diff --git a/daemons/based/based_io.c b/daemons/based/based_io.c index f55722b284..7ea3cf83ec 100644 --- a/daemons/based/based_io.c +++ b/daemons/based/based_io.c @@ -1,462 +1,463 @@ /* * Copyright 2004-2024 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 #include #include #include #include #include #include crm_trigger_t *cib_writer = NULL; int write_cib_contents(gpointer p); static void cib_rename(const char *old) { int new_fd; char *new = crm_strdup_printf("%s/cib.auto.XXXXXX", cib_root); umask(S_IWGRP | S_IWOTH | S_IROTH); new_fd = mkstemp(new); if ((new_fd < 0) || (rename(old, new) < 0)) { crm_err("Couldn't archive unusable file %s (disabling disk writes and continuing)", old); cib_writes_enabled = FALSE; } else { crm_err("Archived unusable file %s as %s", old, new); } if (new_fd > 0) { close(new_fd); } free(new); } /* * It is the callers responsibility to free the output of this function */ static xmlNode * retrieveCib(const char *filename, const char *sigfile) { xmlNode *root = NULL; crm_info("Reading cluster configuration file %s (digest: %s)", filename, sigfile); switch (cib_file_read_and_verify(filename, sigfile, &root)) { case -pcmk_err_cib_corrupt: crm_warn("Continuing but %s will NOT be used.", filename); break; case -pcmk_err_cib_modified: /* Archive the original files so the contents are not lost */ crm_warn("Continuing but %s will NOT be used.", filename); cib_rename(filename); cib_rename(sigfile); break; } return root; } /* * for OSs without support for direntry->d_type, like Solaris */ #ifndef DT_UNKNOWN # define DT_UNKNOWN 0 # define DT_FIFO 1 # define DT_CHR 2 # define DT_DIR 4 # define DT_BLK 6 # define DT_REG 8 # define DT_LNK 10 # define DT_SOCK 12 # define DT_WHT 14 #endif /*DT_UNKNOWN*/ static int cib_archive_filter(const struct dirent * a) { int rc = 0; /* Looking for regular files (d_type = 8) starting with 'cib-' and not ending in .sig */ struct stat s; char *a_path = crm_strdup_printf("%s/%s", cib_root, a->d_name); if(stat(a_path, &s) != 0) { rc = errno; crm_trace("%s - stat failed: %s (%d)", a->d_name, pcmk_rc_str(rc), rc); rc = 0; } else if ((s.st_mode & S_IFREG) != S_IFREG) { unsigned char dtype; #ifdef HAVE_STRUCT_DIRENT_D_TYPE dtype = a->d_type; #else switch (s.st_mode & S_IFMT) { case S_IFREG: dtype = DT_REG; break; case S_IFDIR: dtype = DT_DIR; break; case S_IFCHR: dtype = DT_CHR; break; case S_IFBLK: dtype = DT_BLK; break; case S_IFLNK: dtype = DT_LNK; break; case S_IFIFO: dtype = DT_FIFO; break; case S_IFSOCK: dtype = DT_SOCK; break; default: dtype = DT_UNKNOWN; break; } #endif crm_trace("%s - wrong type (%d)", a->d_name, dtype); } else if(strstr(a->d_name, "cib-") != a->d_name) { crm_trace("%s - wrong prefix", a->d_name); } else if (pcmk__ends_with_ext(a->d_name, ".sig")) { crm_trace("%s - wrong suffix", a->d_name); } else { crm_debug("%s - candidate", a->d_name); rc = 1; } free(a_path); return rc; } static int cib_archive_sort(const struct dirent ** a, const struct dirent **b) { /* Order by creation date - most recently created file first */ int rc = 0; struct stat buf; time_t a_age = 0; time_t b_age = 0; char *a_path = crm_strdup_printf("%s/%s", cib_root, a[0]->d_name); char *b_path = crm_strdup_printf("%s/%s", cib_root, b[0]->d_name); if(stat(a_path, &buf) == 0) { a_age = buf.st_ctime; } if(stat(b_path, &buf) == 0) { b_age = buf.st_ctime; } free(a_path); free(b_path); if(a_age > b_age) { rc = 1; } else if(a_age < b_age) { rc = -1; } crm_trace("%s (%lu) vs. %s (%lu) : %d", a[0]->d_name, (unsigned long)a_age, b[0]->d_name, (unsigned long)b_age, rc); return rc; } xmlNode * readCibXmlFile(const char *dir, const char *file, gboolean discard_status) { struct dirent **namelist = NULL; int lpc = 0; char *sigfile = NULL; char *sigfilepath = NULL; char *filename = NULL; const char *name = NULL; const char *value = NULL; const char *use_valgrind = pcmk__env_option(PCMK__ENV_VALGRIND_ENABLED); xmlNode *root = NULL; xmlNode *status = NULL; sigfile = crm_strdup_printf("%s.sig", file); if (pcmk__daemon_can_write(dir, file) == FALSE || pcmk__daemon_can_write(dir, sigfile) == FALSE) { cib_status = -EACCES; return NULL; } filename = crm_strdup_printf("%s/%s", dir, file); sigfilepath = crm_strdup_printf("%s/%s", dir, sigfile); free(sigfile); cib_status = pcmk_ok; root = retrieveCib(filename, sigfilepath); free(filename); free(sigfilepath); if (root == NULL) { crm_warn("Primary configuration corrupt or unusable, trying backups in %s", cib_root); lpc = scandir(cib_root, &namelist, cib_archive_filter, cib_archive_sort); if (lpc < 0) { crm_err("scandir(%s) failed: %s", cib_root, pcmk_rc_str(errno)); } } while (root == NULL && lpc > 1) { crm_debug("Testing %d candidates", lpc); lpc--; filename = crm_strdup_printf("%s/%s", cib_root, namelist[lpc]->d_name); sigfile = crm_strdup_printf("%s.sig", filename); crm_info("Reading cluster configuration file %s (digest: %s)", filename, sigfile); if (cib_file_read_and_verify(filename, sigfile, &root) < 0) { crm_warn("Continuing but %s will NOT be used.", filename); } else { crm_notice("Continuing with last valid configuration archive: %s", filename); } free(namelist[lpc]); free(filename); free(sigfile); } free(namelist); if (root == NULL) { root = createEmptyCib(0); crm_warn("Continuing with an empty configuration."); } if (cib_writes_enabled && (use_valgrind != NULL) && (crm_is_true(use_valgrind) || (strstr(use_valgrind, PCMK__SERVER_BASED) != NULL))) { cib_writes_enabled = FALSE; crm_err("*** Disabling disk writes to avoid confusing Valgrind ***"); } status = pcmk__xe_first_child(root, PCMK_XE_STATUS, NULL, NULL); if (discard_status && status != NULL) { // Strip out the PCMK_XE_STATUS section if there is one pcmk__xml_free(status); status = NULL; } if (status == NULL) { pcmk__xe_create(root, PCMK_XE_STATUS); } /* Do this before schema validation happens */ /* fill in some defaults */ name = PCMK_XA_ADMIN_EPOCH; value = crm_element_value(root, name); if (value == NULL) { crm_warn("No value for %s was specified in the configuration.", name); crm_warn("The recommended course of action is to shutdown," " run crm_verify and fix any errors it reports."); crm_warn("We will default to zero and continue but may get" " confused about which configuration to use if" " multiple nodes are powered up at the same time."); crm_xml_add_int(root, name, 0); } name = PCMK_XA_EPOCH; value = crm_element_value(root, name); if (value == NULL) { crm_xml_add_int(root, name, 0); } name = PCMK_XA_NUM_UPDATES; value = crm_element_value(root, name); if (value == NULL) { crm_xml_add_int(root, name, 0); } // Unset (DC should set appropriate value) pcmk__xe_remove_attr(root, PCMK_XA_DC_UUID); if (discard_status) { crm_log_xml_trace(root, "[on-disk]"); } if (!pcmk__configured_schema_validates(root)) { cib_status = -pcmk_err_schema_validation; } return root; } gboolean uninitializeCib(void) { xmlNode *tmp_cib = the_cib; if (tmp_cib == NULL) { crm_debug("The CIB has already been deallocated."); return FALSE; } the_cib = NULL; crm_debug("Deallocating the CIB."); pcmk__xml_free(tmp_cib); crm_debug("The CIB has been deallocated."); return TRUE; } /* * This method will free the old CIB pointer on success and the new one * on failure. */ int activateCibXml(xmlNode * new_cib, gboolean to_disk, const char *op) { if (new_cib) { xmlNode *saved_cib = the_cib; pcmk__assert(new_cib != saved_cib); the_cib = new_cib; pcmk__xml_free(saved_cib); if (cib_writes_enabled && cib_status == pcmk_ok && to_disk) { crm_debug("Triggering CIB write for %s op", op); mainloop_set_trigger(cib_writer); } return pcmk_ok; } crm_err("Ignoring invalid CIB"); if (the_cib) { crm_warn("Reverting to last known CIB"); } else { crm_crit("Could not write out new CIB and no saved version to revert to"); } return -ENODATA; } static void cib_diskwrite_complete(mainloop_child_t * p, pid_t pid, int core, int signo, int exitcode) { const char *errmsg = "Could not write CIB to disk"; if ((exitcode != 0) && cib_writes_enabled) { cib_writes_enabled = FALSE; errmsg = "Disabling CIB disk writes after failure"; } if ((signo == 0) && (exitcode == 0)) { crm_trace("Disk write [%d] succeeded", (int) pid); } else if (signo == 0) { crm_err("%s: process %d exited %d", errmsg, (int) pid, exitcode); } else { crm_err("%s: process %d terminated with signal %d (%s)%s", errmsg, (int) pid, signo, strsignal(signo), (core? " and dumped core" : "")); } mainloop_trigger_complete(cib_writer); } int write_cib_contents(gpointer p) { int exit_rc = pcmk_ok; xmlNode *cib_local = NULL; /* Make a copy of the CIB to write (possibly in a forked child) */ if (p) { /* Synchronous write out */ cib_local = pcmk__xml_copy(NULL, p); } else { int pid = 0; int bb_state = qb_log_ctl(QB_LOG_BLACKBOX, QB_LOG_CONF_STATE_GET, 0); /* Turn it off before the fork() to avoid: * - 2 processes writing to the same shared mem * - the child needing to disable it * (which would close it from underneath the parent) * This way, the shared mem files are already closed */ qb_log_ctl(QB_LOG_BLACKBOX, QB_LOG_CONF_ENABLED, QB_FALSE); pid = fork(); if (pid < 0) { crm_err("Disabling disk writes after fork failure: %s", pcmk_rc_str(errno)); cib_writes_enabled = FALSE; return FALSE; } if (pid) { /* Parent */ mainloop_child_add(pid, 0, "disk-writer", NULL, cib_diskwrite_complete); if (bb_state == QB_LOG_STATE_ENABLED) { /* Re-enable now that it it safe */ qb_log_ctl(QB_LOG_BLACKBOX, QB_LOG_CONF_ENABLED, QB_TRUE); } return -1; /* -1 means 'still work to do' */ } /* Asynchronous write-out after a fork() */ /* In theory, we can scribble on the_cib here and not affect the parent, * but let's be safe anyway. */ cib_local = pcmk__xml_copy(NULL, the_cib); } /* Write the CIB */ exit_rc = cib_file_write_with_digest(cib_local, cib_root, "cib.xml"); /* A nonzero exit code will cause further writes to be disabled */ pcmk__xml_free(cib_local); if (p == NULL) { crm_exit_t exit_code = CRM_EX_OK; switch (exit_rc) { case pcmk_ok: exit_code = CRM_EX_OK; break; case pcmk_err_cib_modified: exit_code = CRM_EX_DIGEST; // Existing CIB doesn't match digest break; case pcmk_err_cib_backup: // Existing CIB couldn't be backed up case pcmk_err_cib_save: // New CIB couldn't be saved exit_code = CRM_EX_CANTCREAT; break; default: exit_code = CRM_EX_ERROR; break; } /* Use _exit() because exit() could affect the parent adversely */ + pcmk_common_cleanup(); _exit(exit_code); } return exit_rc; } diff --git a/daemons/controld/controld_join_dc.c b/daemons/controld/controld_join_dc.c index dfb4ae6cc3..7ada26949d 100644 --- a/daemons/controld/controld_join_dc.c +++ b/daemons/controld/controld_join_dc.c @@ -1,1092 +1,1095 @@ /* * Copyright 2004-2024 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 // PRIu32 #include // bool, true, false #include // NULL #include // free(), etc. #include // gboolean, etc. #include // xmlNode #include #include #include #include static char *max_generation_from = NULL; static xmlNodePtr max_generation_xml = NULL; /*! * \internal * \brief Nodes from which a CIB sync has failed since the peer joined * * This table is of the form (node_name -> join_id). \p node_name is * the name of a client node from which a CIB \p sync_from() call has failed in * \p do_dc_join_finalize() since the client joined the cluster as a peer. * \p join_id is the ID of the join round in which the \p sync_from() failed, * and is intended for use in nack log messages. */ static GHashTable *failed_sync_nodes = NULL; void finalize_join_for(gpointer key, gpointer value, gpointer user_data); void finalize_sync_callback(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data); gboolean check_join_state(enum crmd_fsa_state cur_state, const char *source); /* Numeric counter used to identify join rounds (an unsigned int would be * appropriate, except we get and set it in XML as int) */ static int current_join_id = 0; /*! * \internal * \brief Get log-friendly string equivalent of a controller group join phase * * \param[in] phase Join phase * * \return Log-friendly string equivalent of \p phase */ static const char * join_phase_text(enum controld_join_phase phase) { switch (phase) { case controld_join_nack: return "nack"; case controld_join_none: return "none"; case controld_join_welcomed: return "welcomed"; case controld_join_integrated: return "integrated"; case controld_join_finalized: return "finalized"; case controld_join_confirmed: return "confirmed"; default: return "invalid"; } } /*! * \internal * \brief Destroy the hash table containing failed sync nodes */ void controld_destroy_failed_sync_table(void) { if (failed_sync_nodes != NULL) { g_hash_table_destroy(failed_sync_nodes); failed_sync_nodes = NULL; } } /*! * \internal * \brief Remove a node from the failed sync nodes table if present * * \param[in] node_name Node name to remove */ void controld_remove_failed_sync_node(const char *node_name) { if (failed_sync_nodes != NULL) { g_hash_table_remove(failed_sync_nodes, (gchar *) node_name); } } /*! * \internal * \brief Add to a hash table a node whose CIB failed to sync * * \param[in] node_name Name of node whose CIB failed to sync * \param[in] join_id Join round when the failure occurred */ static void record_failed_sync_node(const char *node_name, gint join_id) { if (failed_sync_nodes == NULL) { failed_sync_nodes = pcmk__strikey_table(g_free, NULL); } /* If the node is already in the table then we failed to nack it during the * filter offer step */ CRM_LOG_ASSERT(g_hash_table_insert(failed_sync_nodes, g_strdup(node_name), GINT_TO_POINTER(join_id))); } /*! * \internal * \brief Look up a node name in the failed sync table * * \param[in] node_name Name of node to look up * \param[out] join_id Where to store the join ID of when the sync failed * * \return Standard Pacemaker return code. Specifically, \p pcmk_rc_ok if the * node name was found, or \p pcmk_rc_node_unknown otherwise. * \note \p *join_id is set to -1 if the node is not found. */ static int lookup_failed_sync_node(const char *node_name, gint *join_id) { *join_id = -1; if (failed_sync_nodes != NULL) { gpointer result = g_hash_table_lookup(failed_sync_nodes, (gchar *) node_name); if (result != NULL) { *join_id = GPOINTER_TO_INT(result); return pcmk_rc_ok; } } return pcmk_rc_node_unknown; } void crm_update_peer_join(const char *source, pcmk__node_status_t *node, enum controld_join_phase phase) { enum controld_join_phase last = controld_get_join_phase(node); CRM_CHECK(node != NULL, return); /* Remote nodes do not participate in joins */ if (pcmk_is_set(node->flags, pcmk__node_status_remote)) { return; } if (phase == last) { crm_trace("Node %s join-%d phase is still %s " QB_XS " nodeid=%" PRIu32 " source=%s", node->name, current_join_id, join_phase_text(last), node->cluster_layer_id, source); return; } if ((phase <= controld_join_none) || (phase == (last + 1))) { - struct controld_node_status_data *data = - pcmk__assert_alloc(1, sizeof(struct controld_node_status_data)); + struct controld_node_status_data *data = NULL; + if (node->user_data == NULL) { + node->user_data = + pcmk__assert_alloc(1, sizeof(struct controld_node_status_data)); + } + data = node->user_data; data->join_phase = phase; - node->user_data = data; crm_trace("Node %s join-%d phase is now %s (was %s) " QB_XS " nodeid=%" PRIu32 " source=%s", node->name, current_join_id, join_phase_text(phase), join_phase_text(last), node->cluster_layer_id, source); return; } crm_warn("Rejecting join-%d phase update for node %s because can't go from " "%s to %s " QB_XS " nodeid=%" PRIu32 " source=%s", current_join_id, node->name, join_phase_text(last), join_phase_text(phase), node->cluster_layer_id, source); } static void start_join_round(void) { GHashTableIter iter; pcmk__node_status_t *peer = NULL; crm_debug("Starting new join round join-%d", current_join_id); g_hash_table_iter_init(&iter, pcmk__peer_cache); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &peer)) { crm_update_peer_join(__func__, peer, controld_join_none); } if (max_generation_from != NULL) { free(max_generation_from); max_generation_from = NULL; } if (max_generation_xml != NULL) { pcmk__xml_free(max_generation_xml); max_generation_xml = NULL; } controld_clear_fsa_input_flags(R_HAVE_CIB); } /*! * \internal * \brief Create a join message from the DC * * \param[in] join_op Join operation name * \param[in] host_to Recipient of message */ static xmlNode * create_dc_message(const char *join_op, const char *host_to) { xmlNode *msg = pcmk__new_request(pcmk_ipc_controld, CRM_SYSTEM_DC, host_to, CRM_SYSTEM_CRMD, join_op, NULL); /* Identify which election this is a part of */ crm_xml_add_int(msg, PCMK__XA_JOIN_ID, current_join_id); /* Add a field specifying whether the DC is shutting down. This keeps the * joining node from fencing the old DC if it becomes the new DC. */ pcmk__xe_set_bool_attr(msg, PCMK__XA_DC_LEAVING, pcmk_is_set(controld_globals.fsa_input_register, R_SHUTDOWN)); return msg; } static void join_make_offer(gpointer key, gpointer value, gpointer user_data) { /* @TODO We don't use user_data except to distinguish one particular call * from others. Make this clearer. */ xmlNode *offer = NULL; pcmk__node_status_t *member = (pcmk__node_status_t *) value; pcmk__assert(member != NULL); if (!pcmk__cluster_is_node_active(member)) { crm_info("Not making join-%d offer to inactive node %s", current_join_id, pcmk__s(member->name, "with unknown name")); if ((member->expected == NULL) && pcmk__str_eq(member->state, PCMK__VALUE_LOST, pcmk__str_none)) { /* You would think this unsafe, but in fact this plus an * active resource is what causes it to be fenced. * * Yes, this does mean that any node that dies at the same * time as the old DC and is not running resource (still) * won't be fenced. * * I'm not happy about this either. */ pcmk__update_peer_expected(__func__, member, CRMD_JOINSTATE_DOWN); } return; } if (member->name == NULL) { crm_info("Not making join-%d offer to node uuid %s with unknown name", current_join_id, member->xml_id); return; } if (controld_globals.membership_id != controld_globals.peer_seq) { controld_globals.membership_id = controld_globals.peer_seq; crm_info("Making join-%d offers based on membership event %llu", current_join_id, controld_globals.peer_seq); } if (user_data != NULL) { enum controld_join_phase phase = controld_get_join_phase(member); if (phase > controld_join_none) { crm_info("Not making join-%d offer to already known node %s (%s)", current_join_id, member->name, join_phase_text(phase)); return; } } crm_update_peer_join(__func__, (pcmk__node_status_t*) member, controld_join_none); offer = create_dc_message(CRM_OP_JOIN_OFFER, member->name); // Advertise our feature set so the joining node can bail if not compatible crm_xml_add(offer, PCMK_XA_CRM_FEATURE_SET, CRM_FEATURE_SET); crm_info("Sending join-%d offer to %s", current_join_id, member->name); pcmk__cluster_send_message(member, pcmk_ipc_controld, offer); pcmk__xml_free(offer); crm_update_peer_join(__func__, member, controld_join_welcomed); } /* A_DC_JOIN_OFFER_ALL */ void do_dc_join_offer_all(long long action, enum crmd_fsa_cause cause, enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data) { int count; /* Reset everyone's status back to down or in_ccm in the CIB. * Any nodes that are active in the CIB but not in the cluster membership * will be seen as offline by the scheduler anyway. */ current_join_id++; start_join_round(); update_dc(NULL); if (cause == C_HA_MESSAGE && current_input == I_NODE_JOIN) { crm_info("A new node joined the cluster"); } g_hash_table_foreach(pcmk__peer_cache, join_make_offer, NULL); count = crmd_join_phase_count(controld_join_welcomed); crm_info("Waiting on join-%d requests from %d outstanding node%s", current_join_id, count, pcmk__plural_s(count)); // Don't waste time by invoking the scheduler yet } /* A_DC_JOIN_OFFER_ONE */ void do_dc_join_offer_one(long long action, enum crmd_fsa_cause cause, enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data) { pcmk__node_status_t *member = NULL; ha_msg_input_t *welcome = NULL; int count; const char *join_to = NULL; if (msg_data->data == NULL) { crm_info("Making join-%d offers to any unconfirmed nodes " "because an unknown node joined", current_join_id); g_hash_table_foreach(pcmk__peer_cache, join_make_offer, &member); check_join_state(cur_state, __func__); return; } welcome = fsa_typed_data(fsa_dt_ha_msg); if (welcome == NULL) { // fsa_typed_data() already logged an error return; } join_to = crm_element_value(welcome->msg, PCMK__XA_SRC); if (join_to == NULL) { crm_err("Can't make join-%d offer to unknown node", current_join_id); return; } member = pcmk__get_node(0, join_to, NULL, pcmk__node_search_cluster_member); /* It is possible that a node will have been sick or starting up when the * original offer was made. However, it will either re-announce itself in * due course, or we can re-store the original offer on the client. */ crm_update_peer_join(__func__, member, controld_join_none); join_make_offer(NULL, member, NULL); /* If the offer isn't to the local node, make an offer to the local node as * well, to ensure the correct value for max_generation_from. */ if (!controld_is_local_node(join_to)) { member = controld_get_local_node_status(); join_make_offer(NULL, member, NULL); } /* This was a genuine join request; cancel any existing transition and * invoke the scheduler. */ abort_transition(PCMK_SCORE_INFINITY, pcmk__graph_restart, "Node join", NULL); count = crmd_join_phase_count(controld_join_welcomed); crm_info("Waiting on join-%d requests from %d outstanding node%s", current_join_id, count, pcmk__plural_s(count)); // Don't waste time by invoking the scheduler yet } static int compare_int_fields(xmlNode * left, xmlNode * right, const char *field) { const char *elem_l = crm_element_value(left, field); const char *elem_r = crm_element_value(right, field); long long int_elem_l; long long int_elem_r; int rc = pcmk_rc_ok; rc = pcmk__scan_ll(elem_l, &int_elem_l, -1LL); if (rc != pcmk_rc_ok) { // Shouldn't be possible crm_warn("Comparing current CIB %s as -1 " "because '%s' is not an integer", field, elem_l); } rc = pcmk__scan_ll(elem_r, &int_elem_r, -1LL); if (rc != pcmk_rc_ok) { // Shouldn't be possible crm_warn("Comparing joining node's CIB %s as -1 " "because '%s' is not an integer", field, elem_r); } if (int_elem_l < int_elem_r) { return -1; } else if (int_elem_l > int_elem_r) { return 1; } return 0; } /* A_DC_JOIN_PROCESS_REQ */ void do_dc_join_filter_offer(long long action, enum crmd_fsa_cause cause, enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data) { xmlNode *generation = NULL; int cmp = 0; int join_id = -1; int count = 0; gint value = 0; gboolean ack_nack_bool = TRUE; ha_msg_input_t *join_ack = fsa_typed_data(fsa_dt_ha_msg); const char *join_from = crm_element_value(join_ack->msg, PCMK__XA_SRC); const char *ref = crm_element_value(join_ack->msg, PCMK_XA_REFERENCE); const char *join_version = crm_element_value(join_ack->msg, PCMK_XA_CRM_FEATURE_SET); pcmk__node_status_t *join_node = NULL; if (join_from == NULL) { crm_err("Ignoring invalid join request without node name"); return; } join_node = pcmk__get_node(0, join_from, NULL, pcmk__node_search_cluster_member); crm_element_value_int(join_ack->msg, PCMK__XA_JOIN_ID, &join_id); if (join_id != current_join_id) { crm_debug("Ignoring join-%d request from %s because we are on join-%d", join_id, join_from, current_join_id); check_join_state(cur_state, __func__); return; } generation = join_ack->xml; if (max_generation_xml != NULL && generation != NULL) { int lpc = 0; const char *attributes[] = { PCMK_XA_ADMIN_EPOCH, PCMK_XA_EPOCH, PCMK_XA_NUM_UPDATES, }; /* It's not obvious that join_ack->xml is the PCMK__XE_GENERATION_TUPLE * element from the join client. The "if" guard is for clarity. */ if (pcmk__xe_is(generation, PCMK__XE_GENERATION_TUPLE)) { for (lpc = 0; cmp == 0 && lpc < PCMK__NELEM(attributes); lpc++) { cmp = compare_int_fields(max_generation_xml, generation, attributes[lpc]); } } else { // Should always be PCMK__XE_GENERATION_TUPLE CRM_LOG_ASSERT(false); } } if (ref == NULL) { ref = "none"; // for logging only } if (lookup_failed_sync_node(join_from, &value) == pcmk_rc_ok) { crm_err("Rejecting join-%d request from node %s because we failed to " "sync its CIB in join-%d " QB_XS " ref=%s", join_id, join_from, value, ref); ack_nack_bool = FALSE; } else if (!pcmk__cluster_is_node_active(join_node)) { if (match_down_event(join_from) != NULL) { /* The join request was received after the node was fenced or * otherwise shutdown in a way that we're aware of. No need to log * an error in this rare occurrence; we know the client was recently * shut down, and receiving a lingering in-flight request is not * cause for alarm. */ crm_debug("Rejecting join-%d request from inactive node %s " QB_XS " ref=%s", join_id, join_from, ref); } else { crm_err("Rejecting join-%d request from inactive node %s " QB_XS " ref=%s", join_id, join_from, ref); } ack_nack_bool = FALSE; } else if (generation == NULL) { crm_err("Rejecting invalid join-%d request from node %s " "missing CIB generation " QB_XS " ref=%s", join_id, join_from, ref); ack_nack_bool = FALSE; } else if ((join_version == NULL) || !feature_set_compatible(CRM_FEATURE_SET, join_version)) { crm_err("Rejecting join-%d request from node %s because feature set %s" " is incompatible with ours (%s) " QB_XS " ref=%s", join_id, join_from, (join_version? join_version : "pre-3.1.0"), CRM_FEATURE_SET, ref); ack_nack_bool = FALSE; } else if (max_generation_xml == NULL) { const char *validation = crm_element_value(generation, PCMK_XA_VALIDATE_WITH); if (pcmk__get_schema(validation) == NULL) { crm_err("Rejecting join-%d request from %s (with first CIB " "generation) due to %s schema version %s " QB_XS " ref=%s", join_id, join_from, ((validation == NULL)? "missing" : "unknown"), pcmk__s(validation, ""), ref); ack_nack_bool = FALSE; } else { crm_debug("Accepting join-%d request from %s (with first CIB " "generation) " QB_XS " ref=%s", join_id, join_from, ref); max_generation_xml = pcmk__xml_copy(NULL, generation); pcmk__str_update(&max_generation_from, join_from); } } else if ((cmp < 0) || ((cmp == 0) && controld_is_local_node(join_from))) { const char *validation = crm_element_value(generation, PCMK_XA_VALIDATE_WITH); if (pcmk__get_schema(validation) == NULL) { crm_err("Rejecting join-%d request from %s (with better CIB " "generation than current best from %s) due to %s " "schema version %s " QB_XS " ref=%s", join_id, join_from, max_generation_from, ((validation == NULL)? "missing" : "unknown"), pcmk__s(validation, ""), ref); ack_nack_bool = FALSE; } else { crm_debug("Accepting join-%d request from %s (with better CIB " "generation than current best from %s) " QB_XS " ref=%s", join_id, join_from, max_generation_from, ref); crm_log_xml_debug(max_generation_xml, "Old max generation"); crm_log_xml_debug(generation, "New max generation"); pcmk__xml_free(max_generation_xml); max_generation_xml = pcmk__xml_copy(NULL, join_ack->xml); pcmk__str_update(&max_generation_from, join_from); } } else { crm_debug("Accepting join-%d request from %s " QB_XS " ref=%s", join_id, join_from, ref); } if (!ack_nack_bool) { crm_update_peer_join(__func__, join_node, controld_join_nack); pcmk__update_peer_expected(__func__, join_node, CRMD_JOINSTATE_NACK); } else { crm_update_peer_join(__func__, join_node, controld_join_integrated); pcmk__update_peer_expected(__func__, join_node, CRMD_JOINSTATE_MEMBER); } count = crmd_join_phase_count(controld_join_integrated); crm_debug("%d node%s currently integrated in join-%d", count, pcmk__plural_s(count), join_id); if (check_join_state(cur_state, __func__) == FALSE) { // Don't waste time by invoking the scheduler yet count = crmd_join_phase_count(controld_join_welcomed); crm_debug("Waiting on join-%d requests from %d outstanding node%s", join_id, count, pcmk__plural_s(count)); } } /* A_DC_JOIN_FINALIZE */ void do_dc_join_finalize(long long action, enum crmd_fsa_cause cause, enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data) { char *sync_from = NULL; int rc = pcmk_ok; int count_welcomed = crmd_join_phase_count(controld_join_welcomed); int count_finalizable = crmd_join_phase_count(controld_join_integrated) + crmd_join_phase_count(controld_join_nack); /* This we can do straight away and avoid clients timing us out * while we compute the latest CIB */ if (count_welcomed != 0) { crm_debug("Waiting on join-%d requests from %d outstanding node%s " "before finalizing join", current_join_id, count_welcomed, pcmk__plural_s(count_welcomed)); crmd_join_phase_log(LOG_DEBUG); /* crmd_fsa_stall(FALSE); Needed? */ return; } else if (count_finalizable == 0) { crm_debug("Finalization not needed for join-%d at the current time", current_join_id); crmd_join_phase_log(LOG_DEBUG); check_join_state(controld_globals.fsa_state, __func__); return; } controld_clear_fsa_input_flags(R_HAVE_CIB); if ((max_generation_from == NULL) || controld_is_local_node(max_generation_from)) { controld_set_fsa_input_flags(R_HAVE_CIB); } if (!controld_globals.transition_graph->complete) { crm_warn("Delaying join-%d finalization while transition in progress", current_join_id); crmd_join_phase_log(LOG_DEBUG); crmd_fsa_stall(FALSE); return; } if (pcmk_is_set(controld_globals.fsa_input_register, R_HAVE_CIB)) { // Send our CIB out to everyone sync_from = pcmk__str_copy(controld_globals.cluster->priv->node_name); crm_debug("Finalizing join-%d for %d node%s (sync'ing from local CIB)", current_join_id, count_finalizable, pcmk__plural_s(count_finalizable)); crm_log_xml_debug(max_generation_xml, "Requested CIB version"); } else { // Ask for the agreed best CIB sync_from = pcmk__str_copy(max_generation_from); crm_notice("Finalizing join-%d for %d node%s (sync'ing CIB from %s)", current_join_id, count_finalizable, pcmk__plural_s(count_finalizable), sync_from); crm_log_xml_notice(max_generation_xml, "Requested CIB version"); } crmd_join_phase_log(LOG_DEBUG); rc = controld_globals.cib_conn->cmds->sync_from(controld_globals.cib_conn, sync_from, NULL, cib_none); fsa_register_cib_callback(rc, sync_from, finalize_sync_callback); } void free_max_generation(void) { free(max_generation_from); max_generation_from = NULL; pcmk__xml_free(max_generation_xml); max_generation_xml = NULL; } void finalize_sync_callback(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data) { CRM_LOG_ASSERT(-EPERM != rc); if (rc != pcmk_ok) { const char *sync_from = (const char *) user_data; do_crm_log(((rc == -pcmk_err_old_data)? LOG_WARNING : LOG_ERR), "Could not sync CIB from %s in join-%d: %s", sync_from, current_join_id, pcmk_strerror(rc)); if (rc != -pcmk_err_old_data) { record_failed_sync_node(sync_from, current_join_id); } /* restart the whole join process */ register_fsa_error_adv(C_FSA_INTERNAL, I_ELECTION_DC, NULL, NULL, __func__); } else if (!AM_I_DC) { crm_debug("Sync'ed CIB for join-%d but no longer DC", current_join_id); } else if (controld_globals.fsa_state != S_FINALIZE_JOIN) { crm_debug("Sync'ed CIB for join-%d but no longer in S_FINALIZE_JOIN " "(%s)", current_join_id, fsa_state2string(controld_globals.fsa_state)); } else { controld_set_fsa_input_flags(R_HAVE_CIB); /* make sure dc_uuid is re-set to us */ if (!check_join_state(controld_globals.fsa_state, __func__)) { int count_finalizable = 0; count_finalizable = crmd_join_phase_count(controld_join_integrated) + crmd_join_phase_count(controld_join_nack); crm_debug("Notifying %d node%s of join-%d results", count_finalizable, pcmk__plural_s(count_finalizable), current_join_id); g_hash_table_foreach(pcmk__peer_cache, finalize_join_for, NULL); } } } static void join_node_state_commit_callback(xmlNode *msg, int call_id, int rc, xmlNode *output, void *user_data) { const char *node = user_data; if (rc != pcmk_ok) { fsa_data_t *msg_data = NULL; // for register_fsa_error() macro crm_crit("join-%d node history update (via CIB call %d) for node %s " "failed: %s", current_join_id, call_id, node, pcmk_strerror(rc)); crm_log_xml_debug(msg, "failed"); register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); } crm_debug("join-%d node history update (via CIB call %d) for node %s " "complete", current_join_id, call_id, node); check_join_state(controld_globals.fsa_state, __func__); } /* A_DC_JOIN_PROCESS_ACK */ void do_dc_join_ack(long long action, enum crmd_fsa_cause cause, enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data) { int join_id = -1; ha_msg_input_t *join_ack = fsa_typed_data(fsa_dt_ha_msg); const char *op = crm_element_value(join_ack->msg, PCMK__XA_CRM_TASK); char *join_from = crm_element_value_copy(join_ack->msg, PCMK__XA_SRC); pcmk__node_status_t *peer = NULL; enum controld_join_phase phase = controld_join_none; enum controld_section_e section = controld_section_lrm; char *xpath = NULL; xmlNode *state = join_ack->xml; xmlNode *execd_state = NULL; cib_t *cib = controld_globals.cib_conn; int rc = pcmk_ok; // Sanity checks if (join_from == NULL) { crm_warn("Ignoring message received without node identification"); goto done; } if (op == NULL) { crm_warn("Ignoring message received from %s without task", join_from); goto done; } if (strcmp(op, CRM_OP_JOIN_CONFIRM)) { crm_debug("Ignoring '%s' message from %s while waiting for '%s'", op, join_from, CRM_OP_JOIN_CONFIRM); goto done; } if (crm_element_value_int(join_ack->msg, PCMK__XA_JOIN_ID, &join_id) != 0) { crm_warn("Ignoring join confirmation from %s without valid join ID", join_from); goto done; } peer = pcmk__get_node(0, join_from, NULL, pcmk__node_search_cluster_member); phase = controld_get_join_phase(peer); if (phase != controld_join_finalized) { crm_info("Ignoring out-of-sequence join-%d confirmation from %s " "(currently %s not %s)", join_id, join_from, join_phase_text(phase), join_phase_text(controld_join_finalized)); goto done; } if (join_id != current_join_id) { crm_err("Rejecting join-%d confirmation from %s " "because currently on join-%d", join_id, join_from, current_join_id); crm_update_peer_join(__func__, peer, controld_join_nack); goto done; } crm_update_peer_join(__func__, peer, controld_join_confirmed); /* Update CIB with node's current executor state. A new transition will be * triggered later, when the CIB manager notifies us of the change. * * The delete and modify requests are part of an atomic transaction. */ rc = cib->cmds->init_transaction(cib); if (rc != pcmk_ok) { goto done; } // Delete relevant parts of node's current executor state from CIB if (pcmk_is_set(controld_globals.flags, controld_shutdown_lock_enabled)) { section = controld_section_lrm_unlocked; } controld_node_state_deletion_strings(join_from, section, &xpath, NULL); rc = cib->cmds->remove(cib, xpath, NULL, cib_xpath|cib_multiple|cib_transaction); if (rc != pcmk_ok) { goto done; } // Update CIB with node's latest known executor state if (controld_is_local_node(join_from)) { // Use the latest possible state if processing our own join ack execd_state = controld_query_executor_state(); if (execd_state != NULL) { crm_debug("Updating local node history for join-%d from query " "result", current_join_id); state = execd_state; } else { crm_warn("Updating local node history from join-%d confirmation " "because query failed", current_join_id); } } else { crm_debug("Updating node history for %s from join-%d confirmation", join_from, current_join_id); } rc = cib->cmds->modify(cib, PCMK_XE_STATUS, state, cib_can_create|cib_transaction); pcmk__xml_free(execd_state); if (rc != pcmk_ok) { goto done; } // Commit the transaction rc = cib->cmds->end_transaction(cib, true, cib_none); fsa_register_cib_callback(rc, join_from, join_node_state_commit_callback); if (rc > 0) { // join_from will be freed after callback join_from = NULL; rc = pcmk_ok; } done: if (rc != pcmk_ok) { crm_crit("join-%d node history update for node %s failed: %s", current_join_id, join_from, pcmk_strerror(rc)); register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); } free(join_from); free(xpath); } void finalize_join_for(gpointer key, gpointer value, gpointer user_data) { xmlNode *acknak = NULL; xmlNode *tmp1 = NULL; pcmk__node_status_t *join_node = value; const char *join_to = join_node->name; enum controld_join_phase phase = controld_get_join_phase(join_node); bool integrated = false; switch (phase) { case controld_join_integrated: integrated = true; break; case controld_join_nack: break; default: crm_trace("Not updating non-integrated and non-nacked node %s (%s) " "for join-%d", join_to, join_phase_text(phase), current_join_id); return; } /* Update the element with the node's name and UUID, in case they * weren't known before */ crm_trace("Updating node name and UUID in CIB for %s", join_to); tmp1 = pcmk__xe_create(NULL, PCMK_XE_NODE); crm_xml_add(tmp1, PCMK_XA_ID, pcmk__cluster_node_uuid(join_node)); crm_xml_add(tmp1, PCMK_XA_UNAME, join_to); fsa_cib_anon_update(PCMK_XE_NODES, tmp1); pcmk__xml_free(tmp1); join_node = pcmk__get_node(0, join_to, NULL, pcmk__node_search_cluster_member); if (!pcmk__cluster_is_node_active(join_node)) { /* * NACK'ing nodes that the membership layer doesn't know about yet * simply creates more churn * * Better to leave them waiting and let the join restart when * the new membership event comes in * * All other NACKs (due to versions etc) should still be processed */ pcmk__update_peer_expected(__func__, join_node, CRMD_JOINSTATE_PENDING); return; } // Acknowledge or nack node's join request crm_debug("%sing join-%d request from %s", integrated? "Acknowledg" : "Nack", current_join_id, join_to); acknak = create_dc_message(CRM_OP_JOIN_ACKNAK, join_to); pcmk__xe_set_bool_attr(acknak, CRM_OP_JOIN_ACKNAK, integrated); if (integrated) { // No change needed for a nacked node crm_update_peer_join(__func__, join_node, controld_join_finalized); pcmk__update_peer_expected(__func__, join_node, CRMD_JOINSTATE_MEMBER); /* Iterate through the remote peer cache and add information on which * node hosts each to the ACK message. This keeps new controllers in * sync with what has already happened. */ if (pcmk__cluster_num_remote_nodes() > 0) { GHashTableIter iter; pcmk__node_status_t *node = NULL; xmlNode *remotes = pcmk__xe_create(acknak, PCMK_XE_NODES); g_hash_table_iter_init(&iter, pcmk__remote_peer_cache); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &node)) { xmlNode *remote = NULL; if (!node->conn_host) { continue; } remote = pcmk__xe_create(remotes, PCMK_XE_NODE); pcmk__xe_set_props(remote, PCMK_XA_ID, node->name, PCMK__XA_NODE_STATE, node->state, PCMK__XA_CONNECTION_HOST, node->conn_host, NULL); } } } pcmk__cluster_send_message(join_node, pcmk_ipc_controld, acknak); pcmk__xml_free(acknak); return; } gboolean check_join_state(enum crmd_fsa_state cur_state, const char *source) { static unsigned long long highest_seq = 0; if (controld_globals.membership_id != controld_globals.peer_seq) { crm_debug("join-%d: Membership changed from %llu to %llu " QB_XS " highest=%llu state=%s for=%s", current_join_id, controld_globals.membership_id, controld_globals.peer_seq, highest_seq, fsa_state2string(cur_state), source); if (highest_seq < controld_globals.peer_seq) { /* Don't spam the FSA with duplicates */ highest_seq = controld_globals.peer_seq; register_fsa_input_before(C_FSA_INTERNAL, I_NODE_JOIN, NULL); } } else if (cur_state == S_INTEGRATION) { if (crmd_join_phase_count(controld_join_welcomed) == 0) { int count = crmd_join_phase_count(controld_join_integrated); crm_debug("join-%d: Integration of %d peer%s complete " QB_XS " state=%s for=%s", current_join_id, count, pcmk__plural_s(count), fsa_state2string(cur_state), source); register_fsa_input_before(C_FSA_INTERNAL, I_INTEGRATED, NULL); return TRUE; } } else if (cur_state == S_FINALIZE_JOIN) { if (!pcmk_is_set(controld_globals.fsa_input_register, R_HAVE_CIB)) { crm_debug("join-%d: Delaying finalization until we have CIB " QB_XS " state=%s for=%s", current_join_id, fsa_state2string(cur_state), source); return TRUE; } else if (crmd_join_phase_count(controld_join_welcomed) != 0) { int count = crmd_join_phase_count(controld_join_welcomed); crm_debug("join-%d: Still waiting on %d welcomed node%s " QB_XS " state=%s for=%s", current_join_id, count, pcmk__plural_s(count), fsa_state2string(cur_state), source); crmd_join_phase_log(LOG_DEBUG); } else if (crmd_join_phase_count(controld_join_integrated) != 0) { int count = crmd_join_phase_count(controld_join_integrated); crm_debug("join-%d: Still waiting on %d integrated node%s " QB_XS " state=%s for=%s", current_join_id, count, pcmk__plural_s(count), fsa_state2string(cur_state), source); crmd_join_phase_log(LOG_DEBUG); } else if (crmd_join_phase_count(controld_join_finalized) != 0) { int count = crmd_join_phase_count(controld_join_finalized); crm_debug("join-%d: Still waiting on %d finalized node%s " QB_XS " state=%s for=%s", current_join_id, count, pcmk__plural_s(count), fsa_state2string(cur_state), source); crmd_join_phase_log(LOG_DEBUG); } else { crm_debug("join-%d: Complete " QB_XS " state=%s for=%s", current_join_id, fsa_state2string(cur_state), source); register_fsa_input_later(C_FSA_INTERNAL, I_FINALIZED, NULL); return TRUE; } } return FALSE; } void do_dc_join_final(long long action, enum crmd_fsa_cause cause, enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data) { crm_debug("Ensuring DC, quorum and node attributes are up-to-date"); crm_update_quorum(pcmk__cluster_has_quorum(), TRUE); } int crmd_join_phase_count(enum controld_join_phase phase) { int count = 0; pcmk__node_status_t *peer; GHashTableIter iter; g_hash_table_iter_init(&iter, pcmk__peer_cache); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &peer)) { if (controld_get_join_phase(peer) == phase) { count++; } } return count; } void crmd_join_phase_log(int level) { pcmk__node_status_t *peer; GHashTableIter iter; g_hash_table_iter_init(&iter, pcmk__peer_cache); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &peer)) { do_crm_log(level, "join-%d: %s=%s", current_join_id, peer->name, join_phase_text(controld_get_join_phase(peer))); } } diff --git a/daemons/execd/remoted_schemas.c b/daemons/execd/remoted_schemas.c index e553551659..15258ff486 100644 --- a/daemons/execd/remoted_schemas.c +++ b/daemons/execd/remoted_schemas.c @@ -1,286 +1,290 @@ /* * Copyright 2023-2024 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 "pacemaker-execd.h" static pid_t schema_fetch_pid = 0; static int rm_files(const char *pathname, const struct stat *sbuf, int type, struct FTW *ftwb) { /* Don't delete PCMK__REMOTE_SCHEMA_DIR . */ if (ftwb->level == 0) { return 0; } if (remove(pathname) != 0) { int rc = errno; crm_err("Could not remove %s: %s", pathname, pcmk_rc_str(rc)); return -1; } return 0; } static void clean_up_extra_schema_files(void) { const char *remote_schema_dir = pcmk__remote_schema_dir(); struct stat sb; int rc; rc = stat(remote_schema_dir, &sb); if (rc == -1) { if (errno == ENOENT) { /* If the directory doesn't exist, try to make it first. */ if (mkdir(remote_schema_dir, 0755) != 0) { rc = errno; crm_err("Could not create directory for schemas: %s", pcmk_rc_str(rc)); } } else { rc = errno; crm_err("Could not create directory for schemas: %s", pcmk_rc_str(rc)); } } else if (!S_ISDIR(sb.st_mode)) { /* If something exists with the same name that's not a directory, that's * an error. */ crm_err("%s already exists but is not a directory", remote_schema_dir); } else { /* It's a directory - clear it out so we can download potentially new * schema files. */ rc = nftw(remote_schema_dir, rm_files, 10, FTW_DEPTH|FTW_MOUNT|FTW_PHYS); if (rc != 0) { crm_err("Could not remove %s: %s", remote_schema_dir, pcmk_rc_str(rc)); } } } static void write_extra_schema_file(xmlNode *xml, void *user_data) { const char *remote_schema_dir = pcmk__remote_schema_dir(); const char *file = NULL; char *path = NULL; int rc; file = crm_element_value(xml, PCMK_XA_PATH); if (file == NULL) { crm_warn("No destination path given in schema request"); return; } path = crm_strdup_printf("%s/%s", remote_schema_dir, file); /* The schema is a CDATA node, which is a child of the node. Traverse * all children and look for the first CDATA child. There can't be more than * one because we only have one file attribute on the parent. */ for (xmlNode *child = xml->children; child != NULL; child = child->next) { FILE *stream = NULL; if (child->type != XML_CDATA_SECTION_NODE) { continue; } stream = fopen(path, "w+"); if (stream == NULL) { crm_warn("Could not write schema file %s: %s", path, strerror(errno)); } else { rc = fprintf(stream, "%s", child->content); if (rc < 0) { crm_warn("Could not write schema file %s: %s", path, strerror(errno)); } fclose(stream); } break; } free(path); } static void get_schema_files(void) { int rc = pcmk_rc_ok; cib_t *cib = NULL; xmlNode *reply; cib = cib_new(); if (cib == NULL) { - _exit(ENOTCONN); + pcmk_common_cleanup(); + _exit(CRM_EX_OSERR); } rc = cib->cmds->signon(cib, crm_system_name, cib_query); - if (rc != pcmk_ok) { - crm_err("Could not connect to the CIB manager: %s", pcmk_strerror(rc)); + rc = pcmk_legacy2rc(rc); + if (rc != pcmk_rc_ok) { + crm_err("Could not connect to the CIB manager: %s", pcmk_rc_str(rc)); + pcmk_common_cleanup(); _exit(pcmk_rc2exitc(rc)); } rc = cib->cmds->fetch_schemas(cib, &reply, pcmk__highest_schema_name(), cib_sync_call); if (rc != pcmk_ok) { crm_err("Could not get schema files: %s", pcmk_strerror(rc)); rc = pcmk_legacy2rc(rc); } else if (reply->children != NULL) { /* The returned document looks something like this: * * * * * * * * * ... * * * * ... * * * * * All the and tags are really just there for organizing * the XML a little better. What we really care about are the nodes, * and specifically the path attributes and the CDATA children (not shown) * of each. We can use an xpath query to reach down and get all the * nodes at once. * * If we already have the latest schema version, or we asked for one later * than what the cluster supports, we'll get back an empty node, * so all this will continue to work. It just won't do anything. */ crm_foreach_xpath_result(reply, "//" PCMK_XA_FILE, write_extra_schema_file, NULL); } pcmk__xml_free(reply); cib__clean_up_connection(&cib); + pcmk_common_cleanup(); _exit(pcmk_rc2exitc(rc)); } /* Load any additional schema files when the child is finished fetching and * saving them to disk. */ static void get_schema_files_complete(mainloop_child_t *p, pid_t pid, int core, int signo, int exitcode) { const char *errmsg = "Could not load additional schema files"; if ((signo == 0) && (exitcode == 0)) { const char *remote_schema_dir = pcmk__remote_schema_dir(); /* Don't just call pcmk__schema_init() here because that will load the * base schemas again too. Instead just load the things we fetched. */ pcmk__load_schemas_from_dir(remote_schema_dir); pcmk__sort_schemas(); crm_info("Fetching extra schema files completed successfully"); } else { if (signo == 0) { crm_err("%s: process %d exited %d", errmsg, (int) pid, exitcode); } else { crm_err("%s: process %d terminated with signal %d (%s)%s", errmsg, (int) pid, signo, strsignal(signo), (core? " and dumped core" : "")); } /* Clean up any incomplete schema data we might have been downloading * when the process timed out or crashed. We don't need to do any extra * cleanup because we never loaded the extra schemas, and we don't need * to call pcmk__schema_init() because that was called in * remoted_request_cib_schema_files() before this function. */ clean_up_extra_schema_files(); } } void remoted_request_cib_schema_files(void) { pid_t pid; int rc; /* If a previous schema-fetch process is still running when we're called * again, it's hung. Attempt to kill it before cleaning up the extra * directory. */ if (schema_fetch_pid != 0) { if (mainloop_child_kill(schema_fetch_pid) == FALSE) { crm_warn("Unable to kill pre-existing schema-fetch process"); return; } schema_fetch_pid = 0; } /* Clean up any extra schema files we downloaded from a previous cluster * connection. After the files are gone, we need to wipe them from * known_schemas, but there's no opposite operation for add_schema(). * * Instead, unload all the schemas. This means we'll also forget about all * installed schemas as well, which means that pcmk__highest_schema_name() * would fail. So we need to load the base schemas right now. */ clean_up_extra_schema_files(); pcmk__schema_cleanup(); pcmk__schema_init(); crm_info("Fetching extra schema files from cluster"); pid = fork(); switch (pid) { case -1: { rc = errno; crm_warn("Could not spawn process to get schema files: %s", pcmk_rc_str(rc)); break; } case 0: /* child */ get_schema_files(); break; default: /* parent */ schema_fetch_pid = pid; mainloop_child_add_with_flags(pid, 5 * 60 * 1000, "schema-fetch", NULL, mainloop_leave_pid_group, get_schema_files_complete); break; } } diff --git a/include/crm/common/actions_internal.h b/include/crm/common/actions_internal.h index f05e9d59d5..3d025cd5b4 100644 --- a/include/crm/common/actions_internal.h +++ b/include/crm/common/actions_internal.h @@ -1,269 +1,269 @@ /* * Copyright 2004-2024 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_ACTIONS_INTERNAL__H #define PCMK__CRM_COMMON_ACTIONS_INTERNAL__H #include // bool #include // uint32_t, UINT32_C() #include // guint, GList, GHashTable #include // xmlNode #include // PCMK_ACTION_MONITOR #include // enum rsc_role_e #include // pcmk_resource_t, pcmk_node_t #include // pcmk__str_eq() #ifdef __cplusplus extern "C" { #endif //! printf-style format to create operation key from resource, action, interval #define PCMK__OP_FMT "%s_%s_%u" /*! * \internal * \brief Set action flags for an action * * \param[in,out] action Action to set flags for * \param[in] flags_to_set Group of enum pcmk__action_flags to set */ #define pcmk__set_action_flags(action, flags_to_set) do { \ (action)->flags = pcmk__set_flags_as(__func__, __LINE__, \ LOG_TRACE, \ "Action", (action)->uuid, \ (action)->flags, \ (flags_to_set), \ #flags_to_set); \ } while (0) /*! * \internal * \brief Clear action flags for an action * * \param[in,out] action Action to clear flags for * \param[in] flags_to_clear Group of enum pcmk__action_flags to clear */ #define pcmk__clear_action_flags(action, flags_to_clear) do { \ (action)->flags = pcmk__clear_flags_as(__func__, __LINE__, \ LOG_TRACE, \ "Action", (action)->uuid, \ (action)->flags, \ (flags_to_clear), \ #flags_to_clear); \ } while (0) /*! * \internal * \brief Set action flags for a flag group * * \param[in,out] action_flags Flag group to set flags for * \param[in] action_name Name of action being modified (for logging) * \param[in] to_set Group of enum pcmk__action_flags to set */ #define pcmk__set_raw_action_flags(action_flags, action_name, to_set) do { \ action_flags = pcmk__set_flags_as(__func__, __LINE__, \ LOG_TRACE, "Action", action_name, \ (action_flags), \ (to_set), #to_set); \ } while (0) /*! * \internal * \brief Clear action flags for a flag group * * \param[in,out] action_flags Flag group to clear flags for * \param[in] action_name Name of action being modified (for logging) * \param[in] to_clear Group of enum pcmk__action_flags to clear */ #define pcmk__clear_raw_action_flags(action_flags, action_name, to_clear) \ do { \ action_flags = pcmk__clear_flags_as(__func__, __LINE__, LOG_TRACE, \ "Action", action_name, \ (action_flags), \ (to_clear), #to_clear); \ } while (0) // Possible actions (including some pseudo-actions) enum pcmk__action_type { pcmk__action_unspecified = 0, // Unspecified or unknown action pcmk__action_monitor, // Monitor // Each "completed" action must be the regular action plus 1 pcmk__action_stop, // Stop pcmk__action_stopped, // Stop completed pcmk__action_start, // Start pcmk__action_started, // Start completed pcmk__action_notify, // Notify pcmk__action_notified, // Notify completed pcmk__action_promote, // Promote pcmk__action_promoted, // Promoted pcmk__action_demote, // Demote pcmk__action_demoted, // Demoted pcmk__action_shutdown, // Shut down node pcmk__action_fence, // Fence node }; // Action scheduling flags enum pcmk__action_flags { // No action flags set (compare with equality rather than bit set) pcmk__no_action_flags = 0, // Whether action does not require invoking an agent pcmk__action_pseudo = (UINT32_C(1) << 0), // Whether action is runnable pcmk__action_runnable = (UINT32_C(1) << 1), // Whether action should not be executed pcmk__action_optional = (UINT32_C(1) << 2), // Whether action should be added to transition graph even if optional pcmk__action_always_in_graph = (UINT32_C(1) << 3), // Whether operation-specific instance attributes have been unpacked yet pcmk__action_attrs_evaluated = (UINT32_C(1) << 4), // Whether action is allowed to be part of a live migration pcmk__action_migratable = (UINT32_C(1) << 7), // Whether action has been added to transition graph pcmk__action_added_to_graph = (UINT32_C(1) << 8), // Whether action is a stop to abort a dangling migration pcmk__action_migration_abort = (UINT32_C(1) << 11), // Whether action is recurring monitor that must be rescheduled if active pcmk__action_reschedule = (UINT32_C(1) << 13), // Whether action has already been processed by a recursive procedure pcmk__action_detect_loop = (UINT32_C(1) << 14), // Whether action's inputs have been de-duplicated yet pcmk__action_inputs_deduplicated = (UINT32_C(1) << 15), // Whether action can be executed on DC rather than own node pcmk__action_on_dc = (UINT32_C(1) << 16), }; /* Possible responses to a resource action failure * * The order is significant; the values are in order of increasing severity so * that they can be compared with less than and greater than. */ enum pcmk__on_fail { pcmk__on_fail_ignore, // Act as if failure didn't happen pcmk__on_fail_demote, // Demote if promotable, else stop pcmk__on_fail_restart, // Restart resource /* Fence the remote node created by the resource if fencing is enabled, * otherwise attempt to restart the resource (used internally for some * remote connection failures). */ pcmk__on_fail_reset_remote, pcmk__on_fail_restart_container, // Restart resource's container pcmk__on_fail_ban, // Ban resource from current node pcmk__on_fail_block, // Treat resource as unmanaged pcmk__on_fail_stop, // Stop resource and leave stopped pcmk__on_fail_standby_node, // Put resource's node in standby pcmk__on_fail_fence_node, // Fence resource's node }; // What resource needs before it can be recovered from a failed node enum pcmk__requires { pcmk__requires_nothing = 0, // Resource can be recovered immediately pcmk__requires_quorum = 1, // Resource can be recovered if quorate pcmk__requires_fencing = 2, // Resource can be recovered after fencing }; // Implementation of pcmk_action_t struct pcmk__action { int id; // Counter to identify action /* * When the controller aborts a transition graph, it sets an abort priority. * If this priority is higher, the action will still be executed anyway. * Pseudo-actions are always allowed, so this is irrelevant for them. */ int priority; pcmk_resource_t *rsc; // Resource to apply action to, if any - pcmk_node_t *node; // Node to execute action on, if any + pcmk_node_t *node; // Copy of node to execute action on, if any xmlNode *op_entry; // Action XML configuration, if any char *task; // Action name char *uuid; // Action key char *cancel_task; // If task is "cancel", the action being cancelled char *reason; // Readable description of why action is needed uint32_t flags; // Group of enum pcmk__action_flags enum pcmk__requires needs; // Prerequisite for recovery enum pcmk__on_fail on_fail; // Response to failure enum rsc_role_e fail_role; // Resource role if action fails GHashTable *meta; // Meta-attributes relevant to action GHashTable *extra; // Action-specific instance attributes pcmk_scheduler_t *scheduler; // Scheduler data this action is part of /* Current count of runnable instance actions for "first" action in an * ordering dependency with pcmk__ar_min_runnable set. */ int runnable_before; /* * Number of instance actions for "first" action in an ordering dependency * with pcmk__ar_min_runnable set that must be runnable before this action * can be runnable. */ int required_runnable_before; // Actions in a relation with this one (as pcmk__related_action_t *) GList *actions_before; GList *actions_after; }; char *pcmk__op_key(const char *rsc_id, const char *op_type, guint interval_ms); char *pcmk__notify_key(const char *rsc_id, const char *notify_type, const char *op_type); char *pcmk__transition_key(int transition_id, int action_id, int target_rc, const char *node); void pcmk__filter_op_for_digest(xmlNode *param_set); bool pcmk__is_fencing_action(const char *action); enum pcmk__action_type pcmk__parse_action(const char *action_name); const char *pcmk__action_text(enum pcmk__action_type action); const char *pcmk__on_fail_text(enum pcmk__on_fail on_fail); /*! * \internal * \brief Get a human-friendly action name * * \param[in] action_name Actual action name * \param[in] interval_ms Action interval (in milliseconds) * * \return Action name suitable for display */ static inline const char * pcmk__readable_action(const char *action_name, guint interval_ms) { if ((interval_ms == 0) && pcmk__str_eq(action_name, PCMK_ACTION_MONITOR, pcmk__str_none)) { return "probe"; } return action_name; } #ifdef __cplusplus } #endif #endif // PCMK__CRM_COMMON_ACTIONS_INTERNAL__H diff --git a/include/crm/common/bundles_internal.h b/include/crm/common/bundles_internal.h index a0abdeed58..27fe053c0e 100644 --- a/include/crm/common/bundles_internal.h +++ b/include/crm/common/bundles_internal.h @@ -1,91 +1,91 @@ /* * Copyright 2017-2024 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_BUNDLES_INTERNAL__H #define PCMK__CRM_COMMON_BUNDLES_INTERNAL__H #include // NULL #include // bool, false #include // struct pcmk__node_private #include // pcmk__is_guest_or_bundle_node() #include // pcmk__rsc_variant_bundle etc. #include // pcmk_resource_t, pcmk_node_t #ifdef __cplusplus extern "C" { #endif //! A single instance of a bundle typedef struct { int offset; //!< 0-origin index of this instance in bundle char *ipaddr; //!< IP address associated with this instance - pcmk_node_t *node; //!< Node created for this instance + pcmk_node_t *node; //!< Copy of node created for this instance pcmk_resource_t *ip; //!< IP address resource for ipaddr pcmk_resource_t *child; //!< Instance of bundled resource pcmk_resource_t *container; //!< Container associated with this instance pcmk_resource_t *remote; //!< Pacemaker Remote connection into container } pcmk__bundle_replica_t; /*! * \internal * \brief Check whether a resource is a bundle resource * * \param[in] rsc Resource to check * * \return true if \p rsc is a bundle, otherwise false * \note This does not return true if \p rsc is part of a bundle * (see pcmk__is_bundled()). */ static inline bool pcmk__is_bundle(const pcmk_resource_t *rsc) { return (rsc != NULL) && (rsc->priv->variant == pcmk__rsc_variant_bundle); } /*! * \internal * \brief Check whether a resource is part of a bundle * * \param[in] rsc Resource to check * * \return true if \p rsc is part of a bundle, otherwise false */ static inline bool pcmk__is_bundled(const pcmk_resource_t *rsc) { if (rsc == NULL) { return false; } while (rsc->priv->parent != NULL) { rsc = rsc->priv->parent; } return rsc->priv->variant == pcmk__rsc_variant_bundle; } /*! * \internal * \brief Check whether a node is a bundle node * * \param[in] node Node to check * * \return true if \p node is a bundle node, otherwise false */ static inline bool pcmk__is_bundle_node(const pcmk_node_t *node) { return pcmk__is_guest_or_bundle_node(node) && pcmk__is_bundled(node->priv->remote); } #ifdef __cplusplus } #endif #endif // PCMK__CRM_COMMON_BUNDLES_INTERNAL__H diff --git a/include/crm/common/location_internal.h b/include/crm/common/location_internal.h index 5991feea98..39aba91c4b 100644 --- a/include/crm/common/location_internal.h +++ b/include/crm/common/location_internal.h @@ -1,36 +1,36 @@ /* * Copyright 2004-2024 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_LOCATION_INTERNAL__H #define PCMK__CRM_COMMON_LOCATION_INTERNAL__H #include // GList #include // enum pcmk__probe_mode #include // enum rsc_role_e #include // pcmk_resource_t #ifdef __cplusplus extern "C" { #endif //! Location constraint object typedef struct { char *id; // XML ID of location constraint pcmk_resource_t *rsc; // Resource with location preference enum rsc_role_e role_filter; // Limit to instances with this role enum pcmk__probe_mode probe_mode; // How to probe resource on node - GList *nodes; // Affected nodes, with preference score + GList *nodes; // Copies of affected nodes, with score } pcmk__location_t; #ifdef __cplusplus } #endif #endif // PCMK__CRM_COMMON_LOCATION_INTERNAL__H diff --git a/include/crm/common/nodes_internal.h b/include/crm/common/nodes_internal.h index 176b0dbaae..4e0dd86691 100644 --- a/include/crm/common/nodes_internal.h +++ b/include/crm/common/nodes_internal.h @@ -1,201 +1,202 @@ /* * Copyright 2024 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_NODES_INTERNAL__H #define PCMK__CRM_COMMON_NODES_INTERNAL__H #include // NULL #include // bool #include // uint32_t, UINT32_C() #include #include #ifdef __cplusplus extern "C" { #endif /* * Special node attributes */ #define PCMK__NODE_ATTR_SHUTDOWN "shutdown" /* @COMPAT Deprecated since 2.1.8. Use a location constraint with * PCMK_XA_RSC_PATTERN=".*" and PCMK_XA_RESOURCE_DISCOVERY="never" instead of * PCMK__NODE_ATTR_RESOURCE_DISCOVERY_ENABLED="false". */ #define PCMK__NODE_ATTR_RESOURCE_DISCOVERY_ENABLED "resource-discovery-enabled" enum pcmk__node_variant { // Possible node types pcmk__node_variant_cluster = 1, // Cluster layer node pcmk__node_variant_remote = 2, // Pacemaker Remote node }; enum pcmk__node_flags { pcmk__node_none = UINT32_C(0), // Whether node is in standby mode pcmk__node_standby = (UINT32_C(1) << 0), // Whether node is in standby mode due to PCMK_META_ON_FAIL pcmk__node_fail_standby = (UINT32_C(1) << 1), // Whether node has ever joined cluster (and thus has node state in CIB) pcmk__node_seen = (UINT32_C(1) << 2), // Whether expected join state is member pcmk__node_expected_up = (UINT32_C(1) << 3), // Whether probes are allowed on node pcmk__node_probes_allowed = (UINT32_C(1) << 4), /* Whether this either is a guest node whose guest resource must be * recovered or a remote node that must be fenced */ pcmk__node_remote_reset = (UINT32_C(1) << 5), /* Whether this is a Pacemaker Remote node that was fenced since it was last * connected by the cluster */ pcmk__node_remote_fenced = (UINT32_C(1) << 6), /* * Whether this is a Pacemaker Remote node previously marked in its * node state as being in maintenance mode */ pcmk__node_remote_maint = (UINT32_C(1) << 7), // Whether node history has been unpacked pcmk__node_unpacked = (UINT32_C(1) << 8), }; // When to probe a resource on a node (as specified in location constraints) enum pcmk__probe_mode { pcmk__probe_always = 0, // Always probe resource on node pcmk__probe_never = 1, // Never probe resource on node pcmk__probe_exclusive = 2, // Probe only on designated nodes }; /* Per-node data used in resource assignment * * @COMPAT When we can make the pcmk_node_t implementation internal, move these * there and drop this struct. */ struct pcmk__node_assignment { int score; // Node's score for relevant resource int count; // Counter reused by assignment and promotion code enum pcmk__probe_mode probe_mode; // When to probe resource on this node }; /* Implementation of pcmk__node_private_t (pcmk_node_t objects are shallow * copies, so all pcmk_node_t objects for the same node will share the same * private data) */ struct pcmk__node_private { /* Node's XML ID in the CIB (the cluster layer ID for cluster nodes, * the node name for Pacemaker Remote nodes) */ const char *id; /* * Sum of priorities of all resources active on node and on any guest nodes * connected to this node, with +1 for promoted instances (used to compare * nodes for PCMK_OPT_PRIORITY_FENCING_DELAY) */ int priority; const char *name; // Node name in cluster enum pcmk__node_variant variant; // Node variant uint32_t flags; // Group of enum pcmk__node_flags GHashTable *attrs; // Node attributes GHashTable *utilization; // Node utilization attributes int num_resources; // Number of active resources on node GList *assigned_resources; // List of resources assigned to node GHashTable *digest_cache; // Cache of calculated resource digests pcmk_resource_t *remote; // Pacemaker Remote connection (if any) pcmk_scheduler_t *scheduler; // Scheduler data that node is part of }; +void pcmk__free_node_copy(void *data); pcmk_node_t *pcmk__find_node_in_list(const GList *nodes, const char *node_name); /*! * \internal * \brief Set node flags * * \param[in,out] node Node to set flags for * \param[in] flags_to_set Group of enum pcmk_node_flags to set */ #define pcmk__set_node_flags(node, flags_to_set) do { \ (node)->priv->flags = pcmk__set_flags_as(__func__, __LINE__, \ LOG_TRACE, "Node", pcmk__node_name(node), \ (node)->priv->flags, (flags_to_set), #flags_to_set); \ } while (0) /*! * \internal * \brief Clear node flags * * \param[in,out] node Node to clear flags for * \param[in] flags_to_clear Group of enum pcmk_node_flags to clear */ #define pcmk__clear_node_flags(node, flags_to_clear) do { \ (node)->priv->flags = pcmk__clear_flags_as(__func__, __LINE__, \ LOG_TRACE, "Node", pcmk__node_name(node), \ (node)->priv->flags, (flags_to_clear), #flags_to_clear); \ } while (0) /*! * \internal * \brief Return a string suitable for logging as a node name * * \param[in] node Node to return a node name string for * * \return Node name if available, otherwise node ID if available, * otherwise "unspecified node" if node is NULL or "unidentified node" * if node has neither a name nor ID. */ static inline const char * pcmk__node_name(const pcmk_node_t *node) { if (node == NULL) { return "unspecified node"; } else if (node->priv->name != NULL) { return node->priv->name; } else if (node->priv->id != NULL) { return node->priv->id; } else { return "unidentified node"; } } /*! * \internal * \brief Check whether two node objects refer to the same node * * \param[in] node1 First node object to compare * \param[in] node2 Second node object to compare * * \return true if \p node1 and \p node2 refer to the same node */ static inline bool pcmk__same_node(const pcmk_node_t *node1, const pcmk_node_t *node2) { return (node1 != NULL) && (node2 != NULL) && (node1->priv == node2->priv); } #ifdef __cplusplus } #endif #endif // PCMK__CRM_COMMON_NODES_INTERNAL__H diff --git a/include/crm/common/resources_internal.h b/include/crm/common/resources_internal.h index b499a808a7..87ac4c609c 100644 --- a/include/crm/common/resources_internal.h +++ b/include/crm/common/resources_internal.h @@ -1,462 +1,468 @@ /* * Copyright 2024 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_RESOURCES_INTERNAL__H #define PCMK__CRM_COMMON_RESOURCES_INTERNAL__H #include // uint32_t #include // gboolean, guint, GHashTable, GList #include // xmlNode #include // pcmk_resource_t #include // enum rsc_role_e #include // pcmk_node_t, etc. #ifdef __cplusplus extern "C" { #endif /*! * \internal * \brief Set resource flags * * \param[in,out] resource Resource to set flags for * \param[in] flags_to_set Group of enum pcmk_rsc_flags to set */ #define pcmk__set_rsc_flags(resource, flags_to_set) do { \ (resource)->flags = pcmk__set_flags_as(__func__, __LINE__, \ LOG_TRACE, "Resource", (resource)->id, (resource)->flags, \ (flags_to_set), #flags_to_set); \ } while (0) /*! * \internal * \brief Clear resource flags * * \param[in,out] resource Resource to clear flags for * \param[in] flags_to_clear Group of enum pcmk_rsc_flags to clear */ #define pcmk__clear_rsc_flags(resource, flags_to_clear) do { \ (resource)->flags = pcmk__clear_flags_as(__func__, __LINE__, \ LOG_TRACE, "Resource", (resource)->id, (resource)->flags, \ (flags_to_clear), #flags_to_clear); \ } while (0) //! Resource variants supported by Pacemaker enum pcmk__rsc_variant { // Order matters: some code compares greater or lesser than pcmk__rsc_variant_unknown = -1, //!< Unknown resource variant pcmk__rsc_variant_primitive = 0, //!< Primitive resource pcmk__rsc_variant_group = 1, //!< Group resource pcmk__rsc_variant_clone = 2, //!< Clone resource pcmk__rsc_variant_bundle = 3, //!< Bundle resource }; //! How to recover a resource that is incorrectly active on multiple nodes enum pcmk__multiply_active { pcmk__multiply_active_restart, //!< Stop on all, start on desired pcmk__multiply_active_stop, //!< Stop on all and leave stopped pcmk__multiply_active_block, //!< Do nothing to resource pcmk__multiply_active_unexpected, //!< Stop unexpected instances }; //! Resource scheduling flags enum pcmk__rsc_flags { // No resource flags set (compare with equality rather than bit set) pcmk__no_rsc_flags = 0ULL, // Whether resource has been removed from the configuration pcmk__rsc_removed = (1ULL << 0), /* NOTE: sbd (at least as of 1.5.2) uses pe_rsc_managed which equates to * this value, so the value should not be changed */ // Whether resource is managed pcmk__rsc_managed = (1ULL << 1), // Whether resource is blocked from further action pcmk__rsc_blocked = (1ULL << 2), // Whether resource has been removed but was launched pcmk__rsc_removed_launched = (1ULL << 3), // Whether resource has clone notifications enabled pcmk__rsc_notify = (1ULL << 4), // Whether resource is not an anonymous clone instance pcmk__rsc_unique = (1ULL << 5), // Whether resource's class is "stonith" pcmk__rsc_fence_device = (1ULL << 6), // Whether resource can be promoted and demoted pcmk__rsc_promotable = (1ULL << 7), // Whether resource has not yet been assigned to a node pcmk__rsc_unassigned = (1ULL << 8), // Whether resource is in the process of being assigned to a node pcmk__rsc_assigning = (1ULL << 9), // Whether resource is in the process of modifying allowed node scores pcmk__rsc_updating_nodes = (1ULL << 10), // Whether resource is in the process of scheduling actions to restart pcmk__rsc_restarting = (1ULL << 11), // Whether resource must be stopped (instead of demoted) if it is failed pcmk__rsc_stop_if_failed = (1ULL << 12), // Whether a reload action has been scheduled for resource pcmk__rsc_reload = (1ULL << 13), // Whether resource is a remote connection allowed to run on a remote node pcmk__rsc_remote_nesting_allowed = (1ULL << 14), // Whether resource has \c PCMK_META_CRITICAL meta-attribute enabled pcmk__rsc_critical = (1ULL << 15), // Whether resource is considered failed pcmk__rsc_failed = (1ULL << 16), // Flag for non-scheduler code to use to detect recursion loops pcmk__rsc_detect_loop = (1ULL << 17), // Whether resource is a Pacemaker Remote connection pcmk__rsc_is_remote_connection = (1ULL << 18), // Whether resource has pending start action in history pcmk__rsc_start_pending = (1ULL << 19), // Whether resource is probed only on nodes marked exclusive pcmk__rsc_exclusive_probes = (1ULL << 20), /* * Whether resource is multiply active with recovery set to * \c PCMK_VALUE_STOP_UNEXPECTED */ pcmk__rsc_stop_unexpected = (1ULL << 22), // Whether resource is allowed to live-migrate pcmk__rsc_migratable = (1ULL << 23), // Whether resource has an ignorable failure pcmk__rsc_ignore_failure = (1ULL << 24), // Whether resource is an implicit container resource for a bundle replica pcmk__rsc_replica_container = (1ULL << 25), // Whether resource, its node, or entire cluster is in maintenance mode pcmk__rsc_maintenance = (1ULL << 26), // Whether resource can be started or promoted only on quorate nodes pcmk__rsc_needs_quorum = (1ULL << 28), // Whether resource requires fencing before recovery if on unclean node pcmk__rsc_needs_fencing = (1ULL << 29), // Whether resource can be started or promoted only on unfenced nodes pcmk__rsc_needs_unfencing = (1ULL << 30), }; // Where to look for a resource enum pcmk__rsc_node { pcmk__rsc_node_none = 0U, // Nowhere pcmk__rsc_node_assigned = (1U << 0), // Where resource is assigned pcmk__rsc_node_current = (1U << 1), // Where resource is running pcmk__rsc_node_pending = (1U << 2), // Where resource is pending }; //! Resource assignment methods (implementation defined by libpacemaker) typedef struct pcmk__assignment_methods pcmk__assignment_methods_t; //! Resource object methods typedef struct { /*! * \internal * \brief Parse variant-specific resource XML from CIB into struct members * * \param[in,out] rsc Partially unpacked resource * \param[in,out] scheduler Scheduler data * * \return TRUE if resource was unpacked successfully, otherwise FALSE */ gboolean (*unpack)(pcmk_resource_t *rsc, pcmk_scheduler_t *scheduler); /*! * \internal * \brief Search for a resource ID in a resource and its children * * \param[in] rsc Search this resource and its children * \param[in] id Search for this resource ID * \param[in] on_node If not NULL, limit search to resources on this node * \param[in] flags Group of enum pe_find flags * * \return Resource that matches search criteria if any, otherwise NULL */ pcmk_resource_t *(*find_rsc)(pcmk_resource_t *rsc, const char *search, const pcmk_node_t *node, int flags); /*! * \internal * \brief Get value of a resource instance attribute * * \param[in,out] rsc Resource to check * \param[in] node Node to use to evaluate rules * \param[in] create Ignored * \param[in] name Name of instance attribute to check * \param[in,out] scheduler Scheduler data * * \return Value of requested attribute if available, otherwise NULL * \note The caller is responsible for freeing the result using free(). */ char *(*parameter)(pcmk_resource_t *rsc, pcmk_node_t *node, gboolean create, const char *name, pcmk_scheduler_t *scheduler); /*! * \internal * \brief Check whether a resource is active * * \param[in] rsc Resource to check * \param[in] all If \p rsc is collective, all instances must be active * * \return TRUE if \p rsc is active, otherwise FALSE */ gboolean (*active)(pcmk_resource_t *rsc, gboolean all); /*! * \internal * \brief Get resource's current or assigned role * * \param[in] rsc Resource to check * \param[in] current If TRUE, check current role, otherwise assigned role * * \return Current or assigned role of \p rsc */ enum rsc_role_e (*state)(const pcmk_resource_t *rsc, gboolean current); /*! * \internal * \brief List nodes where a resource (or any of its children) is * * \param[in] rsc Resource to check * \param[out] list List to add result to * \param[in] target Which resource conditions to target (group of * enum pcmk__rsc_node flags) * * \return If list contains only one node, that node, otherwise NULL */ pcmk_node_t *(*location)(const pcmk_resource_t *rsc, GList **list, uint32_t target); /*! * \internal * \brief Free all memory used by a resource * * \param[in,out] rsc Resource to free */ void (*free)(pcmk_resource_t *rsc); /*! * \internal * \brief Increment cluster's instance counts for a resource * * Given a resource, increment its cluster's ninstances, disabled_resources, * and blocked_resources counts for the resource and its descendants. * * \param[in,out] rsc Resource to count */ void (*count)(pcmk_resource_t *rsc); /*! * \internal * \brief Check whether a given resource is in a list of resources * * \param[in] rsc Resource ID to check for * \param[in] only_rsc List of resource IDs to check * \param[in] check_parent If TRUE, check top ancestor as well * * \return TRUE if \p rsc, its top parent if requested, or '*' is in * \p only_rsc, otherwise FALSE */ gboolean (*is_filtered)(const pcmk_resource_t *rsc, GList *only_rsc, gboolean check_parent); /*! * \internal * \brief Find a node (and optionally count all) where resource is active * * \param[in] rsc Resource to check * \param[out] count_all If not NULL, set this to count of active nodes * \param[out] count_clean If not NULL, set this to count of clean nodes * * \return A node where the resource is active, preferring the source node * if the resource is involved in a partial migration, or a clean, * online node if the resource's \c PCMK_META_REQUIRES is * \c PCMK_VALUE_QUORUM or \c PCMK_VALUE_NOTHING, otherwise \c NULL. */ pcmk_node_t *(*active_node)(const pcmk_resource_t *rsc, unsigned int *count_all, unsigned int *count_clean); /*! * \internal * \brief Get maximum resource instances per node * * \param[in] rsc Resource to check * * \return Maximum number of \p rsc instances that can be active on one node */ unsigned int (*max_per_node)(const pcmk_resource_t *rsc); } pcmk__rsc_methods_t; // Implementation of pcmk__resource_private_t struct pcmk__resource_private { enum pcmk__rsc_variant variant; // Resource variant void *variant_opaque; // Variant-specific data char *history_id; // Resource instance ID in history GHashTable *meta; // Resource meta-attributes GHashTable *utilization; // Resource utilization attributes int priority; // Priority relative other resources int promotion_priority; // Promotion priority on assigned node enum rsc_role_e orig_role; // Resource's role at start of transition enum rsc_role_e next_role; // Resource's role at end of transition int stickiness; // Extra preference for current node guint failure_expiration_ms; // Failures expire after this much time int ban_after_failures; // Ban from node after this many failures guint remote_reconnect_ms; // Retry interval for remote connections char *pending_action; // Pending action in history, if any const pcmk_node_t *pending_node;// Node on which pending_action is happening time_t lock_time; // When shutdown lock started const pcmk_node_t *lock_node; // Node that resource is shutdown-locked to GList *actions; // Actions scheduled for resource GList *children; // Resource's child resources, if any pcmk_resource_t *parent; // Resource's parent resource, if any pcmk_scheduler_t *scheduler; // Scheduler data containing resource // Resource configuration (possibly expanded from template) xmlNode *xml; // Original resource configuration, if using template xmlNode *orig_xml; // Configuration of resource operations (possibly expanded from template) xmlNode *ops_xml; /* * Resource parameters may have node-attribute-based rules, which means the * values can vary by node. This table has node names as keys and parameter * name/value tables as values. Use pe_rsc_params() to get the table for a * given node rather than use this directly. */ GHashTable *parameter_cache; /* A "launcher" is defined in one of these ways: * * - A Pacemaker Remote connection for a guest node or bundle node has its * launcher set to the resource that starts the guest or the bundle * replica's container. * * - If the user configures the PCMK__META_CONTAINER meta-attribute for this * resource, the launcher is set to that. * * If the launcher is a Pacemaker Remote connection resource, this * resource may run only on the node created by that connection. * * Otherwise, this resource will be colocated with and ordered after the * launcher, and failures of this resource will cause the launcher to be * recovered instead of this one. This is appropriate for monitoring-only * resources that represent a service launched by the other resource. */ pcmk_resource_t *launcher; // Resources launched by this one, if any (pcmk_resource_t *) GList *launched; // What to do if the resource is incorrectly active on multiple nodes enum pcmk__multiply_active multiply_active_policy; /* The assigned node (if not NULL) is the one where the resource *should* * be active by the end of the current scheduler transition. Only primitive - * resources have an assigned node. + * resources have an assigned node. This is a node copy (created by + * pe__copy_node()) and so must be freed using pcmk__free_node_copy(). * * @TODO This should probably be part of the primitive variant data. */ pcmk_node_t *assigned_node; /* The active nodes are ones where the resource is (or might be, if * insufficient information is available to be sure) already active at the * start of the current scheduler transition. * * For primitive resources, there should be at most one, but could be more * if it is (incorrectly) multiply active. For collective resources, this * combines active nodes of all descendants. */ GList *active_nodes; + /* The next two tables store node copies (created by pe__copy_node()), which + * share some members with the original node objects and must be freed with + * pcmk__free_node_copy(). + */ + // Nodes where resource has been probed (key is node ID, not name) GHashTable *probed_nodes; // Nodes where resource is allowed to run (key is node ID, not name) GHashTable *allowed_nodes; // The source node, if migrate_to completed but migrate_from has not pcmk_node_t *partial_migration_source; // The destination node, if migrate_to completed but migrate_from has not pcmk_node_t *partial_migration_target; // Source nodes where stop is needed after migrate_from and migrate_to GList *dangling_migration_sources; /* Pay special attention to whether you want to use with_this_colocations * and this_with_colocations directly, which include only colocations * explicitly involving this resource, or call libpacemaker's * pcmk__with_this_colocations() and pcmk__this_with_colocations() * functions, which may return relevant colocations involving the resource's * ancestors as well. */ // Colocations of other resources with this one GList *with_this_colocations; // Colocations of this resource with others GList *this_with_colocations; GList *location_constraints; // Location constraints for resource GList *ticket_constraints; // Ticket constraints for resource const pcmk__rsc_methods_t *fns; // Resource object methods const pcmk__assignment_methods_t *cmds; // Resource assignment methods }; const char *pcmk__multiply_active_text(const pcmk_resource_t *rsc); /*! * \internal * \brief Get node where resource is currently active (if any) * * \param[in] rsc Resource to check * * \return Node that \p rsc is active on, if any, otherwise NULL */ static inline pcmk_node_t * pcmk__current_node(const pcmk_resource_t *rsc) { if (rsc == NULL) { return NULL; } return rsc->priv->fns->active_node(rsc, NULL, NULL); } #ifdef __cplusplus } #endif #endif // PCMK__CRM_COMMON_RESOURCES_INTERNAL__H diff --git a/include/crm/common/util.h b/include/crm/common/util.h index 11388b028f..868ab96d0f 100644 --- a/include/crm/common/util.h +++ b/include/crm/common/util.h @@ -1,99 +1,97 @@ /* * Copyright 2004-2024 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_UTIL__H #define PCMK__CRM_COMMON_UTIL__H #include // gid_t, mode_t, size_t, time_t, uid_t #include #include #include // uint32_t #include #include #include #include #include #include #include #include #include #include #ifdef __cplusplus extern "C" { #endif /** * \file * \brief Utility functions * \ingroup core */ -/* public node attribute functions (from attrd_client.c) */ +/* public node attribute functions (from attrs.c) */ char *pcmk_promotion_score_name(const char *rsc_id); /* public Pacemaker Remote functions (from remote.c) */ int crm_default_remote_port(void); int compare_version(const char *version1, const char *version2); /*! * \brief Check whether any of specified flags are set in a flag group * * \param[in] flag_group The flag group being examined * \param[in] flags_to_check Which flags in flag_group should be checked * * \return true if \p flags_to_check is nonzero and any of its flags are set in * \p flag_group, or false otherwise */ static inline bool pcmk_any_flags_set(uint64_t flag_group, uint64_t flags_to_check) { return (flag_group & flags_to_check) != 0; } /*! * \brief Check whether all of specified flags are set in a flag group * * \param[in] flag_group The flag group being examined * \param[in] flags_to_check Which flags in flag_group should be checked * * \return true if \p flags_to_check is zero or all of its flags are set in * \p flag_group, or false otherwise */ static inline bool pcmk_all_flags_set(uint64_t flag_group, uint64_t flags_to_check) { return (flag_group & flags_to_check) == flags_to_check; } /*! * \brief Convenience alias for pcmk_all_flags_set(), to check single flag */ #define pcmk_is_set(g, f) pcmk_all_flags_set((g), (f)) +void pcmk_common_cleanup(void); char *crm_md5sum(const char *buffer); - char *crm_generate_uuid(void); - int crm_user_lookup(const char *name, uid_t * uid, gid_t * gid); int pcmk_daemon_user(uid_t *uid, gid_t *gid); - void crm_gnutls_global_init(void); #ifdef __cplusplus } #endif #if !defined(PCMK_ALLOW_DEPRECATED) || (PCMK_ALLOW_DEPRECATED == 1) #include #endif #endif diff --git a/lib/common/mainloop.c b/lib/common/mainloop.c index a65a5bbf1d..9529d523af 100644 --- a/lib/common/mainloop.c +++ b/lib/common/mainloop.c @@ -1,1463 +1,1465 @@ /* * Copyright 2004-2024 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)); 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)); return SIG_ERR; } return old.sa_handler; } static void mainloop_destroy_signal_entry(int sig) { crm_signal_t *tmp = crm_signals[sig]; - crm_signals[sig] = NULL; - - crm_trace("Destroying signal %d", sig); - mainloop_destroy_trigger((crm_trigger_t *) tmp); + 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); 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) { + 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)); } // All clients should use at least PCMK_ipc_buffer as their buffer size 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); } 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; } } } 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, max_size); 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); } 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/nodes.c b/lib/common/nodes.c index 0ef472bb38..fed3159464 100644 --- a/lib/common/nodes.c +++ b/lib/common/nodes.c @@ -1,192 +1,212 @@ /* * Copyright 2022-2024 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 // xmlNode #include +/*! + * \internal + * \brief Free a copy of a node object + * + * \param[in] data Node copy (created by pe__copy_node()) to free + */ +void +pcmk__free_node_copy(void *data) +{ + if (data != NULL) { + pcmk_node_t *node = data; + + if (node->assign != NULL) { + // This is the only member allocated separately for a node copy + free(node->assign); + } + free(node); + } +} + /*! * \internal * \brief Check whether a node is online * * \param[in] node Node to check * * \return true if \p node is online, otherwise false */ bool pcmk_node_is_online(const pcmk_node_t *node) { return (node != NULL) && node->details->online; } /*! * \internal * \brief Check whether a node is pending * * Check whether a node is pending. A node is pending if it is a member of the * cluster but not the controller group, which means it is in the process of * either joining or leaving the cluster. * * \param[in] node Node to check * * \return true if \p node is pending, otherwise false */ bool pcmk_node_is_pending(const pcmk_node_t *node) { return (node != NULL) && node->details->pending; } /*! * \internal * \brief Check whether a node is clean * * Check whether a node is clean. A node is clean if it is a cluster node or * remote node that has been seen by the cluster at least once, or the * startup-fencing cluster option is false; and the node, and its host if a * guest or bundle node, are not scheduled to be fenced. * * \param[in] node Node to check * * \return true if \p node is clean, otherwise false */ bool pcmk_node_is_clean(const pcmk_node_t *node) { return (node != NULL) && !(node->details->unclean); } /*! * \internal * \brief Check whether a node is shutting down * * \param[in] node Node to check * * \return true if \p node is shutting down, otherwise false */ bool pcmk_node_is_shutting_down(const pcmk_node_t *node) { return (node != NULL) && node->details->shutdown; } /*! * \internal * \brief Check whether a node is in maintenance mode * * \param[in] node Node to check * * \return true if \p node is in maintenance mode, otherwise false */ bool pcmk_node_is_in_maintenance(const pcmk_node_t *node) { return (node != NULL) && node->details->maintenance; } /*! * \internal * \brief Call a function for each resource active on a node * * Call a caller-supplied function with a caller-supplied argument for each * resource that is active on a given node. If the function returns false, this * function will return immediately without processing any remaining resources. * * \param[in] node Node to check * * \return Result of last call of \p fn (or false if none) */ bool pcmk_foreach_active_resource(pcmk_node_t *node, bool (*fn)(pcmk_resource_t *, void *), void *user_data) { bool result = false; if ((node != NULL) && (fn != NULL)) { for (GList *item = node->details->running_rsc; item != NULL; item = item->next) { result = fn((pcmk_resource_t *) item->data, user_data); if (!result) { break; } } } return result; } void pcmk__xe_add_node(xmlNode *xml, const char *node, int nodeid) { pcmk__assert(xml != NULL); if (node != NULL) { crm_xml_add(xml, PCMK__XA_ATTR_HOST, node); } if (nodeid > 0) { crm_xml_add_int(xml, PCMK__XA_ATTR_HOST_ID, nodeid); } } /*! * \internal * \brief Find a node by name in a list of nodes * * \param[in] nodes List of nodes (as pcmk_node_t*) * \param[in] node_name Name of node to find * * \return Node from \p nodes that matches \p node_name if any, otherwise NULL */ pcmk_node_t * pcmk__find_node_in_list(const GList *nodes, const char *node_name) { if (node_name != NULL) { for (const GList *iter = nodes; iter != NULL; iter = iter->next) { pcmk_node_t *node = (pcmk_node_t *) iter->data; if (pcmk__str_eq(node->priv->name, node_name, pcmk__str_casei)) { return node; } } } return NULL; } #define XP_SHUTDOWN "//" PCMK__XE_NODE_STATE "[@" PCMK_XA_UNAME "='%s']/" \ PCMK__XE_TRANSIENT_ATTRIBUTES "/" PCMK_XE_INSTANCE_ATTRIBUTES "/" \ PCMK_XE_NVPAIR "[@" PCMK_XA_NAME "='" PCMK__NODE_ATTR_SHUTDOWN "']" /*! * \brief Get value of a node's shutdown attribute from CIB, if present * * \param[in] cib CIB to check * \param[in] node Name of node to check * * \return Value of shutdown attribute for \p node in \p cib if any, * otherwise NULL * \note The return value is a pointer into \p cib and so is valid only for the * lifetime of that object. */ const char * pcmk_cib_node_shutdown(xmlNode *cib, const char *node) { if ((cib != NULL) && (node != NULL)) { char *xpath = crm_strdup_printf(XP_SHUTDOWN, node); xmlNode *match = get_xpath_object(xpath, cib, LOG_TRACE); free(xpath); if (match != NULL) { return crm_element_value(match, PCMK_XA_VALUE); } } return NULL; } diff --git a/lib/common/results.c b/lib/common/results.c index bad074b6fd..507280492c 100644 --- a/lib/common/results.c +++ b/lib/common/results.c @@ -1,1203 +1,1195 @@ /* * Copyright 2004-2024 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_DEFINE_QUARK(pcmk-rc-error-quark, pcmk__rc_error) G_DEFINE_QUARK(pcmk-exitc-error-quark, pcmk__exitc_error) // General (all result code types) /*! * \brief Get the name and description of a given result code * * A result code can be interpreted as a member of any one of several families. * * \param[in] code The result code to look up * \param[in] type How \p code should be interpreted * \param[out] name Where to store the result code's name * \param[out] desc Where to store the result code's description * * \return Standard Pacemaker return code */ int pcmk_result_get_strings(int code, enum pcmk_result_type type, const char **name, const char **desc) { const char *code_name = NULL; const char *code_desc = NULL; switch (type) { case pcmk_result_legacy: code_name = pcmk_errorname(code); code_desc = pcmk_strerror(code); break; case pcmk_result_rc: code_name = pcmk_rc_name(code); code_desc = pcmk_rc_str(code); break; case pcmk_result_exitcode: code_name = crm_exit_name(code); code_desc = crm_exit_str((crm_exit_t) code); break; default: return pcmk_rc_undetermined; } if (name != NULL) { *name = code_name; } if (desc != NULL) { *desc = code_desc; } return pcmk_rc_ok; } /*! * \internal * \brief Get the lower and upper bounds of a result code family * * \param[in] type Type of result code * \param[out] lower Where to store the lower bound * \param[out] upper Where to store the upper bound * * \return Standard Pacemaker return code * * \note There is no true upper bound on standard Pacemaker return codes or * legacy return codes. All system \p errno values are valid members of * these result code families, and there is no global upper limit nor a * constant by which to refer to the highest \p errno value on a given * system. */ int pcmk__result_bounds(enum pcmk_result_type type, int *lower, int *upper) { pcmk__assert((lower != NULL) && (upper != NULL)); switch (type) { case pcmk_result_legacy: *lower = pcmk_ok; *upper = 256; // should be enough for almost any system error code break; case pcmk_result_rc: *lower = pcmk_rc_error - pcmk__n_rc + 1; *upper = 256; break; case pcmk_result_exitcode: *lower = CRM_EX_OK; *upper = CRM_EX_MAX; break; default: *lower = 0; *upper = -1; return pcmk_rc_undetermined; } return pcmk_rc_ok; } /*! * \internal * \brief Log a failed assertion * * \param[in] file File making the assertion * \param[in] function Function making the assertion * \param[in] line Line of file making the assertion * \param[in] assert_condition String representation of assertion */ static void log_assertion_as(const char *file, const char *function, int line, const char *assert_condition) { if (!pcmk__is_daemon) { crm_enable_stderr(TRUE); // Make sure command-line user sees message } crm_err("%s: Triggered fatal assertion at %s:%d : %s", function, file, line, assert_condition); } /* coverity[+kill] */ /*! * \internal * \brief Log a failed assertion and abort * * \param[in] file File making the assertion * \param[in] function Function making the assertion * \param[in] line Line of file making the assertion * \param[in] assert_condition String representation of assertion * * \note This does not return */ _Noreturn void pcmk__abort_as(const char *file, const char *function, int line, const char *assert_condition) { log_assertion_as(file, function, line, assert_condition); abort(); } /* coverity[+kill] */ /*! * \internal * \brief Handle a failed assertion * * When called by a daemon, fork a child that aborts (to dump core), otherwise * abort the current process. * * \param[in] file File making the assertion * \param[in] function Function making the assertion * \param[in] line Line of file making the assertion * \param[in] assert_condition String representation of assertion */ static void fail_assert_as(const char *file, const char *function, int line, const char *assert_condition) { int status = 0; pid_t pid = 0; if (!pcmk__is_daemon) { pcmk__abort_as(file, function, line, assert_condition); // No return } pid = fork(); switch (pid) { case -1: // Fork failed crm_warn("%s: Cannot dump core for non-fatal assertion at %s:%d " ": %s", function, file, line, assert_condition); break; case 0: // Child process: just abort to dump core abort(); break; default: // Parent process: wait for child crm_err("%s: Forked child [%d] to record non-fatal assertion at " "%s:%d : %s", function, pid, file, line, assert_condition); crm_write_blackbox(SIGTRAP, NULL); do { if (waitpid(pid, &status, 0) == pid) { return; // Child finished dumping core } } while (errno == EINTR); if (errno == ECHILD) { // crm_mon ignores SIGCHLD crm_trace("Cannot wait on forked child [%d] " "(SIGCHLD is probably ignored)", pid); } else { crm_err("Cannot wait on forked child [%d]: %s", pid, pcmk_rc_str(errno)); } break; } } /* coverity[+kill] */ void crm_abort(const char *file, const char *function, int line, const char *assert_condition, gboolean do_core, gboolean do_fork) { if (!do_fork) { pcmk__abort_as(file, function, line, assert_condition); // No return } else if (do_core) { fail_assert_as(file, function, line, assert_condition); } else { log_assertion_as(file, function, line, assert_condition); } } // @COMPAT Legacy function return codes //! \deprecated Use standard return codes and pcmk_rc_name() instead const char * pcmk_errorname(int rc) { rc = abs(rc); switch (rc) { case pcmk_err_generic: return "pcmk_err_generic"; case pcmk_err_no_quorum: return "pcmk_err_no_quorum"; case pcmk_err_schema_validation: return "pcmk_err_schema_validation"; case pcmk_err_transform_failed: return "pcmk_err_transform_failed"; case pcmk_err_old_data: return "pcmk_err_old_data"; case pcmk_err_diff_failed: return "pcmk_err_diff_failed"; case pcmk_err_diff_resync: return "pcmk_err_diff_resync"; case pcmk_err_cib_modified: return "pcmk_err_cib_modified"; case pcmk_err_cib_backup: return "pcmk_err_cib_backup"; case pcmk_err_cib_save: return "pcmk_err_cib_save"; case pcmk_err_cib_corrupt: return "pcmk_err_cib_corrupt"; case pcmk_err_multiple: return "pcmk_err_multiple"; case pcmk_err_node_unknown: return "pcmk_err_node_unknown"; case pcmk_err_already: return "pcmk_err_already"; case pcmk_err_bad_nvpair: return "pcmk_err_bad_nvpair"; case pcmk_err_unknown_format: return "pcmk_err_unknown_format"; default: return pcmk_rc_name(rc); // system errno } } //! \deprecated Use standard return codes and pcmk_rc_str() instead const char * pcmk_strerror(int rc) { return pcmk_rc_str(pcmk_legacy2rc(rc)); } // Standard Pacemaker API return codes /* This array is used only for nonzero values of pcmk_rc_e. Its values must be * kept in the exact reverse order of the enum value numbering (i.e. add new * values to the end of the array). */ static const struct pcmk__rc_info { const char *name; const char *desc; int legacy_rc; } pcmk__rcs[] = { { "pcmk_rc_error", "Error", -pcmk_err_generic, }, { "pcmk_rc_unknown_format", "Unknown output format", -pcmk_err_unknown_format, }, { "pcmk_rc_bad_nvpair", "Bad name/value pair given", -pcmk_err_bad_nvpair, }, { "pcmk_rc_already", "Already in requested state", -pcmk_err_already, }, { "pcmk_rc_node_unknown", "Node not found", -pcmk_err_node_unknown, }, { "pcmk_rc_multiple", "Resource active on multiple nodes", -pcmk_err_multiple, }, { "pcmk_rc_cib_corrupt", "Could not parse on-disk configuration", -pcmk_err_cib_corrupt, }, { "pcmk_rc_cib_save", "Could not save new configuration to disk", -pcmk_err_cib_save, }, { "pcmk_rc_cib_backup", "Could not archive previous configuration", -pcmk_err_cib_backup, }, { "pcmk_rc_cib_modified", "On-disk configuration was manually modified", -pcmk_err_cib_modified, }, { "pcmk_rc_diff_resync", "Application of update diff failed, requesting full refresh", -pcmk_err_diff_resync, }, { "pcmk_rc_diff_failed", "Application of update diff failed", -pcmk_err_diff_failed, }, { "pcmk_rc_old_data", "Update was older than existing configuration", -pcmk_err_old_data, }, { "pcmk_rc_transform_failed", "Schema transform failed", -pcmk_err_transform_failed, }, { "pcmk_rc_schema_unchanged", "Schema is already the latest available", -pcmk_err_schema_unchanged, }, { "pcmk_rc_schema_validation", "Update does not conform to the configured schema", -pcmk_err_schema_validation, }, { "pcmk_rc_no_quorum", "Operation requires quorum", -pcmk_err_no_quorum, }, { "pcmk_rc_ipc_unauthorized", "IPC server is blocked by unauthorized process", -pcmk_err_generic, }, { "pcmk_rc_ipc_unresponsive", "IPC server is unresponsive", -pcmk_err_generic, }, { "pcmk_rc_ipc_pid_only", "IPC server process is active but not accepting connections", -pcmk_err_generic, }, { "pcmk_rc_op_unsatisfied", "Not applicable under current conditions", -pcmk_err_generic, }, { "pcmk_rc_undetermined", "Result undetermined", -pcmk_err_generic, }, { "pcmk_rc_before_range", "Result occurs before given range", -pcmk_err_generic, }, { "pcmk_rc_within_range", "Result occurs within given range", -pcmk_err_generic, }, { "pcmk_rc_after_range", "Result occurs after given range", -pcmk_err_generic, }, { "pcmk_rc_no_output", "Output message produced no output", -pcmk_err_generic, }, { "pcmk_rc_no_input", "Input file not available", -pcmk_err_generic, }, { "pcmk_rc_underflow", "Value too small to be stored in data type", -pcmk_err_generic, }, { "pcmk_rc_dot_error", "Error writing dot(1) file", -pcmk_err_generic, }, { "pcmk_rc_graph_error", "Error writing graph file", -pcmk_err_generic, }, { "pcmk_rc_invalid_transition", "Cluster simulation produced invalid transition", -pcmk_err_generic, }, { "pcmk_rc_unpack_error", "Unable to parse CIB XML", -pcmk_err_generic, }, { "pcmk_rc_duplicate_id", "Two or more XML elements have the same ID", -pcmk_err_generic, }, { "pcmk_rc_disabled", "Disabled", -pcmk_err_generic, }, { "pcmk_rc_bad_input", "Bad input value provided", -pcmk_err_generic, }, { "pcmk_rc_bad_xml_patch", "Bad XML patch format", -pcmk_err_generic, }, { "pcmk_rc_no_transaction", "No active transaction found", -pcmk_err_generic, }, { "pcmk_rc_ns_resolution", "Nameserver resolution error", -pcmk_err_generic, }, { "pcmk_rc_compression", "Compression/decompression error", -pcmk_err_generic, }, }; /*! * \internal * \brief The number of enum pcmk_rc_e values, excluding \c pcmk_rc_ok * * This constant stores the number of negative standard Pacemaker return codes. * These represent Pacemaker-custom error codes. The count does not include * positive system error numbers, nor does it include \c pcmk_rc_ok (success). */ const size_t pcmk__n_rc = PCMK__NELEM(pcmk__rcs); /*! * \brief Get a return code constant name as a string * * \param[in] rc Integer return code to convert * * \return String of constant name corresponding to rc */ const char * pcmk_rc_name(int rc) { if ((rc <= pcmk_rc_error) && ((pcmk_rc_error - rc) < pcmk__n_rc)) { return pcmk__rcs[pcmk_rc_error - rc].name; } switch (rc) { case pcmk_rc_ok: return "pcmk_rc_ok"; case E2BIG: return "E2BIG"; case EACCES: return "EACCES"; case EADDRINUSE: return "EADDRINUSE"; case EADDRNOTAVAIL: return "EADDRNOTAVAIL"; case EAFNOSUPPORT: return "EAFNOSUPPORT"; case EAGAIN: return "EAGAIN"; case EALREADY: return "EALREADY"; case EBADF: return "EBADF"; case EBADMSG: return "EBADMSG"; case EBUSY: return "EBUSY"; case ECANCELED: return "ECANCELED"; case ECHILD: return "ECHILD"; case ECOMM: return "ECOMM"; case ECONNABORTED: return "ECONNABORTED"; case ECONNREFUSED: return "ECONNREFUSED"; case ECONNRESET: return "ECONNRESET"; /* case EDEADLK: return "EDEADLK"; */ case EDESTADDRREQ: return "EDESTADDRREQ"; case EDOM: return "EDOM"; case EDQUOT: return "EDQUOT"; case EEXIST: return "EEXIST"; case EFAULT: return "EFAULT"; case EFBIG: return "EFBIG"; case EHOSTDOWN: return "EHOSTDOWN"; case EHOSTUNREACH: return "EHOSTUNREACH"; case EIDRM: return "EIDRM"; case EILSEQ: return "EILSEQ"; case EINPROGRESS: return "EINPROGRESS"; case EINTR: return "EINTR"; case EINVAL: return "EINVAL"; case EIO: return "EIO"; case EISCONN: return "EISCONN"; case EISDIR: return "EISDIR"; case ELIBACC: return "ELIBACC"; case ELOOP: return "ELOOP"; case EMFILE: return "EMFILE"; case EMLINK: return "EMLINK"; case EMSGSIZE: return "EMSGSIZE"; #ifdef EMULTIHOP // Not available on OpenBSD case EMULTIHOP: return "EMULTIHOP"; #endif case ENAMETOOLONG: return "ENAMETOOLONG"; case ENETDOWN: return "ENETDOWN"; case ENETRESET: return "ENETRESET"; case ENETUNREACH: return "ENETUNREACH"; case ENFILE: return "ENFILE"; case ENOBUFS: return "ENOBUFS"; case ENODATA: return "ENODATA"; case ENODEV: return "ENODEV"; case ENOENT: return "ENOENT"; case ENOEXEC: return "ENOEXEC"; case ENOKEY: return "ENOKEY"; case ENOLCK: return "ENOLCK"; #ifdef ENOLINK // Not available on OpenBSD case ENOLINK: return "ENOLINK"; #endif case ENOMEM: return "ENOMEM"; case ENOMSG: return "ENOMSG"; case ENOPROTOOPT: return "ENOPROTOOPT"; case ENOSPC: return "ENOSPC"; #ifdef ENOSR case ENOSR: return "ENOSR"; #endif #ifdef ENOSTR case ENOSTR: return "ENOSTR"; #endif case ENOSYS: return "ENOSYS"; case ENOTBLK: return "ENOTBLK"; case ENOTCONN: return "ENOTCONN"; case ENOTDIR: return "ENOTDIR"; case ENOTEMPTY: return "ENOTEMPTY"; case ENOTSOCK: return "ENOTSOCK"; #if ENOTSUP != EOPNOTSUPP case ENOTSUP: return "ENOTSUP"; #endif case ENOTTY: return "ENOTTY"; case ENOTUNIQ: return "ENOTUNIQ"; case ENXIO: return "ENXIO"; case EOPNOTSUPP: return "EOPNOTSUPP"; case EOVERFLOW: return "EOVERFLOW"; case EPERM: return "EPERM"; case EPFNOSUPPORT: return "EPFNOSUPPORT"; case EPIPE: return "EPIPE"; case EPROTO: return "EPROTO"; case EPROTONOSUPPORT: return "EPROTONOSUPPORT"; case EPROTOTYPE: return "EPROTOTYPE"; case ERANGE: return "ERANGE"; case EREMOTE: return "EREMOTE"; case EREMOTEIO: return "EREMOTEIO"; case EROFS: return "EROFS"; case ESHUTDOWN: return "ESHUTDOWN"; case ESPIPE: return "ESPIPE"; case ESOCKTNOSUPPORT: return "ESOCKTNOSUPPORT"; case ESRCH: return "ESRCH"; case ESTALE: return "ESTALE"; case ETIME: return "ETIME"; case ETIMEDOUT: return "ETIMEDOUT"; case ETXTBSY: return "ETXTBSY"; #ifdef EUNATCH case EUNATCH: return "EUNATCH"; #endif case EUSERS: return "EUSERS"; /* case EWOULDBLOCK: return "EWOULDBLOCK"; */ case EXDEV: return "EXDEV"; #ifdef EBADE // Not available on OS X case EBADE: return "EBADE"; case EBADFD: return "EBADFD"; case EBADSLT: return "EBADSLT"; case EDEADLOCK: return "EDEADLOCK"; case EBADR: return "EBADR"; case EBADRQC: return "EBADRQC"; case ECHRNG: return "ECHRNG"; #ifdef EISNAM // Not available on OS X, Illumos, Solaris case EISNAM: return "EISNAM"; case EKEYEXPIRED: return "EKEYEXPIRED"; case EKEYREVOKED: return "EKEYREVOKED"; #endif case EKEYREJECTED: return "EKEYREJECTED"; case EL2HLT: return "EL2HLT"; case EL2NSYNC: return "EL2NSYNC"; case EL3HLT: return "EL3HLT"; case EL3RST: return "EL3RST"; case ELIBBAD: return "ELIBBAD"; case ELIBMAX: return "ELIBMAX"; case ELIBSCN: return "ELIBSCN"; case ELIBEXEC: return "ELIBEXEC"; #ifdef ENOMEDIUM // Not available on OS X, Illumos, Solaris case ENOMEDIUM: return "ENOMEDIUM"; case EMEDIUMTYPE: return "EMEDIUMTYPE"; #endif case ENONET: return "ENONET"; case ENOPKG: return "ENOPKG"; case EREMCHG: return "EREMCHG"; case ERESTART: return "ERESTART"; case ESTRPIPE: return "ESTRPIPE"; #ifdef EUCLEAN // Not available on OS X, Illumos, Solaris case EUCLEAN: return "EUCLEAN"; #endif case EXFULL: return "EXFULL"; #endif // EBADE default: return "Unknown"; } } /*! * \brief Get a user-friendly description of a return code * * \param[in] rc Integer return code to convert * * \return String description of rc */ const char * pcmk_rc_str(int rc) { if (rc == pcmk_rc_ok) { return "OK"; } if ((rc <= pcmk_rc_error) && ((pcmk_rc_error - rc) < pcmk__n_rc)) { return pcmk__rcs[pcmk_rc_error - rc].desc; } if (rc < 0) { return "Error"; } // Handle values that could be defined by system or by portability.h switch (rc) { #ifdef PCMK__ENOTUNIQ case ENOTUNIQ: return "Name not unique on network"; #endif #ifdef PCMK__ECOMM case ECOMM: return "Communication error on send"; #endif #ifdef PCMK__ELIBACC case ELIBACC: return "Can not access a needed shared library"; #endif #ifdef PCMK__EREMOTEIO case EREMOTEIO: return "Remote I/O error"; #endif #ifdef PCMK__ENOKEY case ENOKEY: return "Required key not available"; #endif #ifdef PCMK__ENODATA case ENODATA: return "No data available"; #endif #ifdef PCMK__ETIME case ETIME: return "Timer expired"; #endif #ifdef PCMK__EKEYREJECTED case EKEYREJECTED: return "Key was rejected by service"; #endif default: return strerror(rc); } } // This returns negative values for errors //! \deprecated Use standard return codes instead int pcmk_rc2legacy(int rc) { if (rc >= 0) { return -rc; // OK or system errno } if ((rc <= pcmk_rc_error) && ((pcmk_rc_error - rc) < pcmk__n_rc)) { return pcmk__rcs[pcmk_rc_error - rc].legacy_rc; } return -pcmk_err_generic; } //! \deprecated Use standard return codes instead int pcmk_legacy2rc(int legacy_rc) { legacy_rc = abs(legacy_rc); switch (legacy_rc) { case pcmk_err_no_quorum: return pcmk_rc_no_quorum; case pcmk_err_schema_validation: return pcmk_rc_schema_validation; case pcmk_err_schema_unchanged: return pcmk_rc_schema_unchanged; case pcmk_err_transform_failed: return pcmk_rc_transform_failed; case pcmk_err_old_data: return pcmk_rc_old_data; case pcmk_err_diff_failed: return pcmk_rc_diff_failed; case pcmk_err_diff_resync: return pcmk_rc_diff_resync; case pcmk_err_cib_modified: return pcmk_rc_cib_modified; case pcmk_err_cib_backup: return pcmk_rc_cib_backup; case pcmk_err_cib_save: return pcmk_rc_cib_save; case pcmk_err_cib_corrupt: return pcmk_rc_cib_corrupt; case pcmk_err_multiple: return pcmk_rc_multiple; case pcmk_err_node_unknown: return pcmk_rc_node_unknown; case pcmk_err_already: return pcmk_rc_already; case pcmk_err_bad_nvpair: return pcmk_rc_bad_nvpair; case pcmk_err_unknown_format: return pcmk_rc_unknown_format; case pcmk_err_generic: return pcmk_rc_error; case pcmk_ok: return pcmk_rc_ok; default: return legacy_rc; // system errno } } // Exit status codes const char * crm_exit_name(crm_exit_t exit_code) { switch (exit_code) { case CRM_EX_OK: return "CRM_EX_OK"; case CRM_EX_ERROR: return "CRM_EX_ERROR"; case CRM_EX_INVALID_PARAM: return "CRM_EX_INVALID_PARAM"; case CRM_EX_UNIMPLEMENT_FEATURE: return "CRM_EX_UNIMPLEMENT_FEATURE"; case CRM_EX_INSUFFICIENT_PRIV: return "CRM_EX_INSUFFICIENT_PRIV"; case CRM_EX_NOT_INSTALLED: return "CRM_EX_NOT_INSTALLED"; case CRM_EX_NOT_CONFIGURED: return "CRM_EX_NOT_CONFIGURED"; case CRM_EX_NOT_RUNNING: return "CRM_EX_NOT_RUNNING"; case CRM_EX_PROMOTED: return "CRM_EX_PROMOTED"; case CRM_EX_FAILED_PROMOTED: return "CRM_EX_FAILED_PROMOTED"; case CRM_EX_USAGE: return "CRM_EX_USAGE"; case CRM_EX_DATAERR: return "CRM_EX_DATAERR"; case CRM_EX_NOINPUT: return "CRM_EX_NOINPUT"; case CRM_EX_NOUSER: return "CRM_EX_NOUSER"; case CRM_EX_NOHOST: return "CRM_EX_NOHOST"; case CRM_EX_UNAVAILABLE: return "CRM_EX_UNAVAILABLE"; case CRM_EX_SOFTWARE: return "CRM_EX_SOFTWARE"; case CRM_EX_OSERR: return "CRM_EX_OSERR"; case CRM_EX_OSFILE: return "CRM_EX_OSFILE"; case CRM_EX_CANTCREAT: return "CRM_EX_CANTCREAT"; case CRM_EX_IOERR: return "CRM_EX_IOERR"; case CRM_EX_TEMPFAIL: return "CRM_EX_TEMPFAIL"; case CRM_EX_PROTOCOL: return "CRM_EX_PROTOCOL"; case CRM_EX_NOPERM: return "CRM_EX_NOPERM"; case CRM_EX_CONFIG: return "CRM_EX_CONFIG"; case CRM_EX_FATAL: return "CRM_EX_FATAL"; case CRM_EX_PANIC: return "CRM_EX_PANIC"; case CRM_EX_DISCONNECT: return "CRM_EX_DISCONNECT"; case CRM_EX_DIGEST: return "CRM_EX_DIGEST"; case CRM_EX_NOSUCH: return "CRM_EX_NOSUCH"; case CRM_EX_QUORUM: return "CRM_EX_QUORUM"; case CRM_EX_UNSAFE: return "CRM_EX_UNSAFE"; case CRM_EX_EXISTS: return "CRM_EX_EXISTS"; case CRM_EX_MULTIPLE: return "CRM_EX_MULTIPLE"; case CRM_EX_EXPIRED: return "CRM_EX_EXPIRED"; case CRM_EX_NOT_YET_IN_EFFECT: return "CRM_EX_NOT_YET_IN_EFFECT"; case CRM_EX_INDETERMINATE: return "CRM_EX_INDETERMINATE"; case CRM_EX_UNSATISFIED: return "CRM_EX_UNSATISFIED"; case CRM_EX_OLD: return "CRM_EX_OLD"; case CRM_EX_TIMEOUT: return "CRM_EX_TIMEOUT"; case CRM_EX_DEGRADED: return "CRM_EX_DEGRADED"; case CRM_EX_DEGRADED_PROMOTED: return "CRM_EX_DEGRADED_PROMOTED"; case CRM_EX_NONE: return "CRM_EX_NONE"; case CRM_EX_MAX: return "CRM_EX_UNKNOWN"; } return "CRM_EX_UNKNOWN"; } const char * crm_exit_str(crm_exit_t exit_code) { switch (exit_code) { case CRM_EX_OK: return "OK"; case CRM_EX_ERROR: return "Error occurred"; case CRM_EX_INVALID_PARAM: return "Invalid parameter"; case CRM_EX_UNIMPLEMENT_FEATURE: return "Unimplemented"; case CRM_EX_INSUFFICIENT_PRIV: return "Insufficient privileges"; case CRM_EX_NOT_INSTALLED: return "Not installed"; case CRM_EX_NOT_CONFIGURED: return "Not configured"; case CRM_EX_NOT_RUNNING: return "Not running"; case CRM_EX_PROMOTED: return "Promoted"; case CRM_EX_FAILED_PROMOTED: return "Failed in promoted role"; case CRM_EX_USAGE: return "Incorrect usage"; case CRM_EX_DATAERR: return "Invalid data given"; case CRM_EX_NOINPUT: return "Input file not available"; case CRM_EX_NOUSER: return "User does not exist"; case CRM_EX_NOHOST: return "Host does not exist"; case CRM_EX_UNAVAILABLE: return "Necessary service unavailable"; case CRM_EX_SOFTWARE: return "Internal software bug"; case CRM_EX_OSERR: return "Operating system error occurred"; case CRM_EX_OSFILE: return "System file not available"; case CRM_EX_CANTCREAT: return "Cannot create output file"; case CRM_EX_IOERR: return "I/O error occurred"; case CRM_EX_TEMPFAIL: return "Temporary failure, try again"; case CRM_EX_PROTOCOL: return "Protocol violated"; case CRM_EX_NOPERM: return "Insufficient privileges"; case CRM_EX_CONFIG: return "Invalid configuration"; case CRM_EX_FATAL: return "Fatal error occurred, will not respawn"; case CRM_EX_PANIC: return "System panic required"; case CRM_EX_DISCONNECT: return "Not connected"; case CRM_EX_DIGEST: return "Digest mismatch"; case CRM_EX_NOSUCH: return "No such object"; case CRM_EX_QUORUM: return "Quorum required"; case CRM_EX_UNSAFE: return "Operation not safe"; case CRM_EX_EXISTS: return "Requested item already exists"; case CRM_EX_MULTIPLE: return "Multiple items match request"; case CRM_EX_EXPIRED: return "Requested item has expired"; case CRM_EX_NOT_YET_IN_EFFECT: return "Requested item is not yet in effect"; case CRM_EX_INDETERMINATE: return "Could not determine status"; case CRM_EX_UNSATISFIED: return "Not applicable under current conditions"; case CRM_EX_OLD: return "Update was older than existing configuration"; case CRM_EX_TIMEOUT: return "Timeout occurred"; case CRM_EX_DEGRADED: return "Service is active but might fail soon"; case CRM_EX_DEGRADED_PROMOTED: return "Service is promoted but might fail soon"; case CRM_EX_NONE: return "No exit status available"; case CRM_EX_MAX: return "Error occurred"; } if ((exit_code > 128) && (exit_code < CRM_EX_MAX)) { return "Interrupted by signal"; } return "Unknown exit status"; } /*! * \brief Map a function return code to the most similar exit code * * \param[in] rc Function return code * * \return Most similar exit code */ crm_exit_t pcmk_rc2exitc(int rc) { switch (rc) { case pcmk_rc_ok: case pcmk_rc_no_output: // quiet mode, or nothing to output return CRM_EX_OK; case pcmk_rc_no_quorum: return CRM_EX_QUORUM; case pcmk_rc_old_data: return CRM_EX_OLD; case pcmk_rc_schema_validation: case pcmk_rc_transform_failed: case pcmk_rc_unpack_error: return CRM_EX_CONFIG; case pcmk_rc_bad_nvpair: return CRM_EX_INVALID_PARAM; case EACCES: return CRM_EX_INSUFFICIENT_PRIV; case EBADF: case EINVAL: case EFAULT: case ENOSYS: case EOVERFLOW: case pcmk_rc_underflow: case pcmk_rc_compression: return CRM_EX_SOFTWARE; case EBADMSG: case EMSGSIZE: case ENOMSG: case ENOPROTOOPT: case EPROTO: case EPROTONOSUPPORT: case EPROTOTYPE: return CRM_EX_PROTOCOL; case ECOMM: case ENOMEM: return CRM_EX_OSERR; case ECONNABORTED: case ECONNREFUSED: case ECONNRESET: case ENOTCONN: return CRM_EX_DISCONNECT; case EEXIST: case pcmk_rc_already: return CRM_EX_EXISTS; case EIO: case pcmk_rc_dot_error: case pcmk_rc_graph_error: return CRM_EX_IOERR; case ENOTSUP: #if EOPNOTSUPP != ENOTSUP case EOPNOTSUPP: #endif return CRM_EX_UNIMPLEMENT_FEATURE; case ENOTUNIQ: case pcmk_rc_multiple: return CRM_EX_MULTIPLE; case ENODEV: case ENOENT: case ENXIO: case pcmk_rc_no_transaction: case pcmk_rc_unknown_format: return CRM_EX_NOSUCH; case pcmk_rc_node_unknown: case pcmk_rc_ns_resolution: return CRM_EX_NOHOST; case ETIME: case ETIMEDOUT: return CRM_EX_TIMEOUT; case EAGAIN: case EBUSY: return CRM_EX_UNSATISFIED; case pcmk_rc_before_range: return CRM_EX_NOT_YET_IN_EFFECT; case pcmk_rc_after_range: return CRM_EX_EXPIRED; case pcmk_rc_undetermined: return CRM_EX_INDETERMINATE; case pcmk_rc_op_unsatisfied: return CRM_EX_UNSATISFIED; case pcmk_rc_within_range: return CRM_EX_OK; case pcmk_rc_no_input: return CRM_EX_NOINPUT; case pcmk_rc_duplicate_id: return CRM_EX_MULTIPLE; case pcmk_rc_bad_input: case pcmk_rc_bad_xml_patch: return CRM_EX_DATAERR; default: return CRM_EX_ERROR; } } /*! * \brief Map a function return code to the most similar OCF exit code * * \param[in] rc Function return code * * \return Most similar OCF exit code */ enum ocf_exitcode pcmk_rc2ocf(int rc) { switch (rc) { case pcmk_rc_ok: return PCMK_OCF_OK; case pcmk_rc_bad_nvpair: return PCMK_OCF_INVALID_PARAM; case EACCES: return PCMK_OCF_INSUFFICIENT_PRIV; case ENOTSUP: #if EOPNOTSUPP != ENOTSUP case EOPNOTSUPP: #endif return PCMK_OCF_UNIMPLEMENT_FEATURE; default: return PCMK_OCF_UNKNOWN_ERROR; } } // Other functions /*! * \brief Map a getaddrinfo() return code to the most similar Pacemaker * return code * * \param[in] gai getaddrinfo() return code * * \return Most similar Pacemaker return code */ int pcmk__gaierror2rc(int gai) { switch (gai) { case 0: return pcmk_rc_ok; case EAI_AGAIN: return EAGAIN; case EAI_BADFLAGS: case EAI_SERVICE: return EINVAL; case EAI_FAMILY: return EAFNOSUPPORT; case EAI_MEMORY: return ENOMEM; case EAI_NONAME: return pcmk_rc_node_unknown; case EAI_SOCKTYPE: return ESOCKTNOSUPPORT; case EAI_SYSTEM: return errno; default: return pcmk_rc_ns_resolution; } } /*! * \brief Map a bz2 return code to the most similar Pacemaker return code * * \param[in] bz2 bz2 return code * * \return Most similar Pacemaker return code */ int pcmk__bzlib2rc(int bz2) { switch (bz2) { case BZ_OK: case BZ_RUN_OK: case BZ_FLUSH_OK: case BZ_FINISH_OK: case BZ_STREAM_END: return pcmk_rc_ok; case BZ_MEM_ERROR: return ENOMEM; case BZ_DATA_ERROR: case BZ_DATA_ERROR_MAGIC: case BZ_UNEXPECTED_EOF: return pcmk_rc_bad_input; case BZ_IO_ERROR: return EIO; case BZ_OUTBUFF_FULL: return EFBIG; default: return pcmk_rc_compression; } } crm_exit_t -crm_exit(crm_exit_t rc) +crm_exit(crm_exit_t exit_status) { /* A compiler could theoretically use any type for crm_exit_t, but an int * should always hold it, so cast to int to keep static analysis happy. */ - if ((((int) rc) < 0) || (((int) rc) > CRM_EX_MAX)) { - rc = CRM_EX_ERROR; + if ((((int) exit_status) < 0) || (((int) exit_status) > CRM_EX_MAX)) { + exit_status = CRM_EX_ERROR; } - mainloop_cleanup(); - pcmk__xml_cleanup(); - - if (crm_system_name) { - crm_info("Exiting %s " QB_XS " with status %d", crm_system_name, rc); - free(crm_system_name); - } else { - crm_trace("Exiting with status %d", rc); - } - pcmk__free_common_logger(); - qb_log_fini(); // Don't log anything after this point - - exit(rc); + crm_info("Exiting %s " QB_XS " with status %d (%s: %s)", + pcmk__s(crm_system_name, "process"), exit_status, + crm_exit_name(exit_status), crm_exit_str(exit_status)); + pcmk_common_cleanup(); + exit(exit_status); } /* * External action results */ /*! * \internal * \brief Set the result of an action * * \param[out] result Where to set action result * \param[in] exit_status OCF exit status to set * \param[in] exec_status Execution status to set * \param[in] exit_reason Human-friendly description of event to set */ void pcmk__set_result(pcmk__action_result_t *result, int exit_status, enum pcmk_exec_status exec_status, const char *exit_reason) { if (result == NULL) { return; } result->exit_status = exit_status; result->execution_status = exec_status; if (!pcmk__str_eq(result->exit_reason, exit_reason, pcmk__str_none)) { free(result->exit_reason); result->exit_reason = (exit_reason == NULL)? NULL : strdup(exit_reason); } } /*! * \internal * \brief Set the result of an action, with a formatted exit reason * * \param[out] result Where to set action result * \param[in] exit_status OCF exit status to set * \param[in] exec_status Execution status to set * \param[in] format printf-style format for a human-friendly * description of reason for result * \param[in] ... arguments for \p format */ G_GNUC_PRINTF(4, 5) void pcmk__format_result(pcmk__action_result_t *result, int exit_status, enum pcmk_exec_status exec_status, const char *format, ...) { va_list ap; int len = 0; char *reason = NULL; if (result == NULL) { return; } result->exit_status = exit_status; result->execution_status = exec_status; if (format != NULL) { va_start(ap, format); len = vasprintf(&reason, format, ap); pcmk__assert(len > 0); va_end(ap); } free(result->exit_reason); result->exit_reason = reason; } /*! * \internal * \brief Set the output of an action * * \param[out] result Action result to set output for * \param[in] out Action output to set (must be dynamically * allocated) * \param[in] err Action error output to set (must be dynamically * allocated) * * \note \p result will take ownership of \p out and \p err, so the caller * should not free them. */ void pcmk__set_result_output(pcmk__action_result_t *result, char *out, char *err) { if (result == NULL) { return; } free(result->action_stdout); result->action_stdout = out; free(result->action_stderr); result->action_stderr = err; } /*! * \internal * \brief Clear a result's exit reason, output, and error output * * \param[in,out] result Result to reset */ void pcmk__reset_result(pcmk__action_result_t *result) { if (result == NULL) { return; } free(result->exit_reason); result->exit_reason = NULL; free(result->action_stdout); result->action_stdout = NULL; free(result->action_stderr); result->action_stderr = NULL; } /*! * \internal * \brief Copy the result of an action * * \param[in] src Result to copy * \param[out] dst Where to copy \p src to */ void pcmk__copy_result(const pcmk__action_result_t *src, pcmk__action_result_t *dst) { CRM_CHECK((src != NULL) && (dst != NULL), return); dst->exit_status = src->exit_status; dst->execution_status = src->execution_status; dst->exit_reason = pcmk__str_copy(src->exit_reason); dst->action_stdout = pcmk__str_copy(src->action_stdout); dst->action_stderr = pcmk__str_copy(src->action_stderr); } diff --git a/lib/common/utils.c b/lib/common/utils.c index 16f543711a..c137330a95 100644 --- a/lib/common/utils.c +++ b/lib/common/utils.c @@ -1,472 +1,492 @@ /* * Copyright 2004-2024 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 #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__xml_cleanup(); + pcmk__free_common_logger(); + qb_log_fini(); // Don't log anything after this point + + free(crm_system_name); + crm_system_name = NULL; +} + 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; } void crm_gnutls_global_init(void) { signal(SIGPIPE, SIG_IGN); gnutls_global_init(); } /*! * \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 /*! * \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/pacemaker/pcmk_sched_instances.c b/lib/pacemaker/pcmk_sched_instances.c index 1c95a1333e..a204a65525 100644 --- a/lib/pacemaker/pcmk_sched_instances.c +++ b/lib/pacemaker/pcmk_sched_instances.c @@ -1,1714 +1,1714 @@ /* * Copyright 2004-2024 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. */ /* This file is intended for code usable with both clone instances and bundle * replica containers. */ #include #include #include #include "libpacemaker_private.h" /*! * \internal * \brief Check whether a node is allowed to run an instance * * \param[in] instance Clone instance or bundle container to check * \param[in] node Node to check * \param[in] max_per_node Maximum number of instances allowed to run on a node * * \return true if \p node is allowed to run \p instance, otherwise false */ static bool can_run_instance(const pcmk_resource_t *instance, const pcmk_node_t *node, int max_per_node) { pcmk_node_t *allowed_node = NULL; if (pcmk_is_set(instance->flags, pcmk__rsc_removed)) { pcmk__rsc_trace(instance, "%s cannot run on %s: orphaned", instance->id, pcmk__node_name(node)); return false; } if (!pcmk__node_available(node, false, false)) { pcmk__rsc_trace(instance, "%s cannot run on %s: node cannot run resources", instance->id, pcmk__node_name(node)); return false; } allowed_node = pcmk__top_allowed_node(instance, node); if (allowed_node == NULL) { crm_warn("%s cannot run on %s: node not allowed", instance->id, pcmk__node_name(node)); return false; } if (allowed_node->assign->score < 0) { pcmk__rsc_trace(instance, "%s cannot run on %s: parent score is %s there", instance->id, pcmk__node_name(node), pcmk_readable_score(allowed_node->assign->score)); return false; } if (allowed_node->assign->count >= max_per_node) { pcmk__rsc_trace(instance, "%s cannot run on %s: node already has %d instance%s", instance->id, pcmk__node_name(node), max_per_node, pcmk__plural_s(max_per_node)); return false; } pcmk__rsc_trace(instance, "%s can run on %s (%d already running)", instance->id, pcmk__node_name(node), allowed_node->assign->count); return true; } /*! * \internal * \brief Ban a clone instance or bundle replica from unavailable allowed nodes * * \param[in,out] instance Clone instance or bundle replica to ban * \param[in] max_per_node Maximum instances allowed to run on a node */ static void ban_unavailable_allowed_nodes(pcmk_resource_t *instance, int max_per_node) { if (instance->priv->allowed_nodes != NULL) { GHashTableIter iter; pcmk_node_t *node = NULL; g_hash_table_iter_init(&iter, instance->priv->allowed_nodes); while (g_hash_table_iter_next(&iter, NULL, (void **) &node)) { if (!can_run_instance(instance, node, max_per_node)) { pcmk__rsc_trace(instance, "Banning %s from unavailable node %s", instance->id, pcmk__node_name(node)); node->assign->score = -PCMK_SCORE_INFINITY; for (GList *child_iter = instance->priv->children; child_iter != NULL; child_iter = child_iter->next) { pcmk_resource_t *child = child_iter->data; pcmk_node_t *child_node = NULL; child_node = g_hash_table_lookup(child->priv->allowed_nodes, node->priv->id); if (child_node != NULL) { pcmk__rsc_trace(instance, "Banning %s child %s " "from unavailable node %s", instance->id, child->id, pcmk__node_name(node)); child_node->assign->score = -PCMK_SCORE_INFINITY; } } } } } } /*! * \internal * \brief Create a hash table with a single node in it * * \param[in] node Node to copy into new table * * \return Newly created hash table containing a copy of \p node * \note The caller is responsible for freeing the result with * g_hash_table_destroy(). */ static GHashTable * new_node_table(pcmk_node_t *node) { - GHashTable *table = pcmk__strkey_table(NULL, free); + GHashTable *table = pcmk__strkey_table(NULL, pcmk__free_node_copy); node = pe__copy_node(node); g_hash_table_insert(table, (gpointer) node->priv->id, node); return table; } /*! * \internal * \brief Apply a resource's parent's colocation scores to a node table * * \param[in] rsc Resource whose colocations should be applied * \param[in,out] nodes Node table to apply colocations to */ static void apply_parent_colocations(const pcmk_resource_t *rsc, GHashTable **nodes) { GList *colocations = pcmk__this_with_colocations(rsc); for (const GList *iter = colocations; iter != NULL; iter = iter->next) { const pcmk__colocation_t *colocation = iter->data; pcmk_resource_t *other = colocation->primary; float factor = colocation->score / (float) PCMK_SCORE_INFINITY; other->priv->cmds->add_colocated_node_scores(other, rsc, rsc->id, nodes, colocation, factor, pcmk__coloc_select_default); } g_list_free(colocations); colocations = pcmk__with_this_colocations(rsc); for (const GList *iter = colocations; iter != NULL; iter = iter->next) { const pcmk__colocation_t *colocation = iter->data; pcmk_resource_t *other = colocation->dependent; float factor = colocation->score / (float) PCMK_SCORE_INFINITY; if (!pcmk__colocation_has_influence(colocation, rsc)) { continue; } other->priv->cmds->add_colocated_node_scores(other, rsc, rsc->id, nodes, colocation, factor, pcmk__coloc_select_nonnegative); } g_list_free(colocations); } /*! * \internal * \brief Compare clone or bundle instances based on colocation scores * * Determine the relative order in which two clone or bundle instances should be * assigned to nodes, considering the scores of colocation constraints directly * or indirectly involving them. * * \param[in] instance1 First instance to compare * \param[in] instance2 Second instance to compare * * \return A negative number if \p instance1 should be assigned first, * a positive number if \p instance2 should be assigned first, * or 0 if assignment order doesn't matter */ static int cmp_instance_by_colocation(const pcmk_resource_t *instance1, const pcmk_resource_t *instance2) { int rc = 0; pcmk_node_t *node1 = NULL; pcmk_node_t *node2 = NULL; pcmk_node_t *current_node1 = pcmk__current_node(instance1); pcmk_node_t *current_node2 = pcmk__current_node(instance2); GHashTable *colocated_scores1 = NULL; GHashTable *colocated_scores2 = NULL; pcmk__assert((instance1 != NULL) && (instance1->priv->parent != NULL) && (instance2 != NULL) && (instance2->priv->parent != NULL) && (current_node1 != NULL) && (current_node2 != NULL)); // Create node tables initialized with each node colocated_scores1 = new_node_table(current_node1); colocated_scores2 = new_node_table(current_node2); // Apply parental colocations apply_parent_colocations(instance1, &colocated_scores1); apply_parent_colocations(instance2, &colocated_scores2); // Find original nodes again, with scores updated for colocations node1 = g_hash_table_lookup(colocated_scores1, current_node1->priv->id); node2 = g_hash_table_lookup(colocated_scores2, current_node2->priv->id); // Compare nodes by updated scores if (node1->assign->score < node2->assign->score) { crm_trace("Assign %s (%d on %s) after %s (%d on %s)", instance1->id, node1->assign->score, pcmk__node_name(node1), instance2->id, node2->assign->score, pcmk__node_name(node2)); rc = 1; } else if (node1->assign->score > node2->assign->score) { crm_trace("Assign %s (%d on %s) before %s (%d on %s)", instance1->id, node1->assign->score, pcmk__node_name(node1), instance2->id, node2->assign->score, pcmk__node_name(node2)); rc = -1; } g_hash_table_destroy(colocated_scores1); g_hash_table_destroy(colocated_scores2); return rc; } /*! * \internal * \brief Check whether a resource or any of its children are failed * * \param[in] rsc Resource to check * * \return true if \p rsc or any of its children are failed, otherwise false */ static bool did_fail(const pcmk_resource_t *rsc) { if (pcmk_is_set(rsc->flags, pcmk__rsc_failed)) { return true; } for (GList *iter = rsc->priv->children; iter != NULL; iter = iter->next) { if (did_fail((const pcmk_resource_t *) iter->data)) { return true; } } return false; } /*! * \internal * \brief Check whether a node is allowed to run a resource * * \param[in] rsc Resource to check * \param[in,out] node Node to check (will be set NULL if not allowed) * * \return true if *node is either NULL or allowed for \p rsc, otherwise false */ static bool node_is_allowed(const pcmk_resource_t *rsc, pcmk_node_t **node) { if (*node != NULL) { pcmk_node_t *allowed = g_hash_table_lookup(rsc->priv->allowed_nodes, (*node)->priv->id); if ((allowed == NULL) || (allowed->assign->score < 0)) { pcmk__rsc_trace(rsc, "%s: current location (%s) is unavailable", rsc->id, pcmk__node_name(*node)); *node = NULL; return false; } } return true; } /*! * \internal * \brief Compare two clone or bundle instances' instance numbers * * \param[in] a First instance to compare * \param[in] b Second instance to compare * * \return A negative number if \p a's instance number is lower, * a positive number if \p b's instance number is lower, * or 0 if their instance numbers are the same */ gint pcmk__cmp_instance_number(gconstpointer a, gconstpointer b) { const pcmk_resource_t *instance1 = (const pcmk_resource_t *) a; const pcmk_resource_t *instance2 = (const pcmk_resource_t *) b; char *div1 = NULL; char *div2 = NULL; pcmk__assert((instance1 != NULL) && (instance2 != NULL)); // Clone numbers are after a colon, bundle numbers after a dash div1 = strrchr(instance1->id, ':'); if (div1 == NULL) { div1 = strrchr(instance1->id, '-'); } div2 = strrchr(instance2->id, ':'); if (div2 == NULL) { div2 = strrchr(instance2->id, '-'); } pcmk__assert((div1 != NULL) && (div2 != NULL)); return (gint) (strtol(div1 + 1, NULL, 10) - strtol(div2 + 1, NULL, 10)); } /*! * \internal * \brief Compare clone or bundle instances according to assignment order * * Compare two clone or bundle instances according to the order they should be * assigned to nodes, preferring (in order): * * - Active instance that is less multiply active * - Instance that is not active on a disallowed node * - Instance with higher configured priority * - Active instance whose current node can run resources * - Active instance whose parent is allowed on current node * - Active instance whose current node has fewer other instances * - Active instance * - Instance that isn't failed * - Instance whose colocations result in higher score on current node * - Instance with lower ID in lexicographic order * * \param[in] a First instance to compare * \param[in] b Second instance to compare * * \return A negative number if \p a should be assigned first, * a positive number if \p b should be assigned first, * or 0 if assignment order doesn't matter */ gint pcmk__cmp_instance(gconstpointer a, gconstpointer b) { int rc = 0; pcmk_node_t *node1 = NULL; pcmk_node_t *node2 = NULL; unsigned int nnodes1 = 0; unsigned int nnodes2 = 0; bool can1 = true; bool can2 = true; const pcmk_resource_t *instance1 = (const pcmk_resource_t *) a; const pcmk_resource_t *instance2 = (const pcmk_resource_t *) b; pcmk__assert((instance1 != NULL) && (instance2 != NULL)); node1 = instance1->priv->fns->active_node(instance1, &nnodes1, NULL); node2 = instance2->priv->fns->active_node(instance2, &nnodes2, NULL); /* If both instances are running and at least one is multiply * active, prefer instance that's running on fewer nodes. */ if ((nnodes1 > 0) && (nnodes2 > 0)) { if (nnodes1 < nnodes2) { crm_trace("Assign %s (active on %d) before %s (active on %d): " "less multiply active", instance1->id, nnodes1, instance2->id, nnodes2); return -1; } else if (nnodes1 > nnodes2) { crm_trace("Assign %s (active on %d) after %s (active on %d): " "more multiply active", instance1->id, nnodes1, instance2->id, nnodes2); return 1; } } /* An instance that is either inactive or active on an allowed node is * preferred over an instance that is active on a no-longer-allowed node. */ can1 = node_is_allowed(instance1, &node1); can2 = node_is_allowed(instance2, &node2); if (can1 && !can2) { crm_trace("Assign %s before %s: not active on a disallowed node", instance1->id, instance2->id); return -1; } else if (!can1 && can2) { crm_trace("Assign %s after %s: active on a disallowed node", instance1->id, instance2->id); return 1; } // Prefer instance with higher configured priority if (instance1->priv->priority > instance2->priv->priority) { crm_trace("Assign %s before %s: priority (%d > %d)", instance1->id, instance2->id, instance1->priv->priority, instance2->priv->priority); return -1; } else if (instance1->priv->priority < instance2->priv->priority) { crm_trace("Assign %s after %s: priority (%d < %d)", instance1->id, instance2->id, instance1->priv->priority, instance2->priv->priority); return 1; } // Prefer active instance if ((node1 == NULL) && (node2 == NULL)) { crm_trace("No assignment preference for %s vs. %s: inactive", instance1->id, instance2->id); return 0; } else if (node1 == NULL) { crm_trace("Assign %s after %s: active", instance1->id, instance2->id); return 1; } else if (node2 == NULL) { crm_trace("Assign %s before %s: active", instance1->id, instance2->id); return -1; } // Prefer instance whose current node can run resources can1 = pcmk__node_available(node1, false, false); can2 = pcmk__node_available(node2, false, false); if (can1 && !can2) { crm_trace("Assign %s before %s: current node can run resources", instance1->id, instance2->id); return -1; } else if (!can1 && can2) { crm_trace("Assign %s after %s: current node can't run resources", instance1->id, instance2->id); return 1; } // Prefer instance whose parent is allowed to run on instance's current node node1 = pcmk__top_allowed_node(instance1, node1); node2 = pcmk__top_allowed_node(instance2, node2); if ((node1 == NULL) && (node2 == NULL)) { crm_trace("No assignment preference for %s vs. %s: " "parent not allowed on either instance's current node", instance1->id, instance2->id); return 0; } else if (node1 == NULL) { crm_trace("Assign %s after %s: parent not allowed on current node", instance1->id, instance2->id); return 1; } else if (node2 == NULL) { crm_trace("Assign %s before %s: parent allowed on current node", instance1->id, instance2->id); return -1; } // Prefer instance whose current node is running fewer other instances if (node1->assign->count < node2->assign->count) { crm_trace("Assign %s before %s: fewer active instances on current node", instance1->id, instance2->id); return -1; } else if (node1->assign->count > node2->assign->count) { crm_trace("Assign %s after %s: more active instances on current node", instance1->id, instance2->id); return 1; } // Prefer instance that isn't failed can1 = did_fail(instance1); can2 = did_fail(instance2); if (!can1 && can2) { crm_trace("Assign %s before %s: not failed", instance1->id, instance2->id); return -1; } else if (can1 && !can2) { crm_trace("Assign %s after %s: failed", instance1->id, instance2->id); return 1; } // Prefer instance with higher cumulative colocation score on current node rc = cmp_instance_by_colocation(instance1, instance2); if (rc != 0) { return rc; } // Prefer instance with lower instance number rc = pcmk__cmp_instance_number(instance1, instance2); if (rc < 0) { crm_trace("Assign %s before %s: instance number", instance1->id, instance2->id); } else if (rc > 0) { crm_trace("Assign %s after %s: instance number", instance1->id, instance2->id); } else { crm_trace("No assignment preference for %s vs. %s", instance1->id, instance2->id); } return rc; } /*! * \internal * \brief Increment the parent's instance count after assigning an instance * * An instance's parent tracks how many instances have been assigned to each * node via its pcmk_node_t:count member. After assigning an instance to a node, * find the corresponding node in the parent's allowed table and increment it. * * \param[in,out] instance Instance whose parent to update * \param[in] assigned_to Node to which the instance was assigned */ static void increment_parent_count(pcmk_resource_t *instance, const pcmk_node_t *assigned_to) { pcmk_node_t *allowed = NULL; if (assigned_to == NULL) { return; } allowed = pcmk__top_allowed_node(instance, assigned_to); if (allowed == NULL) { /* The instance is allowed on the node, but its parent isn't. This * shouldn't be possible if the resource is managed, and we won't be * able to limit the number of instances assigned to the node. */ CRM_LOG_ASSERT(!pcmk_is_set(instance->flags, pcmk__rsc_managed)); } else { allowed->assign->count++; } } /*! * \internal * \brief Assign an instance to a node * * \param[in,out] instance Clone instance or bundle replica container * \param[in] prefer If not NULL, attempt early assignment to this * node, if still the best choice; otherwise, * perform final assignment * \param[in] max_per_node Assign at most this many instances to one node * * \return Node to which \p instance is assigned */ static const pcmk_node_t * assign_instance(pcmk_resource_t *instance, const pcmk_node_t *prefer, int max_per_node) { pcmk_node_t *chosen = NULL; pcmk__rsc_trace(instance, "Assigning %s (preferring %s)", instance->id, ((prefer == NULL)? "no node" : prefer->priv->name)); if (pcmk_is_set(instance->flags, pcmk__rsc_assigning)) { pcmk__rsc_debug(instance, "Assignment loop detected involving %s colocations", instance->id); return NULL; } ban_unavailable_allowed_nodes(instance, max_per_node); // Failed early assignments are reversible (stop_if_fail=false) chosen = instance->priv->cmds->assign(instance, prefer, (prefer == NULL)); increment_parent_count(instance, chosen); return chosen; } /*! * \internal * \brief Try to assign an instance to its current node early * * \param[in] rsc Clone or bundle being assigned (for logs only) * \param[in] instance Clone instance or bundle replica container * \param[in] current Instance's current node * \param[in] max_per_node Maximum number of instances per node * \param[in] available Number of instances still available for assignment * * \return \c true if \p instance was successfully assigned to its current node, * or \c false otherwise */ static bool assign_instance_early(const pcmk_resource_t *rsc, pcmk_resource_t *instance, const pcmk_node_t *current, int max_per_node, int available) { const pcmk_node_t *chosen = NULL; int reserved = 0; pcmk_resource_t *parent = instance->priv->parent; GHashTable *allowed_orig = NULL; GHashTable *allowed_orig_parent = parent->priv->allowed_nodes; const pcmk_node_t *allowed_node = NULL; pcmk__rsc_trace(instance, "Trying to assign %s to its current node %s", instance->id, pcmk__node_name(current)); allowed_node = g_hash_table_lookup(instance->priv->allowed_nodes, current->priv->id); if (!pcmk__node_available(allowed_node, true, false)) { pcmk__rsc_info(instance, "Not assigning %s to current node %s: unavailable", instance->id, pcmk__node_name(current)); return false; } /* On each iteration, if instance gets assigned to a node other than its * current one, we reserve one instance for the chosen node, unassign * instance, restore instance's original node tables, and try again. This * way, instances are proportionally assigned to nodes based on preferences, * but shuffling of specific instances is minimized. If a node will be * assigned instances at all, it preferentially receives instances that are * currently active there. * * parent->private->allowed_nodes tracks the number of instances assigned to * each node. If a node already has max_per_node instances assigned, * ban_unavailable_allowed_nodes() marks it as unavailable. * * In the end, we restore the original parent->private->allowed_nodes to * undo the changes to counts during tentative assignments. If we * successfully assigned an instance to its current node, we increment that * node's counter. */ // Back up the allowed node tables of instance and its children recursively pcmk__copy_node_tables(instance, &allowed_orig); // Update instances-per-node counts in a scratch table parent->priv->allowed_nodes = pcmk__copy_node_table(allowed_orig_parent); while (reserved < available) { chosen = assign_instance(instance, current, max_per_node); if (pcmk__same_node(chosen, current)) { // Successfully assigned to current node break; } // Assignment updates scores, so restore to original state pcmk__rsc_debug(instance, "Rolling back node scores for %s", instance->id); pcmk__restore_node_tables(instance, allowed_orig); if (chosen == NULL) { // Assignment failed, so give up pcmk__rsc_info(instance, "Not assigning %s to current node %s: unavailable", instance->id, pcmk__node_name(current)); pcmk__set_rsc_flags(instance, pcmk__rsc_unassigned); break; } // We prefer more strongly to assign an instance to the chosen node pcmk__rsc_debug(instance, "Not assigning %s to current node %s: %s is better", instance->id, pcmk__node_name(current), pcmk__node_name(chosen)); // Reserve one instance for the chosen node and try again if (++reserved >= available) { pcmk__rsc_info(instance, "Not assigning %s to current node %s: " "other assignments are more important", instance->id, pcmk__node_name(current)); } else { pcmk__rsc_debug(instance, "Reserved an instance of %s for %s. Retrying " "assignment of %s to %s", rsc->id, pcmk__node_name(chosen), instance->id, pcmk__node_name(current)); } // Clear this assignment (frees chosen); leave instance counts in parent pcmk__unassign_resource(instance); chosen = NULL; } g_hash_table_destroy(allowed_orig); // Restore original instances-per-node counts g_hash_table_destroy(parent->priv->allowed_nodes); parent->priv->allowed_nodes = allowed_orig_parent; if (chosen == NULL) { // Couldn't assign instance to current node return false; } pcmk__rsc_trace(instance, "Assigned %s to current node %s", instance->id, pcmk__node_name(current)); increment_parent_count(instance, chosen); return true; } /*! * \internal * \brief Reset the node counts of a resource's allowed nodes to zero * * \param[in,out] rsc Resource to reset * * \return Number of nodes that are available to run resources */ static unsigned int reset_allowed_node_counts(pcmk_resource_t *rsc) { unsigned int available_nodes = 0; pcmk_node_t *node = NULL; GHashTableIter iter; g_hash_table_iter_init(&iter, rsc->priv->allowed_nodes); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &node)) { node->assign->count = 0; if (pcmk__node_available(node, false, false)) { available_nodes++; } } return available_nodes; } /*! * \internal * \brief Check whether an instance has a preferred node * * \param[in] instance Clone instance or bundle replica container * \param[in] optimal_per_node Optimal number of instances per node * * \return Instance's current node if still available, otherwise NULL */ static const pcmk_node_t * preferred_node(const pcmk_resource_t *instance, int optimal_per_node) { const pcmk_node_t *node = NULL; const pcmk_node_t *parent_node = NULL; // Check whether instance is active, healthy, and not yet assigned if ((instance->priv->active_nodes == NULL) || !pcmk_is_set(instance->flags, pcmk__rsc_unassigned) || pcmk_is_set(instance->flags, pcmk__rsc_failed)) { return NULL; } // Check whether instance's current node can run resources node = pcmk__current_node(instance); if (!pcmk__node_available(node, true, false)) { pcmk__rsc_trace(instance, "Not assigning %s to %s early (unavailable)", instance->id, pcmk__node_name(node)); return NULL; } // Check whether node already has optimal number of instances assigned parent_node = pcmk__top_allowed_node(instance, node); if ((parent_node != NULL) && (parent_node->assign->count >= optimal_per_node)) { pcmk__rsc_trace(instance, "Not assigning %s to %s early " "(optimal instances already assigned)", instance->id, pcmk__node_name(node)); return NULL; } return node; } /*! * \internal * \brief Assign collective instances to nodes * * \param[in,out] collective Clone or bundle resource being assigned * \param[in,out] instances List of clone instances or bundle containers * \param[in] max_total Maximum instances to assign in total * \param[in] max_per_node Maximum instances to assign to any one node */ void pcmk__assign_instances(pcmk_resource_t *collective, GList *instances, int max_total, int max_per_node) { // Reuse node count to track number of assigned instances unsigned int available_nodes = reset_allowed_node_counts(collective); int optimal_per_node = 0; int assigned = 0; GList *iter = NULL; pcmk_resource_t *instance = NULL; const pcmk_node_t *current = NULL; if (available_nodes > 0) { optimal_per_node = max_total / available_nodes; } if (optimal_per_node < 1) { optimal_per_node = 1; } pcmk__rsc_debug(collective, "Assigning up to %d %s instance%s to up to %u node%s " "(at most %d per host, %d optimal)", max_total, collective->id, pcmk__plural_s(max_total), available_nodes, pcmk__plural_s(available_nodes), max_per_node, optimal_per_node); // Assign as many instances as possible to their current location for (iter = instances; (iter != NULL) && (assigned < max_total); iter = iter->next) { int available = max_total - assigned; instance = iter->data; if (!pcmk_is_set(instance->flags, pcmk__rsc_unassigned)) { continue; // Already assigned } current = preferred_node(instance, optimal_per_node); if ((current != NULL) && assign_instance_early(collective, instance, current, max_per_node, available)) { assigned++; } } pcmk__rsc_trace(collective, "Assigned %d of %d instance%s to current node", assigned, max_total, pcmk__plural_s(max_total)); for (iter = instances; iter != NULL; iter = iter->next) { instance = (pcmk_resource_t *) iter->data; if (!pcmk_is_set(instance->flags, pcmk__rsc_unassigned)) { continue; // Already assigned } if (instance->priv->active_nodes != NULL) { current = pcmk__current_node(instance); if (pcmk__top_allowed_node(instance, current) == NULL) { const char *unmanaged = ""; if (!pcmk_is_set(instance->flags, pcmk__rsc_managed)) { unmanaged = "Unmanaged resource "; } crm_notice("%s%s is running on %s which is no longer allowed", unmanaged, instance->id, pcmk__node_name(current)); } } if (assigned >= max_total) { pcmk__rsc_debug(collective, "Not assigning %s because maximum %d instances " "already assigned", instance->id, max_total); resource_location(instance, NULL, -PCMK_SCORE_INFINITY, "collective_limit_reached", collective->priv->scheduler); } else if (assign_instance(instance, NULL, max_per_node) != NULL) { assigned++; } } pcmk__rsc_debug(collective, "Assigned %d of %d possible instance%s of %s", assigned, max_total, pcmk__plural_s(max_total), collective->id); } enum instance_state { instance_starting = (1 << 0), instance_stopping = (1 << 1), /* This indicates that some instance is restarting. It's not the same as * instance_starting|instance_stopping, which would indicate that some * instance is starting, and some instance (not necessarily the same one) is * stopping. */ instance_restarting = (1 << 2), instance_active = (1 << 3), instance_all = instance_starting|instance_stopping |instance_restarting|instance_active, }; /*! * \internal * \brief Check whether an instance is active, starting, and/or stopping * * \param[in] instance Clone instance or bundle replica container * \param[in,out] state Whether any instance is starting, stopping, etc. */ static void check_instance_state(const pcmk_resource_t *instance, uint32_t *state) { const GList *iter = NULL; uint32_t instance_state = 0; // State of just this instance // No need to check further if all conditions have already been detected if (pcmk_all_flags_set(*state, instance_all)) { return; } // If instance is a collective (a cloned group), check its children instead if (instance->priv->variant > pcmk__rsc_variant_primitive) { for (iter = instance->priv->children; (iter != NULL) && !pcmk_all_flags_set(*state, instance_all); iter = iter->next) { check_instance_state((const pcmk_resource_t *) iter->data, state); } return; } // If we get here, instance is a primitive if (instance->priv->active_nodes != NULL) { instance_state |= instance_active; } // Check each of the instance's actions for runnable start or stop for (iter = instance->priv->actions; (iter != NULL) && !pcmk_all_flags_set(instance_state, instance_starting |instance_stopping); iter = iter->next) { const pcmk_action_t *action = (const pcmk_action_t *) iter->data; const bool optional = pcmk_is_set(action->flags, pcmk__action_optional); if (pcmk__str_eq(PCMK_ACTION_START, action->task, pcmk__str_none)) { if (!optional && pcmk_is_set(action->flags, pcmk__action_runnable)) { pcmk__rsc_trace(instance, "Instance is starting due to %s", action->uuid); instance_state |= instance_starting; } else { pcmk__rsc_trace(instance, "%s doesn't affect %s state (%s)", action->uuid, instance->id, (optional? "optional" : "unrunnable")); } } else if (pcmk__str_eq(PCMK_ACTION_STOP, action->task, pcmk__str_none)) { /* Only stop actions can be pseudo-actions for primitives. That * indicates that the node they are on is being fenced, so the stop * is implied rather than actually executed. */ if (!optional && pcmk_any_flags_set(action->flags, pcmk__action_pseudo |pcmk__action_runnable)) { pcmk__rsc_trace(instance, "Instance is stopping due to %s", action->uuid); instance_state |= instance_stopping; } else { pcmk__rsc_trace(instance, "%s doesn't affect %s state (%s)", action->uuid, instance->id, (optional? "optional" : "unrunnable")); } } } if (pcmk_all_flags_set(instance_state, instance_starting|instance_stopping)) { instance_state |= instance_restarting; } *state |= instance_state; } /*! * \internal * \brief Create actions for collective resource instances * * \param[in,out] collective Clone or bundle resource to create actions for * \param[in,out] instances List of clone instances or bundle containers */ void pcmk__create_instance_actions(pcmk_resource_t *collective, GList *instances) { uint32_t state = 0; pcmk_action_t *stop = NULL; pcmk_action_t *stopped = NULL; pcmk_action_t *start = NULL; pcmk_action_t *started = NULL; pcmk__rsc_trace(collective, "Creating collective instance actions for %s", collective->id); // Create actions for each instance appropriate to its variant for (GList *iter = instances; iter != NULL; iter = iter->next) { pcmk_resource_t *instance = (pcmk_resource_t *) iter->data; instance->priv->cmds->create_actions(instance); check_instance_state(instance, &state); } // Create pseudo-actions for rsc start and started start = pe__new_rsc_pseudo_action(collective, PCMK_ACTION_START, !pcmk_is_set(state, instance_starting), true); started = pe__new_rsc_pseudo_action(collective, PCMK_ACTION_RUNNING, !pcmk_is_set(state, instance_starting), false); started->priority = PCMK_SCORE_INFINITY; if (pcmk_any_flags_set(state, instance_active|instance_starting)) { pcmk__set_action_flags(started, pcmk__action_runnable); } // Create pseudo-actions for rsc stop and stopped stop = pe__new_rsc_pseudo_action(collective, PCMK_ACTION_STOP, !pcmk_is_set(state, instance_stopping), true); stopped = pe__new_rsc_pseudo_action(collective, PCMK_ACTION_STOPPED, !pcmk_is_set(state, instance_stopping), true); stopped->priority = PCMK_SCORE_INFINITY; if (!pcmk_is_set(state, instance_restarting)) { pcmk__set_action_flags(stop, pcmk__action_migratable); } if (pcmk__is_clone(collective)) { pe__create_clone_notif_pseudo_ops(collective, start, started, stop, stopped); } } /*! * \internal * \brief Get a list of clone instances or bundle replica containers * * \param[in] rsc Clone or bundle resource * * \return Clone instances if \p rsc is a clone, or a newly created list of * \p rsc's replica containers if \p rsc is a bundle * \note The caller must call free_instance_list() on the result when the list * is no longer needed. */ static inline GList * get_instance_list(const pcmk_resource_t *rsc) { if (pcmk__is_bundle(rsc)) { return pe__bundle_containers(rsc); } else { return rsc->priv->children; } } /*! * \internal * \brief Free any memory created by get_instance_list() * * \param[in] rsc Clone or bundle resource passed to get_instance_list() * \param[in,out] list Return value of get_instance_list() for \p rsc */ static inline void free_instance_list(const pcmk_resource_t *rsc, GList *list) { if (list != rsc->priv->children) { g_list_free(list); } } /*! * \internal * \brief Check whether an instance is compatible with a role and node * * \param[in] instance Clone instance or bundle replica container * \param[in] node Instance must match this node * \param[in] role If not pcmk_role_unknown, instance must match this role * \param[in] current If true, compare instance's original node and role, * otherwise compare assigned next node and role * * \return true if \p instance is compatible with \p node and \p role, * otherwise false */ bool pcmk__instance_matches(const pcmk_resource_t *instance, const pcmk_node_t *node, enum rsc_role_e role, bool current) { pcmk_node_t *instance_node = NULL; CRM_CHECK((instance != NULL) && (node != NULL), return false); if ((role != pcmk_role_unknown) && (role != instance->priv->fns->state(instance, current))) { pcmk__rsc_trace(instance, "%s is not a compatible instance (role is not %s)", instance->id, pcmk_role_text(role)); return false; } if (!is_set_recursive(instance, pcmk__rsc_blocked, true)) { uint32_t target = pcmk__rsc_node_assigned; if (current) { target = pcmk__rsc_node_current; } // We only want instances that haven't failed instance_node = instance->priv->fns->location(instance, NULL, target); } if (instance_node == NULL) { pcmk__rsc_trace(instance, "%s is not a compatible instance " "(not assigned to a node)", instance->id); return false; } if (!pcmk__same_node(instance_node, node)) { pcmk__rsc_trace(instance, "%s is not a compatible instance " "(assigned to %s not %s)", instance->id, pcmk__node_name(instance_node), pcmk__node_name(node)); return false; } return true; } #define display_role(r) \ (((r) == pcmk_role_unknown)? "matching" : pcmk_role_text(r)) /*! * \internal * \brief Find an instance that matches a given resource by node and role * * \param[in] match_rsc Resource that instance must match (for logging only) * \param[in] rsc Clone or bundle resource to check for matching instance * \param[in] node Instance must match this node * \param[in] role If not pcmk_role_unknown, instance must match this role * \param[in] current If true, compare instance's original node and role, * otherwise compare assigned next node and role * * \return \p rsc instance matching \p node and \p role if any, otherwise NULL */ static pcmk_resource_t * find_compatible_instance_on_node(const pcmk_resource_t *match_rsc, const pcmk_resource_t *rsc, const pcmk_node_t *node, enum rsc_role_e role, bool current) { GList *instances = NULL; instances = get_instance_list(rsc); for (GList *iter = instances; iter != NULL; iter = iter->next) { pcmk_resource_t *instance = (pcmk_resource_t *) iter->data; if (pcmk__instance_matches(instance, node, role, current)) { pcmk__rsc_trace(match_rsc, "Found %s %s instance %s compatible with %s on %s", display_role(role), rsc->id, instance->id, match_rsc->id, pcmk__node_name(node)); free_instance_list(rsc, instances); // Only frees list, not contents return instance; } } free_instance_list(rsc, instances); pcmk__rsc_trace(match_rsc, "No %s %s instance found compatible with %s on %s", display_role(role), rsc->id, match_rsc->id, pcmk__node_name(node)); return NULL; } /*! * \internal * \brief Find a clone instance or bundle container compatible with a resource * * \param[in] match_rsc Resource that instance must match * \param[in] rsc Clone or bundle resource to check for matching instance * \param[in] role If not pcmk_role_unknown, instance must match this role * \param[in] current If true, compare instance's original node and role, * otherwise compare assigned next node and role * * \return Compatible (by \p role and \p match_rsc location) instance of \p rsc * if any, otherwise NULL */ pcmk_resource_t * pcmk__find_compatible_instance(const pcmk_resource_t *match_rsc, const pcmk_resource_t *rsc, enum rsc_role_e role, bool current) { pcmk_resource_t *instance = NULL; GList *nodes = NULL; const pcmk_node_t *node = NULL; GHashTable *allowed_nodes = match_rsc->priv->allowed_nodes; uint32_t target = pcmk__rsc_node_assigned; if (current) { target = pcmk__rsc_node_current; } // If match_rsc has a node, check only that node node = match_rsc->priv->fns->location(match_rsc, NULL, target); if (node != NULL) { return find_compatible_instance_on_node(match_rsc, rsc, node, role, current); } // Otherwise check for an instance matching any of match_rsc's allowed nodes nodes = pcmk__sort_nodes(g_hash_table_get_values(allowed_nodes), NULL); for (GList *iter = nodes; (iter != NULL) && (instance == NULL); iter = iter->next) { instance = find_compatible_instance_on_node(match_rsc, rsc, (pcmk_node_t *) iter->data, role, current); } if (instance == NULL) { pcmk__rsc_debug(rsc, "No %s instance found compatible with %s", rsc->id, match_rsc->id); } g_list_free(nodes); return instance; } /*! * \internal * \brief Unassign an instance if mandatory ordering has no interleave match * * \param[in] first 'First' action in an ordering * \param[in] then 'Then' action in an ordering * \param[in,out] then_instance 'Then' instance that has no interleave match * \param[in] type Group of enum pcmk__action_relation_flags * \param[in] current If true, "then" action is stopped or demoted * * \return true if \p then_instance was unassigned, otherwise false */ static bool unassign_if_mandatory(const pcmk_action_t *first, const pcmk_action_t *then, pcmk_resource_t *then_instance, uint32_t type, bool current) { // Allow "then" instance to go down even without an interleave match if (current) { pcmk__rsc_trace(then->rsc, "%s has no instance to order before stopping " "or demoting %s", first->rsc->id, then_instance->id); /* If the "first" action must be runnable, but there is no "first" * instance, the "then" instance must not be allowed to come up. */ } else if (pcmk_any_flags_set(type, pcmk__ar_unrunnable_first_blocks |pcmk__ar_first_implies_then)) { pcmk__rsc_info(then->rsc, "Inhibiting %s from being active " "because there is no %s instance to interleave", then_instance->id, first->rsc->id); return pcmk__assign_resource(then_instance, NULL, true, true); } return false; } /*! * \internal * \brief Find first matching action for a clone instance or bundle container * * \param[in] action Action in an interleaved ordering * \param[in] instance Clone instance or bundle container being interleaved * \param[in] action_name Action to look for * \param[in] node If not NULL, require action to be on this node * \param[in] for_first If true, \p instance is the 'first' resource in the * ordering, otherwise it is the 'then' resource * * \return First action for \p instance (or in some cases if \p instance is a * bundle container, its containerized resource) that matches * \p action_name and \p node if any, otherwise NULL */ static pcmk_action_t * find_instance_action(const pcmk_action_t *action, const pcmk_resource_t *instance, const char *action_name, const pcmk_node_t *node, bool for_first) { const pcmk_resource_t *rsc = NULL; pcmk_action_t *matching_action = NULL; /* If instance is a bundle container, sometimes we should interleave the * action for the container itself, and sometimes for the containerized * resource. * * For example, given "start bundle A then bundle B", B likely requires the * service inside A's container to be active, rather than just the * container, so we should interleave the action for A's containerized * resource. On the other hand, it's possible B's container itself requires * something from A, so we should interleave the action for B's container. * * Essentially, for 'first', we should use the containerized resource for * everything except stop, and for 'then', we should use the container for * everything except promote and demote (which can only be performed on the * containerized resource). */ if ((for_first && !pcmk__str_any_of(action->task, PCMK_ACTION_STOP, PCMK_ACTION_STOPPED, NULL)) || (!for_first && pcmk__str_any_of(action->task, PCMK_ACTION_PROMOTE, PCMK_ACTION_PROMOTED, PCMK_ACTION_DEMOTE, PCMK_ACTION_DEMOTED, NULL))) { rsc = pe__get_rsc_in_container(instance); } if (rsc == NULL) { rsc = instance; // No containerized resource, use instance itself } else { node = NULL; // Containerized actions are on bundle-created guest } matching_action = find_first_action(rsc->priv->actions, NULL, action_name, node); if (matching_action != NULL) { return matching_action; } if (pcmk_is_set(instance->flags, pcmk__rsc_removed) || pcmk__str_any_of(action_name, PCMK_ACTION_STOP, PCMK_ACTION_DEMOTE, NULL)) { crm_trace("No %s action found for %s%s", action_name, pcmk_is_set(instance->flags, pcmk__rsc_removed)? "orphan " : "", instance->id); } else { crm_err("No %s action found for %s to interleave (bug?)", action_name, instance->id); } return NULL; } /*! * \internal * \brief Get the original action name of a bundle or clone action * * Given an action for a bundle or clone, get the original action name, * mapping notify to the action being notified, and if the instances are * primitives, mapping completion actions to the action that was completed * (for example, stopped to stop). * * \param[in] action Clone or bundle action to check * * \return Original action name for \p action */ static const char * orig_action_name(const pcmk_action_t *action) { // Any instance will do const pcmk_resource_t *instance = action->rsc->priv->children->data; char *action_type = NULL; const char *action_name = action->task; enum pcmk__action_type orig_task = pcmk__action_unspecified; if (pcmk__strcase_any_of(action->task, PCMK_ACTION_NOTIFY, PCMK_ACTION_NOTIFIED, NULL)) { // action->uuid is RSC_(confirmed-){pre,post}_notify_ACTION_INTERVAL CRM_CHECK(parse_op_key(action->uuid, NULL, &action_type, NULL), return pcmk__action_text(pcmk__action_unspecified)); action_name = strstr(action_type, "_notify_"); CRM_CHECK(action_name != NULL, return pcmk__action_text(pcmk__action_unspecified)); action_name += strlen("_notify_"); } orig_task = get_complex_task(instance, action_name); free(action_type); return pcmk__action_text(orig_task); } /*! * \internal * \brief Update two interleaved actions according to an ordering between them * * Given information about an ordering of two interleaved actions, update the * actions' flags (and runnable_before members if appropriate) as appropriate * for the ordering. Effects may cascade to other orderings involving the * actions as well. * * \param[in,out] first 'First' action in an ordering * \param[in,out] then 'Then' action in an ordering * \param[in] node If not NULL, limit scope of ordering to this node * \param[in] filter Action flags to limit scope of certain updates (may * include pcmk__action_optional to affect only * mandatory actions, and pcmk__action_runnable to * affect only runnable actions) * \param[in] type Group of enum pcmk__action_relation_flags to apply * * \return Group of enum pcmk__updated flags indicating what was updated */ static uint32_t update_interleaved_actions(pcmk_action_t *first, pcmk_action_t *then, const pcmk_node_t *node, uint32_t filter, uint32_t type) { GList *instances = NULL; uint32_t changed = pcmk__updated_none; const char *orig_first_task = orig_action_name(first); // Stops and demotes must be interleaved with instance on current node bool current = pcmk__ends_with(first->uuid, "_" PCMK_ACTION_STOPPED "_0") || pcmk__ends_with(first->uuid, "_" PCMK_ACTION_DEMOTED "_0"); // Update the specified actions for each "then" instance individually instances = get_instance_list(then->rsc); for (GList *iter = instances; iter != NULL; iter = iter->next) { pcmk_resource_t *first_instance = NULL; pcmk_resource_t *then_instance = iter->data; pcmk_action_t *first_action = NULL; pcmk_action_t *then_action = NULL; // Find a "first" instance to interleave with this "then" instance first_instance = pcmk__find_compatible_instance(then_instance, first->rsc, pcmk_role_unknown, current); if (first_instance == NULL) { // No instance can be interleaved if (unassign_if_mandatory(first, then, then_instance, type, current)) { pcmk__set_updated_flags(changed, first, pcmk__updated_then); } continue; } first_action = find_instance_action(first, first_instance, orig_first_task, node, true); if (first_action == NULL) { continue; } then_action = find_instance_action(then, then_instance, then->task, node, false); if (then_action == NULL) { continue; } if (order_actions(first_action, then_action, type)) { pcmk__set_updated_flags(changed, first, pcmk__updated_first|pcmk__updated_then); } changed |= then_instance->priv->cmds->update_ordered_actions( first_action, then_action, node, first_instance->priv->cmds->action_flags(first_action, node), filter, type, then->rsc->priv->scheduler); } free_instance_list(then->rsc, instances); return changed; } /*! * \internal * \brief Check whether two actions in an ordering can be interleaved * * \param[in] first 'First' action in the ordering * \param[in] then 'Then' action in the ordering * * \return true if \p first and \p then can be interleaved, otherwise false */ static bool can_interleave_actions(const pcmk_action_t *first, const pcmk_action_t *then) { bool interleave = false; pcmk_resource_t *rsc = NULL; if ((first->rsc == NULL) || (then->rsc == NULL)) { crm_trace("Not interleaving %s with %s: not resource actions", first->uuid, then->uuid); return false; } if (first->rsc == then->rsc) { crm_trace("Not interleaving %s with %s: same resource", first->uuid, then->uuid); return false; } if ((first->rsc->priv->variant < pcmk__rsc_variant_clone) || (then->rsc->priv->variant < pcmk__rsc_variant_clone)) { crm_trace("Not interleaving %s with %s: not clones or bundles", first->uuid, then->uuid); return false; } if (pcmk__ends_with(then->uuid, "_stop_0") || pcmk__ends_with(then->uuid, "_demote_0")) { rsc = first->rsc; } else { rsc = then->rsc; } interleave = crm_is_true(g_hash_table_lookup(rsc->priv->meta, PCMK_META_INTERLEAVE)); pcmk__rsc_trace(rsc, "'%s then %s' will %sbe interleaved (based on %s)", first->uuid, then->uuid, (interleave? "" : "not "), rsc->id); return interleave; } /*! * \internal * \brief Update non-interleaved instance actions according to an ordering * * Given information about an ordering of two non-interleaved actions, update * the actions' flags (and runnable_before members if appropriate) as * appropriate for the ordering. Effects may cascade to other orderings * involving the actions as well. * * \param[in,out] instance Clone instance or bundle container * \param[in,out] first "First" action in ordering * \param[in] then "Then" action in ordering (for \p instance's parent) * \param[in] node If not NULL, limit scope of ordering to this node * \param[in] flags Action flags for \p first for ordering purposes * \param[in] filter Action flags to limit scope of certain updates (may * include pcmk__action_optional to affect only * mandatory actions, and pcmk__action_runnable to * affect only runnable actions) * \param[in] type Group of enum pcmk__action_relation_flags to apply * * \return Group of enum pcmk__updated flags indicating what was updated */ static uint32_t update_noninterleaved_actions(pcmk_resource_t *instance, pcmk_action_t *first, const pcmk_action_t *then, const pcmk_node_t *node, uint32_t flags, uint32_t filter, uint32_t type) { pcmk_action_t *instance_action = NULL; pcmk_scheduler_t *scheduler = instance->priv->scheduler; uint32_t instance_flags = 0; uint32_t changed = pcmk__updated_none; // Check whether instance has an equivalent of "then" action instance_action = find_first_action(instance->priv->actions, NULL, then->task, node); if (instance_action == NULL) { return changed; } // Check whether action is runnable instance_flags = instance->priv->cmds->action_flags(instance_action, node); if (!pcmk_is_set(instance_flags, pcmk__action_runnable)) { return changed; } // If so, update actions for the instance changed = instance->priv->cmds->update_ordered_actions(first, instance_action, node, flags, filter, type, scheduler); // Propagate any changes to later actions if (pcmk_is_set(changed, pcmk__updated_then)) { for (GList *after_iter = instance_action->actions_after; after_iter != NULL; after_iter = after_iter->next) { pcmk__related_action_t *after = after_iter->data; pcmk__update_action_for_orderings(after->action, scheduler); } } return changed; } /*! * \internal * \brief Update two actions according to an ordering between them * * Given information about an ordering of two clone or bundle actions, update * the actions' flags (and runnable_before members if appropriate) as * appropriate for the ordering. Effects may cascade to other orderings * involving the actions as well. * * \param[in,out] first 'First' action in an ordering * \param[in,out] then 'Then' action in an ordering * \param[in] node If not NULL, limit scope of ordering to this node * (only used when interleaving instances) * \param[in] flags Action flags for \p first for ordering purposes * \param[in] filter Action flags to limit scope of certain updates (may * include pcmk__action_optional to affect only * mandatory actions, and pcmk__action_runnable to * affect only runnable actions) * \param[in] type Group of enum pcmk__action_relation_flags to apply * \param[in,out] scheduler Scheduler data * * \return Group of enum pcmk__updated flags indicating what was updated */ uint32_t pcmk__instance_update_ordered_actions(pcmk_action_t *first, pcmk_action_t *then, const pcmk_node_t *node, uint32_t flags, uint32_t filter, uint32_t type, pcmk_scheduler_t *scheduler) { pcmk__assert((first != NULL) && (then != NULL) && (scheduler != NULL)); if (then->rsc == NULL) { return pcmk__updated_none; } else if (can_interleave_actions(first, then)) { return update_interleaved_actions(first, then, node, filter, type); } else { uint32_t changed = pcmk__updated_none; GList *instances = get_instance_list(then->rsc); // Update actions for the clone or bundle resource itself changed |= pcmk__update_ordered_actions(first, then, node, flags, filter, type, scheduler); // Update the 'then' clone instances or bundle containers individually for (GList *iter = instances; iter != NULL; iter = iter->next) { pcmk_resource_t *instance = iter->data; changed |= update_noninterleaved_actions(instance, first, then, node, flags, filter, type); } free_instance_list(then->rsc, instances); return changed; } } #define pe__clear_action_summary_flags(flags, action, flag) do { \ flags = pcmk__clear_flags_as(__func__, __LINE__, LOG_TRACE, \ "Action summary", action->rsc->id, \ flags, flag, #flag); \ } while (0) /*! * \internal * \brief Return action flags for a given clone or bundle action * * \param[in,out] action Action for a clone or bundle * \param[in] instances Clone instances or bundle containers * \param[in] node If not NULL, limit effects to this node * * \return Flags appropriate to \p action on \p node */ uint32_t pcmk__collective_action_flags(pcmk_action_t *action, const GList *instances, const pcmk_node_t *node) { bool any_runnable = false; const char *action_name = orig_action_name(action); // Set original assumptions (optional and runnable may be cleared below) uint32_t flags = pcmk__action_optional |pcmk__action_runnable |pcmk__action_pseudo; for (const GList *iter = instances; iter != NULL; iter = iter->next) { const pcmk_resource_t *instance = iter->data; const pcmk_node_t *instance_node = NULL; pcmk_action_t *instance_action = NULL; uint32_t instance_flags; // Node is relevant only to primitive instances if (pcmk__is_primitive(instance)) { instance_node = node; } instance_action = find_first_action(instance->priv->actions, NULL, action_name, instance_node); if (instance_action == NULL) { pcmk__rsc_trace(action->rsc, "%s has no %s action on %s", instance->id, action_name, pcmk__node_name(node)); continue; } pcmk__rsc_trace(action->rsc, "%s has %s for %s on %s", instance->id, instance_action->uuid, action_name, pcmk__node_name(node)); instance_flags = instance->priv->cmds->action_flags(instance_action, node); // If any instance action is mandatory, so is the collective action if (pcmk_is_set(flags, pcmk__action_optional) && !pcmk_is_set(instance_flags, pcmk__action_optional)) { pcmk__rsc_trace(instance, "%s is mandatory because %s is", action->uuid, instance_action->uuid); pe__clear_action_summary_flags(flags, action, pcmk__action_optional); pcmk__clear_action_flags(action, pcmk__action_optional); } // If any instance action is runnable, so is the collective action if (pcmk_is_set(instance_flags, pcmk__action_runnable)) { any_runnable = true; } } if (!any_runnable) { pcmk__rsc_trace(action->rsc, "%s is not runnable because no instance can run %s", action->uuid, action_name); pe__clear_action_summary_flags(flags, action, pcmk__action_runnable); if (node == NULL) { pcmk__clear_action_flags(action, pcmk__action_runnable); } } return flags; } diff --git a/lib/pacemaker/pcmk_sched_nodes.c b/lib/pacemaker/pcmk_sched_nodes.c index 503215c22b..544a8ced6b 100644 --- a/lib/pacemaker/pcmk_sched_nodes.c +++ b/lib/pacemaker/pcmk_sched_nodes.c @@ -1,445 +1,446 @@ /* * Copyright 2004-2024 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 "libpacemaker_private.h" /*! * \internal * \brief Check whether a node is available to run resources * * \param[in] node Node to check * \param[in] consider_score If true, consider a negative score unavailable * \param[in] consider_guest If true, consider a guest node unavailable whose * resource will not be active * * \return true if node is online and not shutting down, unclean, or in standby * or maintenance mode, otherwise false */ bool pcmk__node_available(const pcmk_node_t *node, bool consider_score, bool consider_guest) { if ((node == NULL) || (node->details == NULL) || !node->details->online || node->details->shutdown || node->details->unclean || pcmk_is_set(node->priv->flags, pcmk__node_standby) || node->details->maintenance) { return false; } if (consider_score && (node->assign->score < 0)) { return false; } // @TODO Go through all callers to see which should set consider_guest if (consider_guest && pcmk__is_guest_or_bundle_node(node)) { pcmk_resource_t *guest = node->priv->remote->priv->launcher; if (guest->priv->fns->location(guest, NULL, pcmk__rsc_node_assigned) == NULL) { return false; } } return true; } /*! * \internal - * \brief Copy a hash table of node objects + * \brief Create a hash table with copies of another table's nodes * * \param[in] nodes Hash table to copy * - * \return New copy of nodes (or NULL if nodes is NULL) + * \return New table with copies of nodes in \p nodes, or \c NULL if \p nodes is + * \c NULL */ GHashTable * pcmk__copy_node_table(GHashTable *nodes) { GHashTable *new_table = NULL; GHashTableIter iter; pcmk_node_t *node = NULL; if (nodes == NULL) { return NULL; } - new_table = pcmk__strkey_table(NULL, free); + new_table = pcmk__strkey_table(NULL, pcmk__free_node_copy); g_hash_table_iter_init(&iter, nodes); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &node)) { pcmk_node_t *new_node = pe__copy_node(node); g_hash_table_insert(new_table, (gpointer) new_node->priv->id, new_node); } return new_table; } /*! * \internal * \brief Free a table of node tables * * \param[in,out] data Table to free * * \note This is a \c GDestroyNotify wrapper for \c g_hash_table_destroy(). */ static void destroy_node_tables(gpointer data) { g_hash_table_destroy((GHashTable *) data); } /*! * \internal * \brief Recursively copy the node tables of a resource * * Build a hash table containing copies of the allowed nodes tables of \p rsc * and its entire tree of descendants. The key is the resource ID, and the value * is a copy of the resource's node table. * * \param[in] rsc Resource whose node table to copy * \param[in,out] copy Where to store the copied node tables * * \note \p *copy should be \c NULL for the top-level call. * \note The caller is responsible for freeing \p copy using * \c g_hash_table_destroy(). */ void pcmk__copy_node_tables(const pcmk_resource_t *rsc, GHashTable **copy) { pcmk__assert((rsc != NULL) && (copy != NULL)); if (*copy == NULL) { *copy = pcmk__strkey_table(NULL, destroy_node_tables); } g_hash_table_insert(*copy, rsc->id, pcmk__copy_node_table(rsc->priv->allowed_nodes)); for (const GList *iter = rsc->priv->children; iter != NULL; iter = iter->next) { pcmk__copy_node_tables((const pcmk_resource_t *) iter->data, copy); } } /*! * \internal * \brief Recursively restore the node tables of a resource from backup * * Given a hash table containing backup copies of the allowed nodes tables of * \p rsc and its entire tree of descendants, replace the resources' current * node tables with the backed-up copies. * * \param[in,out] rsc Resource whose node tables to restore * \param[in] backup Table of backup node tables (created by * \c pcmk__copy_node_tables()) * * \note This function frees the resources' current node tables. */ void pcmk__restore_node_tables(pcmk_resource_t *rsc, GHashTable *backup) { pcmk__assert((rsc != NULL) && (backup != NULL)); g_hash_table_destroy(rsc->priv->allowed_nodes); // Copy to avoid danger with multiple restores rsc->priv->allowed_nodes = pcmk__copy_node_table(g_hash_table_lookup(backup, rsc->id)); for (GList *iter = rsc->priv->children; iter != NULL; iter = iter->next) { pcmk__restore_node_tables((pcmk_resource_t *) iter->data, backup); } } /*! * \internal * \brief Copy a list of node objects * * \param[in] list List to copy * \param[in] reset Set copies' scores to 0 * * \return New list of shallow copies of nodes in original list */ GList * pcmk__copy_node_list(const GList *list, bool reset) { GList *result = NULL; for (const GList *iter = list; iter != NULL; iter = iter->next) { pcmk_node_t *new_node = NULL; pcmk_node_t *this_node = iter->data; new_node = pe__copy_node(this_node); if (reset) { new_node->assign->score = 0; } result = g_list_prepend(result, new_node); } return result; } /*! * \internal * \brief Compare two nodes for assignment preference * * Given two nodes, check which one is more preferred by assignment criteria * such as node score and utilization. * * \param[in] a First node to compare * \param[in] b Second node to compare * \param[in] data Node to prefer if all else equal * * \return -1 if \p a is preferred, +1 if \p b is preferred, or 0 if they are * equally preferred */ static gint compare_nodes(gconstpointer a, gconstpointer b, gpointer data) { const pcmk_node_t *node1 = (const pcmk_node_t *) a; const pcmk_node_t *node2 = (const pcmk_node_t *) b; const pcmk_node_t *preferred = (const pcmk_node_t *) data; int node1_score = -PCMK_SCORE_INFINITY; int node2_score = -PCMK_SCORE_INFINITY; int result = 0; if (a == NULL) { return 1; } if (b == NULL) { return -1; } // Compare node scores if (pcmk__node_available(node1, false, false)) { node1_score = node1->assign->score; } if (pcmk__node_available(node2, false, false)) { node2_score = node2->assign->score; } if (node1_score > node2_score) { crm_trace("%s before %s (score %d > %d)", pcmk__node_name(node1), pcmk__node_name(node2), node1_score, node2_score); return -1; } if (node1_score < node2_score) { crm_trace("%s after %s (score %d < %d)", pcmk__node_name(node1), pcmk__node_name(node2), node1_score, node2_score); return 1; } // If appropriate, compare node utilization if (pcmk__str_eq(node1->priv->scheduler->priv->placement_strategy, PCMK_VALUE_MINIMAL, pcmk__str_casei)) { goto equal; } if (pcmk__str_eq(node1->priv->scheduler->priv->placement_strategy, PCMK_VALUE_BALANCED, pcmk__str_casei)) { result = pcmk__compare_node_capacities(node1, node2); if (result < 0) { crm_trace("%s before %s (greater capacity by %d attributes)", pcmk__node_name(node1), pcmk__node_name(node2), result * -1); return -1; } else if (result > 0) { crm_trace("%s after %s (lower capacity by %d attributes)", pcmk__node_name(node1), pcmk__node_name(node2), result); return 1; } } // Compare number of resources already assigned to node if (node1->priv->num_resources < node2->priv->num_resources) { crm_trace("%s before %s (%d resources < %d)", pcmk__node_name(node1), pcmk__node_name(node2), node1->priv->num_resources, node2->priv->num_resources); return -1; } else if (node1->priv->num_resources > node2->priv->num_resources) { crm_trace("%s after %s (%d resources > %d)", pcmk__node_name(node1), pcmk__node_name(node2), node1->priv->num_resources, node2->priv->num_resources); return 1; } // Check whether one node is already running desired resource if (preferred != NULL) { if (pcmk__same_node(preferred, node1)) { crm_trace("%s before %s (preferred node)", pcmk__node_name(node1), pcmk__node_name(node2)); return -1; } else if (pcmk__same_node(preferred, node2)) { crm_trace("%s after %s (not preferred node)", pcmk__node_name(node1), pcmk__node_name(node2)); return 1; } } // If all else is equal, prefer node with lowest-sorting name equal: result = strcmp(node1->priv->name, node2->priv->name); if (result < 0) { crm_trace("%s before %s (name)", pcmk__node_name(node1), pcmk__node_name(node2)); return -1; } else if (result > 0) { crm_trace("%s after %s (name)", pcmk__node_name(node1), pcmk__node_name(node2)); return 1; } crm_trace("%s == %s", pcmk__node_name(node1), pcmk__node_name(node2)); return 0; } /*! * \internal * \brief Sort a list of nodes by assigment preference * * \param[in,out] nodes Node list to sort * \param[in] active_node Node where resource being assigned is active * * \return New head of sorted list */ GList * pcmk__sort_nodes(GList *nodes, pcmk_node_t *active_node) { return g_list_sort_with_data(nodes, compare_nodes, active_node); } /*! * \internal * \brief Check whether any node is available to run resources * * \param[in] nodes Nodes to check * * \return true if any node in \p nodes is available to run resources, * otherwise false */ bool pcmk__any_node_available(GHashTable *nodes) { GHashTableIter iter; const pcmk_node_t *node = NULL; if (nodes == NULL) { return false; } g_hash_table_iter_init(&iter, nodes); while (g_hash_table_iter_next(&iter, NULL, (void **) &node)) { if (pcmk__node_available(node, true, false)) { return true; } } return false; } /*! * \internal * \brief Apply node health values for all nodes in cluster * * \param[in,out] scheduler Scheduler data */ void pcmk__apply_node_health(pcmk_scheduler_t *scheduler) { int base_health = 0; enum pcmk__health_strategy strategy; const char *strategy_str = pcmk__cluster_option(scheduler->priv->options, PCMK_OPT_NODE_HEALTH_STRATEGY); strategy = pcmk__parse_health_strategy(strategy_str); if (strategy == pcmk__health_strategy_none) { return; } crm_info("Applying node health strategy '%s'", strategy_str); // The progressive strategy can use a base health score if (strategy == pcmk__health_strategy_progressive) { base_health = pcmk__health_score(PCMK_OPT_NODE_HEALTH_BASE, scheduler); } for (GList *iter = scheduler->nodes; iter != NULL; iter = iter->next) { pcmk_node_t *node = (pcmk_node_t *) iter->data; int health = pe__sum_node_health_scores(node, base_health); // An overall health score of 0 has no effect if (health == 0) { continue; } crm_info("Overall system health of %s is %d", pcmk__node_name(node), health); // Use node health as a location score for each resource on the node for (GList *r = scheduler->priv->resources; r != NULL; r = r->next) { pcmk_resource_t *rsc = (pcmk_resource_t *) r->data; bool constrain = true; if (health < 0) { /* Negative health scores do not apply to resources with * PCMK_META_ALLOW_UNHEALTHY_NODES=true. */ constrain = !crm_is_true(g_hash_table_lookup(rsc->priv->meta, PCMK_META_ALLOW_UNHEALTHY_NODES)); } if (constrain) { pcmk__new_location(strategy_str, rsc, health, NULL, node); } else { pcmk__rsc_trace(rsc, "%s is immune from health ban on %s", rsc->id, pcmk__node_name(node)); } } } } /*! * \internal * \brief Check for a node in a resource's parent's allowed nodes * * \param[in] rsc Resource whose parent should be checked * \param[in] node Node to check for * * \return Equivalent of \p node from \p rsc's parent's allowed nodes if any, * otherwise NULL */ pcmk_node_t * pcmk__top_allowed_node(const pcmk_resource_t *rsc, const pcmk_node_t *node) { GHashTable *allowed_nodes = NULL; if ((rsc == NULL) || (node == NULL)) { return NULL; } if (rsc->priv->parent == NULL) { allowed_nodes = rsc->priv->allowed_nodes; } else { allowed_nodes = rsc->priv->parent->priv->allowed_nodes; } return g_hash_table_lookup(allowed_nodes, node->priv->id); } diff --git a/lib/pacemaker/pcmk_sched_resource.c b/lib/pacemaker/pcmk_sched_resource.c index 17487f5095..2bc843f05f 100644 --- a/lib/pacemaker/pcmk_sched_resource.c +++ b/lib/pacemaker/pcmk_sched_resource.c @@ -1,800 +1,800 @@ /* * Copyright 2014-2024 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 "libpacemaker_private.h" // Resource assignment methods by resource variant static pcmk__assignment_methods_t assignment_methods[] = { { pcmk__primitive_assign, pcmk__primitive_create_actions, pcmk__probe_rsc_on_node, pcmk__primitive_internal_constraints, pcmk__primitive_apply_coloc_score, pcmk__colocated_resources, pcmk__with_primitive_colocations, pcmk__primitive_with_colocations, pcmk__add_colocated_node_scores, pcmk__apply_location, pcmk__primitive_action_flags, pcmk__update_ordered_actions, pcmk__output_resource_actions, pcmk__add_rsc_actions_to_graph, pcmk__primitive_add_graph_meta, pcmk__primitive_add_utilization, pcmk__primitive_shutdown_lock, }, { pcmk__group_assign, pcmk__group_create_actions, pcmk__probe_rsc_on_node, pcmk__group_internal_constraints, pcmk__group_apply_coloc_score, pcmk__group_colocated_resources, pcmk__with_group_colocations, pcmk__group_with_colocations, pcmk__group_add_colocated_node_scores, pcmk__group_apply_location, pcmk__group_action_flags, pcmk__group_update_ordered_actions, pcmk__output_resource_actions, pcmk__add_rsc_actions_to_graph, pcmk__noop_add_graph_meta, pcmk__group_add_utilization, pcmk__group_shutdown_lock, }, { pcmk__clone_assign, pcmk__clone_create_actions, pcmk__clone_create_probe, pcmk__clone_internal_constraints, pcmk__clone_apply_coloc_score, pcmk__colocated_resources, pcmk__with_clone_colocations, pcmk__clone_with_colocations, pcmk__add_colocated_node_scores, pcmk__clone_apply_location, pcmk__clone_action_flags, pcmk__instance_update_ordered_actions, pcmk__output_resource_actions, pcmk__clone_add_actions_to_graph, pcmk__clone_add_graph_meta, pcmk__clone_add_utilization, pcmk__clone_shutdown_lock, }, { pcmk__bundle_assign, pcmk__bundle_create_actions, pcmk__bundle_create_probe, pcmk__bundle_internal_constraints, pcmk__bundle_apply_coloc_score, pcmk__colocated_resources, pcmk__with_bundle_colocations, pcmk__bundle_with_colocations, pcmk__add_colocated_node_scores, pcmk__bundle_apply_location, pcmk__bundle_action_flags, pcmk__instance_update_ordered_actions, pcmk__output_bundle_actions, pcmk__bundle_add_actions_to_graph, pcmk__noop_add_graph_meta, pcmk__bundle_add_utilization, pcmk__bundle_shutdown_lock, } }; /*! * \internal * \brief Check whether a resource's agent standard, provider, or type changed * * \param[in,out] rsc Resource to check * \param[in,out] node Node needing unfencing if agent changed * \param[in] rsc_entry XML with previously known agent information * \param[in] active_on_node Whether \p rsc is active on \p node * * \return true if agent for \p rsc changed, otherwise false */ bool pcmk__rsc_agent_changed(pcmk_resource_t *rsc, pcmk_node_t *node, const xmlNode *rsc_entry, bool active_on_node) { bool changed = false; const char *attr_list[] = { PCMK_XA_TYPE, PCMK_XA_CLASS, PCMK_XA_PROVIDER, }; for (int i = 0; i < PCMK__NELEM(attr_list); i++) { const char *value = crm_element_value(rsc->priv->xml, attr_list[i]); const char *old_value = crm_element_value(rsc_entry, attr_list[i]); if (!pcmk__str_eq(value, old_value, pcmk__str_none)) { changed = true; trigger_unfencing(rsc, node, "Device definition changed", NULL, rsc->priv->scheduler); if (active_on_node) { crm_notice("Forcing restart of %s on %s " "because %s changed from '%s' to '%s'", rsc->id, pcmk__node_name(node), attr_list[i], pcmk__s(old_value, ""), pcmk__s(value, "")); } } } if (changed && active_on_node) { // Make sure the resource is restarted custom_action(rsc, stop_key(rsc), PCMK_ACTION_STOP, node, FALSE, rsc->priv->scheduler); pcmk__set_rsc_flags(rsc, pcmk__rsc_start_pending); } return changed; } /*! * \internal * \brief Add resource (and any matching children) to list if it matches ID * * \param[in] result List to add resource to * \param[in] rsc Resource to check * \param[in] id ID to match * * \return (Possibly new) head of list */ static GList * add_rsc_if_matching(GList *result, pcmk_resource_t *rsc, const char *id) { if (pcmk__str_eq(id, rsc->id, pcmk__str_none) || pcmk__str_eq(id, rsc->priv->history_id, pcmk__str_none)) { result = g_list_prepend(result, rsc); } for (GList *iter = rsc->priv->children; iter != NULL; iter = iter->next) { pcmk_resource_t *child = (pcmk_resource_t *) iter->data; result = add_rsc_if_matching(result, child, id); } return result; } /*! * \internal * \brief Find all resources matching a given ID by either ID or clone name * * \param[in] id Resource ID to check * \param[in] scheduler Scheduler data * * \return List of all resources that match \p id * \note The caller is responsible for freeing the return value with * g_list_free(). */ GList * pcmk__rscs_matching_id(const char *id, const pcmk_scheduler_t *scheduler) { GList *result = NULL; CRM_CHECK((id != NULL) && (scheduler != NULL), return NULL); for (GList *iter = scheduler->priv->resources; iter != NULL; iter = iter->next) { result = add_rsc_if_matching(result, (pcmk_resource_t *) iter->data, id); } return result; } /*! * \internal * \brief Set the variant-appropriate assignment methods for a resource * * \param[in,out] data Resource to set assignment methods for * \param[in] user_data Ignored */ static void set_assignment_methods_for_rsc(gpointer data, gpointer user_data) { pcmk_resource_t *rsc = data; rsc->priv->cmds = &assignment_methods[rsc->priv->variant]; g_list_foreach(rsc->priv->children, set_assignment_methods_for_rsc, NULL); } /*! * \internal * \brief Set the variant-appropriate assignment methods for all resources * * \param[in,out] scheduler Scheduler data */ void pcmk__set_assignment_methods(pcmk_scheduler_t *scheduler) { g_list_foreach(scheduler->priv->resources, set_assignment_methods_for_rsc, NULL); } /*! * \internal * \brief Wrapper for colocated_resources() method for readability * * \param[in] rsc Resource to add to colocated list * \param[in] orig_rsc Resource originally requested * \param[in,out] list Pointer to list to add to * * \return (Possibly new) head of list */ static inline void add_colocated_resources(const pcmk_resource_t *rsc, const pcmk_resource_t *orig_rsc, GList **list) { *list = rsc->priv->cmds->colocated_resources(rsc, orig_rsc, *list); } // Shared implementation of pcmk__assignment_methods_t:colocated_resources() GList * pcmk__colocated_resources(const pcmk_resource_t *rsc, const pcmk_resource_t *orig_rsc, GList *colocated_rscs) { const GList *iter = NULL; GList *colocations = NULL; if (orig_rsc == NULL) { orig_rsc = rsc; } if ((rsc == NULL) || (g_list_find(colocated_rscs, rsc) != NULL)) { return colocated_rscs; } pcmk__rsc_trace(orig_rsc, "%s is in colocation chain with %s", rsc->id, orig_rsc->id); colocated_rscs = g_list_prepend(colocated_rscs, (gpointer) rsc); // Follow colocations where this resource is the dependent resource colocations = pcmk__this_with_colocations(rsc); for (iter = colocations; iter != NULL; iter = iter->next) { const pcmk__colocation_t *constraint = iter->data; const pcmk_resource_t *primary = constraint->primary; if (primary == orig_rsc) { continue; // Break colocation loop } if ((constraint->score == PCMK_SCORE_INFINITY) && (pcmk__colocation_affects(rsc, primary, constraint, true) == pcmk__coloc_affects_location)) { add_colocated_resources(primary, orig_rsc, &colocated_rscs); } } g_list_free(colocations); // Follow colocations where this resource is the primary resource colocations = pcmk__with_this_colocations(rsc); for (iter = colocations; iter != NULL; iter = iter->next) { const pcmk__colocation_t *constraint = iter->data; const pcmk_resource_t *dependent = constraint->dependent; if (dependent == orig_rsc) { continue; // Break colocation loop } if (pcmk__is_clone(rsc) && !pcmk__is_clone(dependent)) { continue; // We can't be sure whether dependent will be colocated } if ((constraint->score == PCMK_SCORE_INFINITY) && (pcmk__colocation_affects(dependent, rsc, constraint, true) == pcmk__coloc_affects_location)) { add_colocated_resources(dependent, orig_rsc, &colocated_rscs); } } g_list_free(colocations); return colocated_rscs; } // No-op function for variants that don't need to implement add_graph_meta() void pcmk__noop_add_graph_meta(const pcmk_resource_t *rsc, xmlNode *xml) { } /*! * \internal * \brief Output a summary of scheduled actions for a resource * * \param[in,out] rsc Resource to output actions for */ void pcmk__output_resource_actions(pcmk_resource_t *rsc) { pcmk_node_t *next = NULL; pcmk_node_t *current = NULL; pcmk__output_t *out = NULL; pcmk__assert(rsc != NULL); out = rsc->priv->scheduler->priv->out; if (rsc->priv->children != NULL) { for (GList *iter = rsc->priv->children; iter != NULL; iter = iter->next) { pcmk_resource_t *child = (pcmk_resource_t *) iter->data; child->priv->cmds->output_actions(child); } return; } next = rsc->priv->assigned_node; if (rsc->priv->active_nodes != NULL) { current = pcmk__current_node(rsc); if (rsc->priv->orig_role == pcmk_role_stopped) { /* This can occur when resources are being recovered because * the current role can change in pcmk__primitive_create_actions() */ rsc->priv->orig_role = pcmk_role_started; } } if ((current == NULL) && pcmk_is_set(rsc->flags, pcmk__rsc_removed)) { /* Don't log stopped orphans */ return; } out->message(out, "rsc-action", rsc, current, next); } /*! * \internal * \brief Add a resource to a node's list of assigned resources * * \param[in,out] node Node to add resource to * \param[in] rsc Resource to add */ static inline void add_assigned_resource(pcmk_node_t *node, pcmk_resource_t *rsc) { node->priv->assigned_resources = g_list_prepend(node->priv->assigned_resources, rsc); } /*! * \internal * \brief Assign a specified resource (of any variant) to a node * * Assign a specified resource and its children (if any) to a specified node, if * the node can run the resource (or unconditionally, if \p force is true). Mark * the resources as no longer provisional. * * If a resource can't be assigned (or \p node is \c NULL), unassign any * previous assignment. If \p stop_if_fail is \c true, set next role to stopped * and update any existing actions scheduled for the resource. * * \param[in,out] rsc Resource to assign * \param[in,out] node Node to assign \p rsc to * \param[in] force If true, assign to \p node even if unavailable * \param[in] stop_if_fail If \c true and either \p rsc can't be assigned * or \p chosen is \c NULL, set next role to * stopped and update existing actions (if \p rsc * is not a primitive, this applies to its * primitive descendants instead) * * \return \c true if the assignment of \p rsc changed, or \c false otherwise * * \note Assigning a resource to the NULL node using this function is different * from calling pcmk__unassign_resource(), in that it may also update any * actions created for the resource. * \note The \c pcmk__assignment_methods_t:assign() method is preferred, unless * a resource should be assigned to the \c NULL node or every resource in * a tree should be assigned to the same node. * \note If \p stop_if_fail is \c false, then \c pcmk__unassign_resource() can * completely undo the assignment. A successful assignment can be either * undone or left alone as final. A failed assignment has the same effect * as calling pcmk__unassign_resource(); there are no side effects on * roles or actions. */ bool pcmk__assign_resource(pcmk_resource_t *rsc, pcmk_node_t *node, bool force, bool stop_if_fail) { bool changed = false; pcmk_scheduler_t *scheduler = NULL; pcmk__assert(rsc != NULL); scheduler = rsc->priv->scheduler; if (rsc->priv->children != NULL) { for (GList *iter = rsc->priv->children; iter != NULL; iter = iter->next) { pcmk_resource_t *child_rsc = iter->data; changed |= pcmk__assign_resource(child_rsc, node, force, stop_if_fail); } return changed; } // Assigning a primitive if (!force && (node != NULL) && ((node->assign->score < 0) // Allow graph to assume that guest node connections will come up || (!pcmk__node_available(node, true, false) && !pcmk__is_guest_or_bundle_node(node)))) { pcmk__rsc_debug(rsc, "All nodes for resource %s are unavailable, unclean or " "shutting down (%s can%s run resources, with score %s)", rsc->id, pcmk__node_name(node), (pcmk__node_available(node, true, false)? "" : "not"), pcmk_readable_score(node->assign->score)); if (stop_if_fail) { pe__set_next_role(rsc, pcmk_role_stopped, "node availability"); } node = NULL; } if (rsc->priv->assigned_node != NULL) { changed = !pcmk__same_node(rsc->priv->assigned_node, node); } else { changed = (node != NULL); } pcmk__unassign_resource(rsc); pcmk__clear_rsc_flags(rsc, pcmk__rsc_unassigned); if (node == NULL) { char *rc_stopped = NULL; pcmk__rsc_debug(rsc, "Could not assign %s to a node", rsc->id); if (!stop_if_fail) { return changed; } pe__set_next_role(rsc, pcmk_role_stopped, "unable to assign"); for (GList *iter = rsc->priv->actions; iter != NULL; iter = iter->next) { pcmk_action_t *op = (pcmk_action_t *) iter->data; pcmk__rsc_debug(rsc, "Updating %s for %s assignment failure", op->uuid, rsc->id); if (pcmk__str_eq(op->task, PCMK_ACTION_STOP, pcmk__str_none)) { pcmk__clear_action_flags(op, pcmk__action_optional); } else if (pcmk__str_eq(op->task, PCMK_ACTION_START, pcmk__str_none)) { pcmk__clear_action_flags(op, pcmk__action_runnable); } else { // Cancel recurring actions, unless for stopped state const char *interval_ms_s = NULL; const char *target_rc_s = NULL; interval_ms_s = g_hash_table_lookup(op->meta, PCMK_META_INTERVAL); target_rc_s = g_hash_table_lookup(op->meta, PCMK__META_OP_TARGET_RC); if (rc_stopped == NULL) { rc_stopped = pcmk__itoa(PCMK_OCF_NOT_RUNNING); } if (!pcmk__str_eq(interval_ms_s, "0", pcmk__str_null_matches) && !pcmk__str_eq(rc_stopped, target_rc_s, pcmk__str_none)) { pcmk__clear_action_flags(op, pcmk__action_runnable); } } } free(rc_stopped); return changed; } pcmk__rsc_debug(rsc, "Assigning %s to %s", rsc->id, pcmk__node_name(node)); rsc->priv->assigned_node = pe__copy_node(node); add_assigned_resource(node, rsc); node->priv->num_resources++; node->assign->count++; pcmk__consume_node_capacity(node->priv->utilization, rsc); if (pcmk_is_set(scheduler->flags, pcmk__sched_show_utilization)) { pcmk__output_t *out = scheduler->priv->out; out->message(out, "resource-util", rsc, node, __func__); } return changed; } /*! * \internal * \brief Remove any node assignment from a specified resource and its children * * If a specified resource has been assigned to a node, remove that assignment * and mark the resource as provisional again. * * \param[in,out] rsc Resource to unassign * * \note This function is called recursively on \p rsc and its children. */ void pcmk__unassign_resource(pcmk_resource_t *rsc) { pcmk_node_t *old = rsc->priv->assigned_node; if (old == NULL) { crm_info("Unassigning %s", rsc->id); } else { crm_info("Unassigning %s from %s", rsc->id, pcmk__node_name(old)); } pcmk__set_rsc_flags(rsc, pcmk__rsc_unassigned); if (rsc->priv->children == NULL) { if (old == NULL) { return; } rsc->priv->assigned_node = NULL; - /* We're going to free the pcmk_node_t, but its details member is shared - * and will remain, so update that appropriately first. + /* We're going to free the pcmk_node_t copy, but its priv member is + * shared and will remain, so update that appropriately first. */ old->priv->assigned_resources = g_list_remove(old->priv->assigned_resources, rsc); old->priv->num_resources--; pcmk__release_node_capacity(old->priv->utilization, rsc); - free(old); + pcmk__free_node_copy(old); return; } for (GList *iter = rsc->priv->children; iter != NULL; iter = iter->next) { pcmk__unassign_resource((pcmk_resource_t *) iter->data); } } /*! * \internal * \brief Check whether a resource has reached its migration threshold on a node * * \param[in,out] rsc Resource to check * \param[in] node Node to check * \param[out] failed If threshold has been reached, this will be set to * resource that failed (possibly a parent of \p rsc) * * \return true if the migration threshold has been reached, false otherwise */ bool pcmk__threshold_reached(pcmk_resource_t *rsc, const pcmk_node_t *node, pcmk_resource_t **failed) { int fail_count, remaining_tries; pcmk_resource_t *rsc_to_ban = rsc; // Migration threshold of 0 means never force away if (rsc->priv->ban_after_failures == 0) { return false; } // If we're ignoring failures, also ignore the migration threshold if (pcmk_is_set(rsc->flags, pcmk__rsc_ignore_failure)) { return false; } // If there are no failures, there's no need to force away fail_count = pe_get_failcount(node, rsc, NULL, pcmk__fc_effective|pcmk__fc_launched, NULL); if (fail_count <= 0) { return false; } // If failed resource is anonymous clone instance, we'll force clone away if (!pcmk_is_set(rsc->flags, pcmk__rsc_unique)) { rsc_to_ban = uber_parent(rsc); } // How many more times recovery will be tried on this node remaining_tries = rsc->priv->ban_after_failures - fail_count; if (remaining_tries <= 0) { pcmk__sched_warn(rsc->priv->scheduler, "%s cannot run on %s due to reaching migration " "threshold (clean up resource to allow again) " QB_XS " failures=%d " PCMK_META_MIGRATION_THRESHOLD "=%d", rsc_to_ban->id, pcmk__node_name(node), fail_count, rsc->priv->ban_after_failures); if (failed != NULL) { *failed = rsc_to_ban; } return true; } crm_info("%s can fail %d more time%s on " "%s before reaching migration threshold (%d)", rsc_to_ban->id, remaining_tries, pcmk__plural_s(remaining_tries), pcmk__node_name(node), rsc->priv->ban_after_failures); return false; } /*! * \internal * \brief Get a node's score * * \param[in] node Node with ID to check * \param[in] nodes List of nodes to look for \p node score in * * \return Node's score, or -INFINITY if not found */ static int get_node_score(const pcmk_node_t *node, GHashTable *nodes) { pcmk_node_t *found_node = NULL; if ((node != NULL) && (nodes != NULL)) { found_node = g_hash_table_lookup(nodes, node->priv->id); } if (found_node == NULL) { return -PCMK_SCORE_INFINITY; } return found_node->assign->score; } /*! * \internal * \brief Compare two resources according to which should be assigned first * * \param[in] a First resource to compare * \param[in] b Second resource to compare * \param[in] data Sorted list of all nodes in cluster * * \return -1 if \p a should be assigned before \b, 0 if they are equal, * or +1 if \p a should be assigned after \b */ static gint cmp_resources(gconstpointer a, gconstpointer b, gpointer data) { /* GLib insists that this function require gconstpointer arguments, but we * make a small, temporary change to each argument (setting the * pe_rsc_merging flag) during comparison */ pcmk_resource_t *resource1 = (pcmk_resource_t *) a; pcmk_resource_t *resource2 = (pcmk_resource_t *) b; const GList *nodes = data; int rc = 0; int r1_score = -PCMK_SCORE_INFINITY; int r2_score = -PCMK_SCORE_INFINITY; pcmk_node_t *r1_node = NULL; pcmk_node_t *r2_node = NULL; GHashTable *r1_nodes = NULL; GHashTable *r2_nodes = NULL; const char *reason = NULL; // Resources with highest priority should be assigned first reason = "priority"; r1_score = resource1->priv->priority; r2_score = resource2->priv->priority; if (r1_score > r2_score) { rc = -1; goto done; } if (r1_score < r2_score) { rc = 1; goto done; } // We need nodes to make any other useful comparisons reason = "no node list"; if (nodes == NULL) { goto done; } // Calculate and log node scores resource1->priv->cmds->add_colocated_node_scores(resource1, NULL, resource1->id, &r1_nodes, NULL, 1, pcmk__coloc_select_this_with); resource2->priv->cmds->add_colocated_node_scores(resource2, NULL, resource2->id, &r2_nodes, NULL, 1, pcmk__coloc_select_this_with); pe__show_node_scores(true, NULL, resource1->id, r1_nodes, resource1->priv->scheduler); pe__show_node_scores(true, NULL, resource2->id, r2_nodes, resource2->priv->scheduler); // The resource with highest score on its current node goes first reason = "current location"; if (resource1->priv->active_nodes != NULL) { r1_node = pcmk__current_node(resource1); } if (resource2->priv->active_nodes != NULL) { r2_node = pcmk__current_node(resource2); } r1_score = get_node_score(r1_node, r1_nodes); r2_score = get_node_score(r2_node, r2_nodes); if (r1_score > r2_score) { rc = -1; goto done; } if (r1_score < r2_score) { rc = 1; goto done; } // Otherwise a higher score on any node will do reason = "score"; for (const GList *iter = nodes; iter != NULL; iter = iter->next) { const pcmk_node_t *node = (const pcmk_node_t *) iter->data; r1_score = get_node_score(node, r1_nodes); r2_score = get_node_score(node, r2_nodes); if (r1_score > r2_score) { rc = -1; goto done; } if (r1_score < r2_score) { rc = 1; goto done; } } done: crm_trace("%s (%d)%s%s %c %s (%d)%s%s: %s", resource1->id, r1_score, ((r1_node == NULL)? "" : " on "), ((r1_node == NULL)? "" : r1_node->priv->id), ((rc < 0)? '>' : ((rc > 0)? '<' : '=')), resource2->id, r2_score, ((r2_node == NULL)? "" : " on "), ((r2_node == NULL)? "" : r2_node->priv->id), reason); if (r1_nodes != NULL) { g_hash_table_destroy(r1_nodes); } if (r2_nodes != NULL) { g_hash_table_destroy(r2_nodes); } return rc; } /*! * \internal * \brief Sort resources in the order they should be assigned to nodes * * \param[in,out] scheduler Scheduler data */ void pcmk__sort_resources(pcmk_scheduler_t *scheduler) { GList *nodes = g_list_copy(scheduler->nodes); nodes = pcmk__sort_nodes(nodes, NULL); scheduler->priv->resources = g_list_sort_with_data(scheduler->priv->resources, cmp_resources, nodes); g_list_free(nodes); } diff --git a/lib/pengine/bundle.c b/lib/pengine/bundle.c index dd2752ebca..43aef213f6 100644 --- a/lib/pengine/bundle.c +++ b/lib/pengine/bundle.c @@ -1,2093 +1,2091 @@ /* * Copyright 2004-2024 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 enum pe__bundle_mount_flags { pe__bundle_mount_none = 0x00, // mount instance-specific subdirectory rather than source directly pe__bundle_mount_subdir = 0x01 }; typedef struct { char *source; char *target; char *options; uint32_t flags; // bitmask of pe__bundle_mount_flags } pe__bundle_mount_t; typedef struct { char *source; char *target; } pe__bundle_port_t; enum pe__container_agent { PE__CONTAINER_AGENT_UNKNOWN, PE__CONTAINER_AGENT_DOCKER, PE__CONTAINER_AGENT_PODMAN, }; #define PE__CONTAINER_AGENT_UNKNOWN_S "unknown" #define PE__CONTAINER_AGENT_DOCKER_S "docker" #define PE__CONTAINER_AGENT_PODMAN_S "podman" typedef struct pe__bundle_variant_data_s { int promoted_max; int nreplicas; int nreplicas_per_host; char *prefix; char *image; const char *ip_last; char *host_network; char *host_netmask; char *control_port; char *container_network; char *ip_range_start; gboolean add_host; gchar *container_host_options; char *container_command; char *launcher_options; const char *attribute_target; pcmk_resource_t *child; GList *replicas; // pcmk__bundle_replica_t * GList *ports; // pe__bundle_port_t * GList *mounts; // pe__bundle_mount_t * enum pe__container_agent agent_type; } pe__bundle_variant_data_t; #define get_bundle_variant_data(data, rsc) do { \ pcmk__assert(pcmk__is_bundle(rsc)); \ data = rsc->priv->variant_opaque; \ } while (0) /*! * \internal * \brief Get maximum number of bundle replicas allowed to run * * \param[in] rsc Bundle or bundled resource to check * * \return Maximum replicas for bundle corresponding to \p rsc */ int pe__bundle_max(const pcmk_resource_t *rsc) { const pe__bundle_variant_data_t *bundle_data = NULL; get_bundle_variant_data(bundle_data, pe__const_top_resource(rsc, true)); return bundle_data->nreplicas; } /*! * \internal * \brief Get the resource inside a bundle * * \param[in] bundle Bundle to check * * \return Resource inside \p bundle if any, otherwise NULL */ pcmk_resource_t * pe__bundled_resource(const pcmk_resource_t *rsc) { const pe__bundle_variant_data_t *bundle_data = NULL; get_bundle_variant_data(bundle_data, pe__const_top_resource(rsc, true)); return bundle_data->child; } /*! * \internal * \brief Get containerized resource corresponding to a given bundle container * * \param[in] instance Collective instance that might be a bundle container * * \return Bundled resource instance inside \p instance if it is a bundle * container instance, otherwise NULL */ const pcmk_resource_t * pe__get_rsc_in_container(const pcmk_resource_t *instance) { const pe__bundle_variant_data_t *data = NULL; const pcmk_resource_t *top = pe__const_top_resource(instance, true); if (!pcmk__is_bundle(top)) { return NULL; } get_bundle_variant_data(data, top); for (const GList *iter = data->replicas; iter != NULL; iter = iter->next) { const pcmk__bundle_replica_t *replica = iter->data; if (instance == replica->container) { return replica->child; } } return NULL; } /*! * \internal * \brief Check whether a given node is created by a bundle * * \param[in] bundle Bundle resource to check * \param[in] node Node to check * * \return true if \p node is an instance of \p bundle, otherwise false */ bool pe__node_is_bundle_instance(const pcmk_resource_t *bundle, const pcmk_node_t *node) { pe__bundle_variant_data_t *bundle_data = NULL; get_bundle_variant_data(bundle_data, bundle); for (GList *iter = bundle_data->replicas; iter != NULL; iter = iter->next) { pcmk__bundle_replica_t *replica = iter->data; if (pcmk__same_node(node, replica->node)) { return true; } } return false; } /*! * \internal * \brief Get the container of a bundle's first replica * * \param[in] bundle Bundle resource to get container for * * \return Container resource from first replica of \p bundle if any, * otherwise NULL */ pcmk_resource_t * pe__first_container(const pcmk_resource_t *bundle) { const pe__bundle_variant_data_t *bundle_data = NULL; const pcmk__bundle_replica_t *replica = NULL; get_bundle_variant_data(bundle_data, bundle); if (bundle_data->replicas == NULL) { return NULL; } replica = bundle_data->replicas->data; return replica->container; } /*! * \internal * \brief Iterate over bundle replicas * * \param[in,out] bundle Bundle to iterate over * \param[in] fn Function to call for each replica (its return value * indicates whether to continue iterating) * \param[in,out] user_data Pointer to pass to \p fn */ void pe__foreach_bundle_replica(pcmk_resource_t *bundle, bool (*fn)(pcmk__bundle_replica_t *, void *), void *user_data) { const pe__bundle_variant_data_t *bundle_data = NULL; get_bundle_variant_data(bundle_data, bundle); for (GList *iter = bundle_data->replicas; iter != NULL; iter = iter->next) { if (!fn((pcmk__bundle_replica_t *) iter->data, user_data)) { break; } } } /*! * \internal * \brief Iterate over const bundle replicas * * \param[in] bundle Bundle to iterate over * \param[in] fn Function to call for each replica (its return value * indicates whether to continue iterating) * \param[in,out] user_data Pointer to pass to \p fn */ void pe__foreach_const_bundle_replica(const pcmk_resource_t *bundle, bool (*fn)(const pcmk__bundle_replica_t *, void *), void *user_data) { const pe__bundle_variant_data_t *bundle_data = NULL; get_bundle_variant_data(bundle_data, bundle); for (const GList *iter = bundle_data->replicas; iter != NULL; iter = iter->next) { if (!fn((const pcmk__bundle_replica_t *) iter->data, user_data)) { break; } } } static char * next_ip(const char *last_ip) { unsigned int oct1 = 0; unsigned int oct2 = 0; unsigned int oct3 = 0; unsigned int oct4 = 0; int rc = sscanf(last_ip, "%u.%u.%u.%u", &oct1, &oct2, &oct3, &oct4); if (rc != 4) { /*@ TODO check for IPv6 */ return NULL; } else if (oct3 > 253) { return NULL; } else if (oct4 > 253) { ++oct3; oct4 = 1; } else { ++oct4; } return crm_strdup_printf("%u.%u.%u.%u", oct1, oct2, oct3, oct4); } static void allocate_ip(pe__bundle_variant_data_t *data, pcmk__bundle_replica_t *replica, GString *buffer) { if(data->ip_range_start == NULL) { return; } else if(data->ip_last) { replica->ipaddr = next_ip(data->ip_last); } else { replica->ipaddr = strdup(data->ip_range_start); } data->ip_last = replica->ipaddr; switch (data->agent_type) { case PE__CONTAINER_AGENT_DOCKER: case PE__CONTAINER_AGENT_PODMAN: if (data->add_host) { g_string_append_printf(buffer, " --add-host=%s-%d:%s", data->prefix, replica->offset, replica->ipaddr); } else { g_string_append_printf(buffer, " --hosts-entry=%s=%s-%d", replica->ipaddr, data->prefix, replica->offset); } break; default: // PE__CONTAINER_AGENT_UNKNOWN break; } } static xmlNode * create_resource(const char *name, const char *provider, const char *kind) { xmlNode *rsc = pcmk__xe_create(NULL, PCMK_XE_PRIMITIVE); crm_xml_add(rsc, PCMK_XA_ID, name); crm_xml_add(rsc, PCMK_XA_CLASS, PCMK_RESOURCE_CLASS_OCF); crm_xml_add(rsc, PCMK_XA_PROVIDER, provider); crm_xml_add(rsc, PCMK_XA_TYPE, kind); return rsc; } /*! * \internal * \brief Check whether cluster can manage resource inside container * * \param[in,out] data Container variant data * * \return TRUE if networking configuration is acceptable, FALSE otherwise * * \note The resource is manageable if an IP range or control port has been * specified. If a control port is used without an IP range, replicas per * host must be 1. */ static bool valid_network(pe__bundle_variant_data_t *data) { if(data->ip_range_start) { return TRUE; } if(data->control_port) { if(data->nreplicas_per_host > 1) { pcmk__config_err("Specifying the '" PCMK_XA_CONTROL_PORT "' for %s " "requires '" PCMK_XA_REPLICAS_PER_HOST "=1'", data->prefix); data->nreplicas_per_host = 1; // @TODO to be sure: // pcmk__clear_rsc_flags(rsc, pcmk__rsc_unique); } return TRUE; } return FALSE; } static int create_ip_resource(pcmk_resource_t *parent, pe__bundle_variant_data_t *data, pcmk__bundle_replica_t *replica) { if(data->ip_range_start) { char *id = NULL; xmlNode *xml_ip = NULL; xmlNode *xml_obj = NULL; id = crm_strdup_printf("%s-ip-%s", data->prefix, replica->ipaddr); pcmk__xml_sanitize_id(id); xml_ip = create_resource(id, "heartbeat", "IPaddr2"); free(id); xml_obj = pcmk__xe_create(xml_ip, PCMK_XE_INSTANCE_ATTRIBUTES); pcmk__xe_set_id(xml_obj, "%s-attributes-%d", data->prefix, replica->offset); crm_create_nvpair_xml(xml_obj, NULL, "ip", replica->ipaddr); if(data->host_network) { crm_create_nvpair_xml(xml_obj, NULL, "nic", data->host_network); } if(data->host_netmask) { crm_create_nvpair_xml(xml_obj, NULL, "cidr_netmask", data->host_netmask); } else { crm_create_nvpair_xml(xml_obj, NULL, "cidr_netmask", "32"); } xml_obj = pcmk__xe_create(xml_ip, PCMK_XE_OPERATIONS); crm_create_op_xml(xml_obj, pcmk__xe_id(xml_ip), PCMK_ACTION_MONITOR, "60s", NULL); // TODO: Other ops? Timeouts and intervals from underlying resource? if (pe__unpack_resource(xml_ip, &replica->ip, parent, parent->priv->scheduler) != pcmk_rc_ok) { return pcmk_rc_unpack_error; } parent->priv->children = g_list_append(parent->priv->children, replica->ip); } return pcmk_rc_ok; } static const char* container_agent_str(enum pe__container_agent t) { switch (t) { case PE__CONTAINER_AGENT_DOCKER: return PE__CONTAINER_AGENT_DOCKER_S; case PE__CONTAINER_AGENT_PODMAN: return PE__CONTAINER_AGENT_PODMAN_S; default: // PE__CONTAINER_AGENT_UNKNOWN break; } return PE__CONTAINER_AGENT_UNKNOWN_S; } static int create_container_resource(pcmk_resource_t *parent, const pe__bundle_variant_data_t *data, pcmk__bundle_replica_t *replica) { char *id = NULL; xmlNode *xml_container = NULL; xmlNode *xml_obj = NULL; // Agent-specific const char *hostname_opt = NULL; const char *env_opt = NULL; const char *agent_str = NULL; GString *buffer = NULL; GString *dbuffer = NULL; // Where syntax differences are drop-in replacements, set them now switch (data->agent_type) { case PE__CONTAINER_AGENT_DOCKER: case PE__CONTAINER_AGENT_PODMAN: hostname_opt = "-h "; env_opt = "-e "; break; default: // PE__CONTAINER_AGENT_UNKNOWN return pcmk_rc_unpack_error; } agent_str = container_agent_str(data->agent_type); buffer = g_string_sized_new(4096); id = crm_strdup_printf("%s-%s-%d", data->prefix, agent_str, replica->offset); pcmk__xml_sanitize_id(id); xml_container = create_resource(id, "heartbeat", agent_str); free(id); xml_obj = pcmk__xe_create(xml_container, PCMK_XE_INSTANCE_ATTRIBUTES); pcmk__xe_set_id(xml_obj, "%s-attributes-%d", data->prefix, replica->offset); crm_create_nvpair_xml(xml_obj, NULL, "image", data->image); crm_create_nvpair_xml(xml_obj, NULL, "allow_pull", PCMK_VALUE_TRUE); crm_create_nvpair_xml(xml_obj, NULL, "force_kill", PCMK_VALUE_FALSE); crm_create_nvpair_xml(xml_obj, NULL, "reuse", PCMK_VALUE_FALSE); if (data->agent_type == PE__CONTAINER_AGENT_DOCKER) { g_string_append(buffer, " --restart=no"); } /* Set a container hostname only if we have an IP to map it to. The user can * set -h or --uts=host themselves if they want a nicer name for logs, but * this makes applications happy who need their hostname to match the IP * they bind to. */ if (data->ip_range_start != NULL) { g_string_append_printf(buffer, " %s%s-%d", hostname_opt, data->prefix, replica->offset); } pcmk__g_strcat(buffer, " ", env_opt, "PCMK_stderr=1", NULL); if (data->container_network != NULL) { pcmk__g_strcat(buffer, " --net=", data->container_network, NULL); } if (data->control_port != NULL) { pcmk__g_strcat(buffer, " ", env_opt, "PCMK_" PCMK__ENV_REMOTE_PORT "=", data->control_port, NULL); } else { g_string_append_printf(buffer, " %sPCMK_" PCMK__ENV_REMOTE_PORT "=%d", env_opt, DEFAULT_REMOTE_PORT); } for (GList *iter = data->mounts; iter != NULL; iter = iter->next) { pe__bundle_mount_t *mount = (pe__bundle_mount_t *) iter->data; char *source = NULL; if (pcmk_is_set(mount->flags, pe__bundle_mount_subdir)) { source = crm_strdup_printf("%s/%s-%d", mount->source, data->prefix, replica->offset); pcmk__add_separated_word(&dbuffer, 1024, source, ","); } switch (data->agent_type) { case PE__CONTAINER_AGENT_DOCKER: case PE__CONTAINER_AGENT_PODMAN: pcmk__g_strcat(buffer, " -v ", pcmk__s(source, mount->source), ":", mount->target, NULL); if (mount->options != NULL) { pcmk__g_strcat(buffer, ":", mount->options, NULL); } break; default: break; } free(source); } for (GList *iter = data->ports; iter != NULL; iter = iter->next) { pe__bundle_port_t *port = (pe__bundle_port_t *) iter->data; switch (data->agent_type) { case PE__CONTAINER_AGENT_DOCKER: case PE__CONTAINER_AGENT_PODMAN: if (replica->ipaddr != NULL) { pcmk__g_strcat(buffer, " -p ", replica->ipaddr, ":", port->source, ":", port->target, NULL); } else if (!pcmk__str_eq(data->container_network, PCMK_VALUE_HOST, pcmk__str_none)) { // No need to do port mapping if net == host pcmk__g_strcat(buffer, " -p ", port->source, ":", port->target, NULL); } break; default: break; } } /* @COMPAT: We should use pcmk__add_word() here, but we can't yet, because * it would cause restarts during rolling upgrades. * * In a previous version of the container resource creation logic, if * data->launcher_options is not NULL, we append * (" %s", data->launcher_options) even if data->launcher_options is an * empty string. Likewise for data->container_host_options. Using * * pcmk__add_word(buffer, 0, data->launcher_options) * * removes that extra trailing space, causing a resource definition change. */ if (data->launcher_options != NULL) { pcmk__g_strcat(buffer, " ", data->launcher_options, NULL); } if (data->container_host_options != NULL) { pcmk__g_strcat(buffer, " ", data->container_host_options, NULL); } crm_create_nvpair_xml(xml_obj, NULL, "run_opts", (const char *) buffer->str); g_string_free(buffer, TRUE); crm_create_nvpair_xml(xml_obj, NULL, "mount_points", (dbuffer != NULL)? (const char *) dbuffer->str : ""); if (dbuffer != NULL) { g_string_free(dbuffer, TRUE); } if (replica->child != NULL) { if (data->container_command != NULL) { crm_create_nvpair_xml(xml_obj, NULL, "run_cmd", data->container_command); } else { crm_create_nvpair_xml(xml_obj, NULL, "run_cmd", SBIN_DIR "/" PCMK__SERVER_REMOTED); } /* TODO: Allow users to specify their own? * * We just want to know if the container is alive; we'll monitor the * child independently. */ crm_create_nvpair_xml(xml_obj, NULL, "monitor_cmd", "/bin/true"); #if 0 /* @TODO Consider supporting the use case where we can start and stop * resources, but not proxy local commands (such as setting node * attributes), by running the local executor in stand-alone mode. * However, this would probably be better done via ACLs as with other * Pacemaker Remote nodes. */ } else if ((child != NULL) && data->untrusted) { crm_create_nvpair_xml(xml_obj, NULL, "run_cmd", CRM_DAEMON_DIR "/" PCMK__SERVER_EXECD); crm_create_nvpair_xml(xml_obj, NULL, "monitor_cmd", CRM_DAEMON_DIR "/pacemaker/cts-exec-helper -c poke"); #endif } else { if (data->container_command != NULL) { crm_create_nvpair_xml(xml_obj, NULL, "run_cmd", data->container_command); } /* TODO: Allow users to specify their own? * * We don't know what's in the container, so we just want to know if it * is alive. */ crm_create_nvpair_xml(xml_obj, NULL, "monitor_cmd", "/bin/true"); } xml_obj = pcmk__xe_create(xml_container, PCMK_XE_OPERATIONS); crm_create_op_xml(xml_obj, pcmk__xe_id(xml_container), PCMK_ACTION_MONITOR, "60s", NULL); // TODO: Other ops? Timeouts and intervals from underlying resource? if (pe__unpack_resource(xml_container, &replica->container, parent, parent->priv->scheduler) != pcmk_rc_ok) { return pcmk_rc_unpack_error; } pcmk__set_rsc_flags(replica->container, pcmk__rsc_replica_container); parent->priv->children = g_list_append(parent->priv->children, replica->container); return pcmk_rc_ok; } /*! * \brief Ban a node from a resource's (and its children's) allowed nodes list * * \param[in,out] rsc Resource to modify * \param[in] uname Name of node to ban */ static void disallow_node(pcmk_resource_t *rsc, const char *uname) { gpointer match = g_hash_table_lookup(rsc->priv->allowed_nodes, uname); if (match) { ((pcmk_node_t *) match)->assign->score = -PCMK_SCORE_INFINITY; ((pcmk_node_t *) match)->assign->probe_mode = pcmk__probe_never; } g_list_foreach(rsc->priv->children, (GFunc) disallow_node, (gpointer) uname); } static int create_remote_resource(pcmk_resource_t *parent, pe__bundle_variant_data_t *data, pcmk__bundle_replica_t *replica) { if (replica->child && valid_network(data)) { GHashTableIter gIter; pcmk_node_t *node = NULL; xmlNode *xml_remote = NULL; char *id = crm_strdup_printf("%s-%d", data->prefix, replica->offset); char *port_s = NULL; const char *uname = NULL; const char *connect_name = NULL; pcmk_scheduler_t *scheduler = parent->priv->scheduler; if (pe_find_resource(scheduler->priv->resources, id) != NULL) { free(id); // The biggest hammer we have id = crm_strdup_printf("pcmk-internal-%s-remote-%d", replica->child->id, replica->offset); //@TODO return error instead of asserting? pcmk__assert(pe_find_resource(scheduler->priv->resources, id) == NULL); } /* REMOTE_CONTAINER_HACK: Using "#uname" as the server name when the * connection does not have its own IP is a magic string that we use to * support nested remotes (i.e. a bundle running on a remote node). */ connect_name = (replica->ipaddr? replica->ipaddr : "#uname"); if (data->control_port == NULL) { port_s = pcmk__itoa(DEFAULT_REMOTE_PORT); } /* This sets replica->container as replica->remote's container, which is * similar to what happens with guest nodes. This is how the scheduler * knows that the bundle node is fenced by recovering the container, and * that remote should be ordered relative to the container. */ xml_remote = pe_create_remote_xml(NULL, id, replica->container->id, NULL, NULL, NULL, connect_name, (data->control_port? data->control_port : port_s)); free(port_s); /* Abandon our created ID, and pull the copy from the XML, because we * need something that will get freed during scheduler data cleanup to * use as the node ID and uname. */ free(id); id = NULL; uname = pcmk__xe_id(xml_remote); /* Ensure a node has been created for the guest (it may have already * been, if it has a permanent node attribute), and ensure its weight is * -INFINITY so no other resources can run on it. */ node = pcmk_find_node(scheduler, uname); if (node == NULL) { node = pe_create_node(uname, uname, PCMK_VALUE_REMOTE, -PCMK_SCORE_INFINITY, scheduler); } else { node->assign->score = -PCMK_SCORE_INFINITY; } node->assign->probe_mode = pcmk__probe_never; /* unpack_remote_nodes() ensures that each remote node and guest node * has a pcmk_node_t entry. Ideally, it would do the same for bundle * nodes. Unfortunately, a bundle has to be mostly unpacked before it's * obvious what nodes will be needed, so we do it just above. * * Worse, that means that the node may have been utilized while * unpacking other resources, without our weight correction. The most * likely place for this to happen is when pe__unpack_resource() calls * resource_location() to set a default score in symmetric clusters. * This adds a node *copy* to each resource's allowed nodes, and these * copies will have the wrong weight. * * As a hacky workaround, fix those copies here. * * @TODO Possible alternative: ensure bundles are unpacked before other * resources, so the weight is correct before any copies are made. */ g_list_foreach(scheduler->priv->resources, (GFunc) disallow_node, (gpointer) uname); replica->node = pe__copy_node(node); replica->node->assign->score = 500; replica->node->assign->probe_mode = pcmk__probe_exclusive; /* Ensure the node shows up as allowed and with the correct discovery set */ if (replica->child->priv->allowed_nodes != NULL) { g_hash_table_destroy(replica->child->priv->allowed_nodes); } replica->child->priv->allowed_nodes = pcmk__strkey_table(NULL, free); g_hash_table_insert(replica->child->priv->allowed_nodes, (gpointer) replica->node->priv->id, pe__copy_node(replica->node)); { const pcmk_resource_t *parent = replica->child->priv->parent; pcmk_node_t *copy = pe__copy_node(replica->node); copy->assign->score = -PCMK_SCORE_INFINITY; g_hash_table_insert(parent->priv->allowed_nodes, (gpointer) replica->node->priv->id, copy); } if (pe__unpack_resource(xml_remote, &replica->remote, parent, scheduler) != pcmk_rc_ok) { return pcmk_rc_unpack_error; } g_hash_table_iter_init(&gIter, replica->remote->priv->allowed_nodes); while (g_hash_table_iter_next(&gIter, NULL, (void **)&node)) { if (pcmk__is_pacemaker_remote_node(node)) { /* Remote resources can only run on 'normal' cluster node */ node->assign->score = -PCMK_SCORE_INFINITY; } } replica->node->priv->remote = replica->remote; // Ensure pcmk__is_guest_or_bundle_node() functions correctly replica->remote->priv->launcher = replica->container; /* A bundle's #kind is closer to "container" (guest node) than the * "remote" set by pe_create_node(). */ pcmk__insert_dup(replica->node->priv->attrs, CRM_ATTR_KIND, "container"); /* One effect of this is that unpack_launcher() will add * replica->remote to replica->container's launched resources, which * will make pe__resource_contains_guest_node() true for * replica->container. * * replica->child does NOT get added to replica->container's launched * resources. The only noticeable effect if it did would be for its * fail count to be taken into account when checking * replica->container's migration threshold. */ parent->priv->children = g_list_append(parent->priv->children, replica->remote); } return pcmk_rc_ok; } static int create_replica_resources(pcmk_resource_t *parent, pe__bundle_variant_data_t *data, pcmk__bundle_replica_t *replica) { int rc = pcmk_rc_ok; rc = create_container_resource(parent, data, replica); if (rc != pcmk_rc_ok) { return rc; } rc = create_ip_resource(parent, data, replica); if (rc != pcmk_rc_ok) { return rc; } rc = create_remote_resource(parent, data, replica); if (rc != pcmk_rc_ok) { return rc; } if ((replica->child != NULL) && (replica->ipaddr != NULL)) { pcmk__insert_meta(replica->child->priv, "external-ip", replica->ipaddr); } if (replica->remote != NULL) { /* * Allow the remote connection resource to be allocated to a * different node than the one on which the container is active. * * This makes it possible to have Pacemaker Remote nodes running * containers with the remote executor inside in order to start * services inside those containers. */ pcmk__set_rsc_flags(replica->remote, pcmk__rsc_remote_nesting_allowed); } return rc; } static void mount_add(pe__bundle_variant_data_t *bundle_data, const char *source, const char *target, const char *options, uint32_t flags) { pe__bundle_mount_t *mount = pcmk__assert_alloc(1, sizeof(pe__bundle_mount_t)); mount->source = pcmk__str_copy(source); mount->target = pcmk__str_copy(target); mount->options = pcmk__str_copy(options); mount->flags = flags; bundle_data->mounts = g_list_append(bundle_data->mounts, mount); } static void mount_free(pe__bundle_mount_t *mount) { free(mount->source); free(mount->target); free(mount->options); free(mount); } static void port_free(pe__bundle_port_t *port) { free(port->source); free(port->target); free(port); } static pcmk__bundle_replica_t * replica_for_remote(pcmk_resource_t *remote) { pcmk_resource_t *top = remote; pe__bundle_variant_data_t *bundle_data = NULL; if (top == NULL) { return NULL; } while (top->priv->parent != NULL) { top = top->priv->parent; } get_bundle_variant_data(bundle_data, top); for (GList *gIter = bundle_data->replicas; gIter != NULL; gIter = gIter->next) { pcmk__bundle_replica_t *replica = gIter->data; if (replica->remote == remote) { return replica; } } CRM_LOG_ASSERT(FALSE); return NULL; } bool pe__bundle_needs_remote_name(pcmk_resource_t *rsc) { const char *value; GHashTable *params = NULL; if (rsc == NULL) { return false; } // Use NULL node since pcmk__bundle_expand() uses that to set value params = pe_rsc_params(rsc, NULL, rsc->priv->scheduler); value = g_hash_table_lookup(params, PCMK_REMOTE_RA_ADDR); return pcmk__str_eq(value, "#uname", pcmk__str_casei) && xml_contains_remote_node(rsc->priv->xml); } const char * pe__add_bundle_remote_name(pcmk_resource_t *rsc, xmlNode *xml, const char *field) { // REMOTE_CONTAINER_HACK: Allow remote nodes that start containers with pacemaker remote inside pcmk_node_t *node = NULL; pcmk__bundle_replica_t *replica = NULL; if (!pe__bundle_needs_remote_name(rsc)) { return NULL; } replica = replica_for_remote(rsc); if (replica == NULL) { return NULL; } node = replica->container->priv->assigned_node; if (node == NULL) { /* If it won't be running anywhere after the * transition, go with where it's running now. */ node = pcmk__current_node(replica->container); } if(node == NULL) { crm_trace("Cannot determine address for bundle connection %s", rsc->id); return NULL; } crm_trace("Setting address for bundle connection %s to bundle host %s", rsc->id, pcmk__node_name(node)); if(xml != NULL && field != NULL) { crm_xml_add(xml, field, node->priv->name); } return node->priv->name; } #define pe__set_bundle_mount_flags(mount_xml, flags, flags_to_set) do { \ flags = pcmk__set_flags_as(__func__, __LINE__, LOG_TRACE, \ "Bundle mount", pcmk__xe_id(mount_xml), \ flags, (flags_to_set), #flags_to_set); \ } while (0) gboolean pe__unpack_bundle(pcmk_resource_t *rsc, pcmk_scheduler_t *scheduler) { const char *value = NULL; xmlNode *xml_obj = NULL; const xmlNode *xml_child = NULL; xmlNode *xml_resource = NULL; pe__bundle_variant_data_t *bundle_data = NULL; bool need_log_mount = TRUE; pcmk__assert(rsc != NULL); pcmk__rsc_trace(rsc, "Processing resource %s...", rsc->id); bundle_data = pcmk__assert_alloc(1, sizeof(pe__bundle_variant_data_t)); rsc->priv->variant_opaque = bundle_data; bundle_data->prefix = strdup(rsc->id); xml_obj = pcmk__xe_first_child(rsc->priv->xml, PCMK_XE_DOCKER, NULL, NULL); if (xml_obj != NULL) { bundle_data->agent_type = PE__CONTAINER_AGENT_DOCKER; } if (xml_obj == NULL) { xml_obj = pcmk__xe_first_child(rsc->priv->xml, PCMK_XE_PODMAN, NULL, NULL); if (xml_obj != NULL) { bundle_data->agent_type = PE__CONTAINER_AGENT_PODMAN; } } if (xml_obj == NULL) { return FALSE; } // Use 0 for default, minimum, and invalid PCMK_XA_PROMOTED_MAX value = crm_element_value(xml_obj, PCMK_XA_PROMOTED_MAX); pcmk__scan_min_int(value, &bundle_data->promoted_max, 0); /* Default replicas to PCMK_XA_PROMOTED_MAX if it was specified and 1 * otherwise */ value = crm_element_value(xml_obj, PCMK_XA_REPLICAS); if ((value == NULL) && (bundle_data->promoted_max > 0)) { bundle_data->nreplicas = bundle_data->promoted_max; } else { pcmk__scan_min_int(value, &bundle_data->nreplicas, 1); } /* * Communication between containers on the same host via the * floating IPs only works if the container is started with: * --userland-proxy=false --ip-masq=false */ value = crm_element_value(xml_obj, PCMK_XA_REPLICAS_PER_HOST); pcmk__scan_min_int(value, &bundle_data->nreplicas_per_host, 1); if (bundle_data->nreplicas_per_host == 1) { pcmk__clear_rsc_flags(rsc, pcmk__rsc_unique); } bundle_data->container_command = crm_element_value_copy(xml_obj, PCMK_XA_RUN_COMMAND); bundle_data->launcher_options = crm_element_value_copy(xml_obj, PCMK_XA_OPTIONS); bundle_data->image = crm_element_value_copy(xml_obj, PCMK_XA_IMAGE); bundle_data->container_network = crm_element_value_copy(xml_obj, PCMK_XA_NETWORK); xml_obj = pcmk__xe_first_child(rsc->priv->xml, PCMK_XE_NETWORK, NULL, NULL); if(xml_obj) { bundle_data->ip_range_start = crm_element_value_copy(xml_obj, PCMK_XA_IP_RANGE_START); bundle_data->host_netmask = crm_element_value_copy(xml_obj, PCMK_XA_HOST_NETMASK); bundle_data->host_network = crm_element_value_copy(xml_obj, PCMK_XA_HOST_INTERFACE); bundle_data->control_port = crm_element_value_copy(xml_obj, PCMK_XA_CONTROL_PORT); value = crm_element_value(xml_obj, PCMK_XA_ADD_HOST); if (crm_str_to_boolean(value, &bundle_data->add_host) != 1) { bundle_data->add_host = TRUE; } for (xml_child = pcmk__xe_first_child(xml_obj, PCMK_XE_PORT_MAPPING, NULL, NULL); xml_child != NULL; xml_child = pcmk__xe_next(xml_child, PCMK_XE_PORT_MAPPING)) { pe__bundle_port_t *port = pcmk__assert_alloc(1, sizeof(pe__bundle_port_t)); port->source = crm_element_value_copy(xml_child, PCMK_XA_PORT); if(port->source == NULL) { port->source = crm_element_value_copy(xml_child, PCMK_XA_RANGE); } else { port->target = crm_element_value_copy(xml_child, PCMK_XA_INTERNAL_PORT); } if(port->source != NULL && strlen(port->source) > 0) { if(port->target == NULL) { port->target = strdup(port->source); } bundle_data->ports = g_list_append(bundle_data->ports, port); } else { pcmk__config_err("Invalid " PCMK_XA_PORT " directive %s", pcmk__xe_id(xml_child)); port_free(port); } } } xml_obj = pcmk__xe_first_child(rsc->priv->xml, PCMK_XE_STORAGE, NULL, NULL); for (xml_child = pcmk__xe_first_child(xml_obj, PCMK_XE_STORAGE_MAPPING, NULL, NULL); xml_child != NULL; xml_child = pcmk__xe_next(xml_child, PCMK_XE_STORAGE_MAPPING)) { const char *source = crm_element_value(xml_child, PCMK_XA_SOURCE_DIR); const char *target = crm_element_value(xml_child, PCMK_XA_TARGET_DIR); const char *options = crm_element_value(xml_child, PCMK_XA_OPTIONS); int flags = pe__bundle_mount_none; if (source == NULL) { source = crm_element_value(xml_child, PCMK_XA_SOURCE_DIR_ROOT); pe__set_bundle_mount_flags(xml_child, flags, pe__bundle_mount_subdir); } if (source && target) { mount_add(bundle_data, source, target, options, flags); if (strcmp(target, "/var/log") == 0) { need_log_mount = FALSE; } } else { pcmk__config_err("Invalid mount directive %s", pcmk__xe_id(xml_child)); } } xml_obj = pcmk__xe_first_child(rsc->priv->xml, PCMK_XE_PRIMITIVE, NULL, NULL); if (xml_obj && valid_network(bundle_data)) { const char *suffix = NULL; char *value = NULL; xmlNode *xml_set = NULL; xml_resource = pcmk__xe_create(NULL, PCMK_XE_CLONE); /* @COMPAT We no longer use the tag, but we need to keep it as * part of the resource name, so that bundles don't restart in a rolling * upgrade. (It also avoids needing to change regression tests.) */ suffix = (const char *) xml_resource->name; if (bundle_data->promoted_max > 0) { suffix = "master"; } pcmk__xe_set_id(xml_resource, "%s-%s", bundle_data->prefix, suffix); xml_set = pcmk__xe_create(xml_resource, PCMK_XE_META_ATTRIBUTES); pcmk__xe_set_id(xml_set, "%s-%s-meta", bundle_data->prefix, xml_resource->name); crm_create_nvpair_xml(xml_set, NULL, PCMK_META_ORDERED, PCMK_VALUE_TRUE); value = pcmk__itoa(bundle_data->nreplicas); crm_create_nvpair_xml(xml_set, NULL, PCMK_META_CLONE_MAX, value); free(value); value = pcmk__itoa(bundle_data->nreplicas_per_host); crm_create_nvpair_xml(xml_set, NULL, PCMK_META_CLONE_NODE_MAX, value); free(value); crm_create_nvpair_xml(xml_set, NULL, PCMK_META_GLOBALLY_UNIQUE, pcmk__btoa(bundle_data->nreplicas_per_host > 1)); if (bundle_data->promoted_max) { crm_create_nvpair_xml(xml_set, NULL, PCMK_META_PROMOTABLE, PCMK_VALUE_TRUE); value = pcmk__itoa(bundle_data->promoted_max); crm_create_nvpair_xml(xml_set, NULL, PCMK_META_PROMOTED_MAX, value); free(value); } //crm_xml_add(xml_obj, PCMK_XA_ID, bundle_data->prefix); pcmk__xml_copy(xml_resource, xml_obj); } else if(xml_obj) { pcmk__config_err("Cannot control %s inside %s without either " PCMK_XA_IP_RANGE_START " or " PCMK_XA_CONTROL_PORT, rsc->id, pcmk__xe_id(xml_obj)); return FALSE; } if(xml_resource) { int lpc = 0; GList *childIter = NULL; pe__bundle_port_t *port = NULL; GString *buffer = NULL; if (pe__unpack_resource(xml_resource, &(bundle_data->child), rsc, scheduler) != pcmk_rc_ok) { return FALSE; } /* Currently, we always map the default authentication key location * into the same location inside the container. * * Ideally, we would respect the host's PCMK_authkey_location, but: * - it may be different on different nodes; * - the actual connection will do extra checking to make sure the key * file exists and is readable, that we can't do here on the DC * - tools such as crm_resource and crm_simulate may not have the same * environment variables as the cluster, causing operation digests to * differ * * Always using the default location inside the container is fine, * because we control the pacemaker_remote environment, and it avoids * having to pass another environment variable to the container. * * @TODO A better solution may be to have only pacemaker_remote use the * environment variable, and have the cluster nodes use a new * cluster option for key location. This would introduce the limitation * of the location being the same on all cluster nodes, but that's * reasonable. */ mount_add(bundle_data, DEFAULT_REMOTE_KEY_LOCATION, DEFAULT_REMOTE_KEY_LOCATION, NULL, pe__bundle_mount_none); if (need_log_mount) { mount_add(bundle_data, CRM_BUNDLE_DIR, "/var/log", NULL, pe__bundle_mount_subdir); } port = pcmk__assert_alloc(1, sizeof(pe__bundle_port_t)); if(bundle_data->control_port) { port->source = strdup(bundle_data->control_port); } else { /* If we wanted to respect PCMK_remote_port, we could use * crm_default_remote_port() here and elsewhere in this file instead * of DEFAULT_REMOTE_PORT. * * However, it gains nothing, since we control both the container * environment and the connection resource parameters, and the user * can use a different port if desired by setting * PCMK_XA_CONTROL_PORT. */ port->source = pcmk__itoa(DEFAULT_REMOTE_PORT); } port->target = strdup(port->source); bundle_data->ports = g_list_append(bundle_data->ports, port); buffer = g_string_sized_new(1024); for (childIter = bundle_data->child->priv->children; childIter != NULL; childIter = childIter->next) { pcmk__bundle_replica_t *replica = NULL; replica = pcmk__assert_alloc(1, sizeof(pcmk__bundle_replica_t)); replica->child = childIter->data; pcmk__set_rsc_flags(replica->child, pcmk__rsc_exclusive_probes); replica->offset = lpc++; // Ensure the child's notify gets set based on the underlying primitive's value if (pcmk_is_set(replica->child->flags, pcmk__rsc_notify)) { pcmk__set_rsc_flags(bundle_data->child, pcmk__rsc_notify); } allocate_ip(bundle_data, replica, buffer); bundle_data->replicas = g_list_append(bundle_data->replicas, replica); bundle_data->attribute_target = g_hash_table_lookup(replica->child->priv->meta, PCMK_META_CONTAINER_ATTRIBUTE_TARGET); } bundle_data->container_host_options = g_string_free(buffer, FALSE); if (bundle_data->attribute_target) { pcmk__insert_dup(rsc->priv->meta, PCMK_META_CONTAINER_ATTRIBUTE_TARGET, bundle_data->attribute_target); pcmk__insert_dup(bundle_data->child->priv->meta, PCMK_META_CONTAINER_ATTRIBUTE_TARGET, bundle_data->attribute_target); } } else { // Just a naked container, no pacemaker-remote GString *buffer = g_string_sized_new(1024); for (int lpc = 0; lpc < bundle_data->nreplicas; lpc++) { pcmk__bundle_replica_t *replica = NULL; replica = pcmk__assert_alloc(1, sizeof(pcmk__bundle_replica_t)); replica->offset = lpc; allocate_ip(bundle_data, replica, buffer); bundle_data->replicas = g_list_append(bundle_data->replicas, replica); } bundle_data->container_host_options = g_string_free(buffer, FALSE); } for (GList *gIter = bundle_data->replicas; gIter != NULL; gIter = gIter->next) { pcmk__bundle_replica_t *replica = gIter->data; if (create_replica_resources(rsc, bundle_data, replica) != pcmk_rc_ok) { pcmk__config_err("Failed unpacking resource %s", rsc->id); rsc->priv->fns->free(rsc); return FALSE; } /* Utilization needs special handling for bundles. It makes no sense for * the inner primitive to have utilization, because it is tied * one-to-one to the guest node created by the container resource -- and * there's no way to set capacities for that guest node anyway. * * What the user really wants is to configure utilization for the * container. However, the schema only allows utilization for * primitives, and the container resource is implicit anyway, so the * user can *only* configure utilization for the inner primitive. If * they do, move the primitive's utilization values to the container. * * @TODO This means that bundles without an inner primitive can't have * utilization. An alternative might be to allow utilization values in * the top-level bundle XML in the schema, and copy those to each * container. */ if (replica->child != NULL) { GHashTable *empty = replica->container->priv->utilization; replica->container->priv->utilization = replica->child->priv->utilization; replica->child->priv->utilization = empty; } } if (bundle_data->child) { rsc->priv->children = g_list_append(rsc->priv->children, bundle_data->child); } return TRUE; } static int replica_resource_active(pcmk_resource_t *rsc, gboolean all) { if (rsc) { gboolean child_active = rsc->priv->fns->active(rsc, all); if (child_active && !all) { return TRUE; } else if (!child_active && all) { return FALSE; } } return -1; } gboolean pe__bundle_active(pcmk_resource_t *rsc, gboolean all) { pe__bundle_variant_data_t *bundle_data = NULL; GList *iter = NULL; get_bundle_variant_data(bundle_data, rsc); for (iter = bundle_data->replicas; iter != NULL; iter = iter->next) { pcmk__bundle_replica_t *replica = iter->data; int rsc_active; rsc_active = replica_resource_active(replica->ip, all); if (rsc_active >= 0) { return (gboolean) rsc_active; } rsc_active = replica_resource_active(replica->child, all); if (rsc_active >= 0) { return (gboolean) rsc_active; } rsc_active = replica_resource_active(replica->container, all); if (rsc_active >= 0) { return (gboolean) rsc_active; } rsc_active = replica_resource_active(replica->remote, all); if (rsc_active >= 0) { return (gboolean) rsc_active; } } /* If "all" is TRUE, we've already checked that no resources were inactive, * so return TRUE; if "all" is FALSE, we didn't find any active resources, * so return FALSE. */ return all; } /*! * \internal * \brief Find the bundle replica corresponding to a given node * * \param[in] bundle Top-level bundle resource * \param[in] node Node to search for * * \return Bundle replica if found, NULL otherwise */ pcmk_resource_t * pe__find_bundle_replica(const pcmk_resource_t *bundle, const pcmk_node_t *node) { pe__bundle_variant_data_t *bundle_data = NULL; pcmk__assert((bundle != NULL) && (node != NULL)); get_bundle_variant_data(bundle_data, bundle); for (GList *gIter = bundle_data->replicas; gIter != NULL; gIter = gIter->next) { pcmk__bundle_replica_t *replica = gIter->data; pcmk__assert((replica != NULL) && (replica->node != NULL)); if (pcmk__same_node(replica->node, node)) { return replica->child; } } return NULL; } PCMK__OUTPUT_ARGS("bundle", "uint32_t", "pcmk_resource_t *", "GList *", "GList *") int pe__bundle_xml(pcmk__output_t *out, va_list args) { uint32_t show_opts = va_arg(args, uint32_t); pcmk_resource_t *rsc = va_arg(args, pcmk_resource_t *); GList *only_node = va_arg(args, GList *); GList *only_rsc = va_arg(args, GList *); pe__bundle_variant_data_t *bundle_data = NULL; int rc = pcmk_rc_no_output; gboolean printed_header = FALSE; gboolean print_everything = TRUE; const char *desc = NULL; pcmk__assert(rsc != NULL); get_bundle_variant_data(bundle_data, rsc); if (rsc->priv->fns->is_filtered(rsc, only_rsc, TRUE)) { return rc; } print_everything = pcmk__str_in_list(rsc->id, only_rsc, pcmk__str_star_matches); for (GList *gIter = bundle_data->replicas; gIter != NULL; gIter = gIter->next) { pcmk__bundle_replica_t *replica = gIter->data; pcmk_resource_t *ip = replica->ip; pcmk_resource_t *child = replica->child; pcmk_resource_t *container = replica->container; pcmk_resource_t *remote = replica->remote; char *id = NULL; gboolean print_ip, print_child, print_ctnr, print_remote; pcmk__assert(replica != NULL); if (pcmk__rsc_filtered_by_node(container, only_node)) { continue; } print_ip = (ip != NULL) && !ip->priv->fns->is_filtered(ip, only_rsc, print_everything); print_child = (child != NULL) && !child->priv->fns->is_filtered(child, only_rsc, print_everything); print_ctnr = !container->priv->fns->is_filtered(container, only_rsc, print_everything); print_remote = (remote != NULL) && !remote->priv->fns->is_filtered(remote, only_rsc, print_everything); if (!print_everything && !print_ip && !print_child && !print_ctnr && !print_remote) { continue; } if (!printed_header) { const char *type = container_agent_str(bundle_data->agent_type); const char *unique = pcmk__flag_text(rsc->flags, pcmk__rsc_unique); const char *maintenance = pcmk__flag_text(rsc->flags, pcmk__rsc_maintenance); const char *managed = pcmk__flag_text(rsc->flags, pcmk__rsc_managed); const char *failed = pcmk__flag_text(rsc->flags, pcmk__rsc_failed); printed_header = TRUE; desc = pe__resource_description(rsc, show_opts); rc = pe__name_and_nvpairs_xml(out, true, PCMK_XE_BUNDLE, PCMK_XA_ID, rsc->id, PCMK_XA_TYPE, type, PCMK_XA_IMAGE, bundle_data->image, PCMK_XA_UNIQUE, unique, PCMK_XA_MAINTENANCE, maintenance, PCMK_XA_MANAGED, managed, PCMK_XA_FAILED, failed, PCMK_XA_DESCRIPTION, desc, NULL); pcmk__assert(rc == pcmk_rc_ok); } id = pcmk__itoa(replica->offset); rc = pe__name_and_nvpairs_xml(out, true, PCMK_XE_REPLICA, PCMK_XA_ID, id, NULL); free(id); pcmk__assert(rc == pcmk_rc_ok); if (print_ip) { out->message(out, (const char *) ip->priv->xml->name, show_opts, ip, only_node, only_rsc); } if (print_child) { out->message(out, (const char *) child->priv->xml->name, show_opts, child, only_node, only_rsc); } if (print_ctnr) { out->message(out, (const char *) container->priv->xml->name, show_opts, container, only_node, only_rsc); } if (print_remote) { out->message(out, (const char *) remote->priv->xml->name, show_opts, remote, only_node, only_rsc); } pcmk__output_xml_pop_parent(out); // replica } if (printed_header) { pcmk__output_xml_pop_parent(out); // bundle } return rc; } static void pe__bundle_replica_output_html(pcmk__output_t *out, pcmk__bundle_replica_t *replica, pcmk_node_t *node, uint32_t show_opts) { pcmk_resource_t *rsc = replica->child; int offset = 0; char buffer[LINE_MAX]; if(rsc == NULL) { rsc = replica->container; } if (replica->remote) { offset += snprintf(buffer + offset, LINE_MAX - offset, "%s", rsc_printable_id(replica->remote)); } else { offset += snprintf(buffer + offset, LINE_MAX - offset, "%s", rsc_printable_id(replica->container)); } if (replica->ipaddr) { offset += snprintf(buffer + offset, LINE_MAX - offset, " (%s)", replica->ipaddr); } pe__common_output_html(out, rsc, buffer, node, show_opts); } /*! * \internal * \brief Get a string describing a resource's unmanaged state or lack thereof * * \param[in] rsc Resource to describe * * \return A string indicating that a resource is in maintenance mode or * otherwise unmanaged, or an empty string otherwise */ static const char * get_unmanaged_str(const pcmk_resource_t *rsc) { if (pcmk_is_set(rsc->flags, pcmk__rsc_maintenance)) { return " (maintenance)"; } if (!pcmk_is_set(rsc->flags, pcmk__rsc_managed)) { return " (unmanaged)"; } return ""; } PCMK__OUTPUT_ARGS("bundle", "uint32_t", "pcmk_resource_t *", "GList *", "GList *") int pe__bundle_html(pcmk__output_t *out, va_list args) { uint32_t show_opts = va_arg(args, uint32_t); pcmk_resource_t *rsc = va_arg(args, pcmk_resource_t *); GList *only_node = va_arg(args, GList *); GList *only_rsc = va_arg(args, GList *); const char *desc = NULL; pe__bundle_variant_data_t *bundle_data = NULL; int rc = pcmk_rc_no_output; gboolean print_everything = TRUE; pcmk__assert(rsc != NULL); get_bundle_variant_data(bundle_data, rsc); desc = pe__resource_description(rsc, show_opts); if (rsc->priv->fns->is_filtered(rsc, only_rsc, TRUE)) { return rc; } print_everything = pcmk__str_in_list(rsc->id, only_rsc, pcmk__str_star_matches); for (GList *gIter = bundle_data->replicas; gIter != NULL; gIter = gIter->next) { pcmk__bundle_replica_t *replica = gIter->data; pcmk_resource_t *ip = replica->ip; pcmk_resource_t *child = replica->child; pcmk_resource_t *container = replica->container; pcmk_resource_t *remote = replica->remote; gboolean print_ip, print_child, print_ctnr, print_remote; pcmk__assert(replica != NULL); if (pcmk__rsc_filtered_by_node(container, only_node)) { continue; } print_ip = (ip != NULL) && !ip->priv->fns->is_filtered(ip, only_rsc, print_everything); print_child = (child != NULL) && !child->priv->fns->is_filtered(child, only_rsc, print_everything); print_ctnr = !container->priv->fns->is_filtered(container, only_rsc, print_everything); print_remote = (remote != NULL) && !remote->priv->fns->is_filtered(remote, only_rsc, print_everything); if (pcmk_is_set(show_opts, pcmk_show_implicit_rscs) || (print_everything == FALSE && (print_ip || print_child || print_ctnr || print_remote))) { /* The text output messages used below require pe_print_implicit to * be set to do anything. */ uint32_t new_show_opts = show_opts | pcmk_show_implicit_rscs; PCMK__OUTPUT_LIST_HEADER(out, FALSE, rc, "Container bundle%s: %s [%s]%s%s%s%s%s", (bundle_data->nreplicas > 1)? " set" : "", rsc->id, bundle_data->image, pcmk_is_set(rsc->flags, pcmk__rsc_unique)? " (unique)" : "", desc ? " (" : "", desc ? desc : "", desc ? ")" : "", get_unmanaged_str(rsc)); if (pcmk__list_of_multiple(bundle_data->replicas)) { out->begin_list(out, NULL, NULL, "Replica[%d]", replica->offset); } if (print_ip) { out->message(out, (const char *) ip->priv->xml->name, new_show_opts, ip, only_node, only_rsc); } if (print_child) { out->message(out, (const char *) child->priv->xml->name, new_show_opts, child, only_node, only_rsc); } if (print_ctnr) { out->message(out, (const char *) container->priv->xml->name, new_show_opts, container, only_node, only_rsc); } if (print_remote) { out->message(out, (const char *) remote->priv->xml->name, new_show_opts, remote, only_node, only_rsc); } if (pcmk__list_of_multiple(bundle_data->replicas)) { out->end_list(out); } } else if (print_everything == FALSE && !(print_ip || print_child || print_ctnr || print_remote)) { continue; } else { PCMK__OUTPUT_LIST_HEADER(out, FALSE, rc, "Container bundle%s: %s [%s]%s%s%s%s%s", (bundle_data->nreplicas > 1)? " set" : "", rsc->id, bundle_data->image, pcmk_is_set(rsc->flags, pcmk__rsc_unique)? " (unique)" : "", desc ? " (" : "", desc ? desc : "", desc ? ")" : "", get_unmanaged_str(rsc)); pe__bundle_replica_output_html(out, replica, pcmk__current_node(container), show_opts); } } PCMK__OUTPUT_LIST_FOOTER(out, rc); return rc; } static void pe__bundle_replica_output_text(pcmk__output_t *out, pcmk__bundle_replica_t *replica, pcmk_node_t *node, uint32_t show_opts) { const pcmk_resource_t *rsc = replica->child; int offset = 0; char buffer[LINE_MAX]; if(rsc == NULL) { rsc = replica->container; } if (replica->remote) { offset += snprintf(buffer + offset, LINE_MAX - offset, "%s", rsc_printable_id(replica->remote)); } else { offset += snprintf(buffer + offset, LINE_MAX - offset, "%s", rsc_printable_id(replica->container)); } if (replica->ipaddr) { offset += snprintf(buffer + offset, LINE_MAX - offset, " (%s)", replica->ipaddr); } pe__common_output_text(out, rsc, buffer, node, show_opts); } PCMK__OUTPUT_ARGS("bundle", "uint32_t", "pcmk_resource_t *", "GList *", "GList *") int pe__bundle_text(pcmk__output_t *out, va_list args) { uint32_t show_opts = va_arg(args, uint32_t); pcmk_resource_t *rsc = va_arg(args, pcmk_resource_t *); GList *only_node = va_arg(args, GList *); GList *only_rsc = va_arg(args, GList *); const char *desc = NULL; pe__bundle_variant_data_t *bundle_data = NULL; int rc = pcmk_rc_no_output; gboolean print_everything = TRUE; desc = pe__resource_description(rsc, show_opts); pcmk__assert(rsc != NULL); get_bundle_variant_data(bundle_data, rsc); if (rsc->priv->fns->is_filtered(rsc, only_rsc, TRUE)) { return rc; } print_everything = pcmk__str_in_list(rsc->id, only_rsc, pcmk__str_star_matches); for (GList *gIter = bundle_data->replicas; gIter != NULL; gIter = gIter->next) { pcmk__bundle_replica_t *replica = gIter->data; pcmk_resource_t *ip = replica->ip; pcmk_resource_t *child = replica->child; pcmk_resource_t *container = replica->container; pcmk_resource_t *remote = replica->remote; gboolean print_ip, print_child, print_ctnr, print_remote; pcmk__assert(replica != NULL); if (pcmk__rsc_filtered_by_node(container, only_node)) { continue; } print_ip = (ip != NULL) && !ip->priv->fns->is_filtered(ip, only_rsc, print_everything); print_child = (child != NULL) && !child->priv->fns->is_filtered(child, only_rsc, print_everything); print_ctnr = !container->priv->fns->is_filtered(container, only_rsc, print_everything); print_remote = (remote != NULL) && !remote->priv->fns->is_filtered(remote, only_rsc, print_everything); if (pcmk_is_set(show_opts, pcmk_show_implicit_rscs) || (print_everything == FALSE && (print_ip || print_child || print_ctnr || print_remote))) { /* The text output messages used below require pe_print_implicit to * be set to do anything. */ uint32_t new_show_opts = show_opts | pcmk_show_implicit_rscs; PCMK__OUTPUT_LIST_HEADER(out, FALSE, rc, "Container bundle%s: %s [%s]%s%s%s%s%s", (bundle_data->nreplicas > 1)? " set" : "", rsc->id, bundle_data->image, pcmk_is_set(rsc->flags, pcmk__rsc_unique)? " (unique)" : "", desc ? " (" : "", desc ? desc : "", desc ? ")" : "", get_unmanaged_str(rsc)); if (pcmk__list_of_multiple(bundle_data->replicas)) { out->list_item(out, NULL, "Replica[%d]", replica->offset); } out->begin_list(out, NULL, NULL, NULL); if (print_ip) { out->message(out, (const char *) ip->priv->xml->name, new_show_opts, ip, only_node, only_rsc); } if (print_child) { out->message(out, (const char *) child->priv->xml->name, new_show_opts, child, only_node, only_rsc); } if (print_ctnr) { out->message(out, (const char *) container->priv->xml->name, new_show_opts, container, only_node, only_rsc); } if (print_remote) { out->message(out, (const char *) remote->priv->xml->name, new_show_opts, remote, only_node, only_rsc); } out->end_list(out); } else if (print_everything == FALSE && !(print_ip || print_child || print_ctnr || print_remote)) { continue; } else { PCMK__OUTPUT_LIST_HEADER(out, FALSE, rc, "Container bundle%s: %s [%s]%s%s%s%s%s", (bundle_data->nreplicas > 1)? " set" : "", rsc->id, bundle_data->image, pcmk_is_set(rsc->flags, pcmk__rsc_unique)? " (unique)" : "", desc ? " (" : "", desc ? desc : "", desc ? ")" : "", get_unmanaged_str(rsc)); pe__bundle_replica_output_text(out, replica, pcmk__current_node(container), show_opts); } } PCMK__OUTPUT_LIST_FOOTER(out, rc); return rc; } static void free_bundle_replica(pcmk__bundle_replica_t *replica) { if (replica == NULL) { return; } - if (replica->node) { - free(replica->node); - replica->node = NULL; - } + pcmk__free_node_copy(replica->node); + replica->node = NULL; if (replica->ip) { pcmk__xml_free(replica->ip->priv->xml); replica->ip->priv->xml = NULL; replica->ip->priv->fns->free(replica->ip); } if (replica->container) { pcmk__xml_free(replica->container->priv->xml); replica->container->priv->xml = NULL; replica->container->priv->fns->free(replica->container); } if (replica->remote) { pcmk__xml_free(replica->remote->priv->xml); replica->remote->priv->xml = NULL; replica->remote->priv->fns->free(replica->remote); } free(replica->ipaddr); free(replica); } void pe__free_bundle(pcmk_resource_t *rsc) { pe__bundle_variant_data_t *bundle_data = NULL; CRM_CHECK(rsc != NULL, return); get_bundle_variant_data(bundle_data, rsc); pcmk__rsc_trace(rsc, "Freeing %s", rsc->id); free(bundle_data->prefix); free(bundle_data->image); free(bundle_data->control_port); free(bundle_data->host_network); free(bundle_data->host_netmask); free(bundle_data->ip_range_start); free(bundle_data->container_network); free(bundle_data->launcher_options); free(bundle_data->container_command); g_free(bundle_data->container_host_options); g_list_free_full(bundle_data->replicas, (GDestroyNotify) free_bundle_replica); g_list_free_full(bundle_data->mounts, (GDestroyNotify)mount_free); g_list_free_full(bundle_data->ports, (GDestroyNotify)port_free); g_list_free(rsc->priv->children); if(bundle_data->child) { pcmk__xml_free(bundle_data->child->priv->xml); bundle_data->child->priv->xml = NULL; bundle_data->child->priv->fns->free(bundle_data->child); } common_free(rsc); } enum rsc_role_e pe__bundle_resource_state(const pcmk_resource_t *rsc, gboolean current) { enum rsc_role_e container_role = pcmk_role_unknown; return container_role; } /*! * \brief Get the number of configured replicas in a bundle * * \param[in] rsc Bundle resource * * \return Number of configured replicas, or 0 on error */ int pe_bundle_replicas(const pcmk_resource_t *rsc) { if (pcmk__is_bundle(rsc)) { pe__bundle_variant_data_t *bundle_data = NULL; get_bundle_variant_data(bundle_data, rsc); return bundle_data->nreplicas; } return 0; } void pe__count_bundle(pcmk_resource_t *rsc) { pe__bundle_variant_data_t *bundle_data = NULL; get_bundle_variant_data(bundle_data, rsc); for (GList *item = bundle_data->replicas; item != NULL; item = item->next) { pcmk__bundle_replica_t *replica = item->data; if (replica->ip) { replica->ip->priv->fns->count(replica->ip); } if (replica->child) { replica->child->priv->fns->count(replica->child); } if (replica->container) { replica->container->priv->fns->count(replica->container); } if (replica->remote) { replica->remote->priv->fns->count(replica->remote); } } } gboolean pe__bundle_is_filtered(const pcmk_resource_t *rsc, GList *only_rsc, gboolean check_parent) { gboolean passes = FALSE; pe__bundle_variant_data_t *bundle_data = NULL; if (pcmk__str_in_list(rsc_printable_id(rsc), only_rsc, pcmk__str_star_matches)) { passes = TRUE; } else { get_bundle_variant_data(bundle_data, rsc); for (GList *gIter = bundle_data->replicas; gIter != NULL; gIter = gIter->next) { pcmk__bundle_replica_t *replica = gIter->data; pcmk_resource_t *ip = replica->ip; pcmk_resource_t *child = replica->child; pcmk_resource_t *container = replica->container; pcmk_resource_t *remote = replica->remote; if ((ip != NULL) && !ip->priv->fns->is_filtered(ip, only_rsc, FALSE)) { passes = TRUE; break; } if ((child != NULL) && !child->priv->fns->is_filtered(child, only_rsc, FALSE)) { passes = TRUE; break; } if (!container->priv->fns->is_filtered(container, only_rsc, FALSE)) { passes = TRUE; break; } if ((remote != NULL) && !remote->priv->fns->is_filtered(remote, only_rsc, FALSE)) { passes = TRUE; break; } } } return !passes; } /*! * \internal * \brief Get a list of a bundle's containers * * \param[in] bundle Bundle resource * * \return Newly created list of \p bundle's containers * \note It is the caller's responsibility to free the result with * g_list_free(). */ GList * pe__bundle_containers(const pcmk_resource_t *bundle) { GList *containers = NULL; const pe__bundle_variant_data_t *data = NULL; get_bundle_variant_data(data, bundle); for (GList *iter = data->replicas; iter != NULL; iter = iter->next) { pcmk__bundle_replica_t *replica = iter->data; containers = g_list_append(containers, replica->container); } return containers; } // Bundle implementation of pcmk__rsc_methods_t:active_node() pcmk_node_t * pe__bundle_active_node(const pcmk_resource_t *rsc, unsigned int *count_all, unsigned int *count_clean) { pcmk_node_t *active = NULL; pcmk_node_t *node = NULL; pcmk_resource_t *container = NULL; GList *containers = NULL; GList *iter = NULL; GHashTable *nodes = NULL; const pe__bundle_variant_data_t *data = NULL; if (count_all != NULL) { *count_all = 0; } if (count_clean != NULL) { *count_clean = 0; } if (rsc == NULL) { return NULL; } /* For the purposes of this method, we only care about where the bundle's * containers are active, so build a list of active containers. */ get_bundle_variant_data(data, rsc); for (iter = data->replicas; iter != NULL; iter = iter->next) { pcmk__bundle_replica_t *replica = iter->data; if (replica->container->priv->active_nodes != NULL) { containers = g_list_append(containers, replica->container); } } if (containers == NULL) { return NULL; } /* If the bundle has only a single active container, just use that * container's method. If live migration is ever supported for bundle * containers, this will allow us to prefer the migration source when there * is only one container and it is migrating. For now, this just lets us * avoid creating the nodes table. */ if (pcmk__list_of_1(containers)) { container = containers->data; node = container->priv->fns->active_node(container, count_all, count_clean); g_list_free(containers); return node; } // Add all containers' active nodes to a hash table (for uniqueness) nodes = g_hash_table_new(NULL, NULL); for (iter = containers; iter != NULL; iter = iter->next) { container = iter->data; for (GList *node_iter = container->priv->active_nodes; node_iter != NULL; node_iter = node_iter->next) { node = node_iter->data; // If insert returns true, we haven't counted this node yet if (g_hash_table_insert(nodes, (gpointer) node->details, (gpointer) node) && !pe__count_active_node(rsc, node, &active, count_all, count_clean)) { goto done; } } } done: g_list_free(containers); g_hash_table_destroy(nodes); return active; } /*! * \internal * \brief Get maximum bundle resource instances per node * * \param[in] rsc Bundle resource to check * * \return Maximum number of \p rsc instances that can be active on one node */ unsigned int pe__bundle_max_per_node(const pcmk_resource_t *rsc) { pe__bundle_variant_data_t *bundle_data = NULL; get_bundle_variant_data(bundle_data, rsc); pcmk__assert(bundle_data->nreplicas_per_host >= 0); return (unsigned int) bundle_data->nreplicas_per_host; } diff --git a/lib/pengine/complex.c b/lib/pengine/complex.c index d47a3ad8c7..2e9ec18574 100644 --- a/lib/pengine/complex.c +++ b/lib/pengine/complex.c @@ -1,1274 +1,1274 @@ /* * Copyright 2004-2024 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" void populate_hash(xmlNode * nvpair_list, GHashTable * hash, const char **attrs, int attrs_length); static pcmk_node_t *active_node(const pcmk_resource_t *rsc, unsigned int *count_all, unsigned int *count_clean); static pcmk__rsc_methods_t resource_class_functions[] = { { native_unpack, native_find_rsc, native_parameter, native_active, native_resource_state, native_location, native_free, pe__count_common, pe__native_is_filtered, active_node, pe__primitive_max_per_node, }, { group_unpack, native_find_rsc, native_parameter, group_active, group_resource_state, native_location, group_free, pe__count_common, pe__group_is_filtered, active_node, pe__group_max_per_node, }, { clone_unpack, native_find_rsc, native_parameter, clone_active, clone_resource_state, native_location, clone_free, pe__count_common, pe__clone_is_filtered, active_node, pe__clone_max_per_node, }, { pe__unpack_bundle, native_find_rsc, native_parameter, pe__bundle_active, pe__bundle_resource_state, native_location, pe__free_bundle, pe__count_bundle, pe__bundle_is_filtered, pe__bundle_active_node, pe__bundle_max_per_node, } }; static enum pcmk__rsc_variant get_resource_type(const char *name) { if (pcmk__str_eq(name, PCMK_XE_PRIMITIVE, pcmk__str_casei)) { return pcmk__rsc_variant_primitive; } else if (pcmk__str_eq(name, PCMK_XE_GROUP, pcmk__str_casei)) { return pcmk__rsc_variant_group; } else if (pcmk__str_eq(name, PCMK_XE_CLONE, pcmk__str_casei)) { return pcmk__rsc_variant_clone; } else if (pcmk__str_eq(name, PCMK_XE_BUNDLE, pcmk__str_casei)) { return pcmk__rsc_variant_bundle; } return pcmk__rsc_variant_unknown; } /*! * \internal * \brief Insert a meta-attribute if not already present * * \param[in] key Meta-attribute name * \param[in] value Meta-attribute value to add if not already present * \param[in,out] table Meta-attribute hash table to insert into * * \note This is like pcmk__insert_meta() except it won't overwrite existing * values. */ static void dup_attr(gpointer key, gpointer value, gpointer user_data) { GHashTable *table = user_data; CRM_CHECK((key != NULL) && (table != NULL), return); if (pcmk__str_eq((const char *) value, "#default", pcmk__str_casei)) { // @COMPAT Deprecated since 2.1.8 pcmk__config_warn("Support for setting meta-attributes (such as %s) to " "the explicit value '#default' is deprecated and " "will be removed in a future release", (const char *) key); } else if ((value != NULL) && (g_hash_table_lookup(table, key) == NULL)) { pcmk__insert_dup(table, (const char *) key, (const char *) value); } } static void expand_parents_fixed_nvpairs(pcmk_resource_t *rsc, pe_rule_eval_data_t *rule_data, GHashTable *meta_hash, pcmk_scheduler_t *scheduler) { GHashTable *parent_orig_meta = pcmk__strkey_table(free, free); pcmk_resource_t *p = rsc->priv->parent; if (p == NULL) { return ; } /* Search all parent resources, get the fixed value of * PCMK_XE_META_ATTRIBUTES set only in the original xml, and stack it in the * hash table. The fixed value of the lower parent resource takes precedence * and is not overwritten. */ while(p != NULL) { /* A hash table for comparison is generated, including the id-ref. */ pe__unpack_dataset_nvpairs(p->priv->xml, PCMK_XE_META_ATTRIBUTES, rule_data, parent_orig_meta, NULL, scheduler); p = p->priv->parent; } if (parent_orig_meta != NULL) { // This will not overwrite any values already existing for child g_hash_table_foreach(parent_orig_meta, dup_attr, meta_hash); } if (parent_orig_meta != NULL) { g_hash_table_destroy(parent_orig_meta); } return ; } /* * \brief Get fully evaluated resource meta-attributes * * \param[in,out] meta_hash Where to store evaluated meta-attributes * \param[in] rsc Resource to get meta-attributes for * \param[in] node Ignored * \param[in,out] scheduler Scheduler data */ void get_meta_attributes(GHashTable * meta_hash, pcmk_resource_t * rsc, pcmk_node_t *node, pcmk_scheduler_t *scheduler) { pe_rsc_eval_data_t rsc_rule_data = { .standard = crm_element_value(rsc->priv->xml, PCMK_XA_CLASS), .provider = crm_element_value(rsc->priv->xml, PCMK_XA_PROVIDER), .agent = crm_element_value(rsc->priv->xml, PCMK_XA_TYPE) }; pe_rule_eval_data_t rule_data = { .node_hash = NULL, .now = scheduler->priv->now, .match_data = NULL, .rsc_data = &rsc_rule_data, .op_data = NULL }; for (xmlAttrPtr a = pcmk__xe_first_attr(rsc->priv->xml); a != NULL; a = a->next) { if (a->children != NULL) { dup_attr((gpointer) a->name, (gpointer) a->children->content, meta_hash); } } pe__unpack_dataset_nvpairs(rsc->priv->xml, PCMK_XE_META_ATTRIBUTES, &rule_data, meta_hash, NULL, scheduler); /* Set the PCMK_XE_META_ATTRIBUTES explicitly set in the parent resource to * the hash table of the child resource. If it is already explicitly set as * a child, it will not be overwritten. */ if (rsc->priv->parent != NULL) { expand_parents_fixed_nvpairs(rsc, &rule_data, meta_hash, scheduler); } /* check the defaults */ pe__unpack_dataset_nvpairs(scheduler->priv->rsc_defaults, PCMK_XE_META_ATTRIBUTES, &rule_data, meta_hash, NULL, scheduler); /* If there is PCMK_XE_META_ATTRIBUTES that the parent resource has not * explicitly set, set a value that is not set from PCMK_XE_RSC_DEFAULTS * either. The values already set up to this point will not be overwritten. */ if (rsc->priv->parent != NULL) { g_hash_table_foreach(rsc->priv->parent->priv->meta, dup_attr, meta_hash); } } /*! * \brief Get final values of a resource's instance attributes * * \param[in,out] instance_attrs Where to store the instance attributes * \param[in] rsc Resource to get instance attributes for * \param[in] node If not NULL, evaluate rules for this node * \param[in,out] scheduler Scheduler data */ void get_rsc_attributes(GHashTable *instance_attrs, const pcmk_resource_t *rsc, const pcmk_node_t *node, pcmk_scheduler_t *scheduler) { pe_rule_eval_data_t rule_data = { .node_hash = NULL, .now = NULL, .match_data = NULL, .rsc_data = NULL, .op_data = NULL }; CRM_CHECK((instance_attrs != NULL) && (rsc != NULL) && (scheduler != NULL), return); rule_data.now = scheduler->priv->now; if (node != NULL) { rule_data.node_hash = node->priv->attrs; } // Evaluate resource's own values, then its ancestors' values pe__unpack_dataset_nvpairs(rsc->priv->xml, PCMK_XE_INSTANCE_ATTRIBUTES, &rule_data, instance_attrs, NULL, scheduler); if (rsc->priv->parent != NULL) { get_rsc_attributes(instance_attrs, rsc->priv->parent, node, scheduler); } } static char * template_op_key(xmlNode * op) { const char *name = crm_element_value(op, PCMK_XA_NAME); const char *role = crm_element_value(op, PCMK_XA_ROLE); char *key = NULL; if ((role == NULL) || pcmk__strcase_any_of(role, PCMK_ROLE_STARTED, PCMK_ROLE_UNPROMOTED, PCMK__ROLE_UNPROMOTED_LEGACY, NULL)) { role = PCMK__ROLE_UNKNOWN; } key = crm_strdup_printf("%s-%s", name, role); return key; } static gboolean unpack_template(xmlNode *xml_obj, xmlNode **expanded_xml, pcmk_scheduler_t *scheduler) { xmlNode *cib_resources = NULL; xmlNode *template = NULL; xmlNode *new_xml = NULL; xmlNode *child_xml = NULL; xmlNode *rsc_ops = NULL; xmlNode *template_ops = NULL; const char *template_ref = NULL; const char *id = NULL; if (xml_obj == NULL) { pcmk__config_err("No resource object for template unpacking"); return FALSE; } template_ref = crm_element_value(xml_obj, PCMK_XA_TEMPLATE); if (template_ref == NULL) { return TRUE; } id = pcmk__xe_id(xml_obj); if (id == NULL) { pcmk__config_err("'%s' object must have a id", xml_obj->name); return FALSE; } if (pcmk__str_eq(template_ref, id, pcmk__str_none)) { pcmk__config_err("The resource object '%s' should not reference itself", id); return FALSE; } cib_resources = get_xpath_object("//" PCMK_XE_RESOURCES, scheduler->input, LOG_TRACE); if (cib_resources == NULL) { pcmk__config_err("No resources configured"); return FALSE; } template = pcmk__xe_first_child(cib_resources, PCMK_XE_TEMPLATE, PCMK_XA_ID, template_ref); if (template == NULL) { pcmk__config_err("No template named '%s'", template_ref); return FALSE; } new_xml = pcmk__xml_copy(NULL, template); xmlNodeSetName(new_xml, xml_obj->name); crm_xml_add(new_xml, PCMK_XA_ID, id); crm_xml_add(new_xml, PCMK__META_CLONE, crm_element_value(xml_obj, PCMK__META_CLONE)); template_ops = pcmk__xe_first_child(new_xml, PCMK_XE_OPERATIONS, NULL, NULL); for (child_xml = pcmk__xe_first_child(xml_obj, NULL, NULL, NULL); child_xml != NULL; child_xml = pcmk__xe_next(child_xml, NULL)) { xmlNode *new_child = pcmk__xml_copy(new_xml, child_xml); if (pcmk__xe_is(new_child, PCMK_XE_OPERATIONS)) { rsc_ops = new_child; } } if (template_ops && rsc_ops) { xmlNode *op = NULL; GHashTable *rsc_ops_hash = pcmk__strkey_table(free, NULL); for (op = pcmk__xe_first_child(rsc_ops, NULL, NULL, NULL); op != NULL; op = pcmk__xe_next(op, NULL)) { char *key = template_op_key(op); g_hash_table_insert(rsc_ops_hash, key, op); } for (op = pcmk__xe_first_child(template_ops, NULL, NULL, NULL); op != NULL; op = pcmk__xe_next(op, NULL)) { char *key = template_op_key(op); if (g_hash_table_lookup(rsc_ops_hash, key) == NULL) { pcmk__xml_copy(rsc_ops, op); } free(key); } if (rsc_ops_hash) { g_hash_table_destroy(rsc_ops_hash); } pcmk__xml_free(template_ops); } /*pcmk__xml_free(*expanded_xml); */ *expanded_xml = new_xml; #if 0 /* Disable multi-level templates for now */ if (!unpack_template(new_xml, expanded_xml, scheduler)) { pcmk__xml_free(*expanded_xml); *expanded_xml = NULL; return FALSE; } #endif return TRUE; } static gboolean add_template_rsc(xmlNode *xml_obj, pcmk_scheduler_t *scheduler) { const char *template_ref = NULL; const char *id = NULL; if (xml_obj == NULL) { pcmk__config_err("No resource object for processing resource list " "of template"); return FALSE; } template_ref = crm_element_value(xml_obj, PCMK_XA_TEMPLATE); if (template_ref == NULL) { return TRUE; } id = pcmk__xe_id(xml_obj); if (id == NULL) { pcmk__config_err("'%s' object must have a id", xml_obj->name); return FALSE; } if (pcmk__str_eq(template_ref, id, pcmk__str_none)) { pcmk__config_err("The resource object '%s' should not reference itself", id); return FALSE; } pcmk__add_idref(scheduler->priv->templates, template_ref, id); return TRUE; } /*! * \internal * \brief Check whether a clone or instance being unpacked is globally unique * * \param[in] rsc Clone or clone instance to check * * \return \c true if \p rsc is globally unique according to its * meta-attributes, otherwise \c false */ static bool detect_unique(const pcmk_resource_t *rsc) { const char *value = g_hash_table_lookup(rsc->priv->meta, PCMK_META_GLOBALLY_UNIQUE); if (value == NULL) { // Default to true if clone-node-max > 1 value = g_hash_table_lookup(rsc->priv->meta, PCMK_META_CLONE_NODE_MAX); if (value != NULL) { int node_max = 1; if ((pcmk__scan_min_int(value, &node_max, 0) == pcmk_rc_ok) && (node_max > 1)) { return true; } } return false; } return crm_is_true(value); } static void free_params_table(gpointer data) { g_hash_table_destroy((GHashTable *) data); } /*! * \brief Get a table of resource parameters * * \param[in,out] rsc Resource to query * \param[in] node Node for evaluating rules (NULL for defaults) * \param[in,out] scheduler Scheduler data * * \return Hash table containing resource parameter names and values * (or NULL if \p rsc or \p scheduler is NULL) * \note The returned table will be destroyed when the resource is freed, so * callers should not destroy it. */ GHashTable * pe_rsc_params(pcmk_resource_t *rsc, const pcmk_node_t *node, pcmk_scheduler_t *scheduler) { GHashTable *params_on_node = NULL; /* A NULL node is used to request the resource's default parameters * (not evaluated for node), but we always want something non-NULL * as a hash table key. */ const char *node_name = ""; // Sanity check if ((rsc == NULL) || (scheduler == NULL)) { return NULL; } if ((node != NULL) && (node->priv->name != NULL)) { node_name = node->priv->name; } // Find the parameter table for given node if (rsc->priv->parameter_cache == NULL) { rsc->priv->parameter_cache = pcmk__strikey_table(free, free_params_table); } else { params_on_node = g_hash_table_lookup(rsc->priv->parameter_cache, node_name); } // If none exists yet, create one with parameters evaluated for node if (params_on_node == NULL) { params_on_node = pcmk__strkey_table(free, free); get_rsc_attributes(params_on_node, rsc, node, scheduler); g_hash_table_insert(rsc->priv->parameter_cache, strdup(node_name), params_on_node); } return params_on_node; } /*! * \internal * \brief Unpack a resource's \c PCMK_META_REQUIRES meta-attribute * * \param[in,out] rsc Resource being unpacked * \param[in] value Value of \c PCMK_META_REQUIRES meta-attribute * \param[in] is_default Whether \p value was selected by default */ static void unpack_requires(pcmk_resource_t *rsc, const char *value, bool is_default) { const pcmk_scheduler_t *scheduler = rsc->priv->scheduler; if (pcmk__str_eq(value, PCMK_VALUE_NOTHING, pcmk__str_casei)) { } else if (pcmk__str_eq(value, PCMK_VALUE_QUORUM, pcmk__str_casei)) { pcmk__set_rsc_flags(rsc, pcmk__rsc_needs_quorum); } else if (pcmk__str_eq(value, PCMK_VALUE_FENCING, pcmk__str_casei)) { pcmk__set_rsc_flags(rsc, pcmk__rsc_needs_fencing); if (!pcmk_is_set(scheduler->flags, pcmk__sched_fencing_enabled)) { pcmk__config_warn("%s requires fencing but fencing is disabled", rsc->id); } } else if (pcmk__str_eq(value, PCMK_VALUE_UNFENCING, pcmk__str_casei)) { if (pcmk_is_set(rsc->flags, pcmk__rsc_fence_device)) { pcmk__config_warn("Resetting \"" PCMK_META_REQUIRES "\" for %s " "to \"" PCMK_VALUE_QUORUM "\" because fencing " "devices cannot require unfencing", rsc->id); unpack_requires(rsc, PCMK_VALUE_QUORUM, true); return; } else if (!pcmk_is_set(scheduler->flags, pcmk__sched_fencing_enabled)) { pcmk__config_warn("Resetting \"" PCMK_META_REQUIRES "\" for %s " "to \"" PCMK_VALUE_QUORUM "\" because fencing is " "disabled", rsc->id); unpack_requires(rsc, PCMK_VALUE_QUORUM, true); return; } else { pcmk__set_rsc_flags(rsc, pcmk__rsc_needs_fencing |pcmk__rsc_needs_unfencing); } } else { const char *orig_value = value; if (pcmk_is_set(rsc->flags, pcmk__rsc_fence_device)) { value = PCMK_VALUE_QUORUM; } else if (pcmk__is_primitive(rsc) && xml_contains_remote_node(rsc->priv->xml)) { value = PCMK_VALUE_QUORUM; } else if (pcmk_is_set(scheduler->flags, pcmk__sched_enable_unfencing)) { value = PCMK_VALUE_UNFENCING; } else if (pcmk_is_set(scheduler->flags, pcmk__sched_fencing_enabled)) { value = PCMK_VALUE_FENCING; } else if (scheduler->no_quorum_policy == pcmk_no_quorum_ignore) { value = PCMK_VALUE_NOTHING; } else { value = PCMK_VALUE_QUORUM; } if (orig_value != NULL) { pcmk__config_err("Resetting '" PCMK_META_REQUIRES "' for %s " "to '%s' because '%s' is not valid", rsc->id, value, orig_value); } unpack_requires(rsc, value, true); return; } pcmk__rsc_trace(rsc, "\tRequired to start: %s%s", value, (is_default? " (default)" : "")); } /*! * \internal * \brief Parse resource priority from meta-attribute * * \param[in,out] rsc Resource being unpacked */ static void unpack_priority(pcmk_resource_t *rsc) { const char *value = g_hash_table_lookup(rsc->priv->meta, PCMK_META_PRIORITY); int rc = pcmk_parse_score(value, &(rsc->priv->priority), 0); if (rc != pcmk_rc_ok) { pcmk__config_warn("Using default (0) for resource %s " PCMK_META_PRIORITY " because '%s' is not a valid value: %s", rsc->id, value, pcmk_rc_str(rc)); } } /*! * \internal * \brief Parse resource stickiness from meta-attribute * * \param[in,out] rsc Resource being unpacked */ static void unpack_stickiness(pcmk_resource_t *rsc) { const char *value = g_hash_table_lookup(rsc->priv->meta, PCMK_META_RESOURCE_STICKINESS); if (pcmk__str_eq(value, PCMK_VALUE_DEFAULT, pcmk__str_casei)) { // @COMPAT Deprecated since 2.1.8 pcmk__config_warn("Support for setting " PCMK_META_RESOURCE_STICKINESS " to the explicit value '" PCMK_VALUE_DEFAULT "' is deprecated and will be removed in a " "future release (just leave it unset)"); } else { int rc = pcmk_parse_score(value, &(rsc->priv->stickiness), 0); if (rc != pcmk_rc_ok) { pcmk__config_warn("Using default (0) for resource %s " PCMK_META_RESOURCE_STICKINESS " because '%s' is not a valid value: %s", rsc->id, value, pcmk_rc_str(rc)); } } } /*! * \internal * \brief Parse resource migration threshold from meta-attribute * * \param[in,out] rsc Resource being unpacked */ static void unpack_migration_threshold(pcmk_resource_t *rsc) { const char *value = g_hash_table_lookup(rsc->priv->meta, PCMK_META_MIGRATION_THRESHOLD); if (pcmk__str_eq(value, PCMK_VALUE_DEFAULT, pcmk__str_casei)) { // @COMPAT Deprecated since 2.1.8 pcmk__config_warn("Support for setting " PCMK_META_MIGRATION_THRESHOLD " to the explicit value '" PCMK_VALUE_DEFAULT "' is deprecated and will be removed in a " "future release (just leave it unset)"); rsc->priv->ban_after_failures = PCMK_SCORE_INFINITY; } else { int rc = pcmk_parse_score(value, &(rsc->priv->ban_after_failures), PCMK_SCORE_INFINITY); if ((rc != pcmk_rc_ok) || (rsc->priv->ban_after_failures < 0)) { pcmk__config_warn("Using default (" PCMK_VALUE_INFINITY ") for resource %s meta-attribute " PCMK_META_MIGRATION_THRESHOLD " because '%s' is not a valid value: %s", rsc->id, value, pcmk_rc_str(rc)); rsc->priv->ban_after_failures = PCMK_SCORE_INFINITY; } } } /*! * \internal * \brief Unpack configuration XML for a given resource * * Unpack the XML object containing a resource's configuration into a new * \c pcmk_resource_t object. * * \param[in] xml_obj XML node containing the resource's configuration * \param[out] rsc Where to store the unpacked resource information * \param[in] parent Resource's parent, if any * \param[in,out] scheduler Scheduler data * * \return Standard Pacemaker return code * \note If pcmk_rc_ok is returned, \p *rsc is guaranteed to be non-NULL, and * the caller is responsible for freeing it using its variant-specific * free() method. Otherwise, \p *rsc is guaranteed to be NULL. */ int pe__unpack_resource(xmlNode *xml_obj, pcmk_resource_t **rsc, pcmk_resource_t *parent, pcmk_scheduler_t *scheduler) { xmlNode *expanded_xml = NULL; xmlNode *ops = NULL; const char *value = NULL; const char *id = NULL; bool guest_node = false; bool remote_node = false; pcmk__resource_private_t *rsc_private = NULL; pe_rule_eval_data_t rule_data = { .node_hash = NULL, .now = NULL, .match_data = NULL, .rsc_data = NULL, .op_data = NULL }; CRM_CHECK(rsc != NULL, return EINVAL); CRM_CHECK((xml_obj != NULL) && (scheduler != NULL), *rsc = NULL; return EINVAL); rule_data.now = scheduler->priv->now; crm_log_xml_trace(xml_obj, "[raw XML]"); id = crm_element_value(xml_obj, PCMK_XA_ID); if (id == NULL) { pcmk__config_err("Ignoring <%s> configuration without " PCMK_XA_ID, xml_obj->name); return pcmk_rc_unpack_error; } if (unpack_template(xml_obj, &expanded_xml, scheduler) == FALSE) { return pcmk_rc_unpack_error; } *rsc = calloc(1, sizeof(pcmk_resource_t)); if (*rsc == NULL) { pcmk__sched_err(scheduler, "Unable to allocate memory for resource '%s'", id); return ENOMEM; } (*rsc)->priv = calloc(1, sizeof(pcmk__resource_private_t)); if ((*rsc)->priv == NULL) { pcmk__sched_err(scheduler, "Unable to allocate memory for resource '%s'", id); free(*rsc); return ENOMEM; } rsc_private = (*rsc)->priv; rsc_private->scheduler = scheduler; if (expanded_xml) { crm_log_xml_trace(expanded_xml, "[expanded XML]"); rsc_private->xml = expanded_xml; rsc_private->orig_xml = xml_obj; } else { rsc_private->xml = xml_obj; rsc_private->orig_xml = NULL; } /* Do not use xml_obj from here on, use (*rsc)->xml in case templates are involved */ rsc_private->parent = parent; ops = pcmk__xe_first_child(rsc_private->xml, PCMK_XE_OPERATIONS, NULL, NULL); rsc_private->ops_xml = pcmk__xe_resolve_idref(ops, scheduler->input); rsc_private->variant = get_resource_type((const char *) rsc_private->xml->name); if (rsc_private->variant == pcmk__rsc_variant_unknown) { pcmk__config_err("Ignoring resource '%s' of unknown type '%s'", id, rsc_private->xml->name); common_free(*rsc); *rsc = NULL; return pcmk_rc_unpack_error; } rsc_private->meta = pcmk__strkey_table(free, free); rsc_private->utilization = pcmk__strkey_table(free, free); - rsc_private->probed_nodes = pcmk__strkey_table(NULL, free); - rsc_private->allowed_nodes = pcmk__strkey_table(NULL, free); + rsc_private->probed_nodes = pcmk__strkey_table(NULL, pcmk__free_node_copy); + rsc_private->allowed_nodes = pcmk__strkey_table(NULL, pcmk__free_node_copy); value = crm_element_value(rsc_private->xml, PCMK__META_CLONE); if (value) { (*rsc)->id = crm_strdup_printf("%s:%s", id, value); pcmk__insert_meta(rsc_private, PCMK__META_CLONE, value); } else { (*rsc)->id = strdup(id); } rsc_private->fns = &resource_class_functions[rsc_private->variant]; get_meta_attributes(rsc_private->meta, *rsc, NULL, scheduler); (*rsc)->flags = 0; pcmk__set_rsc_flags(*rsc, pcmk__rsc_unassigned); if (!pcmk_is_set(scheduler->flags, pcmk__sched_in_maintenance)) { pcmk__set_rsc_flags(*rsc, pcmk__rsc_managed); } rsc_private->orig_role = pcmk_role_stopped; rsc_private->next_role = pcmk_role_unknown; unpack_priority(*rsc); value = g_hash_table_lookup(rsc_private->meta, PCMK_META_CRITICAL); if ((value == NULL) || crm_is_true(value)) { pcmk__set_rsc_flags(*rsc, pcmk__rsc_critical); } value = g_hash_table_lookup(rsc_private->meta, PCMK_META_NOTIFY); if (crm_is_true(value)) { pcmk__set_rsc_flags(*rsc, pcmk__rsc_notify); } if (xml_contains_remote_node(rsc_private->xml)) { pcmk__set_rsc_flags(*rsc, pcmk__rsc_is_remote_connection); if (g_hash_table_lookup(rsc_private->meta, PCMK__META_CONTAINER)) { guest_node = true; } else { remote_node = true; } } value = g_hash_table_lookup(rsc_private->meta, PCMK_META_ALLOW_MIGRATE); if (crm_is_true(value)) { pcmk__set_rsc_flags(*rsc, pcmk__rsc_migratable); } else if ((value == NULL) && remote_node) { /* By default, we want remote nodes to be able * to float around the cluster without having to stop all the * resources within the remote-node before moving. Allowing * migration support enables this feature. If this ever causes * problems, migration support can be explicitly turned off with * PCMK_META_ALLOW_MIGRATE=false. */ pcmk__set_rsc_flags(*rsc, pcmk__rsc_migratable); } value = g_hash_table_lookup(rsc_private->meta, PCMK_META_IS_MANAGED); if (value != NULL) { 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_IS_MANAGED " to the explicit value '" PCMK_VALUE_DEFAULT "' is deprecated and will be removed in a " "future release (just leave it unset)"); } else if (crm_is_true(value)) { pcmk__set_rsc_flags(*rsc, pcmk__rsc_managed); } else { pcmk__clear_rsc_flags(*rsc, pcmk__rsc_managed); } } value = g_hash_table_lookup(rsc_private->meta, PCMK_META_MAINTENANCE); if (crm_is_true(value)) { pcmk__clear_rsc_flags(*rsc, pcmk__rsc_managed); pcmk__set_rsc_flags(*rsc, pcmk__rsc_maintenance); } if (pcmk_is_set(scheduler->flags, pcmk__sched_in_maintenance)) { pcmk__clear_rsc_flags(*rsc, pcmk__rsc_managed); pcmk__set_rsc_flags(*rsc, pcmk__rsc_maintenance); } if (pcmk__is_clone(pe__const_top_resource(*rsc, false))) { if (detect_unique(*rsc)) { pcmk__set_rsc_flags(*rsc, pcmk__rsc_unique); } if (crm_is_true(g_hash_table_lookup((*rsc)->priv->meta, PCMK_META_PROMOTABLE))) { pcmk__set_rsc_flags(*rsc, pcmk__rsc_promotable); } } else { pcmk__set_rsc_flags(*rsc, pcmk__rsc_unique); } value = g_hash_table_lookup(rsc_private->meta, PCMK_META_MULTIPLE_ACTIVE); if (pcmk__str_eq(value, PCMK_VALUE_STOP_ONLY, pcmk__str_casei)) { rsc_private->multiply_active_policy = pcmk__multiply_active_stop; pcmk__rsc_trace(*rsc, "%s multiple running resource recovery: stop only", (*rsc)->id); } else if (pcmk__str_eq(value, PCMK_VALUE_BLOCK, pcmk__str_casei)) { rsc_private->multiply_active_policy = pcmk__multiply_active_block; pcmk__rsc_trace(*rsc, "%s multiple running resource recovery: block", (*rsc)->id); } else if (pcmk__str_eq(value, PCMK_VALUE_STOP_UNEXPECTED, pcmk__str_casei)) { rsc_private->multiply_active_policy = pcmk__multiply_active_unexpected; pcmk__rsc_trace(*rsc, "%s multiple running resource recovery: " "stop unexpected instances", (*rsc)->id); } else { // PCMK_VALUE_STOP_START if (!pcmk__str_eq(value, PCMK_VALUE_STOP_START, pcmk__str_casei|pcmk__str_null_matches)) { pcmk__config_warn("%s is not a valid value for " PCMK_META_MULTIPLE_ACTIVE ", using default of " "\"" PCMK_VALUE_STOP_START "\"", value); } rsc_private->multiply_active_policy = pcmk__multiply_active_restart; pcmk__rsc_trace(*rsc, "%s multiple running resource recovery: stop/start", (*rsc)->id); } unpack_stickiness(*rsc); unpack_migration_threshold(*rsc); if (pcmk__str_eq(crm_element_value(rsc_private->xml, PCMK_XA_CLASS), PCMK_RESOURCE_CLASS_STONITH, pcmk__str_casei)) { pcmk__set_scheduler_flags(scheduler, pcmk__sched_have_fencing); pcmk__set_rsc_flags(*rsc, pcmk__rsc_fence_device); } value = g_hash_table_lookup(rsc_private->meta, PCMK_META_REQUIRES); unpack_requires(*rsc, value, false); value = g_hash_table_lookup(rsc_private->meta, PCMK_META_FAILURE_TIMEOUT); if (value != NULL) { pcmk_parse_interval_spec(value, &(rsc_private->failure_expiration_ms)); } if (remote_node) { GHashTable *params = pe_rsc_params(*rsc, NULL, scheduler); /* Grabbing the value now means that any rules based on node attributes * will evaluate to false, so such rules should not be used with * PCMK_REMOTE_RA_RECONNECT_INTERVAL. * * @TODO Evaluate per node before using */ value = g_hash_table_lookup(params, PCMK_REMOTE_RA_RECONNECT_INTERVAL); if (value) { /* reconnect delay works by setting failure_timeout and preventing the * connection from starting until the failure is cleared. */ pcmk_parse_interval_spec(value, &(rsc_private->remote_reconnect_ms)); /* We want to override any default failure_timeout in use when remote * PCMK_REMOTE_RA_RECONNECT_INTERVAL is in use. */ rsc_private->failure_expiration_ms = rsc_private->remote_reconnect_ms; } } get_target_role(*rsc, &(rsc_private->next_role)); pcmk__rsc_trace(*rsc, "%s desired next state: %s", (*rsc)->id, (rsc_private->next_role == pcmk_role_unknown)? "default" : pcmk_role_text(rsc_private->next_role)); if (rsc_private->fns->unpack(*rsc, scheduler) == FALSE) { rsc_private->fns->free(*rsc); *rsc = NULL; return pcmk_rc_unpack_error; } if (pcmk_is_set(scheduler->flags, pcmk__sched_symmetric_cluster)) { // This tag must stay exactly the same because it is tested elsewhere resource_location(*rsc, NULL, 0, "symmetric_default", scheduler); } else if (guest_node) { /* remote resources tied to a container resource must always be allowed * to opt-in to the cluster. Whether the connection resource is actually * allowed to be placed on a node is dependent on the container resource */ resource_location(*rsc, NULL, 0, "remote_connection_default", scheduler); } pcmk__rsc_trace(*rsc, "%s action notification: %s", (*rsc)->id, pcmk_is_set((*rsc)->flags, pcmk__rsc_notify)? "required" : "not required"); pe__unpack_dataset_nvpairs(rsc_private->xml, PCMK_XE_UTILIZATION, &rule_data, rsc_private->utilization, NULL, scheduler); if (expanded_xml) { if (add_template_rsc(xml_obj, scheduler) == FALSE) { rsc_private->fns->free(*rsc); *rsc = NULL; return pcmk_rc_unpack_error; } } return pcmk_rc_ok; } gboolean is_parent(pcmk_resource_t *child, pcmk_resource_t *rsc) { pcmk_resource_t *parent = child; if (parent == NULL || rsc == NULL) { return FALSE; } while (parent->priv->parent != NULL) { if (parent->priv->parent == rsc) { return TRUE; } parent = parent->priv->parent; } return FALSE; } pcmk_resource_t * uber_parent(pcmk_resource_t *rsc) { pcmk_resource_t *parent = rsc; if (parent == NULL) { return NULL; } while ((parent->priv->parent != NULL) && !pcmk__is_bundle(parent->priv->parent)) { parent = parent->priv->parent; } return parent; } /*! * \internal * \brief Get the topmost parent of a resource as a const pointer * * \param[in] rsc Resource to check * \param[in] include_bundle If true, go all the way to bundle * * \return \p NULL if \p rsc is NULL, \p rsc if \p rsc has no parent, * the bundle if \p rsc is bundled and \p include_bundle is true, * otherwise the topmost parent of \p rsc up to a clone */ const pcmk_resource_t * pe__const_top_resource(const pcmk_resource_t *rsc, bool include_bundle) { const pcmk_resource_t *parent = rsc; if (parent == NULL) { return NULL; } while (parent->priv->parent != NULL) { if (!include_bundle && pcmk__is_bundle(parent->priv->parent)) { break; } parent = parent->priv->parent; } return parent; } void common_free(pcmk_resource_t * rsc) { if (rsc == NULL) { return; } pcmk__rsc_trace(rsc, "Freeing %s", rsc->id); if (rsc->priv->parameter_cache != NULL) { g_hash_table_destroy(rsc->priv->parameter_cache); } if ((rsc->priv->parent == NULL) && pcmk_is_set(rsc->flags, pcmk__rsc_removed)) { pcmk__xml_free(rsc->priv->xml); rsc->priv->xml = NULL; pcmk__xml_free(rsc->priv->orig_xml); rsc->priv->orig_xml = NULL; } else if (rsc->priv->orig_xml != NULL) { // rsc->private->xml was expanded from a template pcmk__xml_free(rsc->priv->xml); rsc->priv->xml = NULL; } free(rsc->id); free(rsc->priv->variant_opaque); free(rsc->priv->history_id); free(rsc->priv->pending_action); - free(rsc->priv->assigned_node); + pcmk__free_node_copy(rsc->priv->assigned_node); g_list_free(rsc->priv->actions); g_list_free(rsc->priv->active_nodes); g_list_free(rsc->priv->launched); g_list_free(rsc->priv->dangling_migration_sources); g_list_free(rsc->priv->with_this_colocations); g_list_free(rsc->priv->this_with_colocations); g_list_free(rsc->priv->location_constraints); g_list_free(rsc->priv->ticket_constraints); if (rsc->priv->meta != NULL) { g_hash_table_destroy(rsc->priv->meta); } if (rsc->priv->utilization != NULL) { g_hash_table_destroy(rsc->priv->utilization); } if (rsc->priv->probed_nodes != NULL) { g_hash_table_destroy(rsc->priv->probed_nodes); } if (rsc->priv->allowed_nodes != NULL) { g_hash_table_destroy(rsc->priv->allowed_nodes); } free(rsc->priv); free(rsc); } /*! * \internal * \brief Count a node and update most preferred to it as appropriate * * \param[in] rsc An active resource * \param[in] node A node that \p rsc is active on * \param[in,out] active This will be set to \p node if \p node is more * preferred than the current value * \param[in,out] count_all If not NULL, this will be incremented * \param[in,out] count_clean If not NULL, this will be incremented if \p node * is online and clean * * \return true if the count should continue, or false if sufficiently known */ bool pe__count_active_node(const pcmk_resource_t *rsc, pcmk_node_t *node, pcmk_node_t **active, unsigned int *count_all, unsigned int *count_clean) { bool keep_looking = false; bool is_happy = false; CRM_CHECK((rsc != NULL) && (node != NULL) && (active != NULL), return false); is_happy = node->details->online && !node->details->unclean; if (count_all != NULL) { ++*count_all; } if ((count_clean != NULL) && is_happy) { ++*count_clean; } if ((count_all != NULL) || (count_clean != NULL)) { keep_looking = true; // We're counting, so go through entire list } if (rsc->priv->partial_migration_source != NULL) { if (pcmk__same_node(node, rsc->priv->partial_migration_source)) { *active = node; // This is the migration source } else { keep_looking = true; } } else if (!pcmk_is_set(rsc->flags, pcmk__rsc_needs_fencing)) { if (is_happy && ((*active == NULL) || !(*active)->details->online || (*active)->details->unclean)) { *active = node; // This is the first clean node } else { keep_looking = true; } } if (*active == NULL) { *active = node; // This is the first node checked } return keep_looking; } // Shared implementation of pcmk__rsc_methods_t:active_node() static pcmk_node_t * active_node(const pcmk_resource_t *rsc, unsigned int *count_all, unsigned int *count_clean) { pcmk_node_t *active = NULL; if (count_all != NULL) { *count_all = 0; } if (count_clean != NULL) { *count_clean = 0; } if (rsc == NULL) { return NULL; } for (GList *iter = rsc->priv->active_nodes; iter != NULL; iter = iter->next) { if (!pe__count_active_node(rsc, (pcmk_node_t *) iter->data, &active, count_all, count_clean)) { break; // Don't waste time iterating if we don't have to } } return active; } /*! * \brief * \internal Find and count active nodes according to \c PCMK_META_REQUIRES * * \param[in] rsc Resource to check * \param[out] count If not NULL, will be set to count of active nodes * * \return An active node (or NULL if resource is not active anywhere) * * \note This is a convenience wrapper for active_node() where the count of all * active nodes or only clean active nodes is desired according to the * \c PCMK_META_REQUIRES meta-attribute. */ pcmk_node_t * pe__find_active_requires(const pcmk_resource_t *rsc, unsigned int *count) { if (rsc == NULL) { if (count != NULL) { *count = 0; } return NULL; } if (pcmk_is_set(rsc->flags, pcmk__rsc_needs_fencing)) { return rsc->priv->fns->active_node(rsc, count, NULL); } else { return rsc->priv->fns->active_node(rsc, NULL, count); } } void pe__count_common(pcmk_resource_t *rsc) { if (rsc->priv->children != NULL) { for (GList *item = rsc->priv->children; item != NULL; item = item->next) { pcmk_resource_t *child = item->data; child->priv->fns->count(item->data); } } else if (!pcmk_is_set(rsc->flags, pcmk__rsc_removed) || (rsc->priv->orig_role > pcmk_role_stopped)) { rsc->priv->scheduler->priv->ninstances++; if (pe__resource_is_disabled(rsc)) { rsc->priv->scheduler->priv->disabled_resources++; } if (pcmk_is_set(rsc->flags, pcmk__rsc_blocked)) { rsc->priv->scheduler->priv->blocked_resources++; } } } /*! * \internal * \brief Update a resource's next role * * \param[in,out] rsc Resource to be updated * \param[in] role Resource's new next role * \param[in] why Human-friendly reason why role is changing (for logs) */ void pe__set_next_role(pcmk_resource_t *rsc, enum rsc_role_e role, const char *why) { pcmk__assert((rsc != NULL) && (why != NULL)); if (rsc->priv->next_role != role) { pcmk__rsc_trace(rsc, "Resetting next role for %s from %s to %s (%s)", rsc->id, pcmk_role_text(rsc->priv->next_role), pcmk_role_text(role), why); rsc->priv->next_role = role; } } diff --git a/lib/pengine/pe_actions.c b/lib/pengine/pe_actions.c index cbe7bda25f..eab9304879 100644 --- a/lib/pengine/pe_actions.c +++ b/lib/pengine/pe_actions.c @@ -1,1782 +1,1782 @@ /* * Copyright 2004-2024 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 "pe_status_private.h" static void unpack_operation(pcmk_action_t *action, const xmlNode *xml_obj, guint interval_ms); static void add_singleton(pcmk_scheduler_t *scheduler, pcmk_action_t *action) { if (scheduler->priv->singletons == NULL) { scheduler->priv->singletons = pcmk__strkey_table(NULL, NULL); } g_hash_table_insert(scheduler->priv->singletons, action->uuid, action); } static pcmk_action_t * lookup_singleton(pcmk_scheduler_t *scheduler, const char *action_uuid) { /* @TODO This is the only use of the pcmk_scheduler_t:singletons hash table. * Compare the performance of this approach to keeping the * pcmk_scheduler_t:actions list sorted by action key and just searching * that instead. */ if (scheduler->priv->singletons == NULL) { return NULL; } return g_hash_table_lookup(scheduler->priv->singletons, action_uuid); } /*! * \internal * \brief Find an existing action that matches arguments * * \param[in] key Action key to match * \param[in] rsc Resource to match (if any) * \param[in] node Node to match (if any) * \param[in] scheduler Scheduler data * * \return Existing action that matches arguments (or NULL if none) */ static pcmk_action_t * find_existing_action(const char *key, const pcmk_resource_t *rsc, const pcmk_node_t *node, const pcmk_scheduler_t *scheduler) { /* When rsc is NULL, it would be quicker to check * scheduler->priv->singletons, but checking all scheduler->priv->actions * takes the node into account. */ GList *actions = (rsc == NULL)? scheduler->priv->actions : rsc->priv->actions; GList *matches = find_actions(actions, key, node); pcmk_action_t *action = NULL; if (matches == NULL) { return NULL; } CRM_LOG_ASSERT(!pcmk__list_of_multiple(matches)); action = matches->data; g_list_free(matches); return action; } /*! * \internal * \brief Find the XML configuration corresponding to a specific action key * * \param[in] rsc Resource to find action configuration for * \param[in] key "RSC_ACTION_INTERVAL" of action to find * \param[in] include_disabled If false, do not return disabled actions * * \return XML configuration of desired action if any, otherwise NULL */ static xmlNode * find_exact_action_config(const pcmk_resource_t *rsc, const char *action_name, guint interval_ms, bool include_disabled) { for (xmlNode *operation = pcmk__xe_first_child(rsc->priv->ops_xml, PCMK_XE_OP, NULL, NULL); operation != NULL; operation = pcmk__xe_next(operation, PCMK_XE_OP)) { bool enabled = false; const char *config_name = NULL; const char *interval_spec = NULL; guint tmp_ms = 0U; // @TODO This does not consider meta-attributes, rules, defaults, etc. if (!include_disabled && (pcmk__xe_get_bool_attr(operation, PCMK_META_ENABLED, &enabled) == pcmk_rc_ok) && !enabled) { continue; } interval_spec = crm_element_value(operation, PCMK_META_INTERVAL); pcmk_parse_interval_spec(interval_spec, &tmp_ms); if (tmp_ms != interval_ms) { continue; } config_name = crm_element_value(operation, PCMK_XA_NAME); if (pcmk__str_eq(action_name, config_name, pcmk__str_none)) { return operation; } } return NULL; } /*! * \internal * \brief Find the XML configuration of a resource action * * \param[in] rsc Resource to find action configuration for * \param[in] action_name Action name to search for * \param[in] interval_ms Action interval (in milliseconds) to search for * \param[in] include_disabled If false, do not return disabled actions * * \return XML configuration of desired action if any, otherwise NULL */ xmlNode * pcmk__find_action_config(const pcmk_resource_t *rsc, const char *action_name, guint interval_ms, bool include_disabled) { xmlNode *action_config = NULL; // Try requested action first action_config = find_exact_action_config(rsc, action_name, interval_ms, include_disabled); // For migrate_to and migrate_from actions, retry with "migrate" // @TODO This should be either documented or deprecated if ((action_config == NULL) && pcmk__str_any_of(action_name, PCMK_ACTION_MIGRATE_TO, PCMK_ACTION_MIGRATE_FROM, NULL)) { action_config = find_exact_action_config(rsc, "migrate", 0, include_disabled); } return action_config; } /*! * \internal * \brief Create a new action object * * \param[in] key Action key * \param[in] task Action name * \param[in,out] rsc Resource that action is for (if any) * \param[in] node Node that action is on (if any) * \param[in] optional Whether action should be considered optional * \param[in,out] scheduler Scheduler data * * \return Newly allocated action * \note This function takes ownership of \p key. It is the caller's * responsibility to free the return value with pe_free_action(). */ static pcmk_action_t * new_action(char *key, const char *task, pcmk_resource_t *rsc, const pcmk_node_t *node, bool optional, pcmk_scheduler_t *scheduler) { pcmk_action_t *action = pcmk__assert_alloc(1, sizeof(pcmk_action_t)); action->rsc = rsc; action->task = pcmk__str_copy(task); action->uuid = key; action->scheduler = scheduler; if (node) { action->node = pe__copy_node(node); } if (pcmk__str_eq(task, PCMK_ACTION_LRM_DELETE, pcmk__str_casei)) { // Resource history deletion for a node can be done on the DC pcmk__set_action_flags(action, pcmk__action_on_dc); } pcmk__set_action_flags(action, pcmk__action_runnable); if (optional) { pcmk__set_action_flags(action, pcmk__action_optional); } else { pcmk__clear_action_flags(action, pcmk__action_optional); } if (rsc == NULL) { action->meta = pcmk__strkey_table(free, free); } else { guint interval_ms = 0; parse_op_key(key, NULL, NULL, &interval_ms); action->op_entry = pcmk__find_action_config(rsc, task, interval_ms, true); /* If the given key is for one of the many notification pseudo-actions * (pre_notify_promote, etc.), the actual action name is "notify" */ if ((action->op_entry == NULL) && (strstr(key, "_notify_") != NULL)) { action->op_entry = find_exact_action_config(rsc, PCMK_ACTION_NOTIFY, 0, true); } unpack_operation(action, action->op_entry, interval_ms); } pcmk__rsc_trace(rsc, "Created %s action %d (%s): %s for %s on %s", (optional? "optional" : "required"), scheduler->priv->next_action_id, key, task, ((rsc == NULL)? "no resource" : rsc->id), pcmk__node_name(node)); action->id = scheduler->priv->next_action_id++; scheduler->priv->actions = g_list_prepend(scheduler->priv->actions, action); if (rsc == NULL) { add_singleton(scheduler, action); } else { rsc->priv->actions = g_list_prepend(rsc->priv->actions, action); } return action; } /*! * \internal * \brief Unpack a resource's action-specific instance parameters * * \param[in] action_xml XML of action's configuration in CIB (if any) * \param[in,out] node_attrs Table of node attributes (for rule evaluation) * \param[in,out] scheduler Cluster working set (for rule evaluation) * * \return Newly allocated hash table of action-specific instance parameters */ GHashTable * pcmk__unpack_action_rsc_params(const xmlNode *action_xml, GHashTable *node_attrs, pcmk_scheduler_t *scheduler) { GHashTable *params = pcmk__strkey_table(free, free); pe_rule_eval_data_t rule_data = { .node_hash = node_attrs, .now = scheduler->priv->now, .match_data = NULL, .rsc_data = NULL, .op_data = NULL }; pe__unpack_dataset_nvpairs(action_xml, PCMK_XE_INSTANCE_ATTRIBUTES, &rule_data, params, NULL, scheduler); return params; } /*! * \internal * \brief Update an action's optional flag * * \param[in,out] action Action to update * \param[in] optional Requested optional status */ static void update_action_optional(pcmk_action_t *action, gboolean optional) { // Force a non-recurring action to be optional if its resource is unmanaged if ((action->rsc != NULL) && (action->node != NULL) && !pcmk_is_set(action->flags, pcmk__action_pseudo) && !pcmk_is_set(action->rsc->flags, pcmk__rsc_managed) && (g_hash_table_lookup(action->meta, PCMK_META_INTERVAL) == NULL)) { pcmk__rsc_debug(action->rsc, "%s on %s is optional (%s is unmanaged)", action->uuid, pcmk__node_name(action->node), action->rsc->id); pcmk__set_action_flags(action, pcmk__action_optional); // We shouldn't clear runnable here because ... something // Otherwise require the action if requested } else if (!optional) { pcmk__clear_action_flags(action, pcmk__action_optional); } } static enum pe_quorum_policy effective_quorum_policy(pcmk_resource_t *rsc, pcmk_scheduler_t *scheduler) { enum pe_quorum_policy policy = scheduler->no_quorum_policy; if (pcmk_is_set(scheduler->flags, pcmk__sched_quorate)) { policy = pcmk_no_quorum_ignore; } else if (scheduler->no_quorum_policy == pcmk_no_quorum_demote) { switch (rsc->priv->orig_role) { case pcmk_role_promoted: case pcmk_role_unpromoted: if (rsc->priv->next_role > pcmk_role_unpromoted) { pe__set_next_role(rsc, pcmk_role_unpromoted, PCMK_OPT_NO_QUORUM_POLICY "=demote"); } policy = pcmk_no_quorum_ignore; break; default: policy = pcmk_no_quorum_stop; break; } } return policy; } /*! * \internal * \brief Update a resource action's runnable flag * * \param[in,out] action Action to update * \param[in,out] scheduler Scheduler data * * \note This may also schedule fencing if a stop is unrunnable. */ static void update_resource_action_runnable(pcmk_action_t *action, pcmk_scheduler_t *scheduler) { pcmk_resource_t *rsc = action->rsc; if (pcmk_is_set(action->flags, pcmk__action_pseudo)) { return; } if (action->node == NULL) { pcmk__rsc_trace(rsc, "%s is unrunnable (unallocated)", action->uuid); pcmk__clear_action_flags(action, pcmk__action_runnable); } else if (!pcmk_is_set(action->flags, pcmk__action_on_dc) && !(action->node->details->online) && (!pcmk__is_guest_or_bundle_node(action->node) || pcmk_is_set(action->node->priv->flags, pcmk__node_remote_reset))) { pcmk__clear_action_flags(action, pcmk__action_runnable); do_crm_log(LOG_WARNING, "%s on %s is unrunnable (node is offline)", action->uuid, pcmk__node_name(action->node)); if (pcmk_is_set(rsc->flags, pcmk__rsc_managed) && pcmk__str_eq(action->task, PCMK_ACTION_STOP, pcmk__str_casei) && !(action->node->details->unclean)) { pe_fence_node(scheduler, action->node, "stop is unrunnable", false); } } else if (!pcmk_is_set(action->flags, pcmk__action_on_dc) && action->node->details->pending) { pcmk__clear_action_flags(action, pcmk__action_runnable); do_crm_log(LOG_WARNING, "Action %s on %s is unrunnable (node is pending)", action->uuid, pcmk__node_name(action->node)); } else if (action->needs == pcmk__requires_nothing) { pe_action_set_reason(action, NULL, TRUE); if (pcmk__is_guest_or_bundle_node(action->node) && !pe_can_fence(scheduler, action->node)) { /* An action that requires nothing usually does not require any * fencing in order to be runnable. However, there is an exception: * such an action cannot be completed if it is on a guest node whose * host is unclean and cannot be fenced. */ pcmk__rsc_debug(rsc, "%s on %s is unrunnable " "(node's host cannot be fenced)", action->uuid, pcmk__node_name(action->node)); pcmk__clear_action_flags(action, pcmk__action_runnable); } else { pcmk__rsc_trace(rsc, "%s on %s does not require fencing or quorum", action->uuid, pcmk__node_name(action->node)); pcmk__set_action_flags(action, pcmk__action_runnable); } } else { switch (effective_quorum_policy(rsc, scheduler)) { case pcmk_no_quorum_stop: pcmk__rsc_debug(rsc, "%s on %s is unrunnable (no quorum)", action->uuid, pcmk__node_name(action->node)); pcmk__clear_action_flags(action, pcmk__action_runnable); pe_action_set_reason(action, "no quorum", true); break; case pcmk_no_quorum_freeze: if (!rsc->priv->fns->active(rsc, TRUE) || (rsc->priv->next_role > rsc->priv->orig_role)) { pcmk__rsc_debug(rsc, "%s on %s is unrunnable (no quorum)", action->uuid, pcmk__node_name(action->node)); pcmk__clear_action_flags(action, pcmk__action_runnable); pe_action_set_reason(action, "quorum freeze", true); } break; default: //pe_action_set_reason(action, NULL, TRUE); pcmk__set_action_flags(action, pcmk__action_runnable); break; } } } static bool valid_stop_on_fail(const char *value) { return !pcmk__strcase_any_of(value, PCMK_VALUE_STANDBY, PCMK_VALUE_DEMOTE, PCMK_VALUE_STOP, NULL); } /*! * \internal * \brief Validate (and possibly reset) resource action's on_fail meta-attribute * * \param[in] rsc Resource that action is for * \param[in] action_name Action name * \param[in] action_config Action configuration XML from CIB (if any) * \param[in,out] meta Table of action meta-attributes */ static void validate_on_fail(const pcmk_resource_t *rsc, const char *action_name, const xmlNode *action_config, GHashTable *meta) { const char *name = NULL; const char *role = NULL; const char *interval_spec = NULL; const char *value = g_hash_table_lookup(meta, PCMK_META_ON_FAIL); guint interval_ms = 0U; // Stop actions can only use certain on-fail values if (pcmk__str_eq(action_name, PCMK_ACTION_STOP, pcmk__str_none) && !valid_stop_on_fail(value)) { pcmk__config_err("Resetting '" PCMK_META_ON_FAIL "' for %s stop " "action to default value because '%s' is not " "allowed for stop", rsc->id, value); g_hash_table_remove(meta, PCMK_META_ON_FAIL); return; } /* Demote actions default on-fail to the on-fail value for the first * recurring monitor for the promoted role (if any). */ if (pcmk__str_eq(action_name, PCMK_ACTION_DEMOTE, pcmk__str_none) && (value == NULL)) { /* @TODO This does not consider promote options set in a meta-attribute * block (which may have rules that need to be evaluated) rather than * XML properties. */ for (xmlNode *operation = pcmk__xe_first_child(rsc->priv->ops_xml, PCMK_XE_OP, NULL, NULL); operation != NULL; operation = pcmk__xe_next(operation, PCMK_XE_OP)) { bool enabled = false; const char *promote_on_fail = NULL; /* We only care about explicit on-fail (if promote uses default, so * can demote) */ promote_on_fail = crm_element_value(operation, PCMK_META_ON_FAIL); if (promote_on_fail == NULL) { continue; } // We only care about recurring monitors for the promoted role name = crm_element_value(operation, PCMK_XA_NAME); role = crm_element_value(operation, PCMK_XA_ROLE); if (!pcmk__str_eq(name, PCMK_ACTION_MONITOR, pcmk__str_none) || !pcmk__strcase_any_of(role, PCMK_ROLE_PROMOTED, PCMK__ROLE_PROMOTED_LEGACY, NULL)) { continue; } interval_spec = crm_element_value(operation, PCMK_META_INTERVAL); pcmk_parse_interval_spec(interval_spec, &interval_ms); if (interval_ms == 0U) { continue; } // We only care about enabled monitors if ((pcmk__xe_get_bool_attr(operation, PCMK_META_ENABLED, &enabled) == pcmk_rc_ok) && !enabled) { continue; } /* Demote actions can't default to * PCMK_META_ON_FAIL=PCMK_VALUE_DEMOTE */ if (pcmk__str_eq(promote_on_fail, PCMK_VALUE_DEMOTE, pcmk__str_casei)) { continue; } // Use value from first applicable promote action found pcmk__insert_dup(meta, PCMK_META_ON_FAIL, promote_on_fail); } return; } if (pcmk__str_eq(action_name, PCMK_ACTION_LRM_DELETE, pcmk__str_none) && !pcmk__str_eq(value, PCMK_VALUE_IGNORE, pcmk__str_casei)) { pcmk__insert_dup(meta, PCMK_META_ON_FAIL, PCMK_VALUE_IGNORE); return; } // PCMK_META_ON_FAIL=PCMK_VALUE_DEMOTE is allowed only for certain actions if (pcmk__str_eq(value, PCMK_VALUE_DEMOTE, pcmk__str_casei)) { name = crm_element_value(action_config, PCMK_XA_NAME); role = crm_element_value(action_config, PCMK_XA_ROLE); interval_spec = crm_element_value(action_config, PCMK_META_INTERVAL); pcmk_parse_interval_spec(interval_spec, &interval_ms); if (!pcmk__str_eq(name, PCMK_ACTION_PROMOTE, pcmk__str_none) && ((interval_ms == 0U) || !pcmk__str_eq(name, PCMK_ACTION_MONITOR, pcmk__str_none) || !pcmk__strcase_any_of(role, PCMK_ROLE_PROMOTED, PCMK__ROLE_PROMOTED_LEGACY, NULL))) { pcmk__config_err("Resetting '" PCMK_META_ON_FAIL "' for %s %s " "action to default value because 'demote' is not " "allowed for it", rsc->id, name); g_hash_table_remove(meta, PCMK_META_ON_FAIL); return; } } } static int unpack_timeout(const char *value) { long long timeout_ms = crm_get_msec(value); if (timeout_ms <= 0) { timeout_ms = PCMK_DEFAULT_ACTION_TIMEOUT_MS; } return (int) QB_MIN(timeout_ms, INT_MAX); } // true if value contains valid, non-NULL interval origin for recurring op static bool unpack_interval_origin(const char *value, const xmlNode *xml_obj, guint interval_ms, const crm_time_t *now, long long *start_delay) { long long result = 0; guint interval_sec = pcmk__timeout_ms2s(interval_ms); crm_time_t *origin = NULL; // Ignore unspecified values and non-recurring operations if ((value == NULL) || (interval_ms == 0) || (now == NULL)) { return false; } // Parse interval origin from text origin = crm_time_new(value); if (origin == NULL) { pcmk__config_err("Ignoring '" PCMK_META_INTERVAL_ORIGIN "' for " "operation '%s' because '%s' is not valid", pcmk__s(pcmk__xe_id(xml_obj), "(missing ID)"), value); return false; } // Get seconds since origin (negative if origin is in the future) result = crm_time_get_seconds(now) - crm_time_get_seconds(origin); crm_time_free(origin); // Calculate seconds from closest interval to now result = result % interval_sec; // Calculate seconds remaining until next interval result = ((result <= 0)? 0 : interval_sec) - result; crm_info("Calculated a start delay of %llds for operation '%s'", result, pcmk__s(pcmk__xe_id(xml_obj), "(unspecified)")); if (start_delay != NULL) { *start_delay = result * 1000; // milliseconds } return true; } static int unpack_start_delay(const char *value, GHashTable *meta) { long long start_delay_ms = 0; if (value == NULL) { return 0; } start_delay_ms = crm_get_msec(value); start_delay_ms = QB_MIN(start_delay_ms, INT_MAX); if (start_delay_ms < 0) { start_delay_ms = 0; } if (meta != NULL) { g_hash_table_replace(meta, strdup(PCMK_META_START_DELAY), pcmk__itoa(start_delay_ms)); } return (int) start_delay_ms; } /*! * \internal * \brief Find a resource's most frequent recurring monitor * * \param[in] rsc Resource to check * * \return Operation XML configured for most frequent recurring monitor for * \p rsc (if any) */ static xmlNode * most_frequent_monitor(const pcmk_resource_t *rsc) { guint min_interval_ms = G_MAXUINT; xmlNode *op = NULL; for (xmlNode *operation = pcmk__xe_first_child(rsc->priv->ops_xml, PCMK_XE_OP, NULL, NULL); operation != NULL; operation = pcmk__xe_next(operation, PCMK_XE_OP)) { bool enabled = false; guint interval_ms = 0U; const char *interval_spec = crm_element_value(operation, PCMK_META_INTERVAL); // We only care about enabled recurring monitors if (!pcmk__str_eq(crm_element_value(operation, PCMK_XA_NAME), PCMK_ACTION_MONITOR, pcmk__str_none)) { continue; } pcmk_parse_interval_spec(interval_spec, &interval_ms); if (interval_ms == 0U) { continue; } // @TODO This does not consider meta-attributes, rules, defaults, etc. if ((pcmk__xe_get_bool_attr(operation, PCMK_META_ENABLED, &enabled) == pcmk_rc_ok) && !enabled) { continue; } if (interval_ms < min_interval_ms) { min_interval_ms = interval_ms; op = operation; } } return op; } /*! * \internal * \brief Unpack action meta-attributes * * \param[in,out] rsc Resource that action is for * \param[in] node Node that action is on * \param[in] action_name Action name * \param[in] interval_ms Action interval (in milliseconds) * \param[in] action_config Action XML configuration from CIB (if any) * * Unpack a resource action's meta-attributes (normalizing the interval, * timeout, and start delay values as integer milliseconds) from its CIB XML * configuration (including defaults). * * \return Newly allocated hash table with normalized action meta-attributes */ GHashTable * pcmk__unpack_action_meta(pcmk_resource_t *rsc, const pcmk_node_t *node, const char *action_name, guint interval_ms, const xmlNode *action_config) { GHashTable *meta = NULL; const char *timeout_spec = NULL; const char *str = NULL; pe_rsc_eval_data_t rsc_rule_data = { .standard = crm_element_value(rsc->priv->xml, PCMK_XA_CLASS), .provider = crm_element_value(rsc->priv->xml, PCMK_XA_PROVIDER), .agent = crm_element_value(rsc->priv->xml, PCMK_XA_TYPE), }; pe_op_eval_data_t op_rule_data = { .op_name = action_name, .interval = interval_ms, }; pe_rule_eval_data_t rule_data = { /* Node attributes are not set because node expressions are not allowed * for meta-attributes */ .now = rsc->priv->scheduler->priv->now, .match_data = NULL, .rsc_data = &rsc_rule_data, .op_data = &op_rule_data, }; meta = pcmk__strkey_table(free, free); if (action_config != NULL) { // take precedence over defaults pe__unpack_dataset_nvpairs(action_config, PCMK_XE_META_ATTRIBUTES, &rule_data, meta, NULL, rsc->priv->scheduler); /* Anything set as an XML property has highest precedence. * This ensures we use the name and interval from the tag. * (See below for the only exception, fence device start/probe timeout.) */ for (xmlAttrPtr attr = action_config->properties; attr != NULL; attr = attr->next) { pcmk__insert_dup(meta, (const char *) attr->name, pcmk__xml_attr_value(attr)); } } // Derive default timeout for probes from recurring monitor timeouts if (pcmk_is_probe(action_name, interval_ms) && (g_hash_table_lookup(meta, PCMK_META_TIMEOUT) == NULL)) { xmlNode *min_interval_mon = most_frequent_monitor(rsc); if (min_interval_mon != NULL) { /* @TODO This does not consider timeouts set in * PCMK_XE_META_ATTRIBUTES blocks (which may also have rules that * need to be evaluated). */ timeout_spec = crm_element_value(min_interval_mon, PCMK_META_TIMEOUT); if (timeout_spec != NULL) { pcmk__rsc_trace(rsc, "Setting default timeout for %s probe to " "most frequent monitor's timeout '%s'", rsc->id, timeout_spec); pcmk__insert_dup(meta, PCMK_META_TIMEOUT, timeout_spec); } } } // Cluster-wide pe__unpack_dataset_nvpairs(rsc->priv->scheduler->priv->op_defaults, PCMK_XE_META_ATTRIBUTES, &rule_data, meta, NULL, rsc->priv->scheduler); g_hash_table_remove(meta, PCMK_XA_ID); // Normalize interval to milliseconds if (interval_ms > 0) { g_hash_table_insert(meta, pcmk__str_copy(PCMK_META_INTERVAL), crm_strdup_printf("%u", interval_ms)); } else { g_hash_table_remove(meta, PCMK_META_INTERVAL); } /* Timeout order of precedence (highest to lowest): * 1. pcmk_monitor_timeout resource parameter (only for starts and probes * when rsc has pcmk_ra_cap_fence_params; this gets used for recurring * monitors via the executor instead) * 2. timeout configured in (with taking precedence over * ) * 3. timeout configured in * 4. PCMK_DEFAULT_ACTION_TIMEOUT_MS */ // Check for pcmk_monitor_timeout if (pcmk_is_set(pcmk_get_ra_caps(rsc_rule_data.standard), pcmk_ra_cap_fence_params) && (pcmk__str_eq(action_name, PCMK_ACTION_START, pcmk__str_none) || pcmk_is_probe(action_name, interval_ms))) { GHashTable *params = pe_rsc_params(rsc, node, rsc->priv->scheduler); timeout_spec = g_hash_table_lookup(params, "pcmk_monitor_timeout"); if (timeout_spec != NULL) { pcmk__rsc_trace(rsc, "Setting timeout for %s %s to " "pcmk_monitor_timeout (%s)", rsc->id, action_name, timeout_spec); pcmk__insert_dup(meta, PCMK_META_TIMEOUT, timeout_spec); } } // Normalize timeout to positive milliseconds timeout_spec = g_hash_table_lookup(meta, PCMK_META_TIMEOUT); g_hash_table_insert(meta, pcmk__str_copy(PCMK_META_TIMEOUT), pcmk__itoa(unpack_timeout(timeout_spec))); // Ensure on-fail has a valid value validate_on_fail(rsc, action_name, action_config, meta); // Normalize PCMK_META_START_DELAY str = g_hash_table_lookup(meta, PCMK_META_START_DELAY); if (str != NULL) { unpack_start_delay(str, meta); } else { long long start_delay = 0; str = g_hash_table_lookup(meta, PCMK_META_INTERVAL_ORIGIN); if (unpack_interval_origin(str, action_config, interval_ms, rsc->priv->scheduler->priv->now, &start_delay)) { g_hash_table_insert(meta, pcmk__str_copy(PCMK_META_START_DELAY), crm_strdup_printf("%lld", start_delay)); } } return meta; } /*! * \internal * \brief Determine an action's quorum and fencing dependency * * \param[in] rsc Resource that action is for * \param[in] action_name Name of action being unpacked * * \return Quorum and fencing dependency appropriate to action */ enum pcmk__requires pcmk__action_requires(const pcmk_resource_t *rsc, const char *action_name) { const char *value = NULL; enum pcmk__requires requires = pcmk__requires_nothing; CRM_CHECK((rsc != NULL) && (action_name != NULL), return requires); if (!pcmk__strcase_any_of(action_name, PCMK_ACTION_START, PCMK_ACTION_PROMOTE, NULL)) { value = "nothing (not start or promote)"; } else if (pcmk_is_set(rsc->flags, pcmk__rsc_needs_fencing)) { requires = pcmk__requires_fencing; value = "fencing"; } else if (pcmk_is_set(rsc->flags, pcmk__rsc_needs_quorum)) { requires = pcmk__requires_quorum; value = "quorum"; } else { value = "nothing"; } pcmk__rsc_trace(rsc, "%s of %s requires %s", action_name, rsc->id, value); return requires; } /*! * \internal * \brief Parse action failure response from a user-provided string * * \param[in] rsc Resource that action is for * \param[in] action_name Name of action * \param[in] interval_ms Action interval (in milliseconds) * \param[in] value User-provided configuration value for on-fail * * \return Action failure response parsed from \p text */ enum pcmk__on_fail pcmk__parse_on_fail(const pcmk_resource_t *rsc, const char *action_name, guint interval_ms, const char *value) { const char *desc = NULL; bool needs_remote_reset = false; enum pcmk__on_fail on_fail = pcmk__on_fail_ignore; const pcmk_scheduler_t *scheduler = NULL; // There's no enum value for unknown or invalid, so assert pcmk__assert((rsc != NULL) && (action_name != NULL)); scheduler = rsc->priv->scheduler; if (value == NULL) { // Use default } else if (pcmk__str_eq(value, PCMK_VALUE_BLOCK, pcmk__str_casei)) { on_fail = pcmk__on_fail_block; desc = "block"; } else if (pcmk__str_eq(value, PCMK_VALUE_FENCE, pcmk__str_casei)) { if (pcmk_is_set(scheduler->flags, pcmk__sched_fencing_enabled)) { on_fail = pcmk__on_fail_fence_node; desc = "node fencing"; } else { pcmk__config_err("Resetting '" PCMK_META_ON_FAIL "' for " "%s of %s to 'stop' because 'fence' is not " "valid when fencing is disabled", action_name, rsc->id); on_fail = pcmk__on_fail_stop; desc = "stop resource"; } } else if (pcmk__str_eq(value, PCMK_VALUE_STANDBY, pcmk__str_casei)) { on_fail = pcmk__on_fail_standby_node; desc = "node standby"; } else if (pcmk__strcase_any_of(value, PCMK_VALUE_IGNORE, PCMK_VALUE_NOTHING, NULL)) { desc = "ignore"; } else if (pcmk__str_eq(value, "migrate", pcmk__str_casei)) { on_fail = pcmk__on_fail_ban; desc = "force migration"; } else if (pcmk__str_eq(value, PCMK_VALUE_STOP, pcmk__str_casei)) { on_fail = pcmk__on_fail_stop; desc = "stop resource"; } else if (pcmk__str_eq(value, PCMK_VALUE_RESTART, pcmk__str_casei)) { on_fail = pcmk__on_fail_restart; desc = "restart (and possibly migrate)"; } else if (pcmk__str_eq(value, PCMK_VALUE_RESTART_CONTAINER, pcmk__str_casei)) { if (rsc->priv->launcher == NULL) { pcmk__rsc_debug(rsc, "Using default " PCMK_META_ON_FAIL " for %s " "of %s because it does not have a launcher", action_name, rsc->id); } else { on_fail = pcmk__on_fail_restart_container; desc = "restart container (and possibly migrate)"; } } else if (pcmk__str_eq(value, PCMK_VALUE_DEMOTE, pcmk__str_casei)) { on_fail = pcmk__on_fail_demote; desc = "demote instance"; } else { pcmk__config_err("Using default '" PCMK_META_ON_FAIL "' for " "%s of %s because '%s' is not valid", action_name, rsc->id, value); } /* Remote node connections are handled specially. Failures that result * in dropping an active connection must result in fencing. The only * failures that don't are probes and starts. The user can explicitly set * PCMK_META_ON_FAIL=PCMK_VALUE_FENCE to fence after start failures. */ if (pcmk_is_set(rsc->flags, pcmk__rsc_is_remote_connection) && pcmk__is_remote_node(pcmk_find_node(scheduler, rsc->id)) && !pcmk_is_probe(action_name, interval_ms) && !pcmk__str_eq(action_name, PCMK_ACTION_START, pcmk__str_none)) { needs_remote_reset = true; if (!pcmk_is_set(rsc->flags, pcmk__rsc_managed)) { desc = NULL; // Force default for unmanaged connections } } if (desc != NULL) { // Explicit value used, default not needed } else if (rsc->priv->launcher != NULL) { on_fail = pcmk__on_fail_restart_container; desc = "restart container (and possibly migrate) (default)"; } else if (needs_remote_reset) { if (pcmk_is_set(rsc->flags, pcmk__rsc_managed)) { if (pcmk_is_set(scheduler->flags, pcmk__sched_fencing_enabled)) { desc = "fence remote node (default)"; } else { desc = "recover remote node connection (default)"; } on_fail = pcmk__on_fail_reset_remote; } else { on_fail = pcmk__on_fail_stop; desc = "stop unmanaged remote node (enforcing default)"; } } else if (pcmk__str_eq(action_name, PCMK_ACTION_STOP, pcmk__str_none)) { if (pcmk_is_set(scheduler->flags, pcmk__sched_fencing_enabled)) { on_fail = pcmk__on_fail_fence_node; desc = "resource fence (default)"; } else { on_fail = pcmk__on_fail_block; desc = "resource block (default)"; } } else { on_fail = pcmk__on_fail_restart; desc = "restart (and possibly migrate) (default)"; } pcmk__rsc_trace(rsc, "Failure handling for %s-interval %s of %s: %s", pcmk__readable_interval(interval_ms), action_name, rsc->id, desc); return on_fail; } /*! * \internal * \brief Determine a resource's role after failure of an action * * \param[in] rsc Resource that action is for * \param[in] action_name Action name * \param[in] on_fail Failure handling for action * \param[in] meta Unpacked action meta-attributes * * \return Resource role that results from failure of action */ enum rsc_role_e pcmk__role_after_failure(const pcmk_resource_t *rsc, const char *action_name, enum pcmk__on_fail on_fail, GHashTable *meta) { enum rsc_role_e role = pcmk_role_unknown; // Set default for role after failure specially in certain circumstances switch (on_fail) { case pcmk__on_fail_stop: role = pcmk_role_stopped; break; case pcmk__on_fail_reset_remote: if (rsc->priv->remote_reconnect_ms != 0U) { role = pcmk_role_stopped; } break; default: break; } if (role == pcmk_role_unknown) { // Use default if (pcmk__str_eq(action_name, PCMK_ACTION_PROMOTE, pcmk__str_none)) { role = pcmk_role_unpromoted; } else { role = pcmk_role_started; } } pcmk__rsc_trace(rsc, "Role after %s %s failure is: %s", rsc->id, action_name, pcmk_role_text(role)); return role; } /*! * \internal * \brief Unpack action configuration * * Unpack a resource action's meta-attributes (normalizing the interval, * timeout, and start delay values as integer milliseconds), requirements, and * failure policy from its CIB XML configuration (including defaults). * * \param[in,out] action Resource action to unpack into * \param[in] xml_obj Action configuration XML (NULL for defaults only) * \param[in] interval_ms How frequently to perform the operation */ static void unpack_operation(pcmk_action_t *action, const xmlNode *xml_obj, guint interval_ms) { const char *value = NULL; action->meta = pcmk__unpack_action_meta(action->rsc, action->node, action->task, interval_ms, xml_obj); action->needs = pcmk__action_requires(action->rsc, action->task); value = g_hash_table_lookup(action->meta, PCMK_META_ON_FAIL); action->on_fail = pcmk__parse_on_fail(action->rsc, action->task, interval_ms, value); action->fail_role = pcmk__role_after_failure(action->rsc, action->task, action->on_fail, action->meta); } /*! * \brief Create or update an action object * * \param[in,out] rsc Resource that action is for (if any) * \param[in,out] key Action key (must be non-NULL) * \param[in] task Action name (must be non-NULL) * \param[in] on_node Node that action is on (if any) * \param[in] optional Whether action should be considered optional * \param[in,out] scheduler Scheduler data * * \return Action object corresponding to arguments (guaranteed not to be * \c NULL) * \note This function takes ownership of (and might free) \p key, and * \p scheduler takes ownership of the returned action (the caller should * not free it). */ pcmk_action_t * custom_action(pcmk_resource_t *rsc, char *key, const char *task, const pcmk_node_t *on_node, gboolean optional, pcmk_scheduler_t *scheduler) { pcmk_action_t *action = NULL; pcmk__assert((key != NULL) && (task != NULL) && (scheduler != NULL)); action = find_existing_action(key, rsc, on_node, scheduler); if (action == NULL) { action = new_action(key, task, rsc, on_node, optional, scheduler); } else { free(key); } update_action_optional(action, optional); if (rsc != NULL) { /* An action can be initially created with a NULL node, and later have * the node added via find_existing_action() (above) -> find_actions(). * That is why the extra parameters are unpacked here rather than in * new_action(). */ if ((action->node != NULL) && (action->op_entry != NULL) && !pcmk_is_set(action->flags, pcmk__action_attrs_evaluated)) { GHashTable *attrs = action->node->priv->attrs; if (action->extra != NULL) { g_hash_table_destroy(action->extra); } action->extra = pcmk__unpack_action_rsc_params(action->op_entry, attrs, scheduler); pcmk__set_action_flags(action, pcmk__action_attrs_evaluated); } update_resource_action_runnable(action, scheduler); } if (action->extra == NULL) { action->extra = pcmk__strkey_table(free, free); } return action; } pcmk_action_t * get_pseudo_op(const char *name, pcmk_scheduler_t *scheduler) { pcmk_action_t *op = lookup_singleton(scheduler, name); if (op == NULL) { op = custom_action(NULL, strdup(name), name, NULL, TRUE, scheduler); pcmk__set_action_flags(op, pcmk__action_pseudo|pcmk__action_runnable); } return op; } static GList * find_unfencing_devices(GList *candidates, GList *matches) { for (GList *gIter = candidates; gIter != NULL; gIter = gIter->next) { pcmk_resource_t *candidate = gIter->data; if (candidate->priv->children != NULL) { matches = find_unfencing_devices(candidate->priv->children, matches); } else if (!pcmk_is_set(candidate->flags, pcmk__rsc_fence_device)) { continue; } else if (pcmk_is_set(candidate->flags, pcmk__rsc_needs_unfencing)) { matches = g_list_prepend(matches, candidate); } else if (pcmk__str_eq(g_hash_table_lookup(candidate->priv->meta, PCMK_STONITH_PROVIDES), PCMK_VALUE_UNFENCING, pcmk__str_casei)) { matches = g_list_prepend(matches, candidate); } } return matches; } static int node_priority_fencing_delay(const pcmk_node_t *node, const pcmk_scheduler_t *scheduler) { int member_count = 0; int online_count = 0; int top_priority = 0; int lowest_priority = 0; GList *gIter = NULL; // PCMK_OPT_PRIORITY_FENCING_DELAY is disabled if (scheduler->priv->priority_fencing_ms == 0U) { return 0; } /* No need to request a delay if the fencing target is not a normal cluster * member, for example if it's a remote node or a guest node. */ if (node->priv->variant != pcmk__node_variant_cluster) { return 0; } // No need to request a delay if the fencing target is in our partition if (node->details->online) { return 0; } for (gIter = scheduler->nodes; gIter != NULL; gIter = gIter->next) { pcmk_node_t *n = gIter->data; if (n->priv->variant != pcmk__node_variant_cluster) { continue; } member_count ++; if (n->details->online) { online_count++; } if (member_count == 1 || n->priv->priority > top_priority) { top_priority = n->priv->priority; } if (member_count == 1 || n->priv->priority < lowest_priority) { lowest_priority = n->priv->priority; } } // No need to delay if we have more than half of the cluster members if (online_count > member_count / 2) { return 0; } /* All the nodes have equal priority. * Any configured corresponding `pcmk_delay_base/max` will be applied. */ if (lowest_priority == top_priority) { return 0; } if (node->priv->priority < top_priority) { return 0; } return pcmk__timeout_ms2s(scheduler->priv->priority_fencing_ms); } pcmk_action_t * pe_fence_op(pcmk_node_t *node, const char *op, bool optional, const char *reason, bool priority_delay, pcmk_scheduler_t *scheduler) { char *op_key = NULL; pcmk_action_t *stonith_op = NULL; if(op == NULL) { op = scheduler->priv->fence_action; } op_key = crm_strdup_printf("%s-%s-%s", PCMK_ACTION_STONITH, node->priv->name, op); stonith_op = lookup_singleton(scheduler, op_key); if(stonith_op == NULL) { stonith_op = custom_action(NULL, op_key, PCMK_ACTION_STONITH, node, TRUE, scheduler); pcmk__insert_meta(stonith_op, PCMK__META_ON_NODE, node->priv->name); pcmk__insert_meta(stonith_op, PCMK__META_ON_NODE_UUID, node->priv->id); pcmk__insert_meta(stonith_op, PCMK__META_STONITH_ACTION, op); if (pcmk_is_set(scheduler->flags, pcmk__sched_enable_unfencing)) { /* Extra work to detect device changes */ GString *digests_all = g_string_sized_new(1024); GString *digests_secure = g_string_sized_new(1024); GList *matches = find_unfencing_devices(scheduler->priv->resources, NULL); for (GList *gIter = matches; gIter != NULL; gIter = gIter->next) { pcmk_resource_t *match = gIter->data; const char *agent = g_hash_table_lookup(match->priv->meta, PCMK_XA_TYPE); pcmk__op_digest_t *data = NULL; data = pe__compare_fencing_digest(match, agent, node, scheduler); if (data->rc == pcmk__digest_mismatch) { optional = FALSE; crm_notice("Unfencing node %s because the definition of " "%s changed", pcmk__node_name(node), match->id); if (!pcmk__is_daemon && (scheduler->priv->out != NULL)) { pcmk__output_t *out = scheduler->priv->out; out->info(out, "notice: Unfencing node %s because the " "definition of %s changed", pcmk__node_name(node), match->id); } } pcmk__g_strcat(digests_all, match->id, ":", agent, ":", data->digest_all_calc, ",", NULL); pcmk__g_strcat(digests_secure, match->id, ":", agent, ":", data->digest_secure_calc, ",", NULL); } pcmk__insert_dup(stonith_op->meta, PCMK__META_DIGESTS_ALL, digests_all->str); g_string_free(digests_all, TRUE); pcmk__insert_dup(stonith_op->meta, PCMK__META_DIGESTS_SECURE, digests_secure->str); g_string_free(digests_secure, TRUE); g_list_free(matches); } } else { free(op_key); } if ((scheduler->priv->priority_fencing_ms > 0U) /* It's a suitable case where PCMK_OPT_PRIORITY_FENCING_DELAY * applies. At least add PCMK_OPT_PRIORITY_FENCING_DELAY field as * an indicator. */ && (priority_delay /* The priority delay needs to be recalculated if this function has * been called by schedule_fencing_and_shutdowns() after node * priority has already been calculated by native_add_running(). */ || g_hash_table_lookup(stonith_op->meta, PCMK_OPT_PRIORITY_FENCING_DELAY) != NULL)) { /* Add PCMK_OPT_PRIORITY_FENCING_DELAY to the fencing op even if * it's 0 for the targeting node. So that it takes precedence over * any possible `pcmk_delay_base/max`. */ char *delay_s = pcmk__itoa(node_priority_fencing_delay(node, scheduler)); g_hash_table_insert(stonith_op->meta, strdup(PCMK_OPT_PRIORITY_FENCING_DELAY), delay_s); } if(optional == FALSE && pe_can_fence(scheduler, node)) { pcmk__clear_action_flags(stonith_op, pcmk__action_optional); pe_action_set_reason(stonith_op, reason, false); } else if(reason && stonith_op->reason == NULL) { stonith_op->reason = strdup(reason); } return stonith_op; } void pe_free_action(pcmk_action_t *action) { if (action == NULL) { return; } g_list_free_full(action->actions_before, free); g_list_free_full(action->actions_after, free); if (action->extra) { g_hash_table_destroy(action->extra); } if (action->meta) { g_hash_table_destroy(action->meta); } + pcmk__free_node_copy(action->node); free(action->cancel_task); free(action->reason); free(action->task); free(action->uuid); - free(action->node); free(action); } enum pcmk__action_type get_complex_task(const pcmk_resource_t *rsc, const char *name) { enum pcmk__action_type task = pcmk__parse_action(name); if (pcmk__is_primitive(rsc)) { switch (task) { case pcmk__action_stopped: case pcmk__action_started: case pcmk__action_demoted: case pcmk__action_promoted: crm_trace("Folding %s back into its atomic counterpart for %s", name, rsc->id); --task; break; default: break; } } return task; } /*! * \internal * \brief Find first matching action in a list * * \param[in] input List of actions to search * \param[in] uuid If not NULL, action must have this UUID * \param[in] task If not NULL, action must have this action name * \param[in] on_node If not NULL, action must be on this node * * \return First action in list that matches criteria, or NULL if none */ pcmk_action_t * find_first_action(const GList *input, const char *uuid, const char *task, const pcmk_node_t *on_node) { CRM_CHECK(uuid || task, return NULL); for (const GList *gIter = input; gIter != NULL; gIter = gIter->next) { pcmk_action_t *action = (pcmk_action_t *) gIter->data; if (uuid != NULL && !pcmk__str_eq(uuid, action->uuid, pcmk__str_casei)) { continue; } else if (task != NULL && !pcmk__str_eq(task, action->task, pcmk__str_casei)) { continue; } else if (on_node == NULL) { return action; } else if (action->node == NULL) { continue; } else if (pcmk__same_node(on_node, action->node)) { return action; } } return NULL; } GList * find_actions(GList *input, const char *key, const pcmk_node_t *on_node) { GList *gIter = input; GList *result = NULL; CRM_CHECK(key != NULL, return NULL); for (; gIter != NULL; gIter = gIter->next) { pcmk_action_t *action = (pcmk_action_t *) gIter->data; if (!pcmk__str_eq(key, action->uuid, pcmk__str_casei)) { continue; } else if (on_node == NULL) { crm_trace("Action %s matches (ignoring node)", key); result = g_list_prepend(result, action); } else if (action->node == NULL) { crm_trace("Action %s matches (unallocated, assigning to %s)", key, pcmk__node_name(on_node)); action->node = pe__copy_node(on_node); result = g_list_prepend(result, action); } else if (pcmk__same_node(on_node, action->node)) { crm_trace("Action %s on %s matches", key, pcmk__node_name(on_node)); result = g_list_prepend(result, action); } } return result; } GList * find_actions_exact(GList *input, const char *key, const pcmk_node_t *on_node) { GList *result = NULL; CRM_CHECK(key != NULL, return NULL); if (on_node == NULL) { return NULL; } for (GList *gIter = input; gIter != NULL; gIter = gIter->next) { pcmk_action_t *action = (pcmk_action_t *) gIter->data; if ((action->node != NULL) && pcmk__str_eq(key, action->uuid, pcmk__str_casei) && pcmk__same_node(on_node, action->node)) { crm_trace("Action %s on %s matches", key, pcmk__node_name(on_node)); result = g_list_prepend(result, action); } } return result; } /*! * \brief Find all actions of given type for a resource * * \param[in] rsc Resource to search * \param[in] node Find only actions scheduled on this node * \param[in] task Action name to search for * \param[in] require_node If TRUE, NULL node or action node will not match * * \return List of actions found (or NULL if none) * \note If node is not NULL and require_node is FALSE, matching actions * without a node will be assigned to node. */ GList * pe__resource_actions(const pcmk_resource_t *rsc, const pcmk_node_t *node, const char *task, bool require_node) { GList *result = NULL; char *key = pcmk__op_key(rsc->id, task, 0); if (require_node) { result = find_actions_exact(rsc->priv->actions, key, node); } else { result = find_actions(rsc->priv->actions, key, node); } free(key); return result; } /*! * \internal * \brief Create an action reason string based on the action itself * * \param[in] action Action to create reason string for * \param[in] flag Action flag that was cleared * * \return Newly allocated string suitable for use as action reason * \note It is the caller's responsibility to free() the result. */ char * pe__action2reason(const pcmk_action_t *action, enum pcmk__action_flags flag) { const char *change = NULL; switch (flag) { case pcmk__action_runnable: change = "unrunnable"; break; case pcmk__action_migratable: change = "unmigrateable"; break; case pcmk__action_optional: change = "required"; break; default: // Bug: caller passed unsupported flag CRM_CHECK(change != NULL, change = ""); break; } return crm_strdup_printf("%s%s%s %s", change, (action->rsc == NULL)? "" : " ", (action->rsc == NULL)? "" : action->rsc->id, action->task); } void pe_action_set_reason(pcmk_action_t *action, const char *reason, bool overwrite) { if (action->reason != NULL && overwrite) { pcmk__rsc_trace(action->rsc, "Changing %s reason from '%s' to '%s'", action->uuid, action->reason, pcmk__s(reason, "(none)")); } else if (action->reason == NULL) { pcmk__rsc_trace(action->rsc, "Set %s reason to '%s'", action->uuid, pcmk__s(reason, "(none)")); } else { // crm_assert(action->reason != NULL && !overwrite); return; } pcmk__str_update(&action->reason, reason); } /*! * \internal * \brief Create an action to clear a resource's history from CIB * * \param[in,out] rsc Resource to clear * \param[in] node Node to clear history on */ void pe__clear_resource_history(pcmk_resource_t *rsc, const pcmk_node_t *node) { pcmk__assert((rsc != NULL) && (node != NULL)); custom_action(rsc, pcmk__op_key(rsc->id, PCMK_ACTION_LRM_DELETE, 0), PCMK_ACTION_LRM_DELETE, node, FALSE, rsc->priv->scheduler); } #define sort_return(an_int, why) do { \ free(a_uuid); \ free(b_uuid); \ crm_trace("%s (%d) %c %s (%d) : %s", \ a_xml_id, a_call_id, an_int>0?'>':an_int<0?'<':'=', \ b_xml_id, b_call_id, why); \ return an_int; \ } while(0) int pe__is_newer_op(const xmlNode *xml_a, const xmlNode *xml_b) { int a_call_id = -1; int b_call_id = -1; char *a_uuid = NULL; char *b_uuid = NULL; const char *a_xml_id = crm_element_value(xml_a, PCMK_XA_ID); const char *b_xml_id = crm_element_value(xml_b, PCMK_XA_ID); const char *a_node = crm_element_value(xml_a, PCMK__META_ON_NODE); const char *b_node = crm_element_value(xml_b, PCMK__META_ON_NODE); bool same_node = pcmk__str_eq(a_node, b_node, pcmk__str_casei); if (same_node && pcmk__str_eq(a_xml_id, b_xml_id, pcmk__str_none)) { /* We have duplicate PCMK__XE_LRM_RSC_OP entries in the status * section which is unlikely to be a good thing * - we can handle it easily enough, but we need to get * to the bottom of why it's happening. */ pcmk__config_err("Duplicate " PCMK__XE_LRM_RSC_OP " entries named %s", a_xml_id); sort_return(0, "duplicate"); } crm_element_value_int(xml_a, PCMK__XA_CALL_ID, &a_call_id); crm_element_value_int(xml_b, PCMK__XA_CALL_ID, &b_call_id); if (a_call_id == -1 && b_call_id == -1) { /* both are pending ops so it doesn't matter since * stops are never pending */ sort_return(0, "pending"); } else if (same_node && a_call_id >= 0 && a_call_id < b_call_id) { sort_return(-1, "call id"); } else if (same_node && b_call_id >= 0 && a_call_id > b_call_id) { sort_return(1, "call id"); } else if (a_call_id >= 0 && b_call_id >= 0 && (!same_node || a_call_id == b_call_id)) { /* The op and last_failed_op are the same. Order on * PCMK_XA_LAST_RC_CHANGE. */ time_t last_a = -1; time_t last_b = -1; crm_element_value_epoch(xml_a, PCMK_XA_LAST_RC_CHANGE, &last_a); crm_element_value_epoch(xml_b, PCMK_XA_LAST_RC_CHANGE, &last_b); crm_trace("rc-change: %lld vs %lld", (long long) last_a, (long long) last_b); if (last_a >= 0 && last_a < last_b) { sort_return(-1, "rc-change"); } else if (last_b >= 0 && last_a > last_b) { sort_return(1, "rc-change"); } sort_return(0, "rc-change"); } else { /* One of the inputs is a pending operation. * Attempt to use PCMK__XA_TRANSITION_MAGIC to determine its age relative * to the other. */ int a_id = -1; int b_id = -1; const char *a_magic = crm_element_value(xml_a, PCMK__XA_TRANSITION_MAGIC); const char *b_magic = crm_element_value(xml_b, PCMK__XA_TRANSITION_MAGIC); CRM_CHECK(a_magic != NULL && b_magic != NULL, sort_return(0, "No magic")); if (!decode_transition_magic(a_magic, &a_uuid, &a_id, NULL, NULL, NULL, NULL)) { sort_return(0, "bad magic a"); } if (!decode_transition_magic(b_magic, &b_uuid, &b_id, NULL, NULL, NULL, NULL)) { sort_return(0, "bad magic b"); } /* try to determine the relative age of the operation... * some pending operations (e.g. a start) may have been superseded * by a subsequent stop * * [a|b]_id == -1 means it's a shutdown operation and _always_ comes last */ if (!pcmk__str_eq(a_uuid, b_uuid, pcmk__str_casei) || a_id == b_id) { /* * some of the logic in here may be redundant... * * if the UUID from the TE doesn't match then one better * be a pending operation. * pending operations don't survive between elections and joins * because we query the LRM directly */ if (b_call_id == -1) { sort_return(-1, "transition + call"); } else if (a_call_id == -1) { sort_return(1, "transition + call"); } } else if ((a_id >= 0 && a_id < b_id) || b_id == -1) { sort_return(-1, "transition"); } else if ((b_id >= 0 && a_id > b_id) || a_id == -1) { sort_return(1, "transition"); } } /* we should never end up here */ CRM_CHECK(FALSE, sort_return(0, "default")); } gint sort_op_by_callid(gconstpointer a, gconstpointer b) { return pe__is_newer_op((const xmlNode *) a, (const xmlNode *) b); } /*! * \internal * \brief Create a new pseudo-action for a resource * * \param[in,out] rsc Resource to create action for * \param[in] task Action name * \param[in] optional Whether action should be considered optional * \param[in] runnable Whethe action should be considered runnable * * \return New action object corresponding to arguments */ pcmk_action_t * pe__new_rsc_pseudo_action(pcmk_resource_t *rsc, const char *task, bool optional, bool runnable) { pcmk_action_t *action = NULL; pcmk__assert((rsc != NULL) && (task != NULL)); action = custom_action(rsc, pcmk__op_key(rsc->id, task, 0), task, NULL, optional, rsc->priv->scheduler); pcmk__set_action_flags(action, pcmk__action_pseudo); if (runnable) { pcmk__set_action_flags(action, pcmk__action_runnable); } return action; } /*! * \internal * \brief Add the expected result to an action * * \param[in,out] action Action to add expected result to * \param[in] expected_result Expected result to add * * \note This is more efficient than calling pcmk__insert_meta(). */ void pe__add_action_expected_result(pcmk_action_t *action, int expected_result) { pcmk__assert((action != NULL) && (action->meta != NULL)); g_hash_table_insert(action->meta, pcmk__str_copy(PCMK__META_OP_TARGET_RC), pcmk__itoa(expected_result)); } diff --git a/lib/pengine/status.c b/lib/pengine/status.c index 1fce0b32af..2e4deb07e1 100644 --- a/lib/pengine/status.c +++ b/lib/pengine/status.c @@ -1,546 +1,546 @@ /* * Copyright 2004-2024 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 /*! * \brief Create a new object to hold scheduler data * * \return New, initialized scheduler data on success, else NULL (and set errno) * \note Only pcmk_scheduler_t objects created with this function (as opposed * to statically declared or directly allocated) should be used with the * functions in this library, to allow for future extensions to the * data type. The caller is responsible for freeing the memory with * pe_free_working_set() when the instance is no longer needed. */ pcmk_scheduler_t * pe_new_working_set(void) { pcmk_scheduler_t *scheduler = calloc(1, sizeof(pcmk_scheduler_t)); if (scheduler == NULL) { return NULL; } scheduler->priv = calloc(1, sizeof(pcmk__scheduler_private_t)); if (scheduler->priv == NULL) { free(scheduler); return NULL; } set_working_set_defaults(scheduler); return scheduler; } /*! * \brief Free scheduler data * * \param[in,out] scheduler Scheduler data to free */ void pe_free_working_set(pcmk_scheduler_t *scheduler) { if (scheduler != NULL) { pe_reset_working_set(scheduler); free(scheduler->priv->local_node_name); free(scheduler->priv); free(scheduler); } } #define XPATH_DEPRECATED_RULES \ "//" PCMK_XE_OP_DEFAULTS "//" PCMK_XE_EXPRESSION \ "|//" PCMK_XE_OP "//" PCMK_XE_EXPRESSION /*! * \internal * \brief Log a warning for deprecated rule syntax in operations * * \param[in] scheduler Scheduler data */ static void check_for_deprecated_rules(pcmk_scheduler_t *scheduler) { // @COMPAT Drop this function when support for the syntax is dropped xmlNode *deprecated = get_xpath_object(XPATH_DEPRECATED_RULES, scheduler->input, LOG_NEVER); if (deprecated != NULL) { pcmk__warn_once(pcmk__wo_op_attr_expr, "Support for rules with node attribute expressions in " PCMK_XE_OP " or " PCMK_XE_OP_DEFAULTS " is deprecated " "and will be dropped in a future release"); } } /* * Unpack everything * At the end you'll have: * - A list of nodes * - A list of resources (each with any dependencies on other resources) * - A list of constraints between resources and nodes * - A list of constraints between start/stop actions * - A list of nodes that need to be stonith'd * - A list of nodes that need to be shutdown * - A list of the possible stop/start actions (without dependencies) */ gboolean cluster_status(pcmk_scheduler_t * scheduler) { const char *new_version = NULL; xmlNode *section = NULL; if ((scheduler == NULL) || (scheduler->input == NULL)) { return FALSE; } new_version = crm_element_value(scheduler->input, PCMK_XA_CRM_FEATURE_SET); if (pcmk__check_feature_set(new_version) != pcmk_rc_ok) { pcmk__config_err("Can't process CIB with feature set '%s' greater than our own '%s'", new_version, CRM_FEATURE_SET); return FALSE; } crm_trace("Beginning unpack"); if (scheduler->priv->failed != NULL) { pcmk__xml_free(scheduler->priv->failed); } scheduler->priv->failed = pcmk__xe_create(NULL, "failed-ops"); if (scheduler->priv->now == NULL) { scheduler->priv->now = crm_time_new(NULL); } if (pcmk__xe_attr_is_true(scheduler->input, PCMK_XA_HAVE_QUORUM)) { pcmk__set_scheduler_flags(scheduler, pcmk__sched_quorate); } else { pcmk__clear_scheduler_flags(scheduler, pcmk__sched_quorate); } scheduler->priv->op_defaults = get_xpath_object("//" PCMK_XE_OP_DEFAULTS, scheduler->input, LOG_NEVER); check_for_deprecated_rules(scheduler); scheduler->priv->rsc_defaults = get_xpath_object("//" PCMK_XE_RSC_DEFAULTS, scheduler->input, LOG_NEVER); section = get_xpath_object("//" PCMK_XE_CRM_CONFIG, scheduler->input, LOG_TRACE); unpack_config(section, scheduler); if (!pcmk_any_flags_set(scheduler->flags, pcmk__sched_location_only|pcmk__sched_quorate) && (scheduler->no_quorum_policy != pcmk_no_quorum_ignore)) { pcmk__sched_warn(scheduler, "Fencing and resource management disabled " "due to lack of quorum"); } section = get_xpath_object("//" PCMK_XE_NODES, scheduler->input, LOG_TRACE); unpack_nodes(section, scheduler); section = get_xpath_object("//" PCMK_XE_RESOURCES, scheduler->input, LOG_TRACE); if (!pcmk_is_set(scheduler->flags, pcmk__sched_location_only)) { unpack_remote_nodes(section, scheduler); } unpack_resources(section, scheduler); section = get_xpath_object("//" PCMK_XE_FENCING_TOPOLOGY, scheduler->input, LOG_TRACE); pcmk__validate_fencing_topology(section); section = get_xpath_object("//" PCMK_XE_TAGS, scheduler->input, LOG_NEVER); unpack_tags(section, scheduler); if (!pcmk_is_set(scheduler->flags, pcmk__sched_location_only)) { section = get_xpath_object("//" PCMK_XE_STATUS, scheduler->input, LOG_TRACE); unpack_status(section, scheduler); } if (!pcmk_is_set(scheduler->flags, pcmk__sched_no_counts)) { for (GList *item = scheduler->priv->resources; item != NULL; item = item->next) { pcmk_resource_t *rsc = item->data; rsc->priv->fns->count(item->data); } crm_trace("Cluster resource count: %d (%d disabled, %d blocked)", scheduler->priv->ninstances, scheduler->priv->disabled_resources, scheduler->priv->blocked_resources); } if ((scheduler->priv->local_node_name != NULL) && (pcmk_find_node(scheduler, scheduler->priv->local_node_name) == NULL)) { crm_info("Creating a fake local node for %s", scheduler->priv->local_node_name); pe_create_node(scheduler->priv->local_node_name, scheduler->priv->local_node_name, NULL, 0, scheduler); } pcmk__set_scheduler_flags(scheduler, pcmk__sched_have_status); return TRUE; } /*! * \internal * \brief Free a list of pcmk_resource_t * * \param[in,out] resources List to free * * \note When the scheduler's resource list is freed, that includes the original * storage for the uname and id of any Pacemaker Remote nodes in the * scheduler's node list, so take care not to use those afterward. * \todo Refactor pcmk_node_t to strdup() the node name. */ static void pe_free_resources(GList *resources) { pcmk_resource_t *rsc = NULL; GList *iterator = resources; while (iterator != NULL) { rsc = (pcmk_resource_t *) iterator->data; iterator = iterator->next; rsc->priv->fns->free(rsc); } if (resources != NULL) { g_list_free(resources); } } static void pe_free_actions(GList *actions) { GList *iterator = actions; while (iterator != NULL) { pe_free_action(iterator->data); iterator = iterator->next; } if (actions != NULL) { g_list_free(actions); } } static void pe_free_nodes(GList *nodes) { for (GList *iterator = nodes; iterator != NULL; iterator = iterator->next) { pcmk_node_t *node = (pcmk_node_t *) iterator->data; // Shouldn't be possible, but to be safe ... if (node == NULL) { continue; } if (node->details == NULL) { free(node); continue; } /* This is called after pe_free_resources(), which means that we can't * use node->private->name for Pacemaker Remote nodes. */ crm_trace("Freeing node %s", (pcmk__is_pacemaker_remote_node(node)? "(guest or remote)" : pcmk__node_name(node))); if (node->priv->attrs != NULL) { g_hash_table_destroy(node->priv->attrs); } if (node->priv->utilization != NULL) { g_hash_table_destroy(node->priv->utilization); } if (node->priv->digest_cache != NULL) { g_hash_table_destroy(node->priv->digest_cache); } g_list_free(node->details->running_rsc); g_list_free(node->priv->assigned_resources); free(node->priv); free(node->details); free(node->assign); free(node); } if (nodes != NULL) { g_list_free(nodes); } } static void pe__free_ordering(GList *constraints) { GList *iterator = constraints; while (iterator != NULL) { pcmk__action_relation_t *order = iterator->data; iterator = iterator->next; free(order->task1); free(order->task2); free(order); } if (constraints != NULL) { g_list_free(constraints); } } static void pe__free_location(GList *constraints) { GList *iterator = constraints; while (iterator != NULL) { pcmk__location_t *cons = iterator->data; iterator = iterator->next; - g_list_free_full(cons->nodes, free); + g_list_free_full(cons->nodes, pcmk__free_node_copy); free(cons->id); free(cons); } if (constraints != NULL) { g_list_free(constraints); } } /*! * \brief Reset scheduler data to defaults without freeing it or constraints * * \param[in,out] scheduler Scheduler data to reset * * \deprecated This function is deprecated as part of the API; * pe_reset_working_set() should be used instead. */ void cleanup_calculations(pcmk_scheduler_t *scheduler) { if (scheduler == NULL) { return; } pcmk__clear_scheduler_flags(scheduler, pcmk__sched_have_status); if (scheduler->priv->options != NULL) { g_hash_table_destroy(scheduler->priv->options); } if (scheduler->priv->singletons != NULL) { g_hash_table_destroy(scheduler->priv->singletons); } if (scheduler->priv->ticket_constraints != NULL) { g_hash_table_destroy(scheduler->priv->ticket_constraints); } if (scheduler->priv->templates != NULL) { g_hash_table_destroy(scheduler->priv->templates); } if (scheduler->priv->tags != NULL) { g_hash_table_destroy(scheduler->priv->tags); } crm_trace("deleting resources"); pe_free_resources(scheduler->priv->resources); crm_trace("deleting actions"); pe_free_actions(scheduler->priv->actions); crm_trace("deleting nodes"); pe_free_nodes(scheduler->nodes); pe__free_param_checks(scheduler); g_list_free(scheduler->priv->stop_needed); crm_time_free(scheduler->priv->now); pcmk__xml_free(scheduler->input); pcmk__xml_free(scheduler->priv->failed); pcmk__xml_free(scheduler->priv->graph); set_working_set_defaults(scheduler); CRM_LOG_ASSERT((scheduler->priv->location_constraints == NULL) && (scheduler->priv->ordering_constraints == NULL)); } /*! * \brief Reset scheduler data to default state without freeing it * * \param[in,out] scheduler Scheduler data to reset */ void pe_reset_working_set(pcmk_scheduler_t *scheduler) { if (scheduler == NULL) { return; } crm_trace("Deleting %d ordering constraints", g_list_length(scheduler->priv->ordering_constraints)); pe__free_ordering(scheduler->priv->ordering_constraints); scheduler->priv->ordering_constraints = NULL; crm_trace("Deleting %d location constraints", g_list_length(scheduler->priv->location_constraints)); pe__free_location(scheduler->priv->location_constraints); scheduler->priv->location_constraints = NULL; crm_trace("Deleting %d colocation constraints", g_list_length(scheduler->priv->colocation_constraints)); g_list_free_full(scheduler->priv->colocation_constraints, free); scheduler->priv->colocation_constraints = NULL; cleanup_calculations(scheduler); } void set_working_set_defaults(pcmk_scheduler_t *scheduler) { // These members must be preserved pcmk__scheduler_private_t *priv = scheduler->priv; pcmk__output_t *out = priv->out; char *local_node_name = scheduler->priv->local_node_name; // Wipe the main structs (any other members must have previously been freed) memset(scheduler, 0, sizeof(pcmk_scheduler_t)); memset(priv, 0, sizeof(pcmk__scheduler_private_t)); // Restore the members to preserve scheduler->priv = priv; scheduler->priv->out = out; scheduler->priv->local_node_name = local_node_name; // Set defaults for everything else scheduler->priv->next_ordering_id = 1; scheduler->priv->next_action_id = 1; scheduler->no_quorum_policy = pcmk_no_quorum_stop; #if PCMK__CONCURRENT_FENCING_DEFAULT_TRUE pcmk__set_scheduler_flags(scheduler, pcmk__sched_symmetric_cluster |pcmk__sched_concurrent_fencing |pcmk__sched_stop_removed_resources |pcmk__sched_cancel_removed_actions); #else pcmk__set_scheduler_flags(scheduler, pcmk__sched_symmetric_cluster |pcmk__sched_stop_removed_resources |pcmk__sched_cancel_removed_actions); #endif } pcmk_resource_t * pe_find_resource(GList *rsc_list, const char *id) { return pe_find_resource_with_flags(rsc_list, id, pcmk_rsc_match_history); } pcmk_resource_t * pe_find_resource_with_flags(GList *rsc_list, const char *id, enum pe_find flags) { GList *rIter = NULL; for (rIter = rsc_list; id && rIter; rIter = rIter->next) { pcmk_resource_t *parent = rIter->data; pcmk_resource_t *match = parent->priv->fns->find_rsc(parent, id, NULL, flags); if (match != NULL) { return match; } } crm_trace("No match for %s", id); return NULL; } /*! * \brief Find a node by name or ID in a list of nodes * * \param[in] nodes List of nodes (as pcmk_node_t*) * \param[in] id If not NULL, ID of node to find * \param[in] node_name If not NULL, name of node to find * * \return Node from \p nodes that matches \p id if any, * otherwise node from \p nodes that matches \p uname if any, * otherwise NULL */ pcmk_node_t * pe_find_node_any(const GList *nodes, const char *id, const char *uname) { pcmk_node_t *match = NULL; if (id != NULL) { match = pe_find_node_id(nodes, id); } if ((match == NULL) && (uname != NULL)) { match = pcmk__find_node_in_list(nodes, uname); } return match; } /*! * \brief Find a node by ID in a list of nodes * * \param[in] nodes List of nodes (as pcmk_node_t*) * \param[in] id ID of node to find * * \return Node from \p nodes that matches \p id if any, otherwise NULL */ pcmk_node_t * pe_find_node_id(const GList *nodes, const char *id) { for (const GList *iter = nodes; iter != NULL; iter = iter->next) { pcmk_node_t *node = (pcmk_node_t *) iter->data; /* @TODO Whether node IDs should be considered case-sensitive should * probably depend on the node type, so functionizing the comparison * would be worthwhile */ if (pcmk__str_eq(node->priv->id, id, pcmk__str_casei)) { return node; } } return NULL; } // Deprecated functions kept only for backward API compatibility // LCOV_EXCL_START #include /*! * \brief Find a node by name in a list of nodes * * \param[in] nodes List of nodes (as pcmk_node_t*) * \param[in] node_name Name of node to find * * \return Node from \p nodes that matches \p node_name if any, otherwise NULL */ pcmk_node_t * pe_find_node(const GList *nodes, const char *node_name) { return pcmk__find_node_in_list(nodes, node_name); } // LCOV_EXCL_STOP // End deprecated API diff --git a/lib/pengine/utils.c b/lib/pengine/utils.c index 4e7cf5e6ac..1197819c18 100644 --- a/lib/pengine/utils.c +++ b/lib/pengine/utils.c @@ -1,925 +1,927 @@ /* * Copyright 2004-2024 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; 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) { 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 node hash table from a node list + * \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, free); + 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)); - free(rsc->priv->assigned_node); + pcmk__free_node_copy(rsc->priv->assigned_node); rsc->priv->assigned_node = NULL; } } time_t get_effective_time(pcmk_scheduler_t *scheduler) { if(scheduler) { if (scheduler->priv->now == NULL) { crm_trace("Recording a new 'now'"); scheduler->priv->now = crm_time_new(NULL); } return crm_time_get_seconds_since_epoch(scheduler->priv->now); } crm_trace("Defaulting to 'now'"); return time(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 Update "recheck by" time in scheduler data * * \param[in] recheck Epoch time when recheck should happen * \param[in,out] scheduler Scheduler data * \param[in] reason What time is being updated for (for logs) */ void pe__update_recheck_time(time_t recheck, pcmk_scheduler_t *scheduler, const char *reason) { if ((recheck > get_effective_time(scheduler)) && ((scheduler->priv->recheck_by == 0) || (scheduler->priv->recheck_by > recheck))) { scheduler->priv->recheck_by = recheck; crm_debug("Updated next scheduler recheck to %s for %s", pcmk__trim(ctime(&recheck)), reason); } } /*! * \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_data Matching parameters to use when unpacking * (node_hash 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 pe_rule_eval_data_t *rule_data, GHashTable *hash, const char *always_first, pcmk_scheduler_t *scheduler) { crm_time_t *next_change = NULL; CRM_CHECK((set_name != NULL) && (rule_data != NULL) && (hash != NULL) && (scheduler != NULL), return); // Node attribute expressions are not allowed for meta-attributes CRM_CHECK((rule_data->node_hash == NULL) || (strcmp(set_name, PCMK_XE_META_ATTRIBUTES) != 0), return); if (xml_obj == NULL) { return; } next_change = crm_time_new_undefined(); pe_eval_nvpairs(scheduler->input, xml_obj, set_name, rule_data, hash, always_first, FALSE, next_change); if (crm_time_is_defined(next_change)) { time_t recheck = (time_t) crm_time_get_seconds_since_epoch(next_change); pe__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 = crm_element_value(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 0e991e4533..ff4763a20c 100644 --- a/lib/services/services_linux.c +++ b/lib/services/services_linux.c @@ -1,1477 +1,1478 @@ /* * Copyright 2010-2024 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 %lld", op->id, (long long) 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 %lld", op->id, (long long) 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 %lld bytes of %s %s: %.80s", (long long) 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 %lld bytes (discarded %lld)", op->id, out_type(is_stderr), (long long) len, (long long) 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) { if (setenv(key, value, 1) != 0) { crm_perror(LOG_ERR, "setenv failed for key:%s and value:%s", key, value); } } static void set_ocf_env_with_prefix(gpointer key, gpointer value, gpointer user_data) { char buffer[500]; snprintf(buffer, sizeof(buffer), strcmp(key, "OCF_CHECK_LEVEL") != 0 ? "OCF_RESKEY_%s" : "%s", (char *)key); set_ocf_env(buffer, value, user_data); } 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 : "")); } else { crm_trace("setenv %s=%s", (char*)key, (value? (char*)value : "")); } } /*! * \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); /* 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. */ #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; int entries = 0, lpc = 0; char buffer[PATH_MAX]; entries = scandir(root, &namelist, NULL, alphasort); if (entries <= 0) { return list; } for (lpc = 0; lpc < entries; lpc++) { struct stat sb; if ('.' == namelist[lpc]->d_name[0]) { free(namelist[lpc]); continue; } snprintf(buffer, sizeof(buffer), "%s/%s", root, namelist[lpc]->d_name); if (stat(buffer, &sb)) { 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/maint/Makefile.am b/maint/Makefile.am index 4d3174409e..fd5373f75d 100644 --- a/maint/Makefile.am +++ b/maint/Makefile.am @@ -1,82 +1,82 @@ # # Copyright 2019-2024 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. # # Define release-related variables include $(top_srcdir)/mk/release.mk include $(top_srcdir)/mk/common.mk noinst_SCRIPTS = bumplibs EXTRA_DIST = README # # Change log generation # # Count changes in these directories CHANGELOG_DIRS = ../include \ ../lib \ ../daemons \ ../tools \ ../xml .PHONY: require_last_release require_last_release: @if [ -z "$(CHECKOUT)" ]; then \ echo "This target must be run from a git checkout"; \ exit 1; \ elif ! "$(GIT)" rev-parse $(LAST_RELEASE) >/dev/null 2>&1; then \ echo "LAST_RELEASE must be set to a valid git tag"; \ exit 1; \ fi .PHONY: summary summary: require_last_release - @printf "# Pacemaker-%s (%s)\n* %d commits with%s\n" \ + @printf "# %s (%s)\n* %d commits with%s\n" \ "$(NEXT_RELEASE)" "$$(date +'%d %b %Y')" \ "$$("$(GIT)" log --pretty=oneline --no-merges \ $(LAST_RELEASE)..HEAD | wc -l)" \ "$$("$(GIT)" diff $(LAST_RELEASE)..HEAD --shortstat \ $(CHANGELOG_DIRS))" .PHONY: changes changes: summary @printf "\n## Features added since $(LAST_RELEASE)\n\n" @"$(GIT)" log --pretty=format:'%s' --no-merges \ --abbrev-commit $(LAST_RELEASE)..HEAD \ | sed -n -e 's/^ *Feature: */* /p' | sort -uf \ | sed -e 's/^\( *[-+*] \)\([^:]*:\)\(.*\)$$/\1**\2**\3/' \ -e 's/\([ (]\)\([A-Za-z0-9]*_[^ ,]*\)/\1`\2`/g' @printf "\n## Fixes since $(LAST_RELEASE)\n\n" @"$(GIT)" log --pretty=format:'%s' --no-merges \ --abbrev-commit $(LAST_RELEASE)..HEAD \ | sed -n -e 's/^ *\(Fix\|High\|Bug\): */* /p' \ | sed -e 's/\(\(pacemaker-\)?based\):/CIB:/' \ -e 's/\(\(pacemaker-\)?execd\):/executor:/' \ -e 's/\(\(pacemaker-\)?controld\):/controller:/' \ -e 's/\(\(pacemaker-\)?fenced\):/fencing:/' \ | sort -uf \ | sed -e 's/^\( *[-+*] \)\([^:]*:\)\(.*\)$$/\1**\2**\3/' \ -e 's/\([ (]\)\([A-Za-z0-9]*_[^ ,]*\)/\1`\2`/g' @printf "\n## Public API changes since $(LAST_RELEASE)\n\n" @"$(GIT)" log --pretty=format:'%s' --no-merges \ --abbrev-commit $(LAST_RELEASE)..HEAD \ | sed -n -e 's/^ *API: */* /p' | sort -uf \ | sed -e 's/^\( *[-+*] \)\([^:]*:\)\(.*\)$$/\1**\2**\3/' \ -e 's/\([ (]\)\([A-Za-z0-9]*_[^ ,]*\)/\1`\2`/g' .PHONY: changelog changelog: require_last_release @printf "%s\n\n%s\n" \ "$$($(MAKE) $(AM_MAKEFLAGS) changes \ | grep -v 'make\(\[[0-9]*\]\)\?:')" \ "$$(cat ../ChangeLog.md)" > ../ChangeLog.md .PHONY: authors authors: require_last_release "$(GIT)" log $(LAST_RELEASE)..$(COMMIT) --format='%an' | sort -u