diff --git a/.gitignore b/.gitignore index 847e48490c..03df200f01 100644 --- a/.gitignore +++ b/.gitignore @@ -1,243 +1,242 @@ # Common \#* .\#* GPATH GRTAGS GTAGS TAGS Makefile Makefile.in .deps .dirstamp .libs *.pc *.pyc *.bz2 *.tar.gz *.tgz *.la *.lo *.o *~ *.gcda *.gcno # Autobuild aclocal.m4 autoconf autoheader autom4te.cache/ automake build.counter compile /confdefs.h config.guess config.log config.status config.sub configure /conftest* depcomp install-sh include/stamp-* libtool libtool.m4 ltdl.m4 libltdl ltmain.sh missing py-compile /m4/argz.m4 /m4/ltargz.m4 /m4/ltoptions.m4 /m4/ltsugar.m4 /m4/ltversion.m4 /m4/lt~obsolete.m4 test-driver ylwrap # Configure targets /cts/CTS.py /cts/CTSlab.py /cts/CTSvars.py /cts/LSBDummy /cts/OCFIPraTest.py /cts/benchmark/clubench /cts/cluster_test /cts/cts /cts/cts-cli /cts/cts-coverage /cts/cts-exec /cts/cts-fencing /cts/cts-log-watcher /cts/cts-regression /cts/cts-scheduler /cts/cts-support /cts/fence_dummy /cts/lxc_autogen.sh /cts/pacemaker-cts-dummyd /cts/pacemaker-cts-dummyd@.service /daemons/execd/pacemaker_remote /daemons/execd/pacemaker_remote.service /daemons/fenced/fence_legacy /daemons/pacemakerd/pacemaker /daemons/pacemakerd/pacemaker.combined.upstart /daemons/pacemakerd/pacemaker.service /daemons/pacemakerd/pacemaker.upstart /doc/Doxyfile /extra/logrotate/pacemaker /extra/resources/ClusterMon /extra/resources/HealthSMART /extra/resources/SysInfo /extra/resources/ifspeed /extra/resources/o2cb include/config.h include/config.h.in include/crm_config.h publican.cfg /tools/cibsecret /tools/crm_error /tools/crm_failcount /tools/crm_master /tools/crm_mon.service /tools/crm_mon.upstart /tools/crm_report /tools/crm_rule /tools/crm_standby /tools/pcmk_simtimes /tools/report.collector /tools/report.common # Build targets *.7 *.7.xml *.7.html *.8 *.8.xml *.8.html /daemons/attrd/pacemaker-attrd /daemons/based/pacemaker-based /daemons/based/cibmon /daemons/controld/pacemaker-controld /daemons/execd/cts-exec-helper /daemons/execd/pacemaker-execd /daemons/execd/pacemaker-remoted /daemons/fenced/cts-fence-helper /daemons/fenced/pacemaker-fenced /daemons/fenced/pacemaker-fenced.xml /daemons/pacemakerd/pacemakerd /daemons/schedulerd/pacemaker-schedulerd /daemons/schedulerd/pacemaker-schedulerd.xml /doc/*/tmp/** /doc/*/publish /doc/*.build /doc/*/en-US/Ap-*.xml /doc/*/en-US/Ch-*.xml /doc/.ABI-build /doc/HTML /doc/abi_dumps /doc/abi-check /doc/api/* /doc/compat_reports /doc/crm_fencing.html /doc/publican-catalog* /doc/shared/en-US/*.xml /doc/shared/en-US/images/pcmk-*.png /doc/shared/en-US/images/Policy-Engine-*.png /doc/sphinx/*/_build /doc/sphinx/*/conf.py /lib/common/md5.c /maint/testcc_helper.cc /maint/testcc_*_h /maint/mocked/based scratch /tools/attrd_updater /tools/cibadmin /tools/crmadmin /tools/crm_attribute /tools/crm_diff /tools/crm_mon /tools/crm_node /tools/crm_resource /tools/crm_shadow /tools/crm_simulate /tools/crm_ticket /tools/crm_verify /tools/iso8601 /tools/stonith_admin xml/crm.dtd xml/pacemaker*.rng xml/versions.rng xml/api/api-result*.rng lib/gnu/libgnu.a lib/gnu/stdalign.h *.coverity # Packager artifacts *.rpm /mock /pacemaker.spec /rpm/[A-Z]* # make dist/export working directory pacemaker-[a-f0-9][a-f0-9][a-f0-9][a-f0-9][a-f0-9][a-f0-9][a-f0-9] # Test detritus /cts/.regression.failed.diff /cts/scheduler/*.ref /cts/scheduler/*.up /cts/scheduler/*.up.err /cts/scheduler/bug-rh-1097457.log /cts/scheduler/bug-rh-1097457.trs /cts/scheduler/shadow.* /cts/test-suite.log /xml/test-*/*.up /xml/test-*/*.up.err /xml/assets/*.rng /xml/assets/diffview.js /xml/assets/xmlcatalog # Test results *.log *.trs /lib/common/tests/strings/pcmk__parse_ll_range /lib/common/tests/strings/pcmk__str_any_of -/lib/common/tests/strings/pcmk__str_none_of /lib/common/tests/utils/pcmk_str_is_infinity /lib/common/tests/utils/pcmk_str_is_minus_infinity /lib/pengine/tests/rules/pe_cron_range_satisfied # Release maintenance detritus /maint/gnulib # Formerly built files (helps when jumping back and forth in checkout) /.ABI-build /Doxyfile /HTML /abi_dumps /abi-check /compat_reports /attrd /cib /coverage.sh /crmd /cts/HBDummy /doc/Clusters_from_Scratch.txt /doc/Pacemaker_Explained.txt /doc/acls.html /fencing /lrmd /mcp /pacemaker-*.spec /pengine #Other coverity-* logs *.patch *.diff *.sed *.orig *.rej *.swp diff --git a/daemons/controld/controld_based.c b/daemons/controld/controld_based.c index 919f590841..f2b25897c8 100644 --- a/daemons/controld/controld_based.c +++ b/daemons/controld/controld_based.c @@ -1,331 +1,331 @@ /* * Copyright 2004-2020 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include /* sleep */ #include #include #include #include #include int cib_retries = 0; static void do_cib_updated(const char *event, xmlNode * msg) { if (pcmk__alert_in_patchset(msg, TRUE)) { mainloop_set_trigger(config_read); } } static void do_cib_replaced(const char *event, xmlNode * msg) { crm_debug("Updating the CIB after a replace: DC=%s", AM_I_DC ? "true" : "false"); if (AM_I_DC == FALSE) { return; } else if (fsa_state == S_FINALIZE_JOIN && is_set(fsa_input_register, R_CIB_ASKED)) { /* no need to restart the join - we asked for this replace op */ return; } /* start the join process again so we get everyone's LRM status */ populate_cib_nodes(node_update_quick|node_update_all, __FUNCTION__); register_fsa_input(C_FSA_INTERNAL, I_ELECTION, NULL); } /* A_CIB_STOP, A_CIB_START, O_CIB_RESTART */ void do_cib_control(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_ASSERT(fsa_cib_conn != NULL); if (action & A_CIB_STOP) { if (fsa_cib_conn->state != cib_disconnected && last_resource_update != 0) { crm_info("Waiting for resource update %d to complete", last_resource_update); crmd_fsa_stall(FALSE); return; } crm_info("Disconnecting from the CIB manager"); clear_bit(fsa_input_register, R_CIB_CONNECTED); fsa_cib_conn->cmds->del_notify_callback(fsa_cib_conn, T_CIB_DIFF_NOTIFY, do_cib_updated); if (fsa_cib_conn->state != cib_disconnected) { fsa_cib_conn->cmds->set_slave(fsa_cib_conn, cib_scope_local); fsa_cib_conn->cmds->signoff(fsa_cib_conn); } crm_notice("Disconnected from the CIB manager"); } if (action & A_CIB_START) { int rc = pcmk_ok; if (cur_state == S_STOPPING) { crm_err("Ignoring request to connect to the CIB manager after shutdown"); return; } rc = fsa_cib_conn->cmds->signon(fsa_cib_conn, CRM_SYSTEM_CRMD, cib_command_nonblocking); if (rc != pcmk_ok) { /* a short wait that usually avoids stalling the FSA */ sleep(1); rc = fsa_cib_conn->cmds->signon(fsa_cib_conn, CRM_SYSTEM_CRMD, cib_command_nonblocking); } if (rc != pcmk_ok) { crm_info("Could not connect to the CIB manager: %s", pcmk_strerror(rc)); } else if (pcmk_ok != fsa_cib_conn->cmds->set_connection_dnotify(fsa_cib_conn, crmd_cib_connection_destroy)) { crm_err("Could not set dnotify callback"); } else if (pcmk_ok != fsa_cib_conn->cmds->add_notify_callback(fsa_cib_conn, T_CIB_REPLACE_NOTIFY, do_cib_replaced)) { crm_err("Could not set CIB notification callback (replace)"); } else if (pcmk_ok != fsa_cib_conn->cmds->add_notify_callback(fsa_cib_conn, T_CIB_DIFF_NOTIFY, do_cib_updated)) { crm_err("Could not set CIB notification callback (update)"); } else { set_bit(fsa_input_register, R_CIB_CONNECTED); cib_retries = 0; } if (is_not_set(fsa_input_register, R_CIB_CONNECTED)) { cib_retries++; crm_warn("Couldn't complete CIB registration %d" " times... pause and retry", cib_retries); if (cib_retries < 30) { controld_start_timer(wait_timer); crmd_fsa_stall(FALSE); } else { crm_err("Could not complete CIB" " registration %d times..." " hard error", cib_retries); register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); } } } } /*! * \internal * \brief Get CIB call options to use local scope if master unavailable * * \return CIB call options */ int crmd_cib_smart_opt() { int call_opt = cib_quorum_override; if (fsa_state == S_ELECTION || fsa_state == S_PENDING) { crm_info("Sending update to local CIB in state: %s", fsa_state2string(fsa_state)); call_opt |= cib_scope_local; } return call_opt; } /*! * \internal * \brief Check whether an action type should be recorded in the CIB * * \param[in] action Action type * * \return TRUE if action should be recorded, FALSE otherwise */ bool controld_action_is_recordable(const char *action) { - return pcmk__str_none_of(action, CRMD_ACTION_CANCEL, CRMD_ACTION_DELETE, + return !pcmk__str_any_of(action, CRMD_ACTION_CANCEL, CRMD_ACTION_DELETE, CRMD_ACTION_NOTIFY, CRMD_ACTION_METADATA, NULL); } static void cib_delete_callback(xmlNode *msg, int call_id, int rc, xmlNode *output, void *user_data) { char *desc = user_data; if (rc == 0) { crm_debug("Deletion of %s (via CIB call %d) succeeded", desc, call_id); } else { crm_warn("Deletion of %s (via CIB call %d) failed: %s " CRM_XS " rc=%d", desc, call_id, pcmk_strerror(rc), rc); } } // Searches for various portions of node_state to delete // Match a particular node's node_state (takes node name 1x) #define XPATH_NODE_STATE "//" XML_CIB_TAG_STATE "[@" XML_ATTR_UNAME "='%s']" // Node's lrm section (name 1x) #define XPATH_NODE_LRM XPATH_NODE_STATE "/" XML_CIB_TAG_LRM // Node's lrm_rsc_op entries and lrm_resource entries without lock (name 2x) #define XPATH_NODE_LRM_UNLOCKED XPATH_NODE_STATE "//" XML_LRM_TAG_RSC_OP \ "|" XPATH_NODE_STATE \ "//" XML_LRM_TAG_RESOURCE \ "[not(@" XML_CONFIG_ATTR_SHUTDOWN_LOCK ")]" // Node's transient_attributes section (name 1x) #define XPATH_NODE_ATTRS XPATH_NODE_STATE "/" XML_TAG_TRANSIENT_NODEATTRS // Everything under node_state (name 1x) #define XPATH_NODE_ALL XPATH_NODE_STATE "/*" // Unlocked history + transient attributes (name 3x) #define XPATH_NODE_ALL_UNLOCKED XPATH_NODE_LRM_UNLOCKED "|" XPATH_NODE_ATTRS /*! * \internal * \brief Delete subsection of a node's CIB node_state * * \param[in] uname Desired node * \param[in] section Subsection of node_state to delete * \param[in] options CIB call options to use */ void controld_delete_node_state(const char *uname, enum controld_section_e section, int options) { char *xpath = NULL; char *desc = NULL; CRM_CHECK(uname != NULL, return); switch (section) { case controld_section_lrm: xpath = crm_strdup_printf(XPATH_NODE_LRM, uname); desc = crm_strdup_printf("resource history for node %s", uname); break; case controld_section_lrm_unlocked: xpath = crm_strdup_printf(XPATH_NODE_LRM_UNLOCKED, uname, uname); desc = crm_strdup_printf("resource history (other than shutdown " "locks) for node %s", uname); break; case controld_section_attrs: xpath = crm_strdup_printf(XPATH_NODE_ATTRS, uname); desc = crm_strdup_printf("transient attributes for node %s", uname); break; case controld_section_all: xpath = crm_strdup_printf(XPATH_NODE_ALL, uname); desc = crm_strdup_printf("all state for node %s", uname); break; case controld_section_all_unlocked: xpath = crm_strdup_printf(XPATH_NODE_ALL_UNLOCKED, uname, uname, uname); desc = crm_strdup_printf("all state (other than shutdown locks) " "for node %s", uname); break; } if (fsa_cib_conn == NULL) { crm_warn("Unable to delete %s: no CIB connection", desc); free(desc); } else { int call_id; options |= cib_quorum_override|cib_xpath|cib_multiple; call_id = fsa_cib_conn->cmds->remove(fsa_cib_conn, xpath, NULL, options); crm_info("Deleting %s (via CIB call %d) " CRM_XS " xpath=%s", desc, call_id, xpath); fsa_register_cib_callback(call_id, FALSE, desc, cib_delete_callback); // CIB library handles freeing desc } free(xpath); } // Takes node name and resource ID #define XPATH_RESOURCE_HISTORY "//" XML_CIB_TAG_STATE \ "[@" XML_ATTR_UNAME "='%s']/" \ XML_CIB_TAG_LRM "/" XML_LRM_TAG_RESOURCES \ "/" XML_LRM_TAG_RESOURCE \ "[@" XML_ATTR_ID "='%s']" // @TODO could add "and @XML_CONFIG_ATTR_SHUTDOWN_LOCK" to limit to locks /*! * \internal * \brief Clear resource history from CIB for a given resource and node * * \param[in] rsc_id ID of resource to be cleared * \param[in] node Node whose resource history should be cleared * \param[in] user_name ACL user name to use * \param[in] call_options CIB call options * * \return Standard Pacemaker return code */ int controld_delete_resource_history(const char *rsc_id, const char *node, const char *user_name, int call_options) { char *desc = NULL; char *xpath = NULL; int rc = pcmk_rc_ok; CRM_CHECK((rsc_id != NULL) && (node != NULL), return EINVAL); desc = crm_strdup_printf("resource history for %s on %s", rsc_id, node); if (fsa_cib_conn == NULL) { crm_err("Unable to clear %s: no CIB connection", desc); free(desc); return ENOTCONN; } // Ask CIB to delete the entry xpath = crm_strdup_printf(XPATH_RESOURCE_HISTORY, node, rsc_id); rc = cib_internal_op(fsa_cib_conn, CIB_OP_DELETE, NULL, xpath, NULL, NULL, call_options|cib_xpath, user_name); if (rc < 0) { rc = pcmk_legacy2rc(rc); crm_err("Could not delete resource status of %s on %s%s%s: %s " CRM_XS " rc=%d", rsc_id, node, (user_name? " for user " : ""), (user_name? user_name : ""), pcmk_rc_str(rc), rc); free(desc); free(xpath); return rc; } if (is_set(call_options, cib_sync_call)) { if (is_set(call_options, cib_dryrun)) { crm_debug("Deletion of %s would succeed", desc); } else { crm_debug("Deletion of %s succeeded", desc); } free(desc); } else { crm_info("Clearing %s (via CIB call %d) " CRM_XS " xpath=%s", desc, rc, xpath); fsa_register_cib_callback(rc, FALSE, desc, cib_delete_callback); // CIB library handles freeing desc } free(xpath); return pcmk_rc_ok; } diff --git a/daemons/controld/controld_execd.c b/daemons/controld/controld_execd.c index 1407b86f84..0c26202bc7 100644 --- a/daemons/controld/controld_execd.c +++ b/daemons/controld/controld_execd.c @@ -1,2855 +1,2855 @@ /* * Copyright 2004-2020 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include // lrmd_event_data_t, lrmd_rsc_info_t, etc. #include #include #include #include #include #include #define START_DELAY_THRESHOLD 5 * 60 * 1000 #define MAX_LRM_REG_FAILS 30 struct delete_event_s { int rc; const char *rsc; lrm_state_t *lrm_state; }; static gboolean is_rsc_active(lrm_state_t * lrm_state, const char *rsc_id); static gboolean build_active_RAs(lrm_state_t * lrm_state, xmlNode * rsc_list); static gboolean stop_recurring_actions(gpointer key, gpointer value, gpointer user_data); static lrmd_event_data_t *construct_op(lrm_state_t * lrm_state, xmlNode * rsc_op, const char *rsc_id, const char *operation); static void do_lrm_rsc_op(lrm_state_t *lrm_state, lrmd_rsc_info_t *rsc, const char *operation, xmlNode *msg); static gboolean lrm_state_verify_stopped(lrm_state_t * lrm_state, enum crmd_fsa_state cur_state, int log_level); static int do_update_resource(const char *node_name, lrmd_rsc_info_t *rsc, lrmd_event_data_t *op, time_t lock_time); static void lrm_connection_destroy(void) { if (is_set(fsa_input_register, R_LRM_CONNECTED)) { crm_crit("Connection to executor failed"); register_fsa_input(C_FSA_INTERNAL, I_ERROR, NULL); clear_bit(fsa_input_register, R_LRM_CONNECTED); } else { crm_info("Disconnected from executor"); } } static char * make_stop_id(const char *rsc, int call_id) { return crm_strdup_printf("%s:%d", rsc, call_id); } static void copy_instance_keys(gpointer key, gpointer value, gpointer user_data) { if (strstr(key, CRM_META "_") == NULL) { g_hash_table_replace(user_data, strdup((const char *)key), strdup((const char *)value)); } } static void copy_meta_keys(gpointer key, gpointer value, gpointer user_data) { if (strstr(key, CRM_META "_") != NULL) { g_hash_table_replace(user_data, strdup((const char *)key), strdup((const char *)value)); } } /*! * \internal * \brief Remove a recurring operation from a resource's history * * \param[in,out] history Resource history to modify * \param[in] op Operation to remove * * \return TRUE if the operation was found and removed, FALSE otherwise */ static gboolean history_remove_recurring_op(rsc_history_t *history, const lrmd_event_data_t *op) { GList *iter; for (iter = history->recurring_op_list; iter != NULL; iter = iter->next) { lrmd_event_data_t *existing = iter->data; if ((op->interval_ms == existing->interval_ms) && crm_str_eq(op->rsc_id, existing->rsc_id, TRUE) && safe_str_eq(op->op_type, existing->op_type)) { history->recurring_op_list = g_list_delete_link(history->recurring_op_list, iter); lrmd_free_event(existing); return TRUE; } } return FALSE; } /*! * \internal * \brief Free all recurring operations in resource history * * \param[in,out] history Resource history to modify */ static void history_free_recurring_ops(rsc_history_t *history) { GList *iter; for (iter = history->recurring_op_list; iter != NULL; iter = iter->next) { lrmd_free_event(iter->data); } g_list_free(history->recurring_op_list); history->recurring_op_list = NULL; } /*! * \internal * \brief Free resource history * * \param[in,out] history Resource history to free */ void history_free(gpointer data) { rsc_history_t *history = (rsc_history_t*)data; if (history->stop_params) { g_hash_table_destroy(history->stop_params); } /* Don't need to free history->rsc.id because it's set to history->id */ free(history->rsc.type); free(history->rsc.standard); free(history->rsc.provider); lrmd_free_event(history->failed); lrmd_free_event(history->last); free(history->id); history_free_recurring_ops(history); free(history); } static void update_history_cache(lrm_state_t * lrm_state, lrmd_rsc_info_t * rsc, lrmd_event_data_t * op) { int target_rc = 0; rsc_history_t *entry = NULL; if (op->rsc_deleted) { crm_debug("Purged history for '%s' after %s", op->rsc_id, op->op_type); controld_delete_resource_history(op->rsc_id, lrm_state->node_name, NULL, crmd_cib_smart_opt()); return; } if (safe_str_eq(op->op_type, RSC_NOTIFY)) { return; } crm_debug("Updating history for '%s' with %s op", op->rsc_id, op->op_type); entry = g_hash_table_lookup(lrm_state->resource_history, op->rsc_id); if (entry == NULL && rsc) { entry = calloc(1, sizeof(rsc_history_t)); entry->id = strdup(op->rsc_id); g_hash_table_insert(lrm_state->resource_history, entry->id, entry); entry->rsc.id = entry->id; entry->rsc.type = strdup(rsc->type); entry->rsc.standard = strdup(rsc->standard); if (rsc->provider) { entry->rsc.provider = strdup(rsc->provider); } else { entry->rsc.provider = NULL; } } else if (entry == NULL) { crm_info("Resource %s no longer exists, not updating cache", op->rsc_id); return; } entry->last_callid = op->call_id; target_rc = rsc_op_expected_rc(op); if (op->op_status == PCMK_LRM_OP_CANCELLED) { if (op->interval_ms > 0) { crm_trace("Removing cancelled recurring op: " PCMK__OP_FMT, op->rsc_id, op->op_type, op->interval_ms); history_remove_recurring_op(entry, op); return; } else { crm_trace("Skipping " PCMK__OP_FMT " rc=%d, status=%d", op->rsc_id, op->op_type, op->interval_ms, op->rc, op->op_status); } } else if (did_rsc_op_fail(op, target_rc)) { /* Store failed monitors here, otherwise the block below will cause them * to be forgotten when a stop happens. */ if (entry->failed) { lrmd_free_event(entry->failed); } entry->failed = lrmd_copy_event(op); } else if (op->interval_ms == 0) { if (entry->last) { lrmd_free_event(entry->last); } entry->last = lrmd_copy_event(op); if (op->params && pcmk__str_any_of(op->op_type, CRMD_ACTION_START, "reload", CRMD_ACTION_STATUS, NULL)) { if (entry->stop_params) { g_hash_table_destroy(entry->stop_params); } entry->stop_params = crm_str_table_new(); g_hash_table_foreach(op->params, copy_instance_keys, entry->stop_params); } } if (op->interval_ms > 0) { /* Ensure there are no duplicates */ history_remove_recurring_op(entry, op); crm_trace("Adding recurring op: " PCMK__OP_FMT, op->rsc_id, op->op_type, op->interval_ms); entry->recurring_op_list = g_list_prepend(entry->recurring_op_list, lrmd_copy_event(op)); } else if (entry->recurring_op_list && safe_str_eq(op->op_type, RSC_STATUS) == FALSE) { crm_trace("Dropping %d recurring ops because of: " PCMK__OP_FMT, g_list_length(entry->recurring_op_list), op->rsc_id, op->op_type, op->interval_ms); history_free_recurring_ops(entry); } } /*! * \internal * \brief Send a direct OK ack for a resource task * * \param[in] lrm_state LRM connection * \param[in] input Input message being ack'ed * \param[in] rsc_id ID of affected resource * \param[in] rsc Affected resource (if available) * \param[in] task Operation task being ack'ed * \param[in] ack_host Name of host to send ack to * \param[in] ack_sys IPC system name to ack */ static void send_task_ok_ack(lrm_state_t *lrm_state, ha_msg_input_t *input, const char *rsc_id, lrmd_rsc_info_t *rsc, const char *task, const char *ack_host, const char *ack_sys) { lrmd_event_data_t *op = construct_op(lrm_state, input->xml, rsc_id, task); op->rc = PCMK_OCF_OK; op->op_status = PCMK_LRM_OP_DONE; controld_ack_event_directly(ack_host, ack_sys, rsc, op, rsc_id); lrmd_free_event(op); } static inline const char * op_node_name(lrmd_event_data_t *op) { return op->remote_nodename? op->remote_nodename : fsa_our_uname; } void lrm_op_callback(lrmd_event_data_t * op) { CRM_CHECK(op != NULL, return); switch (op->type) { case lrmd_event_disconnect: if (op->remote_nodename == NULL) { /* If this is the local executor IPC connection, set the right * bits in the controller when the connection goes down. */ lrm_connection_destroy(); } break; case lrmd_event_exec_complete: { lrm_state_t *lrm_state = lrm_state_find(op_node_name(op)); CRM_ASSERT(lrm_state != NULL); process_lrm_event(lrm_state, op, NULL, NULL); } break; default: break; } } /* A_LRM_CONNECT */ void do_lrm_control(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) { /* This only pertains to local executor connections. Remote connections are * handled as resources within the scheduler. Connecting and disconnecting * from remote executor instances is handled differently. */ lrm_state_t *lrm_state = NULL; if(fsa_our_uname == NULL) { return; /* Nothing to do */ } lrm_state = lrm_state_find_or_create(fsa_our_uname); if (lrm_state == NULL) { register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); return; } if (action & A_LRM_DISCONNECT) { if (lrm_state_verify_stopped(lrm_state, cur_state, LOG_INFO) == FALSE) { if (action == A_LRM_DISCONNECT) { crmd_fsa_stall(FALSE); return; } } clear_bit(fsa_input_register, R_LRM_CONNECTED); crm_info("Disconnecting from the executor"); lrm_state_disconnect(lrm_state); lrm_state_reset_tables(lrm_state, FALSE); crm_notice("Disconnected from the executor"); } if (action & A_LRM_CONNECT) { int ret = pcmk_ok; crm_debug("Connecting to the executor"); ret = lrm_state_ipc_connect(lrm_state); if (ret != pcmk_ok) { if (lrm_state->num_lrm_register_fails < MAX_LRM_REG_FAILS) { crm_warn("Failed to connect to the executor %d time%s (%d max)", lrm_state->num_lrm_register_fails, pcmk__plural_s(lrm_state->num_lrm_register_fails), MAX_LRM_REG_FAILS); controld_start_timer(wait_timer); crmd_fsa_stall(FALSE); return; } } if (ret != pcmk_ok) { crm_err("Failed to connect to the executor the max allowed %d time%s", lrm_state->num_lrm_register_fails, pcmk__plural_s(lrm_state->num_lrm_register_fails)); register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); return; } set_bit(fsa_input_register, R_LRM_CONNECTED); crm_info("Connection to the executor established"); } if (action & ~(A_LRM_CONNECT | A_LRM_DISCONNECT)) { crm_err("Unexpected action %s in %s", fsa_action2string(action), __FUNCTION__); } } static gboolean lrm_state_verify_stopped(lrm_state_t * lrm_state, enum crmd_fsa_state cur_state, int log_level) { int counter = 0; gboolean rc = TRUE; const char *when = "lrm disconnect"; GHashTableIter gIter; const char *key = NULL; rsc_history_t *entry = NULL; active_op_t *pending = NULL; crm_debug("Checking for active resources before exit"); if (cur_state == S_TERMINATE) { log_level = LOG_ERR; when = "shutdown"; } else if (is_set(fsa_input_register, R_SHUTDOWN)) { when = "shutdown... waiting"; } if (lrm_state->pending_ops && lrm_state_is_connected(lrm_state) == TRUE) { guint removed = g_hash_table_foreach_remove( lrm_state->pending_ops, stop_recurring_actions, lrm_state); guint nremaining = g_hash_table_size(lrm_state->pending_ops); if (removed || nremaining) { crm_notice("Stopped %u recurring operation%s at %s (%u remaining)", removed, pcmk__plural_s(removed), when, nremaining); } } if (lrm_state->pending_ops) { g_hash_table_iter_init(&gIter, lrm_state->pending_ops); while (g_hash_table_iter_next(&gIter, NULL, (void **)&pending)) { /* Ignore recurring actions in the shutdown calculations */ if (pending->interval_ms == 0) { counter++; } } } if (counter > 0) { do_crm_log(log_level, "%d pending executor operation%s at %s", counter, pcmk__plural_s(counter), when); if (cur_state == S_TERMINATE || !is_set(fsa_input_register, R_SENT_RSC_STOP)) { g_hash_table_iter_init(&gIter, lrm_state->pending_ops); while (g_hash_table_iter_next(&gIter, (gpointer*)&key, (gpointer*)&pending)) { do_crm_log(log_level, "Pending action: %s (%s)", key, pending->op_key); } } else { rc = FALSE; } return rc; } if (lrm_state->resource_history == NULL) { return rc; } if (is_set(fsa_input_register, R_SHUTDOWN)) { /* At this point we're not waiting, we're just shutting down */ when = "shutdown"; } counter = 0; g_hash_table_iter_init(&gIter, lrm_state->resource_history); while (g_hash_table_iter_next(&gIter, NULL, (gpointer*)&entry)) { if (is_rsc_active(lrm_state, entry->id) == FALSE) { continue; } counter++; if (log_level == LOG_ERR) { crm_info("Found %s active at %s", entry->id, when); } else { crm_trace("Found %s active at %s", entry->id, when); } if (lrm_state->pending_ops) { GHashTableIter hIter; g_hash_table_iter_init(&hIter, lrm_state->pending_ops); while (g_hash_table_iter_next(&hIter, (gpointer*)&key, (gpointer*)&pending)) { if (crm_str_eq(entry->id, pending->rsc_id, TRUE)) { crm_notice("%sction %s (%s) incomplete at %s", pending->interval_ms == 0 ? "A" : "Recurring a", key, pending->op_key, when); } } } } if (counter) { crm_err("%d resource%s active at %s", counter, (counter == 1)? " was" : "s were", when); } return rc; } static char * build_parameter_list(const lrmd_event_data_t *op, const struct ra_metadata_s *metadata, xmlNode *result, enum ra_param_flags_e param_type, bool invert_for_xml) { int len = 0; int max = 0; char *list = NULL; GList *iter = NULL; /* Newer resource agents support the "private" parameter attribute to * indicate sensitive parameters. For backward compatibility with older * agents, this list is used if the agent doesn't specify any as "private". */ const char *secure_terms[] = { "password", "passwd", "user", }; if (is_not_set(metadata->ra_flags, ra_uses_private) && (param_type == ra_param_private)) { max = DIMOF(secure_terms); } for (iter = metadata->ra_params; iter != NULL; iter = iter->next) { struct ra_param_s *param = (struct ra_param_s *) iter->data; bool accept = FALSE; if (is_set(param->rap_flags, param_type)) { accept = TRUE; } else if (max) { for (int lpc = 0; lpc < max; lpc++) { if (safe_str_eq(secure_terms[lpc], param->rap_name)) { accept = TRUE; break; } } } if (accept) { int start = len; crm_trace("Attr %s is %s", param->rap_name, ra_param_flag2text(param_type)); len += strlen(param->rap_name) + 2; // include spaces around list = realloc_safe(list, len + 1); // include null terminator // spaces before and after make parsing simpler sprintf(list + start, " %s ", param->rap_name); } else { crm_trace("Rejecting %s for %s", param->rap_name, ra_param_flag2text(param_type)); } if (result && (invert_for_xml? !accept : accept)) { const char *v = g_hash_table_lookup(op->params, param->rap_name); if (v != NULL) { crm_trace("Adding attr %s=%s to the xml result", param->rap_name, v); crm_xml_add(result, param->rap_name, v); } } } return list; } static void append_restart_list(lrmd_event_data_t *op, struct ra_metadata_s *metadata, xmlNode *update, const char *version) { char *list = NULL; char *digest = NULL; xmlNode *restart = NULL; CRM_LOG_ASSERT(op->params != NULL); if (op->interval_ms > 0) { /* monitors are not reloadable */ return; } if (is_set(metadata->ra_flags, ra_supports_reload)) { restart = create_xml_node(NULL, XML_TAG_PARAMS); /* Add any parameters with unique="1" to the "op-force-restart" list. * * (Currently, we abuse "unique=0" to indicate reloadability. This is * nonstandard and should eventually be replaced once the OCF standard * is updated with something better.) */ list = build_parameter_list(op, metadata, restart, ra_param_unique, FALSE); } else { /* Resource does not support reloads */ return; } digest = calculate_operation_digest(restart, version); /* Add "op-force-restart" and "op-restart-digest" to indicate the resource supports reload, * no matter if it actually supports any parameters with unique="1"). */ crm_xml_add(update, XML_LRM_ATTR_OP_RESTART, list? list: ""); crm_xml_add(update, XML_LRM_ATTR_RESTART_DIGEST, digest); crm_trace("%s: %s, %s", op->rsc_id, digest, list); crm_log_xml_trace(restart, "restart digest source"); free_xml(restart); free(digest); free(list); } static void append_secure_list(lrmd_event_data_t *op, struct ra_metadata_s *metadata, xmlNode *update, const char *version) { char *list = NULL; char *digest = NULL; xmlNode *secure = NULL; CRM_LOG_ASSERT(op->params != NULL); /* * To keep XML_LRM_ATTR_OP_SECURE short, we want it to contain the * secure parameters but XML_LRM_ATTR_SECURE_DIGEST to be based on * the insecure ones */ secure = create_xml_node(NULL, XML_TAG_PARAMS); list = build_parameter_list(op, metadata, secure, ra_param_private, TRUE); if (list != NULL) { digest = calculate_operation_digest(secure, version); crm_xml_add(update, XML_LRM_ATTR_OP_SECURE, list); crm_xml_add(update, XML_LRM_ATTR_SECURE_DIGEST, digest); crm_trace("%s: %s, %s", op->rsc_id, digest, list); crm_log_xml_trace(secure, "secure digest source"); } else { crm_trace("%s: no secure parameters", op->rsc_id); } free_xml(secure); free(digest); free(list); } static gboolean build_operation_update(xmlNode * parent, lrmd_rsc_info_t * rsc, lrmd_event_data_t * op, const char *node_name, const char *src) { int target_rc = 0; xmlNode *xml_op = NULL; struct ra_metadata_s *metadata = NULL; const char *caller_version = NULL; lrm_state_t *lrm_state = NULL; if (op == NULL) { return FALSE; } target_rc = rsc_op_expected_rc(op); /* there is a small risk in formerly mixed clusters that it will * be sub-optimal. * * however with our upgrade policy, the update we send should * still be completely supported anyway */ caller_version = g_hash_table_lookup(op->params, XML_ATTR_CRM_VERSION); CRM_LOG_ASSERT(caller_version != NULL); if(caller_version == NULL) { caller_version = CRM_FEATURE_SET; } crm_trace("Building %s operation update with originator version: %s", op->rsc_id, caller_version); xml_op = pcmk__create_history_xml(parent, op, caller_version, target_rc, fsa_our_uname, src, LOG_DEBUG); if (xml_op == NULL) { return TRUE; } if ((rsc == NULL) || (op->params == NULL) || !crm_op_needs_metadata(rsc->standard, op->op_type)) { crm_trace("No digests needed for %s action on %s (params=%p rsc=%p)", op->op_type, op->rsc_id, op->params, rsc); return TRUE; } lrm_state = lrm_state_find(node_name); if (lrm_state == NULL) { crm_warn("Cannot calculate digests for operation " PCMK__OP_FMT " because we have no connection to executor for %s", op->rsc_id, op->op_type, op->interval_ms, node_name); return TRUE; } metadata = metadata_cache_get(lrm_state->metadata_cache, rsc); if (metadata == NULL) { /* For now, we always collect resource agent meta-data via a local, * synchronous, direct execution of the agent. This has multiple issues: * the executor should execute agents, not the controller; meta-data for * Pacemaker Remote nodes should be collected on those nodes, not * locally; and the meta-data call shouldn't eat into the timeout of the * real action being performed. * * These issues are planned to be addressed by having the scheduler * schedule a meta-data cache check at the beginning of each transition. * Once that is working, this block will only be a fallback in case the * initial collection fails. */ char *metadata_str = NULL; int rc = lrm_state_get_metadata(lrm_state, rsc->standard, rsc->provider, rsc->type, &metadata_str, 0); if (rc != pcmk_ok) { crm_warn("Failed to get metadata for %s (%s:%s:%s)", rsc->id, rsc->standard, rsc->provider, rsc->type); return TRUE; } metadata = metadata_cache_update(lrm_state->metadata_cache, rsc, metadata_str); free(metadata_str); if (metadata == NULL) { crm_warn("Failed to update metadata for %s (%s:%s:%s)", rsc->id, rsc->standard, rsc->provider, rsc->type); return TRUE; } } #if ENABLE_VERSIONED_ATTRS crm_xml_add(xml_op, XML_ATTR_RA_VERSION, metadata->ra_version); #endif crm_trace("Including additional digests for %s::%s:%s", rsc->standard, rsc->provider, rsc->type); append_restart_list(op, metadata, xml_op, caller_version); append_secure_list(op, metadata, xml_op, caller_version); return TRUE; } static gboolean is_rsc_active(lrm_state_t * lrm_state, const char *rsc_id) { rsc_history_t *entry = NULL; entry = g_hash_table_lookup(lrm_state->resource_history, rsc_id); if (entry == NULL || entry->last == NULL) { return FALSE; } crm_trace("Processing %s: %s.%d=%d", rsc_id, entry->last->op_type, entry->last->interval_ms, entry->last->rc); if (entry->last->rc == PCMK_OCF_OK && safe_str_eq(entry->last->op_type, CRMD_ACTION_STOP)) { return FALSE; } else if (entry->last->rc == PCMK_OCF_OK && safe_str_eq(entry->last->op_type, CRMD_ACTION_MIGRATE)) { // A stricter check is too complex ... leave that to the scheduler return FALSE; } else if (entry->last->rc == PCMK_OCF_NOT_RUNNING) { return FALSE; } else if ((entry->last->interval_ms == 0) && (entry->last->rc == PCMK_OCF_NOT_CONFIGURED)) { /* Badly configured resources can't be reliably stopped */ return FALSE; } return TRUE; } static gboolean build_active_RAs(lrm_state_t * lrm_state, xmlNode * rsc_list) { GHashTableIter iter; rsc_history_t *entry = NULL; g_hash_table_iter_init(&iter, lrm_state->resource_history); while (g_hash_table_iter_next(&iter, NULL, (void **)&entry)) { GList *gIter = NULL; xmlNode *xml_rsc = create_xml_node(rsc_list, XML_LRM_TAG_RESOURCE); crm_xml_add(xml_rsc, XML_ATTR_ID, entry->id); crm_xml_add(xml_rsc, XML_ATTR_TYPE, entry->rsc.type); crm_xml_add(xml_rsc, XML_AGENT_ATTR_CLASS, entry->rsc.standard); crm_xml_add(xml_rsc, XML_AGENT_ATTR_PROVIDER, entry->rsc.provider); if (entry->last && entry->last->params) { const char *container = g_hash_table_lookup(entry->last->params, CRM_META"_"XML_RSC_ATTR_CONTAINER); if (container) { crm_trace("Resource %s is a part of container resource %s", entry->id, container); crm_xml_add(xml_rsc, XML_RSC_ATTR_CONTAINER, container); } } build_operation_update(xml_rsc, &(entry->rsc), entry->failed, lrm_state->node_name, __FUNCTION__); build_operation_update(xml_rsc, &(entry->rsc), entry->last, lrm_state->node_name, __FUNCTION__); for (gIter = entry->recurring_op_list; gIter != NULL; gIter = gIter->next) { build_operation_update(xml_rsc, &(entry->rsc), gIter->data, lrm_state->node_name, __FUNCTION__); } } return FALSE; } static xmlNode * do_lrm_query_internal(lrm_state_t *lrm_state, int update_flags) { xmlNode *xml_state = NULL; xmlNode *xml_data = NULL; xmlNode *rsc_list = NULL; crm_node_t *peer = NULL; peer = crm_get_peer_full(0, lrm_state->node_name, CRM_GET_PEER_ANY); CRM_CHECK(peer != NULL, return NULL); xml_state = create_node_state_update(peer, update_flags, NULL, __FUNCTION__); if (xml_state == NULL) { return NULL; } xml_data = create_xml_node(xml_state, XML_CIB_TAG_LRM); crm_xml_add(xml_data, XML_ATTR_ID, peer->uuid); rsc_list = create_xml_node(xml_data, XML_LRM_TAG_RESOURCES); /* Build a list of active (not always running) resources */ build_active_RAs(lrm_state, rsc_list); crm_log_xml_trace(xml_state, "Current executor state"); return xml_state; } xmlNode * controld_query_executor_state(const char *node_name) { lrm_state_t *lrm_state = lrm_state_find(node_name); if (!lrm_state) { crm_err("Could not find executor state for node %s", node_name); return NULL; } return do_lrm_query_internal(lrm_state, node_update_cluster|node_update_peer); } /*! * \internal * \brief Map standard Pacemaker return code to operation status and OCF code * * \param[out] event Executor event whose status and return code should be set * \param[in] rc Standard Pacemaker return code */ void controld_rc2event(lrmd_event_data_t *event, int rc) { switch (rc) { case pcmk_rc_ok: event->rc = PCMK_OCF_OK; event->op_status = PCMK_LRM_OP_DONE; break; case EACCES: event->rc = PCMK_OCF_INSUFFICIENT_PRIV; event->op_status = PCMK_LRM_OP_ERROR; break; default: event->rc = PCMK_OCF_UNKNOWN_ERROR; event->op_status = PCMK_LRM_OP_ERROR; break; } } /*! * \internal * \brief Trigger a new transition after CIB status was deleted * * If a CIB status delete was not expected (as part of the transition graph), * trigger a new transition by updating the (arbitrary) "last-lrm-refresh" * cluster property. * * \param[in] from_sys IPC name that requested the delete * \param[in] rsc_id Resource whose status was deleted (for logging only) */ void controld_trigger_delete_refresh(const char *from_sys, const char *rsc_id) { if (safe_str_neq(from_sys, CRM_SYSTEM_TENGINE)) { char *now_s = crm_strdup_printf("%lld", (long long) time(NULL)); crm_debug("Triggering a refresh after %s cleaned %s", from_sys, rsc_id); update_attr_delegate(fsa_cib_conn, cib_none, XML_CIB_TAG_CRMCONFIG, NULL, NULL, NULL, NULL, "last-lrm-refresh", now_s, FALSE, NULL, NULL); free(now_s); } } static void notify_deleted(lrm_state_t * lrm_state, ha_msg_input_t * input, const char *rsc_id, int rc) { lrmd_event_data_t *op = NULL; const char *from_sys = crm_element_value(input->msg, F_CRM_SYS_FROM); const char *from_host = crm_element_value(input->msg, F_CRM_HOST_FROM); crm_info("Notifying %s on %s that %s was%s deleted", from_sys, (from_host? from_host : "localhost"), rsc_id, ((rc == pcmk_ok)? "" : " not")); op = construct_op(lrm_state, input->xml, rsc_id, CRMD_ACTION_DELETE); controld_rc2event(op, pcmk_legacy2rc(rc)); controld_ack_event_directly(from_host, from_sys, NULL, op, rsc_id); lrmd_free_event(op); controld_trigger_delete_refresh(from_sys, rsc_id); } static gboolean lrm_remove_deleted_rsc(gpointer key, gpointer value, gpointer user_data) { struct delete_event_s *event = user_data; struct pending_deletion_op_s *op = value; if (crm_str_eq(event->rsc, op->rsc, TRUE)) { notify_deleted(event->lrm_state, op->input, event->rsc, event->rc); return TRUE; } return FALSE; } static gboolean lrm_remove_deleted_op(gpointer key, gpointer value, gpointer user_data) { const char *rsc = user_data; active_op_t *pending = value; if (crm_str_eq(rsc, pending->rsc_id, TRUE)) { crm_info("Removing op %s:%d for deleted resource %s", pending->op_key, pending->call_id, rsc); return TRUE; } return FALSE; } static void delete_rsc_entry(lrm_state_t * lrm_state, ha_msg_input_t * input, const char *rsc_id, GHashTableIter * rsc_gIter, int rc, const char *user_name) { struct delete_event_s event; CRM_CHECK(rsc_id != NULL, return); if (rc == pcmk_ok) { char *rsc_id_copy = strdup(rsc_id); if (rsc_gIter) { g_hash_table_iter_remove(rsc_gIter); } else { g_hash_table_remove(lrm_state->resource_history, rsc_id_copy); } controld_delete_resource_history(rsc_id_copy, lrm_state->node_name, user_name, crmd_cib_smart_opt()); g_hash_table_foreach_remove(lrm_state->pending_ops, lrm_remove_deleted_op, rsc_id_copy); free(rsc_id_copy); } if (input) { notify_deleted(lrm_state, input, rsc_id, rc); } event.rc = rc; event.rsc = rsc_id; event.lrm_state = lrm_state; g_hash_table_foreach_remove(lrm_state->deletion_ops, lrm_remove_deleted_rsc, &event); } /*! * \internal * \brief Erase an LRM history entry from the CIB, given the operation data * * \param[in] lrm_state LRM state of the desired node * \param[in] op Operation whose history should be deleted */ static void erase_lrm_history_by_op(lrm_state_t *lrm_state, lrmd_event_data_t *op) { xmlNode *xml_top = NULL; CRM_CHECK(op != NULL, return); xml_top = create_xml_node(NULL, XML_LRM_TAG_RSC_OP); crm_xml_add_int(xml_top, XML_LRM_ATTR_CALLID, op->call_id); crm_xml_add(xml_top, XML_ATTR_TRANSITION_KEY, op->user_data); if (op->interval_ms > 0) { char *op_id = pcmk__op_key(op->rsc_id, op->op_type, op->interval_ms); /* Avoid deleting last_failure too (if it was a result of this recurring op failing) */ crm_xml_add(xml_top, XML_ATTR_ID, op_id); free(op_id); } crm_debug("Erasing resource operation history for " PCMK__OP_FMT " (call=%d)", op->rsc_id, op->op_type, op->interval_ms, op->call_id); fsa_cib_conn->cmds->remove(fsa_cib_conn, XML_CIB_TAG_STATUS, xml_top, cib_quorum_override); crm_log_xml_trace(xml_top, "op:cancel"); free_xml(xml_top); } /* Define xpath to find LRM resource history entry by node and resource */ #define XPATH_HISTORY \ "/" XML_TAG_CIB "/" XML_CIB_TAG_STATUS \ "/" XML_CIB_TAG_STATE "[@" XML_ATTR_UNAME "='%s']" \ "/" XML_CIB_TAG_LRM "/" XML_LRM_TAG_RESOURCES \ "/" XML_LRM_TAG_RESOURCE "[@" XML_ATTR_ID "='%s']" \ "/" XML_LRM_TAG_RSC_OP /* ... and also by operation key */ #define XPATH_HISTORY_ID XPATH_HISTORY \ "[@" XML_ATTR_ID "='%s']" /* ... and also by operation key and operation call ID */ #define XPATH_HISTORY_CALL XPATH_HISTORY \ "[@" XML_ATTR_ID "='%s' and @" XML_LRM_ATTR_CALLID "='%d']" /* ... and also by operation key and original operation key */ #define XPATH_HISTORY_ORIG XPATH_HISTORY \ "[@" XML_ATTR_ID "='%s' and @" XML_LRM_ATTR_TASK_KEY "='%s']" /*! * \internal * \brief Erase an LRM history entry from the CIB, given operation identifiers * * \param[in] lrm_state LRM state of the node to clear history for * \param[in] rsc_id Name of resource to clear history for * \param[in] key Operation key of operation to clear history for * \param[in] orig_op If specified, delete only if it has this original op * \param[in] call_id If specified, delete entry only if it has this call ID */ static void erase_lrm_history_by_id(lrm_state_t *lrm_state, const char *rsc_id, const char *key, const char *orig_op, int call_id) { char *op_xpath = NULL; CRM_CHECK((rsc_id != NULL) && (key != NULL), return); if (call_id > 0) { op_xpath = crm_strdup_printf(XPATH_HISTORY_CALL, lrm_state->node_name, rsc_id, key, call_id); } else if (orig_op) { op_xpath = crm_strdup_printf(XPATH_HISTORY_ORIG, lrm_state->node_name, rsc_id, key, orig_op); } else { op_xpath = crm_strdup_printf(XPATH_HISTORY_ID, lrm_state->node_name, rsc_id, key); } crm_debug("Erasing resource operation history for %s on %s (call=%d)", key, rsc_id, call_id); fsa_cib_conn->cmds->remove(fsa_cib_conn, op_xpath, NULL, cib_quorum_override | cib_xpath); free(op_xpath); } static inline gboolean last_failed_matches_op(rsc_history_t *entry, const char *op, guint interval_ms) { if (entry == NULL) { return FALSE; } if (op == NULL) { return TRUE; } return (safe_str_eq(op, entry->failed->op_type) && (interval_ms == entry->failed->interval_ms)); } /*! * \internal * \brief Clear a resource's last failure * * Erase a resource's last failure on a particular node from both the * LRM resource history in the CIB, and the resource history remembered * for the LRM state. * * \param[in] rsc_id Resource name * \param[in] node_name Node name * \param[in] operation If specified, only clear if matching this operation * \param[in] interval_ms If operation is specified, it has this interval */ void lrm_clear_last_failure(const char *rsc_id, const char *node_name, const char *operation, guint interval_ms) { char *op_key = NULL; char *orig_op_key = NULL; lrm_state_t *lrm_state = NULL; lrm_state = lrm_state_find(node_name); if (lrm_state == NULL) { return; } /* Erase from CIB */ op_key = pcmk__op_key(rsc_id, "last_failure", 0); if (operation) { orig_op_key = pcmk__op_key(rsc_id, operation, interval_ms); } erase_lrm_history_by_id(lrm_state, rsc_id, op_key, orig_op_key, 0); free(op_key); free(orig_op_key); /* Remove from memory */ if (lrm_state->resource_history) { rsc_history_t *entry = g_hash_table_lookup(lrm_state->resource_history, rsc_id); if (last_failed_matches_op(entry, operation, interval_ms)) { lrmd_free_event(entry->failed); entry->failed = NULL; } } } /* Returns: gboolean - cancellation is in progress */ static gboolean cancel_op(lrm_state_t * lrm_state, const char *rsc_id, const char *key, int op, gboolean remove) { int rc = pcmk_ok; char *local_key = NULL; active_op_t *pending = NULL; CRM_CHECK(op != 0, return FALSE); CRM_CHECK(rsc_id != NULL, return FALSE); if (key == NULL) { local_key = make_stop_id(rsc_id, op); key = local_key; } pending = g_hash_table_lookup(lrm_state->pending_ops, key); if (pending) { if (remove && is_not_set(pending->flags, active_op_remove)) { set_bit(pending->flags, active_op_remove); crm_debug("Scheduling %s for removal", key); } if (is_set(pending->flags, active_op_cancelled)) { crm_debug("Operation %s already cancelled", key); free(local_key); return FALSE; } set_bit(pending->flags, active_op_cancelled); } else { crm_info("No pending op found for %s", key); free(local_key); return FALSE; } crm_debug("Cancelling op %d for %s (%s)", op, rsc_id, key); rc = lrm_state_cancel(lrm_state, pending->rsc_id, pending->op_type, pending->interval_ms); if (rc == pcmk_ok) { crm_debug("Op %d for %s (%s): cancelled", op, rsc_id, key); free(local_key); return TRUE; } crm_debug("Op %d for %s (%s): Nothing to cancel", op, rsc_id, key); /* The caller needs to make sure the entry is * removed from the pending_ops list * * Usually by returning TRUE inside the worker function * supplied to g_hash_table_foreach_remove() * * Not removing the entry from pending_ops will block * the node from shutting down */ free(local_key); return FALSE; } struct cancel_data { gboolean done; gboolean remove; const char *key; lrmd_rsc_info_t *rsc; lrm_state_t *lrm_state; }; static gboolean cancel_action_by_key(gpointer key, gpointer value, gpointer user_data) { gboolean remove = FALSE; struct cancel_data *data = user_data; active_op_t *op = value; if (crm_str_eq(op->op_key, data->key, TRUE)) { data->done = TRUE; remove = !cancel_op(data->lrm_state, data->rsc->id, key, op->call_id, data->remove); } return remove; } static gboolean cancel_op_key(lrm_state_t * lrm_state, lrmd_rsc_info_t * rsc, const char *key, gboolean remove) { guint removed = 0; struct cancel_data data; CRM_CHECK(rsc != NULL, return FALSE); CRM_CHECK(key != NULL, return FALSE); data.key = key; data.rsc = rsc; data.done = FALSE; data.remove = remove; data.lrm_state = lrm_state; removed = g_hash_table_foreach_remove(lrm_state->pending_ops, cancel_action_by_key, &data); crm_trace("Removed %u op cache entries, new size: %u", removed, g_hash_table_size(lrm_state->pending_ops)); return data.done; } /*! * \internal * \brief Retrieve resource information from LRM * * \param[in] lrm_state LRM connection to use * \param[in] rsc_xml XML containing resource configuration * \param[in] do_create If true, register resource with LRM if not already * \param[out] rsc_info Where to store resource information obtained from LRM * * \retval pcmk_ok Success (and rsc_info holds newly allocated result) * \retval -EINVAL Required information is missing from arguments * \retval -ENOTCONN No active connection to LRM * \retval -ENODEV Resource not found * \retval -errno Error communicating with executor when registering resource * * \note Caller is responsible for freeing result on success. */ static int get_lrm_resource(lrm_state_t *lrm_state, xmlNode *rsc_xml, gboolean do_create, lrmd_rsc_info_t **rsc_info) { const char *id = ID(rsc_xml); CRM_CHECK(lrm_state && rsc_xml && rsc_info, return -EINVAL); CRM_CHECK(id, return -EINVAL); if (lrm_state_is_connected(lrm_state) == FALSE) { return -ENOTCONN; } crm_trace("Retrieving resource information for %s from the executor", id); *rsc_info = lrm_state_get_rsc_info(lrm_state, id, 0); // If resource isn't known by ID, try clone name, if provided if (!*rsc_info) { const char *long_id = crm_element_value(rsc_xml, XML_ATTR_ID_LONG); if (long_id) { *rsc_info = lrm_state_get_rsc_info(lrm_state, long_id, 0); } } if ((*rsc_info == NULL) && do_create) { const char *class = crm_element_value(rsc_xml, XML_AGENT_ATTR_CLASS); const char *provider = crm_element_value(rsc_xml, XML_AGENT_ATTR_PROVIDER); const char *type = crm_element_value(rsc_xml, XML_ATTR_TYPE); int rc; crm_trace("Registering resource %s with the executor", id); rc = lrm_state_register_rsc(lrm_state, id, class, provider, type, lrmd_opt_drop_recurring); if (rc != pcmk_ok) { fsa_data_t *msg_data = NULL; crm_err("Could not register resource %s with the executor on %s: %s " CRM_XS " rc=%d", id, lrm_state->node_name, pcmk_strerror(rc), rc); /* Register this as an internal error if this involves the local * executor. Otherwise, we're likely dealing with an unresponsive * remote node, which is not an FSA failure. */ if (lrm_state_is_local(lrm_state) == TRUE) { register_fsa_error(C_FSA_INTERNAL, I_FAIL, NULL); } return rc; } *rsc_info = lrm_state_get_rsc_info(lrm_state, id, 0); } return *rsc_info? pcmk_ok : -ENODEV; } static void delete_resource(lrm_state_t * lrm_state, const char *id, lrmd_rsc_info_t * rsc, GHashTableIter * gIter, const char *sys, const char *user, ha_msg_input_t * request, gboolean unregister) { int rc = pcmk_ok; crm_info("Removing resource %s from executor for %s%s%s", id, sys, (user? " as " : ""), (user? user : "")); if (rsc && unregister) { rc = lrm_state_unregister_rsc(lrm_state, id, 0); } if (rc == pcmk_ok) { crm_trace("Resource %s deleted from executor", id); } else if (rc == -EINPROGRESS) { crm_info("Deletion of resource '%s' from executor is pending", id); if (request) { struct pending_deletion_op_s *op = NULL; char *ref = crm_element_value_copy(request->msg, XML_ATTR_REFERENCE); op = calloc(1, sizeof(struct pending_deletion_op_s)); op->rsc = strdup(rsc->id); op->input = copy_ha_msg_input(request); g_hash_table_insert(lrm_state->deletion_ops, ref, op); } return; } else { crm_warn("Could not delete '%s' from executor for %s%s%s: %s " CRM_XS " rc=%d", id, sys, (user? " as " : ""), (user? user : ""), pcmk_strerror(rc), rc); } delete_rsc_entry(lrm_state, request, id, gIter, rc, user); } static int get_fake_call_id(lrm_state_t *lrm_state, const char *rsc_id) { int call_id = 999999999; rsc_history_t *entry = NULL; if(lrm_state) { entry = g_hash_table_lookup(lrm_state->resource_history, rsc_id); } /* Make sure the call id is greater than the last successful operation, * otherwise the failure will not result in a possible recovery of the resource * as it could appear the failure occurred before the successful start */ if (entry) { call_id = entry->last_callid + 1; } if (call_id < 0) { call_id = 1; } return call_id; } static void fake_op_status(lrm_state_t *lrm_state, lrmd_event_data_t *op, int op_status, enum ocf_exitcode op_exitcode) { op->call_id = get_fake_call_id(lrm_state, op->rsc_id); op->t_run = time(NULL); op->t_rcchange = op->t_run; op->op_status = op_status; op->rc = op_exitcode; } static void force_reprobe(lrm_state_t *lrm_state, const char *from_sys, const char *from_host, const char *user_name, gboolean is_remote_node) { GHashTableIter gIter; rsc_history_t *entry = NULL; crm_info("Clearing resource history on node %s", lrm_state->node_name); g_hash_table_iter_init(&gIter, lrm_state->resource_history); while (g_hash_table_iter_next(&gIter, NULL, (void **)&entry)) { /* only unregister the resource during a reprobe if it is not a remote connection * resource. otherwise unregistering the connection will terminate remote-node * membership */ gboolean unregister = TRUE; if (is_remote_lrmd_ra(NULL, NULL, entry->id)) { lrm_state_t *remote_lrm_state = lrm_state_find(entry->id); if (remote_lrm_state) { /* when forcing a reprobe, make sure to clear remote node before * clearing the remote node's connection resource */ force_reprobe(remote_lrm_state, from_sys, from_host, user_name, TRUE); } unregister = FALSE; } delete_resource(lrm_state, entry->id, &entry->rsc, &gIter, from_sys, user_name, NULL, unregister); } /* Now delete the copy in the CIB */ controld_delete_node_state(lrm_state->node_name, controld_section_lrm, cib_scope_local); /* Finally, _delete_ the value in pacemaker-attrd -- setting it to FALSE * would result in the scheduler sending us back here again */ update_attrd(lrm_state->node_name, CRM_OP_PROBED, NULL, user_name, is_remote_node); } /*! * \internal * \brief Fail a requested action without actually executing it * * For an action that can't be executed, process it similarly to an actual * execution result, with specified error status (except for notify actions, * which will always be treated as successful). * * \param[in] lrm_state Executor connection that action is for * \param[in] action Action XML from request * \param[in] rc Desired return code to use * \param[in] op_status Desired operation status to use */ static void synthesize_lrmd_failure(lrm_state_t *lrm_state, xmlNode *action, int op_status, enum ocf_exitcode rc) { lrmd_event_data_t *op = NULL; const char *operation = crm_element_value(action, XML_LRM_ATTR_TASK); const char *target_node = crm_element_value(action, XML_LRM_ATTR_TARGET); xmlNode *xml_rsc = find_xml_node(action, XML_CIB_TAG_RESOURCE, TRUE); if ((xml_rsc == NULL) || (ID(xml_rsc) == NULL)) { /* @TODO Should we do something else, like direct ack? */ crm_info("Can't fake %s failure (%d) on %s without resource configuration", crm_element_value(action, XML_LRM_ATTR_TASK_KEY), rc, target_node); return; } else if(operation == NULL) { /* This probably came from crm_resource -C, nothing to do */ crm_info("Can't fake %s failure (%d) on %s without operation", ID(xml_rsc), rc, target_node); return; } op = construct_op(lrm_state, action, ID(xml_rsc), operation); if (safe_str_eq(operation, RSC_NOTIFY)) { // Notifications can't fail fake_op_status(lrm_state, op, PCMK_LRM_OP_DONE, PCMK_OCF_OK); } else { fake_op_status(lrm_state, op, op_status, rc); } crm_info("Faking " PCMK__OP_FMT " result (%d) on %s", op->rsc_id, op->op_type, op->interval_ms, op->rc, target_node); // Process the result as if it came from the LRM process_lrm_event(lrm_state, op, NULL, action); lrmd_free_event(op); } /*! * \internal * \brief Get target of an LRM operation * * \param[in] xml LRM operation data XML * * \return LRM operation target node name (local node or Pacemaker Remote node) */ static const char * lrm_op_target(xmlNode *xml) { const char *target = NULL; if (xml) { target = crm_element_value(xml, XML_LRM_ATTR_TARGET); } if (target == NULL) { target = fsa_our_uname; } return target; } static void fail_lrm_resource(xmlNode *xml, lrm_state_t *lrm_state, const char *user_name, const char *from_host, const char *from_sys) { lrmd_event_data_t *op = NULL; lrmd_rsc_info_t *rsc = NULL; xmlNode *xml_rsc = find_xml_node(xml, XML_CIB_TAG_RESOURCE, TRUE); CRM_CHECK(xml_rsc != NULL, return); /* The executor simply executes operations and reports the results, without * any concept of success or failure, so to fail a resource, we must fake * what a failure looks like. * * To do this, we create a fake executor operation event for the resource, * and pass that event to the executor client callback so it will be * processed as if it came from the executor. */ op = construct_op(lrm_state, xml, ID(xml_rsc), "asyncmon"); fake_op_status(lrm_state, op, PCMK_LRM_OP_DONE, PCMK_OCF_UNKNOWN_ERROR); free((char*) op->user_data); op->user_data = NULL; op->interval_ms = 0; #if ENABLE_ACL if (user_name && !pcmk__is_privileged(user_name)) { crm_err("%s does not have permission to fail %s", user_name, ID(xml_rsc)); controld_ack_event_directly(from_host, from_sys, NULL, op, ID(xml_rsc)); lrmd_free_event(op); return; } #endif if (get_lrm_resource(lrm_state, xml_rsc, TRUE, &rsc) == pcmk_ok) { crm_info("Failing resource %s...", rsc->id); op->exit_reason = strdup("Simulated failure"); process_lrm_event(lrm_state, op, NULL, xml); op->op_status = PCMK_LRM_OP_DONE; op->rc = PCMK_OCF_OK; lrmd_free_rsc_info(rsc); } else { crm_info("Cannot find/create resource in order to fail it..."); crm_log_xml_warn(xml, "bad input"); } controld_ack_event_directly(from_host, from_sys, NULL, op, ID(xml_rsc)); lrmd_free_event(op); } static void handle_refresh_op(lrm_state_t *lrm_state, const char *user_name, const char *from_host, const char *from_sys) { int rc = pcmk_ok; xmlNode *fragment = do_lrm_query_internal(lrm_state, node_update_all); fsa_cib_update(XML_CIB_TAG_STATUS, fragment, cib_quorum_override, rc, user_name); crm_info("Forced a local resource history refresh: call=%d", rc); if (safe_str_neq(CRM_SYSTEM_CRMD, from_sys)) { xmlNode *reply = create_request(CRM_OP_INVOKE_LRM, fragment, from_host, from_sys, CRM_SYSTEM_LRMD, fsa_our_uuid); crm_debug("ACK'ing refresh from %s (%s)", from_sys, from_host); if (relay_message(reply, TRUE) == FALSE) { crm_log_xml_err(reply, "Unable to route reply"); } free_xml(reply); } free_xml(fragment); } static void handle_query_op(xmlNode *msg, lrm_state_t *lrm_state) { xmlNode *data = do_lrm_query_internal(lrm_state, node_update_all); xmlNode *reply = create_reply(msg, data); if (relay_message(reply, TRUE) == FALSE) { crm_err("Unable to route reply"); crm_log_xml_err(reply, "reply"); } free_xml(reply); free_xml(data); } static void handle_reprobe_op(lrm_state_t *lrm_state, const char *from_sys, const char *from_host, const char *user_name, gboolean is_remote_node) { crm_notice("Forcing the status of all resources to be redetected"); force_reprobe(lrm_state, from_sys, from_host, user_name, is_remote_node); - if (pcmk__str_none_of(from_sys, CRM_SYSTEM_PENGINE, CRM_SYSTEM_TENGINE, NULL)) { + if (!pcmk__str_any_of(from_sys, CRM_SYSTEM_PENGINE, CRM_SYSTEM_TENGINE, NULL)) { xmlNode *reply = create_request(CRM_OP_INVOKE_LRM, NULL, from_host, from_sys, CRM_SYSTEM_LRMD, fsa_our_uuid); crm_debug("ACK'ing re-probe from %s (%s)", from_sys, from_host); if (relay_message(reply, TRUE) == FALSE) { crm_log_xml_err(reply, "Unable to route reply"); } free_xml(reply); } } static bool do_lrm_cancel(ha_msg_input_t *input, lrm_state_t *lrm_state, lrmd_rsc_info_t *rsc, const char *from_host, const char *from_sys) { char *op_key = NULL; char *meta_key = NULL; int call = 0; const char *call_id = NULL; const char *op_task = NULL; guint interval_ms = 0; gboolean in_progress = FALSE; xmlNode *params = find_xml_node(input->xml, XML_TAG_ATTRS, TRUE); CRM_CHECK(params != NULL, return FALSE); meta_key = crm_meta_name(XML_LRM_ATTR_TASK); op_task = crm_element_value(params, meta_key); free(meta_key); CRM_CHECK(op_task != NULL, return FALSE); meta_key = crm_meta_name(XML_LRM_ATTR_INTERVAL_MS); if (crm_element_value_ms(params, meta_key, &interval_ms) != pcmk_ok) { free(meta_key); return FALSE; } free(meta_key); op_key = pcmk__op_key(rsc->id, op_task, interval_ms); meta_key = crm_meta_name(XML_LRM_ATTR_CALLID); call_id = crm_element_value(params, meta_key); free(meta_key); crm_debug("Scheduler requested op %s (call=%s) be cancelled", op_key, (call_id? call_id : "NA")); call = crm_parse_int(call_id, "0"); if (call == 0) { // Normal case when the scheduler cancels a recurring op in_progress = cancel_op_key(lrm_state, rsc, op_key, TRUE); } else { // Normal case when the scheduler cancels an orphan op in_progress = cancel_op(lrm_state, rsc->id, NULL, call, TRUE); } // Acknowledge cancellation operation if for a remote connection resource if (!in_progress || is_remote_lrmd_ra(NULL, NULL, rsc->id)) { char *op_id = make_stop_id(rsc->id, call); if (is_remote_lrmd_ra(NULL, NULL, rsc->id) == FALSE) { crm_info("Nothing known about operation %d for %s", call, op_key); } erase_lrm_history_by_id(lrm_state, rsc->id, op_key, NULL, call); send_task_ok_ack(lrm_state, input, rsc->id, rsc, op_task, from_host, from_sys); /* needed at least for cancellation of a remote operation */ g_hash_table_remove(lrm_state->pending_ops, op_id); free(op_id); } else { /* No ack is needed since abcdaa8, but peers with older versions * in a rolling upgrade need one. We didn't bump the feature set * at that commit, so we can only compare against the previous * CRM version (3.0.8). If any peers have feature set 3.0.9 but * not abcdaa8, they will time out waiting for the ack (no * released versions of Pacemaker are affected). */ const char *peer_version = crm_element_value(params, XML_ATTR_CRM_VERSION); if (compare_version(peer_version, "3.0.8") <= 0) { crm_info("Sending compatibility ack for %s cancellation to %s (CRM version %s)", op_key, from_host, peer_version); send_task_ok_ack(lrm_state, input, rsc->id, rsc, op_task, from_host, from_sys); } } free(op_key); return TRUE; } static void do_lrm_delete(ha_msg_input_t *input, lrm_state_t *lrm_state, lrmd_rsc_info_t *rsc, const char *from_sys, const char *from_host, bool crm_rsc_delete, const char *user_name) { gboolean unregister = TRUE; #if ENABLE_ACL int cib_rc = controld_delete_resource_history(rsc->id, lrm_state->node_name, user_name, cib_dryrun|cib_sync_call); if (cib_rc != pcmk_rc_ok) { lrmd_event_data_t *op = NULL; op = construct_op(lrm_state, input->xml, rsc->id, CRMD_ACTION_DELETE); op->op_status = PCMK_LRM_OP_ERROR; if (cib_rc == EACCES) { op->rc = PCMK_OCF_INSUFFICIENT_PRIV; } else { op->rc = PCMK_OCF_UNKNOWN_ERROR; } controld_ack_event_directly(from_host, from_sys, NULL, op, rsc->id); lrmd_free_event(op); return; } #endif if (crm_rsc_delete && is_remote_lrmd_ra(NULL, NULL, rsc->id)) { unregister = FALSE; } delete_resource(lrm_state, rsc->id, rsc, NULL, from_sys, user_name, input, unregister); } /* A_LRM_INVOKE */ void do_lrm_invoke(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) { lrm_state_t *lrm_state = NULL; const char *crm_op = NULL; const char *from_sys = NULL; const char *from_host = NULL; const char *operation = NULL; ha_msg_input_t *input = fsa_typed_data(fsa_dt_ha_msg); const char *user_name = NULL; const char *target_node = NULL; gboolean is_remote_node = FALSE; bool crm_rsc_delete = FALSE; target_node = lrm_op_target(input->xml); is_remote_node = safe_str_neq(target_node, fsa_our_uname); lrm_state = lrm_state_find(target_node); if ((lrm_state == NULL) && is_remote_node) { crm_err("Failing action because local node has never had connection to remote node %s", target_node); synthesize_lrmd_failure(NULL, input->xml, PCMK_LRM_OP_NOT_CONNECTED, PCMK_OCF_UNKNOWN_ERROR); return; } CRM_ASSERT(lrm_state != NULL); #if ENABLE_ACL user_name = pcmk__update_acl_user(input->msg, F_CRM_USER, NULL); #endif crm_op = crm_element_value(input->msg, F_CRM_TASK); from_sys = crm_element_value(input->msg, F_CRM_SYS_FROM); if (safe_str_neq(from_sys, CRM_SYSTEM_TENGINE)) { from_host = crm_element_value(input->msg, F_CRM_HOST_FROM); } #if ENABLE_ACL crm_trace("Executor %s command from %s as user %s", crm_op, from_sys, user_name); #else crm_trace("Executor %s command from %s", crm_op, from_sys); #endif if (safe_str_eq(crm_op, CRM_OP_LRM_DELETE)) { if (safe_str_neq(from_sys, CRM_SYSTEM_TENGINE)) { crm_rsc_delete = TRUE; // from crm_resource } operation = CRMD_ACTION_DELETE; } else if (safe_str_eq(crm_op, CRM_OP_LRM_FAIL)) { fail_lrm_resource(input->xml, lrm_state, user_name, from_host, from_sys); return; } else if (input->xml != NULL) { operation = crm_element_value(input->xml, XML_LRM_ATTR_TASK); } if (safe_str_eq(crm_op, CRM_OP_LRM_REFRESH)) { handle_refresh_op(lrm_state, user_name, from_host, from_sys); } else if (safe_str_eq(crm_op, CRM_OP_LRM_QUERY)) { handle_query_op(input->msg, lrm_state); } else if (safe_str_eq(operation, CRM_OP_PROBED)) { update_attrd(lrm_state->node_name, CRM_OP_PROBED, XML_BOOLEAN_TRUE, user_name, is_remote_node); } else if (pcmk__str_any_of(CRM_OP_REPROBE, operation, crm_op, NULL)) { handle_reprobe_op(lrm_state, from_sys, from_host, user_name, is_remote_node); } else if (operation != NULL) { lrmd_rsc_info_t *rsc = NULL; xmlNode *xml_rsc = find_xml_node(input->xml, XML_CIB_TAG_RESOURCE, TRUE); gboolean create_rsc = safe_str_neq(operation, CRMD_ACTION_DELETE); int rc; // We can't return anything meaningful without a resource ID CRM_CHECK(xml_rsc && ID(xml_rsc), return); rc = get_lrm_resource(lrm_state, xml_rsc, create_rsc, &rsc); if (rc == -ENOTCONN) { synthesize_lrmd_failure(lrm_state, input->xml, PCMK_LRM_OP_NOT_CONNECTED, PCMK_OCF_UNKNOWN_ERROR); return; } else if ((rc < 0) && !create_rsc) { /* Delete of malformed or nonexistent resource * (deleting something that does not exist is a success) */ crm_notice("Not registering resource '%s' for a %s event " CRM_XS " get-rc=%d (%s) transition-key=%s", ID(xml_rsc), operation, rc, pcmk_strerror(rc), ID(input->xml)); delete_rsc_entry(lrm_state, input, ID(xml_rsc), NULL, pcmk_ok, user_name); return; } else if (rc == -EINVAL) { // Resource operation on malformed resource crm_err("Invalid resource definition for %s", ID(xml_rsc)); crm_log_xml_warn(input->msg, "invalid resource"); synthesize_lrmd_failure(lrm_state, input->xml, PCMK_LRM_OP_ERROR, PCMK_OCF_NOT_CONFIGURED); // fatal error return; } else if (rc < 0) { // Error communicating with the executor crm_err("Could not register resource '%s' with executor: %s " CRM_XS " rc=%d", ID(xml_rsc), pcmk_strerror(rc), rc); crm_log_xml_warn(input->msg, "failed registration"); synthesize_lrmd_failure(lrm_state, input->xml, PCMK_LRM_OP_ERROR, PCMK_OCF_INVALID_PARAM); // hard error return; } if (safe_str_eq(operation, CRMD_ACTION_CANCEL)) { if (!do_lrm_cancel(input, lrm_state, rsc, from_host, from_sys)) { crm_log_xml_warn(input->xml, "Bad command"); } } else if (safe_str_eq(operation, CRMD_ACTION_DELETE)) { do_lrm_delete(input, lrm_state, rsc, from_sys, from_host, crm_rsc_delete, user_name); } else { do_lrm_rsc_op(lrm_state, rsc, operation, input->xml); } lrmd_free_rsc_info(rsc); } else { crm_err("Cannot perform operation %s of unknown type", crm_str(crm_op)); register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); } } #if ENABLE_VERSIONED_ATTRS static void resolve_versioned_parameters(lrm_state_t *lrm_state, const char *rsc_id, const xmlNode *rsc_op, GHashTable *params) { /* Resource info *should* already be cached, so we don't get * executor call */ lrmd_rsc_info_t *rsc = lrm_state_get_rsc_info(lrm_state, rsc_id, 0); struct ra_metadata_s *metadata; metadata = metadata_cache_get(lrm_state->metadata_cache, rsc); if (metadata) { xmlNode *versioned_attrs = NULL; GHashTable *hash = NULL; char *key = NULL; char *value = NULL; GHashTableIter iter; versioned_attrs = first_named_child(rsc_op, XML_TAG_OP_VER_ATTRS); hash = pe_unpack_versioned_parameters(versioned_attrs, metadata->ra_version); g_hash_table_iter_init(&iter, hash); while (g_hash_table_iter_next(&iter, (gpointer *) &key, (gpointer *) &value)) { g_hash_table_iter_steal(&iter); g_hash_table_replace(params, key, value); } g_hash_table_destroy(hash); versioned_attrs = first_named_child(rsc_op, XML_TAG_OP_VER_META); hash = pe_unpack_versioned_parameters(versioned_attrs, metadata->ra_version); g_hash_table_iter_init(&iter, hash); while (g_hash_table_iter_next(&iter, (gpointer *) &key, (gpointer *) &value)) { g_hash_table_replace(params, crm_meta_name(key), strdup(value)); if (safe_str_eq(key, XML_ATTR_TIMEOUT)) { op->timeout = crm_parse_int(value, "0"); } else if (safe_str_eq(key, XML_OP_ATTR_START_DELAY)) { op->start_delay = crm_parse_int(value, "0"); } } g_hash_table_destroy(hash); versioned_attrs = first_named_child(rsc_op, XML_TAG_RSC_VER_ATTRS); hash = pe_unpack_versioned_parameters(versioned_attrs, metadata->ra_version); g_hash_table_iter_init(&iter, hash); while (g_hash_table_iter_next(&iter, (gpointer *) &key, (gpointer *) &value)) { g_hash_table_iter_steal(&iter); g_hash_table_replace(params, key, value); } g_hash_table_destroy(hash); } lrmd_free_rsc_info(rsc); } #endif static lrmd_event_data_t * construct_op(lrm_state_t *lrm_state, xmlNode *rsc_op, const char *rsc_id, const char *operation) { lrmd_event_data_t *op = NULL; const char *op_delay = NULL; const char *op_timeout = NULL; GHashTable *params = NULL; const char *transition = NULL; CRM_ASSERT(rsc_id && operation); op = lrmd_new_event(rsc_id, operation, 0); op->type = lrmd_event_exec_complete; op->op_status = PCMK_LRM_OP_PENDING; op->rc = -1; op->timeout = 0; op->start_delay = 0; if (rsc_op == NULL) { CRM_LOG_ASSERT(safe_str_eq(CRMD_ACTION_STOP, operation)); op->user_data = NULL; /* the stop_all_resources() case * by definition there is no DC (or they'd be shutting * us down). * So we should put our version here. */ op->params = crm_str_table_new(); g_hash_table_insert(op->params, strdup(XML_ATTR_CRM_VERSION), strdup(CRM_FEATURE_SET)); crm_trace("Constructed %s op for %s", operation, rsc_id); return op; } params = xml2list(rsc_op); g_hash_table_remove(params, CRM_META "_op_target_rc"); op_delay = crm_meta_value(params, XML_OP_ATTR_START_DELAY); op->start_delay = crm_parse_int(op_delay, "0"); op_timeout = crm_meta_value(params, XML_ATTR_TIMEOUT); op->timeout = crm_parse_int(op_timeout, "0"); if (pcmk__guint_from_hash(params, CRM_META "_" XML_LRM_ATTR_INTERVAL_MS, 0, &(op->interval_ms)) != pcmk_rc_ok) { op->interval_ms = 0; } #if ENABLE_VERSIONED_ATTRS if (lrm_state && !is_remote_lrmd_ra(NULL, NULL, rsc_id) - && pcmk__str_none_of(op_type, CRMD_ACTION_METADATA, CRMD_ACTION_DELETE, + && !pcmk__str_any_of(op_type, CRMD_ACTION_METADATA, CRMD_ACTION_DELETE, NULL)) { resolve_versioned_parameters(lrm_state, rsc_id, rsc_op, params); } #endif if (safe_str_neq(operation, RSC_STOP)) { op->params = params; } else { rsc_history_t *entry = NULL; if (lrm_state) { entry = g_hash_table_lookup(lrm_state->resource_history, rsc_id); } /* If we do not have stop parameters cached, use * whatever we are given */ if (!entry || !entry->stop_params) { op->params = params; } else { /* Copy the cached parameter list so that we stop the resource * with the old attributes, not the new ones */ op->params = crm_str_table_new(); g_hash_table_foreach(params, copy_meta_keys, op->params); g_hash_table_foreach(entry->stop_params, copy_instance_keys, op->params); g_hash_table_destroy(params); params = NULL; } } /* sanity */ if (op->timeout <= 0) { op->timeout = op->interval_ms; } if (op->start_delay < 0) { op->start_delay = 0; } transition = crm_element_value(rsc_op, XML_ATTR_TRANSITION_KEY); CRM_CHECK(transition != NULL, return op); op->user_data = strdup(transition); if (op->interval_ms != 0) { if (pcmk__str_any_of(operation, CRMD_ACTION_START, CRMD_ACTION_STOP, NULL)) { crm_err("Start and Stop actions cannot have an interval: %u", op->interval_ms); op->interval_ms = 0; } } crm_trace("Constructed %s op for %s: interval=%u", operation, rsc_id, op->interval_ms); return op; } /*! * \internal * \brief Send a (synthesized) event result * * Reply with a synthesized event result directly, as opposed to going through * the executor. * * \param[in] to_host Host to send result to * \param[in] to_sys IPC name to send result to (NULL for transition engine) * \param[in] rsc Type information about resource the result is for * \param[in] op Event with result to send * \param[in] rsc_id ID of resource the result is for */ void controld_ack_event_directly(const char *to_host, const char *to_sys, lrmd_rsc_info_t *rsc, lrmd_event_data_t *op, const char *rsc_id) { xmlNode *reply = NULL; xmlNode *update, *iter; crm_node_t *peer = NULL; CRM_CHECK(op != NULL, return); if (op->rsc_id == NULL) { CRM_ASSERT(rsc_id != NULL); op->rsc_id = strdup(rsc_id); } if (to_sys == NULL) { to_sys = CRM_SYSTEM_TENGINE; } peer = crm_get_peer(0, fsa_our_uname); update = create_node_state_update(peer, node_update_none, NULL, __FUNCTION__); iter = create_xml_node(update, XML_CIB_TAG_LRM); crm_xml_add(iter, XML_ATTR_ID, fsa_our_uuid); iter = create_xml_node(iter, XML_LRM_TAG_RESOURCES); iter = create_xml_node(iter, XML_LRM_TAG_RESOURCE); crm_xml_add(iter, XML_ATTR_ID, op->rsc_id); build_operation_update(iter, rsc, op, fsa_our_uname, __FUNCTION__); reply = create_request(CRM_OP_INVOKE_LRM, update, to_host, to_sys, CRM_SYSTEM_LRMD, NULL); crm_log_xml_trace(update, "[direct ACK]"); crm_debug("ACK'ing resource op " PCMK__OP_FMT " from %s: %s", op->rsc_id, op->op_type, op->interval_ms, op->user_data, crm_element_value(reply, XML_ATTR_REFERENCE)); if (relay_message(reply, TRUE) == FALSE) { crm_log_xml_err(reply, "Unable to route reply"); } free_xml(update); free_xml(reply); } gboolean verify_stopped(enum crmd_fsa_state cur_state, int log_level) { gboolean res = TRUE; GList *lrm_state_list = lrm_state_get_list(); GList *state_entry; for (state_entry = lrm_state_list; state_entry != NULL; state_entry = state_entry->next) { lrm_state_t *lrm_state = state_entry->data; if (!lrm_state_verify_stopped(lrm_state, cur_state, log_level)) { /* keep iterating through all even when false is returned */ res = FALSE; } } set_bit(fsa_input_register, R_SENT_RSC_STOP); g_list_free(lrm_state_list); lrm_state_list = NULL; return res; } struct stop_recurring_action_s { lrmd_rsc_info_t *rsc; lrm_state_t *lrm_state; }; static gboolean stop_recurring_action_by_rsc(gpointer key, gpointer value, gpointer user_data) { gboolean remove = FALSE; struct stop_recurring_action_s *event = user_data; active_op_t *op = value; if ((op->interval_ms != 0) && crm_str_eq(op->rsc_id, event->rsc->id, TRUE)) { crm_debug("Cancelling op %d for %s (%s)", op->call_id, op->rsc_id, (char*)key); remove = !cancel_op(event->lrm_state, event->rsc->id, key, op->call_id, FALSE); } return remove; } static gboolean stop_recurring_actions(gpointer key, gpointer value, gpointer user_data) { gboolean remove = FALSE; lrm_state_t *lrm_state = user_data; active_op_t *op = value; if (op->interval_ms != 0) { crm_info("Cancelling op %d for %s (%s)", op->call_id, op->rsc_id, (const char *) key); remove = !cancel_op(lrm_state, op->rsc_id, key, op->call_id, FALSE); } return remove; } static void record_pending_op(const char *node_name, lrmd_rsc_info_t *rsc, lrmd_event_data_t *op) { const char *record_pending = NULL; CRM_CHECK(node_name != NULL, return); CRM_CHECK(rsc != NULL, return); CRM_CHECK(op != NULL, return); // Never record certain operation types as pending if ((op->op_type == NULL) || (op->params == NULL) || !controld_action_is_recordable(op->op_type)) { return; } // defaults to true record_pending = crm_meta_value(op->params, XML_OP_ATTR_PENDING); if (record_pending && !crm_is_true(record_pending)) { return; } op->call_id = -1; op->op_status = PCMK_LRM_OP_PENDING; op->rc = PCMK_OCF_UNKNOWN; op->t_run = time(NULL); op->t_rcchange = op->t_run; /* write a "pending" entry to the CIB, inhibit notification */ crm_debug("Recording pending op " PCMK__OP_FMT " on %s in the CIB", op->rsc_id, op->op_type, op->interval_ms, node_name); do_update_resource(node_name, rsc, op, 0); } static void do_lrm_rsc_op(lrm_state_t *lrm_state, lrmd_rsc_info_t *rsc, const char *operation, xmlNode *msg) { int call_id = 0; char *op_id = NULL; lrmd_event_data_t *op = NULL; lrmd_key_value_t *params = NULL; fsa_data_t *msg_data = NULL; const char *transition = NULL; gboolean stop_recurring = FALSE; bool send_nack = FALSE; CRM_CHECK(rsc != NULL, return); CRM_CHECK(operation != NULL, return); if (msg != NULL) { transition = crm_element_value(msg, XML_ATTR_TRANSITION_KEY); if (transition == NULL) { crm_log_xml_err(msg, "Missing transition number"); } } op = construct_op(lrm_state, msg, rsc->id, operation); CRM_CHECK(op != NULL, return); if (is_remote_lrmd_ra(NULL, NULL, rsc->id) && (op->interval_ms == 0) && strcmp(operation, CRMD_ACTION_MIGRATE) == 0) { /* pcmk remote connections are a special use case. * We never ever want to stop monitoring a connection resource until * the entire migration has completed. If the connection is unexpectedly * severed, even during a migration, this is an event we must detect.*/ stop_recurring = FALSE; } else if ((op->interval_ms == 0) && strcmp(operation, CRMD_ACTION_STATUS) != 0 && strcmp(operation, CRMD_ACTION_NOTIFY) != 0) { /* stop any previous monitor operations before changing the resource state */ stop_recurring = TRUE; } if (stop_recurring == TRUE) { guint removed = 0; struct stop_recurring_action_s data; data.rsc = rsc; data.lrm_state = lrm_state; removed = g_hash_table_foreach_remove( lrm_state->pending_ops, stop_recurring_action_by_rsc, &data); if (removed) { crm_debug("Stopped %u recurring operation%s in preparation for " PCMK__OP_FMT, removed, pcmk__plural_s(removed), rsc->id, operation, op->interval_ms); } } /* now do the op */ crm_notice("Requesting local execution of %s operation for %s on %s " CRM_XS " transition_key=%s op_key=" PCMK__OP_FMT, crm_action_str(op->op_type, op->interval_ms), rsc->id, lrm_state->node_name, transition, rsc->id, operation, op->interval_ms); if (is_set(fsa_input_register, R_SHUTDOWN) && safe_str_eq(operation, RSC_START)) { register_fsa_input(C_SHUTDOWN, I_SHUTDOWN, NULL); send_nack = TRUE; } else if (fsa_state != S_NOT_DC && fsa_state != S_POLICY_ENGINE /* Recalculating */ && fsa_state != S_TRANSITION_ENGINE && safe_str_neq(operation, CRMD_ACTION_STOP)) { send_nack = TRUE; } if(send_nack) { crm_notice("Discarding attempt to perform action %s on %s in state %s (shutdown=%s)", operation, rsc->id, fsa_state2string(fsa_state), is_set(fsa_input_register, R_SHUTDOWN)?"true":"false"); op->rc = PCMK_OCF_UNKNOWN_ERROR; op->op_status = PCMK_LRM_OP_INVALID; controld_ack_event_directly(NULL, NULL, rsc, op, rsc->id); lrmd_free_event(op); free(op_id); return; } record_pending_op(lrm_state->node_name, rsc, op); op_id = pcmk__op_key(rsc->id, op->op_type, op->interval_ms); if (op->interval_ms > 0) { /* cancel it so we can then restart it without conflict */ cancel_op_key(lrm_state, rsc, op_id, FALSE); } if (op->params) { char *key = NULL; char *value = NULL; GHashTableIter iter; g_hash_table_iter_init(&iter, op->params); while (g_hash_table_iter_next(&iter, (gpointer *) & key, (gpointer *) & value)) { params = lrmd_key_value_add(params, key, value); } } call_id = lrm_state_exec(lrm_state, rsc->id, op->op_type, op->user_data, op->interval_ms, op->timeout, op->start_delay, params); if (call_id <= 0 && lrm_state_is_local(lrm_state)) { crm_err("Operation %s on %s failed: %d", operation, rsc->id, call_id); register_fsa_error(C_FSA_INTERNAL, I_FAIL, NULL); } else if (call_id <= 0) { crm_err("Operation %s on resource %s failed to execute on remote node %s: %d", operation, rsc->id, lrm_state->node_name, call_id); fake_op_status(lrm_state, op, PCMK_LRM_OP_DONE, PCMK_OCF_UNKNOWN_ERROR); process_lrm_event(lrm_state, op, NULL, NULL); } else { /* record all operations so we can wait * for them to complete during shutdown */ char *call_id_s = make_stop_id(rsc->id, call_id); active_op_t *pending = NULL; pending = calloc(1, sizeof(active_op_t)); crm_trace("Recording pending op: %d - %s %s", call_id, op_id, call_id_s); pending->call_id = call_id; pending->interval_ms = op->interval_ms; pending->op_type = strdup(operation); pending->op_key = strdup(op_id); pending->rsc_id = strdup(rsc->id); pending->start_time = time(NULL); pending->user_data = op->user_data? strdup(op->user_data) : NULL; if (crm_element_value_epoch(msg, XML_CONFIG_ATTR_SHUTDOWN_LOCK, &(pending->lock_time)) != pcmk_ok) { pending->lock_time = 0; } g_hash_table_replace(lrm_state->pending_ops, call_id_s, pending); if ((op->interval_ms > 0) && (op->start_delay > START_DELAY_THRESHOLD)) { int target_rc = 0; crm_info("Faking confirmation of %s: execution postponed for over 5 minutes", op_id); decode_transition_key(op->user_data, NULL, NULL, NULL, &target_rc); op->rc = target_rc; op->op_status = PCMK_LRM_OP_DONE; controld_ack_event_directly(NULL, NULL, rsc, op, rsc->id); } pending->params = op->params; op->params = NULL; } free(op_id); lrmd_free_event(op); return; } int last_resource_update = 0; static void cib_rsc_callback(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data) { switch (rc) { case pcmk_ok: case -pcmk_err_diff_failed: case -pcmk_err_diff_resync: crm_trace("Resource update %d complete: rc=%d", call_id, rc); break; default: crm_warn("Resource update %d failed: (rc=%d) %s", call_id, rc, pcmk_strerror(rc)); } if (call_id == last_resource_update) { last_resource_update = 0; trigger_fsa(fsa_source); } } /* Only successful stops, and probes that found the resource inactive, get locks * recorded in the history. This ensures the resource stays locked to the node * until it is active there again after the node comes back up. */ static bool should_preserve_lock(lrmd_event_data_t *op) { if (!controld_shutdown_lock_enabled) { return false; } if (!strcmp(op->op_type, RSC_STOP) && (op->rc == PCMK_OCF_OK)) { return true; } if (!strcmp(op->op_type, RSC_STATUS) && (op->rc == PCMK_OCF_NOT_RUNNING)) { return true; } return false; } static int do_update_resource(const char *node_name, lrmd_rsc_info_t *rsc, lrmd_event_data_t *op, time_t lock_time) { /* */ int rc = pcmk_ok; xmlNode *update, *iter = NULL; int call_opt = crmd_cib_smart_opt(); const char *uuid = NULL; CRM_CHECK(op != NULL, return 0); iter = create_xml_node(iter, XML_CIB_TAG_STATUS); update = iter; iter = create_xml_node(iter, XML_CIB_TAG_STATE); if (safe_str_eq(node_name, fsa_our_uname)) { uuid = fsa_our_uuid; } else { /* remote nodes uuid and uname are equal */ uuid = node_name; crm_xml_add(iter, XML_NODE_IS_REMOTE, "true"); } CRM_LOG_ASSERT(uuid != NULL); if(uuid == NULL) { rc = -EINVAL; goto done; } crm_xml_add(iter, XML_ATTR_UUID, uuid); crm_xml_add(iter, XML_ATTR_UNAME, node_name); crm_xml_add(iter, XML_ATTR_ORIGIN, __FUNCTION__); iter = create_xml_node(iter, XML_CIB_TAG_LRM); crm_xml_add(iter, XML_ATTR_ID, uuid); iter = create_xml_node(iter, XML_LRM_TAG_RESOURCES); iter = create_xml_node(iter, XML_LRM_TAG_RESOURCE); crm_xml_add(iter, XML_ATTR_ID, op->rsc_id); build_operation_update(iter, rsc, op, node_name, __FUNCTION__); if (rsc) { const char *container = NULL; crm_xml_add(iter, XML_ATTR_TYPE, rsc->type); crm_xml_add(iter, XML_AGENT_ATTR_CLASS, rsc->standard); crm_xml_add(iter, XML_AGENT_ATTR_PROVIDER, rsc->provider); if (lock_time != 0) { /* Actions on a locked resource should either preserve the lock by * recording it with the action result, or clear it. */ if (!should_preserve_lock(op)) { lock_time = 0; } crm_xml_add_ll(iter, XML_CONFIG_ATTR_SHUTDOWN_LOCK, (long long) lock_time); } if (op->params) { container = g_hash_table_lookup(op->params, CRM_META"_"XML_RSC_ATTR_CONTAINER); } if (container) { crm_trace("Resource %s is a part of container resource %s", op->rsc_id, container); crm_xml_add(iter, XML_RSC_ATTR_CONTAINER, container); } } else { crm_warn("Resource %s no longer exists in the executor", op->rsc_id); controld_ack_event_directly(NULL, NULL, rsc, op, op->rsc_id); goto cleanup; } crm_log_xml_trace(update, __FUNCTION__); /* make it an asynchronous call and be done with it * * Best case: * the resource state will be discovered during * the next signup or election. * * Bad case: * we are shutting down and there is no DC at the time, * but then why were we shutting down then anyway? * (probably because of an internal error) * * Worst case: * we get shot for having resources "running" that really weren't * * the alternative however means blocking here for too long, which * isn't acceptable */ fsa_cib_update(XML_CIB_TAG_STATUS, update, call_opt, rc, NULL); if (rc > 0) { last_resource_update = rc; } done: /* the return code is a call number, not an error code */ crm_trace("Sent resource state update message: %d for %s=%u on %s", rc, op->op_type, op->interval_ms, op->rsc_id); fsa_register_cib_callback(rc, FALSE, NULL, cib_rsc_callback); cleanup: free_xml(update); return rc; } void do_lrm_event(long long action, enum crmd_fsa_cause cause, enum crmd_fsa_state cur_state, enum crmd_fsa_input cur_input, fsa_data_t * msg_data) { CRM_CHECK(FALSE, return); } static char * unescape_newlines(const char *string) { char *pch = NULL; char *ret = NULL; static const char *escaped_newline = "\\n"; if (!string) { return NULL; } ret = strdup(string); pch = strstr(ret, escaped_newline); while (pch != NULL) { /* Replace newline escape pattern with actual newline (and a space so we * don't have to shuffle the rest of the buffer) */ pch[0] = '\n'; pch[1] = ' '; pch = strstr(pch, escaped_newline); } return ret; } static bool did_lrm_rsc_op_fail(lrm_state_t *lrm_state, const char * rsc_id, const char * op_type, guint interval_ms) { rsc_history_t *entry = NULL; CRM_CHECK(lrm_state != NULL, return FALSE); CRM_CHECK(rsc_id != NULL, return FALSE); CRM_CHECK(op_type != NULL, return FALSE); entry = g_hash_table_lookup(lrm_state->resource_history, rsc_id); if (entry == NULL || entry->failed == NULL) { return FALSE; } if (crm_str_eq(entry->failed->rsc_id, rsc_id, TRUE) && safe_str_eq(entry->failed->op_type, op_type) && entry->failed->interval_ms == interval_ms) { return TRUE; } return FALSE; } void process_lrm_event(lrm_state_t *lrm_state, lrmd_event_data_t *op, active_op_t *pending, xmlNode *action_xml) { char *op_id = NULL; char *op_key = NULL; int update_id = 0; gboolean remove = FALSE; gboolean removed = FALSE; bool need_direct_ack = FALSE; lrmd_rsc_info_t *rsc = NULL; const char *node_name = NULL; CRM_CHECK(op != NULL, return); CRM_CHECK(op->rsc_id != NULL, return); // Remap new status codes for older DCs if (compare_version(fsa_our_dc_version, "3.2.0") < 0) { switch (op->op_status) { case PCMK_LRM_OP_NOT_CONNECTED: op->op_status = PCMK_LRM_OP_ERROR; op->rc = PCMK_OCF_CONNECTION_DIED; break; case PCMK_LRM_OP_INVALID: op->op_status = PCMK_LRM_OP_ERROR; op->rc = CRM_DIRECT_NACK_RC; break; default: break; } } op_id = make_stop_id(op->rsc_id, op->call_id); op_key = pcmk__op_key(op->rsc_id, op->op_type, op->interval_ms); // Get resource info if available (from executor state or action XML) if (lrm_state) { rsc = lrm_state_get_rsc_info(lrm_state, op->rsc_id, 0); } if ((rsc == NULL) && action_xml) { xmlNode *xml = find_xml_node(action_xml, XML_CIB_TAG_RESOURCE, TRUE); const char *standard = crm_element_value(xml, XML_AGENT_ATTR_CLASS); const char *provider = crm_element_value(xml, XML_AGENT_ATTR_PROVIDER); const char *type = crm_element_value(xml, XML_ATTR_TYPE); if (standard && type) { crm_info("%s agent information not cached, using %s%s%s:%s from action XML", op->rsc_id, standard, (provider? ":" : ""), (provider? provider : ""), type); rsc = lrmd_new_rsc_info(op->rsc_id, standard, provider, type); } else { crm_err("Can't process %s result because %s agent information not cached or in XML", op_key, op->rsc_id); } } // Get node name if available (from executor state or action XML) if (lrm_state) { node_name = lrm_state->node_name; } else if (action_xml) { node_name = crm_element_value(action_xml, XML_LRM_ATTR_TARGET); } if(pending == NULL) { remove = TRUE; if (lrm_state) { pending = g_hash_table_lookup(lrm_state->pending_ops, op_id); } } if (op->op_status == PCMK_LRM_OP_ERROR) { switch(op->rc) { case PCMK_OCF_NOT_RUNNING: case PCMK_OCF_RUNNING_MASTER: case PCMK_OCF_DEGRADED: case PCMK_OCF_DEGRADED_MASTER: // Leave it to the TE/scheduler to decide if this is an error op->op_status = PCMK_LRM_OP_DONE; break; default: /* Nothing to do */ break; } } if (op->op_status != PCMK_LRM_OP_CANCELLED) { /* We might not record the result, so directly acknowledge it to the * originator instead, so it doesn't time out waiting for the result * (especially important if part of a transition). */ need_direct_ack = TRUE; if (controld_action_is_recordable(op->op_type)) { if (node_name && rsc) { // We should record the result, and happily, we can update_id = do_update_resource(node_name, rsc, op, pending? pending->lock_time : 0); need_direct_ack = FALSE; } else if (op->rsc_deleted) { /* We shouldn't record the result (likely the resource was * refreshed, cleaned, or removed while this operation was * in flight). */ crm_notice("Not recording %s result in CIB because " "resource information was removed since it was initiated", op_key); } else { /* This shouldn't be possible; the executor didn't consider the * resource deleted, but we couldn't find resource or node * information. */ crm_err("Unable to record %s result in CIB: %s", op_key, (node_name? "No resource information" : "No node name")); } } } else if (op->interval_ms == 0) { /* A non-recurring operation was cancelled. Most likely, the * never-initiated action was removed from the executor's pending * operations list upon resource removal. */ need_direct_ack = TRUE; } else if (pending == NULL) { /* This recurring operation was cancelled, but was not pending. No * transition actions are waiting on it, nothing needs to be done. */ } else if (op->user_data == NULL) { /* This recurring operation was cancelled and pending, but we don't * have a transition key. This should never happen. */ crm_err("Recurring operation %s was cancelled without transition information", op_key); } else if (is_set(pending->flags, active_op_remove)) { /* This recurring operation was cancelled (by us) and pending, and we * have been waiting for it to finish. */ if (lrm_state) { erase_lrm_history_by_op(lrm_state, op); } /* If the recurring operation had failed, the lrm_rsc_op is recorded as * "last_failure" which won't get erased from the cib given the logic on * purpose in erase_lrm_history_by_op(). So that the cancel action won't * have a chance to get confirmed by DC with process_op_deletion(). * Cluster transition would get stuck waiting for the remaining action * timer to time out. * * Directly acknowledge the cancel operation in this case. */ if (did_lrm_rsc_op_fail(lrm_state, pending->rsc_id, pending->op_type, pending->interval_ms)) { need_direct_ack = TRUE; } } else if (op->rsc_deleted) { /* This recurring operation was cancelled (but not by us, and the * executor does not have resource information, likely due to resource * cleanup, refresh, or removal) and pending. */ crm_debug("Recurring op %s was cancelled due to resource deletion", op_key); need_direct_ack = TRUE; } else { /* This recurring operation was cancelled (but not by us, likely by the * executor before stopping the resource) and pending. We don't need to * do anything special. */ } if (need_direct_ack) { controld_ack_event_directly(NULL, NULL, NULL, op, op->rsc_id); } if(remove == FALSE) { /* The caller will do this afterwards, but keep the logging consistent */ removed = TRUE; } else if (lrm_state && ((op->interval_ms == 0) || (op->op_status == PCMK_LRM_OP_CANCELLED))) { gboolean found = g_hash_table_remove(lrm_state->pending_ops, op_id); if (op->interval_ms != 0) { removed = TRUE; } else if (found) { removed = TRUE; crm_trace("Op %s (call=%d, stop-id=%s, remaining=%u): Confirmed", op_key, op->call_id, op_id, g_hash_table_size(lrm_state->pending_ops)); } } if (node_name == NULL) { node_name = "unknown node"; // for logging } switch (op->op_status) { case PCMK_LRM_OP_CANCELLED: crm_info("Result of %s operation for %s on %s: %s " CRM_XS " call=%d key=%s confirmed=%s", crm_action_str(op->op_type, op->interval_ms), op->rsc_id, node_name, services_lrm_status_str(op->op_status), op->call_id, op_key, (removed? "true" : "false")); break; case PCMK_LRM_OP_DONE: crm_notice("Result of %s operation for %s on %s: %s " CRM_XS " rc=%d call=%d key=%s confirmed=%s cib-update=%d", crm_action_str(op->op_type, op->interval_ms), op->rsc_id, node_name, services_ocf_exitcode_str(op->rc), op->rc, op->call_id, op_key, (removed? "true" : "false"), update_id); break; case PCMK_LRM_OP_TIMEOUT: crm_err("Result of %s operation for %s on %s: %s " CRM_XS " call=%d key=%s timeout=%dms", crm_action_str(op->op_type, op->interval_ms), op->rsc_id, node_name, services_lrm_status_str(op->op_status), op->call_id, op_key, op->timeout); break; default: crm_err("Result of %s operation for %s on %s: %s " CRM_XS " call=%d key=%s confirmed=%s status=%d cib-update=%d", crm_action_str(op->op_type, op->interval_ms), op->rsc_id, node_name, services_lrm_status_str(op->op_status), op->call_id, op_key, (removed? "true" : "false"), op->op_status, update_id); } if (op->output) { char *prefix = crm_strdup_printf("%s-" PCMK__OP_FMT ":%d", node_name, op->rsc_id, op->op_type, op->interval_ms, op->call_id); if (op->rc) { crm_log_output(LOG_NOTICE, prefix, op->output); } else { crm_log_output(LOG_DEBUG, prefix, op->output); } free(prefix); } if (lrm_state) { if (safe_str_neq(op->op_type, RSC_METADATA)) { crmd_alert_resource_op(lrm_state->node_name, op); } else if (rsc && (op->rc == PCMK_OCF_OK)) { char *metadata = unescape_newlines(op->output); metadata_cache_update(lrm_state->metadata_cache, rsc, metadata); free(metadata); } } if (op->rsc_deleted) { crm_info("Deletion of resource '%s' complete after %s", op->rsc_id, op_key); if (lrm_state) { delete_rsc_entry(lrm_state, NULL, op->rsc_id, NULL, pcmk_ok, NULL); } } /* If a shutdown was escalated while operations were pending, * then the FSA will be stalled right now... allow it to continue */ mainloop_set_trigger(fsa_source); if (lrm_state && rsc) { update_history_cache(lrm_state, rsc, op); } lrmd_free_rsc_info(rsc); free(op_key); free(op_id); } diff --git a/daemons/controld/controld_messages.c b/daemons/controld/controld_messages.c index 95f9176deb..7b4901f61f 100644 --- a/daemons/controld/controld_messages.c +++ b/daemons/controld/controld_messages.c @@ -1,1291 +1,1291 @@ /* * Copyright 2004-2020 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #include #include #include GListPtr fsa_message_queue = NULL; extern void crm_shutdown(int nsig); static enum crmd_fsa_input handle_message(xmlNode *msg, enum crmd_fsa_cause cause); static void handle_response(xmlNode *stored_msg); static enum crmd_fsa_input handle_request(xmlNode *stored_msg, enum crmd_fsa_cause cause); static enum crmd_fsa_input handle_shutdown_request(xmlNode *stored_msg); static void send_msg_via_ipc(xmlNode * msg, const char *sys); /* debug only, can wrap all it likes */ int last_data_id = 0; void register_fsa_error_adv(enum crmd_fsa_cause cause, enum crmd_fsa_input input, fsa_data_t * cur_data, void *new_data, const char *raised_from) { /* save the current actions if any */ if (fsa_actions != A_NOTHING) { register_fsa_input_adv(cur_data ? cur_data->fsa_cause : C_FSA_INTERNAL, I_NULL, cur_data ? cur_data->data : NULL, fsa_actions, TRUE, __FUNCTION__); } /* reset the action list */ crm_info("Resetting the current action list"); fsa_dump_actions(fsa_actions, "Drop"); fsa_actions = A_NOTHING; /* register the error */ register_fsa_input_adv(cause, input, new_data, A_NOTHING, TRUE, raised_from); } int register_fsa_input_adv(enum crmd_fsa_cause cause, enum crmd_fsa_input input, void *data, long long with_actions, gboolean prepend, const char *raised_from) { unsigned old_len = g_list_length(fsa_message_queue); fsa_data_t *fsa_data = NULL; if (raised_from == NULL) { raised_from = ""; } if (input == I_NULL && with_actions == A_NOTHING /* && data == NULL */ ) { /* no point doing anything */ crm_err("Cannot add entry to queue: no input and no action"); return 0; } if (input == I_WAIT_FOR_EVENT) { do_fsa_stall = TRUE; crm_debug("Stalling the FSA pending further input: source=%s cause=%s data=%p queue=%d", raised_from, fsa_cause2string(cause), data, old_len); if (old_len > 0) { fsa_dump_queue(LOG_TRACE); prepend = FALSE; } if (data == NULL) { fsa_actions |= with_actions; fsa_dump_actions(with_actions, "Restored"); return 0; } /* Store everything in the new event and reset fsa_actions */ with_actions |= fsa_actions; fsa_actions = A_NOTHING; } last_data_id++; crm_trace("%s %s FSA input %d (%s) due to %s, %s data", raised_from, (prepend? "prepended" : "appended"), last_data_id, fsa_input2string(input), fsa_cause2string(cause), (data? "with" : "without")); fsa_data = calloc(1, sizeof(fsa_data_t)); fsa_data->id = last_data_id; fsa_data->fsa_input = input; fsa_data->fsa_cause = cause; fsa_data->origin = raised_from; fsa_data->data = NULL; fsa_data->data_type = fsa_dt_none; fsa_data->actions = with_actions; if (with_actions != A_NOTHING) { crm_trace("Adding actions %.16llx to input", with_actions); } if (data != NULL) { switch (cause) { case C_FSA_INTERNAL: case C_CRMD_STATUS_CALLBACK: case C_IPC_MESSAGE: case C_HA_MESSAGE: CRM_CHECK(((ha_msg_input_t *) data)->msg != NULL, crm_err("Bogus data from %s", raised_from)); crm_trace("Copying %s data from %s as cluster message data", fsa_cause2string(cause), raised_from); fsa_data->data = copy_ha_msg_input(data); fsa_data->data_type = fsa_dt_ha_msg; break; case C_LRM_OP_CALLBACK: crm_trace("Copying %s data from %s as lrmd_event_data_t", fsa_cause2string(cause), raised_from); fsa_data->data = lrmd_copy_event((lrmd_event_data_t *) data); fsa_data->data_type = fsa_dt_lrm; break; case C_TIMER_POPPED: case C_SHUTDOWN: case C_UNKNOWN: case C_STARTUP: crm_crit("Copying %s data (from %s) is not yet implemented", fsa_cause2string(cause), raised_from); crmd_exit(CRM_EX_SOFTWARE); break; } } /* make sure to free it properly later */ if (prepend) { fsa_message_queue = g_list_prepend(fsa_message_queue, fsa_data); } else { fsa_message_queue = g_list_append(fsa_message_queue, fsa_data); } crm_trace("FSA message queue length is %d", g_list_length(fsa_message_queue)); /* fsa_dump_queue(LOG_TRACE); */ if (old_len == g_list_length(fsa_message_queue)) { crm_err("Couldn't add message to the queue"); } if (fsa_source && input != I_WAIT_FOR_EVENT) { crm_trace("Triggering FSA"); mainloop_set_trigger(fsa_source); } return last_data_id; } void fsa_dump_queue(int log_level) { int offset = 0; GListPtr lpc = NULL; for (lpc = fsa_message_queue; lpc != NULL; lpc = lpc->next) { fsa_data_t *data = (fsa_data_t *) lpc->data; do_crm_log_unlikely(log_level, "queue[%d.%d]: input %s raised by %s(%p.%d)\t(cause=%s)", offset++, data->id, fsa_input2string(data->fsa_input), data->origin, data->data, data->data_type, fsa_cause2string(data->fsa_cause)); } } ha_msg_input_t * copy_ha_msg_input(ha_msg_input_t * orig) { ha_msg_input_t *copy = calloc(1, sizeof(ha_msg_input_t)); CRM_ASSERT(copy != NULL); copy->msg = (orig && orig->msg)? copy_xml(orig->msg) : NULL; copy->xml = get_message_xml(copy->msg, F_CRM_DATA); return copy; } void delete_fsa_input(fsa_data_t * fsa_data) { lrmd_event_data_t *op = NULL; xmlNode *foo = NULL; if (fsa_data == NULL) { return; } crm_trace("About to free %s data", fsa_cause2string(fsa_data->fsa_cause)); if (fsa_data->data != NULL) { switch (fsa_data->data_type) { case fsa_dt_ha_msg: delete_ha_msg_input(fsa_data->data); break; case fsa_dt_xml: foo = fsa_data->data; free_xml(foo); break; case fsa_dt_lrm: op = (lrmd_event_data_t *) fsa_data->data; lrmd_free_event(op); break; case fsa_dt_none: if (fsa_data->data != NULL) { crm_err("Don't know how to free %s data from %s", fsa_cause2string(fsa_data->fsa_cause), fsa_data->origin); crmd_exit(CRM_EX_SOFTWARE); } break; } crm_trace("%s data freed", fsa_cause2string(fsa_data->fsa_cause)); } free(fsa_data); } /* returns the next message */ fsa_data_t * get_message(void) { fsa_data_t *message = g_list_nth_data(fsa_message_queue, 0); fsa_message_queue = g_list_remove(fsa_message_queue, message); crm_trace("Processing input %d", message->id); return message; } void * fsa_typed_data_adv(fsa_data_t * fsa_data, enum fsa_data_type a_type, const char *caller) { void *ret_val = NULL; if (fsa_data == NULL) { crm_err("%s: No FSA data available", caller); } else if (fsa_data->data == NULL) { crm_err("%s: No message data available. Origin: %s", caller, fsa_data->origin); } else if (fsa_data->data_type != a_type) { crm_crit("%s: Message data was the wrong type! %d vs. requested=%d. Origin: %s", caller, fsa_data->data_type, a_type, fsa_data->origin); CRM_ASSERT(fsa_data->data_type == a_type); } else { ret_val = fsa_data->data; } return ret_val; } /* A_MSG_ROUTE */ void do_msg_route(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) { ha_msg_input_t *input = fsa_typed_data(fsa_dt_ha_msg); route_message(msg_data->fsa_cause, input->msg); } void route_message(enum crmd_fsa_cause cause, xmlNode * input) { ha_msg_input_t fsa_input; enum crmd_fsa_input result = I_NULL; fsa_input.msg = input; CRM_CHECK(cause == C_IPC_MESSAGE || cause == C_HA_MESSAGE, return); /* try passing the buck first */ if (relay_message(input, cause == C_IPC_MESSAGE)) { return; } /* handle locally */ result = handle_message(input, cause); /* done or process later? */ switch (result) { case I_NULL: case I_CIB_OP: case I_ROUTER: case I_NODE_JOIN: case I_JOIN_REQUEST: case I_JOIN_RESULT: break; default: /* Defering local processing of message */ register_fsa_input_later(cause, result, &fsa_input); return; } if (result != I_NULL) { /* add to the front of the queue */ register_fsa_input(cause, result, &fsa_input); } } gboolean relay_message(xmlNode * msg, gboolean originated_locally) { int dest = 1; int is_for_dc = 0; int is_for_dcib = 0; int is_for_te = 0; int is_for_crm = 0; int is_for_cib = 0; int is_local = 0; const char *host_to = crm_element_value(msg, F_CRM_HOST_TO); const char *sys_to = crm_element_value(msg, F_CRM_SYS_TO); const char *sys_from = crm_element_value(msg, F_CRM_SYS_FROM); const char *type = crm_element_value(msg, F_TYPE); const char *task = crm_element_value(msg, F_CRM_TASK); const char *ref = crm_element_value(msg, XML_ATTR_REFERENCE); if (ref == NULL) { ref = "without reference ID"; } if (msg == NULL) { crm_warn("Cannot route empty message"); return TRUE; } else if (safe_str_eq(task, CRM_OP_HELLO)) { /* quietly ignore */ crm_trace("No routing needed for hello message %s", ref); return TRUE; } else if (safe_str_neq(type, T_CRM)) { crm_warn("Cannot route message %s: Type is '%s' not '" T_CRM "'", ref, (type? type : "missing")); crm_log_xml_warn(msg, "[bad message type]"); return TRUE; } else if (sys_to == NULL) { crm_warn("Cannot route message %s: No subsystem specified", ref); crm_log_xml_warn(msg, "[no subsystem]"); return TRUE; } is_for_dc = (strcasecmp(CRM_SYSTEM_DC, sys_to) == 0); is_for_dcib = (strcasecmp(CRM_SYSTEM_DCIB, sys_to) == 0); is_for_te = (strcasecmp(CRM_SYSTEM_TENGINE, sys_to) == 0); is_for_cib = (strcasecmp(CRM_SYSTEM_CIB, sys_to) == 0); is_for_crm = (strcasecmp(CRM_SYSTEM_CRMD, sys_to) == 0); is_local = 0; if (pcmk__str_empty(host_to)) { if (is_for_dc || is_for_te) { is_local = 0; } else if (is_for_crm) { if (pcmk__str_any_of(task, CRM_OP_NODE_INFO, PCMK__CONTROLD_CMD_NODES, NULL)) { /* Node info requests do not specify a host, which is normally * treated as "all hosts", because the whole point is that the * client may not know the local node name. Always handle these * requests locally. */ is_local = 1; } else { is_local = !originated_locally; } } else { is_local = 1; } } else if (safe_str_eq(fsa_our_uname, host_to)) { is_local = 1; } else if (is_for_crm && safe_str_eq(task, CRM_OP_LRM_DELETE)) { xmlNode *msg_data = get_message_xml(msg, F_CRM_DATA); const char *mode = crm_element_value(msg_data, PCMK__XA_MODE); if (safe_str_eq(mode, XML_TAG_CIB)) { // Local delete of an offline node's resource history is_local = 1; } } if (is_for_dc || is_for_dcib || is_for_te) { if (AM_I_DC && is_for_te) { crm_trace("Route message %s locally as transition request", ref); send_msg_via_ipc(msg, sys_to); } else if (AM_I_DC) { crm_trace("Route message %s locally as DC request", ref); return FALSE; // More to be done by caller - } else if (originated_locally && pcmk__str_none_of(sys_from, CRM_SYSTEM_PENGINE, - CRM_SYSTEM_TENGINE, NULL)) { + } else if (originated_locally && !pcmk__str_any_of(sys_from, CRM_SYSTEM_PENGINE, + CRM_SYSTEM_TENGINE, NULL)) { #if SUPPORT_COROSYNC if (is_corosync_cluster()) { dest = text2msg_type(sys_to); } #endif crm_trace("Relay message %s to DC", ref); send_cluster_message(host_to ? crm_get_peer(0, host_to) : NULL, dest, msg, TRUE); } else { /* Neither the TE nor the scheduler should be sending messages * to DCs on other nodes. By definition, if we are no longer the DC, * then the scheduler's or TE's data should be discarded. */ crm_trace("Discard message %s because we are not DC", ref); } } else if (is_local && (is_for_crm || is_for_cib)) { crm_trace("Route message %s locally as controller request", ref); return FALSE; // More to be done by caller } else if (is_local) { crm_trace("Relay message %s locally to %s", ref, (sys_to? sys_to : "unknown client")); crm_log_xml_trace(msg, "[IPC relay]"); send_msg_via_ipc(msg, sys_to); } else { crm_node_t *node_to = NULL; #if SUPPORT_COROSYNC if (is_corosync_cluster()) { dest = text2msg_type(sys_to); if (dest == crm_msg_none || dest > crm_msg_stonith_ng) { dest = crm_msg_crmd; } } #endif if (host_to) { node_to = crm_find_peer(0, host_to); if (node_to == NULL) { crm_warn("Cannot route message %s: Unknown node %s", ref, host_to); return TRUE; } crm_trace("Relay message %s to %s", ref, (node_to->uname? node_to->uname : "peer")); } else { crm_trace("Broadcast message %s to all peers", ref); } send_cluster_message(host_to ? node_to : NULL, dest, msg, TRUE); } return TRUE; // No further processing of message is needed } // Return true if field contains a positive integer static bool authorize_version(xmlNode *message_data, const char *field, const char *client_name, const char *ref, const char *uuid) { const char *version = crm_element_value(message_data, field); if (pcmk__str_empty(version)) { crm_warn("IPC hello from %s rejected: No protocol %s", CRM_XS " ref=%s uuid=%s", client_name, field, (ref? ref : "none"), uuid); return false; } else { int version_num = crm_parse_int(version, NULL); if (version_num < 0) { crm_warn("IPC hello from %s rejected: Protocol %s '%s' " "not recognized", CRM_XS " ref=%s uuid=%s", client_name, field, version, (ref? ref : "none"), uuid); return false; } } return true; } /*! * \internal * \brief Check whether a client IPC message is acceptable * * If a given client IPC message is a hello, "authorize" it by ensuring it has * valid information such as a protocol version, and return false indicating * that nothing further needs to be done with the message. If the message is not * a hello, just return true to indicate it needs further processing. * * \param[in] client_msg XML of IPC message * \param[in] curr_client If IPC is not proxied, client that sent message * \param[in] proxy_session If IPC is proxied, the session ID * * \return true if message needs further processing, false if it doesn't */ bool controld_authorize_ipc_message(xmlNode *client_msg, pcmk__client_t *curr_client, const char *proxy_session) { xmlNode *message_data = NULL; const char *client_name = NULL; const char *op = crm_element_value(client_msg, F_CRM_TASK); const char *ref = crm_element_value(client_msg, XML_ATTR_REFERENCE); const char *uuid = (curr_client? curr_client->id : proxy_session); if (uuid == NULL) { crm_warn("IPC message from client rejected: No client identifier " CRM_XS " ref=%s", (ref? ref : "none")); goto rejected; } if (safe_str_neq(CRM_OP_HELLO, op)) { // Only hello messages need to be authorized return true; } message_data = get_message_xml(client_msg, F_CRM_DATA); client_name = crm_element_value(message_data, "client_name"); if (pcmk__str_empty(client_name)) { crm_warn("IPC hello from client rejected: No client name", CRM_XS " ref=%s uuid=%s", (ref? ref : "none"), uuid); goto rejected; } if (!authorize_version(message_data, "major_version", client_name, ref, uuid)) { goto rejected; } if (!authorize_version(message_data, "minor_version", client_name, ref, uuid)) { goto rejected; } crm_trace("Validated IPC hello from client %s", client_name); if (curr_client) { curr_client->userdata = strdup(client_name); } mainloop_set_trigger(fsa_source); return false; rejected: if (curr_client) { qb_ipcs_disconnect(curr_client->ipcs); } return false; } static enum crmd_fsa_input handle_message(xmlNode *msg, enum crmd_fsa_cause cause) { const char *type = NULL; CRM_CHECK(msg != NULL, return I_NULL); type = crm_element_value(msg, F_CRM_MSG_TYPE); if (crm_str_eq(type, XML_ATTR_REQUEST, TRUE)) { return handle_request(msg, cause); } else if (crm_str_eq(type, XML_ATTR_RESPONSE, TRUE)) { handle_response(msg); return I_NULL; } crm_err("Unknown message type: %s", type); return I_NULL; } static enum crmd_fsa_input handle_failcount_op(xmlNode * stored_msg) { const char *rsc = NULL; const char *uname = NULL; const char *op = NULL; char *interval_spec = NULL; guint interval_ms = 0; gboolean is_remote_node = FALSE; xmlNode *xml_op = get_message_xml(stored_msg, F_CRM_DATA); if (xml_op) { xmlNode *xml_rsc = first_named_child(xml_op, XML_CIB_TAG_RESOURCE); xmlNode *xml_attrs = first_named_child(xml_op, XML_TAG_ATTRS); if (xml_rsc) { rsc = ID(xml_rsc); } if (xml_attrs) { op = crm_element_value(xml_attrs, CRM_META "_" XML_RSC_ATTR_CLEAR_OP); crm_element_value_ms(xml_attrs, CRM_META "_" XML_RSC_ATTR_CLEAR_INTERVAL, &interval_ms); } } uname = crm_element_value(xml_op, XML_LRM_ATTR_TARGET); if ((rsc == NULL) || (uname == NULL)) { crm_log_xml_warn(stored_msg, "invalid failcount op"); return I_NULL; } if (crm_element_value(xml_op, XML_LRM_ATTR_ROUTER_NODE)) { is_remote_node = TRUE; } if (interval_ms) { interval_spec = crm_strdup_printf("%ums", interval_ms); } update_attrd_clear_failures(uname, rsc, op, interval_spec, is_remote_node); free(interval_spec); lrm_clear_last_failure(rsc, uname, op, interval_ms); return I_NULL; } static enum crmd_fsa_input handle_lrm_delete(xmlNode *stored_msg) { const char *mode = NULL; xmlNode *msg_data = get_message_xml(stored_msg, F_CRM_DATA); CRM_CHECK(msg_data != NULL, return I_NULL); /* CRM_OP_LRM_DELETE has two distinct modes. The default behavior is to * relay the operation to the affected node, which will unregister the * resource from the local executor, clear the resource's history from the * CIB, and do some bookkeeping in the controller. * * However, if the affected node is offline, the client will specify * mode="cib" which means the controller receiving the operation should * clear the resource's history from the CIB and nothing else. This is used * to clear shutdown locks. */ mode = crm_element_value(msg_data, PCMK__XA_MODE); if ((mode == NULL) || strcmp(mode, XML_TAG_CIB)) { // Relay to affected node crm_xml_add(stored_msg, F_CRM_SYS_TO, CRM_SYSTEM_LRMD); return I_ROUTER; } else { // Delete CIB history locally (compare with do_lrm_delete()) const char *from_sys = NULL; const char *user_name = NULL; const char *rsc_id = NULL; const char *node = NULL; xmlNode *rsc_xml = NULL; int rc = pcmk_rc_ok; rsc_xml = first_named_child(msg_data, XML_CIB_TAG_RESOURCE); CRM_CHECK(rsc_xml != NULL, return I_NULL); rsc_id = ID(rsc_xml); from_sys = crm_element_value(stored_msg, F_CRM_SYS_FROM); node = crm_element_value(msg_data, XML_LRM_ATTR_TARGET); #if ENABLE_ACL user_name = pcmk__update_acl_user(stored_msg, F_CRM_USER, NULL); #endif crm_debug("Handling " CRM_OP_LRM_DELETE " for %s on %s locally%s%s " "(clearing CIB resource history only)", rsc_id, node, (user_name? " for user " : ""), (user_name? user_name : "")); #if ENABLE_ACL rc = controld_delete_resource_history(rsc_id, node, user_name, cib_dryrun|cib_sync_call); #endif if (rc == pcmk_rc_ok) { rc = controld_delete_resource_history(rsc_id, node, user_name, crmd_cib_smart_opt()); } //Notify client and tengine.(Only notify tengine if mode = "cib" and CRM_OP_LRM_DELETE.) if (from_sys) { lrmd_event_data_t *op = NULL; const char *from_host = crm_element_value(stored_msg, F_CRM_HOST_FROM); const char *transition; if (strcmp(from_sys, CRM_SYSTEM_TENGINE)) { transition = crm_element_value(msg_data, XML_ATTR_TRANSITION_KEY); } else { transition = crm_element_value(stored_msg, XML_ATTR_TRANSITION_KEY); } crm_info("Notifying %s on %s that %s was%s deleted", from_sys, (from_host? from_host : "local node"), rsc_id, ((rc == pcmk_rc_ok)? "" : " not")); op = lrmd_new_event(rsc_id, CRMD_ACTION_DELETE, 0); op->type = lrmd_event_exec_complete; op->user_data = strdup(transition? transition : FAKE_TE_ID); op->params = crm_str_table_new(); g_hash_table_insert(op->params, strdup(XML_ATTR_CRM_VERSION), strdup(CRM_FEATURE_SET)); controld_rc2event(op, rc); controld_ack_event_directly(from_host, from_sys, NULL, op, rsc_id); lrmd_free_event(op); controld_trigger_delete_refresh(from_sys, rsc_id); } return I_NULL; } } /*! * \brief Handle a CRM_OP_REMOTE_STATE message by updating remote peer cache * * \param[in] msg Message XML * * \return Next FSA input */ static enum crmd_fsa_input handle_remote_state(xmlNode *msg) { const char *remote_uname = ID(msg); const char *remote_is_up = crm_element_value(msg, XML_NODE_IN_CLUSTER); crm_node_t *remote_peer; CRM_CHECK(remote_uname && remote_is_up, return I_NULL); remote_peer = crm_remote_peer_get(remote_uname); CRM_CHECK(remote_peer, return I_NULL); crm_update_peer_state(__FUNCTION__, remote_peer, crm_is_true(remote_is_up)? CRM_NODE_MEMBER : CRM_NODE_LOST, 0); return I_NULL; } /*! * \brief Handle a CRM_OP_PING message * * \param[in] msg Message XML * * \return Next FSA input */ static enum crmd_fsa_input handle_ping(xmlNode *msg) { const char *value = NULL; xmlNode *ping = NULL; // Build reply ping = create_xml_node(NULL, XML_CRM_TAG_PING); value = crm_element_value(msg, F_CRM_SYS_TO); crm_xml_add(ping, XML_PING_ATTR_SYSFROM, value); // Add controller state value = fsa_state2string(fsa_state); crm_xml_add(ping, XML_PING_ATTR_CRMDSTATE, value); crm_notice("Current ping state: %s", value); // CTS needs this // Add controller health // @TODO maybe do some checks to determine meaningful status crm_xml_add(ping, XML_PING_ATTR_STATUS, "ok"); // Send reply msg = create_reply(msg, ping); free_xml(ping); if (msg) { (void) relay_message(msg, TRUE); free_xml(msg); } // Nothing further to do return I_NULL; } /*! * \brief Handle a PCMK__CONTROLD_CMD_NODES message * * \return Next FSA input */ static enum crmd_fsa_input handle_node_list(xmlNode *request) { GHashTableIter iter; crm_node_t *node = NULL; xmlNode *reply = NULL; xmlNode *reply_data = NULL; // Create message data for reply reply_data = create_xml_node(NULL, XML_CIB_TAG_NODES); g_hash_table_iter_init(&iter, crm_peer_cache); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) & node)) { xmlNode *xml = create_xml_node(reply_data, XML_CIB_TAG_NODE); crm_xml_add_ll(xml, XML_ATTR_ID, (long long) node->id); // uint32_t crm_xml_add(xml, XML_ATTR_UNAME, node->uname); crm_xml_add(xml, XML_NODE_IN_CLUSTER, node->state); } // Create and send reply reply = create_reply(request, reply_data); free_xml(reply_data); if (reply) { (void) relay_message(reply, TRUE); free_xml(reply); } // Nothing further to do return I_NULL; } /*! * \brief Handle a CRM_OP_NODE_INFO request * * \param[in] msg Message XML * * \return Next FSA input */ static enum crmd_fsa_input handle_node_info_request(xmlNode *msg) { const char *value = NULL; crm_node_t *node = NULL; int node_id = 0; xmlNode *reply = NULL; // Build reply reply = create_xml_node(NULL, XML_CIB_TAG_NODE); crm_xml_add(reply, XML_PING_ATTR_SYSFROM, CRM_SYSTEM_CRMD); // Add whether current partition has quorum crm_xml_add_boolean(reply, XML_ATTR_HAVE_QUORUM, fsa_has_quorum); // Check whether client requested node info by ID and/or name crm_element_value_int(msg, XML_ATTR_ID, &node_id); if (node_id < 0) { node_id = 0; } value = crm_element_value(msg, XML_ATTR_UNAME); // Default to local node if none given if ((node_id == 0) && (value == NULL)) { value = fsa_our_uname; } node = crm_find_peer_full(node_id, value, CRM_GET_PEER_ANY); if (node) { crm_xml_add_int(reply, XML_ATTR_ID, node->id); crm_xml_add(reply, XML_ATTR_UUID, node->uuid); crm_xml_add(reply, XML_ATTR_UNAME, node->uname); crm_xml_add(reply, XML_NODE_IS_PEER, node->state); crm_xml_add_boolean(reply, XML_NODE_IS_REMOTE, node->flags & crm_remote_node); } // Send reply msg = create_reply(msg, reply); free_xml(reply); if (msg) { (void) relay_message(msg, TRUE); free_xml(msg); } // Nothing further to do return I_NULL; } static void verify_feature_set(xmlNode *msg) { const char *dc_version = crm_element_value(msg, XML_ATTR_CRM_VERSION); if (dc_version == NULL) { /* All we really know is that the DC feature set is older than 3.1.0, * but that's also all that really matters. */ dc_version = "3.0.14"; } if (feature_set_compatible(dc_version, CRM_FEATURE_SET)) { crm_trace("Local feature set (%s) is compatible with DC's (%s)", CRM_FEATURE_SET, dc_version); } else { crm_err("Local feature set (%s) is incompatible with DC's (%s)", CRM_FEATURE_SET, dc_version); // Nothing is likely to improve without administrator involvement set_bit(fsa_input_register, R_STAYDOWN); crmd_exit(CRM_EX_FATAL); } } // DC gets own shutdown all-clear static enum crmd_fsa_input handle_shutdown_self_ack(xmlNode *stored_msg) { const char *host_from = crm_element_value(stored_msg, F_CRM_HOST_FROM); if (is_set(fsa_input_register, R_SHUTDOWN)) { // The expected case -- we initiated own shutdown sequence crm_info("Shutting down controller"); return I_STOP; } if (safe_str_eq(host_from, fsa_our_dc)) { // Must be logic error -- DC confirming its own unrequested shutdown crm_err("Shutting down controller immediately due to " "unexpected shutdown confirmation"); return I_TERMINATE; } if (fsa_state != S_STOPPING) { // Shouldn't happen -- non-DC confirming unrequested shutdown crm_err("Starting new DC election because %s is " "confirming shutdown we did not request", (host_from? host_from : "another node")); return I_ELECTION; } // Shouldn't happen, but we are already stopping anyway crm_debug("Ignoring unexpected shutdown confirmation from %s", (host_from? host_from : "another node")); return I_NULL; } // Non-DC gets shutdown all-clear from DC static enum crmd_fsa_input handle_shutdown_ack(xmlNode *stored_msg) { const char *host_from = crm_element_value(stored_msg, F_CRM_HOST_FROM); if (host_from == NULL) { crm_warn("Ignoring shutdown request without origin specified"); return I_NULL; } if ((fsa_our_dc == NULL) || (strcmp(host_from, fsa_our_dc) == 0)) { if (is_set(fsa_input_register, R_SHUTDOWN)) { crm_info("Shutting down controller after confirmation from %s", host_from); } else { crm_err("Shutting down controller after unexpected " "shutdown request from %s", host_from); set_bit(fsa_input_register, R_STAYDOWN); } return I_STOP; } crm_warn("Ignoring shutdown request from %s because DC is %s", host_from, fsa_our_dc); return I_NULL; } static enum crmd_fsa_input handle_request(xmlNode *stored_msg, enum crmd_fsa_cause cause) { xmlNode *msg = NULL; const char *op = crm_element_value(stored_msg, F_CRM_TASK); /* Optimize this for the DC - it has the most to do */ if (op == NULL) { crm_log_xml_warn(stored_msg, "[request without " F_CRM_TASK "]"); return I_NULL; } if (strcmp(op, CRM_OP_SHUTDOWN_REQ) == 0) { const char *from = crm_element_value(stored_msg, F_CRM_HOST_FROM); crm_node_t *node = crm_find_peer(0, from); crm_update_peer_expected(__FUNCTION__, node, CRMD_JOINSTATE_DOWN); if(AM_I_DC == FALSE) { return I_NULL; /* Done */ } } /*========== DC-Only Actions ==========*/ if (AM_I_DC) { if (strcmp(op, CRM_OP_JOIN_ANNOUNCE) == 0) { return I_NODE_JOIN; } else if (strcmp(op, CRM_OP_JOIN_REQUEST) == 0) { return I_JOIN_REQUEST; } else if (strcmp(op, CRM_OP_JOIN_CONFIRM) == 0) { return I_JOIN_RESULT; } else if (strcmp(op, CRM_OP_SHUTDOWN) == 0) { return handle_shutdown_self_ack(stored_msg); } else if (strcmp(op, CRM_OP_SHUTDOWN_REQ) == 0) { /* a slave wants to shut down */ /* create cib fragment and add to message */ return handle_shutdown_request(stored_msg); } else if (strcmp(op, CRM_OP_REMOTE_STATE) == 0) { /* a remote connection host is letting us know the node state */ return handle_remote_state(stored_msg); } } /*========== common actions ==========*/ if (strcmp(op, CRM_OP_NOVOTE) == 0) { ha_msg_input_t fsa_input; fsa_input.msg = stored_msg; register_fsa_input_adv(C_HA_MESSAGE, I_NULL, &fsa_input, A_ELECTION_COUNT | A_ELECTION_CHECK, FALSE, __FUNCTION__); } else if (strcmp(op, CRM_OP_THROTTLE) == 0) { throttle_update(stored_msg); if (AM_I_DC && transition_graph != NULL) { if (transition_graph->complete == FALSE) { crm_debug("The throttle changed. Trigger a graph."); trigger_graph(); } } return I_NULL; } else if (strcmp(op, CRM_OP_CLEAR_FAILCOUNT) == 0) { return handle_failcount_op(stored_msg); } else if (strcmp(op, CRM_OP_VOTE) == 0) { /* count the vote and decide what to do after that */ ha_msg_input_t fsa_input; fsa_input.msg = stored_msg; register_fsa_input_adv(C_HA_MESSAGE, I_NULL, &fsa_input, A_ELECTION_COUNT | A_ELECTION_CHECK, FALSE, __FUNCTION__); /* Sometimes we _must_ go into S_ELECTION */ if (fsa_state == S_HALT) { crm_debug("Forcing an election from S_HALT"); return I_ELECTION; #if 0 } else if (AM_I_DC) { /* This is the old way of doing things but what is gained? */ return I_ELECTION; #endif } } else if (strcmp(op, CRM_OP_JOIN_OFFER) == 0) { verify_feature_set(stored_msg); crm_debug("Raising I_JOIN_OFFER: join-%s", crm_element_value(stored_msg, F_CRM_JOIN_ID)); return I_JOIN_OFFER; } else if (strcmp(op, CRM_OP_JOIN_ACKNAK) == 0) { crm_debug("Raising I_JOIN_RESULT: join-%s", crm_element_value(stored_msg, F_CRM_JOIN_ID)); return I_JOIN_RESULT; } else if (strcmp(op, CRM_OP_LRM_DELETE) == 0) { return handle_lrm_delete(stored_msg); } else if ((strcmp(op, CRM_OP_LRM_FAIL) == 0) || (strcmp(op, CRM_OP_LRM_REFRESH) == 0) || (strcmp(op, CRM_OP_REPROBE) == 0)) { crm_xml_add(stored_msg, F_CRM_SYS_TO, CRM_SYSTEM_LRMD); return I_ROUTER; } else if (strcmp(op, CRM_OP_NOOP) == 0) { return I_NULL; } else if (strcmp(op, CRM_OP_LOCAL_SHUTDOWN) == 0) { crm_shutdown(SIGTERM); /*return I_SHUTDOWN; */ return I_NULL; } else if (strcmp(op, CRM_OP_PING) == 0) { return handle_ping(stored_msg); } else if (strcmp(op, CRM_OP_NODE_INFO) == 0) { return handle_node_info_request(stored_msg); } else if (strcmp(op, CRM_OP_RM_NODE_CACHE) == 0) { int id = 0; const char *name = NULL; crm_element_value_int(stored_msg, XML_ATTR_ID, &id); name = crm_element_value(stored_msg, XML_ATTR_UNAME); if(cause == C_IPC_MESSAGE) { msg = create_request(CRM_OP_RM_NODE_CACHE, NULL, NULL, CRM_SYSTEM_CRMD, CRM_SYSTEM_CRMD, NULL); if (send_cluster_message(NULL, crm_msg_crmd, msg, TRUE) == FALSE) { crm_err("Could not instruct peers to remove references to node %s/%u", name, id); } else { crm_notice("Instructing peers to remove references to node %s/%u", name, id); } free_xml(msg); } else { reap_crm_member(id, name); /* If we're forgetting this node, also forget any failures to fence * it, so we don't carry that over to any node added later with the * same name. */ st_fail_count_reset(name); } } else if (strcmp(op, CRM_OP_MAINTENANCE_NODES) == 0) { xmlNode *xml = get_message_xml(stored_msg, F_CRM_DATA); remote_ra_process_maintenance_nodes(xml); } else if (strcmp(op, PCMK__CONTROLD_CMD_NODES) == 0) { return handle_node_list(stored_msg); /*========== (NOT_DC)-Only Actions ==========*/ } else if (!AM_I_DC) { if (strcmp(op, CRM_OP_SHUTDOWN) == 0) { return handle_shutdown_ack(stored_msg); } } else { crm_err("Unexpected request (%s) sent to %s", op, AM_I_DC ? "the DC" : "non-DC node"); crm_log_xml_err(stored_msg, "Unexpected"); } return I_NULL; } static void handle_response(xmlNode *stored_msg) { const char *op = crm_element_value(stored_msg, F_CRM_TASK); if (op == NULL) { crm_log_xml_err(stored_msg, "Bad message"); } else if (AM_I_DC && strcmp(op, CRM_OP_PECALC) == 0) { // Check whether scheduler answer been superseded by subsequent request const char *msg_ref = crm_element_value(stored_msg, XML_ATTR_REFERENCE); if (msg_ref == NULL) { crm_err("%s - Ignoring calculation with no reference", op); } else if (safe_str_eq(msg_ref, fsa_pe_ref)) { ha_msg_input_t fsa_input; controld_stop_sched_timer(); fsa_input.msg = stored_msg; register_fsa_input_later(C_IPC_MESSAGE, I_PE_SUCCESS, &fsa_input); } else { crm_info("%s calculation %s is obsolete", op, msg_ref); } } else if (strcmp(op, CRM_OP_VOTE) == 0 || strcmp(op, CRM_OP_SHUTDOWN_REQ) == 0 || strcmp(op, CRM_OP_SHUTDOWN) == 0) { } else { const char *host_from = crm_element_value(stored_msg, F_CRM_HOST_FROM); crm_err("Unexpected response (op=%s, src=%s) sent to the %s", op, host_from, AM_I_DC ? "DC" : "controller"); } } static enum crmd_fsa_input handle_shutdown_request(xmlNode * stored_msg) { /* handle here to avoid potential version issues * where the shutdown message/procedure may have * been changed in later versions. * * This way the DC is always in control of the shutdown */ char *now_s = NULL; time_t now = time(NULL); const char *host_from = crm_element_value(stored_msg, F_CRM_HOST_FROM); if (host_from == NULL) { /* we're shutting down and the DC */ host_from = fsa_our_uname; } crm_info("Creating shutdown request for %s (state=%s)", host_from, fsa_state2string(fsa_state)); crm_log_xml_trace(stored_msg, "message"); now_s = crm_itoa(now); update_attrd(host_from, XML_CIB_ATTR_SHUTDOWN, now_s, NULL, FALSE); free(now_s); /* will be picked up by the TE as long as its running */ return I_NULL; } /* msg is deleted by the time this returns */ extern gboolean process_te_message(xmlNode * msg, xmlNode * xml_data); static void send_msg_via_ipc(xmlNode * msg, const char *sys) { pcmk__client_t *client_channel = pcmk__find_client_by_id(sys); if (crm_element_value(msg, F_CRM_HOST_FROM) == NULL) { crm_xml_add(msg, F_CRM_HOST_FROM, fsa_our_uname); } if (client_channel != NULL) { /* Transient clients such as crmadmin */ pcmk__ipc_send_xml(client_channel, 0, msg, crm_ipc_server_event); } else if (sys != NULL && strcmp(sys, CRM_SYSTEM_TENGINE) == 0) { xmlNode *data = get_message_xml(msg, F_CRM_DATA); process_te_message(msg, data); } else if (sys != NULL && strcmp(sys, CRM_SYSTEM_LRMD) == 0) { fsa_data_t fsa_data; ha_msg_input_t fsa_input; fsa_input.msg = msg; fsa_input.xml = get_message_xml(msg, F_CRM_DATA); fsa_data.id = 0; fsa_data.actions = 0; fsa_data.data = &fsa_input; fsa_data.fsa_input = I_MESSAGE; fsa_data.fsa_cause = C_IPC_MESSAGE; fsa_data.origin = __FUNCTION__; fsa_data.data_type = fsa_dt_ha_msg; do_lrm_invoke(A_LRM_INVOKE, C_IPC_MESSAGE, fsa_state, I_MESSAGE, &fsa_data); } else if (sys != NULL && crmd_is_proxy_session(sys)) { crmd_proxy_send(sys, msg); } else { crm_debug("Unknown Sub-system (%s)... discarding message.", crm_str(sys)); } } void delete_ha_msg_input(ha_msg_input_t * orig) { if (orig == NULL) { return; } free_xml(orig->msg); free(orig); } /*! * \internal * \brief Notify the DC of a remote node state change * * \param[in] node_name Node's name * \param[in] node_up TRUE if node is up, FALSE if down */ void send_remote_state_message(const char *node_name, gboolean node_up) { /* If we don't have a DC, or the message fails, we have a failsafe: * the DC will eventually pick up the change via the CIB node state. * The message allows it to happen sooner if possible. */ if (fsa_our_dc) { xmlNode *msg = create_request(CRM_OP_REMOTE_STATE, NULL, fsa_our_dc, CRM_SYSTEM_DC, CRM_SYSTEM_CRMD, NULL); crm_info("Notifying DC %s of pacemaker_remote node %s %s", fsa_our_dc, node_name, (node_up? "coming up" : "going down")); crm_xml_add(msg, XML_ATTR_ID, node_name); crm_xml_add_boolean(msg, XML_NODE_IN_CLUSTER, node_up); send_cluster_message(crm_get_peer(0, fsa_our_dc), crm_msg_crmd, msg, TRUE); free_xml(msg); } else { crm_debug("No DC to notify of pacemaker_remote node %s %s", node_name, (node_up? "coming up" : "going down")); } } diff --git a/daemons/controld/controld_remote_ra.c b/daemons/controld/controld_remote_ra.c index 8dead202e9..f243624c70 100644 --- a/daemons/controld/controld_remote_ra.c +++ b/daemons/controld/controld_remote_ra.c @@ -1,1298 +1,1298 @@ /* * Copyright 2013-2020 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #define REMOTE_LRMD_RA "remote" /* The max start timeout before cmd retry */ #define MAX_START_TIMEOUT_MS 10000 typedef struct remote_ra_cmd_s { /*! the local node the cmd is issued from */ char *owner; /*! the remote node the cmd is executed on */ char *rsc_id; /*! the action to execute */ char *action; /*! some string the client wants us to give it back */ char *userdata; char *exit_reason; // descriptive text on error /*! start delay in ms */ int start_delay; /*! timer id used for start delay. */ int delay_id; /*! timeout in ms for cmd */ int timeout; int remaining_timeout; /*! recurring interval in ms */ guint interval_ms; /*! interval timer id */ int interval_id; int reported_success; int monitor_timeout_id; int takeover_timeout_id; /*! action parameters */ lrmd_key_value_t *params; /*! executed rc */ int rc; int op_status; int call_id; time_t start_time; gboolean cancel; } remote_ra_cmd_t; enum remote_migration_status { expect_takeover = 1, takeover_complete, }; typedef struct remote_ra_data_s { crm_trigger_t *work; remote_ra_cmd_t *cur_cmd; GList *cmds; GList *recurring_cmds; enum remote_migration_status migrate_status; gboolean active; /* Maintenance mode is difficult to determine from the controller's context, * so we have it signalled back with the transition from the scheduler. */ gboolean is_maintenance; /* Similar for whether we are controlling a guest node or remote node. * Fortunately there is a meta-attribute in the transition already and * as the situation doesn't change over time we can use the * resource start for noting down the information for later use when * the attributes aren't at hand. */ gboolean controlling_guest; } remote_ra_data_t; static int handle_remote_ra_start(lrm_state_t * lrm_state, remote_ra_cmd_t * cmd, int timeout_ms); static void handle_remote_ra_stop(lrm_state_t * lrm_state, remote_ra_cmd_t * cmd); static GList *fail_all_monitor_cmds(GList * list); static void free_cmd(gpointer user_data) { remote_ra_cmd_t *cmd = user_data; if (!cmd) { return; } if (cmd->delay_id) { g_source_remove(cmd->delay_id); } if (cmd->interval_id) { g_source_remove(cmd->interval_id); } if (cmd->monitor_timeout_id) { g_source_remove(cmd->monitor_timeout_id); } if (cmd->takeover_timeout_id) { g_source_remove(cmd->takeover_timeout_id); } free(cmd->owner); free(cmd->rsc_id); free(cmd->action); free(cmd->userdata); free(cmd->exit_reason); lrmd_key_value_freeall(cmd->params); free(cmd); } static int generate_callid(void) { static int remote_ra_callid = 0; remote_ra_callid++; if (remote_ra_callid <= 0) { remote_ra_callid = 1; } return remote_ra_callid; } static gboolean recurring_helper(gpointer data) { remote_ra_cmd_t *cmd = data; lrm_state_t *connection_rsc = NULL; cmd->interval_id = 0; connection_rsc = lrm_state_find(cmd->rsc_id); if (connection_rsc && connection_rsc->remote_ra_data) { remote_ra_data_t *ra_data = connection_rsc->remote_ra_data; ra_data->recurring_cmds = g_list_remove(ra_data->recurring_cmds, cmd); ra_data->cmds = g_list_append(ra_data->cmds, cmd); mainloop_set_trigger(ra_data->work); } return FALSE; } static gboolean start_delay_helper(gpointer data) { remote_ra_cmd_t *cmd = data; lrm_state_t *connection_rsc = NULL; cmd->delay_id = 0; connection_rsc = lrm_state_find(cmd->rsc_id); if (connection_rsc && connection_rsc->remote_ra_data) { remote_ra_data_t *ra_data = connection_rsc->remote_ra_data; mainloop_set_trigger(ra_data->work); } return FALSE; } /*! * \internal * \brief Handle cluster communication related to pacemaker_remote node joining * * \param[in] node_name Name of newly integrated pacemaker_remote node */ static void remote_node_up(const char *node_name) { int call_opt, call_id = 0; xmlNode *update, *state; crm_node_t *node; enum controld_section_e section = controld_section_all; CRM_CHECK(node_name != NULL, return); crm_info("Announcing pacemaker_remote node %s", node_name); /* Clear node's entire state (resource history and transient attributes) * other than shutdown locks. The transient attributes should and normally * will be cleared when the node leaves, but since remote node state has a * number of corner cases, clear them here as well, to be sure. */ call_opt = crmd_cib_smart_opt(); if (controld_shutdown_lock_enabled) { section = controld_section_all_unlocked; } controld_delete_node_state(node_name, section, call_opt); /* Clear node's probed attribute */ update_attrd(node_name, CRM_OP_PROBED, NULL, NULL, TRUE); /* Ensure node is in the remote peer cache with member status */ node = crm_remote_peer_get(node_name); CRM_CHECK(node != NULL, return); crm_update_peer_state(__FUNCTION__, node, CRM_NODE_MEMBER, 0); /* pacemaker_remote nodes don't participate in the membership layer, * so cluster nodes don't automatically get notified when they come and go. * We send a cluster message to the DC, and update the CIB node state entry, * so the DC will get it sooner (via message) or later (via CIB refresh), * and any other interested parties can query the CIB. */ send_remote_state_message(node_name, TRUE); update = create_xml_node(NULL, XML_CIB_TAG_STATUS); state = create_node_state_update(node, node_update_cluster, update, __FUNCTION__); /* Clear the XML_NODE_IS_FENCED flag in the node state. If the node ever * needs to be fenced, this flag will allow various actions to determine * whether the fencing has happened yet. */ crm_xml_add(state, XML_NODE_IS_FENCED, "0"); /* TODO: If the remote connection drops, and this (async) CIB update either * failed or has not yet completed, later actions could mistakenly think the * node has already been fenced (if the XML_NODE_IS_FENCED attribute was * previously set, because it won't have been cleared). This could prevent * actual fencing or allow recurring monitor failures to be cleared too * soon. Ideally, we wouldn't rely on the CIB for the fenced status. */ fsa_cib_update(XML_CIB_TAG_STATUS, update, call_opt, call_id, NULL); if (call_id < 0) { crm_perror(LOG_WARNING, "%s CIB node state setup", node_name); } free_xml(update); } enum down_opts { DOWN_KEEP_LRM, DOWN_ERASE_LRM }; /*! * \internal * \brief Handle cluster communication related to pacemaker_remote node leaving * * \param[in] node_name Name of lost node * \param[in] opts Whether to keep or erase LRM history */ static void remote_node_down(const char *node_name, const enum down_opts opts) { xmlNode *update; int call_id = 0; int call_opt = crmd_cib_smart_opt(); crm_node_t *node; /* Purge node from attrd's memory */ update_attrd_remote_node_removed(node_name, NULL); /* Normally, only node attributes should be erased, and the resource history * should be kept until the node comes back up. However, after a successful * fence, we want to clear the history as well, so we don't think resources * are still running on the node. */ if (opts == DOWN_ERASE_LRM) { controld_delete_node_state(node_name, controld_section_all, call_opt); } else { controld_delete_node_state(node_name, controld_section_attrs, call_opt); } /* Ensure node is in the remote peer cache with lost state */ node = crm_remote_peer_get(node_name); CRM_CHECK(node != NULL, return); crm_update_peer_state(__FUNCTION__, node, CRM_NODE_LOST, 0); /* Notify DC */ send_remote_state_message(node_name, FALSE); /* Update CIB node state */ update = create_xml_node(NULL, XML_CIB_TAG_STATUS); create_node_state_update(node, node_update_cluster, update, __FUNCTION__); fsa_cib_update(XML_CIB_TAG_STATUS, update, call_opt, call_id, NULL); if (call_id < 0) { crm_perror(LOG_ERR, "%s CIB node state update", node_name); } free_xml(update); } /*! * \internal * \brief Handle effects of a remote RA command on node state * * \param[in] cmd Completed remote RA command */ static void check_remote_node_state(remote_ra_cmd_t *cmd) { /* Only successful actions can change node state */ if (cmd->rc != PCMK_OCF_OK) { return; } if (safe_str_eq(cmd->action, "start")) { remote_node_up(cmd->rsc_id); } else if (safe_str_eq(cmd->action, "migrate_from")) { /* After a successful migration, we don't need to do remote_node_up() * because the DC already knows the node is up, and we don't want to * clear LRM history etc. We do need to add the remote node to this * host's remote peer cache, because (unless it happens to be DC) * it hasn't been tracking the remote node, and other code relies on * the cache to distinguish remote nodes from unseen cluster nodes. */ crm_node_t *node = crm_remote_peer_get(cmd->rsc_id); CRM_CHECK(node != NULL, return); crm_update_peer_state(__FUNCTION__, node, CRM_NODE_MEMBER, 0); } else if (safe_str_eq(cmd->action, "stop")) { lrm_state_t *lrm_state = lrm_state_find(cmd->rsc_id); remote_ra_data_t *ra_data = lrm_state? lrm_state->remote_ra_data : NULL; if (ra_data) { if (ra_data->migrate_status != takeover_complete) { /* Stop means down if we didn't successfully migrate elsewhere */ remote_node_down(cmd->rsc_id, DOWN_KEEP_LRM); } else if (AM_I_DC == FALSE) { /* Only the connection host and DC track node state, * so if the connection migrated elsewhere and we aren't DC, * un-cache the node, so we don't have stale info */ crm_remote_peer_cache_remove(cmd->rsc_id); } } } /* We don't do anything for successful monitors, which is correct for * routine recurring monitors, and for monitors on nodes where the * connection isn't supposed to be (the cluster will stop the connection in * that case). However, if the initial probe finds the connection already * active on the node where we want it, we probably should do * remote_node_up(). Unfortunately, we can't distinguish that case here. * Given that connections have to be initiated by the cluster, the chance of * that should be close to zero. */ } static void report_remote_ra_result(remote_ra_cmd_t * cmd) { lrmd_event_data_t op = { 0, }; check_remote_node_state(cmd); op.type = lrmd_event_exec_complete; op.rsc_id = cmd->rsc_id; op.op_type = cmd->action; op.user_data = cmd->userdata; op.exit_reason = cmd->exit_reason; op.timeout = cmd->timeout; op.interval_ms = cmd->interval_ms; op.rc = cmd->rc; op.op_status = cmd->op_status; op.t_run = (unsigned int) cmd->start_time; op.t_rcchange = (unsigned int) cmd->start_time; if (cmd->reported_success && cmd->rc != PCMK_OCF_OK) { op.t_rcchange = (unsigned int) time(NULL); /* This edge case will likely never ever occur, but if it does the * result is that a failure will not be processed correctly. This is only * remotely possible because we are able to detect a connection resource's tcp * connection has failed at any moment after start has completed. The actual * recurring operation is just a connectivity ping. * * basically, we are not guaranteed that the first successful monitor op and * a subsequent failed monitor op will not occur in the same timestamp. We have to * make it look like the operations occurred at separate times though. */ if (op.t_rcchange == op.t_run) { op.t_rcchange++; } } if (cmd->params) { lrmd_key_value_t *tmp; op.params = crm_str_table_new(); for (tmp = cmd->params; tmp; tmp = tmp->next) { g_hash_table_insert(op.params, strdup(tmp->key), strdup(tmp->value)); } } op.call_id = cmd->call_id; op.remote_nodename = cmd->owner; lrm_op_callback(&op); if (op.params) { g_hash_table_destroy(op.params); } } static void update_remaining_timeout(remote_ra_cmd_t * cmd) { cmd->remaining_timeout = ((cmd->timeout / 1000) - (time(NULL) - cmd->start_time)) * 1000; } static gboolean retry_start_cmd_cb(gpointer data) { lrm_state_t *lrm_state = data; remote_ra_data_t *ra_data = lrm_state->remote_ra_data; remote_ra_cmd_t *cmd = NULL; int rc = -1; if (!ra_data || !ra_data->cur_cmd) { return FALSE; } cmd = ra_data->cur_cmd; - if (pcmk__str_none_of(cmd->action, "start", "migrate_from", NULL)) { + if (!pcmk__str_any_of(cmd->action, "start", "migrate_from", NULL)) { return FALSE; } update_remaining_timeout(cmd); if (cmd->remaining_timeout > 0) { rc = handle_remote_ra_start(lrm_state, cmd, cmd->remaining_timeout); } if (rc != 0) { cmd->rc = PCMK_OCF_UNKNOWN_ERROR; cmd->op_status = PCMK_LRM_OP_ERROR; report_remote_ra_result(cmd); if (ra_data->cmds) { mainloop_set_trigger(ra_data->work); } ra_data->cur_cmd = NULL; free_cmd(cmd); } else { /* wait for connection event */ } return FALSE; } static gboolean connection_takeover_timeout_cb(gpointer data) { lrm_state_t *lrm_state = NULL; remote_ra_cmd_t *cmd = data; crm_info("takeover event timed out for node %s", cmd->rsc_id); cmd->takeover_timeout_id = 0; lrm_state = lrm_state_find(cmd->rsc_id); handle_remote_ra_stop(lrm_state, cmd); free_cmd(cmd); return FALSE; } static gboolean monitor_timeout_cb(gpointer data) { lrm_state_t *lrm_state = NULL; remote_ra_cmd_t *cmd = data; lrm_state = lrm_state_find(cmd->rsc_id); crm_info("Timed out waiting for remote poke response from %s%s", cmd->rsc_id, (lrm_state? "" : " (no LRM state)")); cmd->monitor_timeout_id = 0; cmd->op_status = PCMK_LRM_OP_TIMEOUT; cmd->rc = PCMK_OCF_UNKNOWN_ERROR; if (lrm_state && lrm_state->remote_ra_data) { remote_ra_data_t *ra_data = lrm_state->remote_ra_data; if (ra_data->cur_cmd == cmd) { ra_data->cur_cmd = NULL; } if (ra_data->cmds) { mainloop_set_trigger(ra_data->work); } } report_remote_ra_result(cmd); free_cmd(cmd); if(lrm_state) { lrm_state_disconnect(lrm_state); } return FALSE; } static void synthesize_lrmd_success(lrm_state_t *lrm_state, const char *rsc_id, const char *op_type) { lrmd_event_data_t op = { 0, }; if (lrm_state == NULL) { /* if lrm_state not given assume local */ lrm_state = lrm_state_find(fsa_our_uname); } CRM_ASSERT(lrm_state != NULL); op.type = lrmd_event_exec_complete; op.rsc_id = rsc_id; op.op_type = op_type; op.rc = PCMK_OCF_OK; op.op_status = PCMK_LRM_OP_DONE; op.t_run = (unsigned int) time(NULL); op.t_rcchange = op.t_run; op.call_id = generate_callid(); process_lrm_event(lrm_state, &op, NULL, NULL); } void remote_lrm_op_callback(lrmd_event_data_t * op) { gboolean cmd_handled = FALSE; lrm_state_t *lrm_state = NULL; remote_ra_data_t *ra_data = NULL; remote_ra_cmd_t *cmd = NULL; crm_debug("Processing '%s%s%s' event on remote connection to %s: %s " "(%d) status=%s (%d)", (op->op_type? op->op_type : ""), (op->op_type? " " : ""), lrmd_event_type2str(op->type), op->remote_nodename, services_ocf_exitcode_str(op->rc), op->rc, services_lrm_status_str(op->op_status), op->op_status); lrm_state = lrm_state_find(op->remote_nodename); if (!lrm_state || !lrm_state->remote_ra_data) { crm_debug("No state information found for remote connection event"); return; } ra_data = lrm_state->remote_ra_data; if (op->type == lrmd_event_new_client) { // Another client has connected to the remote daemon if (ra_data->migrate_status == expect_takeover) { // Great, we knew this was coming ra_data->migrate_status = takeover_complete; } else { crm_err("Unexpected pacemaker_remote client takeover for %s. Disconnecting", op->remote_nodename); /* In this case, lrmd_tls_connection_destroy() will be called under the control of mainloop. */ /* Do not free lrm_state->conn yet. */ /* It'll be freed in the following stop action. */ lrm_state_disconnect_only(lrm_state); } return; } /* filter all EXEC events up */ if (op->type == lrmd_event_exec_complete) { if (ra_data->migrate_status == takeover_complete) { crm_debug("ignoring event, this connection is taken over by another node"); } else { lrm_op_callback(op); } return; } if ((op->type == lrmd_event_disconnect) && (ra_data->cur_cmd == NULL)) { if (ra_data->active == FALSE) { crm_debug("Disconnection from Pacemaker Remote node %s complete", lrm_state->node_name); } else if (!remote_ra_is_in_maintenance(lrm_state)) { crm_err("Lost connection to Pacemaker Remote node %s", lrm_state->node_name); ra_data->recurring_cmds = fail_all_monitor_cmds(ra_data->recurring_cmds); ra_data->cmds = fail_all_monitor_cmds(ra_data->cmds); } else { crm_notice("Unmanaged Pacemaker Remote node %s disconnected", lrm_state->node_name); /* Do roughly what a 'stop' on the remote-resource would do */ handle_remote_ra_stop(lrm_state, NULL); remote_node_down(lrm_state->node_name, DOWN_KEEP_LRM); /* now fake the reply of a successful 'stop' */ synthesize_lrmd_success(NULL, lrm_state->node_name, "stop"); } return; } if (!ra_data->cur_cmd) { crm_debug("no event to match"); return; } cmd = ra_data->cur_cmd; /* Start actions and migrate from actions complete after connection * comes back to us. */ if (op->type == lrmd_event_connect && pcmk__str_any_of(cmd->action, "start", "migrate_from", NULL)) { if (op->connection_rc < 0) { update_remaining_timeout(cmd); if (op->connection_rc == -ENOKEY) { // Hard error, don't retry cmd->op_status = PCMK_LRM_OP_ERROR; cmd->rc = PCMK_OCF_INVALID_PARAM; cmd->exit_reason = strdup("Authentication key not readable"); } else if (cmd->remaining_timeout > 3000) { crm_trace("rescheduling start, remaining timeout %d", cmd->remaining_timeout); g_timeout_add(1000, retry_start_cmd_cb, lrm_state); return; } else { crm_trace("can't reschedule start, remaining timeout too small %d", cmd->remaining_timeout); cmd->op_status = PCMK_LRM_OP_TIMEOUT; cmd->rc = PCMK_OCF_UNKNOWN_ERROR; } } else { lrm_state_reset_tables(lrm_state, TRUE); cmd->rc = PCMK_OCF_OK; cmd->op_status = PCMK_LRM_OP_DONE; ra_data->active = TRUE; } crm_debug("Remote connection event matched %s action", cmd->action); report_remote_ra_result(cmd); cmd_handled = TRUE; } else if (op->type == lrmd_event_poke && safe_str_eq(cmd->action, "monitor")) { if (cmd->monitor_timeout_id) { g_source_remove(cmd->monitor_timeout_id); cmd->monitor_timeout_id = 0; } /* Only report success the first time, after that only worry about failures. * For this function, if we get the poke pack, it is always a success. Pokes * only fail if the send fails, or the response times out. */ if (!cmd->reported_success) { cmd->rc = PCMK_OCF_OK; cmd->op_status = PCMK_LRM_OP_DONE; report_remote_ra_result(cmd); cmd->reported_success = 1; } crm_debug("Remote poke event matched %s action", cmd->action); /* success, keep rescheduling if interval is present. */ if (cmd->interval_ms && (cmd->cancel == FALSE)) { ra_data->recurring_cmds = g_list_append(ra_data->recurring_cmds, cmd); cmd->interval_id = g_timeout_add(cmd->interval_ms, recurring_helper, cmd); cmd = NULL; /* prevent free */ } cmd_handled = TRUE; } else if (op->type == lrmd_event_disconnect && safe_str_eq(cmd->action, "monitor")) { if (ra_data->active == TRUE && (cmd->cancel == FALSE)) { cmd->rc = PCMK_OCF_UNKNOWN_ERROR; cmd->op_status = PCMK_LRM_OP_ERROR; report_remote_ra_result(cmd); crm_err("Remote connection to %s unexpectedly dropped during monitor", lrm_state->node_name); } cmd_handled = TRUE; } else if (op->type == lrmd_event_new_client && safe_str_eq(cmd->action, "stop")) { handle_remote_ra_stop(lrm_state, cmd); cmd_handled = TRUE; } else { crm_debug("Event did not match %s action", ra_data->cur_cmd->action); } if (cmd_handled) { ra_data->cur_cmd = NULL; if (ra_data->cmds) { mainloop_set_trigger(ra_data->work); } free_cmd(cmd); } } static void handle_remote_ra_stop(lrm_state_t * lrm_state, remote_ra_cmd_t * cmd) { remote_ra_data_t *ra_data = NULL; CRM_ASSERT(lrm_state); ra_data = lrm_state->remote_ra_data; if (ra_data->migrate_status != takeover_complete) { /* delete pending ops when ever the remote connection is intentionally stopped */ g_hash_table_remove_all(lrm_state->pending_ops); } else { /* we no longer hold the history if this connection has been migrated, * however, we keep metadata cache for future use */ lrm_state_reset_tables(lrm_state, FALSE); } ra_data->active = FALSE; lrm_state_disconnect(lrm_state); if (ra_data->cmds) { g_list_free_full(ra_data->cmds, free_cmd); } if (ra_data->recurring_cmds) { g_list_free_full(ra_data->recurring_cmds, free_cmd); } ra_data->cmds = NULL; ra_data->recurring_cmds = NULL; ra_data->cur_cmd = NULL; if (cmd) { cmd->rc = PCMK_OCF_OK; cmd->op_status = PCMK_LRM_OP_DONE; report_remote_ra_result(cmd); } } static int handle_remote_ra_start(lrm_state_t * lrm_state, remote_ra_cmd_t * cmd, int timeout_ms) { const char *server = NULL; lrmd_key_value_t *tmp = NULL; int port = 0; remote_ra_data_t *ra_data = lrm_state->remote_ra_data; int timeout_used = timeout_ms > MAX_START_TIMEOUT_MS ? MAX_START_TIMEOUT_MS : timeout_ms; for (tmp = cmd->params; tmp; tmp = tmp->next) { if (pcmk__str_any_of(tmp->key, XML_RSC_ATTR_REMOTE_RA_ADDR, XML_RSC_ATTR_REMOTE_RA_SERVER, NULL)) { server = tmp->value; } else if (safe_str_eq(tmp->key, XML_RSC_ATTR_REMOTE_RA_PORT)) { port = atoi(tmp->value); } else if (safe_str_eq(tmp->key, CRM_META"_"XML_RSC_ATTR_CONTAINER)) { ra_data->controlling_guest = TRUE; } } return lrm_state_remote_connect_async(lrm_state, server, port, timeout_used); } static gboolean handle_remote_ra_exec(gpointer user_data) { int rc = 0; lrm_state_t *lrm_state = user_data; remote_ra_data_t *ra_data = lrm_state->remote_ra_data; remote_ra_cmd_t *cmd; GList *first = NULL; if (ra_data->cur_cmd) { /* still waiting on previous cmd */ return TRUE; } while (ra_data->cmds) { first = ra_data->cmds; cmd = first->data; if (cmd->delay_id) { /* still waiting for start delay timer to trip */ return TRUE; } ra_data->cmds = g_list_remove_link(ra_data->cmds, first); g_list_free_1(first); if (!strcmp(cmd->action, "start") || !strcmp(cmd->action, "migrate_from")) { ra_data->migrate_status = 0; rc = handle_remote_ra_start(lrm_state, cmd, cmd->timeout); if (rc == 0) { /* take care of this later when we get async connection result */ crm_debug("Initiated async remote connection, %s action will complete after connect event", cmd->action); ra_data->cur_cmd = cmd; return TRUE; } else { crm_debug("Could not initiate remote connection for %s action", cmd->action); cmd->rc = PCMK_OCF_UNKNOWN_ERROR; cmd->op_status = PCMK_LRM_OP_ERROR; } report_remote_ra_result(cmd); } else if (!strcmp(cmd->action, "monitor")) { if (lrm_state_is_connected(lrm_state) == TRUE) { rc = lrm_state_poke_connection(lrm_state); if (rc < 0) { cmd->rc = PCMK_OCF_UNKNOWN_ERROR; cmd->op_status = PCMK_LRM_OP_ERROR; } } else { rc = -1; cmd->op_status = PCMK_LRM_OP_DONE; cmd->rc = PCMK_OCF_NOT_RUNNING; } if (rc == 0) { crm_debug("Poked Pacemaker Remote at node %s, waiting for async response", cmd->rsc_id); ra_data->cur_cmd = cmd; cmd->monitor_timeout_id = g_timeout_add(cmd->timeout, monitor_timeout_cb, cmd); return TRUE; } report_remote_ra_result(cmd); } else if (!strcmp(cmd->action, "stop")) { if (ra_data->migrate_status == expect_takeover) { /* briefly wait on stop for the takeover event to occur. If the * takeover event does not occur during the wait period, that's fine. * It just means that the remote-node's lrm_status section is going to get * cleared which will require all the resources running in the remote-node * to be explicitly re-detected via probe actions. If the takeover does occur * successfully, then we can leave the status section intact. */ cmd->takeover_timeout_id = g_timeout_add((cmd->timeout/2), connection_takeover_timeout_cb, cmd); ra_data->cur_cmd = cmd; return TRUE; } handle_remote_ra_stop(lrm_state, cmd); } else if (!strcmp(cmd->action, "migrate_to")) { ra_data->migrate_status = expect_takeover; cmd->rc = PCMK_OCF_OK; cmd->op_status = PCMK_LRM_OP_DONE; report_remote_ra_result(cmd); } else if (!strcmp(cmd->action, "reload")) { /* reloads are a no-op right now, add logic here when they become important */ cmd->rc = PCMK_OCF_OK; cmd->op_status = PCMK_LRM_OP_DONE; report_remote_ra_result(cmd); } free_cmd(cmd); } return TRUE; } static void remote_ra_data_init(lrm_state_t * lrm_state) { remote_ra_data_t *ra_data = NULL; if (lrm_state->remote_ra_data) { return; } ra_data = calloc(1, sizeof(remote_ra_data_t)); ra_data->work = mainloop_add_trigger(G_PRIORITY_HIGH, handle_remote_ra_exec, lrm_state); lrm_state->remote_ra_data = ra_data; } void remote_ra_cleanup(lrm_state_t * lrm_state) { remote_ra_data_t *ra_data = lrm_state->remote_ra_data; if (!ra_data) { return; } if (ra_data->cmds) { g_list_free_full(ra_data->cmds, free_cmd); } if (ra_data->recurring_cmds) { g_list_free_full(ra_data->recurring_cmds, free_cmd); } mainloop_destroy_trigger(ra_data->work); free(ra_data); lrm_state->remote_ra_data = NULL; } gboolean is_remote_lrmd_ra(const char *agent, const char *provider, const char *id) { if (agent && provider && !strcmp(agent, REMOTE_LRMD_RA) && !strcmp(provider, "pacemaker")) { return TRUE; } if (id && lrm_state_find(id) && safe_str_neq(id, fsa_our_uname)) { return TRUE; } return FALSE; } lrmd_rsc_info_t * remote_ra_get_rsc_info(lrm_state_t * lrm_state, const char *rsc_id) { lrmd_rsc_info_t *info = NULL; if ((lrm_state_find(rsc_id))) { info = calloc(1, sizeof(lrmd_rsc_info_t)); info->id = strdup(rsc_id); info->type = strdup(REMOTE_LRMD_RA); info->standard = strdup(PCMK_RESOURCE_CLASS_OCF); info->provider = strdup("pacemaker"); } return info; } static gboolean is_remote_ra_supported_action(const char *action) { if (!action) { return FALSE; } else if (strcmp(action, "start") && strcmp(action, "stop") && strcmp(action, "reload") && strcmp(action, "migrate_to") && strcmp(action, "migrate_from") && strcmp(action, "monitor")) { return FALSE; } return TRUE; } static GList * fail_all_monitor_cmds(GList * list) { GList *rm_list = NULL; remote_ra_cmd_t *cmd = NULL; GListPtr gIter = NULL; for (gIter = list; gIter != NULL; gIter = gIter->next) { cmd = gIter->data; if ((cmd->interval_ms > 0) && safe_str_eq(cmd->action, "monitor")) { rm_list = g_list_append(rm_list, cmd); } } for (gIter = rm_list; gIter != NULL; gIter = gIter->next) { cmd = gIter->data; cmd->rc = PCMK_OCF_UNKNOWN_ERROR; cmd->op_status = PCMK_LRM_OP_ERROR; crm_trace("Pre-emptively failing %s %s (interval=%u, %s)", cmd->action, cmd->rsc_id, cmd->interval_ms, cmd->userdata); report_remote_ra_result(cmd); list = g_list_remove(list, cmd); free_cmd(cmd); } /* frees only the list data, not the cmds */ g_list_free(rm_list); return list; } static GList * remove_cmd(GList * list, const char *action, guint interval_ms) { remote_ra_cmd_t *cmd = NULL; GListPtr gIter = NULL; for (gIter = list; gIter != NULL; gIter = gIter->next) { cmd = gIter->data; if ((cmd->interval_ms == interval_ms) && safe_str_eq(cmd->action, action)) { break; } cmd = NULL; } if (cmd) { list = g_list_remove(list, cmd); free_cmd(cmd); } return list; } int remote_ra_cancel(lrm_state_t *lrm_state, const char *rsc_id, const char *action, guint interval_ms) { lrm_state_t *connection_rsc = NULL; remote_ra_data_t *ra_data = NULL; connection_rsc = lrm_state_find(rsc_id); if (!connection_rsc || !connection_rsc->remote_ra_data) { return -EINVAL; } ra_data = connection_rsc->remote_ra_data; ra_data->cmds = remove_cmd(ra_data->cmds, action, interval_ms); ra_data->recurring_cmds = remove_cmd(ra_data->recurring_cmds, action, interval_ms); if (ra_data->cur_cmd && (ra_data->cur_cmd->interval_ms == interval_ms) && (safe_str_eq(ra_data->cur_cmd->action, action))) { ra_data->cur_cmd->cancel = TRUE; } return 0; } static remote_ra_cmd_t * handle_dup_monitor(remote_ra_data_t *ra_data, guint interval_ms, const char *userdata) { GList *gIter = NULL; remote_ra_cmd_t *cmd = NULL; /* there are 3 places a potential duplicate monitor operation * could exist. * 1. recurring_cmds list. where the op is waiting for its next interval * 2. cmds list, where the op is queued to get executed immediately * 3. cur_cmd, which means the monitor op is in flight right now. */ if (interval_ms == 0) { return NULL; } if (ra_data->cur_cmd && ra_data->cur_cmd->cancel == FALSE && (ra_data->cur_cmd->interval_ms == interval_ms) && safe_str_eq(ra_data->cur_cmd->action, "monitor")) { cmd = ra_data->cur_cmd; goto handle_dup; } for (gIter = ra_data->recurring_cmds; gIter != NULL; gIter = gIter->next) { cmd = gIter->data; if ((cmd->interval_ms == interval_ms) && safe_str_eq(cmd->action, "monitor")) { goto handle_dup; } } for (gIter = ra_data->cmds; gIter != NULL; gIter = gIter->next) { cmd = gIter->data; if ((cmd->interval_ms == interval_ms) && safe_str_eq(cmd->action, "monitor")) { goto handle_dup; } } return NULL; handle_dup: crm_trace("merging duplicate monitor cmd " PCMK__OP_FMT, cmd->rsc_id, "monitor", interval_ms); /* update the userdata */ if (userdata) { free(cmd->userdata); cmd->userdata = strdup(userdata); } /* if we've already reported success, generate a new call id */ if (cmd->reported_success) { cmd->start_time = time(NULL); cmd->call_id = generate_callid(); cmd->reported_success = 0; } /* if we have an interval_id set, that means we are in the process of * waiting for this cmd's next interval. instead of waiting, cancel * the timer and execute the action immediately */ if (cmd->interval_id) { g_source_remove(cmd->interval_id); cmd->interval_id = 0; recurring_helper(cmd); } return cmd; } int remote_ra_exec(lrm_state_t *lrm_state, const char *rsc_id, const char *action, const char *userdata, guint interval_ms, int timeout, /* ms */ int start_delay, /* ms */ lrmd_key_value_t * params) { int rc = 0; lrm_state_t *connection_rsc = NULL; remote_ra_cmd_t *cmd = NULL; remote_ra_data_t *ra_data = NULL; if (is_remote_ra_supported_action(action) == FALSE) { rc = -EINVAL; goto exec_done; } connection_rsc = lrm_state_find(rsc_id); if (!connection_rsc) { rc = -EINVAL; goto exec_done; } remote_ra_data_init(connection_rsc); ra_data = connection_rsc->remote_ra_data; cmd = handle_dup_monitor(ra_data, interval_ms, userdata); if (cmd) { rc = cmd->call_id; goto exec_done; } cmd = calloc(1, sizeof(remote_ra_cmd_t)); cmd->owner = strdup(lrm_state->node_name); cmd->rsc_id = strdup(rsc_id); cmd->action = strdup(action); cmd->userdata = strdup(userdata); cmd->interval_ms = interval_ms; cmd->timeout = timeout; cmd->start_delay = start_delay; cmd->params = params; cmd->start_time = time(NULL); cmd->call_id = generate_callid(); if (cmd->start_delay) { cmd->delay_id = g_timeout_add(cmd->start_delay, start_delay_helper, cmd); } ra_data->cmds = g_list_append(ra_data->cmds, cmd); mainloop_set_trigger(ra_data->work); return cmd->call_id; exec_done: lrmd_key_value_freeall(params); return rc; } /*! * \internal * \brief Immediately fail all monitors of a remote node, if proxied here * * \param[in] node_name Name of pacemaker_remote node */ void remote_ra_fail(const char *node_name) { lrm_state_t *lrm_state = lrm_state_find(node_name); if (lrm_state && lrm_state_is_connected(lrm_state)) { remote_ra_data_t *ra_data = lrm_state->remote_ra_data; crm_info("Failing monitors on pacemaker_remote node %s", node_name); ra_data->recurring_cmds = fail_all_monitor_cmds(ra_data->recurring_cmds); ra_data->cmds = fail_all_monitor_cmds(ra_data->cmds); } } /* A guest node fencing implied by host fencing looks like: * * * * * * * */ #define XPATH_PSEUDO_FENCE "//" XML_GRAPH_TAG_PSEUDO_EVENT \ "[@" XML_LRM_ATTR_TASK "='stonith']/" XML_GRAPH_TAG_DOWNED \ "/" XML_CIB_TAG_NODE /*! * \internal * \brief Check a pseudo-action for Pacemaker Remote node side effects * * \param[in] xml XML of pseudo-action to check */ void remote_ra_process_pseudo(xmlNode *xml) { xmlXPathObjectPtr search = xpath_search(xml, XPATH_PSEUDO_FENCE); if (numXpathResults(search) == 1) { xmlNode *result = getXpathResult(search, 0); /* Normally, we handle the necessary side effects of a guest node stop * action when reporting the remote agent's result. However, if the stop * is implied due to fencing, it will be a fencing pseudo-event, and * there won't be a result to report. Handle that case here. * * This will result in a duplicate call to remote_node_down() if the * guest stop was real instead of implied, but that shouldn't hurt. * * There is still one corner case that isn't handled: if a guest node * isn't running any resources when its host is fenced, it will appear * to be cleanly stopped, so there will be no pseudo-fence, and our * peer cache state will be incorrect unless and until the guest is * recovered. */ if (result) { const char *remote = ID(result); if (remote) { remote_node_down(remote, DOWN_ERASE_LRM); } } } freeXpathObject(search); } static void remote_ra_maintenance(lrm_state_t * lrm_state, gboolean maintenance) { remote_ra_data_t *ra_data = lrm_state->remote_ra_data; xmlNode *update, *state; int call_opt, call_id = 0; crm_node_t *node; call_opt = crmd_cib_smart_opt(); node = crm_remote_peer_get(lrm_state->node_name); CRM_CHECK(node != NULL, return); update = create_xml_node(NULL, XML_CIB_TAG_STATUS); state = create_node_state_update(node, node_update_none, update, __FUNCTION__); crm_xml_add(state, XML_NODE_IS_MAINTENANCE, maintenance?"1":"0"); fsa_cib_update(XML_CIB_TAG_STATUS, update, call_opt, call_id, NULL); if (call_id < 0) { crm_perror(LOG_WARNING, "%s CIB node state update failed", lrm_state->node_name); } else { /* TODO: still not 100% sure that async update will succeed ... */ ra_data->is_maintenance = maintenance; } free_xml(update); } #define XPATH_PSEUDO_MAINTENANCE "//" XML_GRAPH_TAG_PSEUDO_EVENT \ "[@" XML_LRM_ATTR_TASK "='" CRM_OP_MAINTENANCE_NODES "']/" \ XML_GRAPH_TAG_MAINTENANCE /*! * \internal * \brief Check a pseudo-action holding updates for maintenance state * * \param[in] xml XML of pseudo-action to check */ void remote_ra_process_maintenance_nodes(xmlNode *xml) { xmlXPathObjectPtr search = xpath_search(xml, XPATH_PSEUDO_MAINTENANCE); if (numXpathResults(search) == 1) { xmlNode *node; int cnt = 0, cnt_remote = 0; for (node = first_named_child(getXpathResult(search, 0), XML_CIB_TAG_NODE); node; node = __xml_next(node)) { lrm_state_t *lrm_state = lrm_state_find(ID(node)); cnt++; if (lrm_state && lrm_state->remote_ra_data && ((remote_ra_data_t *) lrm_state->remote_ra_data)->active) { cnt_remote++; remote_ra_maintenance(lrm_state, crm_atoi(crm_element_value(node, XML_NODE_IS_MAINTENANCE), "0")); } } crm_trace("Action holds %d nodes (%d remotes found) " "adjusting maintenance-mode", cnt, cnt_remote); } freeXpathObject(search); } gboolean remote_ra_is_in_maintenance(lrm_state_t * lrm_state) { remote_ra_data_t *ra_data = lrm_state->remote_ra_data; return ra_data->is_maintenance; } gboolean remote_ra_controlling_guest(lrm_state_t * lrm_state) { remote_ra_data_t *ra_data = lrm_state->remote_ra_data; return ra_data->controlling_guest; } diff --git a/daemons/execd/pacemaker-execd.c b/daemons/execd/pacemaker-execd.c index d2fde151c8..29a6bda235 100644 --- a/daemons/execd/pacemaker-execd.c +++ b/daemons/execd/pacemaker-execd.c @@ -1,517 +1,517 @@ /* * Copyright 2012-2020 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #include #include #include #include #include "pacemaker-execd.h" #if defined(HAVE_GNUTLS_GNUTLS_H) && defined(SUPPORT_REMOTE) # define ENABLE_PCMK_REMOTE #endif static GMainLoop *mainloop = NULL; static qb_ipcs_service_t *ipcs = NULL; static stonith_t *stonith_api = NULL; int lrmd_call_id = 0; #ifdef ENABLE_PCMK_REMOTE /* whether shutdown request has been sent */ static sig_atomic_t shutting_down = FALSE; /* timer for waiting for acknowledgment of shutdown request */ static guint shutdown_ack_timer = 0; static gboolean lrmd_exit(gpointer data); #endif static void stonith_connection_destroy_cb(stonith_t * st, stonith_event_t * e) { stonith_api->state = stonith_disconnected; crm_err("Connection to fencer lost"); stonith_connection_failed(); } stonith_t * get_stonith_connection(void) { if (stonith_api && stonith_api->state == stonith_disconnected) { stonith_api_delete(stonith_api); stonith_api = NULL; } if (stonith_api == NULL) { int rc = pcmk_ok; stonith_api = stonith_api_new(); if (stonith_api == NULL) { crm_err("Could not connect to fencer: API memory allocation failed"); return NULL; } rc = stonith_api_connect_retry(stonith_api, crm_system_name, 10); if (rc != pcmk_ok) { crm_err("Could not connect to fencer in 10 attempts: %s " CRM_XS " rc=%d", pcmk_strerror(rc), rc); stonith_api_delete(stonith_api); stonith_api = NULL; } else { stonith_api->cmds->register_notification(stonith_api, T_STONITH_NOTIFY_DISCONNECT, stonith_connection_destroy_cb); } } return stonith_api; } static int32_t lrmd_ipc_accept(qb_ipcs_connection_t * c, uid_t uid, gid_t gid) { crm_trace("Connection %p", c); if (pcmk__new_client(c, uid, gid) == NULL) { return -EIO; } return 0; } static void lrmd_ipc_created(qb_ipcs_connection_t * c) { pcmk__client_t *new_client = pcmk__find_client(c); crm_trace("Connection %p", c); CRM_ASSERT(new_client != NULL); /* Now that the connection is offically established, alert * the other clients a new connection exists. */ notify_of_new_client(new_client); } static int32_t lrmd_ipc_dispatch(qb_ipcs_connection_t * c, void *data, size_t size) { uint32_t id = 0; uint32_t flags = 0; pcmk__client_t *client = pcmk__find_client(c); xmlNode *request = pcmk__client_data2xml(client, data, &id, &flags); CRM_CHECK(client != NULL, crm_err("Invalid client"); return FALSE); CRM_CHECK(client->id != NULL, crm_err("Invalid client: %p", client); return FALSE); CRM_CHECK(flags & crm_ipc_client_response, crm_err("Invalid client request: %p", client); return FALSE); if (!request) { return 0; } if (!client->name) { const char *value = crm_element_value(request, F_LRMD_CLIENTNAME); if (value == NULL) { client->name = crm_itoa(pcmk__client_pid(c)); } else { client->name = strdup(value); } } lrmd_call_id++; if (lrmd_call_id < 1) { lrmd_call_id = 1; } crm_xml_add(request, F_LRMD_CLIENTID, client->id); crm_xml_add(request, F_LRMD_CLIENTNAME, client->name); crm_xml_add_int(request, F_LRMD_CALLID, lrmd_call_id); process_lrmd_message(client, id, request); free_xml(request); return 0; } /*! * \internal * \brief Free a client connection, and exit if appropriate * * \param[in] client Client connection to free */ void lrmd_client_destroy(pcmk__client_t *client) { pcmk__free_client(client); #ifdef ENABLE_PCMK_REMOTE /* If we were waiting to shut down, we can now safely do so * if there are no more proxied IPC providers */ if (shutting_down && (ipc_proxy_get_provider() == NULL)) { lrmd_exit(NULL); } #endif } static int32_t lrmd_ipc_closed(qb_ipcs_connection_t * c) { pcmk__client_t *client = pcmk__find_client(c); if (client == NULL) { return 0; } crm_trace("Connection %p", c); client_disconnect_cleanup(client->id); #ifdef ENABLE_PCMK_REMOTE ipc_proxy_remove_provider(client); #endif lrmd_client_destroy(client); return 0; } static void lrmd_ipc_destroy(qb_ipcs_connection_t * c) { lrmd_ipc_closed(c); crm_trace("Connection %p", c); } static struct qb_ipcs_service_handlers lrmd_ipc_callbacks = { .connection_accept = lrmd_ipc_accept, .connection_created = lrmd_ipc_created, .msg_process = lrmd_ipc_dispatch, .connection_closed = lrmd_ipc_closed, .connection_destroyed = lrmd_ipc_destroy }; // \return Standard Pacemaker return code int lrmd_server_send_reply(pcmk__client_t *client, uint32_t id, xmlNode *reply) { crm_trace("Sending reply (%d) to client (%s)", id, client->id); switch (client->kind) { case PCMK__CLIENT_IPC: return pcmk__ipc_send_xml(client, id, reply, FALSE); #ifdef ENABLE_PCMK_REMOTE case PCMK__CLIENT_TLS: return lrmd_tls_send_msg(client->remote, reply, id, "reply"); #endif default: crm_err("Could not send reply: unknown client type %d", client->kind); } return ENOTCONN; } // \return Standard Pacemaker return code int lrmd_server_send_notify(pcmk__client_t *client, xmlNode *msg) { crm_trace("Sending notification to client (%s)", client->id); switch (client->kind) { case PCMK__CLIENT_IPC: if (client->ipcs == NULL) { crm_trace("Could not notify local client: disconnected"); return ENOTCONN; } return pcmk__ipc_send_xml(client, 0, msg, crm_ipc_server_event); #ifdef ENABLE_PCMK_REMOTE case PCMK__CLIENT_TLS: if (client->remote == NULL) { crm_trace("Could not notify remote client: disconnected"); return ENOTCONN; } else { return lrmd_tls_send_msg(client->remote, msg, 0, "notify"); } #endif default: crm_err("Could not notify client: unknown type %d", client->kind); } return ENOTCONN; } /*! * \internal * \brief Clean up and exit immediately * * \param[in] data Ignored * * \return Doesn't return * \note This can be used as a timer callback. */ static gboolean lrmd_exit(gpointer data) { crm_info("Terminating with %d clients", pcmk__ipc_client_count()); if (stonith_api) { stonith_api->cmds->remove_notification(stonith_api, T_STONITH_NOTIFY_DISCONNECT); stonith_api->cmds->disconnect(stonith_api); stonith_api_delete(stonith_api); } if (ipcs) { mainloop_del_ipc_server(ipcs); } #ifdef ENABLE_PCMK_REMOTE lrmd_tls_server_destroy(); ipc_proxy_cleanup(); #endif pcmk__client_cleanup(); g_hash_table_destroy(rsc_list); if (mainloop) { lrmd_drain_alerts(mainloop); } crm_exit(CRM_EX_OK); return FALSE; } /*! * \internal * \brief Request cluster shutdown if appropriate, otherwise exit immediately * * \param[in] nsig Signal that caused invocation (ignored) */ static void lrmd_shutdown(int nsig) { #ifdef ENABLE_PCMK_REMOTE pcmk__client_t *ipc_proxy = ipc_proxy_get_provider(); /* If there are active proxied IPC providers, then we may be running * resources, so notify the cluster that we wish to shut down. */ if (ipc_proxy) { if (shutting_down) { crm_notice("Waiting for cluster to stop resources before exiting"); return; } crm_info("Sending shutdown request to cluster"); if (ipc_proxy_shutdown_req(ipc_proxy) < 0) { crm_crit("Shutdown request failed, exiting immediately"); } else { /* We requested a shutdown. Now, we need to wait for an * acknowledgement from the proxy host (which ensures the proxy host * supports shutdown requests), then wait for all proxy hosts to * disconnect (which ensures that all resources have been stopped). */ shutting_down = TRUE; /* Stop accepting new proxy connections */ lrmd_tls_server_destroy(); /* Older controller versions will never acknowledge our request, so * set a fairly short timeout to exit quickly in that case. If we * get the ack, we'll defuse this timer. */ shutdown_ack_timer = g_timeout_add_seconds(20, lrmd_exit, NULL); /* Currently, we let the OS kill us if the clients don't disconnect * in a reasonable time. We could instead set a long timer here * (shorter than what the OS is likely to use) and exit immediately * if it pops. */ return; } } #endif lrmd_exit(NULL); } /*! * \internal * \brief Defuse short exit timer if shutting down */ void handle_shutdown_ack() { #ifdef ENABLE_PCMK_REMOTE if (shutting_down) { crm_info("Received shutdown ack"); if (shutdown_ack_timer > 0) { g_source_remove(shutdown_ack_timer); shutdown_ack_timer = 0; } return; } #endif crm_debug("Ignoring unexpected shutdown ack"); } /*! * \internal * \brief Make short exit timer fire immediately */ void handle_shutdown_nack() { #ifdef ENABLE_PCMK_REMOTE if (shutting_down) { crm_info("Received shutdown nack"); if (shutdown_ack_timer > 0) { g_source_remove(shutdown_ack_timer); shutdown_ack_timer = g_timeout_add(0, lrmd_exit, NULL); } return; } #endif crm_debug("Ignoring unexpected shutdown nack"); } static pcmk__cli_option_t long_options[] = { // long option, argument type, storage, short option, description, flags { "help", no_argument, NULL, '?', "\tThis text", pcmk__option_default }, { "version", no_argument, NULL, '$', "\tVersion information", pcmk__option_default }, { "verbose", no_argument, NULL, 'V', "\tIncrease debug output", pcmk__option_default }, { "logfile", required_argument, NULL, 'l', "\tSend logs to the additional named logfile", pcmk__option_default }, #ifdef ENABLE_PCMK_REMOTE { "port", required_argument, NULL, 'p', "\tPort to listen on", pcmk__option_default }, #endif { 0, 0, 0, 0 } }; #ifdef ENABLE_PCMK_REMOTE # define EXECD_TYPE "remote" # define EXECD_NAME "pacemaker-remoted" # define EXECD_DESC "resource agent executor daemon for Pacemaker Remote nodes" #else # define EXECD_TYPE "local" # define EXECD_NAME "pacemaker-execd" # define EXECD_DESC "resource agent executor daemon for Pacemaker cluster nodes" #endif int main(int argc, char **argv, char **envp) { int flag = 0; int index = 0; int bump_log_num = 0; const char *option = NULL; #ifdef ENABLE_PCMK_REMOTE // If necessary, create PID 1 now before any file descriptors are opened remoted_spawn_pidone(argc, argv, envp); #endif crm_log_preinit(EXECD_NAME, argc, argv); pcmk__set_cli_options(NULL, "[options]", long_options, EXECD_DESC); while (1) { flag = pcmk__next_cli_option(argc, argv, &index, NULL); if (flag == -1) { break; } switch (flag) { case 'l': crm_add_logfile(optarg); break; case 'p': setenv("PCMK_remote_port", optarg, 1); break; case 'V': bump_log_num++; break; case '?': case '$': pcmk__cli_help(flag, CRM_EX_OK); break; default: pcmk__cli_help('?', CRM_EX_USAGE); break; } } crm_log_init(NULL, LOG_INFO, TRUE, FALSE, argc, argv, FALSE); while (bump_log_num > 0) { crm_bump_log_level(argc, argv); bump_log_num--; } option = pcmk__env_option("logfacility"); - if (option && pcmk__str_none_of(option, "none", "/dev/null", NULL)) { + if (option && !pcmk__str_any_of(option, "none", "/dev/null", NULL)) { setenv("HA_LOGFACILITY", option, 1); /* Used by the ocf_log/ha_log OCF macro */ } option = pcmk__env_option("logfile"); if(option && safe_str_neq(option, "none")) { setenv("HA_LOGFILE", option, 1); /* Used by the ocf_log/ha_log OCF macro */ if (pcmk__env_option_enabled(crm_system_name, "debug")) { setenv("HA_DEBUGLOG", option, 1); /* Used by the ocf_log/ha_debug OCF macro */ } } crm_notice("Starting Pacemaker " EXECD_TYPE " executor"); /* The presence of this variable allegedly controls whether child * processes like httpd will try and use Systemd's sd_notify * API */ unsetenv("NOTIFY_SOCKET"); /* Used by RAs - Leave owned by root */ crm_build_path(CRM_RSCTMP_DIR, 0755); rsc_list = g_hash_table_new_full(crm_str_hash, g_str_equal, NULL, free_rsc); ipcs = mainloop_add_ipc_server(CRM_SYSTEM_LRMD, QB_IPC_SHM, &lrmd_ipc_callbacks); if (ipcs == NULL) { crm_err("Failed to create IPC server: shutting down and inhibiting respawn"); crm_exit(CRM_EX_FATAL); } #ifdef ENABLE_PCMK_REMOTE if (lrmd_init_remote_tls_server() < 0) { crm_err("Failed to create TLS listener: shutting down and staying down"); crm_exit(CRM_EX_FATAL); } ipc_proxy_init(); #endif mainloop_add_signal(SIGTERM, lrmd_shutdown); mainloop = g_main_loop_new(NULL, FALSE); crm_notice("Pacemaker " EXECD_TYPE " executor successfully started and accepting connections"); g_main_loop_run(mainloop); /* should never get here */ lrmd_exit(NULL); return CRM_EX_OK; } diff --git a/daemons/fenced/fenced_commands.c b/daemons/fenced/fenced_commands.c index 9c27d6176f..e287541be3 100644 --- a/daemons/fenced/fenced_commands.c +++ b/daemons/fenced/fenced_commands.c @@ -1,2821 +1,2821 @@ /* * Copyright 2009-2020 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include GHashTable *device_list = NULL; GHashTable *topology = NULL; GList *cmd_list = NULL; struct device_search_s { /* target of fence action */ char *host; /* requested fence action */ char *action; /* timeout to use if a device is queried dynamically for possible targets */ int per_device_timeout; /* number of registered fencing devices at time of request */ int replies_needed; /* number of device replies received so far */ int replies_received; /* whether the target is eligible to perform requested action (or off) */ bool allow_suicide; /* private data to pass to search callback function */ void *user_data; /* function to call when all replies have been received */ void (*callback) (GList * devices, void *user_data); /* devices capable of performing requested action (or off if remapping) */ GListPtr capable; }; static gboolean stonith_device_dispatch(gpointer user_data); static void st_child_done(GPid pid, int rc, const char *output, gpointer user_data); static void stonith_send_reply(xmlNode * reply, int call_options, const char *remote_peer, const char *client_id); static void search_devices_record_result(struct device_search_s *search, const char *device, gboolean can_fence); typedef struct async_command_s { int id; int pid; int fd_stdout; int options; int default_timeout; /* seconds */ int timeout; /* seconds */ int start_delay; /* seconds */ int delay_id; char *op; char *origin; char *client; char *client_name; char *remote_op_id; char *victim; uint32_t victim_nodeid; char *action; char *device; char *mode; GListPtr device_list; GListPtr device_next; void *internal_user_data; void (*done_cb) (GPid pid, int rc, const char *output, gpointer user_data); guint timer_sigterm; guint timer_sigkill; /*! If the operation timed out, this is the last signal * we sent to the process to get it to terminate */ int last_timeout_signo; stonith_device_t *active_on; stonith_device_t *activating_on; } async_command_t; static xmlNode *stonith_construct_async_reply(async_command_t * cmd, const char *output, xmlNode * data, int rc); static gboolean is_action_required(const char *action, stonith_device_t *device) { return device && device->automatic_unfencing && safe_str_eq(action, "on"); } static int get_action_delay_max(stonith_device_t * device, const char * action) { const char *value = NULL; int delay_max = 0; - if (pcmk__str_none_of(action, "off", "reboot", NULL)) { + if (!pcmk__str_any_of(action, "off", "reboot", NULL)) { return 0; } value = g_hash_table_lookup(device->params, STONITH_ATTR_DELAY_MAX); if (value) { delay_max = crm_parse_interval_spec(value) / 1000; } return delay_max; } static int get_action_delay_base(stonith_device_t * device, const char * action) { const char *value = NULL; int delay_base = 0; - if (pcmk__str_none_of(action, "off", "reboot", NULL)) { + if (!pcmk__str_any_of(action, "off", "reboot", NULL)) { return 0; } value = g_hash_table_lookup(device->params, STONITH_ATTR_DELAY_BASE); if (value) { delay_base = crm_parse_interval_spec(value) / 1000; } return delay_base; } /*! * \internal * \brief Override STONITH timeout with pcmk_*_timeout if available * * \param[in] device STONITH device to use * \param[in] action STONITH action name * \param[in] default_timeout Timeout to use if device does not have * a pcmk_*_timeout parameter for action * * \return Value of pcmk_(action)_timeout if available, otherwise default_timeout * \note For consistency, it would be nice if reboot/off/on timeouts could be * set the same way as start/stop/monitor timeouts, i.e. with an * entry in the fencing resource configuration. However that * is insufficient because fencing devices may be registered directly via * the fencer's register_device() API instead of going through the CIB * (e.g. stonith_admin uses it for its -R option, and the executor uses it * to ensure a device is registered when a command is issued). As device * properties, pcmk_*_timeout parameters can be grabbed by the fencer when * the device is registered, whether by CIB change or API call. */ static int get_action_timeout(stonith_device_t * device, const char *action, int default_timeout) { if (action && device && device->params) { char buffer[64] = { 0, }; const char *value = NULL; /* If "reboot" was requested but the device does not support it, * we will remap to "off", so check timeout for "off" instead */ if (safe_str_eq(action, "reboot") && is_not_set(device->flags, st_device_supports_reboot)) { crm_trace("%s doesn't support reboot, using timeout for off instead", device->id); action = "off"; } /* If the device config specified an action-specific timeout, use it */ snprintf(buffer, sizeof(buffer), "pcmk_%s_timeout", action); value = g_hash_table_lookup(device->params, buffer); if (value) { return atoi(value); } } return default_timeout; } static void free_async_command(async_command_t * cmd) { if (!cmd) { return; } if (cmd->delay_id) { g_source_remove(cmd->delay_id); } cmd_list = g_list_remove(cmd_list, cmd); g_list_free_full(cmd->device_list, free); free(cmd->device); free(cmd->action); free(cmd->victim); free(cmd->remote_op_id); free(cmd->client); free(cmd->client_name); free(cmd->origin); free(cmd->mode); free(cmd->op); free(cmd); } static async_command_t * create_async_command(xmlNode * msg) { async_command_t *cmd = NULL; xmlNode *op = get_xpath_object("//@" F_STONITH_ACTION, msg, LOG_ERR); const char *action = crm_element_value(op, F_STONITH_ACTION); CRM_CHECK(action != NULL, crm_log_xml_warn(msg, "NoAction"); return NULL); crm_log_xml_trace(msg, "Command"); cmd = calloc(1, sizeof(async_command_t)); crm_element_value_int(msg, F_STONITH_CALLID, &(cmd->id)); crm_element_value_int(msg, F_STONITH_CALLOPTS, &(cmd->options)); crm_element_value_int(msg, F_STONITH_TIMEOUT, &(cmd->default_timeout)); cmd->timeout = cmd->default_timeout; // Value -1 means disable any static/random fencing delays crm_element_value_int(msg, F_STONITH_DELAY, &(cmd->start_delay)); cmd->origin = crm_element_value_copy(msg, F_ORIG); cmd->remote_op_id = crm_element_value_copy(msg, F_STONITH_REMOTE_OP_ID); cmd->client = crm_element_value_copy(msg, F_STONITH_CLIENTID); cmd->client_name = crm_element_value_copy(msg, F_STONITH_CLIENTNAME); cmd->op = crm_element_value_copy(msg, F_STONITH_OPERATION); cmd->action = strdup(action); cmd->victim = crm_element_value_copy(op, F_STONITH_TARGET); cmd->mode = crm_element_value_copy(op, F_STONITH_MODE); cmd->device = crm_element_value_copy(op, F_STONITH_DEVICE); CRM_CHECK(cmd->op != NULL, crm_log_xml_warn(msg, "NoOp"); free_async_command(cmd); return NULL); CRM_CHECK(cmd->client != NULL, crm_log_xml_warn(msg, "NoClient")); cmd->done_cb = st_child_done; cmd_list = g_list_append(cmd_list, cmd); return cmd; } static int get_action_limit(stonith_device_t * device) { const char *value = NULL; int action_limit = 1; value = g_hash_table_lookup(device->params, STONITH_ATTR_ACTION_LIMIT); if (value) { action_limit = crm_parse_int(value, "1"); if (action_limit == 0) { /* pcmk_action_limit should not be 0. Enforce it to be 1. */ action_limit = 1; } } return action_limit; } static int get_active_cmds(stonith_device_t * device) { int counter = 0; GListPtr gIter = NULL; GListPtr gIterNext = NULL; CRM_CHECK(device != NULL, return 0); for (gIter = cmd_list; gIter != NULL; gIter = gIterNext) { async_command_t *cmd = gIter->data; gIterNext = gIter->next; if (cmd->active_on == device) { counter++; } } return counter; } static void fork_cb(GPid pid, gpointer user_data) { async_command_t *cmd = (async_command_t *) user_data; stonith_device_t * device = /* in case of a retry we've done the move from activating_on to active_on already */ cmd->activating_on?cmd->activating_on:cmd->active_on; CRM_ASSERT(device); crm_debug("Operation '%s'%s%s on %s now running with pid=%d, timeout=%ds", cmd->action, cmd->victim ? " targeting " : "", cmd->victim ? cmd->victim : "", device->id, pid, cmd->timeout); cmd->active_on = device; cmd->activating_on = NULL; } static gboolean stonith_device_execute(stonith_device_t * device) { int exec_rc = 0; const char *action_str = NULL; const char *host_arg = NULL; async_command_t *cmd = NULL; stonith_action_t *action = NULL; int active_cmds = 0; int action_limit = 0; GListPtr gIter = NULL; GListPtr gIterNext = NULL; CRM_CHECK(device != NULL, return FALSE); active_cmds = get_active_cmds(device); action_limit = get_action_limit(device); if (action_limit > -1 && active_cmds >= action_limit) { crm_trace("%s is over its action limit of %d (%u active action%s)", device->id, action_limit, active_cmds, active_cmds > 1 ? "s" : ""); return TRUE; } for (gIter = device->pending_ops; gIter != NULL; gIter = gIterNext) { async_command_t *pending_op = gIter->data; gIterNext = gIter->next; if (pending_op && pending_op->delay_id) { crm_trace ("Operation '%s'%s%s on %s was asked to run too early, waiting for start_delay timeout of %ds", pending_op->action, pending_op->victim ? " targeting " : "", pending_op->victim ? pending_op->victim : "", device->id, pending_op->start_delay); continue; } device->pending_ops = g_list_remove_link(device->pending_ops, gIter); g_list_free_1(gIter); cmd = pending_op; break; } if (cmd == NULL) { crm_trace("Nothing further to do for %s for now", device->id); return TRUE; } if(safe_str_eq(device->agent, STONITH_WATCHDOG_AGENT)) { if(safe_str_eq(cmd->action, "reboot")) { pcmk_panic(__FUNCTION__); goto done; } else if(safe_str_eq(cmd->action, "off")) { pcmk_panic(__FUNCTION__); goto done; } else { crm_info("Faking success for %s watchdog operation", cmd->action); cmd->done_cb(0, 0, NULL, cmd); goto done; } } #if SUPPORT_CIBSECRETS if (pcmk__substitute_secrets(device->id, device->params) != pcmk_rc_ok) { /* replacing secrets failed! */ if (safe_str_eq(cmd->action,"stop")) { /* don't fail on stop! */ crm_info("proceeding with the stop operation for %s", device->id); } else { crm_err("failed to get secrets for %s, " "considering resource not configured", device->id); exec_rc = PCMK_OCF_NOT_CONFIGURED; cmd->done_cb(0, exec_rc, NULL, cmd); goto done; } } #endif action_str = cmd->action; if (safe_str_eq(cmd->action, "reboot") && is_not_set(device->flags, st_device_supports_reboot)) { crm_warn("Agent '%s' does not advertise support for 'reboot', performing 'off' action instead", device->agent); action_str = "off"; } if (is_set(device->flags, st_device_supports_parameter_port)) { host_arg = "port"; } else if (is_set(device->flags, st_device_supports_parameter_plug)) { host_arg = "plug"; } action = stonith_action_create(device->agent, action_str, cmd->victim, cmd->victim_nodeid, cmd->timeout, device->params, device->aliases, host_arg); /* for async exec, exec_rc is negative for early error exit otherwise handling of success/errors is done via callbacks */ cmd->activating_on = device; exec_rc = stonith_action_execute_async(action, (void *)cmd, cmd->done_cb, fork_cb); if (exec_rc < 0) { crm_warn("Operation '%s'%s%s on %s failed: %s (%d)", cmd->action, cmd->victim ? " targeting " : "", cmd->victim ? cmd->victim : "", device->id, pcmk_strerror(exec_rc), exec_rc); cmd->activating_on = NULL; cmd->done_cb(0, exec_rc, NULL, cmd); } done: /* Device might get triggered to work by multiple fencing commands * simultaneously. Trigger the device again to make sure any * remaining concurrent commands get executed. */ if (device->pending_ops) { mainloop_set_trigger(device->work); } return TRUE; } static gboolean stonith_device_dispatch(gpointer user_data) { return stonith_device_execute(user_data); } static gboolean start_delay_helper(gpointer data) { async_command_t *cmd = data; stonith_device_t *device = NULL; cmd->delay_id = 0; device = cmd->device ? g_hash_table_lookup(device_list, cmd->device) : NULL; if (device) { mainloop_set_trigger(device->work); } return FALSE; } static void schedule_stonith_command(async_command_t * cmd, stonith_device_t * device) { int delay_max = 0; int delay_base = 0; int requested_delay = cmd->start_delay; CRM_CHECK(cmd != NULL, return); CRM_CHECK(device != NULL, return); if (cmd->device) { free(cmd->device); } if (device->include_nodeid && cmd->victim) { crm_node_t *node = crm_get_peer(0, cmd->victim); cmd->victim_nodeid = node->id; } cmd->device = strdup(device->id); cmd->timeout = get_action_timeout(device, cmd->action, cmd->default_timeout); if (cmd->remote_op_id) { crm_debug("Scheduling '%s' action%s%s on %s for remote peer %s with op id (%s) (timeout=%ds)", cmd->action, cmd->victim ? " targeting " : "", cmd->victim ? cmd->victim : "", device->id, cmd->origin, cmd->remote_op_id, cmd->timeout); } else { crm_debug("Scheduling '%s' action%s%s on %s for %s (timeout=%ds)", cmd->action, cmd->victim ? " targeting " : "", cmd->victim ? cmd->victim : "", device->id, cmd->client, cmd->timeout); } device->pending_ops = g_list_append(device->pending_ops, cmd); mainloop_set_trigger(device->work); // Value -1 means disable any static/random fencing delays if (requested_delay < 0) { return; } delay_max = get_action_delay_max(device, cmd->action); delay_base = get_action_delay_base(device, cmd->action); if (delay_max == 0) { delay_max = delay_base; } if (delay_max < delay_base) { crm_warn("Base-delay (%ds) is larger than max-delay (%ds) " "for %s on %s - limiting to max-delay", delay_base, delay_max, cmd->action, device->id); delay_base = delay_max; } if (delay_max > 0) { // coverity[dont_call] We're not using rand() for security cmd->start_delay += ((delay_max != delay_base)?(rand() % (delay_max - delay_base)):0) + delay_base; } if (cmd->start_delay > 0) { crm_notice("Delaying '%s' action%s%s on %s for %ds (timeout=%ds, " "requested_delay=%ds, base=%ds, max=%ds)", cmd->action, cmd->victim ? " targeting " : "", cmd->victim ? cmd->victim : "", device->id, cmd->start_delay, cmd->timeout, requested_delay, delay_base, delay_max); cmd->delay_id = g_timeout_add_seconds(cmd->start_delay, start_delay_helper, cmd); } } static void free_device(gpointer data) { GListPtr gIter = NULL; stonith_device_t *device = data; g_hash_table_destroy(device->params); g_hash_table_destroy(device->aliases); for (gIter = device->pending_ops; gIter != NULL; gIter = gIter->next) { async_command_t *cmd = gIter->data; crm_warn("Removal of device '%s' purged operation '%s'", device->id, cmd->action); cmd->done_cb(0, -ENODEV, NULL, cmd); } g_list_free(device->pending_ops); g_list_free_full(device->targets, free); mainloop_destroy_trigger(device->work); free_xml(device->agent_metadata); free(device->namespace); free(device->on_target_actions); free(device->agent); free(device->id); free(device); } void free_device_list(void) { if (device_list != NULL) { g_hash_table_destroy(device_list); device_list = NULL; } } void init_device_list(void) { if (device_list == NULL) { device_list = g_hash_table_new_full(crm_str_hash, g_str_equal, NULL, free_device); } } static GHashTable * build_port_aliases(const char *hostmap, GListPtr * targets) { char *name = NULL; int last = 0, lpc = 0, max = 0, added = 0; GHashTable *aliases = crm_strcase_table_new(); if (hostmap == NULL) { return aliases; } max = strlen(hostmap); for (; lpc <= max; lpc++) { switch (hostmap[lpc]) { /* Assignment chars */ case '=': case ':': if (lpc > last) { free(name); name = calloc(1, 1 + lpc - last); memcpy(name, hostmap + last, lpc - last); } last = lpc + 1; break; /* Delimeter chars */ /* case ',': Potentially used to specify multiple ports */ case 0: case ';': case ' ': case '\t': if (name) { char *value = NULL; value = calloc(1, 1 + lpc - last); memcpy(value, hostmap + last, lpc - last); crm_debug("Adding alias '%s'='%s'", name, value); g_hash_table_replace(aliases, name, value); if (targets) { *targets = g_list_append(*targets, strdup(value)); } value = NULL; name = NULL; added++; } else if (lpc > last) { crm_debug("Parse error at offset %d near '%s'", lpc - last, hostmap + last); } last = lpc + 1; break; } if (hostmap[lpc] == 0) { break; } } if (added == 0) { crm_info("No host mappings detected in '%s'", hostmap); } free(name); return aliases; } GHashTable *metadata_cache = NULL; void free_metadata_cache(void) { if (metadata_cache != NULL) { g_hash_table_destroy(metadata_cache); metadata_cache = NULL; } } static void init_metadata_cache(void) { if (metadata_cache == NULL) { metadata_cache = crm_str_table_new(); } } static xmlNode * get_agent_metadata(const char *agent) { xmlNode *xml = NULL; char *buffer = NULL; init_metadata_cache(); buffer = g_hash_table_lookup(metadata_cache, agent); if(safe_str_eq(agent, STONITH_WATCHDOG_AGENT)) { return NULL; } else if(buffer == NULL) { stonith_t *st = stonith_api_new(); int rc; if (st == NULL) { crm_warn("Could not get agent meta-data: " "API memory allocation failed"); return NULL; } rc = st->cmds->metadata(st, st_opt_sync_call, agent, NULL, &buffer, 10); stonith_api_delete(st); if (rc || !buffer) { crm_err("Could not retrieve metadata for fencing agent %s", agent); return NULL; } g_hash_table_replace(metadata_cache, strdup(agent), buffer); } xml = string2xml(buffer); return xml; } static gboolean is_nodeid_required(xmlNode * xml) { xmlXPathObjectPtr xpath = NULL; if (stand_alone) { return FALSE; } if (!xml) { return FALSE; } xpath = xpath_search(xml, "//parameter[@name='nodeid']"); if (numXpathResults(xpath) <= 0) { freeXpathObject(xpath); return FALSE; } freeXpathObject(xpath); return TRUE; } #define MAX_ACTION_LEN 256 static char * add_action(char *actions, const char *action) { int offset = 0; if (actions == NULL) { actions = calloc(1, MAX_ACTION_LEN); } else { offset = strlen(actions); } if (offset > 0) { offset += snprintf(actions+offset, MAX_ACTION_LEN - offset, " "); } offset += snprintf(actions+offset, MAX_ACTION_LEN - offset, "%s", action); return actions; } static void read_action_metadata(stonith_device_t *device) { xmlXPathObjectPtr xpath = NULL; int max = 0; int lpc = 0; if (device->agent_metadata == NULL) { return; } xpath = xpath_search(device->agent_metadata, "//action"); max = numXpathResults(xpath); if (max <= 0) { freeXpathObject(xpath); return; } for (lpc = 0; lpc < max; lpc++) { const char *on_target = NULL; const char *action = NULL; xmlNode *match = getXpathResult(xpath, lpc); CRM_LOG_ASSERT(match != NULL); if(match == NULL) { continue; }; on_target = crm_element_value(match, "on_target"); action = crm_element_value(match, "name"); if(safe_str_eq(action, "list")) { set_bit(device->flags, st_device_supports_list); } else if(safe_str_eq(action, "status")) { set_bit(device->flags, st_device_supports_status); } else if(safe_str_eq(action, "reboot")) { set_bit(device->flags, st_device_supports_reboot); } else if (safe_str_eq(action, "on")) { /* "automatic" means the cluster will unfence node when it joins */ const char *automatic = crm_element_value(match, "automatic"); /* "required" is a deprecated synonym for "automatic" */ const char *required = crm_element_value(match, "required"); if (crm_is_true(automatic) || crm_is_true(required)) { device->automatic_unfencing = TRUE; } } if (action && crm_is_true(on_target)) { device->on_target_actions = add_action(device->on_target_actions, action); } } freeXpathObject(xpath); } /*! * \internal * \brief Set a pcmk_*_action parameter if not already set * * \param[in,out] params Device parameters * \param[in] action Name of action * \param[in] value Value to use if action is not already set */ static void map_action(GHashTable *params, const char *action, const char *value) { char *key = crm_strdup_printf("pcmk_%s_action", action); if (g_hash_table_lookup(params, key)) { crm_warn("Ignoring %s='%s', see %s instead", STONITH_ATTR_ACTION_OP, value, key); free(key); } else { crm_warn("Mapping %s='%s' to %s='%s'", STONITH_ATTR_ACTION_OP, value, key, value); g_hash_table_insert(params, key, strdup(value)); } } /*! * \internal * \brief Create device parameter table from XML * * \param[in] name Device name (used for logging only) * \param[in,out] params Device parameters */ static GHashTable * xml2device_params(const char *name, xmlNode *dev) { GHashTable *params = xml2list(dev); const char *value; /* Action should never be specified in the device configuration, * but we support it for users who are familiar with other software * that worked that way. */ value = g_hash_table_lookup(params, STONITH_ATTR_ACTION_OP); if (value != NULL) { crm_warn("%s has '%s' parameter, which should never be specified in configuration", name, STONITH_ATTR_ACTION_OP); if (*value == '\0') { crm_warn("Ignoring empty '%s' parameter", STONITH_ATTR_ACTION_OP); } else if (strcmp(value, "reboot") == 0) { crm_warn("Ignoring %s='reboot' (see stonith-action cluster property instead)", STONITH_ATTR_ACTION_OP); } else if (strcmp(value, "off") == 0) { map_action(params, "reboot", value); } else { map_action(params, "off", value); map_action(params, "reboot", value); } g_hash_table_remove(params, STONITH_ATTR_ACTION_OP); } return params; } static stonith_device_t * build_device_from_xml(xmlNode * msg) { const char *value; xmlNode *dev = get_xpath_object("//" F_STONITH_DEVICE, msg, LOG_ERR); stonith_device_t *device = NULL; char *agent = crm_element_value_copy(dev, "agent"); CRM_CHECK(agent != NULL, return device); device = calloc(1, sizeof(stonith_device_t)); CRM_CHECK(device != NULL, {free(agent); return device;}); device->id = crm_element_value_copy(dev, XML_ATTR_ID); device->agent = agent; device->namespace = crm_element_value_copy(dev, "namespace"); device->params = xml2device_params(device->id, dev); value = g_hash_table_lookup(device->params, STONITH_ATTR_HOSTLIST); if (value) { device->targets = stonith__parse_targets(value); } value = g_hash_table_lookup(device->params, STONITH_ATTR_HOSTMAP); device->aliases = build_port_aliases(value, &(device->targets)); device->agent_metadata = get_agent_metadata(device->agent); if (device->agent_metadata) { read_action_metadata(device); set_bit(device->flags, stonith__device_parameter_flags(device->agent_metadata)); } value = g_hash_table_lookup(device->params, "nodeid"); if (!value) { device->include_nodeid = is_nodeid_required(device->agent_metadata); } value = crm_element_value(dev, "rsc_provides"); if (safe_str_eq(value, "unfencing")) { device->automatic_unfencing = TRUE; } if (is_action_required("on", device)) { crm_info("The fencing device '%s' requires unfencing", device->id); } if (device->on_target_actions) { crm_info("The fencing device '%s' requires actions (%s) to be executed on the target node", device->id, device->on_target_actions); } device->work = mainloop_add_trigger(G_PRIORITY_HIGH, stonith_device_dispatch, device); /* TODO: Hook up priority */ return device; } static const char * target_list_type(stonith_device_t * dev) { const char *check_type = NULL; check_type = g_hash_table_lookup(dev->params, STONITH_ATTR_HOSTCHECK); if (check_type == NULL) { if (g_hash_table_lookup(dev->params, STONITH_ATTR_HOSTLIST)) { check_type = "static-list"; } else if (g_hash_table_lookup(dev->params, STONITH_ATTR_HOSTMAP)) { check_type = "static-list"; } else if(is_set(dev->flags, st_device_supports_list)){ check_type = "dynamic-list"; } else if(is_set(dev->flags, st_device_supports_status)){ check_type = "status"; } else { check_type = "none"; } } return check_type; } static void schedule_internal_command(const char *origin, stonith_device_t * device, const char *action, const char *victim, int timeout, void *internal_user_data, void (*done_cb) (GPid pid, int rc, const char *output, gpointer user_data)) { async_command_t *cmd = NULL; cmd = calloc(1, sizeof(async_command_t)); cmd->id = -1; cmd->default_timeout = timeout ? timeout : 60; cmd->timeout = cmd->default_timeout; cmd->action = strdup(action); cmd->victim = victim ? strdup(victim) : NULL; cmd->device = strdup(device->id); cmd->origin = strdup(origin); cmd->client = strdup(crm_system_name); cmd->client_name = strdup(crm_system_name); cmd->internal_user_data = internal_user_data; cmd->done_cb = done_cb; /* cmd, not internal_user_data, is passed to 'done_cb' as the userdata */ schedule_stonith_command(cmd, device); } gboolean string_in_list(GListPtr list, const char *item) { int lpc = 0; int max = g_list_length(list); for (lpc = 0; lpc < max; lpc++) { const char *value = g_list_nth_data(list, lpc); if (safe_str_eq(item, value)) { return TRUE; } else { crm_trace("%d: '%s' != '%s'", lpc, item, value); } } return FALSE; } static void status_search_cb(GPid pid, int rc, const char *output, gpointer user_data) { async_command_t *cmd = user_data; struct device_search_s *search = cmd->internal_user_data; stonith_device_t *dev = cmd->device ? g_hash_table_lookup(device_list, cmd->device) : NULL; gboolean can = FALSE; free_async_command(cmd); if (!dev) { search_devices_record_result(search, NULL, FALSE); return; } mainloop_set_trigger(dev->work); if (rc == 1 /* unknown */ ) { crm_trace("Host %s is not known by %s", search->host, dev->id); } else if (rc == 0 /* active */ || rc == 2 /* inactive */ ) { crm_trace("Host %s is known by %s", search->host, dev->id); can = TRUE; } else { crm_notice("Unknown result when testing if %s can fence %s: rc=%d", dev->id, search->host, rc); } search_devices_record_result(search, dev->id, can); } static void dynamic_list_search_cb(GPid pid, int rc, const char *output, gpointer user_data) { async_command_t *cmd = user_data; struct device_search_s *search = cmd->internal_user_data; stonith_device_t *dev = cmd->device ? g_hash_table_lookup(device_list, cmd->device) : NULL; gboolean can_fence = FALSE; free_async_command(cmd); /* Host/alias must be in the list output to be eligible to be fenced * * Will cause problems if down'd nodes aren't listed or (for virtual nodes) * if the guest is still listed despite being moved to another machine */ if (!dev) { search_devices_record_result(search, NULL, FALSE); return; } mainloop_set_trigger(dev->work); /* If we successfully got the targets earlier, don't disable. */ if (rc != 0 && !dev->targets) { crm_notice("Disabling port list queries for %s (%d): %s", dev->id, rc, output); /* Fall back to status */ g_hash_table_replace(dev->params, strdup(STONITH_ATTR_HOSTCHECK), strdup("status")); g_list_free_full(dev->targets, free); dev->targets = NULL; } else if (!rc) { crm_info("Refreshing port list for %s", dev->id); g_list_free_full(dev->targets, free); dev->targets = stonith__parse_targets(output); dev->targets_age = time(NULL); } if (dev->targets) { const char *alias = g_hash_table_lookup(dev->aliases, search->host); if (!alias) { alias = search->host; } if (string_in_list(dev->targets, alias)) { can_fence = TRUE; } } search_devices_record_result(search, dev->id, can_fence); } /*! * \internal * \brief Returns true if any key in first is not in second or second has a different value for key */ static int device_params_diff(GHashTable *first, GHashTable *second) { char *key = NULL; char *value = NULL; GHashTableIter gIter; g_hash_table_iter_init(&gIter, first); while (g_hash_table_iter_next(&gIter, (void **)&key, (void **)&value)) { if(strstr(key, "CRM_meta") == key) { continue; } else if(strcmp(key, "crm_feature_set") == 0) { continue; } else { char *other_value = g_hash_table_lookup(second, key); if (!other_value || safe_str_neq(other_value, value)) { crm_trace("Different value for %s: %s != %s", key, other_value, value); return 1; } } } return 0; } /*! * \internal * \brief Checks to see if an identical device already exists in the device_list */ static stonith_device_t * device_has_duplicate(stonith_device_t * device) { stonith_device_t *dup = g_hash_table_lookup(device_list, device->id); if (!dup) { crm_trace("No match for %s", device->id); return NULL; } else if (safe_str_neq(dup->agent, device->agent)) { crm_trace("Different agent: %s != %s", dup->agent, device->agent); return NULL; } /* Use calculate_operation_digest() here? */ if (device_params_diff(device->params, dup->params) || device_params_diff(dup->params, device->params)) { return NULL; } crm_trace("Match"); return dup; } int stonith_device_register(xmlNode * msg, const char **desc, gboolean from_cib) { stonith_device_t *dup = NULL; stonith_device_t *device = build_device_from_xml(msg); CRM_CHECK(device != NULL, return -ENOMEM); dup = device_has_duplicate(device); if (dup) { crm_debug("Device '%s' already existed in device list (%d active devices)", device->id, g_hash_table_size(device_list)); free_device(device); device = dup; } else { stonith_device_t *old = g_hash_table_lookup(device_list, device->id); if (from_cib && old && old->api_registered) { /* If the cib is writing over an entry that is shared with a stonith client, * copy any pending ops that currently exist on the old entry to the new one. * Otherwise the pending ops will be reported as failures */ crm_info("Overwriting an existing entry for %s from the cib", device->id); device->pending_ops = old->pending_ops; device->api_registered = TRUE; old->pending_ops = NULL; if (device->pending_ops) { mainloop_set_trigger(device->work); } } g_hash_table_replace(device_list, device->id, device); crm_notice("Added '%s' to the device list (%d active devices)", device->id, g_hash_table_size(device_list)); } if (desc) { *desc = device->id; } if (from_cib) { device->cib_registered = TRUE; } else { device->api_registered = TRUE; } return pcmk_ok; } int stonith_device_remove(const char *id, gboolean from_cib) { stonith_device_t *device = g_hash_table_lookup(device_list, id); if (!device) { crm_info("Device '%s' not found (%d active devices)", id, g_hash_table_size(device_list)); return pcmk_ok; } if (from_cib) { device->cib_registered = FALSE; } else { device->verified = FALSE; device->api_registered = FALSE; } if (!device->cib_registered && !device->api_registered) { g_hash_table_remove(device_list, id); crm_info("Removed '%s' from the device list (%d active devices)", id, g_hash_table_size(device_list)); } else { crm_trace("Not removing '%s' from the device list (%d active devices) " "- still %s%s_registered", id, g_hash_table_size(device_list), device->cib_registered?"cib":"", device->api_registered?"api":""); } return pcmk_ok; } /*! * \internal * \brief Return the number of stonith levels registered for a node * * \param[in] tp Node's topology table entry * * \return Number of non-NULL levels in topology entry * \note This function is used only for log messages. */ static int count_active_levels(stonith_topology_t * tp) { int lpc = 0; int count = 0; for (lpc = 0; lpc < ST_LEVEL_MAX; lpc++) { if (tp->levels[lpc] != NULL) { count++; } } return count; } static void free_topology_entry(gpointer data) { stonith_topology_t *tp = data; int lpc = 0; for (lpc = 0; lpc < ST_LEVEL_MAX; lpc++) { if (tp->levels[lpc] != NULL) { g_list_free_full(tp->levels[lpc], free); } } free(tp->target); free(tp->target_value); free(tp->target_pattern); free(tp->target_attribute); free(tp); } void free_topology_list(void) { if (topology != NULL) { g_hash_table_destroy(topology); topology = NULL; } } void init_topology_list(void) { if (topology == NULL) { topology = g_hash_table_new_full(crm_str_hash, g_str_equal, NULL, free_topology_entry); } } char *stonith_level_key(xmlNode *level, int mode) { if(mode == -1) { mode = stonith_level_kind(level); } switch(mode) { case 0: return crm_element_value_copy(level, XML_ATTR_STONITH_TARGET); case 1: return crm_element_value_copy(level, XML_ATTR_STONITH_TARGET_PATTERN); case 2: { const char *name = crm_element_value(level, XML_ATTR_STONITH_TARGET_ATTRIBUTE); const char *value = crm_element_value(level, XML_ATTR_STONITH_TARGET_VALUE); if(name && value) { return crm_strdup_printf("%s=%s", name, value); } } default: return crm_strdup_printf("Unknown-%d-%s", mode, ID(level)); } } int stonith_level_kind(xmlNode * level) { int mode = 0; const char *target = crm_element_value(level, XML_ATTR_STONITH_TARGET); if(target == NULL) { mode++; target = crm_element_value(level, XML_ATTR_STONITH_TARGET_PATTERN); } if(stand_alone == FALSE && target == NULL) { mode++; if(crm_element_value(level, XML_ATTR_STONITH_TARGET_ATTRIBUTE) == NULL) { mode++; } else if(crm_element_value(level, XML_ATTR_STONITH_TARGET_VALUE) == NULL) { mode++; } } return mode; } static stonith_key_value_t * parse_device_list(const char *devices) { int lpc = 0; int max = 0; int last = 0; stonith_key_value_t *output = NULL; if (devices == NULL) { return output; } max = strlen(devices); for (lpc = 0; lpc <= max; lpc++) { if (devices[lpc] == ',' || devices[lpc] == 0) { char *line = strndup(devices + last, lpc - last); output = stonith_key_value_add(output, NULL, line); free(line); last = lpc + 1; } } return output; } /*! * \internal * \brief Register a STONITH level for a target * * Given an XML request specifying the target name, level index, and device IDs * for the level, this will create an entry for the target in the global topology * table if one does not already exist, then append the specified device IDs to * the entry's device list for the specified level. * * \param[in] msg XML request for STONITH level registration * \param[out] desc If not NULL, will be set to string representation ("TARGET[LEVEL]") * * \return pcmk_ok on success, -EINVAL if XML does not specify valid level index */ int stonith_level_register(xmlNode *msg, char **desc) { int id = 0; xmlNode *level; int mode; char *target; stonith_topology_t *tp; stonith_key_value_t *dIter = NULL; stonith_key_value_t *devices = NULL; /* Allow the XML here to point to the level tag directly, or wrapped in * another tag. If directly, don't search by xpath, because it might give * multiple hits (e.g. if the XML is the CIB). */ if (safe_str_eq(TYPE(msg), XML_TAG_FENCING_LEVEL)) { level = msg; } else { level = get_xpath_object("//" XML_TAG_FENCING_LEVEL, msg, LOG_ERR); } CRM_CHECK(level != NULL, return -EINVAL); mode = stonith_level_kind(level); target = stonith_level_key(level, mode); crm_element_value_int(level, XML_ATTR_STONITH_INDEX, &id); if (desc) { *desc = crm_strdup_printf("%s[%d]", target, id); } /* Sanity-check arguments */ if (mode >= 3 || (id <= 0) || (id >= ST_LEVEL_MAX)) { crm_trace("Could not add %s[%d] (%d) to the topology (%d active entries)", target, id, mode, g_hash_table_size(topology)); free(target); crm_log_xml_err(level, "Bad topology"); return -EINVAL; } /* Find or create topology table entry */ tp = g_hash_table_lookup(topology, target); if (tp == NULL) { tp = calloc(1, sizeof(stonith_topology_t)); tp->kind = mode; tp->target = target; tp->target_value = crm_element_value_copy(level, XML_ATTR_STONITH_TARGET_VALUE); tp->target_pattern = crm_element_value_copy(level, XML_ATTR_STONITH_TARGET_PATTERN); tp->target_attribute = crm_element_value_copy(level, XML_ATTR_STONITH_TARGET_ATTRIBUTE); g_hash_table_replace(topology, tp->target, tp); crm_trace("Added %s (%d) to the topology (%d active entries)", target, mode, g_hash_table_size(topology)); } else { free(target); } if (tp->levels[id] != NULL) { crm_info("Adding to the existing %s[%d] topology entry", tp->target, id); } devices = parse_device_list(crm_element_value(level, XML_ATTR_STONITH_DEVICES)); for (dIter = devices; dIter; dIter = dIter->next) { const char *device = dIter->value; crm_trace("Adding device '%s' for %s[%d]", device, tp->target, id); tp->levels[id] = g_list_append(tp->levels[id], strdup(device)); } stonith_key_value_freeall(devices, 1, 1); crm_info("Target %s has %d active fencing levels", tp->target, count_active_levels(tp)); return pcmk_ok; } int stonith_level_remove(xmlNode *msg, char **desc) { int id = 0; stonith_topology_t *tp; char *target; /* Unlike additions, removal requests should always have one level tag */ xmlNode *level = get_xpath_object("//" XML_TAG_FENCING_LEVEL, msg, LOG_ERR); CRM_CHECK(level != NULL, return -EINVAL); target = stonith_level_key(level, -1); crm_element_value_int(level, XML_ATTR_STONITH_INDEX, &id); if (desc) { *desc = crm_strdup_printf("%s[%d]", target, id); } /* Sanity-check arguments */ if (id >= ST_LEVEL_MAX) { free(target); return -EINVAL; } tp = g_hash_table_lookup(topology, target); if (tp == NULL) { crm_info("Topology for %s not found (%d active entries)", target, g_hash_table_size(topology)); } else if (id == 0 && g_hash_table_remove(topology, target)) { crm_info("Removed all %s related entries from the topology (%d active entries)", target, g_hash_table_size(topology)); } else if (id > 0 && tp->levels[id] != NULL) { g_list_free_full(tp->levels[id], free); tp->levels[id] = NULL; crm_info("Removed level '%d' from topology for %s (%d active levels remaining)", id, target, count_active_levels(tp)); } free(target); return pcmk_ok; } /*! * \internal * \brief Schedule an (asynchronous) action directly on a stonith device * * Handle a STONITH_OP_EXEC API message by scheduling a requested agent action * directly on a specified device. Only list, monitor, and status actions are * expected to use this call, though it should work with any agent command. * * \param[in] msg API message XML with desired action * \param[out] output Unused * * \return -EINPROGRESS on success, -errno otherwise * \note If the action is monitor, the device must be registered via the API * (CIB registration is not sufficient), because monitor should not be * possible unless the device is "started" (API registered). */ static int stonith_device_action(xmlNode * msg, char **output) { xmlNode *dev = get_xpath_object("//" F_STONITH_DEVICE, msg, LOG_ERR); xmlNode *op = get_xpath_object("//@" F_STONITH_ACTION, msg, LOG_ERR); const char *id = crm_element_value(dev, F_STONITH_DEVICE); const char *action = crm_element_value(op, F_STONITH_ACTION); async_command_t *cmd = NULL; stonith_device_t *device = NULL; if ((id == NULL) || (action == NULL)) { crm_info("Malformed API action request: device %s, action %s", (id? id : "not specified"), (action? action : "not specified")); return -EPROTO; } device = g_hash_table_lookup(device_list, id); if ((device == NULL) || (!device->api_registered && !strcmp(action, "monitor"))) { // Monitors may run only on "started" (API-registered) devices crm_info("Ignoring API '%s' action request because device %s not found", action, id); return -ENODEV; } cmd = create_async_command(msg); if (cmd == NULL) { return -EPROTO; } schedule_stonith_command(cmd, device); return -EINPROGRESS; } static void search_devices_record_result(struct device_search_s *search, const char *device, gboolean can_fence) { search->replies_received++; if (can_fence && device) { search->capable = g_list_append(search->capable, strdup(device)); } if (search->replies_needed == search->replies_received) { crm_debug("Finished Search. %d devices can perform action (%s) on node %s", g_list_length(search->capable), search->action ? search->action : "", search->host ? search->host : ""); search->callback(search->capable, search->user_data); free(search->host); free(search->action); free(search); } } /*! * \internal * \brief Check whether the local host is allowed to execute a fencing action * * \param[in] device Fence device to check * \param[in] action Fence action to check * \param[in] target Hostname of fence target * \param[in] allow_suicide Whether self-fencing is allowed for this operation * * \return TRUE if local host is allowed to execute action, FALSE otherwise */ static gboolean localhost_is_eligible(const stonith_device_t *device, const char *action, const char *target, gboolean allow_suicide) { gboolean localhost_is_target = safe_str_eq(target, stonith_our_uname); if (device && action && device->on_target_actions && strstr(device->on_target_actions, action)) { if (!localhost_is_target) { crm_trace("'%s' operation with %s can only be executed for localhost not %s", action, device->id, target); return FALSE; } } else if (localhost_is_target && !allow_suicide) { crm_trace("'%s' operation does not support self-fencing", action); return FALSE; } return TRUE; } static void can_fence_host_with_device(stonith_device_t * dev, struct device_search_s *search) { gboolean can = FALSE; const char *check_type = NULL; const char *host = search->host; const char *alias = NULL; CRM_LOG_ASSERT(dev != NULL); if (dev == NULL) { goto search_report_results; } else if (host == NULL) { can = TRUE; goto search_report_results; } /* Short-circuit query if this host is not allowed to perform the action */ if (safe_str_eq(search->action, "reboot")) { /* A "reboot" *might* get remapped to "off" then "on", so short-circuit * only if all three are disallowed. If only one or two are disallowed, * we'll report that with the results. We never allow suicide for * remapped "on" operations because the host is off at that point. */ if (!localhost_is_eligible(dev, "reboot", host, search->allow_suicide) && !localhost_is_eligible(dev, "off", host, search->allow_suicide) && !localhost_is_eligible(dev, "on", host, FALSE)) { goto search_report_results; } } else if (!localhost_is_eligible(dev, search->action, host, search->allow_suicide)) { goto search_report_results; } alias = g_hash_table_lookup(dev->aliases, host); if (alias == NULL) { alias = host; } check_type = target_list_type(dev); if (safe_str_eq(check_type, "none")) { can = TRUE; } else if (safe_str_eq(check_type, "static-list")) { /* Presence in the hostmap is sufficient * Only use if all hosts on which the device can be active can always fence all listed hosts */ if (string_in_list(dev->targets, host)) { can = TRUE; } else if (g_hash_table_lookup(dev->params, STONITH_ATTR_HOSTMAP) && g_hash_table_lookup(dev->aliases, host)) { can = TRUE; } } else if (safe_str_eq(check_type, "dynamic-list")) { time_t now = time(NULL); if (dev->targets == NULL || dev->targets_age + 60 < now) { crm_trace("Running '%s' to check whether %s is eligible to fence %s (%s)", check_type, dev->id, search->host, search->action); schedule_internal_command(__FUNCTION__, dev, "list", NULL, search->per_device_timeout, search, dynamic_list_search_cb); /* we'll respond to this search request async in the cb */ return; } if (string_in_list(dev->targets, alias)) { can = TRUE; } } else if (safe_str_eq(check_type, "status")) { crm_trace("Running '%s' to check whether %s is eligible to fence %s (%s)", check_type, dev->id, search->host, search->action); schedule_internal_command(__FUNCTION__, dev, "status", search->host, search->per_device_timeout, search, status_search_cb); /* we'll respond to this search request async in the cb */ return; } else { crm_err("Invalid value for " STONITH_ATTR_HOSTCHECK ": %s", check_type); check_type = "Invalid " STONITH_ATTR_HOSTCHECK; } if (safe_str_eq(host, alias)) { crm_notice("%s is%s eligible to fence (%s) %s: %s", dev->id, (can? "" : " not"), search->action, host, check_type); } else { crm_notice("%s is%s eligible to fence (%s) %s (aka. '%s'): %s", dev->id, (can? "" : " not"), search->action, host, alias, check_type); } search_report_results: search_devices_record_result(search, dev ? dev->id : NULL, can); } static void search_devices(gpointer key, gpointer value, gpointer user_data) { stonith_device_t *dev = value; struct device_search_s *search = user_data; can_fence_host_with_device(dev, search); } #define DEFAULT_QUERY_TIMEOUT 20 static void get_capable_devices(const char *host, const char *action, int timeout, bool suicide, void *user_data, void (*callback) (GList * devices, void *user_data)) { struct device_search_s *search; int per_device_timeout = DEFAULT_QUERY_TIMEOUT; int devices_needing_async_query = 0; char *key = NULL; const char *check_type = NULL; GHashTableIter gIter; stonith_device_t *device = NULL; if (!g_hash_table_size(device_list)) { callback(NULL, user_data); return; } search = calloc(1, sizeof(struct device_search_s)); if (!search) { callback(NULL, user_data); return; } g_hash_table_iter_init(&gIter, device_list); while (g_hash_table_iter_next(&gIter, (void **)&key, (void **)&device)) { check_type = target_list_type(device); if (pcmk__str_any_of(check_type, "status", "dynamic-list", NULL)) { devices_needing_async_query++; } } /* If we have devices that require an async event in order to know what * nodes they can fence, we have to give the events a timeout. The total * query timeout is divided among those events. */ if (devices_needing_async_query) { per_device_timeout = timeout / devices_needing_async_query; if (!per_device_timeout) { crm_err("STONITH timeout %ds is too low; using %ds, but consider raising to at least %ds", timeout, DEFAULT_QUERY_TIMEOUT, DEFAULT_QUERY_TIMEOUT * devices_needing_async_query); per_device_timeout = DEFAULT_QUERY_TIMEOUT; } else if (per_device_timeout < DEFAULT_QUERY_TIMEOUT) { crm_notice("STONITH timeout %ds is low for the current configuration;" " consider raising to at least %ds", timeout, DEFAULT_QUERY_TIMEOUT * devices_needing_async_query); } } search->host = host ? strdup(host) : NULL; search->action = action ? strdup(action) : NULL; search->per_device_timeout = per_device_timeout; /* We are guaranteed this many replies. Even if a device gets * unregistered some how during the async search, we will get * the correct number of replies. */ search->replies_needed = g_hash_table_size(device_list); search->allow_suicide = suicide; search->callback = callback; search->user_data = user_data; /* kick off the search */ crm_debug("Searching through %d devices to see what is capable of action (%s) for target %s", search->replies_needed, search->action ? search->action : "", search->host ? search->host : ""); g_hash_table_foreach(device_list, search_devices, search); } struct st_query_data { xmlNode *reply; char *remote_peer; char *client_id; char *target; char *action; int call_options; }; /*! * \internal * \brief Add action-specific attributes to query reply XML * * \param[in,out] xml XML to add attributes to * \param[in] action Fence action * \param[in] device Fence device */ static void add_action_specific_attributes(xmlNode *xml, const char *action, stonith_device_t *device) { int action_specific_timeout; int delay_max; int delay_base; CRM_CHECK(xml && action && device, return); if (is_action_required(action, device)) { crm_trace("Action '%s' is required on %s", action, device->id); crm_xml_add_int(xml, F_STONITH_DEVICE_REQUIRED, 1); } action_specific_timeout = get_action_timeout(device, action, 0); if (action_specific_timeout) { crm_trace("Action '%s' has timeout %dms on %s", action, action_specific_timeout, device->id); crm_xml_add_int(xml, F_STONITH_ACTION_TIMEOUT, action_specific_timeout); } delay_max = get_action_delay_max(device, action); if (delay_max > 0) { crm_trace("Action '%s' has maximum random delay %dms on %s", action, delay_max, device->id); crm_xml_add_int(xml, F_STONITH_DELAY_MAX, delay_max / 1000); } delay_base = get_action_delay_base(device, action); if (delay_base > 0) { crm_xml_add_int(xml, F_STONITH_DELAY_BASE, delay_base / 1000); } if ((delay_max > 0) && (delay_base == 0)) { crm_trace("Action '%s' has maximum random delay %dms on %s", action, delay_max, device->id); } else if ((delay_max == 0) && (delay_base > 0)) { crm_trace("Action '%s' has a static delay of %dms on %s", action, delay_base, device->id); } else if ((delay_max > 0) && (delay_base > 0)) { crm_trace("Action '%s' has a minimum delay of %dms and a randomly chosen " "maximum delay of %dms on %s", action, delay_base, delay_max, device->id); } } /*! * \internal * \brief Add "disallowed" attribute to query reply XML if appropriate * * \param[in,out] xml XML to add attribute to * \param[in] action Fence action * \param[in] device Fence device * \param[in] target Fence target * \param[in] allow_suicide Whether self-fencing is allowed */ static void add_disallowed(xmlNode *xml, const char *action, stonith_device_t *device, const char *target, gboolean allow_suicide) { if (!localhost_is_eligible(device, action, target, allow_suicide)) { crm_trace("Action '%s' on %s is disallowed for local host", action, device->id); crm_xml_add(xml, F_STONITH_ACTION_DISALLOWED, XML_BOOLEAN_TRUE); } } /*! * \internal * \brief Add child element with action-specific values to query reply XML * * \param[in,out] xml XML to add attribute to * \param[in] action Fence action * \param[in] device Fence device * \param[in] target Fence target * \param[in] allow_suicide Whether self-fencing is allowed */ static void add_action_reply(xmlNode *xml, const char *action, stonith_device_t *device, const char *target, gboolean allow_suicide) { xmlNode *child = create_xml_node(xml, F_STONITH_ACTION); crm_xml_add(child, XML_ATTR_ID, action); add_action_specific_attributes(child, action, device); add_disallowed(child, action, device, target, allow_suicide); } static void stonith_query_capable_device_cb(GList * devices, void *user_data) { struct st_query_data *query = user_data; int available_devices = 0; xmlNode *dev = NULL; xmlNode *list = NULL; GListPtr lpc = NULL; /* Pack the results into XML */ list = create_xml_node(NULL, __FUNCTION__); crm_xml_add(list, F_STONITH_TARGET, query->target); for (lpc = devices; lpc != NULL; lpc = lpc->next) { stonith_device_t *device = g_hash_table_lookup(device_list, lpc->data); const char *action = query->action; if (!device) { /* It is possible the device got unregistered while * determining who can fence the target */ continue; } available_devices++; dev = create_xml_node(list, F_STONITH_DEVICE); crm_xml_add(dev, XML_ATTR_ID, device->id); crm_xml_add(dev, "namespace", device->namespace); crm_xml_add(dev, "agent", device->agent); crm_xml_add_int(dev, F_STONITH_DEVICE_VERIFIED, device->verified); /* If the originating fencer wants to reboot the node, and we have a * capable device that doesn't support "reboot", remap to "off" instead. */ if (is_not_set(device->flags, st_device_supports_reboot) && safe_str_eq(query->action, "reboot")) { crm_trace("%s doesn't support reboot, using values for off instead", device->id); action = "off"; } /* Add action-specific values if available */ add_action_specific_attributes(dev, action, device); if (safe_str_eq(query->action, "reboot")) { /* A "reboot" *might* get remapped to "off" then "on", so after * sending the "reboot"-specific values in the main element, we add * sub-elements for "off" and "on" values. * * We short-circuited earlier if "reboot", "off" and "on" are all * disallowed for the local host. However if only one or two are * disallowed, we send back the results and mark which ones are * disallowed. If "reboot" is disallowed, this might cause problems * with older fencer versions, which won't check for it. Older * versions will ignore "off" and "on", so they are not a problem. */ add_disallowed(dev, action, device, query->target, is_set(query->call_options, st_opt_allow_suicide)); add_action_reply(dev, "off", device, query->target, is_set(query->call_options, st_opt_allow_suicide)); add_action_reply(dev, "on", device, query->target, FALSE); } /* A query without a target wants device parameters */ if (query->target == NULL) { xmlNode *attrs = create_xml_node(dev, XML_TAG_ATTRS); g_hash_table_foreach(device->params, hash2field, attrs); } } crm_xml_add_int(list, F_STONITH_AVAILABLE_DEVICES, available_devices); if (query->target) { crm_debug("Found %d matching devices for '%s'", available_devices, query->target); } else { crm_debug("%d devices installed", available_devices); } if (list != NULL) { crm_log_xml_trace(list, "Add query results"); add_message_xml(query->reply, F_STONITH_CALLDATA, list); } stonith_send_reply(query->reply, query->call_options, query->remote_peer, query->client_id); free_xml(query->reply); free(query->remote_peer); free(query->client_id); free(query->target); free(query->action); free(query); free_xml(list); g_list_free_full(devices, free); } static void stonith_query(xmlNode * msg, const char *remote_peer, const char *client_id, int call_options) { struct st_query_data *query = NULL; const char *action = NULL; const char *target = NULL; int timeout = 0; xmlNode *dev = get_xpath_object("//@" F_STONITH_ACTION, msg, LOG_NEVER); crm_element_value_int(msg, F_STONITH_TIMEOUT, &timeout); if (dev) { const char *device = crm_element_value(dev, F_STONITH_DEVICE); target = crm_element_value(dev, F_STONITH_TARGET); action = crm_element_value(dev, F_STONITH_ACTION); if (device && safe_str_eq(device, "manual_ack")) { /* No query or reply necessary */ return; } } crm_log_xml_debug(msg, "Query"); query = calloc(1, sizeof(struct st_query_data)); query->reply = stonith_construct_reply(msg, NULL, NULL, pcmk_ok); query->remote_peer = remote_peer ? strdup(remote_peer) : NULL; query->client_id = client_id ? strdup(client_id) : NULL; query->target = target ? strdup(target) : NULL; query->action = action ? strdup(action) : NULL; query->call_options = call_options; get_capable_devices(target, action, timeout, is_set(call_options, st_opt_allow_suicide), query, stonith_query_capable_device_cb); } #define ST_LOG_OUTPUT_MAX 512 static void log_operation(async_command_t * cmd, int rc, int pid, const char *next, const char *output, gboolean op_merged) { if (rc == 0) { next = NULL; } if (cmd->victim != NULL) { do_crm_log(rc == 0 ? LOG_NOTICE : LOG_ERR, "Operation '%s' [%d] (call %d from %s) for host '%s' with device '%s' returned%s: %d (%s)%s%s", cmd->action, pid, cmd->id, cmd->client_name, cmd->victim, cmd->device, (op_merged? " (merged)" : ""), rc, pcmk_strerror(rc), (next? ", retrying with " : ""), (next ? next : "")); } else { do_crm_log_unlikely(rc == 0 ? LOG_DEBUG : LOG_NOTICE, "Operation '%s' [%d] for device '%s' returned%s: %d (%s)%s%s", cmd->action, pid, cmd->device, (op_merged? " (merged)" : ""), rc, pcmk_strerror(rc), (next? ", retrying with " : ""), (next ? next : "")); } if (output) { /* Logging the whole string confuses syslog when the string is xml */ char *prefix = crm_strdup_printf("%s:%d", cmd->device, pid); crm_log_output(rc == 0 ? LOG_DEBUG : LOG_WARNING, prefix, output); free(prefix); } } static void stonith_send_async_reply(async_command_t * cmd, const char *output, int rc, GPid pid, int options) { xmlNode *reply = NULL; gboolean bcast = FALSE; reply = stonith_construct_async_reply(cmd, output, NULL, rc); if (safe_str_eq(cmd->action, "metadata")) { /* Too verbose to log */ crm_trace("Metadata query for %s", cmd->device); output = NULL; } else if (crm_str_eq(cmd->action, "monitor", TRUE) || crm_str_eq(cmd->action, "list", TRUE) || crm_str_eq(cmd->action, "status", TRUE)) { crm_trace("Never broadcast '%s' replies", cmd->action); } else if (!stand_alone && safe_str_eq(cmd->origin, cmd->victim) && safe_str_neq(cmd->action, "on")) { crm_trace("Broadcast '%s' reply for %s", cmd->action, cmd->victim); crm_xml_add(reply, F_SUBTYPE, "broadcast"); bcast = TRUE; } log_operation(cmd, rc, pid, NULL, output, (options & st_reply_opt_merged ? TRUE : FALSE)); crm_log_xml_trace(reply, "Reply"); if (options & st_reply_opt_merged) { crm_xml_add(reply, F_STONITH_MERGED, "true"); } if (bcast) { crm_xml_add(reply, F_STONITH_OPERATION, T_STONITH_NOTIFY); send_cluster_message(NULL, crm_msg_stonith_ng, reply, FALSE); } else if (cmd->origin) { crm_trace("Directed reply to %s", cmd->origin); send_cluster_message(crm_get_peer(0, cmd->origin), crm_msg_stonith_ng, reply, FALSE); } else { crm_trace("Directed local %ssync reply to %s", (cmd->options & st_opt_sync_call) ? "" : "a-", cmd->client_name); do_local_reply(reply, cmd->client, cmd->options & st_opt_sync_call, FALSE); } if (stand_alone) { /* Do notification with a clean data object */ xmlNode *notify_data = create_xml_node(NULL, T_STONITH_NOTIFY_FENCE); crm_xml_add_int(notify_data, F_STONITH_RC, rc); crm_xml_add(notify_data, F_STONITH_TARGET, cmd->victim); crm_xml_add(notify_data, F_STONITH_OPERATION, cmd->op); crm_xml_add(notify_data, F_STONITH_DELEGATE, "localhost"); crm_xml_add(notify_data, F_STONITH_DEVICE, cmd->device); crm_xml_add(notify_data, F_STONITH_REMOTE_OP_ID, cmd->remote_op_id); crm_xml_add(notify_data, F_STONITH_ORIGIN, cmd->client); do_stonith_notify(0, T_STONITH_NOTIFY_FENCE, rc, notify_data); do_stonith_notify(0, T_STONITH_NOTIFY_HISTORY, 0, NULL); } free_xml(reply); } static void cancel_stonith_command(async_command_t * cmd) { stonith_device_t *device; CRM_CHECK(cmd != NULL, return); if (!cmd->device) { return; } device = g_hash_table_lookup(device_list, cmd->device); if (device) { crm_trace("Cancel scheduled '%s' action on %s", cmd->action, device->id); device->pending_ops = g_list_remove(device->pending_ops, cmd); } } static void st_child_done(GPid pid, int rc, const char *output, gpointer user_data) { stonith_device_t *device = NULL; stonith_device_t *next_device = NULL; async_command_t *cmd = user_data; GListPtr gIter = NULL; GListPtr gIterNext = NULL; CRM_CHECK(cmd != NULL, return); cmd->active_on = NULL; /* The device is ready to do something else now */ device = g_hash_table_lookup(device_list, cmd->device); if (device) { if (!device->verified && (rc == pcmk_ok) && (safe_str_eq(cmd->action, "list") || safe_str_eq(cmd->action, "monitor") || safe_str_eq(cmd->action, "status"))) { device->verified = TRUE; } mainloop_set_trigger(device->work); } crm_debug("Operation '%s' on '%s' completed with rc=%d (%d remaining)", cmd->action, cmd->device, rc, g_list_length(cmd->device_next)); if (rc == 0) { GListPtr iter; /* see if there are any required devices left to execute for this op */ for (iter = cmd->device_next; iter != NULL; iter = iter->next) { next_device = g_hash_table_lookup(device_list, iter->data); if (next_device != NULL && is_action_required(cmd->action, next_device)) { cmd->device_next = iter->next; break; } next_device = NULL; } } else if (rc != 0 && cmd->device_next && (is_action_required(cmd->action, device) == FALSE)) { /* if this device didn't work out, see if there are any others we can try. * if the failed device was 'required', we can't pick another device. */ next_device = g_hash_table_lookup(device_list, cmd->device_next->data); cmd->device_next = cmd->device_next->next; } /* this operation requires more fencing, hooray! */ if (next_device) { log_operation(cmd, rc, pid, next_device->id, output, FALSE); schedule_stonith_command(cmd, next_device); /* Prevent cmd from being freed */ cmd = NULL; goto done; } stonith_send_async_reply(cmd, output, rc, pid, st_reply_opt_none); if (rc != 0) { goto done; } /* Check to see if any operations are scheduled to do the exact * same thing that just completed. If so, rather than * performing the same fencing operation twice, return the result * of this operation for all pending commands it matches. */ for (gIter = cmd_list; gIter != NULL; gIter = gIterNext) { async_command_t *cmd_other = gIter->data; gIterNext = gIter->next; if (cmd == cmd_other) { continue; } /* A pending scheduled command matches the command that just finished if. * 1. The client connections are different. * 2. The node victim is the same. * 3. The fencing action is the same. * 4. The device scheduled to execute the action is the same. */ if (safe_str_eq(cmd->client, cmd_other->client) || safe_str_neq(cmd->victim, cmd_other->victim) || safe_str_neq(cmd->action, cmd_other->action) || safe_str_neq(cmd->device, cmd_other->device)) { continue; } /* Duplicate merging will do the right thing for either type of remapped * reboot. If the executing fencer remapped an unsupported reboot to * off, then cmd->action will be reboot and will be merged with any * other reboot requests. If the originating fencer remapped a * topology reboot to off then on, we will get here once with * cmd->action "off" and once with "on", and they will be merged * separately with similar requests. */ crm_notice ("Merging stonith action '%s' targeting %s originating from client %s with identical stonith request from client %s", cmd_other->action, cmd_other->victim, cmd_other->client_name, cmd->client_name); cmd_list = g_list_remove_link(cmd_list, gIter); stonith_send_async_reply(cmd_other, output, rc, pid, st_reply_opt_merged); cancel_stonith_command(cmd_other); free_async_command(cmd_other); g_list_free_1(gIter); } done: free_async_command(cmd); } static gint sort_device_priority(gconstpointer a, gconstpointer b) { const stonith_device_t *dev_a = a; const stonith_device_t *dev_b = b; if (dev_a->priority > dev_b->priority) { return -1; } else if (dev_a->priority < dev_b->priority) { return 1; } return 0; } static void stonith_fence_get_devices_cb(GList * devices, void *user_data) { async_command_t *cmd = user_data; stonith_device_t *device = NULL; crm_info("Found %d matching devices for '%s'", g_list_length(devices), cmd->victim); if (devices != NULL) { /* Order based on priority */ devices = g_list_sort(devices, sort_device_priority); device = g_hash_table_lookup(device_list, devices->data); if (device) { cmd->device_list = devices; cmd->device_next = devices->next; devices = NULL; /* list owned by cmd now */ } } /* we have a device, schedule it for fencing. */ if (device) { schedule_stonith_command(cmd, device); /* in progress */ return; } /* no device found! */ stonith_send_async_reply(cmd, NULL, -ENODEV, 0, st_reply_opt_none); free_async_command(cmd); g_list_free_full(devices, free); } static int stonith_fence(xmlNode * msg) { const char *device_id = NULL; stonith_device_t *device = NULL; async_command_t *cmd = create_async_command(msg); xmlNode *dev = get_xpath_object("//@" F_STONITH_TARGET, msg, LOG_ERR); if (cmd == NULL) { return -EPROTO; } device_id = crm_element_value(dev, F_STONITH_DEVICE); if (device_id) { device = g_hash_table_lookup(device_list, device_id); if (device == NULL) { crm_err("Requested device '%s' is not available", device_id); return -ENODEV; } schedule_stonith_command(cmd, device); } else { const char *host = crm_element_value(dev, F_STONITH_TARGET); if (cmd->options & st_opt_cs_nodeid) { int nodeid = crm_atoi(host, NULL); crm_node_t *node = crm_find_known_peer_full(nodeid, NULL, CRM_GET_PEER_ANY); if (node) { host = node->uname; } } /* If we get to here, then self-fencing is implicitly allowed */ get_capable_devices(host, cmd->action, cmd->default_timeout, TRUE, cmd, stonith_fence_get_devices_cb); } return -EINPROGRESS; } xmlNode * stonith_construct_reply(xmlNode * request, const char *output, xmlNode * data, int rc) { xmlNode *reply = NULL; reply = create_xml_node(NULL, T_STONITH_REPLY); crm_xml_add(reply, "st_origin", __FUNCTION__); crm_xml_add(reply, F_TYPE, T_STONITH_NG); crm_xml_add(reply, "st_output", output); crm_xml_add_int(reply, F_STONITH_RC, rc); if (request == NULL) { /* Most likely, this is the result of a stonith operation that was * initiated before we came up. Unfortunately that means we lack enough * information to provide clients with a full result. * * @TODO Maybe synchronize this information at start-up? */ crm_warn("Missing request information for client notifications for " "operation with result %d (initiated before we came up?)", rc); } else { const char *name = NULL; const char *value = NULL; const char *names[] = { F_STONITH_OPERATION, F_STONITH_CALLID, F_STONITH_CLIENTID, F_STONITH_CLIENTNAME, F_STONITH_REMOTE_OP_ID, F_STONITH_CALLOPTS }; crm_trace("Creating a result reply with%s reply output (rc=%d)", (data? "" : "out"), rc); for (int lpc = 0; lpc < DIMOF(names); lpc++) { name = names[lpc]; value = crm_element_value(request, name); crm_xml_add(reply, name, value); } if (data != NULL) { add_message_xml(reply, F_STONITH_CALLDATA, data); } } return reply; } static xmlNode * stonith_construct_async_reply(async_command_t * cmd, const char *output, xmlNode * data, int rc) { xmlNode *reply = NULL; crm_trace("Creating a basic reply"); reply = create_xml_node(NULL, T_STONITH_REPLY); crm_xml_add(reply, "st_origin", __FUNCTION__); crm_xml_add(reply, F_TYPE, T_STONITH_NG); crm_xml_add(reply, F_STONITH_OPERATION, cmd->op); crm_xml_add(reply, F_STONITH_DEVICE, cmd->device); crm_xml_add(reply, F_STONITH_REMOTE_OP_ID, cmd->remote_op_id); crm_xml_add(reply, F_STONITH_CLIENTID, cmd->client); crm_xml_add(reply, F_STONITH_CLIENTNAME, cmd->client_name); crm_xml_add(reply, F_STONITH_TARGET, cmd->victim); crm_xml_add(reply, F_STONITH_ACTION, cmd->op); crm_xml_add(reply, F_STONITH_ORIGIN, cmd->origin); crm_xml_add_int(reply, F_STONITH_CALLID, cmd->id); crm_xml_add_int(reply, F_STONITH_CALLOPTS, cmd->options); crm_xml_add_int(reply, F_STONITH_RC, rc); crm_xml_add(reply, "st_output", output); if (data != NULL) { crm_info("Attaching reply output"); add_message_xml(reply, F_STONITH_CALLDATA, data); } return reply; } bool fencing_peer_active(crm_node_t *peer) { if (peer == NULL) { return FALSE; } else if (peer->uname == NULL) { return FALSE; } else if (is_set(peer->processes, crm_get_cluster_proc())) { return TRUE; } return FALSE; } /*! * \internal * \brief Determine if we need to use an alternate node to * fence the target. If so return that node's uname * * \retval NULL, no alternate host * \retval uname, uname of alternate host to use */ static const char * check_alternate_host(const char *target) { const char *alternate_host = NULL; crm_trace("Checking if we (%s) can fence %s", stonith_our_uname, target); if (find_topology_for_host(target) && safe_str_eq(target, stonith_our_uname)) { GHashTableIter gIter; crm_node_t *entry = NULL; g_hash_table_iter_init(&gIter, crm_peer_cache); while (g_hash_table_iter_next(&gIter, NULL, (void **)&entry)) { crm_trace("Checking for %s.%d != %s", entry->uname, entry->id, target); if (fencing_peer_active(entry) && safe_str_neq(entry->uname, target)) { alternate_host = entry->uname; break; } } if (alternate_host == NULL) { crm_err("No alternate host available to handle complex self fencing request"); g_hash_table_iter_init(&gIter, crm_peer_cache); while (g_hash_table_iter_next(&gIter, NULL, (void **)&entry)) { crm_notice("Peer[%d] %s", entry->id, entry->uname); } } } return alternate_host; } static void stonith_send_reply(xmlNode * reply, int call_options, const char *remote_peer, const char *client_id) { if (remote_peer) { send_cluster_message(crm_get_peer(0, remote_peer), crm_msg_stonith_ng, reply, FALSE); } else { do_local_reply(reply, client_id, is_set(call_options, st_opt_sync_call), remote_peer != NULL); } } static void remove_relay_op(xmlNode * request) { xmlNode *dev = get_xpath_object("//@" F_STONITH_ACTION, request, LOG_TRACE); const char *relay_op_id = NULL; const char *op_id = NULL; const char *client_name = NULL; const char *target = NULL; remote_fencing_op_t *relay_op = NULL; if (dev) { target = crm_element_value(dev, F_STONITH_TARGET); } relay_op_id = crm_element_value(request, F_STONITH_REMOTE_OP_ID_RELAY); op_id = crm_element_value(request, F_STONITH_REMOTE_OP_ID); client_name = crm_element_value(request, F_STONITH_CLIENTNAME); /* Delete RELAY operation. */ if (relay_op_id && target && safe_str_eq(target, stonith_our_uname)) { relay_op = g_hash_table_lookup(stonith_remote_op_list, relay_op_id); if (relay_op) { GHashTableIter iter; remote_fencing_op_t *list_op = NULL; g_hash_table_iter_init(&iter, stonith_remote_op_list); /* If the operation to be deleted is registered as a duplicate, delete the registration. */ while (g_hash_table_iter_next(&iter, NULL, (void **)&list_op)) { GListPtr dup_iter = NULL; if (list_op != relay_op) { for (dup_iter = list_op->duplicates; dup_iter != NULL; dup_iter = dup_iter->next) { remote_fencing_op_t *other = dup_iter->data; if (other == relay_op) { other->duplicates = g_list_remove(other->duplicates, relay_op); break; } } } } crm_info("Delete the relay op : %s - %s of %s for %s.(replaced by op : %s - %s of %s for %s)", relay_op->id, relay_op->action, relay_op->target, relay_op->client_name, op_id, relay_op->action, target, client_name); g_hash_table_remove(stonith_remote_op_list, relay_op_id); } } } static int handle_request(pcmk__client_t *client, uint32_t id, uint32_t flags, xmlNode *request, const char *remote_peer) { int call_options = 0; int rc = -EOPNOTSUPP; xmlNode *data = NULL; xmlNode *reply = NULL; char *output = NULL; const char *op = crm_element_value(request, F_STONITH_OPERATION); const char *client_id = crm_element_value(request, F_STONITH_CLIENTID); crm_element_value_int(request, F_STONITH_CALLOPTS, &call_options); if (is_set(call_options, st_opt_sync_call)) { CRM_ASSERT(client == NULL || client->request_id == id); } if (crm_str_eq(op, CRM_OP_REGISTER, TRUE)) { xmlNode *reply = create_xml_node(NULL, "reply"); CRM_ASSERT(client); crm_xml_add(reply, F_STONITH_OPERATION, CRM_OP_REGISTER); crm_xml_add(reply, F_STONITH_CLIENTID, client->id); pcmk__ipc_send_xml(client, id, reply, flags); client->request_id = 0; free_xml(reply); return 0; } else if (crm_str_eq(op, STONITH_OP_EXEC, TRUE)) { rc = stonith_device_action(request, &output); } else if (crm_str_eq(op, STONITH_OP_TIMEOUT_UPDATE, TRUE)) { const char *call_id = crm_element_value(request, F_STONITH_CALLID); const char *client_id = crm_element_value(request, F_STONITH_CLIENTID); int op_timeout = 0; crm_element_value_int(request, F_STONITH_TIMEOUT, &op_timeout); do_stonith_async_timeout_update(client_id, call_id, op_timeout); return 0; } else if (crm_str_eq(op, STONITH_OP_QUERY, TRUE)) { if (remote_peer) { create_remote_stonith_op(client_id, request, TRUE); /* Record it for the future notification */ } /* Delete the DC node RELAY operation. */ remove_relay_op(request); stonith_query(request, remote_peer, client_id, call_options); return 0; } else if (crm_str_eq(op, T_STONITH_NOTIFY, TRUE)) { const char *flag_name = NULL; CRM_ASSERT(client); flag_name = crm_element_value(request, F_STONITH_NOTIFY_ACTIVATE); if (flag_name) { crm_debug("Setting %s callbacks for %s (%s): ON", flag_name, client->name, client->id); client->options |= get_stonith_flag(flag_name); } flag_name = crm_element_value(request, F_STONITH_NOTIFY_DEACTIVATE); if (flag_name) { crm_debug("Setting %s callbacks for %s (%s): off", flag_name, client->name, client->id); client->options |= get_stonith_flag(flag_name); } if (flags & crm_ipc_client_response) { pcmk__ipc_send_ack(client, id, flags, "ack"); } return 0; } else if (crm_str_eq(op, STONITH_OP_RELAY, TRUE)) { xmlNode *dev = get_xpath_object("//@" F_STONITH_TARGET, request, LOG_TRACE); crm_notice("Peer %s has received a forwarded fencing request from %s to fence (%s) peer %s", stonith_our_uname, client ? client->name : remote_peer, crm_element_value(dev, F_STONITH_ACTION), crm_element_value(dev, F_STONITH_TARGET)); if (initiate_remote_stonith_op(NULL, request, FALSE) != NULL) { rc = -EINPROGRESS; } } else if (crm_str_eq(op, STONITH_OP_FENCE, TRUE)) { if (remote_peer || stand_alone) { rc = stonith_fence(request); } else if (call_options & st_opt_manual_ack) { remote_fencing_op_t *rop = NULL; xmlNode *dev = get_xpath_object("//@" F_STONITH_TARGET, request, LOG_TRACE); const char *target = crm_element_value(dev, F_STONITH_TARGET); crm_notice("Received manual confirmation that %s is fenced", target); rop = initiate_remote_stonith_op(client, request, TRUE); rc = stonith_manual_ack(request, rop); } else { const char *alternate_host = NULL; xmlNode *dev = get_xpath_object("//@" F_STONITH_TARGET, request, LOG_TRACE); const char *target = crm_element_value(dev, F_STONITH_TARGET); const char *action = crm_element_value(dev, F_STONITH_ACTION); const char *device = crm_element_value(dev, F_STONITH_DEVICE); if (client) { int tolerance = 0; crm_notice("Client %s.%.8s wants to fence (%s) '%s' with device '%s'", client->name, client->id, action, target, device ? device : "(any)"); crm_element_value_int(dev, F_STONITH_TOLERANCE, &tolerance); if (stonith_check_fence_tolerance(tolerance, target, action)) { rc = 0; goto done; } } else { crm_notice("Peer %s wants to fence (%s) '%s' with device '%s'", remote_peer, action, target, device ? device : "(any)"); } alternate_host = check_alternate_host(target); if (alternate_host && client) { const char *client_id = NULL; remote_fencing_op_t *op = NULL; crm_notice("Forwarding complex self fencing request to peer %s", alternate_host); if (client->id) { client_id = client->id; } else { client_id = crm_element_value(request, F_STONITH_CLIENTID); } /* Create an operation for RELAY and send the ID in the RELAY message. */ /* When a QUERY response is received, delete the RELAY operation to avoid the existence of duplicate operations. */ op = create_remote_stonith_op(client_id, request, FALSE); crm_xml_add(request, F_STONITH_OPERATION, STONITH_OP_RELAY); crm_xml_add(request, F_STONITH_CLIENTID, client->id); crm_xml_add(request, F_STONITH_REMOTE_OP_ID, op->id); send_cluster_message(crm_get_peer(0, alternate_host), crm_msg_stonith_ng, request, FALSE); rc = -EINPROGRESS; } else if (initiate_remote_stonith_op(client, request, FALSE) != NULL) { rc = -EINPROGRESS; } } } else if (crm_str_eq(op, STONITH_OP_FENCE_HISTORY, TRUE)) { rc = stonith_fence_history(request, &data, remote_peer, call_options); if (call_options & st_opt_discard_reply) { /* we don't expect answers to the broadcast * we might have sent out */ free_xml(data); return pcmk_ok; } } else if (crm_str_eq(op, STONITH_OP_DEVICE_ADD, TRUE)) { const char *device_id = NULL; rc = stonith_device_register(request, &device_id, FALSE); do_stonith_notify_device(call_options, op, rc, device_id); } else if (crm_str_eq(op, STONITH_OP_DEVICE_DEL, TRUE)) { xmlNode *dev = get_xpath_object("//" F_STONITH_DEVICE, request, LOG_ERR); const char *device_id = crm_element_value(dev, XML_ATTR_ID); rc = stonith_device_remove(device_id, FALSE); do_stonith_notify_device(call_options, op, rc, device_id); } else if (crm_str_eq(op, STONITH_OP_LEVEL_ADD, TRUE)) { char *device_id = NULL; rc = stonith_level_register(request, &device_id); do_stonith_notify_level(call_options, op, rc, device_id); free(device_id); } else if (crm_str_eq(op, STONITH_OP_LEVEL_DEL, TRUE)) { char *device_id = NULL; rc = stonith_level_remove(request, &device_id); do_stonith_notify_level(call_options, op, rc, device_id); } else if(safe_str_eq(op, CRM_OP_RM_NODE_CACHE)) { int node_id = 0; const char *name = NULL; crm_element_value_int(request, XML_ATTR_ID, &node_id); name = crm_element_value(request, XML_ATTR_UNAME); reap_crm_member(node_id, name); return pcmk_ok; } else { crm_err("Unknown IPC request %s from %s", op, (client? client->name : remote_peer)); } done: /* Always reply unless the request is in process still. * If in progress, a reply will happen async after the request * processing is finished */ if (rc != -EINPROGRESS) { crm_trace("Reply handling: %p %u %u %d %d %s", client, client?client->request_id:0, id, is_set(call_options, st_opt_sync_call), call_options, crm_element_value(request, F_STONITH_CALLOPTS)); if (is_set(call_options, st_opt_sync_call)) { CRM_ASSERT(client == NULL || client->request_id == id); } reply = stonith_construct_reply(request, output, data, rc); stonith_send_reply(reply, call_options, remote_peer, client_id); } free(output); free_xml(data); free_xml(reply); return rc; } static void handle_reply(pcmk__client_t *client, xmlNode *request, const char *remote_peer) { const char *op = crm_element_value(request, F_STONITH_OPERATION); if (crm_str_eq(op, STONITH_OP_QUERY, TRUE)) { process_remote_stonith_query(request); } else if (crm_str_eq(op, T_STONITH_NOTIFY, TRUE)) { process_remote_stonith_exec(request); } else if (crm_str_eq(op, STONITH_OP_FENCE, TRUE)) { /* Reply to a complex fencing op */ process_remote_stonith_exec(request); } else { crm_err("Unknown %s reply from %s", op, client ? client->name : remote_peer); crm_log_xml_warn(request, "UnknownOp"); } } void stonith_command(pcmk__client_t *client, uint32_t id, uint32_t flags, xmlNode *request, const char *remote_peer) { int call_options = 0; int rc = 0; gboolean is_reply = FALSE; /* Copy op for reporting. The original might get freed by handle_reply() * before we use it in crm_debug(): * handle_reply() * |- process_remote_stonith_exec() * |-- remote_op_done() * |--- handle_local_reply_and_notify() * |---- crm_xml_add(...F_STONITH_OPERATION...) * |--- free_xml(op->request) */ char *op = crm_element_value_copy(request, F_STONITH_OPERATION); if (get_xpath_object("//" T_STONITH_REPLY, request, LOG_NEVER)) { is_reply = TRUE; } crm_element_value_int(request, F_STONITH_CALLOPTS, &call_options); crm_debug("Processing %s%s %u from %s (%16x)", op, is_reply ? " reply" : "", id, client ? client->name : remote_peer, call_options); if (is_set(call_options, st_opt_sync_call)) { CRM_ASSERT(client == NULL || client->request_id == id); } if (is_reply) { handle_reply(client, request, remote_peer); } else { rc = handle_request(client, id, flags, request, remote_peer); } crm_debug("Processed %s%s from %s: %s (%d)", op, is_reply ? " reply" : "", client ? client->name : remote_peer, rc > 0 ? "" : pcmk_strerror(rc), rc); free(op); } diff --git a/include/crm/common/strings_internal.h b/include/crm/common/strings_internal.h index 681a59f433..8aca4ff9de 100644 --- a/include/crm/common/strings_internal.h +++ b/include/crm/common/strings_internal.h @@ -1,63 +1,62 @@ /* * Copyright 2015-2020 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #ifndef PCMK__STRINGS_INTERNAL__H #define PCMK__STRINGS_INTERNAL__H #include // bool #include // guint, GList, GHashTable /* internal generic string functions (from strings.c) */ int pcmk__guint_from_hash(GHashTable *table, const char *key, guint default_val, guint *result); bool pcmk__starts_with(const char *str, const char *prefix); bool pcmk__ends_with(const char *s, const char *match); bool pcmk__ends_with_ext(const char *s, const char *match); char *pcmk__add_word(char *list, const char *word); int pcmk__compress(const char *data, unsigned int length, unsigned int max, char **result, unsigned int *result_len); int pcmk__parse_ll_range(const char *srcstring, long long *start, long long *end); gboolean pcmk__str_in_list(GList *lst, const gchar *s); bool pcmk__str_any_of(const char *s, ...) G_GNUC_NULL_TERMINATED; -bool pcmk__str_none_of(const char *s, ...) G_GNUC_NULL_TERMINATED; /* Correctly displaying singular or plural is complicated; consider "1 node has" * vs. "2 nodes have". A flexible solution is to pluralize entire strings, e.g. * * if (a == 1) { * crm_info("singular message"): * } else { * crm_info("plural message"); * } * * though even that's not sufficient for all languages besides English (if we * ever desire to do translations of output and log messages). But the following * convenience macros are "good enough" and more concise for many cases. */ /* Example: * crm_info("Found %d %s", nentries, * pcmk__plural_alt(nentries, "entry", "entries")); */ #define pcmk__plural_alt(i, s1, s2) (((i) == 1)? (s1) : (s2)) // Example: crm_info("Found %d node%s", nnodes, pcmk__plural_s(nnodes)); #define pcmk__plural_s(i) pcmk__plural_alt(i, "", "s") static inline int pcmk__str_empty(const char *s) { return (s == NULL) || (s[0] == '\0'); } #endif /* PCMK__STRINGS_INTERNAL__H */ diff --git a/lib/common/logging.c b/lib/common/logging.c index 6f3dc3550f..d2de15e7e3 100644 --- a/lib/common/logging.c +++ b/lib/common/logging.c @@ -1,1002 +1,1002 @@ /* * Copyright 2004-2020 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include unsigned int crm_log_priority = LOG_NOTICE; unsigned int crm_log_level = LOG_INFO; static gboolean crm_tracing_enabled(void); unsigned int crm_trace_nonlog = 0; bool crm_is_daemon = 0; GLogFunc glib_log_default; static void crm_glib_handler(const gchar * log_domain, GLogLevelFlags flags, const gchar * message, gpointer user_data) { int log_level = LOG_WARNING; GLogLevelFlags msg_level = (flags & G_LOG_LEVEL_MASK); static struct qb_log_callsite *glib_cs = NULL; if (glib_cs == NULL) { glib_cs = qb_log_callsite_get(__FUNCTION__, __FILE__, "glib-handler", LOG_DEBUG, __LINE__, crm_trace_nonlog); } switch (msg_level) { case G_LOG_LEVEL_CRITICAL: log_level = LOG_CRIT; if (crm_is_callsite_active(glib_cs, LOG_DEBUG, 0) == FALSE) { /* log and record how we got here */ crm_abort(__FILE__, __FUNCTION__, __LINE__, message, TRUE, TRUE); } break; case G_LOG_LEVEL_ERROR: log_level = LOG_ERR; break; case G_LOG_LEVEL_MESSAGE: log_level = LOG_NOTICE; break; case G_LOG_LEVEL_INFO: log_level = LOG_INFO; break; case G_LOG_LEVEL_DEBUG: log_level = LOG_DEBUG; break; case G_LOG_LEVEL_WARNING: case G_LOG_FLAG_RECURSION: case G_LOG_FLAG_FATAL: case G_LOG_LEVEL_MASK: log_level = LOG_WARNING; break; } do_crm_log(log_level, "%s: %s", log_domain, message); } #ifndef NAME_MAX # define NAME_MAX 256 #endif /*! * \internal * \brief Write out a blackbox (enabling blackboxes if needed) * * \param[in] nsig Signal number that was received * * \note This is a true signal handler, and so must be async-safe. */ static void crm_trigger_blackbox(int nsig) { if(nsig == SIGTRAP) { /* Turn it on if it wasn't already */ crm_enable_blackbox(nsig); } crm_write_blackbox(nsig, NULL); } void crm_log_deinit(void) { g_log_set_default_handler(glib_log_default, NULL); } #define FMT_MAX 256 static void set_format_string(int method, const char *daemon) { if (method == QB_LOG_SYSLOG) { // The system log gets a simplified, user-friendly format crm_extended_logging(method, QB_FALSE); qb_log_format_set(method, "%g %p: %b"); } else { // Everything else gets more detail, for advanced troubleshooting int offset = 0; char fmt[FMT_MAX]; if (method > QB_LOG_STDERR) { struct utsname res; const char *nodename = "localhost"; if (uname(&res) == 0) { nodename = res.nodename; } // If logging to file, prefix with timestamp, node name, daemon ID offset += snprintf(fmt + offset, FMT_MAX - offset, "%%t %s %-20s[%lu] ", nodename, daemon, (unsigned long) getpid()); } // Add function name (in parentheses) offset += snprintf(fmt + offset, FMT_MAX - offset, "(%%n"); if (crm_tracing_enabled()) { // When tracing, add file and line number offset += snprintf(fmt + offset, FMT_MAX - offset, "@%%f:%%l"); } offset += snprintf(fmt + offset, FMT_MAX - offset, ")"); // Add tag (if any), severity, and actual message offset += snprintf(fmt + offset, FMT_MAX - offset, " %%g\t%%p: %%b"); CRM_LOG_ASSERT(offset > 0); qb_log_format_set(method, fmt); } } gboolean crm_add_logfile(const char *filename) { bool is_default = false; static int default_fd = -1; static gboolean have_logfile = FALSE; const char *default_logfile = CRM_LOG_DIR "/pacemaker.log"; struct stat parent; int fd = 0, rc = 0; FILE *logfile = NULL; char *parent_dir = NULL; char *filename_cp; if (filename == NULL && have_logfile == FALSE) { filename = default_logfile; } if (filename == NULL) { return FALSE; /* Nothing to do */ } else if(safe_str_eq(filename, "none")) { return FALSE; /* Nothing to do */ } else if(safe_str_eq(filename, "/dev/null")) { return FALSE; /* Nothing to do */ } else if(safe_str_eq(filename, default_logfile)) { is_default = TRUE; } if(is_default && default_fd >= 0) { return TRUE; /* Nothing to do */ } /* Check the parent directory */ filename_cp = strdup(filename); parent_dir = dirname(filename_cp); rc = stat(parent_dir, &parent); if (rc != 0) { crm_err("Directory '%s' does not exist: logging to '%s' is disabled", parent_dir, filename); free(filename_cp); return FALSE; } free(filename_cp); errno = 0; logfile = fopen(filename, "a"); if(logfile == NULL) { crm_err("%s (%d): Logging to '%s' as uid=%u, gid=%u is disabled", pcmk_strerror(errno), errno, filename, geteuid(), getegid()); return FALSE; } /* Check/Set permissions if we're root */ if (geteuid() == 0) { struct stat st; uid_t pcmk_uid = 0; gid_t pcmk_gid = 0; gboolean fix = FALSE; int logfd = fileno(logfile); const char *modestr; mode_t filemode = S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP; rc = fstat(logfd, &st); if (rc < 0) { crm_perror(LOG_WARNING, "Cannot stat %s", filename); fclose(logfile); return FALSE; } if (pcmk_daemon_user(&pcmk_uid, &pcmk_gid) == 0) { if (st.st_gid != pcmk_gid) { /* Wrong group */ fix = TRUE; } else if ((st.st_mode & S_IRWXG) != (S_IRGRP | S_IWGRP)) { /* Not read/writable by the correct group */ fix = TRUE; } } modestr = getenv("PCMK_logfile_mode"); if (modestr) { long filemode_l = strtol(modestr, NULL, 8); if (filemode_l != LONG_MIN && filemode_l != LONG_MAX) { filemode = (mode_t)filemode_l; } } if (fix) { rc = fchown(logfd, pcmk_uid, pcmk_gid); if (rc < 0) { crm_warn("Cannot change the ownership of %s to user %s and gid %d", filename, CRM_DAEMON_USER, pcmk_gid); } } if (filemode) { rc = fchmod(logfd, filemode); if (rc < 0) { crm_warn("Cannot change the mode of %s to %o", filename, filemode); } fprintf(logfile, "Set r/w permissions for uid=%d, gid=%d on %s\n", pcmk_uid, pcmk_gid, filename); if (fflush(logfile) < 0 || fsync(logfd) < 0) { crm_err("Couldn't write out logfile: %s", filename); } } } /* Close and reopen with libqb */ fclose(logfile); fd = qb_log_file_open(filename); if (fd < 0) { crm_perror(LOG_WARNING, "Couldn't send additional logging to %s", filename); return FALSE; } if(is_default) { default_fd = fd; // Some resource agents will log only if environment variable is set if (pcmk__env_option("logfile") == NULL) { pcmk__set_env_option("logfile", filename); } } else if(default_fd >= 0) { crm_notice("Switching to %s", filename); qb_log_ctl(default_fd, QB_LOG_CONF_ENABLED, QB_FALSE); } crm_notice("Additional logging available in %s", filename); qb_log_ctl(fd, QB_LOG_CONF_ENABLED, QB_TRUE); /* qb_log_ctl(fd, QB_LOG_CONF_FILE_SYNC, 1); Turn on synchronous writes */ #ifdef HAVE_qb_log_conf_QB_LOG_CONF_MAX_LINE_LEN // Longer than default, for logging long XML lines qb_log_ctl(fd, QB_LOG_CONF_MAX_LINE_LEN, 800); #endif /* Enable callsites */ crm_update_callsites(); have_logfile = TRUE; return TRUE; } static int blackbox_trigger = 0; static volatile char *blackbox_file_prefix = NULL; #ifdef QB_FEATURE_LOG_HIRES_TIMESTAMPS typedef struct timespec *log_time_t; #else typedef time_t log_time_t; #endif static void blackbox_logger(int32_t t, struct qb_log_callsite *cs, log_time_t timestamp, const char *msg) { if(cs && cs->priority < LOG_ERR) { crm_write_blackbox(SIGTRAP, cs); /* Bypass the over-dumping logic */ } else { crm_write_blackbox(0, cs); } } static void crm_control_blackbox(int nsig, bool enable) { int lpc = 0; if (blackbox_file_prefix == NULL) { pid_t pid = getpid(); blackbox_file_prefix = crm_strdup_printf("%s/%s-%lu", CRM_BLACKBOX_DIR, crm_system_name, (unsigned long) pid); } if (enable && qb_log_ctl(QB_LOG_BLACKBOX, QB_LOG_CONF_STATE_GET, 0) != QB_LOG_STATE_ENABLED) { qb_log_ctl(QB_LOG_BLACKBOX, QB_LOG_CONF_SIZE, 5 * 1024 * 1024); /* Any size change drops existing entries */ qb_log_ctl(QB_LOG_BLACKBOX, QB_LOG_CONF_ENABLED, QB_TRUE); /* Setting the size seems to disable it */ /* Enable synchronous logging */ for (lpc = QB_LOG_BLACKBOX; lpc < QB_LOG_TARGET_MAX; lpc++) { qb_log_ctl(lpc, QB_LOG_CONF_FILE_SYNC, QB_TRUE); } crm_notice("Initiated blackbox recorder: %s", blackbox_file_prefix); /* Save to disk on abnormal termination */ crm_signal_handler(SIGSEGV, crm_trigger_blackbox); crm_signal_handler(SIGABRT, crm_trigger_blackbox); crm_signal_handler(SIGILL, crm_trigger_blackbox); crm_signal_handler(SIGBUS, crm_trigger_blackbox); crm_signal_handler(SIGFPE, crm_trigger_blackbox); crm_update_callsites(); blackbox_trigger = qb_log_custom_open(blackbox_logger, NULL, NULL, NULL); qb_log_ctl(blackbox_trigger, QB_LOG_CONF_ENABLED, QB_TRUE); crm_trace("Trigger: %d is %d %d", blackbox_trigger, qb_log_ctl(blackbox_trigger, QB_LOG_CONF_STATE_GET, 0), QB_LOG_STATE_ENABLED); crm_update_callsites(); } else if (!enable && qb_log_ctl(QB_LOG_BLACKBOX, QB_LOG_CONF_STATE_GET, 0) == QB_LOG_STATE_ENABLED) { qb_log_ctl(QB_LOG_BLACKBOX, QB_LOG_CONF_ENABLED, QB_FALSE); /* Disable synchronous logging again when the blackbox is disabled */ for (lpc = QB_LOG_BLACKBOX; lpc < QB_LOG_TARGET_MAX; lpc++) { qb_log_ctl(lpc, QB_LOG_CONF_FILE_SYNC, QB_FALSE); } } } void crm_enable_blackbox(int nsig) { crm_control_blackbox(nsig, TRUE); } void crm_disable_blackbox(int nsig) { crm_control_blackbox(nsig, FALSE); } /*! * \internal * \brief Write out a blackbox, if blackboxes are enabled * * \param[in] nsig Signal that was received * \param[in] cs libqb callsite * * \note This may be called via a true signal handler and so must be async-safe. * @TODO actually make this async-safe */ void crm_write_blackbox(int nsig, struct qb_log_callsite *cs) { static volatile int counter = 1; static volatile time_t last = 0; char buffer[NAME_MAX]; time_t now = time(NULL); if (blackbox_file_prefix == NULL) { return; } switch (nsig) { case 0: case SIGTRAP: /* The graceful case - such as assertion failure or user request */ if (nsig == 0 && now == last) { /* Prevent over-dumping */ return; } snprintf(buffer, NAME_MAX, "%s.%d", blackbox_file_prefix, counter++); if (nsig == SIGTRAP) { crm_notice("Blackbox dump requested, please see %s for contents", buffer); } else if (cs) { syslog(LOG_NOTICE, "Problem detected at %s:%d (%s), please see %s for additional details", cs->function, cs->lineno, cs->filename, buffer); } else { crm_notice("Problem detected, please see %s for additional details", buffer); } last = now; qb_log_blackbox_write_to_file(buffer); /* Flush the existing contents * A size change would also work */ qb_log_ctl(QB_LOG_BLACKBOX, QB_LOG_CONF_ENABLED, QB_FALSE); qb_log_ctl(QB_LOG_BLACKBOX, QB_LOG_CONF_ENABLED, QB_TRUE); break; default: /* Do as little as possible, just try to get what we have out * We logged the filename when the blackbox was enabled */ crm_signal_handler(nsig, SIG_DFL); qb_log_blackbox_write_to_file((const char *)blackbox_file_prefix); qb_log_ctl(QB_LOG_BLACKBOX, QB_LOG_CONF_ENABLED, QB_FALSE); raise(nsig); break; } } gboolean crm_log_cli_init(const char *entity) { return crm_log_init(entity, LOG_ERR, FALSE, FALSE, 0, NULL, TRUE); } static const char * crm_quark_to_string(uint32_t tag) { const char *text = g_quark_to_string(tag); if (text) { return text; } return ""; } static void crm_log_filter_source(int source, const char *trace_files, const char *trace_fns, const char *trace_fmts, const char *trace_tags, const char *trace_blackbox, struct qb_log_callsite *cs) { if (qb_log_ctl(source, QB_LOG_CONF_STATE_GET, 0) != QB_LOG_STATE_ENABLED) { return; } else if (cs->tags != crm_trace_nonlog && source == QB_LOG_BLACKBOX) { /* Blackbox gets everything if enabled */ qb_bit_set(cs->targets, source); } else if (source == blackbox_trigger && blackbox_trigger > 0) { /* Should this log message result in the blackbox being dumped */ if (cs->priority <= LOG_ERR) { qb_bit_set(cs->targets, source); } else if (trace_blackbox) { char *key = crm_strdup_printf("%s:%d", cs->function, cs->lineno); if (strstr(trace_blackbox, key) != NULL) { qb_bit_set(cs->targets, source); } free(key); } } else if (source == QB_LOG_SYSLOG) { /* No tracing to syslog */ if (cs->priority <= crm_log_priority && cs->priority <= crm_log_level) { qb_bit_set(cs->targets, source); } /* Log file tracing options... */ } else if (cs->priority <= crm_log_level) { qb_bit_set(cs->targets, source); } else if (trace_files && strstr(trace_files, cs->filename) != NULL) { qb_bit_set(cs->targets, source); } else if (trace_fns && strstr(trace_fns, cs->function) != NULL) { qb_bit_set(cs->targets, source); } else if (trace_fmts && strstr(trace_fmts, cs->format) != NULL) { qb_bit_set(cs->targets, source); } else if (trace_tags && cs->tags != 0 && cs->tags != crm_trace_nonlog && g_quark_to_string(cs->tags) != NULL) { qb_bit_set(cs->targets, source); } } static void crm_log_filter(struct qb_log_callsite *cs) { int lpc = 0; static int need_init = 1; static const char *trace_fns = NULL; static const char *trace_tags = NULL; static const char *trace_fmts = NULL; static const char *trace_files = NULL; static const char *trace_blackbox = NULL; if (need_init) { need_init = 0; trace_fns = getenv("PCMK_trace_functions"); trace_fmts = getenv("PCMK_trace_formats"); trace_tags = getenv("PCMK_trace_tags"); trace_files = getenv("PCMK_trace_files"); trace_blackbox = getenv("PCMK_trace_blackbox"); if (trace_tags != NULL) { uint32_t tag; char token[500]; const char *offset = NULL; const char *next = trace_tags; do { offset = next; next = strchrnul(offset, ','); snprintf(token, sizeof(token), "%.*s", (int)(next - offset), offset); tag = g_quark_from_string(token); crm_info("Created GQuark %u from token '%s' in '%s'", tag, token, trace_tags); if (next[0] != 0) { next++; } } while (next != NULL && next[0] != 0); } } cs->targets = 0; /* Reset then find targets to enable */ for (lpc = QB_LOG_SYSLOG; lpc < QB_LOG_TARGET_MAX; lpc++) { crm_log_filter_source(lpc, trace_files, trace_fns, trace_fmts, trace_tags, trace_blackbox, cs); } } gboolean crm_is_callsite_active(struct qb_log_callsite *cs, uint8_t level, uint32_t tags) { gboolean refilter = FALSE; if (cs == NULL) { return FALSE; } if (cs->priority != level) { cs->priority = level; refilter = TRUE; } if (cs->tags != tags) { cs->tags = tags; refilter = TRUE; } if (refilter) { crm_log_filter(cs); } if (cs->targets == 0) { return FALSE; } return TRUE; } void crm_update_callsites(void) { static gboolean log = TRUE; if (log) { log = FALSE; crm_debug ("Enabling callsites based on priority=%d, files=%s, functions=%s, formats=%s, tags=%s", crm_log_level, getenv("PCMK_trace_files"), getenv("PCMK_trace_functions"), getenv("PCMK_trace_formats"), getenv("PCMK_trace_tags")); } qb_log_filter_fn_set(crm_log_filter); } static gboolean crm_tracing_enabled(void) { if (crm_log_level == LOG_TRACE) { return TRUE; } else if (getenv("PCMK_trace_files") || getenv("PCMK_trace_functions") || getenv("PCMK_trace_formats") || getenv("PCMK_trace_tags")) { return TRUE; } return FALSE; } static int crm_priority2int(const char *name) { struct syslog_names { const char *name; int priority; }; static struct syslog_names p_names[] = { {"emerg", LOG_EMERG}, {"alert", LOG_ALERT}, {"crit", LOG_CRIT}, {"error", LOG_ERR}, {"warning", LOG_WARNING}, {"notice", LOG_NOTICE}, {"info", LOG_INFO}, {"debug", LOG_DEBUG}, {NULL, -1} }; int lpc; for (lpc = 0; name != NULL && p_names[lpc].name != NULL; lpc++) { if (crm_str_eq(p_names[lpc].name, name, TRUE)) { return p_names[lpc].priority; } } return crm_log_priority; } static void crm_identity(const char *entity, int argc, char **argv) { if(crm_system_name != NULL) { /* Nothing to do */ } else if (entity) { free(crm_system_name); crm_system_name = strdup(entity); } else if (argc > 0 && argv != NULL) { char *mutable = strdup(argv[0]); char *modified = basename(mutable); if (strstr(modified, "lt-") == modified) { modified += 3; } free(crm_system_name); crm_system_name = strdup(modified); free(mutable); } else if (crm_system_name == NULL) { crm_system_name = strdup("Unknown"); } setenv("PCMK_service", crm_system_name, 1); } void crm_log_preinit(const char *entity, int argc, char **argv) { /* Configure libqb logging with nothing turned on */ int lpc = 0; int32_t qb_facility = 0; static bool have_logging = FALSE; if(have_logging == FALSE) { have_logging = TRUE; crm_xml_init(); /* Sets buffer allocation strategy */ if (crm_trace_nonlog == 0) { crm_trace_nonlog = g_quark_from_static_string("Pacemaker non-logging tracepoint"); } umask(S_IWGRP | S_IWOTH | S_IROTH); /* Redirect messages from glib functions to our handler */ glib_log_default = g_log_set_default_handler(crm_glib_handler, NULL); /* and for good measure... - this enum is a bit field (!) */ g_log_set_always_fatal((GLogLevelFlags) 0); /*value out of range */ /* Who do we log as */ crm_identity(entity, argc, argv); qb_facility = qb_log_facility2int("local0"); qb_log_init(crm_system_name, qb_facility, LOG_ERR); crm_log_level = LOG_CRIT; /* Nuke any syslog activity until it's asked for */ qb_log_ctl(QB_LOG_SYSLOG, QB_LOG_CONF_ENABLED, QB_FALSE); #ifdef HAVE_qb_log_conf_QB_LOG_CONF_MAX_LINE_LEN // Shorter than default, generous for what we *should* send to syslog qb_log_ctl(QB_LOG_SYSLOG, QB_LOG_CONF_MAX_LINE_LEN, 256); #endif /* Set format strings and disable threading * Pacemaker and threads do not mix well (due to the amount of forking) */ qb_log_tags_stringify_fn_set(crm_quark_to_string); for (lpc = QB_LOG_SYSLOG; lpc < QB_LOG_TARGET_MAX; lpc++) { qb_log_ctl(lpc, QB_LOG_CONF_THREADED, QB_FALSE); #ifdef HAVE_qb_log_conf_QB_LOG_CONF_ELLIPSIS // End truncated lines with '...' qb_log_ctl(lpc, QB_LOG_CONF_ELLIPSIS, QB_TRUE); #endif set_format_string(lpc, crm_system_name); } } } gboolean crm_log_init(const char *entity, uint8_t level, gboolean daemon, gboolean to_stderr, int argc, char **argv, gboolean quiet) { const char *syslog_priority = NULL; const char *logfile = pcmk__env_option("logfile"); const char *facility = pcmk__env_option("logfacility"); const char *f_copy = facility; crm_is_daemon = daemon; crm_log_preinit(entity, argc, argv); if (level > LOG_TRACE) { level = LOG_TRACE; } if(level > crm_log_level) { crm_log_level = level; } /* Should we log to syslog */ if (facility == NULL) { if(crm_is_daemon) { facility = "daemon"; } else { facility = "none"; } pcmk__set_env_option("logfacility", facility); } if (safe_str_eq(facility, "none")) { quiet = TRUE; } else { qb_log_ctl(QB_LOG_SYSLOG, QB_LOG_CONF_FACILITY, qb_log_facility2int(facility)); } if (pcmk__env_option_enabled(crm_system_name, "debug")) { /* Override the default setting */ crm_log_level = LOG_DEBUG; } /* What lower threshold do we have for sending to syslog */ syslog_priority = pcmk__env_option("logpriority"); if(syslog_priority) { int priority = crm_priority2int(syslog_priority); crm_log_priority = priority; qb_log_filter_ctl(QB_LOG_SYSLOG, QB_LOG_FILTER_ADD, QB_LOG_FILTER_FILE, "*", priority); } else { qb_log_filter_ctl(QB_LOG_SYSLOG, QB_LOG_FILTER_ADD, QB_LOG_FILTER_FILE, "*", LOG_NOTICE); } // Log to syslog unless requested to be quiet if (!quiet) { qb_log_ctl(QB_LOG_SYSLOG, QB_LOG_CONF_ENABLED, QB_TRUE); } /* Should we log to stderr */ if (pcmk__env_option_enabled(crm_system_name, "stderr")) { /* Override the default setting */ to_stderr = TRUE; } crm_enable_stderr(to_stderr); /* Should we log to a file */ if (safe_str_eq("none", logfile)) { /* No soup^Hlogs for you! */ } else if(crm_is_daemon) { // Daemons always get a log file, unless explicitly set to "none" crm_add_logfile(logfile); } else if(logfile) { crm_add_logfile(logfile); } if (crm_is_daemon && pcmk__env_option_enabled(crm_system_name, "blackbox")) { crm_enable_blackbox(0); } /* Summary */ crm_trace("Quiet: %d, facility %s", quiet, f_copy); pcmk__env_option("logfile"); pcmk__env_option("logfacility"); crm_update_callsites(); /* Ok, now we can start logging... */ if (quiet == FALSE && crm_is_daemon == FALSE) { crm_log_args(argc, argv); } if (crm_is_daemon) { const char *user = getenv("USER"); - if (user != NULL && pcmk__str_none_of(user, "root", CRM_DAEMON_USER, NULL)) { + if (user != NULL && !pcmk__str_any_of(user, "root", CRM_DAEMON_USER, NULL)) { crm_trace("Not switching to corefile directory for %s", user); crm_is_daemon = FALSE; } } if (crm_is_daemon) { int user = getuid(); const char *base = CRM_CORE_DIR; struct passwd *pwent = getpwuid(user); if (pwent == NULL) { crm_perror(LOG_ERR, "Cannot get name for uid: %d", user); - } else if (pcmk__str_none_of(pwent->pw_name, "root", CRM_DAEMON_USER, NULL)) { + } else if (!pcmk__str_any_of(pwent->pw_name, "root", CRM_DAEMON_USER, NULL)) { crm_trace("Don't change active directory for regular user: %s", pwent->pw_name); } else if (chdir(base) < 0) { crm_perror(LOG_INFO, "Cannot change active directory to %s", base); } else { crm_info("Changed active directory to %s", base); #if 0 { char path[512]; snprintf(path, 512, "%s-%lu", crm_system_name, (unsigned long) getpid()); mkdir(path, 0750); chdir(path); crm_info("Changed active directory to %s/%s/%s", base, pwent->pw_name, path); } #endif } /* Original meanings from signal(7) * * Signal Value Action Comment * SIGTRAP 5 Core Trace/breakpoint trap * SIGUSR1 30,10,16 Term User-defined signal 1 * SIGUSR2 31,12,17 Term User-defined signal 2 * * Our usage is as similar as possible */ mainloop_add_signal(SIGUSR1, crm_enable_blackbox); mainloop_add_signal(SIGUSR2, crm_disable_blackbox); mainloop_add_signal(SIGTRAP, crm_trigger_blackbox); } return TRUE; } /* returns the old value */ unsigned int set_crm_log_level(unsigned int level) { unsigned int old = crm_log_level; if (level > LOG_TRACE) { level = LOG_TRACE; } crm_log_level = level; crm_update_callsites(); crm_trace("New log level: %d", level); return old; } void crm_enable_stderr(int enable) { if (enable && qb_log_ctl(QB_LOG_STDERR, QB_LOG_CONF_STATE_GET, 0) != QB_LOG_STATE_ENABLED) { qb_log_ctl(QB_LOG_STDERR, QB_LOG_CONF_ENABLED, QB_TRUE); crm_update_callsites(); } else if (enable == FALSE) { qb_log_ctl(QB_LOG_STDERR, QB_LOG_CONF_ENABLED, QB_FALSE); } } void crm_bump_log_level(int argc, char **argv) { static int args = TRUE; int level = crm_log_level; if (args && argc > 1) { crm_log_args(argc, argv); } if (qb_log_ctl(QB_LOG_STDERR, QB_LOG_CONF_STATE_GET, 0) == QB_LOG_STATE_ENABLED) { set_crm_log_level(level + 1); } /* Enable after potentially logging the argstring, not before */ crm_enable_stderr(TRUE); } unsigned int get_crm_log_level(void) { return crm_log_level; } #define ARGS_FMT "Invoked: %s" void crm_log_args(int argc, char **argv) { int lpc = 0; int len = 0; int existing_len = 0; int line = __LINE__; static int logged = 0; char *arg_string = NULL; if (argc == 0 || argv == NULL || logged) { return; } logged = 1; // cppcheck seems not to understand the abort logic in realloc_safe // cppcheck-suppress memleak for (; lpc < argc; lpc++) { if (argv[lpc] == NULL) { break; } len = 2 + strlen(argv[lpc]); /* +1 space, +1 EOS */ arg_string = realloc_safe(arg_string, len + existing_len); existing_len += sprintf(arg_string + existing_len, "%s ", argv[lpc]); } qb_log_from_external_source(__func__, __FILE__, ARGS_FMT, LOG_NOTICE, line, 0, arg_string); free(arg_string); } void crm_log_output_fn(const char *file, const char *function, int line, int level, const char *prefix, const char *output) { const char *next = NULL; const char *offset = NULL; if (level == LOG_NEVER) { return; } if (output == NULL) { if (level != LOG_STDOUT) { level = LOG_TRACE; } output = "-- empty --"; } next = output; do { offset = next; next = strchrnul(offset, '\n'); do_crm_log_alias(level, file, function, line, "%s [ %.*s ]", prefix, (int)(next - offset), offset); if (next[0] != 0) { next++; } } while (next != NULL && next[0] != 0); } diff --git a/lib/common/strings.c b/lib/common/strings.c index bd68ccf69f..aaea3c9700 100644 --- a/lib/common/strings.c +++ b/lib/common/strings.c @@ -1,782 +1,759 @@ /* * Copyright 2004-2020 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #include #ifndef _GNU_SOURCE # define _GNU_SOURCE #endif #include #include #include #include #include #include #include char * crm_itoa_stack(int an_int, char *buffer, size_t len) { if (buffer != NULL) { snprintf(buffer, len, "%d", an_int); } return buffer; } /*! * \internal * \brief Scan a long long integer from a string * * \param[in] text String to scan * \param[out] result If not NULL, where to store scanned value * \param[out] end_text If not NULL, where to store pointer to just after value * * \return Standard Pacemaker return code (also set errno on error) */ static int scan_ll(const char *text, long long *result, char **end_text) { long long local_result = -1; char *local_end_text = NULL; int rc = pcmk_rc_ok; errno = 0; if (text != NULL) { #ifdef ANSI_ONLY local_result = (long long) strtol(text, &local_end_text, 10); #else local_result = strtoll(text, &local_end_text, 10); #endif if (errno == ERANGE) { rc = errno; crm_warn("Integer parsed from %s was clipped to %lld", text, local_result); } else if (errno != 0) { rc = errno; local_result = -1; crm_err("Could not parse integer from %s (using -1 instead): %s", text, pcmk_rc_str(rc)); } else if (local_end_text == text) { rc = EINVAL; local_result = -1; crm_err("Could not parse integer from %s (using -1 instead): " "No digits found", text); } if ((end_text == NULL) && (local_end_text != NULL) && (local_end_text[0] != '\0')) { crm_warn("Characters left over after parsing '%s': '%s'", text, local_end_text); } errno = rc; } if (end_text != NULL) { *end_text = local_end_text; } if (result != NULL) { *result = local_result; } return rc; } /*! * \brief Parse a long long integer value from a string * * \param[in] text The string to parse * \param[in] default_text Default string to parse if text is NULL * * \return Parsed value on success, -1 (and set errno) on error */ long long crm_parse_ll(const char *text, const char *default_text) { long long result; if (text == NULL) { text = default_text; if (text == NULL) { crm_err("No default conversion value supplied"); errno = EINVAL; return -1; } } scan_ll(text, &result, NULL); return result; } /*! * \brief Parse an integer value from a string * * \param[in] text The string to parse * \param[in] default_text Default string to parse if text is NULL * * \return Parsed value on success, INT_MIN or INT_MAX (and set errno to ERANGE) * if parsed value is out of integer range, otherwise -1 (and set errno) */ int crm_parse_int(const char *text, const char *default_text) { long long result = crm_parse_ll(text, default_text); if (result < INT_MIN) { // If errno is ERANGE, crm_parse_ll() has already logged a message if (errno != ERANGE) { crm_err("Conversion of %s was clipped: %lld", text, result); errno = ERANGE; } return INT_MIN; } else if (result > INT_MAX) { // If errno is ERANGE, crm_parse_ll() has already logged a message if (errno != ERANGE) { crm_err("Conversion of %s was clipped: %lld", text, result); errno = ERANGE; } return INT_MAX; } return (int) result; } /*! * \internal * \brief Parse a guint from a string stored in a hash table * * \param[in] table Hash table to search * \param[in] key Hash table key to use to retrieve string * \param[in] default_val What to use if key has no entry in table * \param[out] result If not NULL, where to store parsed integer * * \return Standard Pacemaker return code */ int pcmk__guint_from_hash(GHashTable *table, const char *key, guint default_val, guint *result) { const char *value; long long value_ll; CRM_CHECK((table != NULL) && (key != NULL), return EINVAL); value = g_hash_table_lookup(table, key); if (value == NULL) { if (result != NULL) { *result = default_val; } return pcmk_rc_ok; } errno = 0; value_ll = crm_parse_ll(value, NULL); if (errno != 0) { return errno; // Message already logged } if ((value_ll < 0) || (value_ll > G_MAXUINT)) { crm_warn("Could not parse non-negative integer from %s", value); return ERANGE; } if (result != NULL) { *result = (guint) value_ll; } return pcmk_rc_ok; } #ifndef NUMCHARS # define NUMCHARS "0123456789." #endif #ifndef WHITESPACE # define WHITESPACE " \t\n\r\f" #endif /*! * \brief Parse a time+units string and return milliseconds equivalent * * \param[in] input String with a number and units (optionally with whitespace * before and/or after the number) * * \return Milliseconds corresponding to string expression, or -1 on error */ long long crm_get_msec(const char *input) { const char *num_start = NULL; const char *units; long long multiplier = 1000; long long divisor = 1; long long msec = -1; size_t num_len = 0; char *end_text = NULL; if (input == NULL) { return -1; } num_start = input + strspn(input, WHITESPACE); num_len = strspn(num_start, NUMCHARS); if (num_len < 1) { return -1; } units = num_start + num_len; units += strspn(units, WHITESPACE); if (!strncasecmp(units, "ms", 2) || !strncasecmp(units, "msec", 4)) { multiplier = 1; divisor = 1; } else if (!strncasecmp(units, "us", 2) || !strncasecmp(units, "usec", 4)) { multiplier = 1; divisor = 1000; } else if (!strncasecmp(units, "s", 1) || !strncasecmp(units, "sec", 3)) { multiplier = 1000; divisor = 1; } else if (!strncasecmp(units, "m", 1) || !strncasecmp(units, "min", 3)) { multiplier = 60 * 1000; divisor = 1; } else if (!strncasecmp(units, "h", 1) || !strncasecmp(units, "hr", 2)) { multiplier = 60 * 60 * 1000; divisor = 1; } else if ((*units != EOS) && (*units != '\n') && (*units != '\r')) { return -1; } scan_ll(num_start, &msec, &end_text); if (msec > (LLONG_MAX / multiplier)) { // Arithmetics overflow while multiplier/divisor mutually exclusive return LLONG_MAX; } msec *= multiplier; msec /= divisor; return msec; } gboolean safe_str_neq(const char *a, const char *b) { if (a == b) { return FALSE; } else if (a == NULL || b == NULL) { return TRUE; } else if (strcasecmp(a, b) == 0) { return FALSE; } return TRUE; } gboolean crm_is_true(const char *s) { gboolean ret = FALSE; if (s != NULL) { crm_str_to_boolean(s, &ret); } return ret; } int crm_str_to_boolean(const char *s, int *ret) { if (s == NULL) { return -1; } else if (strcasecmp(s, "true") == 0 || strcasecmp(s, "on") == 0 || strcasecmp(s, "yes") == 0 || strcasecmp(s, "y") == 0 || strcasecmp(s, "1") == 0) { *ret = TRUE; return 1; } else if (strcasecmp(s, "false") == 0 || strcasecmp(s, "off") == 0 || strcasecmp(s, "no") == 0 || strcasecmp(s, "n") == 0 || strcasecmp(s, "0") == 0) { *ret = FALSE; return 1; } return -1; } char * crm_strip_trailing_newline(char *str) { int len; if (str == NULL) { return str; } for (len = strlen(str) - 1; len >= 0 && str[len] == '\n'; len--) { str[len] = '\0'; } return str; } gboolean crm_str_eq(const char *a, const char *b, gboolean use_case) { if (use_case) { return g_strcmp0(a, b) == 0; /* TODO - Figure out which calls, if any, really need to be case independent */ } else if (a == b) { return TRUE; } else if (a == NULL || b == NULL) { /* shouldn't be comparing NULLs */ return FALSE; } else if (strcasecmp(a, b) == 0) { return TRUE; } return FALSE; } /*! * \brief Check whether a string starts with a certain sequence * * \param[in] str String to check * \param[in] prefix Sequence to match against beginning of \p str * * \return \c true if \p str begins with match, \c false otherwise * \note This is equivalent to !strncmp(s, prefix, strlen(prefix)) * but is likely less efficient when prefix is a string literal * if the compiler optimizes away the strlen() at compile time, * and more efficient otherwise. */ bool pcmk__starts_with(const char *str, const char *prefix) { const char *s = str; const char *p = prefix; if (!s || !p) { return false; } while (*s && *p) { if (*s++ != *p++) { return false; } } return (*p == 0); } static inline bool ends_with(const char *s, const char *match, bool as_extension) { if (pcmk__str_empty(match)) { return true; } else if (s == NULL) { return false; } else { size_t slen, mlen; /* Besides as_extension, we could also check !strchr(&match[1], match[0]) but that would be inefficient. */ if (as_extension) { s = strrchr(s, match[0]); return (s == NULL)? false : !strcmp(s, match); } mlen = strlen(match); slen = strlen(s); return ((slen >= mlen) && !strcmp(s + slen - mlen, match)); } } /*! * \internal * \brief Check whether a string ends with a certain sequence * * \param[in] s String to check * \param[in] match Sequence to match against end of \p s * * \return \c true if \p s ends case-sensitively with match, \c false otherwise * \note pcmk__ends_with_ext() can be used if the first character of match * does not recur in match. */ bool pcmk__ends_with(const char *s, const char *match) { return ends_with(s, match, false); } /*! * \internal * \brief Check whether a string ends with a certain "extension" * * \param[in] s String to check * \param[in] match Extension to match against end of \p s, that is, * its first character must not occur anywhere * in the rest of that very sequence (example: file * extension where the last dot is its delimiter, * e.g., ".html"); incorrect results may be * returned otherwise. * * \return \c true if \p s ends (verbatim, i.e., case sensitively) * with "extension" designated as \p match (including empty * string), \c false otherwise * * \note Main incentive to prefer this function over \c pcmk__ends_with() * where possible is the efficiency (at the cost of added * restriction on \p match as stated; the complexity class * remains the same, though: BigO(M+N) vs. BigO(M+2N)). */ bool pcmk__ends_with_ext(const char *s, const char *match) { return ends_with(s, match, true); } /* * This re-implements g_str_hash as it was prior to glib2-2.28: * * https://gitlab.gnome.org/GNOME/glib/commit/354d655ba8a54b754cb5a3efb42767327775696c * * Note that the new g_str_hash is presumably a *better* hash (it's actually * a correct implementation of DJB's hash), but we need to preserve existing * behaviour, because the hash key ultimately determines the "sort" order * when iterating through GHashTables, which affects allocation of scores to * clone instances when iterating through rsc->allowed_nodes. It (somehow) * also appears to have some minor impact on the ordering of a few * pseudo_event IDs in the transition graph. */ guint g_str_hash_traditional(gconstpointer v) { const signed char *p; guint32 h = 0; for (p = v; *p != '\0'; p++) h = (h << 5) - h + *p; return h; } /* used with hash tables where case does not matter */ gboolean crm_strcase_equal(gconstpointer a, gconstpointer b) { return crm_str_eq((const char *) a, (const char *) b, FALSE); } guint crm_strcase_hash(gconstpointer v) { const signed char *p; guint32 h = 0; for (p = v; *p != '\0'; p++) h = (h << 5) - h + g_ascii_tolower(*p); return h; } static void copy_str_table_entry(gpointer key, gpointer value, gpointer user_data) { if (key && value && user_data) { g_hash_table_insert((GHashTable*)user_data, strdup(key), strdup(value)); } } GHashTable * crm_str_table_dup(GHashTable *old_table) { GHashTable *new_table = NULL; if (old_table) { new_table = crm_str_table_new(); g_hash_table_foreach(old_table, copy_str_table_entry, new_table); } return new_table; } /*! * \internal * \brief Add a word to a space-separated string list * * \param[in,out] list Pointer to beginning of list * \param[in] word Word to add to list * * \return (Potentially new) beginning of list * \note This dynamically reallocates list as needed. */ char * pcmk__add_word(char *list, const char *word) { if (word != NULL) { size_t len = list? strlen(list) : 0; list = realloc_safe(list, len + strlen(word) + 2); // 2 = space + EOS sprintf(list + len, " %s", word); } return list; } /*! * \internal * \brief Compress data * * \param[in] data Data to compress * \param[in] length Number of characters of data to compress * \param[in] max Maximum size of compressed data (or 0 to estimate) * \param[out] result Where to store newly allocated compressed result * \param[out] result_len Where to store actual compressed length of result * * \return Standard Pacemaker return code */ int pcmk__compress(const char *data, unsigned int length, unsigned int max, char **result, unsigned int *result_len) { int rc; char *compressed = NULL; char *uncompressed = strdup(data); #ifdef CLOCK_MONOTONIC struct timespec after_t; struct timespec before_t; #endif if (max == 0) { max = (length * 1.01) + 601; // Size guaranteed to hold result } #ifdef CLOCK_MONOTONIC clock_gettime(CLOCK_MONOTONIC, &before_t); #endif compressed = calloc((size_t) max, sizeof(char)); CRM_ASSERT(compressed); *result_len = max; rc = BZ2_bzBuffToBuffCompress(compressed, result_len, uncompressed, length, CRM_BZ2_BLOCKS, 0, CRM_BZ2_WORK); free(uncompressed); if (rc != BZ_OK) { crm_err("Compression of %d bytes failed: %s " CRM_XS " bzerror=%d", length, bz2_strerror(rc), rc); free(compressed); return pcmk_rc_error; } #ifdef CLOCK_MONOTONIC clock_gettime(CLOCK_MONOTONIC, &after_t); crm_trace("Compressed %d bytes into %d (ratio %d:1) in %.0fms", length, *result_len, length / (*result_len), (after_t.tv_sec - before_t.tv_sec) * 1000 + (after_t.tv_nsec - before_t.tv_nsec) / 1e6); #else crm_trace("Compressed %d bytes into %d (ratio %d:1)", length, *result_len, length / (*result_len)); #endif *result = compressed; return pcmk_rc_ok; } char * crm_strdup_printf(char const *format, ...) { va_list ap; int len = 0; char *string = NULL; va_start(ap, format); len = vasprintf (&string, format, ap); CRM_ASSERT(len > 0); va_end(ap); return string; } int pcmk__parse_ll_range(const char *srcstring, long long *start, long long *end) { char *remainder = NULL; CRM_ASSERT(start != NULL && end != NULL); *start = -1; *end = -1; crm_trace("Attempting to decode: [%s]", srcstring); if (srcstring == NULL || strcmp(srcstring, "") == 0 || strcmp(srcstring, "-") == 0) { return pcmk_rc_unknown_format; } /* String starts with a dash, so this is either a range with * no beginning or garbage. * */ if (*srcstring == '-') { int rc = scan_ll(srcstring+1, end, &remainder); if (rc != pcmk_rc_ok || *remainder != '\0') { return pcmk_rc_unknown_format; } else { return pcmk_rc_ok; } } if (scan_ll(srcstring, start, &remainder) != pcmk_rc_ok) { return pcmk_rc_unknown_format; } if (*remainder && *remainder == '-') { if (*(remainder+1)) { char *more_remainder = NULL; int rc = scan_ll(remainder+1, end, &more_remainder); if (rc != pcmk_rc_ok || *more_remainder != '\0') { return pcmk_rc_unknown_format; } } } else if (*remainder && *remainder != '-') { *start = -1; return pcmk_rc_unknown_format; } else { /* The input string contained only one number. Set start and end * to the same value and return pcmk_rc_ok. This gives the caller * a way to tell this condition apart from a range with no end. */ *end = *start; } return pcmk_rc_ok; } gboolean pcmk__str_in_list(GList *lst, const gchar *s) { if (lst == NULL) { return FALSE; } if (strcmp(lst->data, "*") == 0 && lst->next == NULL) { return TRUE; } return g_list_find_custom(lst, s, (GCompareFunc) strcmp) != NULL; } bool pcmk__str_any_of(const char *s, ...) { bool rc = false; va_list ap; va_start(ap, s); while (1) { const char *ele = va_arg(ap, const char *); if (ele == NULL) { break; } else if (crm_str_eq(s, ele, FALSE)) { rc = true; break; } } va_end(ap); return rc; } -bool -pcmk__str_none_of(const char *s, ...) -{ - bool rc = true; - va_list ap; - - va_start(ap, s); - - while (1) { - const char *ele = va_arg(ap, const char *); - - if (ele == NULL) { - break; - } else if (crm_str_eq(s, ele, FALSE)) { - rc = false; - break; - } - } - - va_end(ap); - return rc; -} - /* * \brief Sort strings, with numeric portions sorted numerically * * Sort two strings case-insensitively like strcasecmp(), but with any numeric * portions of the string sorted numerically. This is particularly useful for * node names (for example, "node10" will sort higher than "node9" but lower * than "remotenode9"). * * \param[in] s1 First string to compare (must not be NULL) * \param[in] s2 Second string to compare (must not be NULL) * * \retval -1 \p s1 comes before \p s2 * \retval 0 \p s1 and \p s2 are equal * \retval 1 \p s1 comes after \p s2 */ int pcmk_numeric_strcasecmp(const char *s1, const char *s2) { while (*s1 && *s2) { if (isdigit(*s1) && isdigit(*s2)) { // If node names contain a number, sort numerically char *end1 = NULL; char *end2 = NULL; long num1 = strtol(s1, &end1, 10); long num2 = strtol(s2, &end2, 10); // allow ordering e.g. 007 > 7 size_t len1 = end1 - s1; size_t len2 = end2 - s2; if (num1 < num2) { return -1; } else if (num1 > num2) { return 1; } else if (len1 < len2) { return -1; } else if (len1 > len2) { return 1; } s1 = end1; s2 = end2; } else { // Compare non-digits case-insensitively int lower1 = tolower(*s1); int lower2 = tolower(*s2); if (lower1 < lower2) { return -1; } else if (lower1 > lower2) { return 1; } ++s1; ++s2; } } if (!*s1 && *s2) { return -1; } else if (*s1 && !*s2) { return 1; } return 0; } diff --git a/lib/common/tests/strings/Makefile.am b/lib/common/tests/strings/Makefile.am index 047537b227..859559f1f3 100644 --- a/lib/common/tests/strings/Makefile.am +++ b/lib/common/tests/strings/Makefile.am @@ -1,21 +1,20 @@ AM_CPPFLAGS = -I$(top_srcdir)/include -I$(top_builddir)/include LDADD = $(top_builddir)/lib/common/libcrmcommon.la include $(top_srcdir)/mk/glib-tap.mk # Add each test program here. Each test should be written as a little standalone # program using the glib unit testing functions. See the documentation for more # information. # # https://developer.gnome.org/glib/unstable/glib-Testing.html test_programs = pcmk__parse_ll_range \ - pcmk__str_any_of \ - pcmk__str_none_of + pcmk__str_any_of # If any extra data needs to be added to the source distribution, add it to the # following list. dist_test_data = # If any extra data needs to be used by tests but should not be added to the # source distribution, add it to the following list. test_data = diff --git a/lib/common/tests/strings/pcmk__str_none_of.c b/lib/common/tests/strings/pcmk__str_none_of.c deleted file mode 100644 index 5073e8dabf..0000000000 --- a/lib/common/tests/strings/pcmk__str_none_of.c +++ /dev/null @@ -1,37 +0,0 @@ -#include - -#include - -static void -empty_input_list(void) { - g_assert(pcmk__str_none_of("xxx", NULL) == true); - g_assert(pcmk__str_none_of("", NULL) == true); -} - -static void -empty_string(void) { - g_assert(pcmk__str_none_of("", "xxx", "yyy", NULL) == true); - g_assert(pcmk__str_none_of(NULL, "xxx", "yyy", NULL) == true); -} - -static void -in_list(void) { - g_assert(pcmk__str_none_of("xxx", "aaa", "bbb", "xxx", NULL) == false); - g_assert(pcmk__str_none_of("XXX", "aaa", "bbb", "xxx", NULL) == false); -} - -static void -not_in_list(void) { - g_assert(pcmk__str_none_of("xxx", "aaa", "bbb", NULL) == true); -} - -int main(int argc, char **argv) { - g_test_init(&argc, &argv, NULL); - - g_test_add_func("/common/strings/none_of/empty_list", empty_input_list); - g_test_add_func("/common/strings/none_of/empty_string", empty_string); - g_test_add_func("/common/strings/none_of/in", in_list); - g_test_add_func("/common/strings/none_of/not_in", not_in_list); - - return g_test_run(); -} diff --git a/lib/pacemaker/pcmk_sched_transition.c b/lib/pacemaker/pcmk_sched_transition.c index 8fa232d139..7edcc197ec 100644 --- a/lib/pacemaker/pcmk_sched_transition.c +++ b/lib/pacemaker/pcmk_sched_transition.c @@ -1,858 +1,858 @@ /* * Copyright 2009-2020 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #include #include // lrmd_event_data_t, lrmd_free_event() #include #include #include #include #include static bool fake_quiet = FALSE; static cib_t *fake_cib = NULL; static GListPtr fake_resource_list = NULL; static GListPtr fake_op_fail_list = NULL; gboolean bringing_nodes_online = FALSE; #define STATUS_PATH_MAX 512 #define quiet_log(fmt, args...) do { \ if(fake_quiet) { \ crm_trace(fmt, ##args); \ } else { \ printf(fmt , ##args); \ } \ } while(0) #define NEW_NODE_TEMPLATE "//"XML_CIB_TAG_NODE"[@uname='%s']" #define NODE_TEMPLATE "//"XML_CIB_TAG_STATE"[@uname='%s']" #define RSC_TEMPLATE "//"XML_CIB_TAG_STATE"[@uname='%s']//"XML_LRM_TAG_RESOURCE"[@id='%s']" static void inject_transient_attr(xmlNode * cib_node, const char *name, const char *value) { xmlNode *attrs = NULL; xmlNode *instance_attrs = NULL; xmlChar *node_path; const char *node_uuid = ID(cib_node); node_path = xmlGetNodePath(cib_node); quiet_log(" + Injecting attribute %s=%s into %s '%s'\n", name, value, node_path, ID(cib_node)); free(node_path); attrs = first_named_child(cib_node, XML_TAG_TRANSIENT_NODEATTRS); if (attrs == NULL) { attrs = create_xml_node(cib_node, XML_TAG_TRANSIENT_NODEATTRS); crm_xml_add(attrs, XML_ATTR_ID, node_uuid); } instance_attrs = first_named_child(attrs, XML_TAG_ATTR_SETS); if (instance_attrs == NULL) { instance_attrs = create_xml_node(attrs, XML_TAG_ATTR_SETS); crm_xml_add(instance_attrs, XML_ATTR_ID, node_uuid); } crm_create_nvpair_xml(instance_attrs, NULL, name, value); } static void update_failcounts(xmlNode * cib_node, const char *resource, const char *task, guint interval_ms, int rc) { if (rc == 0) { return; } else if ((rc == 7) && (interval_ms == 0)) { return; } else { char *name = NULL; char *now = crm_ttoa(time(NULL)); name = pcmk__failcount_name(resource, task, interval_ms); inject_transient_attr(cib_node, name, "value++"); free(name); name = pcmk__lastfailure_name(resource, task, interval_ms); inject_transient_attr(cib_node, name, now); free(name); free(now); } } static void create_node_entry(cib_t * cib_conn, const char *node) { int rc = pcmk_ok; char *xpath = crm_strdup_printf(NEW_NODE_TEMPLATE, node); rc = cib_conn->cmds->query(cib_conn, xpath, NULL, cib_xpath | cib_sync_call | cib_scope_local); if (rc == -ENXIO) { xmlNode *cib_object = create_xml_node(NULL, XML_CIB_TAG_NODE); crm_xml_add(cib_object, XML_ATTR_ID, node); // Use node name as ID crm_xml_add(cib_object, XML_ATTR_UNAME, node); cib_conn->cmds->create(cib_conn, XML_CIB_TAG_NODES, cib_object, cib_sync_call | cib_scope_local); /* Not bothering with subsequent query to see if it exists, we'll bomb out later in the call to query_node_uuid()... */ free_xml(cib_object); } free(xpath); } static lrmd_event_data_t * create_op(xmlNode *cib_resource, const char *task, guint interval_ms, int outcome) { lrmd_event_data_t *op = NULL; xmlNode *xop = NULL; op = lrmd_new_event(ID(cib_resource), task, interval_ms); op->rc = outcome; op->op_status = 0; op->params = NULL; /* TODO: Fill me in */ op->t_run = (unsigned int) time(NULL); op->t_rcchange = op->t_run; op->call_id = 0; for (xop = __xml_first_child_element(cib_resource); xop != NULL; xop = __xml_next_element(xop)) { int tmp = 0; crm_element_value_int(xop, XML_LRM_ATTR_CALLID, &tmp); if (tmp > op->call_id) { op->call_id = tmp; } } op->call_id++; return op; } static xmlNode * inject_op(xmlNode * cib_resource, lrmd_event_data_t * op, int target_rc) { return pcmk__create_history_xml(cib_resource, op, CRM_FEATURE_SET, target_rc, NULL, crm_system_name, LOG_TRACE); } static xmlNode * inject_node_state(cib_t * cib_conn, const char *node, const char *uuid) { int rc = pcmk_ok; xmlNode *cib_object = NULL; char *xpath = crm_strdup_printf(NODE_TEMPLATE, node); if (bringing_nodes_online) { create_node_entry(cib_conn, node); } rc = cib_conn->cmds->query(cib_conn, xpath, &cib_object, cib_xpath | cib_sync_call | cib_scope_local); if (cib_object && ID(cib_object) == NULL) { crm_err("Detected multiple node_state entries for xpath=%s, bailing", xpath); crm_log_xml_warn(cib_object, "Duplicates"); free(xpath); crm_exit(CRM_EX_SOFTWARE); return NULL; // not reached, but makes static analysis happy } if (rc == -ENXIO) { char *found_uuid = NULL; if (uuid == NULL) { query_node_uuid(cib_conn, node, &found_uuid, NULL); } else { found_uuid = strdup(uuid); } cib_object = create_xml_node(NULL, XML_CIB_TAG_STATE); crm_xml_add(cib_object, XML_ATTR_UUID, found_uuid); crm_xml_add(cib_object, XML_ATTR_UNAME, node); cib_conn->cmds->create(cib_conn, XML_CIB_TAG_STATUS, cib_object, cib_sync_call | cib_scope_local); free_xml(cib_object); free(found_uuid); rc = cib_conn->cmds->query(cib_conn, xpath, &cib_object, cib_xpath | cib_sync_call | cib_scope_local); crm_trace("injecting node state for %s. rc is %d", node, rc); } free(xpath); CRM_ASSERT(rc == pcmk_ok); return cib_object; } static xmlNode * modify_node(cib_t * cib_conn, char *node, gboolean up) { xmlNode *cib_node = inject_node_state(cib_conn, node, NULL); if (up) { crm_xml_add(cib_node, XML_NODE_IN_CLUSTER, XML_BOOLEAN_YES); crm_xml_add(cib_node, XML_NODE_IS_PEER, ONLINESTATUS); crm_xml_add(cib_node, XML_NODE_JOIN_STATE, CRMD_JOINSTATE_MEMBER); crm_xml_add(cib_node, XML_NODE_EXPECTED, CRMD_JOINSTATE_MEMBER); } else { crm_xml_add(cib_node, XML_NODE_IN_CLUSTER, XML_BOOLEAN_NO); crm_xml_add(cib_node, XML_NODE_IS_PEER, OFFLINESTATUS); crm_xml_add(cib_node, XML_NODE_JOIN_STATE, CRMD_JOINSTATE_DOWN); crm_xml_add(cib_node, XML_NODE_EXPECTED, CRMD_JOINSTATE_DOWN); } crm_xml_add(cib_node, XML_ATTR_ORIGIN, crm_system_name); return cib_node; } static xmlNode * find_resource_xml(xmlNode * cib_node, const char *resource) { xmlNode *match = NULL; const char *node = crm_element_value(cib_node, XML_ATTR_UNAME); char *xpath = crm_strdup_printf(RSC_TEMPLATE, node, resource); match = get_xpath_object(xpath, cib_node, LOG_TRACE); free(xpath); return match; } static xmlNode * inject_resource(xmlNode * cib_node, const char *resource, const char *lrm_name, const char *rclass, const char *rtype, const char *rprovider) { xmlNode *lrm = NULL; xmlNode *container = NULL; xmlNode *cib_resource = NULL; char *xpath = NULL; cib_resource = find_resource_xml(cib_node, resource); if (cib_resource != NULL) { /* If an existing LRM history entry uses the resource name, * continue using it, even if lrm_name is different. */ return cib_resource; } // Check for history entry under preferred name if (strcmp(resource, lrm_name)) { cib_resource = find_resource_xml(cib_node, lrm_name); if (cib_resource != NULL) { return cib_resource; } } /* One day, add query for class, provider, type */ if (rclass == NULL || rtype == NULL) { fprintf(stderr, "Resource %s not found in the status section of %s." " Please supply the class and type to continue\n", resource, ID(cib_node)); return NULL; - } else if (pcmk__str_none_of(rclass, PCMK_RESOURCE_CLASS_OCF, PCMK_RESOURCE_CLASS_STONITH, + } else if (!pcmk__str_any_of(rclass, PCMK_RESOURCE_CLASS_OCF, PCMK_RESOURCE_CLASS_STONITH, PCMK_RESOURCE_CLASS_SERVICE, PCMK_RESOURCE_CLASS_UPSTART, PCMK_RESOURCE_CLASS_SYSTEMD, PCMK_RESOURCE_CLASS_LSB, NULL)) { fprintf(stderr, "Invalid class for %s: %s\n", resource, rclass); return NULL; } else if (is_set(pcmk_get_ra_caps(rclass), pcmk_ra_cap_provider) && (rprovider == NULL)) { fprintf(stderr, "Please specify the provider for resource %s\n", resource); return NULL; } xpath = (char *)xmlGetNodePath(cib_node); crm_info("Injecting new resource %s into %s '%s'", lrm_name, xpath, ID(cib_node)); free(xpath); lrm = first_named_child(cib_node, XML_CIB_TAG_LRM); if (lrm == NULL) { const char *node_uuid = ID(cib_node); lrm = create_xml_node(cib_node, XML_CIB_TAG_LRM); crm_xml_add(lrm, XML_ATTR_ID, node_uuid); } container = first_named_child(lrm, XML_LRM_TAG_RESOURCES); if (container == NULL) { container = create_xml_node(lrm, XML_LRM_TAG_RESOURCES); } cib_resource = create_xml_node(container, XML_LRM_TAG_RESOURCE); // If we're creating a new entry, use the preferred name crm_xml_add(cib_resource, XML_ATTR_ID, lrm_name); crm_xml_add(cib_resource, XML_AGENT_ATTR_CLASS, rclass); crm_xml_add(cib_resource, XML_AGENT_ATTR_PROVIDER, rprovider); crm_xml_add(cib_resource, XML_ATTR_TYPE, rtype); return cib_resource; } #define XPATH_MAX 1024 static int find_ticket_state(cib_t * the_cib, const char *ticket_id, xmlNode ** ticket_state_xml) { int offset = 0; int rc = pcmk_ok; xmlNode *xml_search = NULL; char *xpath_string = NULL; CRM_ASSERT(ticket_state_xml != NULL); *ticket_state_xml = NULL; xpath_string = calloc(1, XPATH_MAX); offset += snprintf(xpath_string + offset, XPATH_MAX - offset, "%s", "/cib/status/tickets"); if (ticket_id) { offset += snprintf(xpath_string + offset, XPATH_MAX - offset, "/%s[@id=\"%s\"]", XML_CIB_TAG_TICKET_STATE, ticket_id); } CRM_LOG_ASSERT(offset > 0); rc = the_cib->cmds->query(the_cib, xpath_string, &xml_search, cib_sync_call | cib_scope_local | cib_xpath); if (rc != pcmk_ok) { goto bail; } crm_log_xml_debug(xml_search, "Match"); if (xml_has_children(xml_search)) { if (ticket_id) { fprintf(stdout, "Multiple ticket_states match ticket_id=%s\n", ticket_id); } *ticket_state_xml = xml_search; } else { *ticket_state_xml = xml_search; } bail: free(xpath_string); return rc; } static int set_ticket_state_attr(const char *ticket_id, const char *attr_name, const char *attr_value, cib_t * cib, int cib_options) { int rc = pcmk_ok; xmlNode *xml_top = NULL; xmlNode *ticket_state_xml = NULL; rc = find_ticket_state(cib, ticket_id, &ticket_state_xml); if (rc == pcmk_ok) { crm_debug("Found a match state for ticket: id=%s", ticket_id); xml_top = ticket_state_xml; } else if (rc != -ENXIO) { return rc; } else { xmlNode *xml_obj = NULL; xml_top = create_xml_node(NULL, XML_CIB_TAG_STATUS); xml_obj = create_xml_node(xml_top, XML_CIB_TAG_TICKETS); ticket_state_xml = create_xml_node(xml_obj, XML_CIB_TAG_TICKET_STATE); crm_xml_add(ticket_state_xml, XML_ATTR_ID, ticket_id); } crm_xml_add(ticket_state_xml, attr_name, attr_value); crm_log_xml_debug(xml_top, "Update"); rc = cib->cmds->modify(cib, XML_CIB_TAG_STATUS, xml_top, cib_options); free_xml(xml_top); return rc; } void modify_configuration(pe_working_set_t * data_set, cib_t *cib, const char *quorum, const char *watchdog, GListPtr node_up, GListPtr node_down, GListPtr node_fail, GListPtr op_inject, GListPtr ticket_grant, GListPtr ticket_revoke, GListPtr ticket_standby, GListPtr ticket_activate) { int rc = pcmk_ok; GListPtr gIter = NULL; xmlNode *cib_op = NULL; xmlNode *cib_node = NULL; xmlNode *cib_resource = NULL; lrmd_event_data_t *op = NULL; if (quorum) { xmlNode *top = create_xml_node(NULL, XML_TAG_CIB); quiet_log(" + Setting quorum: %s\n", quorum); /* crm_xml_add(top, XML_ATTR_DC_UUID, dc_uuid); */ crm_xml_add(top, XML_ATTR_HAVE_QUORUM, quorum); rc = cib->cmds->modify(cib, NULL, top, cib_sync_call | cib_scope_local); CRM_ASSERT(rc == pcmk_ok); } if (watchdog) { quiet_log(" + Setting watchdog: %s\n", watchdog); rc = update_attr_delegate(cib, cib_sync_call | cib_scope_local, XML_CIB_TAG_CRMCONFIG, NULL, NULL, NULL, NULL, XML_ATTR_HAVE_WATCHDOG, watchdog, FALSE, NULL, NULL); CRM_ASSERT(rc == pcmk_ok); } for (gIter = node_up; gIter != NULL; gIter = gIter->next) { char *node = (char *)gIter->data; quiet_log(" + Bringing node %s online\n", node); cib_node = modify_node(cib, node, TRUE); CRM_ASSERT(cib_node != NULL); rc = cib->cmds->modify(cib, XML_CIB_TAG_STATUS, cib_node, cib_sync_call | cib_scope_local); CRM_ASSERT(rc == pcmk_ok); free_xml(cib_node); } for (gIter = node_down; gIter != NULL; gIter = gIter->next) { char xpath[STATUS_PATH_MAX]; char *node = (char *)gIter->data; quiet_log(" + Taking node %s offline\n", node); cib_node = modify_node(cib, node, FALSE); CRM_ASSERT(cib_node != NULL); rc = cib->cmds->modify(cib, XML_CIB_TAG_STATUS, cib_node, cib_sync_call | cib_scope_local); CRM_ASSERT(rc == pcmk_ok); free_xml(cib_node); snprintf(xpath, STATUS_PATH_MAX, "//node_state[@uname='%s']/%s", node, XML_CIB_TAG_LRM); cib->cmds->remove(cib, xpath, NULL, cib_xpath | cib_sync_call | cib_scope_local); snprintf(xpath, STATUS_PATH_MAX, "//node_state[@uname='%s']/%s", node, XML_TAG_TRANSIENT_NODEATTRS); cib->cmds->remove(cib, xpath, NULL, cib_xpath | cib_sync_call | cib_scope_local); } for (gIter = node_fail; gIter != NULL; gIter = gIter->next) { char *node = (char *)gIter->data; quiet_log(" + Failing node %s\n", node); cib_node = modify_node(cib, node, TRUE); crm_xml_add(cib_node, XML_NODE_IN_CLUSTER, XML_BOOLEAN_NO); CRM_ASSERT(cib_node != NULL); rc = cib->cmds->modify(cib, XML_CIB_TAG_STATUS, cib_node, cib_sync_call | cib_scope_local); CRM_ASSERT(rc == pcmk_ok); free_xml(cib_node); } for (gIter = ticket_grant; gIter != NULL; gIter = gIter->next) { char *ticket_id = (char *)gIter->data; quiet_log(" + Granting ticket %s\n", ticket_id); rc = set_ticket_state_attr(ticket_id, "granted", "true", cib, cib_sync_call | cib_scope_local); CRM_ASSERT(rc == pcmk_ok); } for (gIter = ticket_revoke; gIter != NULL; gIter = gIter->next) { char *ticket_id = (char *)gIter->data; quiet_log(" + Revoking ticket %s\n", ticket_id); rc = set_ticket_state_attr(ticket_id, "granted", "false", cib, cib_sync_call | cib_scope_local); CRM_ASSERT(rc == pcmk_ok); } for (gIter = ticket_standby; gIter != NULL; gIter = gIter->next) { char *ticket_id = (char *)gIter->data; quiet_log(" + Making ticket %s standby\n", ticket_id); rc = set_ticket_state_attr(ticket_id, "standby", "true", cib, cib_sync_call | cib_scope_local); CRM_ASSERT(rc == pcmk_ok); } for (gIter = ticket_activate; gIter != NULL; gIter = gIter->next) { char *ticket_id = (char *)gIter->data; quiet_log(" + Activating ticket %s\n", ticket_id); rc = set_ticket_state_attr(ticket_id, "standby", "false", cib, cib_sync_call | cib_scope_local); CRM_ASSERT(rc == pcmk_ok); } for (gIter = op_inject; gIter != NULL; gIter = gIter->next) { char *spec = (char *)gIter->data; int rc = 0; int outcome = 0; guint interval_ms = 0; char *key = NULL; char *node = NULL; char *task = NULL; char *resource = NULL; const char *rtype = NULL; const char *rclass = NULL; const char *rprovider = NULL; pe_resource_t *rsc = NULL; quiet_log(" + Injecting %s into the configuration\n", spec); key = calloc(1, strlen(spec) + 1); node = calloc(1, strlen(spec) + 1); rc = sscanf(spec, "%[^@]@%[^=]=%d", key, node, &outcome); if (rc != 3) { fprintf(stderr, "Invalid operation spec: %s. Only found %d fields\n", spec, rc); free(key); free(node); continue; } parse_op_key(key, &resource, &task, &interval_ms); rsc = pe_find_resource(data_set->resources, resource); if (rsc == NULL) { fprintf(stderr, " - Invalid resource name: %s\n", resource); } else { rclass = crm_element_value(rsc->xml, XML_AGENT_ATTR_CLASS); rtype = crm_element_value(rsc->xml, XML_ATTR_TYPE); rprovider = crm_element_value(rsc->xml, XML_AGENT_ATTR_PROVIDER); cib_node = inject_node_state(cib, node, NULL); CRM_ASSERT(cib_node != NULL); update_failcounts(cib_node, resource, task, interval_ms, outcome); cib_resource = inject_resource(cib_node, resource, resource, rclass, rtype, rprovider); CRM_ASSERT(cib_resource != NULL); op = create_op(cib_resource, task, interval_ms, outcome); CRM_ASSERT(op != NULL); cib_op = inject_op(cib_resource, op, 0); CRM_ASSERT(cib_op != NULL); lrmd_free_event(op); rc = cib->cmds->modify(cib, XML_CIB_TAG_STATUS, cib_node, cib_sync_call | cib_scope_local); CRM_ASSERT(rc == pcmk_ok); } free(task); free(node); free(key); } } static gboolean exec_pseudo_action(crm_graph_t * graph, crm_action_t * action) { const char *node = crm_element_value(action->xml, XML_LRM_ATTR_TARGET); const char *task = crm_element_value(action->xml, XML_LRM_ATTR_TASK_KEY); action->confirmed = TRUE; quiet_log(" * Pseudo action: %s%s%s\n", task, node ? " on " : "", node ? node : ""); update_graph(graph, action); return TRUE; } static gboolean exec_rsc_action(crm_graph_t * graph, crm_action_t * action) { int rc = 0; GListPtr gIter = NULL; lrmd_event_data_t *op = NULL; int target_outcome = 0; const char *rtype = NULL; const char *rclass = NULL; const char *resource = NULL; const char *rprovider = NULL; const char *lrm_name = NULL; const char *operation = crm_element_value(action->xml, "operation"); const char *target_rc_s = crm_meta_value(action->params, XML_ATTR_TE_TARGET_RC); xmlNode *cib_node = NULL; xmlNode *cib_resource = NULL; xmlNode *action_rsc = first_named_child(action->xml, XML_CIB_TAG_RESOURCE); char *node = crm_element_value_copy(action->xml, XML_LRM_ATTR_TARGET); char *uuid = crm_element_value_copy(action->xml, XML_LRM_ATTR_TARGET_UUID); const char *router_node = crm_element_value(action->xml, XML_LRM_ATTR_ROUTER_NODE); if (pcmk__str_any_of(operation, CRM_OP_PROBED, CRM_OP_REPROBE, NULL)) { crm_info("Skipping %s op for %s", operation, node); goto done; } if (action_rsc == NULL) { crm_log_xml_err(action->xml, "Bad"); free(node); free(uuid); return FALSE; } /* Look for the preferred name * If not found, try the expected 'local' name * If not found use the preferred name anyway */ resource = crm_element_value(action_rsc, XML_ATTR_ID); CRM_ASSERT(resource != NULL); // makes static analysis happy lrm_name = resource; // Preferred name when writing history if (pe_find_resource(fake_resource_list, resource) == NULL) { const char *longname = crm_element_value(action_rsc, XML_ATTR_ID_LONG); if (longname && pe_find_resource(fake_resource_list, longname)) { resource = longname; } } if (pcmk__str_any_of(operation, "delete", RSC_METADATA, NULL)) { quiet_log(" * Resource action: %-15s %s on %s\n", resource, operation, node); goto done; } rclass = crm_element_value(action_rsc, XML_AGENT_ATTR_CLASS); rtype = crm_element_value(action_rsc, XML_ATTR_TYPE); rprovider = crm_element_value(action_rsc, XML_AGENT_ATTR_PROVIDER); if (target_rc_s != NULL) { target_outcome = crm_parse_int(target_rc_s, "0"); } CRM_ASSERT(fake_cib->cmds->query(fake_cib, NULL, NULL, cib_sync_call | cib_scope_local) == pcmk_ok); cib_node = inject_node_state(fake_cib, node, (router_node? node : uuid)); CRM_ASSERT(cib_node != NULL); cib_resource = inject_resource(cib_node, resource, lrm_name, rclass, rtype, rprovider); if (cib_resource == NULL) { crm_err("invalid resource in transition"); free(node); free(uuid); free_xml(cib_node); return FALSE; } op = convert_graph_action(cib_resource, action, 0, target_outcome); if (op->interval_ms) { quiet_log(" * Resource action: %-15s %s=%u on %s\n", resource, op->op_type, op->interval_ms, node); } else { quiet_log(" * Resource action: %-15s %s on %s\n", resource, op->op_type, node); } for (gIter = fake_op_fail_list; gIter != NULL; gIter = gIter->next) { char *spec = (char *)gIter->data; char *key = NULL; const char *match_name = NULL; // Allow user to specify anonymous clone with or without instance number key = crm_strdup_printf(PCMK__OP_FMT "@%s=", resource, op->op_type, op->interval_ms, node); if (strncasecmp(key, spec, strlen(key)) == 0) { match_name = resource; } free(key); if ((match_name == NULL) && strcmp(resource, lrm_name)) { key = crm_strdup_printf(PCMK__OP_FMT "@%s=", lrm_name, op->op_type, op->interval_ms, node); if (strncasecmp(key, spec, strlen(key)) == 0) { match_name = lrm_name; } free(key); } if (match_name != NULL) { rc = sscanf(spec, "%*[^=]=%d", (int *) &op->rc); // ${match_name}_${task}_${interval_in_ms}@${node}=${rc} if (rc != 1) { fprintf(stderr, "Invalid failed operation spec: %s. Result code must be integer\n", spec); continue; } action->failed = TRUE; graph->abort_priority = INFINITY; printf("\tPretending action %d failed with rc=%d\n", action->id, op->rc); update_failcounts(cib_node, match_name, op->op_type, op->interval_ms, op->rc); break; } } inject_op(cib_resource, op, target_outcome); lrmd_free_event(op); rc = fake_cib->cmds->modify(fake_cib, XML_CIB_TAG_STATUS, cib_node, cib_sync_call | cib_scope_local); CRM_ASSERT(rc == pcmk_ok); done: free(node); free(uuid); free_xml(cib_node); action->confirmed = TRUE; update_graph(graph, action); return TRUE; } static gboolean exec_crmd_action(crm_graph_t * graph, crm_action_t * action) { const char *node = crm_element_value(action->xml, XML_LRM_ATTR_TARGET); const char *task = crm_element_value(action->xml, XML_LRM_ATTR_TASK); xmlNode *rsc = first_named_child(action->xml, XML_CIB_TAG_RESOURCE); action->confirmed = TRUE; if(rsc) { quiet_log(" * Cluster action: %s for %s on %s\n", task, ID(rsc), node); } else { quiet_log(" * Cluster action: %s on %s\n", task, node); } update_graph(graph, action); return TRUE; } static gboolean exec_stonith_action(crm_graph_t * graph, crm_action_t * action) { const char *op = crm_meta_value(action->params, "stonith_action"); char *target = crm_element_value_copy(action->xml, XML_LRM_ATTR_TARGET); quiet_log(" * Fencing %s (%s)\n", target, op); if(safe_str_neq(op, "on")) { int rc = 0; char xpath[STATUS_PATH_MAX]; xmlNode *cib_node = modify_node(fake_cib, target, FALSE); crm_xml_add(cib_node, XML_ATTR_ORIGIN, __FUNCTION__); CRM_ASSERT(cib_node != NULL); rc = fake_cib->cmds->replace(fake_cib, XML_CIB_TAG_STATUS, cib_node, cib_sync_call | cib_scope_local); CRM_ASSERT(rc == pcmk_ok); snprintf(xpath, STATUS_PATH_MAX, "//node_state[@uname='%s']/%s", target, XML_CIB_TAG_LRM); fake_cib->cmds->remove(fake_cib, xpath, NULL, cib_xpath | cib_sync_call | cib_scope_local); snprintf(xpath, STATUS_PATH_MAX, "//node_state[@uname='%s']/%s", target, XML_TAG_TRANSIENT_NODEATTRS); fake_cib->cmds->remove(fake_cib, xpath, NULL, cib_xpath | cib_sync_call | cib_scope_local); free_xml(cib_node); } action->confirmed = TRUE; update_graph(graph, action); free(target); return TRUE; } int run_simulation(pe_working_set_t * data_set, cib_t *cib, GListPtr op_fail_list, bool quiet) { crm_graph_t *transition = NULL; enum transition_status graph_rc = -1; crm_graph_functions_t exec_fns = { exec_pseudo_action, exec_rsc_action, exec_crmd_action, exec_stonith_action, }; fake_cib = cib; fake_quiet = quiet; fake_op_fail_list = op_fail_list; quiet_log("\nExecuting cluster transition:\n"); set_graph_functions(&exec_fns); transition = unpack_graph(data_set->graph, crm_system_name); print_graph(LOG_DEBUG, transition); fake_resource_list = data_set->resources; do { graph_rc = run_graph(transition); } while (graph_rc == transition_active); fake_resource_list = NULL; if (graph_rc != transition_complete) { fprintf(stdout, "Transition failed: %s\n", transition_status(graph_rc)); print_graph(LOG_ERR, transition); } destroy_graph(transition); if (graph_rc != transition_complete) { fprintf(stdout, "An invalid transition was produced\n"); } if (quiet == FALSE) { xmlNode *cib_object = NULL; int rc = fake_cib->cmds->query(fake_cib, NULL, &cib_object, cib_sync_call | cib_scope_local); CRM_ASSERT(rc == pcmk_ok); pe_reset_working_set(data_set); data_set->input = cib_object; } if (graph_rc != transition_complete) { return graph_rc; } return 0; } diff --git a/lib/pengine/utils.c b/lib/pengine/utils.c index 4eebe33699..f21c397b64 100644 --- a/lib/pengine/utils.c +++ b/lib/pengine/utils.c @@ -1,2832 +1,2832 @@ /* * Copyright 2004-2020 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #include extern xmlNode *get_object_root(const char *object_type, xmlNode * the_root); void print_str_str(gpointer key, gpointer value, gpointer user_data); gboolean ghash_free_str_str(gpointer key, gpointer value, gpointer user_data); static void unpack_operation(pe_action_t * action, xmlNode * xml_obj, pe_resource_t * container, pe_working_set_t * data_set, guint interval_ms); static xmlNode *find_rsc_op_entry_helper(pe_resource_t * rsc, const char *key, gboolean include_disabled); #if ENABLE_VERSIONED_ATTRS pe_rsc_action_details_t * pe_rsc_action_details(pe_action_t *action) { pe_rsc_action_details_t *details; CRM_CHECK(action != NULL, return NULL); if (action->action_details == NULL) { action->action_details = calloc(1, sizeof(pe_rsc_action_details_t)); CRM_CHECK(action->action_details != NULL, return NULL); } details = (pe_rsc_action_details_t *) action->action_details; if (details->versioned_parameters == NULL) { details->versioned_parameters = create_xml_node(NULL, XML_TAG_OP_VER_ATTRS); } if (details->versioned_meta == NULL) { details->versioned_meta = create_xml_node(NULL, XML_TAG_OP_VER_META); } return details; } static void pe_free_rsc_action_details(pe_action_t *action) { pe_rsc_action_details_t *details; if ((action == NULL) || (action->action_details == NULL)) { return; } details = (pe_rsc_action_details_t *) action->action_details; if (details->versioned_parameters) { free_xml(details->versioned_parameters); } if (details->versioned_meta) { free_xml(details->versioned_meta); } action->action_details = NULL; } #endif /*! * \internal * \brief Check whether we can fence a particular node * * \param[in] data_set Working set for cluster * \param[in] node Name of node to check * * \return true if node can be fenced, false otherwise */ bool pe_can_fence(pe_working_set_t *data_set, pe_node_t *node) { if (pe__is_guest_node(node)) { /* Guest nodes are fenced by stopping their container resource. We can * do that if the container's host is either online or fenceable. */ pe_resource_t *rsc = node->details->remote_rsc->container; for (GList *n = rsc->running_on; n != NULL; n = n->next) { pe_node_t *container_node = n->data; if (!container_node->details->online && !pe_can_fence(data_set, container_node)) { return false; } } return true; } else if(is_not_set(data_set->flags, pe_flag_stonith_enabled)) { return false; /* Turned off */ } else if (is_not_set(data_set->flags, pe_flag_have_stonith_resource)) { return false; /* No devices */ } else if (is_set(data_set->flags, pe_flag_have_quorum)) { return true; } else if (data_set->no_quorum_policy == 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", node->details->uname); return true; } crm_trace("Cannot fence %s", node->details->uname); 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. */ pe_node_t * pe__copy_node(const pe_node_t *this_node) { pe_node_t *new_node = NULL; CRM_ASSERT(this_node != NULL); new_node = calloc(1, sizeof(pe_node_t)); CRM_ASSERT(new_node != NULL); new_node->rsc_discover_mode = this_node->rsc_discover_mode; new_node->weight = this_node->weight; new_node->fixed = this_node->fixed; new_node->details = this_node->details; return new_node; } /* any node in list1 or list2 and not in the other gets a score of -INFINITY */ void node_list_exclude(GHashTable * hash, GListPtr list, gboolean merge_scores) { GHashTable *result = hash; pe_node_t *other_node = NULL; GListPtr gIter = list; GHashTableIter iter; pe_node_t *node = NULL; g_hash_table_iter_init(&iter, hash); while (g_hash_table_iter_next(&iter, NULL, (void **)&node)) { other_node = pe_find_node_id(list, node->details->id); if (other_node == NULL) { node->weight = -INFINITY; } else if (merge_scores) { node->weight = pe__add_scores(node->weight, other_node->weight); } } for (; gIter != NULL; gIter = gIter->next) { pe_node_t *node = (pe_node_t *) gIter->data; other_node = pe_hash_table_lookup(result, node->details->id); if (other_node == NULL) { pe_node_t *new_node = pe__copy_node(node); new_node->weight = -INFINITY; g_hash_table_insert(result, (gpointer) new_node->details->id, new_node); } } } /*! * \internal * \brief Create a node hash table from a node list * * \param[in] list Node list * * \return Hash table equivalent of node list */ GHashTable * pe__node_list2table(GList *list) { GHashTable *result = NULL; result = g_hash_table_new_full(crm_str_hash, g_str_equal, NULL, free); for (GList *gIter = list; gIter != NULL; gIter = gIter->next) { pe_node_t *new_node = pe__copy_node((pe_node_t *) gIter->data); g_hash_table_insert(result, (gpointer) new_node->details->id, new_node); } return result; } gint sort_node_uname(gconstpointer a, gconstpointer b) { return pcmk_numeric_strcasecmp(((const pe_node_t *) a)->details->uname, ((const pe_node_t *) b)->details->uname); } /*! * \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 */ static void pe__output_node_weights(pe_resource_t *rsc, const char *comment, GHashTable *nodes) { char score[128]; // Stack-allocated since this is called frequently // Sort the nodes so the output is consistent for regression tests GList *list = g_list_sort(g_hash_table_get_values(nodes), sort_node_uname); for (GList *gIter = list; gIter != NULL; gIter = gIter->next) { pe_node_t *node = (pe_node_t *) gIter->data; score2char_stack(node->weight, score, sizeof(score)); if (rsc) { printf("%s: %s allocation score on %s: %s\n", comment, rsc->id, node->details->uname, score); } else { printf("%s: %s = %s\n", comment, node->details->uname, 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 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 */ static void pe__log_node_weights(const char *file, const char *function, int line, pe_resource_t *rsc, const char *comment, GHashTable *nodes) { GHashTableIter iter; pe_node_t *node = NULL; char score[128]; // Stack-allocated since this is called frequently // Don't waste time if we're not tracing at this point pcmk__log_else(LOG_TRACE, return); g_hash_table_iter_init(&iter, nodes); while (g_hash_table_iter_next(&iter, NULL, (void **) &node)) { score2char_stack(node->weight, score, sizeof(score)); if (rsc) { qb_log_from_external_source(function, file, "%s: %s allocation score on %s: %s", LOG_TRACE, line, 0, comment, rsc->id, node->details->uname, score); } else { qb_log_from_external_source(function, file, "%s: %s = %s", LOG_TRACE, line, 0, comment, node->details->uname, 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 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 */ void pe__show_node_weights_as(const char *file, const char *function, int line, bool to_log, pe_resource_t *rsc, const char *comment, GHashTable *nodes) { if (rsc != NULL) { if (is_set(rsc->flags, pe_rsc_orphan)) { // Don't show allocation scores for orphans return; } nodes = rsc->allowed_nodes; } 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); } // If this resource has children, repeat recursively for each if (rsc && rsc->children) { for (GList *gIter = rsc->children; gIter != NULL; gIter = gIter->next) { pe_resource_t *child = (pe_resource_t *) gIter->data; pe__show_node_weights_as(file, function, line, to_log, child, comment, nodes); } } } static void append_dump_text(gpointer key, gpointer value, gpointer user_data) { char **dump_text = user_data; char *new_text = crm_strdup_printf("%s %s=%s", *dump_text, (char *)key, (char *)value); free(*dump_text); *dump_text = new_text; } void dump_node_capacity(int level, const char *comment, pe_node_t * node) { char *dump_text = crm_strdup_printf("%s: %s capacity:", comment, node->details->uname); g_hash_table_foreach(node->details->utilization, append_dump_text, &dump_text); if (level == LOG_STDOUT) { fprintf(stdout, "%s\n", dump_text); } else { crm_trace("%s", dump_text); } free(dump_text); } void dump_rsc_utilization(int level, const char *comment, pe_resource_t * rsc, pe_node_t * node) { char *dump_text = crm_strdup_printf("%s: %s utilization on %s:", comment, rsc->id, node->details->uname); g_hash_table_foreach(rsc->utilization, append_dump_text, &dump_text); switch (level) { case LOG_STDOUT: fprintf(stdout, "%s\n", dump_text); break; case LOG_NEVER: break; default: crm_trace("%s", dump_text); } free(dump_text); } gint sort_rsc_index(gconstpointer a, gconstpointer b) { const pe_resource_t *resource1 = (const pe_resource_t *)a; const pe_resource_t *resource2 = (const pe_resource_t *)b; if (a == NULL && b == NULL) { return 0; } if (a == NULL) { return 1; } if (b == NULL) { return -1; } if (resource1->sort_index > resource2->sort_index) { return -1; } if (resource1->sort_index < resource2->sort_index) { return 1; } return 0; } gint sort_rsc_priority(gconstpointer a, gconstpointer b) { const pe_resource_t *resource1 = (const pe_resource_t *)a; const pe_resource_t *resource2 = (const pe_resource_t *)b; if (a == NULL && b == NULL) { return 0; } if (a == NULL) { return 1; } if (b == NULL) { return -1; } if (resource1->priority > resource2->priority) { return -1; } if (resource1->priority < resource2->priority) { return 1; } return 0; } static enum pe_quorum_policy effective_quorum_policy(pe_resource_t *rsc, pe_working_set_t *data_set) { enum pe_quorum_policy policy = data_set->no_quorum_policy; if (is_set(data_set->flags, pe_flag_have_quorum)) { policy = no_quorum_ignore; } else if (data_set->no_quorum_policy == no_quorum_demote) { switch (rsc->role) { case RSC_ROLE_MASTER: case RSC_ROLE_SLAVE: if (rsc->next_role > RSC_ROLE_SLAVE) { rsc->next_role = RSC_ROLE_SLAVE; } policy = no_quorum_ignore; break; default: policy = no_quorum_stop; break; } } return policy; } pe_action_t * custom_action(pe_resource_t * rsc, char *key, const char *task, pe_node_t * on_node, gboolean optional, gboolean save_action, pe_working_set_t * data_set) { pe_action_t *action = NULL; GListPtr possible_matches = NULL; CRM_CHECK(key != NULL, return NULL); CRM_CHECK(task != NULL, free(key); return NULL); if (save_action && rsc != NULL) { possible_matches = find_actions(rsc->actions, key, on_node); } else if(save_action) { #if 0 action = g_hash_table_lookup(data_set->singletons, key); #else /* More expensive but takes 'node' into account */ possible_matches = find_actions(data_set->actions, key, on_node); #endif } if(data_set->singletons == NULL) { data_set->singletons = g_hash_table_new_full(crm_str_hash, g_str_equal, NULL, NULL); } if (possible_matches != NULL) { if (pcmk__list_of_multiple(possible_matches)) { pe_warn("Action %s for %s on %s exists %d times", task, rsc ? rsc->id : "", on_node ? on_node->details->uname : "", g_list_length(possible_matches)); } action = g_list_nth_data(possible_matches, 0); pe_rsc_trace(rsc, "Found existing action %d (%s) for %s (%s) on %s", action->id, action->uuid, (rsc? rsc->id : "no resource"), task, (on_node? on_node->details->uname : "no node")); g_list_free(possible_matches); } if (action == NULL) { if (save_action) { pe_rsc_trace(rsc, "Creating %s action %d: %s for %s (%s) on %s", (optional? "optional" : "mandatory"), data_set->action_id, key, (rsc? rsc->id : "no resource"), task, (on_node? on_node->details->uname : "no node")); } action = calloc(1, sizeof(pe_action_t)); if (save_action) { action->id = data_set->action_id++; } else { action->id = 0; } action->rsc = rsc; CRM_ASSERT(task != NULL); action->task = strdup(task); if (on_node) { action->node = pe__copy_node(on_node); } action->uuid = strdup(key); if (safe_str_eq(task, CRM_OP_LRM_DELETE)) { // Resource history deletion for a node can be done on the DC pe_set_action_bit(action, pe_action_dc); } pe_set_action_bit(action, pe_action_runnable); if (optional) { pe_set_action_bit(action, pe_action_optional); } else { pe_clear_action_bit(action, pe_action_optional); } action->extra = crm_str_table_new(); action->meta = crm_str_table_new(); if (save_action) { data_set->actions = g_list_prepend(data_set->actions, action); if(rsc == NULL) { g_hash_table_insert(data_set->singletons, action->uuid, action); } } if (rsc != NULL) { guint interval_ms = 0; action->op_entry = find_rsc_op_entry_helper(rsc, key, TRUE); parse_op_key(key, NULL, NULL, &interval_ms); unpack_operation(action, action->op_entry, rsc->container, data_set, interval_ms); if (save_action) { rsc->actions = g_list_prepend(rsc->actions, action); } } if (save_action) { pe_rsc_trace(rsc, "Action %d created", action->id); } } if (!optional && is_set(action->flags, pe_action_optional)) { pe_rsc_trace(rsc, "Unset optional on action %d", action->id); pe_clear_action_bit(action, pe_action_optional); } if (rsc != NULL) { enum action_tasks a_task = text2task(action->task); enum pe_quorum_policy quorum_policy = effective_quorum_policy(rsc, data_set); int warn_level = LOG_TRACE; if (save_action) { warn_level = LOG_WARNING; } if (is_set(action->flags, pe_action_have_node_attrs) == FALSE && action->node != NULL && action->op_entry != NULL) { pe_rule_eval_data_t rule_data = { .node_hash = action->node->details->attrs, .role = RSC_ROLE_UNKNOWN, .now = data_set->now, .match_data = NULL, .rsc_data = NULL, .op_data = NULL }; pe_set_action_bit(action, pe_action_have_node_attrs); pe__unpack_dataset_nvpairs(action->op_entry, XML_TAG_ATTR_SETS, &rule_data, action->extra, NULL, FALSE, data_set); } if (is_set(action->flags, pe_action_pseudo)) { /* leave untouched */ } else if (action->node == NULL) { pe_rsc_trace(rsc, "Unset runnable on %s", action->uuid); pe_clear_action_bit(action, pe_action_runnable); } else if (is_not_set(rsc->flags, pe_rsc_managed) && g_hash_table_lookup(action->meta, XML_LRM_ATTR_INTERVAL_MS) == NULL) { crm_debug("Action %s (unmanaged)", action->uuid); pe_rsc_trace(rsc, "Set optional on %s", action->uuid); pe_set_action_bit(action, pe_action_optional); /* action->runnable = FALSE; */ } else if (is_not_set(action->flags, pe_action_dc) && !(action->node->details->online) && (!pe__is_guest_node(action->node) || action->node->details->remote_requires_reset)) { pe_clear_action_bit(action, pe_action_runnable); do_crm_log(warn_level, "Action %s on %s is unrunnable (offline)", action->uuid, action->node->details->uname); if (is_set(action->rsc->flags, pe_rsc_managed) && save_action && a_task == stop_rsc && action->node->details->unclean == FALSE) { pe_fence_node(data_set, action->node, "resource actions are unrunnable", FALSE); } } else if (is_not_set(action->flags, pe_action_dc) && action->node->details->pending) { pe_clear_action_bit(action, pe_action_runnable); do_crm_log(warn_level, "Action %s on %s is unrunnable (pending)", action->uuid, action->node->details->uname); } else if (action->needs == rsc_req_nothing) { pe_rsc_trace(rsc, "Action %s does not require anything", action->uuid); pe_action_set_reason(action, NULL, TRUE); if (pe__is_guest_node(action->node) && !pe_can_fence(data_set, action->node)) { /* An action that requires nothing usually does not require any * fencing in order to be runnable. However, there is an * exception: an action cannot be completed if it is on a guest * node whose host is unclean and cannot be fenced. */ pe_clear_action_bit(action, pe_action_runnable); crm_debug("%s\t%s (cancelled : host cannot be fenced)", action->node->details->uname, action->uuid); } else { pe_set_action_bit(action, pe_action_runnable); } #if 0 /* * No point checking this * - if we don't have quorum we can't stonith anyway */ } else if (action->needs == rsc_req_stonith) { crm_trace("Action %s requires only stonith", action->uuid); action->runnable = TRUE; #endif } else if (quorum_policy == no_quorum_stop) { pe_action_set_flag_reason(__FUNCTION__, __LINE__, action, NULL, "no quorum", pe_action_runnable, TRUE); crm_debug("%s\t%s (cancelled : quorum)", action->node->details->uname, action->uuid); } else if (quorum_policy == no_quorum_freeze) { pe_rsc_trace(rsc, "Check resource is already active: %s %s %s %s", rsc->id, action->uuid, role2text(rsc->next_role), role2text(rsc->role)); if (rsc->fns->active(rsc, TRUE) == FALSE || rsc->next_role > rsc->role) { pe_action_set_flag_reason(__FUNCTION__, __LINE__, action, NULL, "quorum freeze", pe_action_runnable, TRUE); pe_rsc_debug(rsc, "%s\t%s (cancelled : quorum freeze)", action->node->details->uname, action->uuid); } } else if(is_not_set(action->flags, pe_action_runnable)) { pe_rsc_trace(rsc, "Action %s is runnable", action->uuid); //pe_action_set_reason(action, NULL, TRUE); pe_set_action_bit(action, pe_action_runnable); } if (save_action) { switch (a_task) { case stop_rsc: set_bit(rsc->flags, pe_rsc_stopping); break; case start_rsc: clear_bit(rsc->flags, pe_rsc_starting); if (is_set(action->flags, pe_action_runnable)) { set_bit(rsc->flags, pe_rsc_starting); } break; default: break; } } } free(key); return action; } static bool valid_stop_on_fail(const char *value) { - return pcmk__str_none_of(value, "standby", "demote", "stop", NULL); + return !pcmk__str_any_of(value, "standby", "demote", "stop", NULL); } static const char * unpack_operation_on_fail(pe_action_t * action) { const char *name = NULL; const char *role = NULL; const char *on_fail = NULL; const char *interval_spec = NULL; const char *enabled = NULL; const char *value = g_hash_table_lookup(action->meta, XML_OP_ATTR_ON_FAIL); if (safe_str_eq(action->task, CRMD_ACTION_STOP) && !valid_stop_on_fail(value)) { pcmk__config_err("Resetting '" XML_OP_ATTR_ON_FAIL "' for %s stop " "action to default value because '%s' is not " "allowed for stop", action->rsc->id, value); return NULL; } else if (safe_str_eq(action->task, CRMD_ACTION_DEMOTE) && !value) { /* demote on_fail defaults to master monitor value if present */ xmlNode *operation = NULL; CRM_CHECK(action->rsc != NULL, return NULL); for (operation = __xml_first_child_element(action->rsc->ops_xml); operation && !value; operation = __xml_next_element(operation)) { if (!crm_str_eq((const char *)operation->name, "op", TRUE)) { continue; } name = crm_element_value(operation, "name"); role = crm_element_value(operation, "role"); on_fail = crm_element_value(operation, XML_OP_ATTR_ON_FAIL); enabled = crm_element_value(operation, "enabled"); interval_spec = crm_element_value(operation, XML_LRM_ATTR_INTERVAL); if (!on_fail) { continue; } else if (enabled && !crm_is_true(enabled)) { continue; } else if (safe_str_neq(name, "monitor") || safe_str_neq(role, "Master")) { continue; } else if (crm_parse_interval_spec(interval_spec) == 0) { continue; } else if (safe_str_eq(on_fail, "demote")) { continue; } value = on_fail; } } else if (safe_str_eq(action->task, CRM_OP_LRM_DELETE)) { value = "ignore"; } else if (safe_str_eq(value, "demote")) { name = crm_element_value(action->op_entry, "name"); role = crm_element_value(action->op_entry, "role"); on_fail = crm_element_value(action->op_entry, XML_OP_ATTR_ON_FAIL); interval_spec = crm_element_value(action->op_entry, XML_LRM_ATTR_INTERVAL); if (safe_str_neq(name, CRMD_ACTION_PROMOTE) && (safe_str_neq(name, CRMD_ACTION_STATUS) || safe_str_neq(role, "Master") || (crm_parse_interval_spec(interval_spec) == 0))) { pcmk__config_err("Resetting '" XML_OP_ATTR_ON_FAIL "' for %s %s " "action to default value because 'demote' is not " "allowed for it", action->rsc->id, name); return NULL; } } return value; } static xmlNode * find_min_interval_mon(pe_resource_t * rsc, gboolean include_disabled) { guint interval_ms = 0; guint min_interval_ms = G_MAXUINT; const char *name = NULL; const char *value = NULL; const char *interval_spec = NULL; xmlNode *op = NULL; xmlNode *operation = NULL; for (operation = __xml_first_child_element(rsc->ops_xml); operation != NULL; operation = __xml_next_element(operation)) { if (crm_str_eq((const char *)operation->name, "op", TRUE)) { name = crm_element_value(operation, "name"); interval_spec = crm_element_value(operation, XML_LRM_ATTR_INTERVAL); value = crm_element_value(operation, "enabled"); if (!include_disabled && value && crm_is_true(value) == FALSE) { continue; } if (safe_str_neq(name, RSC_STATUS)) { continue; } interval_ms = crm_parse_interval_spec(interval_spec); if (interval_ms && (interval_ms < min_interval_ms)) { min_interval_ms = interval_ms; op = operation; } } } return op; } static int unpack_start_delay(const char *value, GHashTable *meta) { int start_delay = 0; if (value != NULL) { start_delay = crm_get_msec(value); if (start_delay < 0) { start_delay = 0; } if (meta) { g_hash_table_replace(meta, strdup(XML_OP_ATTR_START_DELAY), crm_itoa(start_delay)); } } return start_delay; } // true if value contains valid, non-NULL interval origin for recurring op static bool unpack_interval_origin(const char *value, xmlNode *xml_obj, guint interval_ms, crm_time_t *now, long long *start_delay) { long long result = 0; guint interval_sec = interval_ms / 1000; 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 '" XML_OP_ATTR_ORIGIN "' for operation " "'%s' because '%s' is not valid", (ID(xml_obj)? 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, (ID(xml_obj)? ID(xml_obj) : "(unspecified)")); if (start_delay != NULL) { *start_delay = result * 1000; // milliseconds } return true; } static int unpack_timeout(const char *value) { int timeout = crm_get_msec(value); if (timeout < 0) { timeout = crm_get_msec(CRM_DEFAULT_OP_TIMEOUT_S); } return timeout; } int pe_get_configured_timeout(pe_resource_t *rsc, const char *action, pe_working_set_t *data_set) { xmlNode *child = NULL; const char *timeout = NULL; GHashTable *action_meta = NULL; int timeout_ms = 0; pe_rule_eval_data_t rule_data = { .node_hash = NULL, .role = RSC_ROLE_UNKNOWN, .now = data_set->now, .match_data = NULL, .rsc_data = NULL, .op_data = NULL }; for (child = first_named_child(rsc->ops_xml, XML_ATTR_OP); child != NULL; child = crm_next_same_xml(child)) { if (safe_str_eq(action, crm_element_value(child, XML_NVPAIR_ATTR_NAME))) { timeout = crm_element_value(child, XML_ATTR_TIMEOUT); break; } } if (timeout == NULL && data_set->op_defaults) { action_meta = crm_str_table_new(); pe__unpack_dataset_nvpairs(data_set->op_defaults, XML_TAG_META_SETS, &rule_data, action_meta, NULL, FALSE, data_set); timeout = g_hash_table_lookup(action_meta, XML_ATTR_TIMEOUT); } // @TODO check meta-attributes (including versioned meta-attributes) // @TODO maybe use min-interval monitor timeout as default for monitors timeout_ms = crm_get_msec(timeout); if (timeout_ms < 0) { timeout_ms = crm_get_msec(CRM_DEFAULT_OP_TIMEOUT_S); } if (action_meta != NULL) { g_hash_table_destroy(action_meta); } return timeout_ms; } #if ENABLE_VERSIONED_ATTRS static void unpack_versioned_meta(xmlNode *versioned_meta, xmlNode *xml_obj, guint interval_ms, crm_time_t *now) { xmlNode *attrs = NULL; xmlNode *attr = NULL; for (attrs = __xml_first_child_element(versioned_meta); attrs != NULL; attrs = __xml_next_element(attrs)) { for (attr = __xml_first_child_element(attrs); attr != NULL; attr = __xml_next_element(attr)) { const char *name = crm_element_value(attr, XML_NVPAIR_ATTR_NAME); const char *value = crm_element_value(attr, XML_NVPAIR_ATTR_VALUE); if (safe_str_eq(name, XML_OP_ATTR_START_DELAY)) { int start_delay = unpack_start_delay(value, NULL); crm_xml_add_int(attr, XML_NVPAIR_ATTR_VALUE, start_delay); } else if (safe_str_eq(name, XML_OP_ATTR_ORIGIN)) { long long start_delay = 0; if (unpack_interval_origin(value, xml_obj, interval_ms, now, &start_delay)) { crm_xml_add(attr, XML_NVPAIR_ATTR_NAME, XML_OP_ATTR_START_DELAY); crm_xml_add_ll(attr, XML_NVPAIR_ATTR_VALUE, start_delay); } } else if (safe_str_eq(name, XML_ATTR_TIMEOUT)) { int timeout = unpack_timeout(value); crm_xml_add_int(attr, XML_NVPAIR_ATTR_VALUE, timeout); } } } } #endif /*! * \brief Unpack operation XML into an action structure * * Unpack an operation's meta-attributes (normalizing the interval, timeout, * and start delay values as integer milliseconds), requirements, and * failure policy. * * \param[in,out] action Action to unpack into * \param[in] xml_obj Operation XML (or NULL if all defaults) * \param[in] container Resource that contains affected resource, if any * \param[in] data_set Cluster state * \param[in] interval_ms How frequently to perform the operation */ static void unpack_operation(pe_action_t * action, xmlNode * xml_obj, pe_resource_t * container, pe_working_set_t * data_set, guint interval_ms) { int timeout = 0; char *value_ms = NULL; const char *value = NULL; const char *field = XML_LRM_ATTR_INTERVAL; char *default_timeout = NULL; #if ENABLE_VERSIONED_ATTRS pe_rsc_action_details_t *rsc_details = NULL; #endif pe_rsc_eval_data_t rsc_rule_data = { .standard = crm_element_value(action->rsc->xml, XML_AGENT_ATTR_CLASS), .provider = crm_element_value(action->rsc->xml, XML_AGENT_ATTR_PROVIDER), .agent = crm_element_value(action->rsc->xml, XML_EXPR_ATTR_TYPE) }; pe_op_eval_data_t op_rule_data = { .op_name = action->task, .interval = interval_ms }; pe_rule_eval_data_t rule_data = { .node_hash = NULL, .role = RSC_ROLE_UNKNOWN, .now = data_set->now, .match_data = NULL, .rsc_data = &rsc_rule_data, .op_data = &op_rule_data }; CRM_CHECK(action && action->rsc, return); // Cluster-wide pe__unpack_dataset_nvpairs(data_set->op_defaults, XML_TAG_META_SETS, &rule_data, action->meta, NULL, FALSE, data_set); // Probe timeouts default differently, so handle timeout default later default_timeout = g_hash_table_lookup(action->meta, XML_ATTR_TIMEOUT); if (default_timeout) { default_timeout = strdup(default_timeout); g_hash_table_remove(action->meta, XML_ATTR_TIMEOUT); } if (xml_obj) { xmlAttrPtr xIter = NULL; // take precedence over defaults pe__unpack_dataset_nvpairs(xml_obj, XML_TAG_META_SETS, &rule_data, action->meta, NULL, TRUE, data_set); #if ENABLE_VERSIONED_ATTRS rsc_details = pe_rsc_action_details(action); pe_eval_versioned_attributes(data_set->input, xml_obj, XML_TAG_ATTR_SETS, &rule_data, rsc_details->versioned_parameters, NULL); pe_eval_versioned_attributes(data_set->input, xml_obj, XML_TAG_META_SETS, &rule_data, rsc_details->versioned_meta, NULL); #endif /* Anything set as an XML property has highest precedence. * This ensures we use the name and interval from the tag. */ for (xIter = xml_obj->properties; xIter; xIter = xIter->next) { const char *prop_name = (const char *)xIter->name; const char *prop_value = crm_element_value(xml_obj, prop_name); g_hash_table_replace(action->meta, strdup(prop_name), strdup(prop_value)); } } g_hash_table_remove(action->meta, "id"); // Normalize interval to milliseconds if (interval_ms > 0) { value_ms = crm_strdup_printf("%u", interval_ms); g_hash_table_replace(action->meta, strdup(field), value_ms); } else if (g_hash_table_lookup(action->meta, field) != NULL) { g_hash_table_remove(action->meta, field); } // Handle timeout default, now that we know the interval if (g_hash_table_lookup(action->meta, XML_ATTR_TIMEOUT)) { free(default_timeout); } else { // Probe timeouts default to minimum-interval monitor's if (safe_str_eq(action->task, RSC_STATUS) && (interval_ms == 0)) { xmlNode *min_interval_mon = find_min_interval_mon(action->rsc, FALSE); if (min_interval_mon) { value = crm_element_value(min_interval_mon, XML_ATTR_TIMEOUT); if (value) { crm_trace("\t%s defaults to minimum-interval monitor's timeout '%s'", action->uuid, value); free(default_timeout); default_timeout = strdup(value); } } } if (default_timeout) { g_hash_table_insert(action->meta, strdup(XML_ATTR_TIMEOUT), default_timeout); } } - if (pcmk__str_none_of(action->task, RSC_START, RSC_PROMOTE, NULL)) { + if (!pcmk__str_any_of(action->task, RSC_START, RSC_PROMOTE, NULL)) { action->needs = rsc_req_nothing; value = "nothing (not start/promote)"; } else if (is_set(action->rsc->flags, pe_rsc_needs_fencing)) { action->needs = rsc_req_stonith; value = "fencing (resource)"; } else if (is_set(action->rsc->flags, pe_rsc_needs_quorum)) { action->needs = rsc_req_quorum; value = "quorum (resource)"; } else { action->needs = rsc_req_nothing; value = "nothing (resource)"; } pe_rsc_trace(action->rsc, "\tAction %s requires: %s", action->uuid, value); value = unpack_operation_on_fail(action); if (value == NULL) { } else if (safe_str_eq(value, "block")) { action->on_fail = action_fail_block; g_hash_table_insert(action->meta, strdup(XML_OP_ATTR_ON_FAIL), strdup("block")); value = "block"; // The above could destroy the original string } else if (safe_str_eq(value, "fence")) { action->on_fail = action_fail_fence; value = "node fencing"; if (is_set(data_set->flags, pe_flag_stonith_enabled) == FALSE) { pcmk__config_err("Resetting '" XML_OP_ATTR_ON_FAIL "' for " "operation '%s' to 'stop' because 'fence' is not " "valid when fencing is disabled", action->uuid); action->on_fail = action_fail_stop; action->fail_role = RSC_ROLE_STOPPED; value = "stop resource"; } } else if (safe_str_eq(value, "standby")) { action->on_fail = action_fail_standby; value = "node standby"; } else if (pcmk__str_any_of(value, "ignore", "nothing", NULL)) { action->on_fail = action_fail_ignore; value = "ignore"; } else if (safe_str_eq(value, "migrate")) { action->on_fail = action_fail_migrate; value = "force migration"; } else if (safe_str_eq(value, "stop")) { action->on_fail = action_fail_stop; action->fail_role = RSC_ROLE_STOPPED; value = "stop resource"; } else if (safe_str_eq(value, "restart")) { action->on_fail = action_fail_recover; value = "restart (and possibly migrate)"; } else if (safe_str_eq(value, "restart-container")) { if (container) { action->on_fail = action_fail_restart_container; value = "restart container (and possibly migrate)"; } else { value = NULL; } } else if (safe_str_eq(value, "demote")) { action->on_fail = action_fail_demote; value = "demote instance"; } else { pe_err("Resource %s: Unknown failure type (%s)", action->rsc->id, value); value = NULL; } /* defaults */ if (value == NULL && container) { action->on_fail = action_fail_restart_container; value = "restart container (and possibly migrate) (default)"; /* For remote nodes, ensure that any failure that results in dropping an * active connection to the node results in fencing of the node. * * There are only two action failures that don't result in fencing. * 1. probes - probe failures are expected. * 2. start - a start failure indicates that an active connection does not already * exist. The user can set op on-fail=fence if they really want to fence start * failures. */ } else if (((value == NULL) || !is_set(action->rsc->flags, pe_rsc_managed)) && (pe__resource_is_remote_conn(action->rsc, data_set) && !(safe_str_eq(action->task, CRMD_ACTION_STATUS) && (interval_ms == 0)) && (safe_str_neq(action->task, CRMD_ACTION_START)))) { if (!is_set(action->rsc->flags, pe_rsc_managed)) { action->on_fail = action_fail_stop; action->fail_role = RSC_ROLE_STOPPED; value = "stop unmanaged remote node (enforcing default)"; } else { if (is_set(data_set->flags, pe_flag_stonith_enabled)) { value = "fence remote node (default)"; } else { value = "recover remote node connection (default)"; } if (action->rsc->remote_reconnect_ms) { action->fail_role = RSC_ROLE_STOPPED; } action->on_fail = action_fail_reset_remote; } } else if (value == NULL && safe_str_eq(action->task, CRMD_ACTION_STOP)) { if (is_set(data_set->flags, pe_flag_stonith_enabled)) { action->on_fail = action_fail_fence; value = "resource fence (default)"; } else { action->on_fail = action_fail_block; value = "resource block (default)"; } } else if (value == NULL) { action->on_fail = action_fail_recover; value = "restart (and possibly migrate) (default)"; } pe_rsc_trace(action->rsc, "\t%s failure handling: %s", action->task, value); value = NULL; if (xml_obj != NULL) { value = g_hash_table_lookup(action->meta, "role_after_failure"); if (value) { pe_warn_once(pe_wo_role_after, "Support for role_after_failure is deprecated and will be removed in a future release"); } } if (value != NULL && action->fail_role == RSC_ROLE_UNKNOWN) { action->fail_role = text2role(value); } /* defaults */ if (action->fail_role == RSC_ROLE_UNKNOWN) { if (safe_str_eq(action->task, CRMD_ACTION_PROMOTE)) { action->fail_role = RSC_ROLE_SLAVE; } else { action->fail_role = RSC_ROLE_STARTED; } } pe_rsc_trace(action->rsc, "\t%s failure results in: %s", action->task, role2text(action->fail_role)); value = g_hash_table_lookup(action->meta, XML_OP_ATTR_START_DELAY); if (value) { unpack_start_delay(value, action->meta); } else { long long start_delay = 0; value = g_hash_table_lookup(action->meta, XML_OP_ATTR_ORIGIN); if (unpack_interval_origin(value, xml_obj, interval_ms, data_set->now, &start_delay)) { g_hash_table_replace(action->meta, strdup(XML_OP_ATTR_START_DELAY), crm_strdup_printf("%lld", start_delay)); } } value = g_hash_table_lookup(action->meta, XML_ATTR_TIMEOUT); timeout = unpack_timeout(value); g_hash_table_replace(action->meta, strdup(XML_ATTR_TIMEOUT), crm_itoa(timeout)); #if ENABLE_VERSIONED_ATTRS unpack_versioned_meta(rsc_details->versioned_meta, xml_obj, interval_ms, data_set->now); #endif } static xmlNode * find_rsc_op_entry_helper(pe_resource_t * rsc, const char *key, gboolean include_disabled) { guint interval_ms = 0; gboolean do_retry = TRUE; char *local_key = NULL; const char *name = NULL; const char *value = NULL; const char *interval_spec = NULL; char *match_key = NULL; xmlNode *op = NULL; xmlNode *operation = NULL; retry: for (operation = __xml_first_child_element(rsc->ops_xml); operation != NULL; operation = __xml_next_element(operation)) { if (crm_str_eq((const char *)operation->name, "op", TRUE)) { name = crm_element_value(operation, "name"); interval_spec = crm_element_value(operation, XML_LRM_ATTR_INTERVAL); value = crm_element_value(operation, "enabled"); if (!include_disabled && value && crm_is_true(value) == FALSE) { continue; } interval_ms = crm_parse_interval_spec(interval_spec); match_key = pcmk__op_key(rsc->id, name, interval_ms); if (safe_str_eq(key, match_key)) { op = operation; } free(match_key); if (rsc->clone_name) { match_key = pcmk__op_key(rsc->clone_name, name, interval_ms); if (safe_str_eq(key, match_key)) { op = operation; } free(match_key); } if (op != NULL) { free(local_key); return op; } } } free(local_key); if (do_retry == FALSE) { return NULL; } do_retry = FALSE; if (strstr(key, CRMD_ACTION_MIGRATE) || strstr(key, CRMD_ACTION_MIGRATED)) { local_key = pcmk__op_key(rsc->id, "migrate", 0); key = local_key; goto retry; } else if (strstr(key, "_notify_")) { local_key = pcmk__op_key(rsc->id, "notify", 0); key = local_key; goto retry; } return NULL; } xmlNode * find_rsc_op_entry(pe_resource_t * rsc, const char *key) { return find_rsc_op_entry_helper(rsc, key, FALSE); } void print_node(const char *pre_text, pe_node_t * node, gboolean details) { if (node == NULL) { crm_trace("%s%s: ", pre_text == NULL ? "" : pre_text, pre_text == NULL ? "" : ": "); return; } CRM_ASSERT(node->details); crm_trace("%s%s%sNode %s: (weight=%d, fixed=%s)", pre_text == NULL ? "" : pre_text, pre_text == NULL ? "" : ": ", node->details->online ? "" : "Unavailable/Unclean ", node->details->uname, node->weight, node->fixed ? "True" : "False"); if (details) { int log_level = LOG_TRACE; char *pe_mutable = strdup("\t\t"); GListPtr gIter = node->details->running_rsc; crm_trace("\t\t===Node Attributes"); g_hash_table_foreach(node->details->attrs, print_str_str, pe_mutable); free(pe_mutable); crm_trace("\t\t=== Resources"); for (; gIter != NULL; gIter = gIter->next) { pe_resource_t *rsc = (pe_resource_t *) gIter->data; rsc->fns->print(rsc, "\t\t", pe_print_log|pe_print_pending, &log_level); } } } /* * Used by the HashTable for-loop */ void print_str_str(gpointer key, gpointer value, gpointer user_data) { crm_trace("%s%s %s ==> %s", user_data == NULL ? "" : (char *)user_data, user_data == NULL ? "" : ": ", (char *)key, (char *)value); } void pe_free_action(pe_action_t * action) { if (action == NULL) { return; } g_list_free_full(action->actions_before, free); /* pe_action_wrapper_t* */ g_list_free_full(action->actions_after, free); /* pe_action_wrapper_t* */ if (action->extra) { g_hash_table_destroy(action->extra); } if (action->meta) { g_hash_table_destroy(action->meta); } #if ENABLE_VERSIONED_ATTRS if (action->rsc) { pe_free_rsc_action_details(action); } #endif free(action->cancel_task); free(action->reason); free(action->task); free(action->uuid); free(action->node); free(action); } GListPtr find_recurring_actions(GListPtr input, pe_node_t * not_on_node) { const char *value = NULL; GListPtr result = NULL; GListPtr gIter = input; CRM_CHECK(input != NULL, return NULL); for (; gIter != NULL; gIter = gIter->next) { pe_action_t *action = (pe_action_t *) gIter->data; value = g_hash_table_lookup(action->meta, XML_LRM_ATTR_INTERVAL_MS); if (value == NULL) { /* skip */ } else if (safe_str_eq(value, "0")) { /* skip */ } else if (safe_str_eq(CRMD_ACTION_CANCEL, action->task)) { /* skip */ } else if (not_on_node == NULL) { crm_trace("(null) Found: %s", action->uuid); result = g_list_prepend(result, action); } else if (action->node == NULL) { /* skip */ } else if (action->node->details != not_on_node->details) { crm_trace("Found: %s", action->uuid); result = g_list_prepend(result, action); } } return result; } enum action_tasks get_complex_task(pe_resource_t * rsc, const char *name, gboolean allow_non_atomic) { enum action_tasks task = text2task(name); if (rsc == NULL) { return task; } else if (allow_non_atomic == FALSE || rsc->variant == pe_native) { switch (task) { case stopped_rsc: case started_rsc: case action_demoted: case action_promoted: crm_trace("Folding %s back into its atomic counterpart for %s", name, rsc->id); return task - 1; default: break; } } return task; } pe_action_t * find_first_action(GListPtr input, const char *uuid, const char *task, pe_node_t * on_node) { GListPtr gIter = NULL; CRM_CHECK(uuid || task, return NULL); for (gIter = input; gIter != NULL; gIter = gIter->next) { pe_action_t *action = (pe_action_t *) gIter->data; if (uuid != NULL && safe_str_neq(uuid, action->uuid)) { continue; } else if (task != NULL && safe_str_neq(task, action->task)) { continue; } else if (on_node == NULL) { return action; } else if (action->node == NULL) { continue; } else if (on_node->details == action->node->details) { return action; } } return NULL; } GListPtr find_actions(GListPtr input, const char *key, const pe_node_t *on_node) { GListPtr gIter = input; GListPtr result = NULL; CRM_CHECK(key != NULL, return NULL); for (; gIter != NULL; gIter = gIter->next) { pe_action_t *action = (pe_action_t *) gIter->data; if (safe_str_neq(key, action->uuid)) { crm_trace("%s does not match action %s", key, action->uuid); 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, on_node->details->uname); action->node = pe__copy_node(on_node); result = g_list_prepend(result, action); } else if (on_node->details == action->node->details) { crm_trace("Action %s on %s matches", key, on_node->details->uname); result = g_list_prepend(result, action); } else { crm_trace("Action %s on node %s does not match requested node %s", key, action->node->details->uname, on_node->details->uname); } } return result; } GList * find_actions_exact(GList *input, const char *key, const pe_node_t *on_node) { GList *result = NULL; CRM_CHECK(key != NULL, return NULL); if (on_node == NULL) { crm_trace("Not searching for action %s because node not specified", key); return NULL; } for (GList *gIter = input; gIter != NULL; gIter = gIter->next) { pe_action_t *action = (pe_action_t *) gIter->data; if (action->node == NULL) { crm_trace("Skipping comparison of %s vs action %s without node", key, action->uuid); } else if (safe_str_neq(key, action->uuid)) { crm_trace("Desired action %s doesn't match %s", key, action->uuid); } else if (safe_str_neq(on_node->details->id, action->node->details->id)) { crm_trace("Action %s desired node ID %s doesn't match %s", key, on_node->details->id, action->node->details->id); } else { crm_trace("Action %s matches", key); 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 pe_resource_t *rsc, const pe_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->actions, key, node); } else { result = find_actions(rsc->actions, key, node); } free(key); return result; } static void resource_node_score(pe_resource_t * rsc, pe_node_t * node, int score, const char *tag) { pe_node_t *match = NULL; if ((rsc->exclusive_discover || (node->rsc_discover_mode == pe_discover_never)) && safe_str_eq(tag, "symmetric_default")) { /* This string comparision may be fragile, but exclusive resources and * exclusive nodes should not have the symmetric_default constraint * applied to them. */ return; } else if (rsc->children) { GListPtr gIter = rsc->children; for (; gIter != NULL; gIter = gIter->next) { pe_resource_t *child_rsc = (pe_resource_t *) gIter->data; resource_node_score(child_rsc, node, score, tag); } } pe_rsc_trace(rsc, "Setting %s for %s on %s: %d", tag, rsc->id, node->details->uname, score); match = pe_hash_table_lookup(rsc->allowed_nodes, node->details->id); if (match == NULL) { match = pe__copy_node(node); g_hash_table_insert(rsc->allowed_nodes, (gpointer) match->details->id, match); } match->weight = pe__add_scores(match->weight, score); } void resource_location(pe_resource_t * rsc, pe_node_t * node, int score, const char *tag, pe_working_set_t * data_set) { if (node != NULL) { resource_node_score(rsc, node, score, tag); } else if (data_set != NULL) { GListPtr gIter = data_set->nodes; for (; gIter != NULL; gIter = gIter->next) { pe_node_t *node_iter = (pe_node_t *) gIter->data; resource_node_score(rsc, node_iter, score, tag); } } else { GHashTableIter iter; pe_node_t *node_iter = NULL; g_hash_table_iter_init(&iter, rsc->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 == -INFINITY) { if (rsc->allocated_to) { crm_info("Deallocating %s from %s", rsc->id, rsc->allocated_to->details->uname); free(rsc->allocated_to); rsc->allocated_to = NULL; } } } #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) gint sort_op_by_callid(gconstpointer a, gconstpointer b) { int a_call_id = -1; int b_call_id = -1; char *a_uuid = NULL; char *b_uuid = NULL; const xmlNode *xml_a = a; const xmlNode *xml_b = b; const char *a_xml_id = crm_element_value(xml_a, XML_ATTR_ID); const char *b_xml_id = crm_element_value(xml_b, XML_ATTR_ID); if (safe_str_eq(a_xml_id, b_xml_id)) { /* We have duplicate 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. */ pe_err("Duplicate lrm_rsc_op entries named %s", a_xml_id); sort_return(0, "duplicate"); } crm_element_value_int(xml_a, XML_LRM_ATTR_CALLID, &a_call_id); crm_element_value_int(xml_b, XML_LRM_ATTR_CALLID, &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 (a_call_id >= 0 && a_call_id < b_call_id) { sort_return(-1, "call id"); } else if (b_call_id >= 0 && a_call_id > b_call_id) { sort_return(1, "call id"); } else if (b_call_id >= 0 && a_call_id == b_call_id) { /* * The op and last_failed_op are the same * Order on last-rc-change */ time_t last_a = -1; time_t last_b = -1; crm_element_value_epoch(xml_a, XML_RSC_OP_LAST_CHANGE, &last_a); crm_element_value_epoch(xml_b, XML_RSC_OP_LAST_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 XML_ATTR_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, XML_ATTR_TRANSITION_MAGIC); const char *b_magic = crm_element_value(xml_b, XML_ATTR_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 (safe_str_neq(a_uuid, b_uuid) || 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")); } time_t get_effective_time(pe_working_set_t * data_set) { if(data_set) { if (data_set->now == NULL) { crm_trace("Recording a new 'now'"); data_set->now = crm_time_new(NULL); } return crm_time_get_seconds_since_epoch(data_set->now); } crm_trace("Defaulting to 'now'"); return time(NULL); } gboolean get_target_role(pe_resource_t * rsc, enum rsc_role_e * role) { enum rsc_role_e local_role = RSC_ROLE_UNKNOWN; const char *value = g_hash_table_lookup(rsc->meta, XML_RSC_ATTR_TARGET_ROLE); CRM_CHECK(role != NULL, return FALSE); if (value == NULL || safe_str_eq("started", value) || safe_str_eq("default", value)) { return FALSE; } local_role = text2role(value); if (local_role == RSC_ROLE_UNKNOWN) { pcmk__config_err("Ignoring '" XML_RSC_ATTR_TARGET_ROLE "' for %s " "because '%s' is not valid", rsc->id, value); return FALSE; } else if (local_role > RSC_ROLE_STARTED) { if (is_set(uber_parent(rsc)->flags, pe_rsc_promotable)) { if (local_role > RSC_ROLE_SLAVE) { /* 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 '" XML_RSC_ATTR_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(pe_action_t * lh_action, pe_action_t * rh_action, enum pe_ordering order) { GListPtr gIter = NULL; pe_action_wrapper_t *wrapper = NULL; GListPtr list = NULL; if (order == pe_order_none) { return FALSE; } if (lh_action == NULL || rh_action == NULL) { return FALSE; } crm_trace("Ordering Action %s before %s", lh_action->uuid, rh_action->uuid); /* Ensure we never create a dependency on ourselves... it's happened */ CRM_ASSERT(lh_action != rh_action); /* Filter dups, otherwise update_action_states() has too much work to do */ gIter = lh_action->actions_after; for (; gIter != NULL; gIter = gIter->next) { pe_action_wrapper_t *after = (pe_action_wrapper_t *) gIter->data; if (after->action == rh_action && (after->type & order)) { return FALSE; } } wrapper = calloc(1, sizeof(pe_action_wrapper_t)); wrapper->action = rh_action; wrapper->type = order; list = lh_action->actions_after; list = g_list_prepend(list, wrapper); lh_action->actions_after = list; wrapper = NULL; /* order |= pe_order_implies_then; */ /* order ^= pe_order_implies_then; */ wrapper = calloc(1, sizeof(pe_action_wrapper_t)); wrapper->action = lh_action; wrapper->type = order; list = rh_action->actions_before; list = g_list_prepend(list, wrapper); rh_action->actions_before = list; return TRUE; } pe_action_t * get_pseudo_op(const char *name, pe_working_set_t * data_set) { pe_action_t *op = NULL; if(data_set->singletons) { op = g_hash_table_lookup(data_set->singletons, name); } if (op == NULL) { op = custom_action(NULL, strdup(name), name, NULL, TRUE, TRUE, data_set); set_bit(op->flags, pe_action_pseudo); set_bit(op->flags, pe_action_runnable); } return op; } void destroy_ticket(gpointer data) { pe_ticket_t *ticket = data; if (ticket->state) { g_hash_table_destroy(ticket->state); } free(ticket->id); free(ticket); } pe_ticket_t * ticket_new(const char *ticket_id, pe_working_set_t * data_set) { pe_ticket_t *ticket = NULL; if (pcmk__str_empty(ticket_id)) { return NULL; } if (data_set->tickets == NULL) { data_set->tickets = g_hash_table_new_full(crm_str_hash, g_str_equal, free, destroy_ticket); } ticket = g_hash_table_lookup(data_set->tickets, ticket_id); if (ticket == NULL) { ticket = calloc(1, sizeof(pe_ticket_t)); if (ticket == NULL) { crm_err("Cannot allocate ticket '%s'", ticket_id); return NULL; } crm_trace("Creaing ticket entry for %s", ticket_id); ticket->id = strdup(ticket_id); ticket->granted = FALSE; ticket->last_granted = -1; ticket->standby = FALSE; ticket->state = crm_str_table_new(); g_hash_table_insert(data_set->tickets, strdup(ticket->id), ticket); } return ticket; } static void filter_parameters(xmlNode * param_set, const char *param_string, bool need_present) { if (param_set && param_string) { xmlAttrPtr xIter = param_set->properties; while (xIter) { const char *prop_name = (const char *)xIter->name; char *name = crm_strdup_printf(" %s ", prop_name); char *match = strstr(param_string, name); free(name); // Do now, because current entry might get removed below xIter = xIter->next; if (need_present && match == NULL) { crm_trace("%s not found in %s", prop_name, param_string); xml_remove_prop(param_set, prop_name); } else if (need_present == FALSE && match) { crm_trace("%s found in %s", prop_name, param_string); xml_remove_prop(param_set, prop_name); } } } } #if ENABLE_VERSIONED_ATTRS static void append_versioned_params(xmlNode *versioned_params, const char *ra_version, xmlNode *params) { GHashTable *hash = pe_unpack_versioned_parameters(versioned_params, ra_version); char *key = NULL; char *value = NULL; GHashTableIter iter; g_hash_table_iter_init(&iter, hash); while (g_hash_table_iter_next(&iter, (gpointer *) &key, (gpointer *) &value)) { crm_xml_add(params, key, value); } g_hash_table_destroy(hash); } #endif /*! * \internal * \brief Calculate action digests and store in node's digest cache * * \param[in] rsc Resource that action was for * \param[in] task Name of action performed * \param[in] key Action's task key * \param[in] node Node action was performed on * \param[in] xml_op XML of operation in CIB status (if available) * \param[in] calc_secure Whether to calculate secure digest * \param[in] data_set Cluster working set * * \return Pointer to node's digest cache entry */ static op_digest_cache_t * rsc_action_digest(pe_resource_t *rsc, const char *task, const char *key, pe_node_t *node, xmlNode *xml_op, bool calc_secure, pe_working_set_t *data_set) { op_digest_cache_t *data = NULL; data = g_hash_table_lookup(node->details->digest_cache, key); if (data == NULL) { GHashTable *local_rsc_params = crm_str_table_new(); pe_action_t *action = custom_action(rsc, strdup(key), task, node, TRUE, FALSE, data_set); #if ENABLE_VERSIONED_ATTRS xmlNode *local_versioned_params = create_xml_node(NULL, XML_TAG_RSC_VER_ATTRS); const char *ra_version = NULL; #endif const char *op_version; const char *restart_list = NULL; const char *secure_list = " passwd password "; data = calloc(1, sizeof(op_digest_cache_t)); CRM_ASSERT(data != NULL); get_rsc_attributes(local_rsc_params, rsc, node, data_set); #if ENABLE_VERSIONED_ATTRS pe_get_versioned_attributes(local_versioned_params, rsc, node, data_set); #endif data->params_all = create_xml_node(NULL, XML_TAG_PARAMS); // REMOTE_CONTAINER_HACK: Allow remote nodes that start containers with pacemaker remote inside if (pe__add_bundle_remote_name(rsc, data->params_all, XML_RSC_ATTR_REMOTE_RA_ADDR)) { crm_trace("Set address for bundle connection %s (on %s)", rsc->id, node->details->uname); } g_hash_table_foreach(local_rsc_params, hash2field, data->params_all); g_hash_table_foreach(action->extra, hash2field, data->params_all); g_hash_table_foreach(rsc->parameters, hash2field, data->params_all); g_hash_table_foreach(action->meta, hash2metafield, data->params_all); if(xml_op) { secure_list = crm_element_value(xml_op, XML_LRM_ATTR_OP_SECURE); restart_list = crm_element_value(xml_op, XML_LRM_ATTR_OP_RESTART); op_version = crm_element_value(xml_op, XML_ATTR_CRM_VERSION); #if ENABLE_VERSIONED_ATTRS ra_version = crm_element_value(xml_op, XML_ATTR_RA_VERSION); #endif } else { op_version = CRM_FEATURE_SET; } #if ENABLE_VERSIONED_ATTRS append_versioned_params(local_versioned_params, ra_version, data->params_all); append_versioned_params(rsc->versioned_parameters, ra_version, data->params_all); { pe_rsc_action_details_t *details = pe_rsc_action_details(action); append_versioned_params(details->versioned_parameters, ra_version, data->params_all); } #endif pcmk__filter_op_for_digest(data->params_all); g_hash_table_destroy(local_rsc_params); pe_free_action(action); data->digest_all_calc = calculate_operation_digest(data->params_all, op_version); if (calc_secure) { data->params_secure = copy_xml(data->params_all); if(secure_list) { filter_parameters(data->params_secure, secure_list, FALSE); } data->digest_secure_calc = calculate_operation_digest(data->params_secure, op_version); } if(xml_op && crm_element_value(xml_op, XML_LRM_ATTR_RESTART_DIGEST) != NULL) { data->params_restart = copy_xml(data->params_all); if (restart_list) { filter_parameters(data->params_restart, restart_list, TRUE); } data->digest_restart_calc = calculate_operation_digest(data->params_restart, op_version); } g_hash_table_insert(node->details->digest_cache, strdup(key), data); } return data; } op_digest_cache_t * rsc_action_digest_cmp(pe_resource_t * rsc, xmlNode * xml_op, pe_node_t * node, pe_working_set_t * data_set) { op_digest_cache_t *data = NULL; char *key = NULL; guint interval_ms = 0; const char *op_version; const char *task = crm_element_value(xml_op, XML_LRM_ATTR_TASK); const char *digest_all; const char *digest_restart; CRM_ASSERT(node != NULL); op_version = crm_element_value(xml_op, XML_ATTR_CRM_VERSION); digest_all = crm_element_value(xml_op, XML_LRM_ATTR_OP_DIGEST); digest_restart = crm_element_value(xml_op, XML_LRM_ATTR_RESTART_DIGEST); crm_element_value_ms(xml_op, XML_LRM_ATTR_INTERVAL_MS, &interval_ms); key = pcmk__op_key(rsc->id, task, interval_ms); data = rsc_action_digest(rsc, task, key, node, xml_op, is_set(data_set->flags, pe_flag_sanitized), data_set); data->rc = RSC_DIGEST_MATCH; if (digest_restart && data->digest_restart_calc && strcmp(data->digest_restart_calc, digest_restart) != 0) { pe_rsc_info(rsc, "Parameters to %s on %s changed: was %s vs. now %s (restart:%s) %s", key, node->details->uname, crm_str(digest_restart), data->digest_restart_calc, op_version, crm_element_value(xml_op, XML_ATTR_TRANSITION_MAGIC)); data->rc = RSC_DIGEST_RESTART; } else if (digest_all == NULL) { /* it is unknown what the previous op digest was */ data->rc = RSC_DIGEST_UNKNOWN; } else if (strcmp(digest_all, data->digest_all_calc) != 0) { pe_rsc_info(rsc, "Parameters to %s on %s changed: was %s vs. now %s (%s:%s) %s", key, node->details->uname, crm_str(digest_all), data->digest_all_calc, (interval_ms > 0)? "reschedule" : "reload", op_version, crm_element_value(xml_op, XML_ATTR_TRANSITION_MAGIC)); data->rc = RSC_DIGEST_ALL; } free(key); return data; } /*! * \internal * \brief Create an unfencing summary for use in special node attribute * * Create a string combining a fence device's resource ID, agent type, and * parameter digest (whether for all parameters or just non-private parameters). * This can be stored in a special node attribute, allowing us to detect changes * in either the agent type or parameters, to know whether unfencing must be * redone or can be safely skipped when the device's history is cleaned. * * \param[in] rsc_id Fence device resource ID * \param[in] agent_type Fence device agent * \param[in] param_digest Fence device parameter digest * * \return Newly allocated string with unfencing digest * \note The caller is responsible for freeing the result. */ static inline char * create_unfencing_summary(const char *rsc_id, const char *agent_type, const char *param_digest) { return crm_strdup_printf("%s:%s:%s", rsc_id, agent_type, param_digest); } /*! * \internal * \brief Check whether a node can skip unfencing * * Check whether a fence device's current definition matches a node's * stored summary of when it was last unfenced by the device. * * \param[in] rsc_id Fence device's resource ID * \param[in] agent Fence device's agent type * \param[in] digest_calc Fence device's current parameter digest * \param[in] node_summary Value of node's special unfencing node attribute * (a comma-separated list of unfencing summaries for * all devices that have unfenced this node) * * \return TRUE if digest matches, FALSE otherwise */ static bool unfencing_digest_matches(const char *rsc_id, const char *agent, const char *digest_calc, const char *node_summary) { bool matches = FALSE; if (rsc_id && agent && digest_calc && node_summary) { char *search_secure = create_unfencing_summary(rsc_id, agent, digest_calc); /* The digest was calculated including the device ID and agent, * so there is no risk of collision using strstr(). */ matches = (strstr(node_summary, search_secure) != NULL); crm_trace("Calculated unfencing digest '%s' %sfound in '%s'", search_secure, matches? "" : "not ", node_summary); free(search_secure); } return matches; } /* Magic string to use as action name for digest cache entries used for * unfencing checks. This is not a real action name (i.e. "on"), so * check_action_definition() won't confuse these entries with real actions. */ #define STONITH_DIGEST_TASK "stonith-on" /*! * \internal * \brief Calculate fence device digests and digest comparison result * * \param[in] rsc Fence device resource * \param[in] agent Fence device's agent type * \param[in] node Node with digest cache to use * \param[in] data_set Cluster working set * * \return Node's digest cache entry */ static op_digest_cache_t * fencing_action_digest_cmp(pe_resource_t *rsc, const char *agent, pe_node_t *node, pe_working_set_t *data_set) { const char *node_summary = NULL; // Calculate device's current parameter digests char *key = pcmk__op_key(rsc->id, STONITH_DIGEST_TASK, 0); op_digest_cache_t *data = rsc_action_digest(rsc, STONITH_DIGEST_TASK, key, node, NULL, TRUE, data_set); free(key); // Check whether node has special unfencing summary node attribute node_summary = pe_node_attribute_raw(node, CRM_ATTR_DIGESTS_ALL); if (node_summary == NULL) { data->rc = RSC_DIGEST_UNKNOWN; return data; } // Check whether full parameter digest matches if (unfencing_digest_matches(rsc->id, agent, data->digest_all_calc, node_summary)) { data->rc = RSC_DIGEST_MATCH; return data; } // Check whether secure parameter digest matches node_summary = pe_node_attribute_raw(node, CRM_ATTR_DIGESTS_SECURE); if (unfencing_digest_matches(rsc->id, agent, data->digest_secure_calc, node_summary)) { data->rc = RSC_DIGEST_MATCH; if (is_set(data_set->flags, pe_flag_stdout)) { printf("Only 'private' parameters to %s for unfencing %s changed\n", rsc->id, node->details->uname); } return data; } // Parameters don't match data->rc = RSC_DIGEST_ALL; if (is_set(data_set->flags, (pe_flag_sanitized|pe_flag_stdout)) && data->digest_secure_calc) { char *digest = create_unfencing_summary(rsc->id, agent, data->digest_secure_calc); printf("Parameters to %s for unfencing %s changed, try '%s'\n", rsc->id, node->details->uname, digest); free(digest); } return data; } const char *rsc_printable_id(pe_resource_t *rsc) { if (is_not_set(rsc->flags, pe_rsc_unique)) { return ID(rsc->xml); } return rsc->id; } void clear_bit_recursive(pe_resource_t * rsc, unsigned long long flag) { GListPtr gIter = rsc->children; clear_bit(rsc->flags, flag); for (; gIter != NULL; gIter = gIter->next) { pe_resource_t *child_rsc = (pe_resource_t *) gIter->data; clear_bit_recursive(child_rsc, flag); } } void set_bit_recursive(pe_resource_t * rsc, unsigned long long flag) { GListPtr gIter = rsc->children; set_bit(rsc->flags, flag); for (; gIter != NULL; gIter = gIter->next) { pe_resource_t *child_rsc = (pe_resource_t *) gIter->data; set_bit_recursive(child_rsc, flag); } } static GListPtr find_unfencing_devices(GListPtr candidates, GListPtr matches) { for (GListPtr gIter = candidates; gIter != NULL; gIter = gIter->next) { pe_resource_t *candidate = gIter->data; const char *provides = g_hash_table_lookup(candidate->meta, XML_RSC_ATTR_PROVIDES); const char *requires = g_hash_table_lookup(candidate->meta, XML_RSC_ATTR_REQUIRES); if(candidate->children) { matches = find_unfencing_devices(candidate->children, matches); } else if (is_not_set(candidate->flags, pe_rsc_fence_device)) { continue; } else if (crm_str_eq(provides, "unfencing", FALSE) || crm_str_eq(requires, "unfencing", FALSE)) { matches = g_list_prepend(matches, candidate); } } return matches; } static int node_priority_fencing_delay(pe_node_t * node, pe_working_set_t * data_set) { int member_count = 0; int online_count = 0; int top_priority = 0; int lowest_priority = 0; GListPtr gIter = NULL; // `priority-fencing-delay` is disabled if (data_set->priority_fencing_delay <= 0) { 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->details->type != node_member) { return 0; } // No need to request a delay if the fencing target is in our partition if (node->details->online) { return 0; } for (gIter = data_set->nodes; gIter != NULL; gIter = gIter->next) { pe_node_t *n = gIter->data; if (n->details->type != node_member) { continue; } member_count ++; if (n->details->online) { online_count++; } if (member_count == 1 || n->details->priority > top_priority) { top_priority = n->details->priority; } if (member_count == 1 || n->details->priority < lowest_priority) { lowest_priority = n->details->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->details->priority < top_priority) { return 0; } return data_set->priority_fencing_delay; } pe_action_t * pe_fence_op(pe_node_t * node, const char *op, bool optional, const char *reason, bool priority_delay, pe_working_set_t * data_set) { char *op_key = NULL; pe_action_t *stonith_op = NULL; if(op == NULL) { op = data_set->stonith_action; } op_key = crm_strdup_printf("%s-%s-%s", CRM_OP_FENCE, node->details->uname, op); if(data_set->singletons) { stonith_op = g_hash_table_lookup(data_set->singletons, op_key); } if(stonith_op == NULL) { stonith_op = custom_action(NULL, op_key, CRM_OP_FENCE, node, TRUE, TRUE, data_set); add_hash_param(stonith_op->meta, XML_LRM_ATTR_TARGET, node->details->uname); add_hash_param(stonith_op->meta, XML_LRM_ATTR_TARGET_UUID, node->details->id); add_hash_param(stonith_op->meta, "stonith_action", op); if (pe__is_guest_or_remote_node(node) && is_set(data_set->flags, pe_flag_enable_unfencing)) { /* Extra work to detect device changes on remotes * * We may do this for all nodes in the future, but for now * the check_action_definition() based stuff works fine. */ long max = 1024; long digests_all_offset = 0; long digests_secure_offset = 0; char *digests_all = calloc(max, sizeof(char)); char *digests_secure = calloc(max, sizeof(char)); GListPtr matches = find_unfencing_devices(data_set->resources, NULL); for (GListPtr gIter = matches; gIter != NULL; gIter = gIter->next) { pe_resource_t *match = gIter->data; const char *agent = g_hash_table_lookup(match->meta, XML_ATTR_TYPE); op_digest_cache_t *data = NULL; data = fencing_action_digest_cmp(match, agent, node, data_set); if(data->rc == RSC_DIGEST_ALL) { optional = FALSE; crm_notice("Unfencing %s (remote): because the definition of %s changed", node->details->uname, match->id); if (is_set(data_set->flags, pe_flag_stdout)) { fprintf(stdout, " notice: Unfencing %s (remote): because the definition of %s changed\n", node->details->uname, match->id); } } digests_all_offset += snprintf( digests_all+digests_all_offset, max-digests_all_offset, "%s:%s:%s,", match->id, agent, data->digest_all_calc); digests_secure_offset += snprintf( digests_secure+digests_secure_offset, max-digests_secure_offset, "%s:%s:%s,", match->id, agent, data->digest_secure_calc); } g_hash_table_insert(stonith_op->meta, strdup(XML_OP_ATTR_DIGESTS_ALL), digests_all); g_hash_table_insert(stonith_op->meta, strdup(XML_OP_ATTR_DIGESTS_SECURE), digests_secure); } } else { free(op_key); } if (data_set->priority_fencing_delay > 0 /* It's a suitable case where `priority-fencing-delay` applies. * At least add `priority-fencing-delay` field as an indicator. */ && (priority_delay /* Re-calculate priority delay for the suitable case when * pe_fence_op() is called again by stage6() after node priority has * been actually calculated with native_add_running() */ || g_hash_table_lookup(stonith_op->meta, XML_CONFIG_ATTR_PRIORITY_FENCING_DELAY) != NULL)) { /* Add `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 = crm_itoa(node_priority_fencing_delay(node, data_set)); g_hash_table_insert(stonith_op->meta, strdup(XML_CONFIG_ATTR_PRIORITY_FENCING_DELAY), delay_s); } if(optional == FALSE && pe_can_fence(data_set, node)) { pe_action_required(stonith_op, NULL, reason); } else if(reason && stonith_op->reason == NULL) { stonith_op->reason = strdup(reason); } return stonith_op; } void trigger_unfencing( pe_resource_t * rsc, pe_node_t *node, const char *reason, pe_action_t *dependency, pe_working_set_t * data_set) { if(is_not_set(data_set->flags, pe_flag_enable_unfencing)) { /* No resources require it */ return; } else if (rsc != NULL && is_not_set(rsc->flags, pe_rsc_fence_device)) { /* Wasn't a stonith device */ return; } else if(node && node->details->online && node->details->unclean == FALSE && node->details->shutdown == FALSE) { pe_action_t *unfence = pe_fence_op(node, "on", FALSE, reason, FALSE, data_set); if(dependency) { order_actions(unfence, dependency, pe_order_optional); } } else if(rsc) { GHashTableIter iter; g_hash_table_iter_init(&iter, rsc->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, data_set); } } } } gboolean add_tag_ref(GHashTable * tags, const char * tag_name, const char * obj_ref) { pe_tag_t *tag = NULL; GListPtr gIter = NULL; gboolean is_existing = FALSE; CRM_CHECK(tags && tag_name && obj_ref, return FALSE); tag = g_hash_table_lookup(tags, tag_name); if (tag == NULL) { tag = calloc(1, sizeof(pe_tag_t)); if (tag == NULL) { return FALSE; } tag->id = strdup(tag_name); tag->refs = NULL; g_hash_table_insert(tags, strdup(tag_name), tag); } for (gIter = tag->refs; gIter != NULL; gIter = gIter->next) { const char *existing_ref = (const char *) gIter->data; if (crm_str_eq(existing_ref, obj_ref, TRUE)){ is_existing = TRUE; break; } } if (is_existing == FALSE) { tag->refs = g_list_append(tag->refs, strdup(obj_ref)); crm_trace("Added: tag=%s ref=%s", tag->id, obj_ref); } return TRUE; } void pe_action_set_flag_reason(const char *function, long line, pe_action_t *action, pe_action_t *reason, const char *text, enum pe_action_flags flags, bool overwrite) { bool unset = FALSE; bool update = FALSE; const char *change = NULL; if(is_set(flags, pe_action_runnable)) { unset = TRUE; change = "unrunnable"; } else if(is_set(flags, pe_action_optional)) { unset = TRUE; change = "required"; } else if(is_set(flags, pe_action_migrate_runnable)) { unset = TRUE; overwrite = TRUE; change = "unrunnable"; } else if(is_set(flags, pe_action_dangle)) { change = "dangling"; } else if(is_set(flags, pe_action_requires_any)) { change = "required"; } else { crm_err("Unknown flag change to %x by %s: 0x%s", flags, action->uuid, (reason? reason->uuid : "0")); } if(unset) { if(is_set(action->flags, flags)) { action->flags = crm_clear_bit(function, line, action->uuid, action->flags, flags); update = TRUE; } } else { if(is_not_set(action->flags, flags)) { action->flags = crm_set_bit(function, line, action->uuid, action->flags, flags); update = TRUE; } } if((change && update) || text) { char *reason_text = NULL; if(reason == NULL) { pe_action_set_reason(action, text, overwrite); } else if(reason->rsc == NULL) { reason_text = crm_strdup_printf("%s %s%c %s", change, reason->task, text?':':0, text?text:""); } else { reason_text = crm_strdup_printf("%s %s %s%c %s", change, reason->rsc->id, reason->task, text?':':0, text?text:"NA"); } if(reason_text && action->rsc != reason->rsc) { pe_action_set_reason(action, reason_text, overwrite); } free(reason_text); } } void pe_action_set_reason(pe_action_t *action, const char *reason, bool overwrite) { if (action->reason != NULL && overwrite) { pe_rsc_trace(action->rsc, "Changing %s reason from '%s' to '%s'", action->uuid, action->reason, crm_str(reason)); free(action->reason); } else if (action->reason == NULL) { pe_rsc_trace(action->rsc, "Set %s reason to '%s'", action->uuid, crm_str(reason)); } else { // crm_assert(action->reason != NULL && !overwrite); return; } if (reason != NULL) { action->reason = strdup(reason); } else { action->reason = NULL; } } /*! * \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(pe_node_t *node) { const char *shutdown = pe_node_attribute_raw(node, XML_CIB_ATTR_SHUTDOWN); return shutdown && strcmp(shutdown, "0"); } /*! * \internal * \brief Update a data set's "recheck by" time * * \param[in] recheck Epoch time when recheck should happen * \param[in,out] data_set Current working set */ void pe__update_recheck_time(time_t recheck, pe_working_set_t *data_set) { if ((recheck > get_effective_time(data_set)) && ((data_set->recheck_by == 0) || (data_set->recheck_by > recheck))) { data_set->recheck_by = recheck; } } /*! * \internal * \brief Wrapper for pe_unpack_nvpairs() using a cluster working set */ void pe__unpack_dataset_nvpairs(xmlNode *xml_obj, const char *set_name, pe_rule_eval_data_t *rule_data, GHashTable *hash, const char *always_first, gboolean overwrite, pe_working_set_t *data_set) { crm_time_t *next_change = crm_time_new_undefined(); pe_eval_nvpairs(data_set->input, xml_obj, set_name, rule_data, hash, always_first, overwrite, 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, data_set); } crm_time_free(next_change); } bool pe__resource_is_disabled(pe_resource_t *rsc) { const char *target_role = NULL; CRM_CHECK(rsc != NULL, return false); target_role = g_hash_table_lookup(rsc->meta, XML_RSC_ATTR_TARGET_ROLE); if (target_role) { enum rsc_role_e target_role_e = text2role(target_role); if ((target_role_e == RSC_ROLE_STOPPED) || ((target_role_e == RSC_ROLE_SLAVE) && is_set(uber_parent(rsc)->flags, pe_rsc_promotable))) { return true; } } return false; } /*! * \internal * \brief Create an action to clear a resource's history from CIB * * \param[in] rsc Resource to clear * \param[in] node Node to clear history on * * \return New action to clear resource history */ pe_action_t * pe__clear_resource_history(pe_resource_t *rsc, pe_node_t *node, pe_working_set_t *data_set) { char *key = NULL; CRM_ASSERT(rsc && node); key = pcmk__op_key(rsc->id, CRM_OP_LRM_DELETE, 0); return custom_action(rsc, key, CRM_OP_LRM_DELETE, node, FALSE, TRUE, data_set); } bool pe__rsc_running_on_any_node_in_list(pe_resource_t *rsc, GListPtr node_list) { for (GListPtr ele = rsc->running_on; ele; ele = ele->next) { pe_node_t *node = (pe_node_t *) ele->data; if (pcmk__str_in_list(node_list, node->details->uname)) { return true; } } return false; } bool pcmk__rsc_filtered_by_node(pe_resource_t *rsc, GListPtr only_node) { return (rsc->fns->active(rsc, FALSE) && !pe__rsc_running_on_any_node_in_list(rsc, only_node)); } GListPtr pe__filter_rsc_list(GListPtr rscs, GListPtr filter) { GListPtr retval = NULL; for (GListPtr gIter = rscs; gIter; gIter = gIter->next) { pe_resource_t *rsc = (pe_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(filter, rsc_printable_id(rsc)) || (rsc->parent && pcmk__str_in_list(filter, rsc_printable_id(rsc->parent)))) { retval = g_list_prepend(retval, rsc); } } return retval; }