == NULL) { crm_debug("No result for %s operation to %s", op, xpath); pcmk__assert(pcmk__str_any_of(op, PCMK_VALUE_MOVE, PCMK_VALUE_DELETE, NULL)); } else if (strcmp(name, PCMK_XE_CIB) == 0) { pcmk__xe_foreach_child(pcmk__xe_first_child(match, PCMK_XE_STATUS, NULL, NULL), NULL, handle_op_for_node, NULL); } else if (strcmp(name, PCMK_XE_STATUS) == 0) { pcmk__xe_foreach_child(match, NULL, handle_op_for_node, NULL); } else if (strcmp(name, PCMK__XE_NODE_STATE) == 0) { node = pcmk__xe_get(match, PCMK_XA_UNAME); if (node == NULL) { node = pcmk__xe_id(match); } handle_rsc_op(match, (void *) node); } else if (strcmp(name, PCMK__XE_LRM) == 0) { node = pcmk__xe_id(match); handle_rsc_op(match, (void *) node); } else if (strcmp(name, PCMK__XE_LRM_RESOURCES) == 0) { char *local_node = pcmk__xpath_node_id(xpath, PCMK__XE_LRM); handle_rsc_op(match, local_node); free(local_node); } else if (strcmp(name, PCMK__XE_LRM_RESOURCE) == 0) { char *local_node = pcmk__xpath_node_id(xpath, PCMK__XE_LRM); handle_rsc_op(match, local_node); free(local_node); } else if (strcmp(name, PCMK__XE_LRM_RSC_OP) == 0) { char *local_node = pcmk__xpath_node_id(xpath, PCMK__XE_LRM); handle_rsc_op(match, local_node); free(local_node); } else { crm_trace("Ignoring %s operation for %s %p, %s", op, xpath, match, name); } return pcmk_rc_ok; } static void crm_diff_update(const char *event, xmlNode * msg) { int rc = -1; static bool stale = FALSE; gboolean cib_updated = FALSE; xmlNode *wrapper = pcmk__xe_first_child(msg, PCMK__XE_CIB_UPDATE_RESULT, NULL, NULL); xmlNode *diff = pcmk__xe_first_child(wrapper, NULL, NULL, NULL); out->progress(out, false); if (current_cib != NULL) { rc = xml_apply_patchset(current_cib, diff, TRUE); switch (rc) { case -pcmk_err_diff_resync: case -pcmk_err_diff_failed: crm_notice("[%s] Patch aborted: %s (%d)", event, pcmk_strerror(rc), rc); pcmk__xml_free(current_cib); current_cib = NULL; break; case pcmk_ok: cib_updated = TRUE; break; default: crm_notice("[%s] ABORTED: %s (%d)", event, pcmk_strerror(rc), rc); pcmk__xml_free(current_cib); current_cib = NULL; } } if (current_cib == NULL) { crm_trace("Re-requesting the full cib"); cib->cmds->query(cib, NULL, ¤t_cib, cib_sync_call); } - if (options.external_agent) { + if (options.external_agent != NULL) { int format = 0; pcmk__xe_get_int(diff, PCMK_XA_FORMAT, &format); if (format == 2) { xmlNode *wrapper = pcmk__xe_first_child(msg, PCMK__XE_CIB_UPDATE_RESULT, NULL, NULL); xmlNode *diff = pcmk__xe_first_child(wrapper, NULL, NULL, NULL); pcmk__xe_foreach_child(diff, NULL, crm_diff_update_element, NULL); } else { crm_err("Unknown patch format: %d", format); } } if (current_cib == NULL) { if(!stale) { out->info(out, "--- Stale data ---"); } stale = TRUE; return; } stale = FALSE; refresh_after_event(cib_updated, FALSE); } static int mon_refresh_display(gpointer user_data) { int rc = pcmk_rc_ok; last_refresh = time(NULL); if (output_format == mon_output_none) { return G_SOURCE_REMOVE; } if (fence_history == pcmk__fence_history_full && !pcmk_all_flags_set(show, pcmk_section_fencing_all) && output_format != mon_output_xml) { fence_history = pcmk__fence_history_reduced; } // Get an up-to-date pacemakerd status for the cluster summary if (cib->variant == cib_native) { pcmk__pacemakerd_status(out, crm_system_name, options.reconnect_ms / 2, false, &pcmkd_state); } if (out->dest != stdout) { out->reset(out); } rc = pcmk__output_cluster_status(scheduler, st, cib, current_cib, pcmkd_state, fence_history, show, show_opts, options.only_node,options.only_rsc, options.neg_location_prefix); if (rc == pcmk_rc_schema_validation) { clean_up(CRM_EX_CONFIG); return G_SOURCE_REMOVE; } if (out->dest != stdout) { out->finish(out, CRM_EX_OK, true, NULL); } return G_SOURCE_CONTINUE; } /* This function is called for fencing events (see setup_fencer_connection() for * which ones) when --watch-fencing is used on the command line */ static void mon_st_callback_event(stonith_t * st, stonith_event_t * e) { if (st->state == stonith_disconnected) { /* disconnect cib as well and have everything reconnect */ mon_cib_connection_destroy(NULL); - } else if (options.external_agent) { + } else if (options.external_agent != NULL) { char *desc = stonith__event_description(e); send_custom_trap(e->target, NULL, e->operation, pcmk_ok, e->result, 0, desc); free(desc); } } /* Cause the screen to be redrawn (via mainloop_set_trigger) when various conditions are met: * * - If the last update occurred more than reconnect_ms ago (defaults to 5s, but * can be changed via the -i command line option), or * - After every 10 CIB updates, or * - If it's been 2s since the last update * * This function sounds like it would be more broadly useful, but it is only called when a * fencing event is received or a CIB diff occurrs. */ static void refresh_after_event(gboolean data_updated, gboolean enforce) { static int updates = 0; time_t now = time(NULL); if (data_updated) { updates++; } if(refresh_timer == NULL) { refresh_timer = mainloop_timer_add("refresh", 2000, FALSE, mon_trigger_refresh, NULL); } if (reconnect_timer > 0) { /* we will receive a refresh request after successful reconnect */ mainloop_timer_stop(refresh_timer); return; } /* as we're not handling initial failure of fencer-connection as * fatal give it a retry here * not getting here if cib-reconnection is already on the way */ setup_fencer_connection(); if (enforce || ((now - last_refresh) > pcmk__timeout_ms2s(options.reconnect_ms)) || updates >= 10) { mainloop_set_trigger((crm_trigger_t *) refresh_trigger); mainloop_timer_stop(refresh_timer); updates = 0; } else { mainloop_timer_start(refresh_timer); } } /* This function is called for fencing events (see setup_fencer_connection() for * which ones) when --watch-fencing is NOT used on the command line */ static void mon_st_callback_display(stonith_t * st, stonith_event_t * e) { if (st->state == stonith_disconnected) { /* disconnect cib as well and have everything reconnect */ mon_cib_connection_destroy(NULL); } else { out->progress(out, false); refresh_after_event(TRUE, FALSE); } } /* * De-init ncurses, disconnect from the CIB manager, disconnect fencing, * deallocate memory and show usage-message if requested. * * We don't actually return, but nominally returning crm_exit_t allows a usage * like "return clean_up(exit_code);" which helps static analysis understand the * code flow. */ static crm_exit_t clean_up(crm_exit_t exit_code) { /* Quitting crm_mon is much more complicated than it ought to be. */ + const bool has_error = (error != NULL); /* (1) Close connections, free things, etc. */ if (io_channel != NULL) { g_io_channel_shutdown(io_channel, TRUE, NULL); } cib__clean_up_connection(&cib); stonith__api_free(st); + g_free(options.external_agent); + g_free(options.external_recipient); free(options.neg_location_prefix); - free(options.only_node); - free(options.only_rsc); - free(options.pid_file); + g_free(options.only_node); + g_free(options.only_rsc); g_slist_free_full(options.includes_excludes, free); g_strfreev(processed_args); pcmk_free_scheduler(scheduler); /* (2) If this is abnormal termination and we're in curses mode, shut down * curses first. Any messages displayed to the screen before curses is shut * down will be lost because doing the shut down will also restore the * screen to whatever it looked like before crm_mon was started. */ - if (((error != NULL) || (exit_code == CRM_EX_USAGE)) + if ((has_error || (exit_code == CRM_EX_USAGE)) && (output_format == mon_output_console) && (out != NULL)) { out->finish(out, exit_code, false, NULL); pcmk__output_free(out); out = NULL; } /* (3) If this is a command line usage related failure, print the usage * message. */ if (exit_code == CRM_EX_USAGE && (output_format == mon_output_console || output_format == mon_output_plain)) { char *help = g_option_context_get_help(context, TRUE, NULL); fprintf(stderr, "%s", help); g_free(help); } pcmk__free_arg_context(context); - /* (4) If this is any kind of error, print the error out and exit. Make - * sure to handle situations both before and after formatted output is - * set up. We want errors to appear formatted if at all possible. - */ - if (error != NULL) { - if (out != NULL) { - out->err(out, "%s: %s", g_get_prgname(), error->message); - out->finish(out, exit_code, true, NULL); - pcmk__output_free(out); - } else { - fprintf(stderr, "%s: %s\n", g_get_prgname(), error->message); - } - - g_clear_error(&error); - crm_exit(exit_code); - } - - /* (5) Print formatted output to the screen if we made it far enough in - * crm_mon to be able to do so. + /* (4) Output the error if one exists. Finish the output unless this is a + * successful daemonized child. Clean up the output object and exit. */ + pcmk__output_and_clear_error(&error, out); if (out != NULL) { - if (options.exec_mode != mon_exec_daemonized) { + if (has_error || (options.exec_mode != mon_exec_daemonized)) { out->finish(out, exit_code, true, NULL); } - pcmk__output_free(out); } pcmk__unregister_formats(); crm_exit(exit_code); }