diff --git a/include/crm/cib/cib_types.h b/include/crm/cib/cib_types.h index 6e25f1851b..9bd49e84db 100644 --- a/include/crm/cib/cib_types.h +++ b/include/crm/cib/cib_types.h @@ -1,379 +1,374 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #ifndef PCMK__CRM_CIB_CIB_TYPES__H # define PCMK__CRM_CIB_CIB_TYPES__H # include // gboolean, GList # include // xmlNode # include # include #ifdef __cplusplus extern "C" { #endif /** * \file * \brief Data types for Cluster Information Base access * \ingroup cib */ enum cib_variant { cib_undefined = 0, cib_native = 1, cib_file = 2, cib_remote = 3, - -#if !defined(PCMK_ALLOW_DEPRECATED) || (PCMK_ALLOW_DEPRECATED == 1) - //! \deprecated This value will be removed in a future release - cib_database = 4, -#endif // !defined(PCMK_ALLOW_DEPRECATED) || (PCMK_ALLOW_DEPRECATED == 1) }; enum cib_state { // NOTE: sbd (as of at least 1.5.2) uses this value cib_connected_command, // NOTE: sbd (as of at least 1.5.2) uses this value cib_connected_query, cib_disconnected }; enum cib_conn_type { cib_command, // NOTE: sbd (as of at least 1.5.2) uses this value cib_query, cib_no_connection, cib_command_nonblocking, }; enum cib_call_options { cib_none = 0, cib_verbose = (1 << 0), //!< Prefer stderr to logs cib_xpath = (1 << 1), cib_multiple = (1 << 2), cib_can_create = (1 << 3), cib_discard_reply = (1 << 4), cib_no_children = (1 << 5), cib_xpath_address = (1 << 6), #if !defined(PCMK_ALLOW_DEPRECATED) || (PCMK_ALLOW_DEPRECATED == 1) // NOTE: sbd (as of at least 1.5.2) uses this value //! \deprecated This value will be removed in a future release cib_scope_local = (1 << 8), #endif // !defined(PCMK_ALLOW_DEPRECATED) || (PCMK_ALLOW_DEPRECATED == 1) cib_dryrun = (1 << 9), /*! * \brief Process request when the client commits the active transaction * * Add the request to the client's active transaction instead of processing * it immediately. If the client has no active transaction, or if the * request is not supported in transactions, the call will fail. * * The request is added to the transaction synchronously, and the return * value indicates whether it was added successfully. * * Refer to \p cib_api_operations_t:init_transaction() and * \p cib_api_operations_t:end_transaction() for more details on CIB * transactions. */ cib_transaction = (1 << 10), /*! * \brief Treat new attribute values as atomic score updates where possible * * This option takes effect when updating XML attributes. For an attribute * named \c "name", if the new value is \c "name++" or \c "name+=X" for some * score \c X, the new value is set as follows: * * If attribute \c "name" is not already set to some value in the element * being updated, the new value is set as a literal string. * * If the new value is \c "name++", then the attribute is set to its * existing value (parsed as a score) plus 1. * * If the new value is \c "name+=X" for some score \c X, then the * attribute is set to its existing value plus \c X, where the existing * value and \c X are parsed and added as scores. * * Scores are integer values capped at \c INFINITY and \c -INFINITY. Refer * to Pacemaker Explained and to the \c char2score() function for more * details on scores, including how they're parsed and added. * * Note: This is implemented only for modify operations. */ cib_score_update = (1 << 11), // NOTE: sbd (as of at least 1.5.2) uses this value cib_sync_call = (1 << 12), cib_no_mtime = (1 << 13), cib_inhibit_notify = (1 << 16), cib_force_diff = (1 << 28), }; typedef struct cib_s cib_t; typedef struct cib_api_operations_s { // NOTE: sbd (as of at least 1.5.2) uses this int (*signon) (cib_t *cib, const char *name, enum cib_conn_type type); //! \deprecated This method will be removed and should not be used int (*signon_raw) (cib_t *cib, const char *name, enum cib_conn_type type, int *event_fd); // NOTE: sbd (as of at least 1.5.2) uses this int (*signoff) (cib_t *cib); int (*free) (cib_t *cib); //! \deprecated This method will be removed and should not be used int (*set_op_callback) (cib_t *cib, void (*callback) (const xmlNode *msg, int callid, int rc, xmlNode *output)); // NOTE: sbd (as of at least 1.5.2) uses this int (*add_notify_callback) (cib_t *cib, const char *event, void (*callback) (const char *event, xmlNode *msg)); // NOTE: sbd (as of at least 1.5.2) uses this int (*del_notify_callback) (cib_t *cib, const char *event, void (*callback) (const char *event, xmlNode *msg)); // NOTE: sbd (as of at least 1.5.2) uses this int (*set_connection_dnotify) (cib_t *cib, void (*dnotify) (gpointer user_data)); //! \deprecated This method will be removed and should not be used int (*inputfd) (cib_t *cib); // NOTE: sbd (as of at least 1.5.2) uses this //! \deprecated This method will be removed and should not be used int (*noop) (cib_t *cib, int call_options); int (*ping) (cib_t *cib, xmlNode **output_data, int call_options); // NOTE: sbd (as of at least 1.5.2) uses this int (*query) (cib_t *cib, const char *section, xmlNode **output_data, int call_options); int (*query_from) (cib_t *cib, const char *host, const char *section, xmlNode **output_data, int call_options); //! \deprecated This method will be removed and should not be used int (*is_master) (cib_t *cib); //! \deprecated Use the set_primary() method instead int (*set_master) (cib_t *cib, int call_options); //! \deprecated Use the set_secondary() method instead int (*set_slave) (cib_t *cib, int call_options); //! \deprecated This method will be removed and should not be used int (*set_slave_all) (cib_t *cib, int call_options); int (*sync) (cib_t *cib, const char *section, int call_options); int (*sync_from) (cib_t *cib, const char *host, const char *section, int call_options); int (*upgrade) (cib_t *cib, int call_options); int (*bump_epoch) (cib_t *cib, int call_options); /*! * The \c element in the reply to a failed creation call is * deprecated since 2.1.8. */ int (*create) (cib_t *cib, const char *section, xmlNode *data, int call_options); int (*modify) (cib_t *cib, const char *section, xmlNode *data, int call_options); //! \deprecated Use the \p modify() method instead int (*update) (cib_t *cib, const char *section, xmlNode *data, int call_options); int (*replace) (cib_t *cib, const char *section, xmlNode *data, int call_options); int (*remove) (cib_t *cib, const char *section, xmlNode *data, int call_options); int (*erase) (cib_t *cib, xmlNode **output_data, int call_options); //! \deprecated This method does nothing and should not be called int (*delete_absolute) (cib_t *cib, const char *section, xmlNode *data, int call_options); //! \deprecated This method is not implemented and should not be used int (*quit) (cib_t *cib, int call_options); int (*register_notification) (cib_t *cib, const char *callback, int enabled); gboolean (*register_callback) (cib_t *cib, int call_id, int timeout, gboolean only_success, void *user_data, const char *callback_name, void (*callback) (xmlNode*, int, int, xmlNode*, void *)); gboolean (*register_callback_full)(cib_t *cib, int call_id, int timeout, gboolean only_success, void *user_data, const char *callback_name, void (*callback)(xmlNode *, int, int, xmlNode *, void *), void (*free_func)(void *)); /*! * \brief Set the local CIB manager as the cluster's primary instance * * \param[in,out] cib CIB connection * \param[in] call_options Group of enum cib_call_options flags * * \return Legacy Pacemaker return code (in particular, pcmk_ok on success) */ int (*set_primary)(cib_t *cib, int call_options); /*! * \brief Set the local CIB manager as a secondary instance * * \param[in,out] cib CIB connection * \param[in] call_options Group of enum cib_call_options flags * * \return Legacy Pacemaker return code (in particular, pcmk_ok on success) */ int (*set_secondary)(cib_t *cib, int call_options); /*! * \brief Get the given CIB connection's unique client identifier(s) * * These can be used to check whether this client requested the action that * triggered a CIB notification. * * \param[in] cib CIB connection * \param[out] async_id If not \p NULL, where to store asynchronous client * ID * \param[out] sync_id If not \p NULL, where to store synchronous client * ID * * \return Legacy Pacemaker return code * * \note Some variants may have only one client for both asynchronous and * synchronous requests. */ int (*client_id)(const cib_t *cib, const char **async_id, const char **sync_id); /*! * \brief Initiate an atomic CIB transaction for this client * * If the client has initiated a transaction and a new request's call * options contain \p cib_transaction, the new request is appended to the * transaction for later processing. * * Supported requests are those that meet the following conditions: * * can be processed synchronously (with any changes applied to a working * CIB copy) * * are not queries * * do not involve other nodes * * do not affect the state of pacemaker-based itself * * Currently supported CIB API functions include: * * \p bump_epoch() * * \p create() * * \p erase() * * \p modify() * * \p remove() * * \p replace() * * \p upgrade() * * Because the transaction is atomic, individual requests do not trigger * callbacks or notifications when they are processed, and they do not * receive output XML. The commit request itself can trigger callbacks and * notifications if any are registered. * * An \c init_transaction() call is always synchronous. * * \param[in,out] cib CIB connection * * \return Legacy Pacemaker return code */ int (*init_transaction)(cib_t *cib); /*! * \brief End and optionally commit this client's CIB transaction * * When a client commits a transaction, all requests in the transaction are * processed in a FIFO manner until either a request fails or all requests * have been processed. Changes are applied to a working copy of the CIB. * If a request fails, the transaction and working CIB copy are discarded, * and an error is returned. If all requests succeed, the working CIB copy * replaces the initial CIB copy. * * Callbacks and notifications can be triggered by the commit request itself * but not by the individual requests in a transaction. * * An \c end_transaction() call with \p commit set to \c false is always * synchronous. * * \param[in,out] cib CIB connection * \param[in] commit If \p true, commit transaction; otherwise, * discard it * \param[in] call_options Group of enum cib_call_options * flags * * \return Legacy Pacemaker return code */ int (*end_transaction)(cib_t *cib, bool commit, int call_options); /*! * \brief Set the user as whom all CIB requests via methods will be executed * * By default, the value of the \c CIB_user environment variable is used if * set. Otherwise, the current effective user is used. * * \param[in,out] cib CIB connection * \param[in] user Name of user whose permissions to use when * processing requests */ void (*set_user)(cib_t *cib, const char *user); int (*fetch_schemas)(cib_t *cib, xmlNode **output_data, const char *after_ver, int call_options); } cib_api_operations_t; struct cib_s { // NOTE: sbd (as of at least 1.5.2) uses this enum cib_state state; enum cib_conn_type type; enum cib_variant variant; int call_id; int call_timeout; void *variant_opaque; void *delegate_fn; GList *notify_list; //! \deprecated This method will be removed in a future release void (*op_callback) (const xmlNode *msg, int call_id, int rc, xmlNode *output); // NOTE: sbd (as of at least 1.5.2) uses this cib_api_operations_t *cmds; xmlNode *transaction; char *user; }; #ifdef __cplusplus } #endif #endif // PCMK__CRM_CIB_CIB_TYPES__H diff --git a/include/crm/common/ipc.h b/include/crm/common/ipc.h index 7a58725453..1400bfdc8f 100644 --- a/include/crm/common/ipc.h +++ b/include/crm/common/ipc.h @@ -1,251 +1,246 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #ifndef PCMK__CRM_COMMON_IPC__H #define PCMK__CRM_COMMON_IPC__H #include #include #include #ifdef __cplusplus extern "C" { #endif /** * \file * \brief IPC interface to Pacemaker daemons * * \ingroup core */ /* * Message creation utilities * * These are used for both IPC messages and cluster layer messages. However, * since this is public API, they stay in this header for backward * compatibility. */ // @COMPAT Make internal when we can break API backward compatibility //! \deprecated Do not use #define create_reply(request, xml_response_data) \ create_reply_adv(request, xml_response_data, __func__) // @COMPAT Make internal when we can break API backward compatibility //! \deprecated Do not use xmlNode *create_reply_adv(const xmlNode *request, xmlNode *xml_response_data, const char *origin); // @COMPAT Make internal when we can break API backward compatibility //! \deprecated Do not use #define create_request(task, xml_data, host_to, sys_to, sys_from, uuid_from) \ create_request_adv(task, xml_data, host_to, sys_to, sys_from, uuid_from, \ __func__) // @COMPAT Make internal when we can break API backward compatibility //! \deprecated Do not use xmlNode *create_request_adv(const char *task, xmlNode *xml_data, const char *host_to, const char *sys_to, const char *sys_from, const char *uuid_from, const char *origin); /* * The library supports two methods of creating IPC connections. The older code * allows connecting to any arbitrary IPC name. The newer code only allows * connecting to one of the Pacemaker daemons. * * As daemons are converted to use the new model, the old functions should be * considered deprecated for use with those daemons. Once all daemons are * converted, the old functions should be officially deprecated as public API * and eventually made internal API. */ /* * Pacemaker daemon IPC */ //! Available IPC interfaces enum pcmk_ipc_server { pcmk_ipc_attrd, //!< Attribute manager pcmk_ipc_based, //!< CIB manager pcmk_ipc_controld, //!< Controller pcmk_ipc_execd, //!< Executor pcmk_ipc_fenced, //!< Fencer pcmk_ipc_pacemakerd, //!< Launcher pcmk_ipc_schedulerd, //!< Scheduler }; // NOTE: sbd (as of at least 1.5.2) uses this enum //! Possible event types that an IPC event callback can be called for enum pcmk_ipc_event { pcmk_ipc_event_connect, //!< Result of asynchronous connection attempt // NOTE: sbd (as of at least 1.5.2) uses this value pcmk_ipc_event_disconnect, //!< Termination of IPC connection // NOTE: sbd (as of at least 1.5.2) uses this value pcmk_ipc_event_reply, //!< Daemon's reply to client IPC request pcmk_ipc_event_notify, //!< Notification from daemon }; //! How IPC replies should be dispatched enum pcmk_ipc_dispatch { pcmk_ipc_dispatch_main, //!< Attach IPC to GMainLoop for dispatch pcmk_ipc_dispatch_poll, //!< Caller will poll and dispatch IPC pcmk_ipc_dispatch_sync, //!< Sending a command will wait for any reply }; // NOTE: sbd (as of at least 1.5.2) uses this //! Client connection to Pacemaker IPC typedef struct pcmk_ipc_api_s pcmk_ipc_api_t; /*! * \brief Callback function type for Pacemaker daemon IPC APIs * * \param[in,out] api IPC API connection * \param[in] event_type The type of event that occurred * \param[in] status Event status * \param[in,out] event_data Event-specific data * \param[in,out] user_data Caller data provided when callback was registered * * \note For connection and disconnection events, event_data may be NULL (for * local IPC) or the name of the connected node (for remote IPC, for * daemons that support that). For reply and notify events, event_data is * defined by the specific daemon API. */ typedef void (*pcmk_ipc_callback_t)(pcmk_ipc_api_t *api, enum pcmk_ipc_event event_type, crm_exit_t status, void *event_data, void *user_data); // NOTE: sbd (as of at least 1.5.2) uses this int pcmk_new_ipc_api(pcmk_ipc_api_t **api, enum pcmk_ipc_server server); // NOTE: sbd (as of at least 1.5.2) uses this void pcmk_free_ipc_api(pcmk_ipc_api_t *api); // NOTE: sbd (as of at least 1.5.2) uses this int pcmk_connect_ipc(pcmk_ipc_api_t *api, enum pcmk_ipc_dispatch dispatch_type); void pcmk_disconnect_ipc(pcmk_ipc_api_t *api); int pcmk_poll_ipc(const pcmk_ipc_api_t *api, int timeout_ms); void pcmk_dispatch_ipc(pcmk_ipc_api_t *api); // NOTE: sbd (as of at least 1.5.2) uses this void pcmk_register_ipc_callback(pcmk_ipc_api_t *api, pcmk_ipc_callback_t cb, void *user_data); const char *pcmk_ipc_name(const pcmk_ipc_api_t *api, bool for_log); bool pcmk_ipc_is_connected(pcmk_ipc_api_t *api); int pcmk_ipc_purge_node(pcmk_ipc_api_t *api, const char *node_name, uint32_t nodeid); /* * Generic IPC API (to eventually be deprecated as public API and made internal) */ /* *INDENT-OFF* */ enum crm_ipc_flags { crm_ipc_flags_none = 0x00000000, crm_ipc_compressed = 0x00000001, /* Message has been compressed */ crm_ipc_proxied = 0x00000100, /* _ALL_ replies to proxied connections need to be sent as events */ crm_ipc_client_response = 0x00000200, /* A Response is expected in reply */ // These are options for Pacemaker's internal use only (pcmk__ipc_send_*()) crm_ipc_server_event = 0x00010000, /* Send an Event instead of a Response */ crm_ipc_server_free = 0x00020000, /* Free the iovec after sending */ crm_ipc_proxied_relay_response = 0x00040000, /* all replies to proxied connections are sent as events, this flag preserves whether the event should be treated as an actual event, or a response.*/ - -#if !defined(PCMK_ALLOW_DEPRECATED) || (PCMK_ALLOW_DEPRECATED == 1) - crm_ipc_server_info = 0x00100000, //!< \deprecated Unused - crm_ipc_server_error = 0x00200000, //!< \deprecated Unused -#endif }; /* *INDENT-ON* */ typedef struct crm_ipc_s crm_ipc_t; crm_ipc_t *crm_ipc_new(const char *name, size_t max_size); bool crm_ipc_connect(crm_ipc_t * client); void crm_ipc_close(crm_ipc_t * client); void crm_ipc_destroy(crm_ipc_t * client); void pcmk_free_ipc_event(struct iovec *event); int crm_ipc_send(crm_ipc_t *client, const xmlNode *message, enum crm_ipc_flags flags, int32_t ms_timeout, xmlNode **reply); int crm_ipc_get_fd(crm_ipc_t * client); bool crm_ipc_connected(crm_ipc_t * client); int crm_ipc_ready(crm_ipc_t * client); long crm_ipc_read(crm_ipc_t * client); const char *crm_ipc_buffer(crm_ipc_t * client); uint32_t crm_ipc_buffer_flags(crm_ipc_t * client); const char *crm_ipc_name(crm_ipc_t * client); unsigned int crm_ipc_default_buffer_size(void); /*! * \brief Check the authenticity of the IPC socket peer process (legacy) * * If everything goes well, peer's authenticity is verified by the means * of comparing against provided referential UID and GID (either satisfies), * and the result of this check can be deduced from the return value. * As an exception, detected UID of 0 ("root") satisfies arbitrary * provided referential daemon's credentials. * * \param[in] sock IPC related, connected Unix socket to check peer of * \param[in] refuid referential UID to check against * \param[in] refgid referential GID to check against * \param[out] gotpid to optionally store obtained PID of the peer * (not available on FreeBSD, special value of 1 * used instead, and the caller is required to * special case this value respectively) * \param[out] gotuid to optionally store obtained UID of the peer * \param[out] gotgid to optionally store obtained GID of the peer * * \return 0 if IPC related socket's peer is not authentic given the * referential credentials (see above), 1 if it is, * negative value on error (generally expressing -errno unless * it was zero even on nonhappy path, -pcmk_err_generic is * returned then; no message is directly emitted) * * \note While this function is tolerant on what constitutes authorized * IPC daemon process (its effective user matches UID=0 or \p refuid, * or at least its group matches \p refgid), either or both (in case * of UID=0) mismatches on the expected credentials of such peer * process \e shall be investigated at the caller when value of 1 * gets returned there, since higher-than-expected privileges in * respect to the expected/intended credentials possibly violate * the least privilege principle and may pose an additional risk * (i.e. such accidental inconsistency shall be eventually fixed). */ int crm_ipc_is_authentic_process(int sock, uid_t refuid, gid_t refgid, pid_t *gotpid, uid_t *gotuid, gid_t *gotgid); // @COMPAT Make internal when we can break API backward compatibility //! \deprecated Do not use xmlNode *create_hello_message(const char *uuid, const char *client_name, const char *major_version, const char *minor_version); #ifdef __cplusplus } #endif #endif diff --git a/include/crm/common/nodes.h b/include/crm/common/nodes.h index 4d9a8d93e5..b9515ba104 100644 --- a/include/crm/common/nodes.h +++ b/include/crm/common/nodes.h @@ -1,102 +1,102 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #ifndef PCMK__CRM_COMMON_NODES__H #define PCMK__CRM_COMMON_NODES__H #include // bool #include // gboolean, GList, GHashTable #include // pcmk_resource_t, pcmk_scheduler_t #ifdef __cplusplus extern "C" { #endif /*! * \file * \brief Scheduler API for nodes * \ingroup core */ // Special node attributes #define PCMK_NODE_ATTR_MAINTENANCE "maintenance" #define PCMK_NODE_ATTR_STANDBY "standby" #define PCMK_NODE_ATTR_TERMINATE "terminate" //! \internal Do not use typedef struct pcmk__node_private pcmk__node_private_t; // Basic node information (all node objects for the same node share this) // @COMPAT Drop this struct once all members are moved to pcmk__node_private_t //!@{ //! \deprecated Do not use (public access will be removed in a future release) -struct pe_node_shared_s { +struct pcmk__node_details { /* @COMPAT Convert these gbooleans into new enum pcmk__node_flags values * when we no longer support versions of sbd that use them */ // NOTE: sbd (as of at least 1.5.2) uses this //! \deprecated Call pcmk_node_is_online() instead gboolean online; // Whether online // NOTE: sbd (as of at least 1.5.2) uses this //! \deprecated Call pcmk_node_is_pending() instead gboolean pending; // Whether controller membership is pending // NOTE: sbd (as of at least 1.5.2) uses this //! \deprecated Call !pcmk_node_is_clean() instead gboolean unclean; // Whether node requires fencing // NOTE: sbd (as of at least 1.5.2) uses this //! \deprecated Call pcmk_node_is_shutting_down() instead gboolean shutdown; // Whether shutting down // NOTE: sbd (as of at least 1.5.2) uses this //! \deprecated Call pcmk_node_is_in_maintenance() instead gboolean maintenance; // Whether in maintenance mode // NOTE: sbd (as of at least 1.5.2) uses this // \deprecated Call pcmk_foreach_active_resource() instead GList *running_rsc; // List of resources active on node }; //!@} // Implementation of pcmk_node_t // @COMPAT Make contents internal when we can break API backward compatibility //!@{ //! \deprecated Do not use (public access will be removed in a future release) struct pcmk__scored_node { struct pcmk__node_assignment *assign; // NOTE: sbd (as of at least 1.5.2) uses this - struct pe_node_shared_s *details; // Basic node information + struct pcmk__node_details *details; // Basic node information //! \internal Do not use pcmk__node_private_t *private; }; //!@} bool pcmk_node_is_online(const pcmk_node_t *node); bool pcmk_node_is_pending(const pcmk_node_t *node); bool pcmk_node_is_clean(const pcmk_node_t *node); bool pcmk_node_is_shutting_down(const pcmk_node_t *node); bool pcmk_node_is_in_maintenance(const pcmk_node_t *node); bool pcmk_foreach_active_resource(pcmk_node_t *node, bool (*fn)(pcmk_resource_t *, void *), void *user_data); #ifdef __cplusplus } #endif #endif // PCMK__CRM_COMMON_NODES__H diff --git a/include/crm/common/resources.h b/include/crm/common/resources.h index 5436394102..f00a6f7bdd 100644 --- a/include/crm/common/resources.h +++ b/include/crm/common/resources.h @@ -1,100 +1,80 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #ifndef PCMK__CRM_COMMON_RESOURCES__H #define PCMK__CRM_COMMON_RESOURCES__H #include // bool #include // time_t #include // xmlNode #include // gboolean, guint, GList, GHashTable #include // enum rsc_role_e #include // pcmk_resource_t, etc. #ifdef __cplusplus extern "C" { #endif /*! * \file * \brief Scheduler API for resources * \ingroup core */ //! Search options for resources (exact resource ID always matches) enum pe_find { //! Also match clone instance ID from resource history pcmk_rsc_match_history = (1 << 0), //! Also match anonymous clone instances by base name pcmk_rsc_match_anon_basename = (1 << 1), //! Match only clones and their instances, by either clone or instance ID pcmk_rsc_match_clone_only = (1 << 2), //! If matching by node, compare current node instead of assigned node pcmk_rsc_match_current_node = (1 << 3), - //! \deprecated Do not use - pe_find_inactive = (1 << 4), - //! Match clone instances (even unique) by base name as well as exact ID pcmk_rsc_match_basename = (1 << 5), - -#if !defined(PCMK_ALLOW_DEPRECATED) || (PCMK_ALLOW_DEPRECATED == 1) - //! \deprecated Use pcmk_rsc_match_history instead - pe_find_renamed = pcmk_rsc_match_history, - - //! \deprecated Use pcmk_rsc_match_anon_basename instead - pe_find_anon = pcmk_rsc_match_anon_basename, - - //! \deprecated Use pcmk_rsc_match_clone_only instead - pe_find_clone = pcmk_rsc_match_clone_only, - - //! \deprecated Use pcmk_rsc_match_current_node instead - pe_find_current = pcmk_rsc_match_current_node, - - //! \deprecated Use pcmk_rsc_match_basename instead - pe_find_any = pcmk_rsc_match_basename, -#endif }; //! \internal Do not use typedef struct pcmk__resource_private pcmk__resource_private_t; // Implementation of pcmk_resource_t // @COMPAT Make this internal when we can break API backward compatibility //!@{ //! \deprecated Do not use (public access will be removed in a future release) -struct pe_resource_s { +struct pcmk__resource { /* @COMPAT Once all members are moved to pcmk__resource_private_t, * We can make that the pcmk_resource_t implementation and drop this * struct altogether, leaving pcmk_resource_t as an opaque public type. */ pcmk__resource_private_t *private; // NOTE: sbd (as of at least 1.5.2) uses this //! \deprecated Call pcmk_resource_id() instead char *id; // Resource ID in configuration // NOTE: sbd (as of at least 1.5.2) uses this //! \deprecated Call pcmk_resource_is_managed() instead unsigned long long flags; // Group of enum pcmk__rsc_flags }; //!@} const char *pcmk_resource_id(const pcmk_resource_t *rsc); bool pcmk_resource_is_managed(const pcmk_resource_t *rsc); #ifdef __cplusplus } #endif #endif // PCMK__CRM_COMMON_RESOURCES__H diff --git a/include/crm/common/results.h b/include/crm/common/results.h index a2d35ddf6b..d64ea57c86 100644 --- a/include/crm/common/results.h +++ b/include/crm/common/results.h @@ -1,419 +1,399 @@ /* * Copyright 2012-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #ifndef PCMK__CRM_COMMON_RESULTS__H #define PCMK__CRM_COMMON_RESULTS__H #ifdef __cplusplus extern "C" { #endif /*! * \file * \brief Function and executable result codes * \ingroup core */ // Lifted from config.h /* The _Noreturn keyword of C11. */ #ifndef _Noreturn # if (defined __cplusplus \ && ((201103 <= __cplusplus && !(__GNUC__ == 4 && __GNUC_MINOR__ == 7)) \ || (defined _MSC_VER && 1900 <= _MSC_VER))) # define _Noreturn [[noreturn]] # elif ((!defined __cplusplus || defined __clang__) \ && (201112 <= (defined __STDC_VERSION__ ? __STDC_VERSION__ : 0) \ || 4 < __GNUC__ + (7 <= __GNUC_MINOR__))) /* _Noreturn works as-is. */ # elif 2 < __GNUC__ + (8 <= __GNUC_MINOR__) || 0x5110 <= __SUNPRO_C # define _Noreturn __attribute__ ((__noreturn__)) # elif 1200 <= (defined _MSC_VER ? _MSC_VER : 0) # define _Noreturn __declspec (noreturn) # else # define _Noreturn # endif #endif #define CRM_ASSERT(expr) do { \ if (!(expr)) { \ crm_abort(__FILE__, __func__, __LINE__, #expr, TRUE, FALSE); \ } \ } while(0) /* * Function return codes * * Most Pacemaker API functions return an integer return code. There are two * alternative interpretations. The legacy interpration is that the absolute * value of the return code is either a system error number or a custom * pcmk_err_* number. This is less than ideal because system error numbers are * constrained only to the positive int range, so there's the possibility that * system errors and custom errors could collide (which did in fact happen * already on one architecture). The new intepretation is that negative values * are from the pcmk_rc_e enum, and positive values are system error numbers. * Both use 0 for success. * * For system error codes, see: * - /usr/include/asm-generic/errno.h * - /usr/include/asm-generic/errno-base.h */ // Legacy custom return codes for Pacemaker API functions (deprecated) // NOTE: sbd (as of at least 1.5.2) uses this #define pcmk_ok 0 #define PCMK_ERROR_OFFSET 190 /* Replacements on non-linux systems, see include/portability.h */ #define PCMK_CUSTOM_OFFSET 200 /* Purely custom codes */ #define pcmk_err_generic 201 #define pcmk_err_no_quorum 202 #define pcmk_err_schema_validation 203 #define pcmk_err_transform_failed 204 #define pcmk_err_old_data 205 // NOTE: sbd (as of at least 1.5.2) uses this #define pcmk_err_diff_failed 206 // NOTE: sbd (as of at least 1.5.2) uses this #define pcmk_err_diff_resync 207 #define pcmk_err_cib_modified 208 #define pcmk_err_cib_backup 209 #define pcmk_err_cib_save 210 #define pcmk_err_schema_unchanged 211 #define pcmk_err_cib_corrupt 212 #define pcmk_err_multiple 213 #define pcmk_err_node_unknown 214 #define pcmk_err_already 215 /* On HPPA 215 is ENOSYM (Unknown error 215), which hopefully never happens. */ #ifdef __hppa__ #define pcmk_err_bad_nvpair 250 /* 216 is ENOTSOCK */ #define pcmk_err_unknown_format 252 /* 217 is EDESTADDRREQ */ #else #define pcmk_err_bad_nvpair 216 #define pcmk_err_unknown_format 217 #endif /*! * \enum pcmk_rc_e * \brief Return codes for Pacemaker API functions * * Any Pacemaker API function documented as returning a "standard Pacemaker * return code" will return pcmk_rc_ok (0) on success, and one of this * enumeration's other (negative) values or a (positive) system error number * otherwise. The custom codes are at -1001 and lower, so that the caller may * use -1 through -1000 for their own custom values if desired. While generally * referred to as "errors", nonzero values simply indicate a result, which might * or might not be an error depending on the calling context. */ enum pcmk_rc_e { /* When adding new values, use consecutively lower numbers, update the array * in lib/common/results.c, and test with crm_error. */ pcmk_rc_compression = -1039, pcmk_rc_ns_resolution = -1038, pcmk_rc_no_transaction = -1037, pcmk_rc_bad_xml_patch = -1036, pcmk_rc_bad_input = -1035, pcmk_rc_disabled = -1034, pcmk_rc_duplicate_id = -1033, pcmk_rc_unpack_error = -1032, pcmk_rc_invalid_transition = -1031, pcmk_rc_graph_error = -1030, pcmk_rc_dot_error = -1029, pcmk_rc_underflow = -1028, pcmk_rc_no_input = -1027, pcmk_rc_no_output = -1026, pcmk_rc_after_range = -1025, pcmk_rc_within_range = -1024, pcmk_rc_before_range = -1023, pcmk_rc_undetermined = -1022, pcmk_rc_op_unsatisfied = -1021, pcmk_rc_ipc_pid_only = -1020, pcmk_rc_ipc_unresponsive = -1019, pcmk_rc_ipc_unauthorized = -1018, pcmk_rc_no_quorum = -1017, pcmk_rc_schema_validation = -1016, pcmk_rc_schema_unchanged = -1015, pcmk_rc_transform_failed = -1014, pcmk_rc_old_data = -1013, pcmk_rc_diff_failed = -1012, pcmk_rc_diff_resync = -1011, pcmk_rc_cib_modified = -1010, pcmk_rc_cib_backup = -1009, pcmk_rc_cib_save = -1008, pcmk_rc_cib_corrupt = -1007, pcmk_rc_multiple = -1006, pcmk_rc_node_unknown = -1005, pcmk_rc_already = -1004, pcmk_rc_bad_nvpair = -1003, pcmk_rc_unknown_format = -1002, // Developers: Use a more specific code than pcmk_rc_error whenever possible pcmk_rc_error = -1001, // Values -1 through -1000 reserved for caller use // NOTE: sbd (as of at least 1.5.2) uses this pcmk_rc_ok = 0 // Positive values reserved for system error numbers }; /*! * \enum ocf_exitcode * \brief Exit status codes for resource agents * * The OCF Resource Agent API standard enumerates the possible exit status codes * that agents should return. Besides being used with OCF agents, these values * are also used by the executor as a universal status for all agent standards; * actual results are mapped to these before returning them to clients. */ enum ocf_exitcode { PCMK_OCF_OK = 0, //!< Success // NOTE: booth (as of at least 1.1) uses this value PCMK_OCF_UNKNOWN_ERROR = 1, //!< Unspecified error PCMK_OCF_INVALID_PARAM = 2, //!< Parameter invalid (in local context) PCMK_OCF_UNIMPLEMENT_FEATURE = 3, //!< Requested action not implemented PCMK_OCF_INSUFFICIENT_PRIV = 4, //!< Insufficient privileges PCMK_OCF_NOT_INSTALLED = 5, //!< Dependencies not available locally PCMK_OCF_NOT_CONFIGURED = 6, //!< Parameter invalid (inherently) // NOTE: booth (as of at least 1.1) uses this value PCMK_OCF_NOT_RUNNING = 7, //!< Service safely stopped PCMK_OCF_RUNNING_PROMOTED = 8, //!< Service active and promoted PCMK_OCF_FAILED_PROMOTED = 9, //!< Service failed and possibly in promoted role PCMK_OCF_DEGRADED = 190, //!< Service active but more likely to fail soon PCMK_OCF_DEGRADED_PROMOTED = 191, //!< Service promoted but more likely to fail soon /* These two are Pacemaker extensions, not in the OCF standard. The * controller records PCMK_OCF_UNKNOWN for pending actions. * PCMK_OCF_CONNECTION_DIED is used only with older DCs that don't support * PCMK_EXEC_NOT_CONNECTED. */ PCMK_OCF_CONNECTION_DIED = 189, //!< \deprecated See PCMK_EXEC_NOT_CONNECTED PCMK_OCF_UNKNOWN = 193, //!< Action is pending - -#if !defined(PCMK_ALLOW_DEPRECATED) || (PCMK_ALLOW_DEPRECATED == 1) - // Former Pacemaker extensions - PCMK_OCF_EXEC_ERROR = 192, //!< \deprecated (Unused) - PCMK_OCF_SIGNAL = 194, //!< \deprecated (Unused) - PCMK_OCF_NOT_SUPPORTED = 195, //!< \deprecated (Unused) - PCMK_OCF_PENDING = 196, //!< \deprecated (Unused) - PCMK_OCF_CANCELLED = 197, //!< \deprecated (Unused) - PCMK_OCF_TIMEOUT = 198, //!< \deprecated (Unused) - PCMK_OCF_OTHER_ERROR = 199, //!< \deprecated (Unused) - - //! \deprecated Use PCMK_OCF_RUNNING_PROMOTED instead - PCMK_OCF_RUNNING_MASTER = PCMK_OCF_RUNNING_PROMOTED, - - //! \deprecated Use PCMK_OCF_FAILED_PROMOTED instead - PCMK_OCF_FAILED_MASTER = PCMK_OCF_FAILED_PROMOTED, - - //! \deprecated Use PCMK_OCF_DEGRADED_PROMOTED instead - PCMK_OCF_DEGRADED_MASTER = PCMK_OCF_DEGRADED_PROMOTED, -#endif }; // NOTE: sbd (as of at least 1.5.2) uses this /*! * \enum crm_exit_e * \brief Exit status codes for tools and daemons * * We want well-specified (i.e. OS-invariant) exit status codes for our daemons * and applications so they can be relied on by callers. (Function return codes * and errno's do not make good exit statuses.) * * The only hard rule is that exit statuses must be between 0 and 255; all else * is convention. Universally, 0 is success, and 1 is generic error (excluding * OSes we don't support -- for example, OpenVMS considers 1 success!). * * For init scripts, the LSB gives meaning to 0-7, and sets aside 150-199 for * application use. OCF adds 8-9 and 190-191. * * sysexits.h was an attempt to give additional meanings, but never really * caught on. It uses 0 and 64-78. * * Bash reserves 2 ("incorrect builtin usage") and 126-255 (126 is "command * found but not executable", 127 is "command not found", 128 + n is * "interrupted by signal n"). * * tldp.org recommends 64-113 for application use. * * We try to overlap with the above conventions when practical. */ typedef enum crm_exit_e { // Common convention CRM_EX_OK = 0, //!< Success CRM_EX_ERROR = 1, //!< Unspecified error // LSB + OCF CRM_EX_INVALID_PARAM = 2, //!< Parameter invalid (in local context) CRM_EX_UNIMPLEMENT_FEATURE = 3, //!< Requested action not implemented CRM_EX_INSUFFICIENT_PRIV = 4, //!< Insufficient privileges CRM_EX_NOT_INSTALLED = 5, //!< Dependencies not available locally CRM_EX_NOT_CONFIGURED = 6, //!< Parameter invalid (inherently) CRM_EX_NOT_RUNNING = 7, //!< Service safely stopped CRM_EX_PROMOTED = 8, //!< Service active and promoted CRM_EX_FAILED_PROMOTED = 9, //!< Service failed and possibly promoted // sysexits.h CRM_EX_USAGE = 64, //!< Command line usage error CRM_EX_DATAERR = 65, //!< User-supplied data incorrect CRM_EX_NOINPUT = 66, //!< Input file not available CRM_EX_NOUSER = 67, //!< User does not exist CRM_EX_NOHOST = 68, //!< Host unknown CRM_EX_UNAVAILABLE = 69, //!< Needed service unavailable CRM_EX_SOFTWARE = 70, //!< Internal software bug CRM_EX_OSERR = 71, //!< External (OS/environmental) problem CRM_EX_OSFILE = 72, //!< System file not usable CRM_EX_CANTCREAT = 73, //!< File couldn't be created CRM_EX_IOERR = 74, //!< File I/O error CRM_EX_TEMPFAIL = 75, //!< Try again CRM_EX_PROTOCOL = 76, //!< Protocol violated CRM_EX_NOPERM = 77, //!< Non-file permission issue CRM_EX_CONFIG = 78, //!< Misconfiguration // Custom CRM_EX_FATAL = 100, //!< Do not respawn CRM_EX_PANIC = 101, //!< Panic the local host CRM_EX_DISCONNECT = 102, //!< Lost connection to something CRM_EX_OLD = 103, //!< Update older than existing config CRM_EX_DIGEST = 104, //!< Digest comparison failed CRM_EX_NOSUCH = 105, //!< Requested item does not exist CRM_EX_QUORUM = 106, //!< Local partition does not have quorum CRM_EX_UNSAFE = 107, //!< Requires --force or new conditions CRM_EX_EXISTS = 108, //!< Requested item already exists CRM_EX_MULTIPLE = 109, //!< Requested item has multiple matches CRM_EX_EXPIRED = 110, //!< Requested item has expired CRM_EX_NOT_YET_IN_EFFECT = 111, //!< Requested item is not in effect CRM_EX_INDETERMINATE = 112, //!< Could not determine status CRM_EX_UNSATISFIED = 113, //!< Requested item does not satisfy constraints // Other CRM_EX_TIMEOUT = 124, //!< Convention from timeout(1) /* Anything above 128 overlaps with some shells' use of these values for * "interrupted by signal N", and so may be unreliable when detected by * shell scripts. */ // OCF Resource Agent API 1.1 CRM_EX_DEGRADED = 190, //!< Service active but more likely to fail soon CRM_EX_DEGRADED_PROMOTED = 191, //!< Service promoted but more likely to fail soon /* Custom * * This can be used to initialize exit status variables or to indicate that * a command is pending (which is what the controller uses it for). */ CRM_EX_NONE = 193, //!< No exit status available CRM_EX_MAX = 255, //!< Ensure crm_exit_t can hold this } crm_exit_t; /*! * \enum pcmk_exec_status * \brief Execution status * * These codes are used to specify the result of the attempt to execute an * agent, rather than the agent's result itself. */ enum pcmk_exec_status { PCMK_EXEC_UNKNOWN = -2, //!< Used only to initialize variables PCMK_EXEC_PENDING = -1, //!< Action is in progress PCMK_EXEC_DONE, //!< Action completed, result is known PCMK_EXEC_CANCELLED, //!< Action was cancelled PCMK_EXEC_TIMEOUT, //!< Action did not complete in time PCMK_EXEC_NOT_SUPPORTED, //!< Agent does not implement requested action PCMK_EXEC_ERROR, //!< Execution failed, may be retried PCMK_EXEC_ERROR_HARD, //!< Execution failed, do not retry on node PCMK_EXEC_ERROR_FATAL, //!< Execution failed, do not retry anywhere PCMK_EXEC_NOT_INSTALLED, //!< Agent or dependency not available locally PCMK_EXEC_NOT_CONNECTED, //!< No connection to executor PCMK_EXEC_INVALID, //!< Action cannot be attempted (e.g. shutdown) PCMK_EXEC_NO_FENCE_DEVICE, //!< No fence device is configured for target PCMK_EXEC_NO_SECRETS, //!< Necessary CIB secrets are unavailable // Add new values above here then update this one below PCMK_EXEC_MAX = PCMK_EXEC_NO_SECRETS, //!< Maximum value for this enum }; /*! * \enum pcmk_result_type * \brief Types of Pacemaker result codes * * A particular integer can have different meanings within different Pacemaker * result code families. It may be interpretable within zero, one, or multiple * families. * * These values are useful for specifying how an integer result code should be * interpreted in situations involving a generic integer value. For example, a * function that can process multiple types of result codes might accept an * arbitrary integer argument along with a \p pcmk_result_type argument that * specifies how to interpret the integer. */ enum pcmk_result_type { pcmk_result_legacy = 0, //!< Legacy API function return code pcmk_result_rc = 1, //!< Standard Pacemaker return code pcmk_result_exitcode = 2, //!< Exit status code pcmk_result_exec_status = 3, //!< Execution status }; int pcmk_result_get_strings(int code, enum pcmk_result_type type, const char **name, const char **desc); const char *pcmk_rc_name(int rc); // NOTE: sbd (as of at least 1.5.2) uses this const char *pcmk_rc_str(int rc); crm_exit_t pcmk_rc2exitc(int rc); enum ocf_exitcode pcmk_rc2ocf(int rc); int pcmk_rc2legacy(int rc); int pcmk_legacy2rc(int legacy_rc); // NOTE: sbd (as of at least 1.5.2) uses this const char *pcmk_strerror(int rc); const char *pcmk_errorname(int rc); const char *crm_exit_name(crm_exit_t exit_code); // NOTE: sbd (as of at least 1.5.2) uses this const char *crm_exit_str(crm_exit_t exit_code); _Noreturn crm_exit_t crm_exit(crm_exit_t rc); static inline const char * pcmk_exec_status_str(enum pcmk_exec_status status) { switch (status) { case PCMK_EXEC_PENDING: return "pending"; case PCMK_EXEC_DONE: return "complete"; case PCMK_EXEC_CANCELLED: return "Cancelled"; case PCMK_EXEC_TIMEOUT: return "Timed Out"; case PCMK_EXEC_NOT_SUPPORTED: return "NOT SUPPORTED"; case PCMK_EXEC_ERROR: return "Error"; case PCMK_EXEC_ERROR_HARD: return "Hard error"; case PCMK_EXEC_ERROR_FATAL: return "Fatal error"; case PCMK_EXEC_NOT_INSTALLED: return "Not installed"; case PCMK_EXEC_NOT_CONNECTED: return "Internal communication failure"; case PCMK_EXEC_INVALID: return "Cannot execute now"; case PCMK_EXEC_NO_FENCE_DEVICE: return "No fence device"; case PCMK_EXEC_NO_SECRETS: return "CIB secrets unavailable"; default: return "UNKNOWN!"; } } #ifdef __cplusplus } #endif #endif diff --git a/include/crm/common/roles.h b/include/crm/common/roles.h index e315d6b6b0..988e0f503d 100644 --- a/include/crm/common/roles.h +++ b/include/crm/common/roles.h @@ -1,72 +1,49 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #ifndef PCMK__CRM_COMMON_ROLES__H #define PCMK__CRM_COMMON_ROLES__H #ifdef __cplusplus extern "C" { #endif /*! * \file * \brief Scheduler API for resource roles * \ingroup core */ // String equivalents of enum rsc_role_e #define PCMK_ROLE_STOPPED "Stopped" #define PCMK_ROLE_STARTED "Started" #define PCMK_ROLE_UNPROMOTED "Unpromoted" #define PCMK_ROLE_PROMOTED "Promoted" /*! * Possible roles that a resource can be in * (order matters; values can be compared with less than and greater than) */ enum rsc_role_e { pcmk_role_unknown = 0, //!< Resource role is unknown pcmk_role_stopped = 1, //!< Stopped pcmk_role_started = 2, //!< Started pcmk_role_unpromoted = 3, //!< Unpromoted pcmk_role_promoted = 4, //!< Promoted - -#if !defined(PCMK_ALLOW_DEPRECATED) || (PCMK_ALLOW_DEPRECATED == 1) - //! \deprecated Use pcmk_role_unknown instead - RSC_ROLE_UNKNOWN = pcmk_role_unknown, - - //! \deprecated Use pcmk_role_stopped instead - RSC_ROLE_STOPPED = pcmk_role_stopped, - - //! \deprecated Use pcmk_role_started instead - RSC_ROLE_STARTED = pcmk_role_started, - - //! \deprecated Use pcmk_role_unpromoted instead - RSC_ROLE_UNPROMOTED = pcmk_role_unpromoted, - - //! \deprecated Use pcmk_role_unpromoted instead - RSC_ROLE_SLAVE = pcmk_role_unpromoted, - - //! \deprecated Use pcmk_role_promoted instead - RSC_ROLE_PROMOTED = pcmk_role_promoted, - - //! \deprecated Use pcmk_role_promoted instead - RSC_ROLE_MASTER = pcmk_role_promoted, -#endif }; const char *pcmk_role_text(enum rsc_role_e role); enum rsc_role_e pcmk_parse_role(const char *role); #ifdef __cplusplus } #endif #endif // PCMK__CRM_COMMON_ROLES__H diff --git a/include/crm/common/scheduler_types.h b/include/crm/common/scheduler_types.h index 1620bd05cc..1f972e2c33 100644 --- a/include/crm/common/scheduler_types.h +++ b/include/crm/common/scheduler_types.h @@ -1,39 +1,39 @@ /* * Copyright 2023-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #ifndef PCMK__CRM_COMMON_SCHEDULER_TYPES__H #define PCMK__CRM_COMMON_SCHEDULER_TYPES__H #ifdef __cplusplus extern "C" { #endif /*! * \file * \brief Type aliases needed to define scheduler objects * \ingroup core */ //! Node object (including information that may vary depending on resource) typedef struct pcmk__scored_node pcmk_node_t; //! Resource object -typedef struct pe_resource_s pcmk_resource_t; +typedef struct pcmk__resource pcmk_resource_t; //! Action object typedef struct pcmk__action pcmk_action_t; //! Scheduler object typedef struct pe_working_set_s pcmk_scheduler_t; #ifdef __cplusplus } #endif #endif // PCMK__CRM_COMMON_SCHEDULER_TYPES__H diff --git a/include/crm/pengine/pe_types_compat.h b/include/crm/pengine/pe_types_compat.h index 8218e2910f..3e39dbc3a4 100644 --- a/include/crm/pengine/pe_types_compat.h +++ b/include/crm/pengine/pe_types_compat.h @@ -1,56 +1,56 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #ifndef PCMK__CRM_PENGINE_PE_TYPES_COMPAT__H # define PCMK__CRM_PENGINE_PE_TYPES_COMPAT__H #include #ifdef __cplusplus extern "C" { #endif /** * \file * \brief Deprecated Pacemaker scheduler API * \ingroup pengine * \deprecated Do not include this header directly. The scheduler APIs in this * header, and the header itself, will be removed in a future * release. */ // NOTE: sbd (as of at least 1.5.2) uses this //! \deprecated Do not use #define pe_rsc_managed (1ULL << 1) // NOTE: sbd (as of at least 1.5.2) uses this //! \deprecated Do not use #define pe_flag_have_quorum pcmk_sched_quorate // NOTE: sbd (as of at least 1.5.2) uses this //! \deprecated Do not use #define pe_flag_have_stonith_resource pcmk_sched_have_fencing // NOTE: sbd (as of at least 1.5.2) uses this //! \deprecated Use pcmk_node_t instead typedef struct pcmk__scored_node node_t; // NOTE: sbd (as of at least 1.5.2) uses this //! \deprecated use pcmk_resource_t instead -typedef struct pe_resource_s resource_t; +typedef struct pcmk__resource resource_t; // NOTE: sbd (as of at least 1.5.2) uses this //! \deprecated Use pcmk_scheduler_t instead typedef struct pe_working_set_s pe_working_set_t; #ifdef __cplusplus } #endif #endif // PCMK__CRM_PENGINE_PE_TYPES_COMPAT__H diff --git a/include/crm/services.h b/include/crm/services.h index c14a0c176c..84bd759abc 100644 --- a/include/crm/services.h +++ b/include/crm/services.h @@ -1,433 +1,418 @@ /* * Copyright 2010-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #ifndef PCMK__CRM_SERVICES__H # define PCMK__CRM_SERVICES__H # include # include # include # include # include # include # include // OCF_ROOT_DIR # include # include #ifdef __cplusplus extern "C" { #endif // NOTE: booth (as of at least 1.1) checks for the existence of this header /*! * \file * \brief Services API * \ingroup core */ /* TODO: Autodetect these two ?*/ # ifndef SYSTEMCTL # define SYSTEMCTL "/bin/systemctl" # endif /* This is the string passed in the OCF_EXIT_REASON_PREFIX environment variable. * The stderr output that occurs after this prefix is encountered is considered * the exit reason for a completed operation. */ #define PCMK_OCF_REASON_PREFIX "ocf-exit-reason:" // Agent version to use if agent doesn't specify one #define PCMK_DEFAULT_AGENT_VERSION "0.1" enum lsb_exitcode { PCMK_LSB_OK = 0, // NOTE: booth (as of at least 1.1) uses this value PCMK_LSB_UNKNOWN_ERROR = 1, PCMK_LSB_INVALID_PARAM = 2, PCMK_LSB_UNIMPLEMENT_FEATURE = 3, PCMK_LSB_INSUFFICIENT_PRIV = 4, PCMK_LSB_NOT_INSTALLED = 5, PCMK_LSB_NOT_CONFIGURED = 6, PCMK_LSB_NOT_RUNNING = 7, }; // LSB uses different return codes for status actions enum lsb_status_exitcode { PCMK_LSB_STATUS_OK = 0, PCMK_LSB_STATUS_VAR_PID = 1, PCMK_LSB_STATUS_VAR_LOCK = 2, PCMK_LSB_STATUS_NOT_RUNNING = 3, PCMK_LSB_STATUS_UNKNOWN = 4, /* custom codes should be in the 150-199 range reserved for application use */ PCMK_LSB_STATUS_NOT_INSTALLED = 150, PCMK_LSB_STATUS_INSUFFICIENT_PRIV = 151, }; //!@{ //! \deprecated Do not use enum nagios_exitcode { NAGIOS_STATE_OK = 0, NAGIOS_STATE_WARNING = 1, NAGIOS_STATE_CRITICAL = 2, NAGIOS_STATE_UNKNOWN = 3, /* This is a custom Pacemaker value (not a nagios convention), used as an * intermediate value between the services library and the executor, so the * executor can map it to the corresponding OCF code. */ NAGIOS_INSUFFICIENT_PRIV = 100, #if !defined(PCMK_ALLOW_DEPRECATED) || (PCMK_ALLOW_DEPRECATED == 1) NAGIOS_STATE_DEPENDENT = 4, NAGIOS_NOT_INSTALLED = 101, #endif }; //!@} enum svc_action_flags { /* On timeout, only kill pid, do not kill entire pid group */ SVC_ACTION_LEAVE_GROUP = 0x01, SVC_ACTION_NON_BLOCKED = 0x02, }; typedef struct svc_action_private_s svc_action_private_t; /*! * \brief Object for executing external actions * * \note This object should never be instantiated directly, but instead created * using one of the constructor functions (resources_action_create() for * resource agents, services_alert_create() for alert agents, or * services_action_create_generic() for generic executables). Similarly, * do not use sizeof() on this struct. */ /* * NOTE: Internally, services__create_resource_action() is preferable to * resources_action_create(). */ typedef struct svc_action_s { /*! Operation key (__) for resource actions, * XML ID for alert actions, or NULL for generic actions */ char *id; //! XML ID of resource being executed for resource actions, otherwise NULL char *rsc; //! Name of action being executed for resource actions, otherwise NULL char *action; //! Action interval for recurring resource actions, otherwise 0 guint interval_ms; //! Resource standard for resource actions, otherwise NULL char *standard; //! Resource provider for resource actions that require it, otherwise NULL char *provider; //! Resource agent name for resource actions, otherwise NULL char *agent; int timeout; //!< Action timeout (in milliseconds) /*! A hash table of name/value pairs to use as parameters for resource and * alert actions, otherwise NULL. These will be used to set environment * variables for non-fencing resource agents and alert agents, and to send * stdin to fence agents. */ GHashTable *params; int rc; //!< Exit status of action (set by library upon completion) //!@{ //! This field should be treated as internal to Pacemaker int pid; // Process ID of child int cancel; // Whether this is a cancellation of a recurring action //!@} int status; //!< Execution status (enum pcmk_exec_status set by library) /*! Action counter (set by library for resource actions, or by caller * otherwise) */ int sequence; //!@{ //! This field should be treated as internal to Pacemaker int expected_rc; // Unused int synchronous; // Whether execution should be synchronous (blocking) //!@} enum svc_action_flags flags; //!< Flag group of enum svc_action_flags char *stderr_data; //!< Action stderr (set by library) char *stdout_data; //!< Action stdout (set by library) void *cb_data; //!< For caller's use (not used by library) //! This field should be treated as internal to Pacemaker svc_action_private_t *opaque; } svc_action_t; /*! * \brief Get a list of files or directories in a given path * * \param[in] root Full path to a directory to read * \param[in] files Return list of files if TRUE or directories if FALSE * \param[in] executable If TRUE and files is TRUE, only return executable files * * \return List of what was found as char * items. * \note The caller is responsibile for freeing the result with * g_list_free_full(list, free). */ GList *get_directory_list(const char *root, gboolean files, gboolean executable); /*! * \brief Get a list of providers * * \param[in] standard List providers of this resource agent standard * * \return List of providers as char * list items (or NULL if standard does not * support providers) * \note The caller is responsible for freeing the result using * g_list_free_full(list, free). */ GList *resources_list_providers(const char *standard); /*! * \brief Get a list of resource agents * * \param[in] standard List agents of this standard (or NULL for all) * \param[in] provider List agents of this provider (or NULL for all) * * \return List of resource agents as char * items. * \note The caller is responsible for freeing the result using * g_list_free_full(list, free). */ GList *resources_list_agents(const char *standard, const char *provider); /*! * Get list of available standards * * \return List of resource standards as char * items. * \note The caller is responsible for freeing the result using * g_list_free_full(list, free). */ GList *resources_list_standards(void); /*! * \brief Check whether a resource agent exists on the local host * * \param[in] standard Resource agent standard of agent to check * \param[in] provider Provider of agent to check (or NULL) * \param[in] agent Name of agent to check * * \return TRUE if agent exists locally, otherwise FALSE */ gboolean resources_agent_exists(const char *standard, const char *provider, const char *agent); /*! * \brief Create a new resource action * * \param[in] name Name of resource that action is for * \param[in] standard Resource agent standard * \param[in] provider Resource agent provider * \param[in] agent Resource agent name * \param[in] action Name of action to create * \param[in] interval_ms How often to repeat action (if 0, execute once) * \param[in] timeout Error if not complete within this time (ms) * \param[in,out] params Action parameters * \param[in] flags Group of enum svc_action_flags * * \return Newly allocated action * \note This function assumes ownership of (and may free) \p params. * \note The caller is responsible for freeing the return value using * services_action_free(). */ svc_action_t *resources_action_create(const char *name, const char *standard, const char *provider, const char *agent, const char *action, guint interval_ms, int timeout, GHashTable *params, enum svc_action_flags flags); /*! * \brief Reschedule a recurring action for immediate execution * * \param[in] name Name of resource that action is for * \param[in] action Action's name * \param[in] interval_ms Action's interval (in milliseconds) * * \return TRUE on success, otherwise FALSE */ gboolean services_action_kick(const char *name, const char *action, guint interval_ms); const char *resources_find_service_class(const char *agent); /*! * \brief Request execution of an arbitrary command * * This API has useful infrastructure in place to be able to run a command * in the background and get notified via a callback when the command finishes. * * \param[in] exec Full path to command executable * \param[in] args NULL-terminated list of arguments to pass to command * * \return Newly allocated action object */ svc_action_t *services_action_create_generic(const char *exec, const char *args[]); void services_action_cleanup(svc_action_t *op); void services_action_free(svc_action_t *op); int services_action_user(svc_action_t *op, const char *user); gboolean services_action_sync(svc_action_t *op); /*! * \brief Run an action asynchronously, with callback after process is forked * * \param[in,out] op Action to run * \param[in] action_callback Function to call when action completes * (if NULL, any previously set callback will * continue to be used) * \param[in] action_fork_callback Function to call after child process is * forked for action (if NULL, any * previously set callback will continue to * be used) * * \retval TRUE if the caller should not free or otherwise use \p op again, * because one of these conditions is true: * * * \p op is NULL. * * The action was successfully initiated, in which case * \p action_fork_callback has been called, but \p action_callback has * not (it will be called when the action completes). * * The action's ID matched an existing recurring action. The existing * action has taken over the callback and callback data from \p op * and has been re-initiated asynchronously, and \p op has been freed. * * Another action for the same resource is in flight, and \p op will * be blocked until it completes. * * The action could not be initiated, and is either non-recurring or * being cancelled. \p action_fork_callback has not been called, but * \p action_callback has, and \p op has been freed. * * \retval FALSE if \op is still valid, because the action cannot be initiated, * and is a recurring action that is not being cancelled. * \p action_fork_callback has not been called, but \p action_callback * has, and a timer has been set for the next invocation of \p op. */ gboolean services_action_async_fork_notify(svc_action_t *op, void (*action_callback) (svc_action_t *), void (*action_fork_callback) (svc_action_t *)); /*! * \brief Request asynchronous execution of an action * * \param[in,out] op Action to execute * \param[in] action_callback Function to call when the action completes * (if NULL, any previously set callback will * continue to be used) * * \retval TRUE if the caller should not free or otherwise use \p op again, * because one of these conditions is true: * * * \p op is NULL. * * The action was successfully initiated, in which case * \p action_callback has not been called (it will be called when the * action completes). * * The action's ID matched an existing recurring action. The existing * action has taken over the callback and callback data from \p op * and has been re-initiated asynchronously, and \p op has been freed. * * Another action for the same resource is in flight, and \p op will * be blocked until it completes. * * The action could not be initiated, and is either non-recurring or * being cancelled. \p action_callback has been called, and \p op has * been freed. * * \retval FALSE if \op is still valid, because the action cannot be initiated, * and is a recurring action that is not being cancelled. * \p action_callback has been called, and a timer has been set for the * next invocation of \p op. */ gboolean services_action_async(svc_action_t *op, void (*action_callback) (svc_action_t *)); gboolean services_action_cancel(const char *name, const char *action, guint interval_ms); /* functions for alert agents */ svc_action_t *services_alert_create(const char *id, const char *exec, int timeout, GHashTable *params, int sequence, void *cb_data); gboolean services_alert_async(svc_action_t *action, void (*cb)(svc_action_t *op)); enum ocf_exitcode services_result2ocf(const char *standard, const char *action, int exit_status); static inline const char *services_ocf_exitcode_str(enum ocf_exitcode code) { switch (code) { case PCMK_OCF_OK: return "ok"; case PCMK_OCF_UNKNOWN_ERROR: return "error"; case PCMK_OCF_INVALID_PARAM: return "invalid parameter"; case PCMK_OCF_UNIMPLEMENT_FEATURE: return "unimplemented feature"; case PCMK_OCF_INSUFFICIENT_PRIV: return "insufficient privileges"; case PCMK_OCF_NOT_INSTALLED: return "not installed"; case PCMK_OCF_NOT_CONFIGURED: return "not configured"; case PCMK_OCF_NOT_RUNNING: return "not running"; case PCMK_OCF_RUNNING_PROMOTED: return "promoted"; case PCMK_OCF_FAILED_PROMOTED: return "promoted (failed)"; case PCMK_OCF_DEGRADED: return "OCF_DEGRADED"; case PCMK_OCF_DEGRADED_PROMOTED: return "promoted (degraded)"; - -#if !defined(PCMK_ALLOW_DEPRECATED) || (PCMK_ALLOW_DEPRECATED == 1) - case PCMK_OCF_NOT_SUPPORTED: - return "not supported (DEPRECATED STATUS)"; - case PCMK_OCF_CANCELLED: - return "cancelled (DEPRECATED STATUS)"; - case PCMK_OCF_OTHER_ERROR: - return "other error (DEPRECATED STATUS)"; - case PCMK_OCF_SIGNAL: - return "interrupted by signal (DEPRECATED STATUS)"; - case PCMK_OCF_PENDING: - return "pending (DEPRECATED STATUS)"; - case PCMK_OCF_TIMEOUT: - return "timeout (DEPRECATED STATUS)"; -#endif default: return "unknown"; } } # ifdef __cplusplus } # endif #endif /* __PCMK_SERVICES__ */ diff --git a/include/crm/stonith-ng.h b/include/crm/stonith-ng.h index 4774d9a0f5..347bec01d5 100644 --- a/include/crm/stonith-ng.h +++ b/include/crm/stonith-ng.h @@ -1,716 +1,693 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #ifndef PCMK__CRM_STONITH_NG__H # define PCMK__CRM_STONITH_NG__H #ifdef __cplusplus extern "C" { #endif /** * \file * \brief Fencing aka. STONITH * \ingroup fencing */ /* IMPORTANT: DLM source code includes this file directly, without having access * to other Pacemaker headers on its include path, so this file should *not* * include any other Pacemaker headers. (DLM might be updated to avoid the * issue, but we should still follow this guideline for a long time after.) */ # include # include # include // bool # include // uint32_t # include // time_t /* *INDENT-OFF* */ enum stonith_state { stonith_connected_command, stonith_connected_query, stonith_disconnected, }; enum stonith_call_options { st_opt_none = 0x00000000, st_opt_verbose = 0x00000001, st_opt_allow_suicide = 0x00000002, st_opt_manual_ack = 0x00000008, st_opt_discard_reply = 0x00000010, /* st_opt_all_replies = 0x00000020, */ st_opt_topology = 0x00000040, st_opt_scope_local = 0x00000100, st_opt_cs_nodeid = 0x00000200, st_opt_sync_call = 0x00001000, /*! Allow the timeout period for a callback to be adjusted * based on the time the server reports the operation will take. */ st_opt_timeout_updates = 0x00002000, /*! Only report back if operation is a success in callback */ st_opt_report_only_success = 0x00004000, /* used where ever apropriate - e.g. cleanup of history */ st_opt_cleanup = 0x000080000, /* used where ever apropriate - e.g. send out a history query to all nodes */ st_opt_broadcast = 0x000100000, }; /*! Order matters here, do not change values */ enum op_state { st_query, st_exec, st_done, st_duplicate, st_failed, }; // Supported fence agent interface standards enum stonith_namespace { st_namespace_invalid, st_namespace_any, st_namespace_internal, // Implemented internally by Pacemaker /* Neither of these projects are active any longer, but the fence agent * interfaces they created are still in use and supported by Pacemaker. */ st_namespace_rhcs, // Red Hat Cluster Suite compatible st_namespace_lha, // Linux-HA compatible }; enum stonith_namespace stonith_text2namespace(const char *namespace_s); const char *stonith_namespace2text(enum stonith_namespace st_namespace); enum stonith_namespace stonith_get_namespace(const char *agent, const char *namespace_s); typedef struct stonith_key_value_s { char *key; char *value; struct stonith_key_value_s *next; } stonith_key_value_t; typedef struct stonith_history_s { char *target; char *action; char *origin; char *delegate; char *client; int state; time_t completed; struct stonith_history_s *next; long completed_nsec; char *exit_reason; } stonith_history_t; typedef struct stonith_s stonith_t; typedef struct stonith_event_s { char *id; char *type; //!< \deprecated Will be removed in future release char *message; //!< \deprecated Will be removed in future release char *operation; int result; char *origin; char *target; char *action; char *executioner; char *device; /*! The name of the client that initiated the action. */ char *client_origin; //! \internal This field should be treated as internal to Pacemaker void *opaque; } stonith_event_t; typedef struct stonith_callback_data_s { int rc; int call_id; void *userdata; //! \internal This field should be treated as internal to Pacemaker void *opaque; } stonith_callback_data_t; typedef struct stonith_api_operations_s { /*! * \brief Destroy a fencer connection * * \param[in,out] st Fencer connection to destroy */ int (*free) (stonith_t *st); /*! * \brief Connect to the local fencer * * \param[in,out] st Fencer connection to connect * \param[in] name Client name to use * \param[out] stonith_fd If NULL, use a main loop, otherwise * store IPC file descriptor here * * \return Legacy Pacemaker return code */ int (*connect) (stonith_t *st, const char *name, int *stonith_fd); /*! * \brief Disconnect from the local stonith daemon. * * \param[in,out] st Fencer connection to disconnect * * \return Legacy Pacemaker return code */ int (*disconnect)(stonith_t *st); /*! * \brief Unregister a fence device with the local fencer * * \param[in,out] st Fencer connection to disconnect * \param[in] options Group of enum stonith_call_options * \param[in] name ID of fence device to unregister * * \return pcmk_ok (if synchronous) or positive call ID (if asynchronous) * on success, otherwise a negative legacy Pacemaker return code */ int (*remove_device)(stonith_t *st, int options, const char *name); /*! * \brief Register a fence device with the local fencer * * \param[in,out] st Fencer connection to use * \param[in] options Group of enum stonith_call_options * \param[in] id ID of fence device to register * \param[in] namespace_s Type of fence agent to search for ("redhat" * or "stonith-ng" for RHCS-style, "internal" * for Pacemaker-internal devices, "heartbeat" * for LHA-style, or "any" or NULL for any) * \param[in] agent Name of fence agent for device * \param[in] params Fence agent parameters for device * * \return pcmk_ok (if synchronous) or positive call ID (if asynchronous) * on success, otherwise a negative legacy Pacemaker return code */ int (*register_device)(stonith_t *st, int options, const char *id, const char *namespace_s, const char *agent, const stonith_key_value_t *params); /*! * \brief Unregister a fencing level for specified node with local fencer * * \param[in,out] st Fencer connection to use * \param[in] options Group of enum stonith_call_options * \param[in] node Target node to unregister level for * \param[in] level Topology level number to unregister * * \return pcmk_ok (if synchronous) or positive call ID (if asynchronous) * on success, otherwise a negative legacy Pacemaker return code */ int (*remove_level)(stonith_t *st, int options, const char *node, int level); /*! * \brief Register a fencing level for specified node with local fencer * * \param[in,out] st Fencer connection to use * \param[in] options Group of enum stonith_call_options * \param[in] node Target node to register level for * \param[in] level Topology level number to register * \param[in] device_list Devices to register in level * * \return pcmk_ok (if synchronous) or positive call ID (if asynchronous) * on success, otherwise a negative legacy Pacemaker return code */ int (*register_level)(stonith_t *st, int options, const char *node, int level, const stonith_key_value_t *device_list); /*! * \brief Retrieve a fence agent's metadata * * \param[in,out] stonith Fencer connection * \param[in] call_options Group of enum stonith_call_options * (currently ignored) * \param[in] agent Fence agent to query * \param[in] namespace_s Type of fence agent to search for ("redhat" * or "stonith-ng" for RHCS-style, "internal" * for Pacemaker-internal devices, "heartbeat" * for LHA-style, or "any" or NULL for any) * \param[out] output Where to store metadata * \param[in] timeout_sec Error if not complete within this time * * \return Legacy Pacemaker return code * \note The caller is responsible for freeing *output using free(). */ int (*metadata)(stonith_t *stonith, int call_options, const char *agent, const char *namespace_s, char **output, int timeout_sec); /*! * \brief Retrieve a list of installed fence agents * * \param[in,out] stonith Fencer connection to use * \param[in] call_options Group of enum stonith_call_options * (currently ignored) * \param[in] namespace_s Type of fence agents to list ("redhat" * or "stonith-ng" for RHCS-style, "internal" for * Pacemaker-internal devices, "heartbeat" for * LHA-style, or "any" or NULL for all) * \param[out] devices Where to store agent list * \param[in] timeout Error if unable to complete within this * (currently ignored) * * \return Number of items in list on success, or negative errno otherwise * \note The caller is responsible for freeing the returned list with * stonith_key_value_freeall(). */ int (*list_agents)(stonith_t *stonith, int call_options, const char *namespace_s, stonith_key_value_t **devices, int timeout); /*! * \brief Get the output of a fence device's list action * * \param[in,out] stonith Fencer connection to use * \param[in] call_options Group of enum stonith_call_options * \param[in] id Fence device ID to run list for * \param[out] list_info Where to store list output * \param[in] timeout Error if unable to complete within this * * \return pcmk_ok (if synchronous) or positive call ID (if asynchronous) * on success, otherwise a negative legacy Pacemaker return code */ int (*list)(stonith_t *stonith, int call_options, const char *id, char **list_info, int timeout); /*! * \brief Check whether a fence device is reachable by monitor action * * \param[in,out] stonith Fencer connection to use * \param[in] call_options Group of enum stonith_call_options * \param[in] id Fence device ID to run monitor for * \param[in] timeout Error if unable to complete within this * * \return pcmk_ok (if synchronous) or positive call ID (if asynchronous) * on success, otherwise a negative legacy Pacemaker return code */ int (*monitor)(stonith_t *stonith, int call_options, const char *id, int timeout); /*! * \brief Check whether a fence device target is reachable by status action * * \param[in,out] stonith Fencer connection to use * \param[in] call_options Group of enum stonith_call_options * \param[in] id Fence device ID to run status for * \param[in] port Fence target to run status for * \param[in] timeout Error if unable to complete within this * * \return pcmk_ok (if synchronous) or positive call ID (if asynchronous) * on success, otherwise a negative legacy Pacemaker return code */ int (*status)(stonith_t *stonith, int call_options, const char *id, const char *port, int timeout); /*! * \brief List registered fence devices * * \param[in,out] stonith Fencer connection to use * \param[in] call_options Group of enum stonith_call_options * \param[in] target Fence target to run status for * \param[out] devices Where to store list of fence devices * \param[in] timeout Error if unable to complete within this * * \note If node is provided, only devices that can fence the node id * will be returned. * * \return Number of items in list on success, or negative errno otherwise */ int (*query)(stonith_t *stonith, int call_options, const char *target, stonith_key_value_t **devices, int timeout); /*! * \brief Request that a target get fenced * * \param[in,out] stonith Fencer connection to use * \param[in] call_options Group of enum stonith_call_options * \param[in] node Fence target * \param[in] action "on", "off", or "reboot" * \param[in] timeout Default per-device timeout to use with * each executed device * \param[in] tolerance Accept result of identical fence action * completed within this time * * \return pcmk_ok (if synchronous) or positive call ID (if asynchronous) * on success, otherwise a negative legacy Pacemaker return code */ int (*fence)(stonith_t *stonith, int call_options, const char *node, const char *action, int timeout, int tolerance); /*! * \brief Manually confirm that a node has been fenced * * \param[in,out] stonith Fencer connection to use * \param[in] call_options Group of enum stonith_call_options * \param[in] target Fence target * * \return pcmk_ok (if synchronous) or positive call ID (if asynchronous) * on success, otherwise a negative legacy Pacemaker return code */ int (*confirm)(stonith_t *stonith, int call_options, const char *target); /*! * \brief List fencing actions that have occurred for a target * * \param[in,out] stonith Fencer connection to use * \param[in] call_options Group of enum stonith_call_options * \param[in] node Fence target * \param[out] history Where to store list of fencing actions * \param[in] timeout Error if unable to complete within this * * \return Legacy Pacemaker return code */ int (*history)(stonith_t *stonith, int call_options, const char *node, stonith_history_t **history, int timeout); /*! * \brief Register a callback for fence notifications * * \param[in,out] stonith Fencer connection to use * \param[in] event Event to register for * \param[in] callback Callback to register * * \return Legacy Pacemaker return code */ int (*register_notification)(stonith_t *stonith, const char *event, void (*callback)(stonith_t *st, stonith_event_t *e)); /*! * \brief Unregister callbacks for fence notifications * * \param[in,out] stonith Fencer connection to use * \param[in] event Event to unregister callbacks for (NULL for all) * * \return Legacy Pacemaker return code */ int (*remove_notification)(stonith_t *stonith, const char *event); /*! * \brief Register a callback for an asynchronous fencing result * * \param[in,out] stonith Fencer connection to use * \param[in] call_id Call ID to register callback for * \param[in] timeout Error if result not received in this time * \param[in] options Group of enum stonith_call_options * (respects \c st_opt_timeout_updates and * \c st_opt_report_only_success) * \param[in,out] user_data Pointer to pass to callback * \param[in] callback_name Unique identifier for callback * \param[in] callback Callback to register (may be called * immediately if \p call_id indicates error) * * \return \c TRUE on success, \c FALSE if call_id indicates error, * or -EINVAL if \p stonith is not valid */ int (*register_callback)(stonith_t *stonith, int call_id, int timeout, int options, void *user_data, const char *callback_name, void (*callback)(stonith_t *st, stonith_callback_data_t *data)); /*! * \brief Unregister callbacks for asynchronous fencing results * * \param[in,out] stonith Fencer connection to use * \param[in] call_id If \p all_callbacks is false, call ID * to unregister callback for * \param[in] all_callbacks If true, unregister all callbacks * * \return pcmk_ok */ int (*remove_callback)(stonith_t *stonith, int call_id, bool all_callbacks); /*! * \brief Unregister fencing level for specified node, pattern or attribute * * \param[in,out] st Fencer connection to use * \param[in] options Group of enum stonith_call_options * \param[in] node If not NULL, unregister level targeting this node * \param[in] pattern If not NULL, unregister level targeting nodes * whose names match this regular expression * \param[in] attr If this and \p value are not NULL, unregister * level targeting nodes with this node attribute * set to \p value * \param[in] value If this and \p attr are not NULL, unregister * level targeting nodes with node attribute \p attr * set to this * \param[in] level Topology level number to remove * * \return pcmk_ok (if synchronous) or positive call ID (if asynchronous) * on success, otherwise a negative legacy Pacemaker return code * \note The caller should set only one of \p node, \p pattern, or \p attr * and \p value. */ int (*remove_level_full)(stonith_t *st, int options, const char *node, const char *pattern, const char *attr, const char *value, int level); /*! * \brief Register fencing level for specified node, pattern or attribute * * \param[in,out] st Fencer connection to use * \param[in] options Group of enum stonith_call_options * \param[in] node If not NULL, register level targeting this * node by name * \param[in] pattern If not NULL, register level targeting nodes * whose names match this regular expression * \param[in] attr If this and \p value are not NULL, register * level targeting nodes with this node * attribute set to \p value * \param[in] value If this and \p attr are not NULL, register * level targeting nodes with node attribute * \p attr set to this * \param[in] level Topology level number to remove * \param[in] device_list Devices to use in level * * \return pcmk_ok (if synchronous) or positive call ID (if asynchronous) * on success, otherwise a negative legacy Pacemaker return code * * \note The caller should set only one of node, pattern or attr/value. */ int (*register_level_full)(stonith_t *st, int options, const char *node, const char *pattern, const char *attr, const char *value, int level, const stonith_key_value_t *device_list); /*! * \brief Validate an arbitrary stonith device configuration * * \param[in,out] st Fencer connection to use * \param[in] call_options Group of enum stonith_call_options * \param[in] rsc_id ID used to replace CIB secrets in \p params * \param[in] namespace_s Type of fence agent to validate ("redhat" * or "stonith-ng" for RHCS-style, "internal" * for Pacemaker-internal devices, "heartbeat" * for LHA-style, or "any" or NULL for any) * \param[in] agent Fence agent to validate * \param[in] params Configuration parameters to pass to agent * \param[in] timeout Fail if no response within this many seconds * \param[out] output If non-NULL, where to store any agent output * \param[out] error_output If non-NULL, where to store agent error output * * \return pcmk_ok if validation succeeds, -errno otherwise * \note If pcmk_ok is returned, the caller is responsible for freeing * the output (if requested) with free(). */ int (*validate)(stonith_t *st, int call_options, const char *rsc_id, const char *namespace_s, const char *agent, const stonith_key_value_t *params, int timeout, char **output, char **error_output); /*! * \brief Request delayed fencing of a target * * \param[in,out] stonith Fencer connection to use * \param[in] call_options Group of enum stonith_call_options * \param[in] node Fence target * \param[in] action "on", "off", or "reboot" * \param[in] timeout Default per-device timeout to use with * each executed device * \param[in] tolerance Accept result of identical fence action * completed within this time * \param[in] delay Execute fencing after this delay (-1 * disables any delay from pcmk_delay_base * and pcmk_delay_max) * * \return pcmk_ok (if synchronous) or positive call ID (if asynchronous) * on success, otherwise a negative legacy Pacemaker return code */ int (*fence_with_delay)(stonith_t *stonith, int call_options, const char *node, const char *action, int timeout, int tolerance, int delay); } stonith_api_operations_t; struct stonith_s { enum stonith_state state; int call_id; int call_timeout; //!< \deprecated Unused void *st_private; stonith_api_operations_t *cmds; }; /* *INDENT-ON* */ /* Core functions */ stonith_t *stonith_api_new(void); void stonith_api_delete(stonith_t * st); void stonith_dump_pending_callbacks(stonith_t * st); bool stonith_dispatch(stonith_t * st); stonith_key_value_t *stonith_key_value_add(stonith_key_value_t * kvp, const char *key, const char *value); void stonith_key_value_freeall(stonith_key_value_t * kvp, int keys, int values); void stonith_history_free(stonith_history_t *history); // Convenience functions int stonith_api_connect_retry(stonith_t *st, const char *name, int max_attempts); const char *stonith_op_state_str(enum op_state state); /* Basic helpers that allows nodes to be fenced and the history to be * queried without mainloop or the caller understanding the full API * * At least one of nodeid and uname are required * * NOTE: DLM uses both of these */ int stonith_api_kick(uint32_t nodeid, const char *uname, int timeout, bool off); time_t stonith_api_time(uint32_t nodeid, const char *uname, bool in_progress); /* * Helpers for using the above functions without install-time dependencies * * Usage: * #include * * To turn a node off by corosync nodeid: * stonith_api_kick_helper(nodeid, 120, 1); * * To check the last fence date/time (also by nodeid): * last = stonith_api_time_helper(nodeid, 0); * * To check if fencing is in progress: * if(stonith_api_time_helper(nodeid, 1) > 0) { ... } * * eg. #include #include #include int main(int argc, char ** argv) { int rc = 0; int nodeid = 102; rc = stonith_api_time_helper(nodeid, 0); printf("%d last fenced at %s\n", nodeid, ctime(rc)); rc = stonith_api_kick_helper(nodeid, 120, 1); printf("%d fence result: %d\n", nodeid, rc); rc = stonith_api_time_helper(nodeid, 0); printf("%d last fenced at %s\n", nodeid, ctime(rc)); return 0; } */ # define STONITH_LIBRARY "libstonithd.so.26" typedef int (*st_api_kick_fn) (int nodeid, const char *uname, int timeout, bool off); typedef time_t (*st_api_time_fn) (int nodeid, const char *uname, bool in_progress); static inline int stonith_api_kick_helper(uint32_t nodeid, int timeout, bool off) { static void *st_library = NULL; static st_api_kick_fn st_kick_fn; if (st_library == NULL) { st_library = dlopen(STONITH_LIBRARY, RTLD_LAZY); } if (st_library && st_kick_fn == NULL) { st_kick_fn = (st_api_kick_fn) dlsym(st_library, "stonith_api_kick"); } if (st_kick_fn == NULL) { #ifdef ELIBACC return -ELIBACC; #else return -ENOSYS; #endif } return (*st_kick_fn) (nodeid, NULL, timeout, off); } static inline time_t stonith_api_time_helper(uint32_t nodeid, bool in_progress) { static void *st_library = NULL; static st_api_time_fn st_time_fn; if (st_library == NULL) { st_library = dlopen(STONITH_LIBRARY, RTLD_LAZY); } if (st_library && st_time_fn == NULL) { st_time_fn = (st_api_time_fn) dlsym(st_library, "stonith_api_time"); } if (st_time_fn == NULL) { return 0; } return (*st_time_fn) (nodeid, NULL, in_progress); } /** * Does the given agent describe a stonith resource that can exist? * * \param[in] agent What is the name of the agent? * \param[in] timeout Timeout to use when querying. If 0 is given, * use a default of 120. * * \return A boolean */ bool stonith_agent_exists(const char *agent, int timeout); /*! * \brief Turn fence action into a more readable string * * \param[in] action Fence action */ const char *stonith_action_str(const char *action); -#if !defined(PCMK_ALLOW_DEPRECATED) || (PCMK_ALLOW_DEPRECATED == 1) -/* Normally we'd put this section in a separate file (crm/fencing/compat.h), but - * we can't do that for the reason noted at the top of this file. That does mean - * we have to duplicate these declarations where they're implemented. - */ - -//! \deprecated Do not use -#define T_STONITH_NOTIFY_DISCONNECT "st_notify_disconnect" - -//! \deprecated Do not use -#define T_STONITH_NOTIFY_FENCE "st_notify_fence" - -//! \deprecated Do not use -#define T_STONITH_NOTIFY_HISTORY "st_notify_history" - -//! \deprecated Do not use -#define T_STONITH_NOTIFY_HISTORY_SYNCED "st_notify_history_synced" - -//! \deprecated Use stonith_get_namespace() instead -const char *get_stonith_provider(const char *agent, const char *provider); - -#endif - #ifdef __cplusplus } #endif #endif diff --git a/lib/common/tests/nodes/pcmk_foreach_active_resource_test.c b/lib/common/tests/nodes/pcmk_foreach_active_resource_test.c index 2402789230..c453316e47 100644 --- a/lib/common/tests/nodes/pcmk_foreach_active_resource_test.c +++ b/lib/common/tests/nodes/pcmk_foreach_active_resource_test.c @@ -1,149 +1,149 @@ /* * Copyright 2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include // NULL #include // GList, TRUE, FALSE #include #include #include static int counter = 1; static int return_false = -1; static char rsc1_id[] = "rsc1"; static char rsc2_id[] = "rsc2"; static char rsc3_id[] = "rsc3"; static pcmk_resource_t rsc1 = { .id = rsc1_id, }; static pcmk_resource_t rsc2 = { .id = rsc2_id, }; static pcmk_resource_t rsc3 = { .id = rsc3_id, }; static bool fn(pcmk_resource_t *rsc, void *user_data) { char *expected_id = crm_strdup_printf("rsc%d", counter); assert_string_equal(rsc->id, expected_id); free(expected_id); return counter++ != return_false; } static void null_args(void **state) { - struct pe_node_shared_s shared = { + struct pcmk__node_details shared = { .running_rsc = NULL, }; pcmk_node_t node = { .details = &shared, }; counter = 1; // These just test that it doesn't crash pcmk_foreach_active_resource(NULL, NULL, NULL); pcmk_foreach_active_resource(&node, NULL, NULL); pcmk_foreach_active_resource(NULL, fn, NULL); assert_int_equal(counter, 1); } static void list_of_0(void **state) { - struct pe_node_shared_s shared = { + struct pcmk__node_details shared = { .running_rsc = NULL, }; pcmk_node_t node = { .details = &shared, }; counter = 1; pcmk_foreach_active_resource(&node, fn, NULL); assert_int_equal(counter, 1); } static void list_of_1(void **state) { - struct pe_node_shared_s shared = { + struct pcmk__node_details shared = { .running_rsc = NULL, }; pcmk_node_t node = { .details = &shared, }; shared.running_rsc = g_list_append(shared.running_rsc, &rsc1); counter = 1; pcmk_foreach_active_resource(&node, fn, NULL); assert_int_equal(counter, 2); g_list_free(shared.running_rsc); } static void list_of_3(void **state) { - struct pe_node_shared_s shared = { + struct pcmk__node_details shared = { .running_rsc = NULL, }; pcmk_node_t node = { .details = &shared, }; shared.running_rsc = g_list_append(shared.running_rsc, &rsc1); shared.running_rsc = g_list_append(shared.running_rsc, &rsc2); shared.running_rsc = g_list_append(shared.running_rsc, &rsc3); counter = 1; pcmk_foreach_active_resource(&node, fn, NULL); assert_int_equal(counter, 4); g_list_free(shared.running_rsc); } static void list_of_3_return_false(void **state) { - struct pe_node_shared_s shared = { + struct pcmk__node_details shared = { .running_rsc = NULL, }; pcmk_node_t node = { .details = &shared, }; shared.running_rsc = g_list_append(shared.running_rsc, &rsc1); shared.running_rsc = g_list_append(shared.running_rsc, &rsc2); shared.running_rsc = g_list_append(shared.running_rsc, &rsc3); counter = 1; return_false = 2; pcmk_foreach_active_resource(&node, fn, NULL); assert_int_equal(counter, 3); g_list_free(shared.running_rsc); } PCMK__UNIT_TEST(NULL, NULL, cmocka_unit_test(null_args), cmocka_unit_test(list_of_0), cmocka_unit_test(list_of_1), cmocka_unit_test(list_of_3), cmocka_unit_test(list_of_3_return_false)) diff --git a/lib/common/tests/nodes/pcmk_node_is_clean_test.c b/lib/common/tests/nodes/pcmk_node_is_clean_test.c index 0534633fdf..d6152ee535 100644 --- a/lib/common/tests/nodes/pcmk_node_is_clean_test.c +++ b/lib/common/tests/nodes/pcmk_node_is_clean_test.c @@ -1,54 +1,54 @@ /* * Copyright 2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include // NULL #include // TRUE, FALSE #include #include static void null_is_unclean(void **state) { assert_false(pcmk_node_is_clean(NULL)); } static void node_is_clean(void **state) { - struct pe_node_shared_s shared = { + struct pcmk__node_details shared = { .unclean = FALSE, }; pcmk_node_t node = { .details = &shared, }; assert_true(pcmk_node_is_clean(&node)); } static void node_is_unclean(void **state) { - struct pe_node_shared_s shared = { + struct pcmk__node_details shared = { .unclean = TRUE, }; pcmk_node_t node = { .details = &shared, }; assert_false(pcmk_node_is_clean(&node)); } PCMK__UNIT_TEST(NULL, NULL, cmocka_unit_test(null_is_unclean), cmocka_unit_test(node_is_clean), cmocka_unit_test(node_is_unclean)) diff --git a/lib/common/tests/nodes/pcmk_node_is_in_maintenance_test.c b/lib/common/tests/nodes/pcmk_node_is_in_maintenance_test.c index 45a3b6fdca..a33e6ef1ac 100644 --- a/lib/common/tests/nodes/pcmk_node_is_in_maintenance_test.c +++ b/lib/common/tests/nodes/pcmk_node_is_in_maintenance_test.c @@ -1,54 +1,54 @@ /* * Copyright 2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include // NULL #include // TRUE, FALSE #include #include static void null_is_not_in_maintenance(void **state) { assert_false(pcmk_node_is_in_maintenance(NULL)); } static void node_is_in_maintenance(void **state) { - struct pe_node_shared_s shared = { + struct pcmk__node_details shared = { .maintenance = TRUE, }; pcmk_node_t node = { .details = &shared, }; assert_true(pcmk_node_is_in_maintenance(&node)); } static void node_is_not_in_maintenance(void **state) { - struct pe_node_shared_s shared = { + struct pcmk__node_details shared = { .maintenance = FALSE, }; pcmk_node_t node = { .details = &shared, }; assert_false(pcmk_node_is_in_maintenance(&node)); } PCMK__UNIT_TEST(NULL, NULL, cmocka_unit_test(null_is_not_in_maintenance), cmocka_unit_test(node_is_in_maintenance), cmocka_unit_test(node_is_not_in_maintenance)) diff --git a/lib/common/tests/nodes/pcmk_node_is_online_test.c b/lib/common/tests/nodes/pcmk_node_is_online_test.c index d22e3b4dd3..209544779d 100644 --- a/lib/common/tests/nodes/pcmk_node_is_online_test.c +++ b/lib/common/tests/nodes/pcmk_node_is_online_test.c @@ -1,54 +1,54 @@ /* * Copyright 2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include // NULL #include // TRUE, FALSE #include #include static void null_is_offline(void **state) { assert_false(pcmk_node_is_online(NULL)); } static void node_is_online(void **state) { - struct pe_node_shared_s shared = { + struct pcmk__node_details shared = { .online = TRUE, }; pcmk_node_t node = { .details = &shared, }; assert_true(pcmk_node_is_online(&node)); } static void node_is_offline(void **state) { - struct pe_node_shared_s shared = { + struct pcmk__node_details shared = { .online = FALSE, }; pcmk_node_t node = { .details = &shared, }; assert_false(pcmk_node_is_online(&node)); } PCMK__UNIT_TEST(NULL, NULL, cmocka_unit_test(null_is_offline), cmocka_unit_test(node_is_online), cmocka_unit_test(node_is_offline)) diff --git a/lib/common/tests/nodes/pcmk_node_is_pending_test.c b/lib/common/tests/nodes/pcmk_node_is_pending_test.c index 9f2abca4e9..6aa2781764 100644 --- a/lib/common/tests/nodes/pcmk_node_is_pending_test.c +++ b/lib/common/tests/nodes/pcmk_node_is_pending_test.c @@ -1,54 +1,54 @@ /* * Copyright 2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include // NULL #include // TRUE, FALSE #include #include static void null_is_not_pending(void **state) { assert_false(pcmk_node_is_pending(NULL)); } static void node_is_pending(void **state) { - struct pe_node_shared_s shared = { + struct pcmk__node_details shared = { .pending = TRUE, }; pcmk_node_t node = { .details = &shared, }; assert_true(pcmk_node_is_pending(&node)); } static void node_is_not_pending(void **state) { - struct pe_node_shared_s shared = { + struct pcmk__node_details shared = { .pending = FALSE, }; pcmk_node_t node = { .details = &shared, }; assert_false(pcmk_node_is_pending(&node)); } PCMK__UNIT_TEST(NULL, NULL, cmocka_unit_test(null_is_not_pending), cmocka_unit_test(node_is_pending), cmocka_unit_test(node_is_not_pending)) diff --git a/lib/common/tests/nodes/pcmk_node_is_shutting_down_test.c b/lib/common/tests/nodes/pcmk_node_is_shutting_down_test.c index b6054b07dd..4af0c16a5b 100644 --- a/lib/common/tests/nodes/pcmk_node_is_shutting_down_test.c +++ b/lib/common/tests/nodes/pcmk_node_is_shutting_down_test.c @@ -1,54 +1,54 @@ /* * Copyright 2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include // NULL #include // TRUE, FALSE #include #include static void null_is_not_shutting_down(void **state) { assert_false(pcmk_node_is_shutting_down(NULL)); } static void node_is_shutting_down(void **state) { - struct pe_node_shared_s shared = { + struct pcmk__node_details shared = { .shutdown = TRUE, }; pcmk_node_t node = { .details = &shared, }; assert_true(pcmk_node_is_shutting_down(&node)); } static void node_is_not_shutting_down(void **state) { - struct pe_node_shared_s shared = { + struct pcmk__node_details shared = { .shutdown = FALSE, }; pcmk_node_t node = { .details = &shared, }; assert_false(pcmk_node_is_shutting_down(&node)); } PCMK__UNIT_TEST(NULL, NULL, cmocka_unit_test(null_is_not_shutting_down), cmocka_unit_test(node_is_shutting_down), cmocka_unit_test(node_is_not_shutting_down)) diff --git a/lib/fencing/st_client.c b/lib/fencing/st_client.c index fcd5225d2e..4047e47e83 100644 --- a/lib/fencing/st_client.c +++ b/lib/fencing/st_client.c @@ -1,2719 +1,2704 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include "fencing_private.h" CRM_TRACE_INIT_DATA(stonith); // Used as stonith_t:st_private typedef struct stonith_private_s { char *token; crm_ipc_t *ipc; mainloop_io_t *source; GHashTable *stonith_op_callback_table; GList *notify_list; int notify_refcnt; bool notify_deletes; void (*op_callback) (stonith_t * st, stonith_callback_data_t * data); } stonith_private_t; // Used as stonith_event_t:opaque struct event_private { pcmk__action_result_t result; }; typedef struct stonith_notify_client_s { const char *event; const char *obj_id; /* implement one day */ const char *obj_type; /* implement one day */ void (*notify) (stonith_t * st, stonith_event_t * e); bool delete; } stonith_notify_client_t; typedef struct stonith_callback_client_s { void (*callback) (stonith_t * st, stonith_callback_data_t * data); const char *id; void *user_data; gboolean only_success; gboolean allow_timeout_updates; struct timer_rec_s *timer; } stonith_callback_client_t; struct notify_blob_s { stonith_t *stonith; xmlNode *xml; }; struct timer_rec_s { int call_id; int timeout; guint ref; stonith_t *stonith; }; typedef int (*stonith_op_t) (const char *, int, const char *, xmlNode *, xmlNode *, xmlNode *, xmlNode **, xmlNode **); bool stonith_dispatch(stonith_t * st); xmlNode *stonith_create_op(int call_id, const char *token, const char *op, xmlNode * data, int call_options); static int stonith_send_command(stonith_t *stonith, const char *op, xmlNode *data, xmlNode **output_data, int call_options, int timeout); static void stonith_connection_destroy(gpointer user_data); static void stonith_send_notification(gpointer data, gpointer user_data); static int stonith_api_del_notification(stonith_t *stonith, const char *event); /*! * \brief Get agent namespace by name * * \param[in] namespace_s Name of namespace as string * * \return Namespace as enum value */ enum stonith_namespace stonith_text2namespace(const char *namespace_s) { if (pcmk__str_eq(namespace_s, "any", pcmk__str_null_matches)) { return st_namespace_any; } else if (!strcmp(namespace_s, "redhat") || !strcmp(namespace_s, "stonith-ng")) { return st_namespace_rhcs; } else if (!strcmp(namespace_s, "internal")) { return st_namespace_internal; } else if (!strcmp(namespace_s, "heartbeat")) { return st_namespace_lha; } return st_namespace_invalid; } /*! * \brief Get agent namespace name * * \param[in] namespace Namespace as enum value * * \return Namespace name as string */ const char * stonith_namespace2text(enum stonith_namespace st_namespace) { switch (st_namespace) { case st_namespace_any: return "any"; case st_namespace_rhcs: return "stonith-ng"; case st_namespace_internal: return "internal"; case st_namespace_lha: return "heartbeat"; default: break; } return "unsupported"; } /*! * \brief Determine namespace of a fence agent * * \param[in] agent Fence agent type * \param[in] namespace_s Name of agent namespace as string, if known * * \return Namespace of specified agent, as enum value */ enum stonith_namespace stonith_get_namespace(const char *agent, const char *namespace_s) { if (pcmk__str_eq(namespace_s, "internal", pcmk__str_none)) { return st_namespace_internal; } if (stonith__agent_is_rhcs(agent)) { return st_namespace_rhcs; } #if HAVE_STONITH_STONITH_H if (stonith__agent_is_lha(agent)) { return st_namespace_lha; } #endif crm_err("Unknown fence agent: %s", agent); return st_namespace_invalid; } gboolean stonith__watchdog_fencing_enabled_for_node_api(stonith_t *st, const char *node) { gboolean rv = FALSE; stonith_t *stonith_api = st?st:stonith_api_new(); char *list = NULL; if(stonith_api) { if (stonith_api->state == stonith_disconnected) { int rc = stonith_api->cmds->connect(stonith_api, "stonith-api", NULL); if (rc != pcmk_ok) { crm_err("Failed connecting to Stonith-API for watchdog-fencing-query."); } } if (stonith_api->state != stonith_disconnected) { /* caveat!!! * this might fail when when stonithd is just updating the device-list * probably something we should fix as well for other api-calls */ int rc = stonith_api->cmds->list(stonith_api, st_opt_sync_call, STONITH_WATCHDOG_ID, &list, 0); if ((rc != pcmk_ok) || (list == NULL)) { /* due to the race described above it can happen that * we drop in here - so as not to make remote nodes * panic on that answer */ if (rc == -ENODEV) { crm_notice("Cluster does not have watchdog fencing device"); } else { crm_warn("Could not check for watchdog fencing device: %s", pcmk_strerror(rc)); } } else if (list[0] == '\0') { rv = TRUE; } else { GList *targets = stonith__parse_targets(list); rv = pcmk__str_in_list(node, targets, pcmk__str_casei); g_list_free_full(targets, free); } free(list); if (!st) { /* if we're provided the api we still might have done the * connection - but let's assume the caller won't bother */ stonith_api->cmds->disconnect(stonith_api); } } if (!st) { stonith_api_delete(stonith_api); } } else { crm_err("Stonith-API for watchdog-fencing-query couldn't be created."); } crm_trace("Pacemaker assumes node %s %sto do watchdog-fencing.", node, rv?"":"not "); return rv; } gboolean stonith__watchdog_fencing_enabled_for_node(const char *node) { return stonith__watchdog_fencing_enabled_for_node_api(NULL, node); } /* when cycling through the list we don't want to delete items so just mark them and when we know nobody is using the list loop over it to remove the marked items */ static void foreach_notify_entry (stonith_private_t *private, GFunc func, gpointer user_data) { private->notify_refcnt++; g_list_foreach(private->notify_list, func, user_data); private->notify_refcnt--; if ((private->notify_refcnt == 0) && private->notify_deletes) { GList *list_item = private->notify_list; private->notify_deletes = FALSE; while (list_item != NULL) { stonith_notify_client_t *list_client = list_item->data; GList *next = g_list_next(list_item); if (list_client->delete) { free(list_client); private->notify_list = g_list_delete_link(private->notify_list, list_item); } list_item = next; } } } static void stonith_connection_destroy(gpointer user_data) { stonith_t *stonith = user_data; stonith_private_t *native = NULL; struct notify_blob_s blob; crm_trace("Sending destroyed notification"); blob.stonith = stonith; blob.xml = pcmk__xe_create(NULL, PCMK__XE_NOTIFY); native = stonith->st_private; native->ipc = NULL; native->source = NULL; free(native->token); native->token = NULL; stonith->state = stonith_disconnected; crm_xml_add(blob.xml, PCMK__XA_T, PCMK__VALUE_ST_NOTIFY); crm_xml_add(blob.xml, PCMK__XA_SUBT, PCMK__VALUE_ST_NOTIFY_DISCONNECT); foreach_notify_entry(native, stonith_send_notification, &blob); pcmk__xml_free(blob.xml); } xmlNode * create_device_registration_xml(const char *id, enum stonith_namespace namespace, const char *agent, const stonith_key_value_t *params, const char *rsc_provides) { xmlNode *data = pcmk__xe_create(NULL, PCMK__XE_ST_DEVICE_ID); xmlNode *args = pcmk__xe_create(data, PCMK__XE_ATTRIBUTES); #if HAVE_STONITH_STONITH_H if (namespace == st_namespace_any) { namespace = stonith_get_namespace(agent, NULL); } if (namespace == st_namespace_lha) { hash2field((gpointer) "plugin", (gpointer) agent, args); agent = "fence_legacy"; } #endif crm_xml_add(data, PCMK_XA_ID, id); crm_xml_add(data, PCMK__XA_ST_ORIGIN, __func__); crm_xml_add(data, PCMK_XA_AGENT, agent); if ((namespace != st_namespace_any) && (namespace != st_namespace_invalid)) { crm_xml_add(data, PCMK__XA_NAMESPACE, stonith_namespace2text(namespace)); } if (rsc_provides) { crm_xml_add(data, PCMK__XA_RSC_PROVIDES, rsc_provides); } for (; params; params = params->next) { hash2field((gpointer) params->key, (gpointer) params->value, args); } return data; } static int stonith_api_register_device(stonith_t *st, int call_options, const char *id, const char *namespace_s, const char *agent, const stonith_key_value_t *params) { int rc = 0; xmlNode *data = NULL; data = create_device_registration_xml(id, stonith_text2namespace(namespace_s), agent, params, NULL); rc = stonith_send_command(st, STONITH_OP_DEVICE_ADD, data, NULL, call_options, 0); pcmk__xml_free(data); return rc; } static int stonith_api_remove_device(stonith_t * st, int call_options, const char *name) { int rc = 0; xmlNode *data = NULL; data = pcmk__xe_create(NULL, PCMK__XE_ST_DEVICE_ID); crm_xml_add(data, PCMK__XA_ST_ORIGIN, __func__); crm_xml_add(data, PCMK_XA_ID, name); rc = stonith_send_command(st, STONITH_OP_DEVICE_DEL, data, NULL, call_options, 0); pcmk__xml_free(data); return rc; } static int stonith_api_remove_level_full(stonith_t *st, int options, const char *node, const char *pattern, const char *attr, const char *value, int level) { int rc = 0; xmlNode *data = NULL; CRM_CHECK(node || pattern || (attr && value), return -EINVAL); data = pcmk__xe_create(NULL, PCMK_XE_FENCING_LEVEL); crm_xml_add(data, PCMK__XA_ST_ORIGIN, __func__); if (node) { crm_xml_add(data, PCMK_XA_TARGET, node); } else if (pattern) { crm_xml_add(data, PCMK_XA_TARGET_PATTERN, pattern); } else { crm_xml_add(data, PCMK_XA_TARGET_ATTRIBUTE, attr); crm_xml_add(data, PCMK_XA_TARGET_VALUE, value); } crm_xml_add_int(data, PCMK_XA_INDEX, level); rc = stonith_send_command(st, STONITH_OP_LEVEL_DEL, data, NULL, options, 0); pcmk__xml_free(data); return rc; } static int stonith_api_remove_level(stonith_t * st, int options, const char *node, int level) { return stonith_api_remove_level_full(st, options, node, NULL, NULL, NULL, level); } /*! * \internal * \brief Create XML for fence topology level registration request * * \param[in] node If not NULL, target level by this node name * \param[in] pattern If not NULL, target by node name using this regex * \param[in] attr If not NULL, target by this node attribute * \param[in] value If not NULL, target by this node attribute value * \param[in] level Index number of level to register * \param[in] device_list List of devices in level * * \return Newly allocated XML tree on success, NULL otherwise * * \note The caller should set only one of node, pattern or attr/value. */ xmlNode * create_level_registration_xml(const char *node, const char *pattern, const char *attr, const char *value, int level, const stonith_key_value_t *device_list) { GString *list = NULL; xmlNode *data; CRM_CHECK(node || pattern || (attr && value), return NULL); data = pcmk__xe_create(NULL, PCMK_XE_FENCING_LEVEL); crm_xml_add(data, PCMK__XA_ST_ORIGIN, __func__); crm_xml_add_int(data, PCMK_XA_ID, level); crm_xml_add_int(data, PCMK_XA_INDEX, level); if (node) { crm_xml_add(data, PCMK_XA_TARGET, node); } else if (pattern) { crm_xml_add(data, PCMK_XA_TARGET_PATTERN, pattern); } else { crm_xml_add(data, PCMK_XA_TARGET_ATTRIBUTE, attr); crm_xml_add(data, PCMK_XA_TARGET_VALUE, value); } for (; device_list; device_list = device_list->next) { pcmk__add_separated_word(&list, 1024, device_list->value, ","); } if (list != NULL) { crm_xml_add(data, PCMK_XA_DEVICES, (const char *) list->str); g_string_free(list, TRUE); } return data; } static int stonith_api_register_level_full(stonith_t *st, int options, const char *node, const char *pattern, const char *attr, const char *value, int level, const stonith_key_value_t *device_list) { int rc = 0; xmlNode *data = create_level_registration_xml(node, pattern, attr, value, level, device_list); CRM_CHECK(data != NULL, return -EINVAL); rc = stonith_send_command(st, STONITH_OP_LEVEL_ADD, data, NULL, options, 0); pcmk__xml_free(data); return rc; } static int stonith_api_register_level(stonith_t * st, int options, const char *node, int level, const stonith_key_value_t * device_list) { return stonith_api_register_level_full(st, options, node, NULL, NULL, NULL, level, device_list); } static int stonith_api_device_list(stonith_t *stonith, int call_options, const char *namespace_s, stonith_key_value_t **devices, int timeout) { int count = 0; enum stonith_namespace ns = stonith_text2namespace(namespace_s); if (devices == NULL) { crm_err("Parameter error: stonith_api_device_list"); return -EFAULT; } #if HAVE_STONITH_STONITH_H // Include Linux-HA agents if requested if ((ns == st_namespace_any) || (ns == st_namespace_lha)) { count += stonith__list_lha_agents(devices); } #endif // Include Red Hat agents if requested if ((ns == st_namespace_any) || (ns == st_namespace_rhcs)) { count += stonith__list_rhcs_agents(devices); } return count; } // See stonith_api_operations_t:metadata() documentation static int stonith_api_device_metadata(stonith_t *stonith, int call_options, const char *agent, const char *namespace_s, char **output, int timeout_sec) { /* By executing meta-data directly, we can get it from stonith_admin when * the cluster is not running, which is important for higher-level tools. */ enum stonith_namespace ns = stonith_get_namespace(agent, namespace_s); if (timeout_sec <= 0) { timeout_sec = PCMK_DEFAULT_METADATA_TIMEOUT_MS; } crm_trace("Looking up metadata for %s agent %s", stonith_namespace2text(ns), agent); switch (ns) { case st_namespace_rhcs: return stonith__rhcs_metadata(agent, timeout_sec, output); #if HAVE_STONITH_STONITH_H case st_namespace_lha: return stonith__lha_metadata(agent, timeout_sec, output); #endif default: crm_err("Can't get fence agent '%s' meta-data: No such agent", agent); break; } return -ENODEV; } static int stonith_api_query(stonith_t * stonith, int call_options, const char *target, stonith_key_value_t ** devices, int timeout) { int rc = 0, lpc = 0, max = 0; xmlNode *data = NULL; xmlNode *output = NULL; xmlXPathObjectPtr xpathObj = NULL; CRM_CHECK(devices != NULL, return -EINVAL); data = pcmk__xe_create(NULL, PCMK__XE_ST_DEVICE_ID); crm_xml_add(data, PCMK__XA_ST_ORIGIN, __func__); crm_xml_add(data, PCMK__XA_ST_TARGET, target); crm_xml_add(data, PCMK__XA_ST_DEVICE_ACTION, PCMK_ACTION_OFF); rc = stonith_send_command(stonith, STONITH_OP_QUERY, data, &output, call_options, timeout); if (rc < 0) { return rc; } xpathObj = xpath_search(output, "//@agent"); if (xpathObj) { max = numXpathResults(xpathObj); for (lpc = 0; lpc < max; lpc++) { xmlNode *match = getXpathResult(xpathObj, lpc); CRM_LOG_ASSERT(match != NULL); if(match != NULL) { xmlChar *match_path = xmlGetNodePath(match); crm_info("%s[%d] = %s", "//@agent", lpc, match_path); free(match_path); *devices = stonith_key_value_add(*devices, NULL, crm_element_value(match, PCMK_XA_ID)); } } freeXpathObject(xpathObj); } pcmk__xml_free(output); pcmk__xml_free(data); return max; } /*! * \internal * \brief Make a STONITH_OP_EXEC request * * \param[in,out] stonith Fencer connection * \param[in] call_options Bitmask of \c stonith_call_options * \param[in] id Fence device ID that request is for * \param[in] action Agent action to request (list, status, monitor) * \param[in] target Name of target node for requested action * \param[in] timeout_sec Error if not completed within this many seconds * \param[out] output Where to set agent output */ static int stonith_api_call(stonith_t *stonith, int call_options, const char *id, const char *action, const char *target, int timeout_sec, xmlNode **output) { int rc = 0; xmlNode *data = NULL; data = pcmk__xe_create(NULL, PCMK__XE_ST_DEVICE_ID); crm_xml_add(data, PCMK__XA_ST_ORIGIN, __func__); crm_xml_add(data, PCMK__XA_ST_DEVICE_ID, id); crm_xml_add(data, PCMK__XA_ST_DEVICE_ACTION, action); crm_xml_add(data, PCMK__XA_ST_TARGET, target); rc = stonith_send_command(stonith, STONITH_OP_EXEC, data, output, call_options, timeout_sec); pcmk__xml_free(data); return rc; } static int stonith_api_list(stonith_t * stonith, int call_options, const char *id, char **list_info, int timeout) { int rc; xmlNode *output = NULL; rc = stonith_api_call(stonith, call_options, id, PCMK_ACTION_LIST, NULL, timeout, &output); if (output && list_info) { const char *list_str; list_str = crm_element_value(output, PCMK__XA_ST_OUTPUT); if (list_str) { *list_info = strdup(list_str); } } if (output) { pcmk__xml_free(output); } return rc; } static int stonith_api_monitor(stonith_t * stonith, int call_options, const char *id, int timeout) { return stonith_api_call(stonith, call_options, id, PCMK_ACTION_MONITOR, NULL, timeout, NULL); } static int stonith_api_status(stonith_t * stonith, int call_options, const char *id, const char *port, int timeout) { return stonith_api_call(stonith, call_options, id, PCMK_ACTION_STATUS, port, timeout, NULL); } static int stonith_api_fence_with_delay(stonith_t * stonith, int call_options, const char *node, const char *action, int timeout, int tolerance, int delay) { int rc = 0; xmlNode *data = NULL; data = pcmk__xe_create(NULL, __func__); crm_xml_add(data, PCMK__XA_ST_TARGET, node); crm_xml_add(data, PCMK__XA_ST_DEVICE_ACTION, action); crm_xml_add_int(data, PCMK__XA_ST_TIMEOUT, timeout); crm_xml_add_int(data, PCMK__XA_ST_TOLERANCE, tolerance); crm_xml_add_int(data, PCMK__XA_ST_DELAY, delay); rc = stonith_send_command(stonith, STONITH_OP_FENCE, data, NULL, call_options, timeout); pcmk__xml_free(data); return rc; } static int stonith_api_fence(stonith_t * stonith, int call_options, const char *node, const char *action, int timeout, int tolerance) { return stonith_api_fence_with_delay(stonith, call_options, node, action, timeout, tolerance, 0); } static int stonith_api_confirm(stonith_t * stonith, int call_options, const char *target) { stonith__set_call_options(call_options, target, st_opt_manual_ack); return stonith_api_fence(stonith, call_options, target, PCMK_ACTION_OFF, 0, 0); } static int stonith_api_history(stonith_t * stonith, int call_options, const char *node, stonith_history_t ** history, int timeout) { int rc = 0; xmlNode *data = NULL; xmlNode *output = NULL; stonith_history_t *last = NULL; *history = NULL; if (node) { data = pcmk__xe_create(NULL, __func__); crm_xml_add(data, PCMK__XA_ST_TARGET, node); } stonith__set_call_options(call_options, node, st_opt_sync_call); rc = stonith_send_command(stonith, STONITH_OP_FENCE_HISTORY, data, &output, call_options, timeout); pcmk__xml_free(data); if (rc == 0) { xmlNode *op = NULL; xmlNode *reply = get_xpath_object("//" PCMK__XE_ST_HISTORY, output, LOG_NEVER); for (op = pcmk__xe_first_child(reply, NULL, NULL, NULL); op != NULL; op = pcmk__xe_next(op)) { stonith_history_t *kvp; long long completed; long long completed_nsec = 0L; kvp = pcmk__assert_alloc(1, sizeof(stonith_history_t)); kvp->target = crm_element_value_copy(op, PCMK__XA_ST_TARGET); kvp->action = crm_element_value_copy(op, PCMK__XA_ST_DEVICE_ACTION); kvp->origin = crm_element_value_copy(op, PCMK__XA_ST_ORIGIN); kvp->delegate = crm_element_value_copy(op, PCMK__XA_ST_DELEGATE); kvp->client = crm_element_value_copy(op, PCMK__XA_ST_CLIENTNAME); crm_element_value_ll(op, PCMK__XA_ST_DATE, &completed); kvp->completed = (time_t) completed; crm_element_value_ll(op, PCMK__XA_ST_DATE_NSEC, &completed_nsec); kvp->completed_nsec = completed_nsec; crm_element_value_int(op, PCMK__XA_ST_STATE, &kvp->state); kvp->exit_reason = crm_element_value_copy(op, PCMK_XA_EXIT_REASON); if (last) { last->next = kvp; } else { *history = kvp; } last = kvp; } } pcmk__xml_free(output); return rc; } void stonith_history_free(stonith_history_t *history) { stonith_history_t *hp, *hp_old; for (hp = history; hp; hp_old = hp, hp = hp->next, free(hp_old)) { free(hp->target); free(hp->action); free(hp->origin); free(hp->delegate); free(hp->client); free(hp->exit_reason); } } static gint stonithlib_GCompareFunc(gconstpointer a, gconstpointer b) { int rc = 0; const stonith_notify_client_t *a_client = a; const stonith_notify_client_t *b_client = b; if (a_client->delete || b_client->delete) { /* make entries marked for deletion not findable */ return -1; } CRM_CHECK(a_client->event != NULL && b_client->event != NULL, return 0); rc = strcmp(a_client->event, b_client->event); if (rc == 0) { if (a_client->notify == NULL || b_client->notify == NULL) { return 0; } else if (a_client->notify == b_client->notify) { return 0; } else if (((long)a_client->notify) < ((long)b_client->notify)) { crm_err("callbacks for %s are not equal: %p vs. %p", a_client->event, a_client->notify, b_client->notify); return -1; } crm_err("callbacks for %s are not equal: %p vs. %p", a_client->event, a_client->notify, b_client->notify); return 1; } return rc; } xmlNode * stonith_create_op(int call_id, const char *token, const char *op, xmlNode * data, int call_options) { xmlNode *op_msg = NULL; CRM_CHECK(token != NULL, return NULL); op_msg = pcmk__xe_create(NULL, PCMK__XE_STONITH_COMMAND); crm_xml_add(op_msg, PCMK__XA_T, PCMK__VALUE_STONITH_NG); crm_xml_add(op_msg, PCMK__XA_ST_OP, op); crm_xml_add_int(op_msg, PCMK__XA_ST_CALLID, call_id); crm_trace("Sending call options: %.8lx, %d", (long)call_options, call_options); crm_xml_add_int(op_msg, PCMK__XA_ST_CALLOPT, call_options); if (data != NULL) { xmlNode *wrapper = pcmk__xe_create(op_msg, PCMK__XE_ST_CALLDATA); pcmk__xml_copy(wrapper, data); } return op_msg; } static void stonith_destroy_op_callback(gpointer data) { stonith_callback_client_t *blob = data; if (blob->timer && blob->timer->ref > 0) { g_source_remove(blob->timer->ref); } free(blob->timer); free(blob); } static int stonith_api_signoff(stonith_t * stonith) { stonith_private_t *native = stonith->st_private; crm_debug("Disconnecting from the fencer"); if (native->source != NULL) { /* Attached to mainloop */ mainloop_del_ipc_client(native->source); native->source = NULL; native->ipc = NULL; } else if (native->ipc) { /* Not attached to mainloop */ crm_ipc_t *ipc = native->ipc; native->ipc = NULL; crm_ipc_close(ipc); crm_ipc_destroy(ipc); } free(native->token); native->token = NULL; stonith->state = stonith_disconnected; return pcmk_ok; } static int stonith_api_del_callback(stonith_t * stonith, int call_id, bool all_callbacks) { stonith_private_t *private = stonith->st_private; if (all_callbacks) { private->op_callback = NULL; g_hash_table_destroy(private->stonith_op_callback_table); private->stonith_op_callback_table = pcmk__intkey_table(stonith_destroy_op_callback); } else if (call_id == 0) { private->op_callback = NULL; } else { pcmk__intkey_table_remove(private->stonith_op_callback_table, call_id); } return pcmk_ok; } /*! * \internal * \brief Invoke a (single) specified fence action callback * * \param[in,out] st Fencer API connection * \param[in] call_id If positive, call ID of completed fence action, * otherwise legacy return code for early failure * \param[in,out] result Full result for action * \param[in,out] userdata User data to pass to callback * \param[in] callback Fence action callback to invoke */ static void invoke_fence_action_callback(stonith_t *st, int call_id, pcmk__action_result_t *result, void *userdata, void (*callback) (stonith_t *st, stonith_callback_data_t *data)) { stonith_callback_data_t data = { 0, }; data.call_id = call_id; data.rc = pcmk_rc2legacy(stonith__result2rc(result)); data.userdata = userdata; data.opaque = (void *) result; callback(st, &data); } /*! * \internal * \brief Invoke any callbacks registered for a specified fence action result * * Given a fence action result from the fencer, invoke any callback registered * for that action, as well as any global callback registered. * * \param[in,out] stonith Fencer API connection * \param[in] msg If non-NULL, fencer reply * \param[in] call_id If \p msg is NULL, call ID of action that timed out */ static void invoke_registered_callbacks(stonith_t *stonith, const xmlNode *msg, int call_id) { stonith_private_t *private = NULL; stonith_callback_client_t *cb_info = NULL; pcmk__action_result_t result = PCMK__UNKNOWN_RESULT; CRM_CHECK(stonith != NULL, return); CRM_CHECK(stonith->st_private != NULL, return); private = stonith->st_private; if (msg == NULL) { // Fencer didn't reply in time pcmk__set_result(&result, CRM_EX_ERROR, PCMK_EXEC_TIMEOUT, "Fencer accepted request but did not reply in time"); CRM_LOG_ASSERT(call_id > 0); } else { // We have the fencer reply if ((crm_element_value_int(msg, PCMK__XA_ST_CALLID, &call_id) != 0) || (call_id <= 0)) { crm_log_xml_warn(msg, "Bad fencer reply"); } stonith__xe_get_result(msg, &result); } if (call_id > 0) { cb_info = pcmk__intkey_table_lookup(private->stonith_op_callback_table, call_id); } if ((cb_info != NULL) && (cb_info->callback != NULL) && (pcmk__result_ok(&result) || !(cb_info->only_success))) { crm_trace("Invoking callback %s for call %d", pcmk__s(cb_info->id, "without ID"), call_id); invoke_fence_action_callback(stonith, call_id, &result, cb_info->user_data, cb_info->callback); } else if ((private->op_callback == NULL) && !pcmk__result_ok(&result)) { crm_warn("Fencing action without registered callback failed: %d (%s%s%s)", result.exit_status, pcmk_exec_status_str(result.execution_status), ((result.exit_reason == NULL)? "" : ": "), ((result.exit_reason == NULL)? "" : result.exit_reason)); crm_log_xml_debug(msg, "Failed fence update"); } if (private->op_callback != NULL) { crm_trace("Invoking global callback for call %d", call_id); invoke_fence_action_callback(stonith, call_id, &result, NULL, private->op_callback); } if (cb_info != NULL) { stonith_api_del_callback(stonith, call_id, FALSE); } pcmk__reset_result(&result); } static gboolean stonith_async_timeout_handler(gpointer data) { struct timer_rec_s *timer = data; crm_err("Async call %d timed out after %dms", timer->call_id, timer->timeout); invoke_registered_callbacks(timer->stonith, NULL, timer->call_id); /* Always return TRUE, never remove the handler * We do that in stonith_del_callback() */ return TRUE; } static void set_callback_timeout(stonith_callback_client_t * callback, stonith_t * stonith, int call_id, int timeout) { struct timer_rec_s *async_timer = callback->timer; if (timeout <= 0) { return; } if (!async_timer) { async_timer = pcmk__assert_alloc(1, sizeof(struct timer_rec_s)); callback->timer = async_timer; } async_timer->stonith = stonith; async_timer->call_id = call_id; /* Allow a fair bit of grace to allow the server to tell us of a timeout * This is only a fallback */ async_timer->timeout = (timeout + 60) * 1000; if (async_timer->ref) { g_source_remove(async_timer->ref); } async_timer->ref = g_timeout_add(async_timer->timeout, stonith_async_timeout_handler, async_timer); } static void update_callback_timeout(int call_id, int timeout, stonith_t * st) { stonith_callback_client_t *callback = NULL; stonith_private_t *private = st->st_private; callback = pcmk__intkey_table_lookup(private->stonith_op_callback_table, call_id); if (!callback || !callback->allow_timeout_updates) { return; } set_callback_timeout(callback, st, call_id, timeout); } static int stonith_dispatch_internal(const char *buffer, ssize_t length, gpointer userdata) { const char *type = NULL; struct notify_blob_s blob; stonith_t *st = userdata; stonith_private_t *private = NULL; CRM_ASSERT(st != NULL); private = st->st_private; blob.stonith = st; blob.xml = pcmk__xml_parse(buffer); if (blob.xml == NULL) { crm_warn("Received malformed message from fencer: %s", buffer); return 0; } /* do callbacks */ type = crm_element_value(blob.xml, PCMK__XA_T); crm_trace("Activating %s callbacks...", type); if (pcmk__str_eq(type, PCMK__VALUE_STONITH_NG, pcmk__str_none)) { invoke_registered_callbacks(st, blob.xml, 0); } else if (pcmk__str_eq(type, PCMK__VALUE_ST_NOTIFY, pcmk__str_none)) { foreach_notify_entry(private, stonith_send_notification, &blob); } else if (pcmk__str_eq(type, PCMK__VALUE_ST_ASYNC_TIMEOUT_VALUE, pcmk__str_none)) { int call_id = 0; int timeout = 0; crm_element_value_int(blob.xml, PCMK__XA_ST_TIMEOUT, &timeout); crm_element_value_int(blob.xml, PCMK__XA_ST_CALLID, &call_id); update_callback_timeout(call_id, timeout, st); } else { crm_err("Unknown message type: %s", type); crm_log_xml_warn(blob.xml, "BadReply"); } pcmk__xml_free(blob.xml); return 1; } static int stonith_api_signon(stonith_t * stonith, const char *name, int *stonith_fd) { int rc = pcmk_ok; stonith_private_t *native = NULL; const char *display_name = name? name : "client"; struct ipc_client_callbacks st_callbacks = { .dispatch = stonith_dispatch_internal, .destroy = stonith_connection_destroy }; CRM_CHECK(stonith != NULL, return -EINVAL); native = stonith->st_private; CRM_ASSERT(native != NULL); crm_debug("Attempting fencer connection by %s with%s mainloop", display_name, (stonith_fd? "out" : "")); stonith->state = stonith_connected_command; if (stonith_fd) { /* No mainloop */ native->ipc = crm_ipc_new("stonith-ng", 0); if (native->ipc != NULL) { rc = pcmk__connect_generic_ipc(native->ipc); if (rc == pcmk_rc_ok) { rc = pcmk__ipc_fd(native->ipc, stonith_fd); if (rc != pcmk_rc_ok) { crm_debug("Couldn't get file descriptor for IPC: %s", pcmk_rc_str(rc)); } } if (rc != pcmk_rc_ok) { crm_ipc_close(native->ipc); crm_ipc_destroy(native->ipc); native->ipc = NULL; } } } else { /* With mainloop */ native->source = mainloop_add_ipc_client("stonith-ng", G_PRIORITY_MEDIUM, 0, stonith, &st_callbacks); native->ipc = mainloop_get_ipc_client(native->source); } if (native->ipc == NULL) { rc = -ENOTCONN; } else { xmlNode *reply = NULL; xmlNode *hello = pcmk__xe_create(NULL, PCMK__XE_STONITH_COMMAND); crm_xml_add(hello, PCMK__XA_T, PCMK__VALUE_STONITH_NG); crm_xml_add(hello, PCMK__XA_ST_OP, CRM_OP_REGISTER); crm_xml_add(hello, PCMK__XA_ST_CLIENTNAME, name); rc = crm_ipc_send(native->ipc, hello, crm_ipc_client_response, -1, &reply); if (rc < 0) { crm_debug("Couldn't register with the fencer: %s " QB_XS " rc=%d", pcmk_strerror(rc), rc); rc = -ECOMM; } else if (reply == NULL) { crm_debug("Couldn't register with the fencer: no reply"); rc = -EPROTO; } else { const char *msg_type = crm_element_value(reply, PCMK__XA_ST_OP); native->token = crm_element_value_copy(reply, PCMK__XA_ST_CLIENTID); if (!pcmk__str_eq(msg_type, CRM_OP_REGISTER, pcmk__str_none)) { crm_debug("Couldn't register with the fencer: invalid reply type '%s'", (msg_type? msg_type : "(missing)")); crm_log_xml_debug(reply, "Invalid fencer reply"); rc = -EPROTO; } else if (native->token == NULL) { crm_debug("Couldn't register with the fencer: no token in reply"); crm_log_xml_debug(reply, "Invalid fencer reply"); rc = -EPROTO; } else { crm_debug("Connection to fencer by %s succeeded (registration token: %s)", display_name, native->token); rc = pcmk_ok; } } pcmk__xml_free(reply); pcmk__xml_free(hello); } if (rc != pcmk_ok) { crm_debug("Connection attempt to fencer by %s failed: %s " QB_XS " rc=%d", display_name, pcmk_strerror(rc), rc); stonith->cmds->disconnect(stonith); } return rc; } static int stonith_set_notification(stonith_t * stonith, const char *callback, int enabled) { int rc = pcmk_ok; xmlNode *notify_msg = pcmk__xe_create(NULL, __func__); stonith_private_t *native = stonith->st_private; if (stonith->state != stonith_disconnected) { crm_xml_add(notify_msg, PCMK__XA_ST_OP, STONITH_OP_NOTIFY); if (enabled) { crm_xml_add(notify_msg, PCMK__XA_ST_NOTIFY_ACTIVATE, callback); } else { crm_xml_add(notify_msg, PCMK__XA_ST_NOTIFY_DEACTIVATE, callback); } rc = crm_ipc_send(native->ipc, notify_msg, crm_ipc_client_response, -1, NULL); if (rc < 0) { crm_perror(LOG_DEBUG, "Couldn't register for fencing notifications: %d", rc); rc = -ECOMM; } else { rc = pcmk_ok; } } pcmk__xml_free(notify_msg); return rc; } static int stonith_api_add_notification(stonith_t * stonith, const char *event, void (*callback) (stonith_t * stonith, stonith_event_t * e)) { GList *list_item = NULL; stonith_notify_client_t *new_client = NULL; stonith_private_t *private = NULL; private = stonith->st_private; crm_trace("Adding callback for %s events (%d)", event, g_list_length(private->notify_list)); new_client = pcmk__assert_alloc(1, sizeof(stonith_notify_client_t)); new_client->event = event; new_client->notify = callback; list_item = g_list_find_custom(private->notify_list, new_client, stonithlib_GCompareFunc); if (list_item != NULL) { crm_warn("Callback already present"); free(new_client); return -ENOTUNIQ; } else { private->notify_list = g_list_append(private->notify_list, new_client); stonith_set_notification(stonith, event, 1); crm_trace("Callback added (%d)", g_list_length(private->notify_list)); } return pcmk_ok; } static void del_notify_entry(gpointer data, gpointer user_data) { stonith_notify_client_t *entry = data; stonith_t * stonith = user_data; if (!entry->delete) { crm_debug("Removing callback for %s events", entry->event); stonith_api_del_notification(stonith, entry->event); } } static int stonith_api_del_notification(stonith_t * stonith, const char *event) { GList *list_item = NULL; stonith_notify_client_t *new_client = NULL; stonith_private_t *private = stonith->st_private; if (event == NULL) { foreach_notify_entry(private, del_notify_entry, stonith); crm_trace("Removed callback"); return pcmk_ok; } crm_debug("Removing callback for %s events", event); new_client = pcmk__assert_alloc(1, sizeof(stonith_notify_client_t)); new_client->event = event; new_client->notify = NULL; list_item = g_list_find_custom(private->notify_list, new_client, stonithlib_GCompareFunc); stonith_set_notification(stonith, event, 0); if (list_item != NULL) { stonith_notify_client_t *list_client = list_item->data; if (private->notify_refcnt) { list_client->delete = TRUE; private->notify_deletes = TRUE; } else { private->notify_list = g_list_remove(private->notify_list, list_client); free(list_client); } crm_trace("Removed callback"); } else { crm_trace("Callback not present"); } free(new_client); return pcmk_ok; } static int stonith_api_add_callback(stonith_t * stonith, int call_id, int timeout, int options, void *user_data, const char *callback_name, void (*callback) (stonith_t * st, stonith_callback_data_t * data)) { stonith_callback_client_t *blob = NULL; stonith_private_t *private = NULL; CRM_CHECK(stonith != NULL, return -EINVAL); CRM_CHECK(stonith->st_private != NULL, return -EINVAL); private = stonith->st_private; if (call_id == 0) { // Add global callback private->op_callback = callback; } else if (call_id < 0) { // Call failed immediately, so call callback now if (!(options & st_opt_report_only_success)) { pcmk__action_result_t result = PCMK__UNKNOWN_RESULT; crm_trace("Call failed, calling %s: %s", callback_name, pcmk_strerror(call_id)); pcmk__set_result(&result, CRM_EX_ERROR, stonith__legacy2status(call_id), NULL); invoke_fence_action_callback(stonith, call_id, &result, user_data, callback); } else { crm_warn("Fencer call failed: %s", pcmk_strerror(call_id)); } return FALSE; } blob = pcmk__assert_alloc(1, sizeof(stonith_callback_client_t)); blob->id = callback_name; blob->only_success = (options & st_opt_report_only_success) ? TRUE : FALSE; blob->user_data = user_data; blob->callback = callback; blob->allow_timeout_updates = (options & st_opt_timeout_updates) ? TRUE : FALSE; if (timeout > 0) { set_callback_timeout(blob, stonith, call_id, timeout); } pcmk__intkey_table_insert(private->stonith_op_callback_table, call_id, blob); crm_trace("Added callback to %s for call %d", callback_name, call_id); return TRUE; } static void stonith_dump_pending_op(gpointer key, gpointer value, gpointer user_data) { int call = GPOINTER_TO_INT(key); stonith_callback_client_t *blob = value; crm_debug("Call %d (%s): pending", call, pcmk__s(blob->id, "no ID")); } void stonith_dump_pending_callbacks(stonith_t * stonith) { stonith_private_t *private = stonith->st_private; if (private->stonith_op_callback_table == NULL) { return; } return g_hash_table_foreach(private->stonith_op_callback_table, stonith_dump_pending_op, NULL); } /*! * \internal * \brief Get the data section of a fencer notification * * \param[in] msg Notification XML * \param[in] ntype Notification type */ static xmlNode * get_event_data_xml(xmlNode *msg, const char *ntype) { char *data_addr = crm_strdup_printf("//%s", ntype); xmlNode *data = get_xpath_object(data_addr, msg, LOG_DEBUG); free(data_addr); return data; } /* */ static stonith_event_t * xml_to_event(xmlNode *msg) { stonith_event_t *event = pcmk__assert_alloc(1, sizeof(stonith_event_t)); struct event_private *event_private = NULL; event->opaque = pcmk__assert_alloc(1, sizeof(struct event_private)); event_private = (struct event_private *) event->opaque; crm_log_xml_trace(msg, "stonith_notify"); // All notification types have the operation result and notification subtype stonith__xe_get_result(msg, &event_private->result); event->operation = crm_element_value_copy(msg, PCMK__XA_ST_OP); // @COMPAT The API originally provided the result as a legacy return code event->result = pcmk_rc2legacy(stonith__result2rc(&event_private->result)); // Some notification subtypes have additional information if (pcmk__str_eq(event->operation, PCMK__VALUE_ST_NOTIFY_FENCE, pcmk__str_none)) { xmlNode *data = get_event_data_xml(msg, event->operation); if (data == NULL) { crm_err("No data for %s event", event->operation); crm_log_xml_notice(msg, "BadEvent"); } else { event->origin = crm_element_value_copy(data, PCMK__XA_ST_ORIGIN); event->action = crm_element_value_copy(data, PCMK__XA_ST_DEVICE_ACTION); event->target = crm_element_value_copy(data, PCMK__XA_ST_TARGET); event->executioner = crm_element_value_copy(data, PCMK__XA_ST_DELEGATE); event->id = crm_element_value_copy(data, PCMK__XA_ST_REMOTE_OP); event->client_origin = crm_element_value_copy(data, PCMK__XA_ST_CLIENTNAME); event->device = crm_element_value_copy(data, PCMK__XA_ST_DEVICE_ID); } } else if (pcmk__str_any_of(event->operation, STONITH_OP_DEVICE_ADD, STONITH_OP_DEVICE_DEL, STONITH_OP_LEVEL_ADD, STONITH_OP_LEVEL_DEL, NULL)) { xmlNode *data = get_event_data_xml(msg, event->operation); if (data == NULL) { crm_err("No data for %s event", event->operation); crm_log_xml_notice(msg, "BadEvent"); } else { event->device = crm_element_value_copy(data, PCMK__XA_ST_DEVICE_ID); } } return event; } static void event_free(stonith_event_t * event) { struct event_private *event_private = event->opaque; free(event->id); free(event->type); free(event->message); free(event->operation); free(event->origin); free(event->action); free(event->target); free(event->executioner); free(event->device); free(event->client_origin); pcmk__reset_result(&event_private->result); free(event->opaque); free(event); } static void stonith_send_notification(gpointer data, gpointer user_data) { struct notify_blob_s *blob = user_data; stonith_notify_client_t *entry = data; stonith_event_t *st_event = NULL; const char *event = NULL; if (blob->xml == NULL) { crm_warn("Skipping callback - NULL message"); return; } event = crm_element_value(blob->xml, PCMK__XA_SUBT); if (entry == NULL) { crm_warn("Skipping callback - NULL callback client"); return; } else if (entry->delete) { crm_trace("Skipping callback - marked for deletion"); return; } else if (entry->notify == NULL) { crm_warn("Skipping callback - NULL callback"); return; } else if (!pcmk__str_eq(entry->event, event, pcmk__str_none)) { crm_trace("Skipping callback - event mismatch %p/%s vs. %s", entry, entry->event, event); return; } st_event = xml_to_event(blob->xml); crm_trace("Invoking callback for %p/%s event...", entry, event); entry->notify(blob->stonith, st_event); crm_trace("Callback invoked..."); event_free(st_event); } /*! * \internal * \brief Create and send an API request * * \param[in,out] stonith Stonith connection * \param[in] op API operation to request * \param[in] data Data to attach to request * \param[out] output_data If not NULL, will be set to reply if synchronous * \param[in] call_options Bitmask of stonith_call_options to use * \param[in] timeout Error if not completed within this many seconds * * \return pcmk_ok (for synchronous requests) or positive call ID * (for asynchronous requests) on success, -errno otherwise */ static int stonith_send_command(stonith_t * stonith, const char *op, xmlNode * data, xmlNode ** output_data, int call_options, int timeout) { int rc = 0; int reply_id = -1; xmlNode *op_msg = NULL; xmlNode *op_reply = NULL; stonith_private_t *native = NULL; CRM_ASSERT(stonith && stonith->st_private && op); native = stonith->st_private; if (output_data != NULL) { *output_data = NULL; } if ((stonith->state == stonith_disconnected) || (native->token == NULL)) { return -ENOTCONN; } /* Increment the call ID, which must be positive to avoid conflicting with * error codes. This shouldn't be a problem unless the client mucked with * it or the counter wrapped around. */ stonith->call_id++; if (stonith->call_id < 1) { stonith->call_id = 1; } op_msg = stonith_create_op(stonith->call_id, native->token, op, data, call_options); if (op_msg == NULL) { return -EINVAL; } crm_xml_add_int(op_msg, PCMK__XA_ST_TIMEOUT, timeout); crm_trace("Sending %s message to fencer with timeout %ds", op, timeout); if (data) { const char *delay_s = crm_element_value(data, PCMK__XA_ST_DELAY); if (delay_s) { crm_xml_add(op_msg, PCMK__XA_ST_DELAY, delay_s); } } { enum crm_ipc_flags ipc_flags = crm_ipc_flags_none; if (call_options & st_opt_sync_call) { pcmk__set_ipc_flags(ipc_flags, "stonith command", crm_ipc_client_response); } rc = crm_ipc_send(native->ipc, op_msg, ipc_flags, 1000 * (timeout + 60), &op_reply); } pcmk__xml_free(op_msg); if (rc < 0) { crm_perror(LOG_ERR, "Couldn't perform %s operation (timeout=%ds): %d", op, timeout, rc); rc = -ECOMM; goto done; } crm_log_xml_trace(op_reply, "Reply"); if (!(call_options & st_opt_sync_call)) { crm_trace("Async call %d, returning", stonith->call_id); pcmk__xml_free(op_reply); return stonith->call_id; } crm_element_value_int(op_reply, PCMK__XA_ST_CALLID, &reply_id); if (reply_id == stonith->call_id) { pcmk__action_result_t result = PCMK__UNKNOWN_RESULT; crm_trace("Synchronous reply %d received", reply_id); stonith__xe_get_result(op_reply, &result); rc = pcmk_rc2legacy(stonith__result2rc(&result)); pcmk__reset_result(&result); if ((call_options & st_opt_discard_reply) || output_data == NULL) { crm_trace("Discarding reply"); } else { *output_data = op_reply; op_reply = NULL; /* Prevent subsequent free */ } } else if (reply_id <= 0) { crm_err("Received bad reply: No id set"); crm_log_xml_err(op_reply, "Bad reply"); pcmk__xml_free(op_reply); op_reply = NULL; rc = -ENOMSG; } else { crm_err("Received bad reply: %d (wanted %d)", reply_id, stonith->call_id); crm_log_xml_err(op_reply, "Old reply"); pcmk__xml_free(op_reply); op_reply = NULL; rc = -ENOMSG; } done: if (!crm_ipc_connected(native->ipc)) { crm_err("Fencer disconnected"); free(native->token); native->token = NULL; stonith->state = stonith_disconnected; } pcmk__xml_free(op_reply); return rc; } /* Not used with mainloop */ bool stonith_dispatch(stonith_t * st) { gboolean stay_connected = TRUE; stonith_private_t *private = NULL; CRM_ASSERT(st != NULL); private = st->st_private; while (crm_ipc_ready(private->ipc)) { if (crm_ipc_read(private->ipc) > 0) { const char *msg = crm_ipc_buffer(private->ipc); stonith_dispatch_internal(msg, strlen(msg), st); } if (!crm_ipc_connected(private->ipc)) { crm_err("Connection closed"); stay_connected = FALSE; } } return stay_connected; } static int stonith_api_free(stonith_t * stonith) { int rc = pcmk_ok; crm_trace("Destroying %p", stonith); if (stonith->state != stonith_disconnected) { crm_trace("Unregistering notifications and disconnecting %p first", stonith); stonith->cmds->remove_notification(stonith, NULL); rc = stonith->cmds->disconnect(stonith); } if (stonith->state == stonith_disconnected) { stonith_private_t *private = stonith->st_private; crm_trace("Removing %d callbacks", g_hash_table_size(private->stonith_op_callback_table)); g_hash_table_destroy(private->stonith_op_callback_table); crm_trace("Destroying %d notification clients", g_list_length(private->notify_list)); g_list_free_full(private->notify_list, free); free(stonith->st_private); free(stonith->cmds); free(stonith); } else { crm_err("Not free'ing active connection: %s (%d)", pcmk_strerror(rc), rc); } return rc; } void stonith_api_delete(stonith_t * stonith) { crm_trace("Destroying %p", stonith); if(stonith) { stonith->cmds->free(stonith); } } static int stonith_api_validate(stonith_t *st, int call_options, const char *rsc_id, const char *namespace_s, const char *agent, const stonith_key_value_t *params, int timeout_sec, char **output, char **error_output) { /* Validation should be done directly via the agent, so we can get it from * stonith_admin when the cluster is not running, which is important for * higher-level tools. */ int rc = pcmk_ok; /* Use a dummy node name in case the agent requires a target. We assume the * actual target doesn't matter for validation purposes (if in practice, * that is incorrect, we will need to allow the caller to pass the target). */ const char *target = "node1"; const char *host_arg = NULL; GHashTable *params_table = pcmk__strkey_table(free, free); // Convert parameter list to a hash table for (; params; params = params->next) { if (pcmk__str_eq(params->key, PCMK_STONITH_HOST_ARGUMENT, pcmk__str_none)) { host_arg = params->value; } if (!pcmk_stonith_param(params->key)) { pcmk__insert_dup(params_table, params->key, params->value); } } #if SUPPORT_CIBSECRETS rc = pcmk__substitute_secrets(rsc_id, params_table); if (rc != pcmk_rc_ok) { crm_warn("Could not replace secret parameters for validation of %s: %s", agent, pcmk_rc_str(rc)); // rc is standard return value, don't return it in this function } #endif if (output) { *output = NULL; } if (error_output) { *error_output = NULL; } if (timeout_sec <= 0) { timeout_sec = PCMK_DEFAULT_METADATA_TIMEOUT_MS; // Questionable } switch (stonith_get_namespace(agent, namespace_s)) { case st_namespace_rhcs: rc = stonith__rhcs_validate(st, call_options, target, agent, params_table, host_arg, timeout_sec, output, error_output); break; #if HAVE_STONITH_STONITH_H case st_namespace_lha: rc = stonith__lha_validate(st, call_options, target, agent, params_table, timeout_sec, output, error_output); break; #endif case st_namespace_invalid: errno = ENOENT; rc = -errno; if (error_output) { *error_output = crm_strdup_printf("Agent %s not found", agent); } else { crm_err("Agent %s not found", agent); } break; default: errno = EOPNOTSUPP; rc = -errno; if (error_output) { *error_output = crm_strdup_printf("Agent %s does not support validation", agent); } else { crm_err("Agent %s does not support validation", agent); } break; } g_hash_table_destroy(params_table); return rc; } stonith_t * stonith_api_new(void) { stonith_t *new_stonith = NULL; stonith_private_t *private = NULL; new_stonith = calloc(1, sizeof(stonith_t)); if (new_stonith == NULL) { return NULL; } private = calloc(1, sizeof(stonith_private_t)); if (private == NULL) { free(new_stonith); return NULL; } new_stonith->st_private = private; private->stonith_op_callback_table = pcmk__intkey_table(stonith_destroy_op_callback); private->notify_list = NULL; private->notify_refcnt = 0; private->notify_deletes = FALSE; new_stonith->call_id = 1; new_stonith->state = stonith_disconnected; new_stonith->cmds = calloc(1, sizeof(stonith_api_operations_t)); if (new_stonith->cmds == NULL) { free(new_stonith->st_private); free(new_stonith); return NULL; } /* *INDENT-OFF* */ new_stonith->cmds->free = stonith_api_free; new_stonith->cmds->connect = stonith_api_signon; new_stonith->cmds->disconnect = stonith_api_signoff; new_stonith->cmds->list = stonith_api_list; new_stonith->cmds->monitor = stonith_api_monitor; new_stonith->cmds->status = stonith_api_status; new_stonith->cmds->fence = stonith_api_fence; new_stonith->cmds->fence_with_delay = stonith_api_fence_with_delay; new_stonith->cmds->confirm = stonith_api_confirm; new_stonith->cmds->history = stonith_api_history; new_stonith->cmds->list_agents = stonith_api_device_list; new_stonith->cmds->metadata = stonith_api_device_metadata; new_stonith->cmds->query = stonith_api_query; new_stonith->cmds->remove_device = stonith_api_remove_device; new_stonith->cmds->register_device = stonith_api_register_device; new_stonith->cmds->remove_level = stonith_api_remove_level; new_stonith->cmds->remove_level_full = stonith_api_remove_level_full; new_stonith->cmds->register_level = stonith_api_register_level; new_stonith->cmds->register_level_full = stonith_api_register_level_full; new_stonith->cmds->remove_callback = stonith_api_del_callback; new_stonith->cmds->register_callback = stonith_api_add_callback; new_stonith->cmds->remove_notification = stonith_api_del_notification; new_stonith->cmds->register_notification = stonith_api_add_notification; new_stonith->cmds->validate = stonith_api_validate; /* *INDENT-ON* */ return new_stonith; } /*! * \brief Make a blocking connection attempt to the fencer * * \param[in,out] st Fencer API object * \param[in] name Client name to use with fencer * \param[in] max_attempts Return error if this many attempts fail * * \return pcmk_ok on success, result of last attempt otherwise */ int stonith_api_connect_retry(stonith_t *st, const char *name, int max_attempts) { int rc = -EINVAL; // if max_attempts is not positive for (int attempt = 1; attempt <= max_attempts; attempt++) { rc = st->cmds->connect(st, name, NULL); if (rc == pcmk_ok) { return pcmk_ok; } else if (attempt < max_attempts) { crm_notice("Fencer connection attempt %d of %d failed (retrying in 2s): %s " QB_XS " rc=%d", attempt, max_attempts, pcmk_strerror(rc), rc); sleep(2); } } crm_notice("Could not connect to fencer: %s " QB_XS " rc=%d", pcmk_strerror(rc), rc); return rc; } stonith_key_value_t * stonith_key_value_add(stonith_key_value_t * head, const char *key, const char *value) { stonith_key_value_t *p, *end; p = pcmk__assert_alloc(1, sizeof(stonith_key_value_t)); p->key = pcmk__str_copy(key); p->value = pcmk__str_copy(value); end = head; while (end && end->next) { end = end->next; } if (end) { end->next = p; } else { head = p; } return head; } void stonith_key_value_freeall(stonith_key_value_t * head, int keys, int values) { stonith_key_value_t *p; while (head) { p = head->next; if (keys) { free(head->key); } if (values) { free(head->value); } free(head); head = p; } } #define api_log_open() openlog("stonith-api", LOG_CONS | LOG_NDELAY | LOG_PID, LOG_DAEMON) #define api_log(level, fmt, args...) syslog(level, "%s: "fmt, __func__, args) int stonith_api_kick(uint32_t nodeid, const char *uname, int timeout, bool off) { int rc = pcmk_ok; stonith_t *st = stonith_api_new(); const char *action = off? PCMK_ACTION_OFF : PCMK_ACTION_REBOOT; api_log_open(); if (st == NULL) { api_log(LOG_ERR, "API initialization failed, could not kick (%s) node %u/%s", action, nodeid, uname); return -EPROTO; } rc = st->cmds->connect(st, "stonith-api", NULL); if (rc != pcmk_ok) { api_log(LOG_ERR, "Connection failed, could not kick (%s) node %u/%s : %s (%d)", action, nodeid, uname, pcmk_strerror(rc), rc); } else { char *name = (uname == NULL)? pcmk__itoa(nodeid) : strdup(uname); int opts = 0; stonith__set_call_options(opts, name, st_opt_sync_call|st_opt_allow_suicide); if ((uname == NULL) && (nodeid > 0)) { stonith__set_call_options(opts, name, st_opt_cs_nodeid); } rc = st->cmds->fence(st, opts, name, action, timeout, 0); free(name); if (rc != pcmk_ok) { api_log(LOG_ERR, "Could not kick (%s) node %u/%s : %s (%d)", action, nodeid, uname, pcmk_strerror(rc), rc); } else { api_log(LOG_NOTICE, "Node %u/%s kicked: %s", nodeid, uname, action); } } stonith_api_delete(st); return rc; } time_t stonith_api_time(uint32_t nodeid, const char *uname, bool in_progress) { int rc = pcmk_ok; time_t when = 0; stonith_t *st = stonith_api_new(); stonith_history_t *history = NULL, *hp = NULL; if (st == NULL) { api_log(LOG_ERR, "Could not retrieve fence history for %u/%s: " "API initialization failed", nodeid, uname); return when; } rc = st->cmds->connect(st, "stonith-api", NULL); if (rc != pcmk_ok) { api_log(LOG_NOTICE, "Connection failed: %s (%d)", pcmk_strerror(rc), rc); } else { int entries = 0; int progress = 0; int completed = 0; int opts = 0; char *name = (uname == NULL)? pcmk__itoa(nodeid) : strdup(uname); stonith__set_call_options(opts, name, st_opt_sync_call); if ((uname == NULL) && (nodeid > 0)) { stonith__set_call_options(opts, name, st_opt_cs_nodeid); } rc = st->cmds->history(st, opts, name, &history, 120); free(name); for (hp = history; hp; hp = hp->next) { entries++; if (in_progress) { progress++; if (hp->state != st_done && hp->state != st_failed) { when = time(NULL); } } else if (hp->state == st_done) { completed++; if (hp->completed > when) { when = hp->completed; } } } stonith_history_free(history); if(rc == pcmk_ok) { api_log(LOG_INFO, "Found %d entries for %u/%s: %d in progress, %d completed", entries, nodeid, uname, progress, completed); } else { api_log(LOG_ERR, "Could not retrieve fence history for %u/%s: %s (%d)", nodeid, uname, pcmk_strerror(rc), rc); } } stonith_api_delete(st); if(when) { api_log(LOG_INFO, "Node %u/%s last kicked at: %ld", nodeid, uname, (long int)when); } return when; } bool stonith_agent_exists(const char *agent, int timeout) { stonith_t *st = NULL; stonith_key_value_t *devices = NULL; stonith_key_value_t *dIter = NULL; bool rc = FALSE; if (agent == NULL) { return rc; } st = stonith_api_new(); if (st == NULL) { crm_err("Could not list fence agents: API memory allocation failed"); return FALSE; } st->cmds->list_agents(st, st_opt_sync_call, NULL, &devices, timeout == 0 ? 120 : timeout); for (dIter = devices; dIter != NULL; dIter = dIter->next) { if (pcmk__str_eq(dIter->value, agent, pcmk__str_none)) { rc = TRUE; break; } } stonith_key_value_freeall(devices, 1, 1); stonith_api_delete(st); return rc; } const char * stonith_action_str(const char *action) { if (action == NULL) { return "fencing"; } else if (strcmp(action, PCMK_ACTION_ON) == 0) { return "unfencing"; } else if (strcmp(action, PCMK_ACTION_OFF) == 0) { return "turning off"; } else { return action; } } /*! * \internal * \brief Parse a target name from one line of a target list string * * \param[in] line One line of a target list string * \param[in] len String length of line * \param[in,out] output List to add newly allocated target name to */ static void parse_list_line(const char *line, int len, GList **output) { size_t i = 0; size_t entry_start = 0; /* Skip complaints about additional parameters device doesn't understand * * @TODO Document or eliminate the implied restriction of target names */ if (strstr(line, "invalid") || strstr(line, "variable")) { crm_debug("Skipping list output line: %s", line); return; } // Process line content, character by character for (i = 0; i <= len; i++) { if (isspace(line[i]) || (line[i] == ',') || (line[i] == ';') || (line[i] == '\0')) { // We've found a separator (i.e. the end of an entry) int rc = 0; char *entry = NULL; if (i == entry_start) { // Skip leading and sequential separators entry_start = i + 1; continue; } entry = pcmk__assert_alloc(i - entry_start + 1, sizeof(char)); /* Read entry, stopping at first separator * * @TODO Document or eliminate these character restrictions */ rc = sscanf(line + entry_start, "%[a-zA-Z0-9_-.]", entry); if (rc != 1) { crm_warn("Could not parse list output entry: %s " QB_XS " entry_start=%d position=%d", line + entry_start, entry_start, i); free(entry); } else if (pcmk__strcase_any_of(entry, PCMK_ACTION_ON, PCMK_ACTION_OFF, NULL)) { /* Some agents print the target status in the list output, * though none are known now (the separate list-status command * is used for this, but it can also print "UNKNOWN"). To handle * this possibility, skip such entries. * * @TODO Document or eliminate the implied restriction of target * names. */ free(entry); } else { // We have a valid entry *output = g_list_append(*output, entry); } entry_start = i + 1; } } } /*! * \internal * \brief Parse a list of targets from a string * * \param[in] list_output Target list as a string * * \return List of target names * \note The target list string format is flexible, to allow for user-specified * lists such pcmk_host_list and the output of an agent's list action * (whether direct or via the API, which escapes newlines). There may be * multiple lines, separated by either a newline or an escaped newline * (backslash n). Each line may have one or more target names, separated * by any combination of whitespace, commas, and semi-colons. Lines * containing "invalid" or "variable" will be ignored entirely. Target * names "on" or "off" (case-insensitive) will be ignored. Target names * may contain only alphanumeric characters, underbars (_), dashes (-), * and dots (.) (if any other character occurs in the name, it and all * subsequent characters in the name will be ignored). * \note The caller is responsible for freeing the result with * g_list_free_full(result, free). */ GList * stonith__parse_targets(const char *target_spec) { GList *targets = NULL; if (target_spec != NULL) { size_t out_len = strlen(target_spec); size_t line_start = 0; // Starting index of line being processed for (size_t i = 0; i <= out_len; ++i) { if ((target_spec[i] == '\n') || (target_spec[i] == '\0') || ((target_spec[i] == '\\') && (target_spec[i + 1] == 'n'))) { // We've reached the end of one line of output int len = i - line_start; if (len > 0) { char *line = strndup(target_spec + line_start, len); line[len] = '\0'; // Because it might be a newline parse_list_line(line, len, &targets); free(line); } if (target_spec[i] == '\\') { ++i; // backslash-n takes up two positions } line_start = i + 1; } } } return targets; } /*! * \internal * \brief Check whether a fencing failure was followed by an equivalent success * * \param[in] event Fencing failure * \param[in] top_history Complete fencing history (must be sorted by * stonith__sort_history() beforehand) * * \return The name of the node that executed the fencing if a later successful * event exists, or NULL if no such event exists */ const char * stonith__later_succeeded(const stonith_history_t *event, const stonith_history_t *top_history) { const char *other = NULL; for (const stonith_history_t *prev_hp = top_history; prev_hp != NULL; prev_hp = prev_hp->next) { if (prev_hp == event) { break; } if ((prev_hp->state == st_done) && pcmk__str_eq(event->target, prev_hp->target, pcmk__str_casei) && pcmk__str_eq(event->action, prev_hp->action, pcmk__str_none) && ((event->completed < prev_hp->completed) || ((event->completed == prev_hp->completed) && (event->completed_nsec < prev_hp->completed_nsec)))) { if ((event->delegate == NULL) || pcmk__str_eq(event->delegate, prev_hp->delegate, pcmk__str_casei)) { // Prefer equivalent fencing by same executioner return prev_hp->delegate; } else if (other == NULL) { // Otherwise remember first successful executioner other = (prev_hp->delegate == NULL)? "some node" : prev_hp->delegate; } } } return other; } /*! * \internal * \brief Sort fencing history, pending first then by most recently completed * * \param[in,out] history List of stonith actions * * \return New head of sorted \p history */ stonith_history_t * stonith__sort_history(stonith_history_t *history) { stonith_history_t *new = NULL, *pending = NULL, *hp, *np, *tmp; for (hp = history; hp; ) { tmp = hp->next; if ((hp->state == st_done) || (hp->state == st_failed)) { /* sort into new */ if ((!new) || (hp->completed > new->completed) || ((hp->completed == new->completed) && (hp->completed_nsec > new->completed_nsec))) { hp->next = new; new = hp; } else { np = new; do { if ((!np->next) || (hp->completed > np->next->completed) || ((hp->completed == np->next->completed) && (hp->completed_nsec > np->next->completed_nsec))) { hp->next = np->next; np->next = hp; break; } np = np->next; } while (1); } } else { /* put into pending */ hp->next = pending; pending = hp; } hp = tmp; } /* pending actions don't have a completed-stamp so make them go front */ if (pending) { stonith_history_t *last_pending = pending; while (last_pending->next) { last_pending = last_pending->next; } last_pending->next = new; new = pending; } return new; } /*! * \brief Return string equivalent of an operation state value * * \param[in] state Fencing operation state value * * \return Human-friendly string equivalent of state */ const char * stonith_op_state_str(enum op_state state) { switch (state) { case st_query: return "querying"; case st_exec: return "executing"; case st_done: return "completed"; case st_duplicate: return "duplicate"; case st_failed: return "failed"; } return "unknown"; } stonith_history_t * stonith__first_matching_event(stonith_history_t *history, bool (*matching_fn)(stonith_history_t *, void *), void *user_data) { for (stonith_history_t *hp = history; hp; hp = hp->next) { if (matching_fn(hp, user_data)) { return hp; } } return NULL; } bool stonith__event_state_pending(stonith_history_t *history, void *user_data) { return history->state != st_failed && history->state != st_done; } bool stonith__event_state_eq(stonith_history_t *history, void *user_data) { return history->state == GPOINTER_TO_INT(user_data); } bool stonith__event_state_neq(stonith_history_t *history, void *user_data) { return history->state != GPOINTER_TO_INT(user_data); } void stonith__device_parameter_flags(uint32_t *device_flags, const char *device_name, xmlNode *metadata) { xmlXPathObjectPtr xpath = NULL; int max = 0; int lpc = 0; CRM_CHECK((device_flags != NULL) && (metadata != NULL), return); xpath = xpath_search(metadata, "//" PCMK_XE_PARAMETER); max = numXpathResults(xpath); if (max <= 0) { freeXpathObject(xpath); return; } for (lpc = 0; lpc < max; lpc++) { const char *parameter = NULL; xmlNode *match = getXpathResult(xpath, lpc); CRM_LOG_ASSERT(match != NULL); if (match == NULL) { continue; } parameter = crm_element_value(match, PCMK_XA_NAME); if (pcmk__str_eq(parameter, "plug", pcmk__str_casei)) { stonith__set_device_flags(*device_flags, device_name, st_device_supports_parameter_plug); } else if (pcmk__str_eq(parameter, "port", pcmk__str_casei)) { stonith__set_device_flags(*device_flags, device_name, st_device_supports_parameter_port); } } freeXpathObject(xpath); } /*! * \internal * \brief Retrieve fence agent meta-data asynchronously * * \param[in] agent Agent to execute * \param[in] timeout_sec Error if not complete within this time * \param[in] callback Function to call with result (this will always be * called, whether by this function directly or * later via the main loop, and on success the * metadata will be in its result argument's * action_stdout) * \param[in,out] user_data User data to pass to callback * * \return Standard Pacemaker return code * \note The caller must use a main loop. This function is not a * stonith_api_operations_t method because it does not need a stonith_t * object and does not go through the fencer, but executes the agent * directly. */ int stonith__metadata_async(const char *agent, int timeout_sec, void (*callback)(int pid, const pcmk__action_result_t *result, void *user_data), void *user_data) { switch (stonith_get_namespace(agent, NULL)) { case st_namespace_rhcs: { stonith_action_t *action = NULL; int rc = pcmk_ok; action = stonith__action_create(agent, PCMK_ACTION_METADATA, NULL, 0, timeout_sec, NULL, NULL, NULL); rc = stonith__execute_async(action, user_data, callback, NULL); if (rc != pcmk_ok) { callback(0, stonith__action_result(action), user_data); stonith__destroy_action(action); } return pcmk_legacy2rc(rc); } #if HAVE_STONITH_STONITH_H case st_namespace_lha: // LHA metadata is simply synthesized, so simulate async { pcmk__action_result_t result = { .exit_status = CRM_EX_OK, .execution_status = PCMK_EXEC_DONE, .exit_reason = NULL, .action_stdout = NULL, .action_stderr = NULL, }; stonith__lha_metadata(agent, timeout_sec, &result.action_stdout); callback(0, &result, user_data); pcmk__reset_result(&result); return pcmk_rc_ok; } #endif default: { pcmk__action_result_t result = { .exit_status = CRM_EX_NOSUCH, .execution_status = PCMK_EXEC_ERROR_HARD, .exit_reason = crm_strdup_printf("No such agent '%s'", agent), .action_stdout = NULL, .action_stderr = NULL, }; callback(0, &result, user_data); pcmk__reset_result(&result); return ENOENT; } } } /*! * \internal * \brief Return the exit status from an async action callback * * \param[in] data Callback data * * \return Exit status from callback data */ int stonith__exit_status(const stonith_callback_data_t *data) { if ((data == NULL) || (data->opaque == NULL)) { return CRM_EX_ERROR; } return ((pcmk__action_result_t *) data->opaque)->exit_status; } /*! * \internal * \brief Return the execution status from an async action callback * * \param[in] data Callback data * * \return Execution status from callback data */ int stonith__execution_status(const stonith_callback_data_t *data) { if ((data == NULL) || (data->opaque == NULL)) { return PCMK_EXEC_UNKNOWN; } return ((pcmk__action_result_t *) data->opaque)->execution_status; } /*! * \internal * \brief Return the exit reason from an async action callback * * \param[in] data Callback data * * \return Exit reason from callback data */ const char * stonith__exit_reason(const stonith_callback_data_t *data) { if ((data == NULL) || (data->opaque == NULL)) { return NULL; } return ((pcmk__action_result_t *) data->opaque)->exit_reason; } /*! * \internal * \brief Return the exit status from an event notification * * \param[in] event Event * * \return Exit status from event */ int stonith__event_exit_status(const stonith_event_t *event) { if ((event == NULL) || (event->opaque == NULL)) { return CRM_EX_ERROR; } else { struct event_private *event_private = event->opaque; return event_private->result.exit_status; } } /*! * \internal * \brief Return the execution status from an event notification * * \param[in] event Event * * \return Execution status from event */ int stonith__event_execution_status(const stonith_event_t *event) { if ((event == NULL) || (event->opaque == NULL)) { return PCMK_EXEC_UNKNOWN; } else { struct event_private *event_private = event->opaque; return event_private->result.execution_status; } } /*! * \internal * \brief Return the exit reason from an event notification * * \param[in] event Event * * \return Exit reason from event */ const char * stonith__event_exit_reason(const stonith_event_t *event) { if ((event == NULL) || (event->opaque == NULL)) { return NULL; } else { struct event_private *event_private = event->opaque; return event_private->result.exit_reason; } } /*! * \internal * \brief Return a human-friendly description of a fencing event * * \param[in] event Event to describe * * \return Newly allocated string with description of \p event * \note The caller is responsible for freeing the return value. * This function asserts on memory errors and never returns NULL. */ char * stonith__event_description(const stonith_event_t *event) { // Use somewhat readable defaults const char *origin = pcmk__s(event->client_origin, "a client"); const char *origin_node = pcmk__s(event->origin, "a node"); const char *executioner = pcmk__s(event->executioner, "the cluster"); const char *device = pcmk__s(event->device, "unknown"); const char *action = pcmk__s(event->action, event->operation); const char *target = pcmk__s(event->target, "no node"); const char *reason = stonith__event_exit_reason(event); const char *status; if (action == NULL) { action = "(unknown)"; } if (stonith__event_execution_status(event) != PCMK_EXEC_DONE) { status = pcmk_exec_status_str(stonith__event_execution_status(event)); } else if (stonith__event_exit_status(event) != CRM_EX_OK) { status = pcmk_exec_status_str(PCMK_EXEC_ERROR); } else { status = crm_exit_str(CRM_EX_OK); } if (pcmk__str_eq(event->operation, PCMK__VALUE_ST_NOTIFY_HISTORY, pcmk__str_none)) { return crm_strdup_printf("Fencing history may have changed"); } else if (pcmk__str_eq(event->operation, STONITH_OP_DEVICE_ADD, pcmk__str_none)) { return crm_strdup_printf("A fencing device (%s) was added", device); } else if (pcmk__str_eq(event->operation, STONITH_OP_DEVICE_DEL, pcmk__str_none)) { return crm_strdup_printf("A fencing device (%s) was removed", device); } else if (pcmk__str_eq(event->operation, STONITH_OP_LEVEL_ADD, pcmk__str_none)) { return crm_strdup_printf("A fencing topology level (%s) was added", device); } else if (pcmk__str_eq(event->operation, STONITH_OP_LEVEL_DEL, pcmk__str_none)) { return crm_strdup_printf("A fencing topology level (%s) was removed", device); } // event->operation should be PCMK__VALUE_ST_NOTIFY_FENCE at this point return crm_strdup_printf("Operation %s of %s by %s for %s@%s: %s%s%s%s (ref=%s)", action, target, executioner, origin, origin_node, status, ((reason == NULL)? "" : " ("), pcmk__s(reason, ""), ((reason == NULL)? "" : ")"), pcmk__s(event->id, "(none)")); } - - -// Deprecated functions kept only for backward API compatibility -// LCOV_EXCL_START - -const char *get_stonith_provider(const char *agent, const char *provider); - -const char * -get_stonith_provider(const char *agent, const char *provider) -{ - return stonith_namespace2text(stonith_get_namespace(agent, provider)); -} - -// LCOV_EXCL_STOP -// End deprecated API diff --git a/lib/pengine/native.c b/lib/pengine/native.c index 8afd67f789..08110a5e42 100644 --- a/lib/pengine/native.c +++ b/lib/pengine/native.c @@ -1,1184 +1,1180 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #include /*! * \internal * \brief Check whether a resource is active on multiple nodes */ static bool is_multiply_active(const pcmk_resource_t *rsc) { unsigned int count = 0; if (pcmk__is_primitive(rsc)) { pe__find_active_requires(rsc, &count); } return count > 1; } static void native_priority_to_node(pcmk_resource_t *rsc, pcmk_node_t *node, gboolean failed) { int priority = 0; const bool promoted = (rsc->private->orig_role == pcmk_role_promoted); if ((rsc->private->priority == 0) || failed) { return; } if (promoted) { // Promoted instance takes base priority + 1 priority = rsc->private->priority + 1; } else { priority = rsc->private->priority; } node->private->priority += priority; pcmk__rsc_trace(rsc, "%s now has priority %d with %s'%s' (priority: %d%s)", pcmk__node_name(node), node->private->priority, (promoted? "promoted " : ""), rsc->id, rsc->private->priority, (promoted? " + 1" : "")); /* Priority of a resource running on a guest node is added to the cluster * node as well. */ if ((node->private->remote != NULL) && (node->private->remote->private->launcher != NULL)) { const pcmk_resource_t *launcher = NULL; launcher = node->private->remote->private->launcher; for (GList *gIter = launcher->private->active_nodes; gIter != NULL; gIter = gIter->next) { pcmk_node_t *a_node = gIter->data; a_node->private->priority += priority; pcmk__rsc_trace(rsc, "%s now has priority %d with %s'%s' " "(priority: %d%s) from guest node %s", pcmk__node_name(a_node), a_node->private->priority, (promoted? "promoted " : ""), rsc->id, rsc->private->priority, (promoted? " + 1" : ""), pcmk__node_name(node)); } } } void native_add_running(pcmk_resource_t *rsc, pcmk_node_t *node, pcmk_scheduler_t *scheduler, gboolean failed) { pcmk_resource_t *parent = rsc->private->parent; CRM_CHECK(node != NULL, return); for (GList *gIter = rsc->private->active_nodes; gIter != NULL; gIter = gIter->next) { pcmk_node_t *a_node = (pcmk_node_t *) gIter->data; if (pcmk__same_node(a_node, node)) { return; } } pcmk__rsc_trace(rsc, "Adding %s to %s %s", rsc->id, pcmk__node_name(node), pcmk_is_set(rsc->flags, pcmk__rsc_managed)? "" : "(unmanaged)"); rsc->private->active_nodes = g_list_append(rsc->private->active_nodes, node); if (pcmk__is_primitive(rsc)) { node->details->running_rsc = g_list_append(node->details->running_rsc, rsc); native_priority_to_node(rsc, node, failed); if (node->details->maintenance) { pcmk__clear_rsc_flags(rsc, pcmk__rsc_managed); pcmk__set_rsc_flags(rsc, pcmk__rsc_maintenance); } } if (!pcmk_is_set(rsc->flags, pcmk__rsc_managed)) { pcmk_resource_t *p = parent; pcmk__rsc_info(rsc, "resource %s isn't managed", rsc->id); resource_location(rsc, node, PCMK_SCORE_INFINITY, "not_managed_default", scheduler); while(p && node->details->online) { /* add without the additional location constraint */ p->private->active_nodes = g_list_append(p->private->active_nodes, node); p = p->private->parent; } return; } if (is_multiply_active(rsc)) { switch (rsc->private->multiply_active_policy) { case pcmk__multiply_active_stop: { GHashTableIter gIter; pcmk_node_t *local_node = NULL; /* make sure it doesn't come up again */ if (rsc->private->allowed_nodes != NULL) { g_hash_table_destroy(rsc->private->allowed_nodes); } rsc->private->allowed_nodes = pe__node_list2table(scheduler->nodes); g_hash_table_iter_init(&gIter, rsc->private->allowed_nodes); while (g_hash_table_iter_next(&gIter, NULL, (void **)&local_node)) { local_node->assign->score = -PCMK_SCORE_INFINITY; } } break; case pcmk__multiply_active_block: pcmk__clear_rsc_flags(rsc, pcmk__rsc_managed); pcmk__set_rsc_flags(rsc, pcmk__rsc_blocked); /* If the resource belongs to a group or bundle configured with * PCMK_META_MULTIPLE_ACTIVE=PCMK_VALUE_BLOCK, block the entire * entity. */ if ((pcmk__is_group(parent) || pcmk__is_bundle(parent)) && (parent->private->multiply_active_policy == pcmk__multiply_active_block)) { for (GList *gIter = parent->private->children; gIter != NULL; gIter = gIter->next) { pcmk_resource_t *child = gIter->data; pcmk__clear_rsc_flags(child, pcmk__rsc_managed); pcmk__set_rsc_flags(child, pcmk__rsc_blocked); } } break; // pcmk__multiply_active_restart, pcmk__multiply_active_unexpected default: /* The scheduler will do the right thing because the relevant * variables and flags are set when unpacking the history. */ break; } crm_debug("%s is active on multiple nodes including %s: %s", rsc->id, pcmk__node_name(node), pcmk__multiply_active_text(rsc)); } else { pcmk__rsc_trace(rsc, "Resource %s is active on %s", rsc->id, pcmk__node_name(node)); } if (parent != NULL) { native_add_running(parent, node, scheduler, FALSE); } } static void recursive_clear_unique(pcmk_resource_t *rsc, gpointer user_data) { pcmk__clear_rsc_flags(rsc, pcmk__rsc_unique); pcmk__insert_meta(rsc->private, PCMK_META_GLOBALLY_UNIQUE, PCMK_VALUE_FALSE); g_list_foreach(rsc->private->children, (GFunc) recursive_clear_unique, NULL); } gboolean native_unpack(pcmk_resource_t *rsc, pcmk_scheduler_t *scheduler) { pcmk_resource_t *parent = uber_parent(rsc); const char *standard = crm_element_value(rsc->private->xml, PCMK_XA_CLASS); uint32_t ra_caps = pcmk_get_ra_caps(standard); pcmk__rsc_trace(rsc, "Processing resource %s...", rsc->id); // Only some agent standards support unique and promotable clones if (!pcmk_is_set(ra_caps, pcmk_ra_cap_unique) && pcmk_is_set(rsc->flags, pcmk__rsc_unique) && pcmk__is_clone(parent)) { /* @COMPAT We should probably reject this situation as an error (as we * do for promotable below) rather than warn and convert, but that would * be a backward-incompatible change that we should probably do with a * transform at a schema major version bump. */ pe__force_anon(standard, parent, rsc->id, scheduler); /* Clear PCMK_META_GLOBALLY_UNIQUE on the parent and all its descendants * unpacked so far (clearing the parent should make any future children * unpacking correct). We have to clear this resource explicitly because * it isn't hooked into the parent's children yet. */ recursive_clear_unique(parent, NULL); recursive_clear_unique(rsc, NULL); } if (!pcmk_is_set(ra_caps, pcmk_ra_cap_promotable) && pcmk_is_set(parent->flags, pcmk__rsc_promotable)) { pcmk__config_err("Resource %s is of type %s and therefore " "cannot be used as a promotable clone resource", rsc->id, standard); return FALSE; } return TRUE; } static bool rsc_is_on_node(pcmk_resource_t *rsc, const pcmk_node_t *node, int flags) { pcmk__rsc_trace(rsc, "Checking whether %s is on %s", rsc->id, pcmk__node_name(node)); if (pcmk_is_set(flags, pcmk_rsc_match_current_node) && (rsc->private->active_nodes != NULL)) { for (GList *iter = rsc->private->active_nodes; iter != NULL; iter = iter->next) { if (pcmk__same_node((pcmk_node_t *) iter->data, node)) { return true; } } - } else if (pcmk_is_set(flags, pe_find_inactive) // @COMPAT deprecated - && (rsc->private->active_nodes == NULL)) { - return true; - } else if (!pcmk_is_set(flags, pcmk_rsc_match_current_node) && (rsc->private->assigned_node != NULL) && pcmk__same_node(rsc->private->assigned_node, node)) { return true; } return false; } pcmk_resource_t * native_find_rsc(pcmk_resource_t *rsc, const char *id, const pcmk_node_t *on_node, int flags) { bool match = false; pcmk_resource_t *result = NULL; CRM_CHECK(id && rsc && rsc->id, return NULL); if (pcmk_is_set(flags, pcmk_rsc_match_clone_only)) { const char *rid = pcmk__xe_id(rsc->private->xml); if (!pcmk__is_clone(pe__const_top_resource(rsc, false))) { match = false; } else if (!strcmp(id, rsc->id) || pcmk__str_eq(id, rid, pcmk__str_none)) { match = true; } } else if (!strcmp(id, rsc->id)) { match = true; } else if (pcmk_is_set(flags, pcmk_rsc_match_history) && pcmk__str_eq(rsc->private->history_id, id, pcmk__str_none)) { match = true; } else if (pcmk_is_set(flags, pcmk_rsc_match_basename) || (pcmk_is_set(flags, pcmk_rsc_match_anon_basename) && !pcmk_is_set(rsc->flags, pcmk__rsc_unique))) { match = pe_base_name_eq(rsc, id); } if (match && on_node) { if (!rsc_is_on_node(rsc, on_node, flags)) { match = false; } } if (match) { return rsc; } for (GList *gIter = rsc->private->children; gIter != NULL; gIter = gIter->next) { pcmk_resource_t *child = (pcmk_resource_t *) gIter->data; result = rsc->private->fns->find_rsc(child, id, on_node, flags); if (result) { return result; } } return NULL; } // create is ignored char * native_parameter(pcmk_resource_t *rsc, pcmk_node_t *node, gboolean create, const char *name, pcmk_scheduler_t *scheduler) { const char *value = NULL; GHashTable *params = NULL; CRM_CHECK(rsc != NULL, return NULL); CRM_CHECK(name != NULL && strlen(name) != 0, return NULL); pcmk__rsc_trace(rsc, "Looking up %s in %s", name, rsc->id); params = pe_rsc_params(rsc, node, scheduler); value = g_hash_table_lookup(params, name); if (value == NULL) { /* try meta attributes instead */ value = g_hash_table_lookup(rsc->private->meta, name); } return pcmk__str_copy(value); } gboolean native_active(pcmk_resource_t * rsc, gboolean all) { for (GList *gIter = rsc->private->active_nodes; gIter != NULL; gIter = gIter->next) { pcmk_node_t *a_node = (pcmk_node_t *) gIter->data; if (a_node->details->unclean) { pcmk__rsc_trace(rsc, "Resource %s: %s is unclean", rsc->id, pcmk__node_name(a_node)); return TRUE; } else if (!a_node->details->online && pcmk_is_set(rsc->flags, pcmk__rsc_managed)) { pcmk__rsc_trace(rsc, "Resource %s: %s is offline", rsc->id, pcmk__node_name(a_node)); } else { pcmk__rsc_trace(rsc, "Resource %s active on %s", rsc->id, pcmk__node_name(a_node)); return TRUE; } } return FALSE; } struct print_data_s { long options; void *print_data; }; static const char * native_pending_state(const pcmk_resource_t *rsc) { const char *pending_state = NULL; if (pcmk__str_eq(rsc->private->pending_action, PCMK_ACTION_START, pcmk__str_none)) { pending_state = "Starting"; } else if (pcmk__str_eq(rsc->private->pending_action, PCMK_ACTION_STOP, pcmk__str_none)) { pending_state = "Stopping"; } else if (pcmk__str_eq(rsc->private->pending_action, PCMK_ACTION_MIGRATE_TO, pcmk__str_none)) { pending_state = "Migrating"; } else if (pcmk__str_eq(rsc->private->pending_action, PCMK_ACTION_MIGRATE_FROM, pcmk__str_none)) { /* Work might be done in here. */ pending_state = "Migrating"; } else if (pcmk__str_eq(rsc->private->pending_action, PCMK_ACTION_PROMOTE, pcmk__str_none)) { pending_state = "Promoting"; } else if (pcmk__str_eq(rsc->private->pending_action, PCMK_ACTION_DEMOTE, pcmk__str_none)) { pending_state = "Demoting"; } return pending_state; } static const char * native_pending_action(const pcmk_resource_t *rsc) { const char *pending_action = NULL; if (pcmk__str_eq(rsc->private->pending_action, PCMK_ACTION_MONITOR, pcmk__str_none)) { pending_action = "Monitoring"; /* Pending probes are not printed, even if pending * operations are requested. If someone ever requests that * behavior, uncomment this and the corresponding part of * unpack.c:unpack_rsc_op(). */ #if 0 } else if (pcmk__str_eq(rsc->private->pending_action, "probe", pcmk__str_none)) { pending_action = "Checking"; #endif } return pending_action; } static enum rsc_role_e native_displayable_role(const pcmk_resource_t *rsc) { enum rsc_role_e role = rsc->private->orig_role; if ((role == pcmk_role_started) && pcmk_is_set(pe__const_top_resource(rsc, false)->flags, pcmk__rsc_promotable)) { role = pcmk_role_unpromoted; } return role; } static const char * native_displayable_state(const pcmk_resource_t *rsc, bool print_pending) { const char *rsc_state = NULL; if (print_pending) { rsc_state = native_pending_state(rsc); } if (rsc_state == NULL) { rsc_state = pcmk_role_text(native_displayable_role(rsc)); } return rsc_state; } // Append a flag to resource description string's flags list static bool add_output_flag(GString *s, const char *flag_desc, bool have_flags) { g_string_append(s, (have_flags? ", " : " (")); g_string_append(s, flag_desc); return true; } // Append a node name to resource description string's node list static bool add_output_node(GString *s, const char *node, bool have_nodes) { g_string_append(s, (have_nodes? " " : " [ ")); g_string_append(s, node); return true; } /*! * \internal * \brief Create a string description of a resource * * \param[in] rsc Resource to describe * \param[in] name Desired identifier for the resource * \param[in] node If not NULL, node that resource is "on" * \param[in] show_opts Bitmask of pcmk_show_opt_e. * \param[in] target_role Resource's target role * \param[in] show_nodes Whether to display nodes when multiply active * * \return Newly allocated string description of resource * \note Caller must free the result with g_free(). */ gchar * pcmk__native_output_string(const pcmk_resource_t *rsc, const char *name, const pcmk_node_t *node, uint32_t show_opts, const char *target_role, bool show_nodes) { const char *class = crm_element_value(rsc->private->xml, PCMK_XA_CLASS); const char *provider = NULL; const char *kind = crm_element_value(rsc->private->xml, PCMK_XA_TYPE); GString *outstr = NULL; bool have_flags = false; if (!pcmk__is_primitive(rsc)) { return NULL; } CRM_CHECK(name != NULL, name = "unknown"); CRM_CHECK(kind != NULL, kind = "unknown"); CRM_CHECK(class != NULL, class = "unknown"); if (pcmk_is_set(pcmk_get_ra_caps(class), pcmk_ra_cap_provider)) { provider = crm_element_value(rsc->private->xml, PCMK_XA_PROVIDER); } if ((node == NULL) && (rsc->private->lock_node != NULL)) { node = rsc->private->lock_node; } if (pcmk_any_flags_set(show_opts, pcmk_show_rsc_only) || pcmk__list_of_multiple(rsc->private->active_nodes)) { node = NULL; } outstr = g_string_sized_new(128); // Resource name and agent pcmk__g_strcat(outstr, name, "\t(", class, ((provider == NULL)? "" : ":"), pcmk__s(provider, ""), ":", kind, "):\t", NULL); // State on node if (pcmk_is_set(rsc->flags, pcmk__rsc_removed)) { g_string_append(outstr, " ORPHANED"); } if (pcmk_is_set(rsc->flags, pcmk__rsc_failed)) { enum rsc_role_e role = native_displayable_role(rsc); g_string_append(outstr, " FAILED"); if (role > pcmk_role_unpromoted) { pcmk__add_word(&outstr, 0, pcmk_role_text(role)); } } else { bool show_pending = pcmk_is_set(show_opts, pcmk_show_pending); pcmk__add_word(&outstr, 0, native_displayable_state(rsc, show_pending)); } if (node) { pcmk__add_word(&outstr, 0, pcmk__node_name(node)); } // Failed probe operation if (native_displayable_role(rsc) == pcmk_role_stopped) { xmlNode *probe_op = pe__failed_probe_for_rsc(rsc, node ? node->private->name : NULL); if (probe_op != NULL) { int rc; pcmk__scan_min_int(crm_element_value(probe_op, PCMK__XA_RC_CODE), &rc, 0); pcmk__g_strcat(outstr, " (", services_ocf_exitcode_str(rc), ") ", NULL); } } // Flags, as: ( [...]) if (node && !(node->details->online) && node->details->unclean) { have_flags = add_output_flag(outstr, "UNCLEAN", have_flags); } if ((node != NULL) && pcmk__same_node(node, rsc->private->lock_node)) { have_flags = add_output_flag(outstr, "LOCKED", have_flags); } if (pcmk_is_set(show_opts, pcmk_show_pending)) { const char *pending_action = native_pending_action(rsc); if (pending_action != NULL) { have_flags = add_output_flag(outstr, pending_action, have_flags); } } if (target_role != NULL) { switch (pcmk_parse_role(target_role)) { case pcmk_role_unknown: pcmk__config_err("Invalid " PCMK_META_TARGET_ROLE " %s for resource %s", target_role, rsc->id); break; case pcmk_role_stopped: have_flags = add_output_flag(outstr, "disabled", have_flags); break; case pcmk_role_unpromoted: if (pcmk_is_set(pe__const_top_resource(rsc, false)->flags, pcmk__rsc_promotable)) { have_flags = add_output_flag(outstr, PCMK_META_TARGET_ROLE ":", have_flags); g_string_append(outstr, target_role); } break; default: /* Only show target role if it limits our abilities (i.e. ignore * Started, as it is the default anyways, and doesn't prevent * the resource from becoming promoted). */ break; } } // Blocked or maintenance implies unmanaged if (pcmk_any_flags_set(rsc->flags, pcmk__rsc_blocked|pcmk__rsc_maintenance)) { if (pcmk_is_set(rsc->flags, pcmk__rsc_blocked)) { have_flags = add_output_flag(outstr, "blocked", have_flags); } else if (pcmk_is_set(rsc->flags, pcmk__rsc_maintenance)) { have_flags = add_output_flag(outstr, "maintenance", have_flags); } } else if (!pcmk_is_set(rsc->flags, pcmk__rsc_managed)) { have_flags = add_output_flag(outstr, "unmanaged", have_flags); } if (pcmk_is_set(rsc->flags, pcmk__rsc_ignore_failure)) { have_flags = add_output_flag(outstr, "failure ignored", have_flags); } if (have_flags) { g_string_append_c(outstr, ')'); } // User-supplied description if (pcmk_any_flags_set(show_opts, pcmk_show_rsc_only|pcmk_show_description) || pcmk__list_of_multiple(rsc->private->active_nodes)) { const char *desc = crm_element_value(rsc->private->xml, PCMK_XA_DESCRIPTION); if (desc) { g_string_append(outstr, " ("); g_string_append(outstr, desc); g_string_append(outstr, ")"); } } if (show_nodes && !pcmk_is_set(show_opts, pcmk_show_rsc_only) && pcmk__list_of_multiple(rsc->private->active_nodes)) { bool have_nodes = false; for (GList *iter = rsc->private->active_nodes; iter != NULL; iter = iter->next) { pcmk_node_t *n = (pcmk_node_t *) iter->data; have_nodes = add_output_node(outstr, n->private->name, have_nodes); } if (have_nodes) { g_string_append(outstr, " ]"); } } return g_string_free(outstr, FALSE); } int pe__common_output_html(pcmk__output_t *out, const pcmk_resource_t *rsc, const char *name, const pcmk_node_t *node, uint32_t show_opts) { const char *kind = crm_element_value(rsc->private->xml, PCMK_XA_TYPE); const char *target_role = NULL; const char *cl = NULL; xmlNode *child = NULL; gchar *content = NULL; CRM_ASSERT((kind != NULL) && pcmk__is_primitive(rsc)); if (crm_is_true(g_hash_table_lookup(rsc->private->meta, PCMK__META_INTERNAL_RSC)) && !pcmk_is_set(show_opts, pcmk_show_implicit_rscs)) { crm_trace("skipping print of internal resource %s", rsc->id); return pcmk_rc_no_output; } target_role = g_hash_table_lookup(rsc->private->meta, PCMK_META_TARGET_ROLE); if (!pcmk_is_set(rsc->flags, pcmk__rsc_managed)) { cl = PCMK__VALUE_RSC_MANAGED; } else if (pcmk_is_set(rsc->flags, pcmk__rsc_failed)) { cl = PCMK__VALUE_RSC_FAILED; } else if (pcmk__is_primitive(rsc) && (rsc->private->active_nodes == NULL)) { cl = PCMK__VALUE_RSC_FAILED; } else if (pcmk__list_of_multiple(rsc->private->active_nodes)) { cl = PCMK__VALUE_RSC_MULTIPLE; } else if (pcmk_is_set(rsc->flags, pcmk__rsc_ignore_failure)) { cl = PCMK__VALUE_RSC_FAILURE_IGNORED; } else { cl = PCMK__VALUE_RSC_OK; } child = pcmk__output_create_html_node(out, "li", NULL, NULL, NULL); child = pcmk__html_create(child, PCMK__XE_SPAN, NULL, cl); content = pcmk__native_output_string(rsc, name, node, show_opts, target_role, true); pcmk__xe_set_content(child, "%s", content); g_free(content); return pcmk_rc_ok; } int pe__common_output_text(pcmk__output_t *out, const pcmk_resource_t *rsc, const char *name, const pcmk_node_t *node, uint32_t show_opts) { const char *target_role = NULL; CRM_ASSERT(pcmk__is_primitive(rsc)); if (crm_is_true(g_hash_table_lookup(rsc->private->meta, PCMK__META_INTERNAL_RSC)) && !pcmk_is_set(show_opts, pcmk_show_implicit_rscs)) { crm_trace("skipping print of internal resource %s", rsc->id); return pcmk_rc_no_output; } target_role = g_hash_table_lookup(rsc->private->meta, PCMK_META_TARGET_ROLE); { gchar *s = pcmk__native_output_string(rsc, name, node, show_opts, target_role, true); out->list_item(out, NULL, "%s", s); g_free(s); } return pcmk_rc_ok; } PCMK__OUTPUT_ARGS("primitive", "uint32_t", "pcmk_resource_t *", "GList *", "GList *") int pe__resource_xml(pcmk__output_t *out, va_list args) { uint32_t show_opts = va_arg(args, uint32_t); pcmk_resource_t *rsc = va_arg(args, pcmk_resource_t *); GList *only_node G_GNUC_UNUSED = va_arg(args, GList *); GList *only_rsc = va_arg(args, GList *); int rc = pcmk_rc_no_output; bool print_pending = pcmk_is_set(show_opts, pcmk_show_pending); const char *class = crm_element_value(rsc->private->xml, PCMK_XA_CLASS); const char *prov = crm_element_value(rsc->private->xml, PCMK_XA_PROVIDER); char ra_name[LINE_MAX]; const char *rsc_state = native_displayable_state(rsc, print_pending); const char *target_role = NULL; const char *active = pcmk__btoa(rsc->private->fns->active(rsc, TRUE)); const char *orphaned = pcmk__flag_text(rsc->flags, pcmk__rsc_removed); const char *blocked = pcmk__flag_text(rsc->flags, pcmk__rsc_blocked); const char *maintenance = pcmk__flag_text(rsc->flags, pcmk__rsc_maintenance); const char *managed = pcmk__flag_text(rsc->flags, pcmk__rsc_managed); const char *failed = pcmk__flag_text(rsc->flags, pcmk__rsc_failed); const char *ignored = pcmk__flag_text(rsc->flags, pcmk__rsc_ignore_failure); char *nodes_running_on = NULL; const char *pending = print_pending? native_pending_action(rsc) : NULL; const char *locked_to = NULL; const char *desc = pe__resource_description(rsc, show_opts); CRM_ASSERT(pcmk__is_primitive(rsc)); if (rsc->private->fns->is_filtered(rsc, only_rsc, TRUE)) { return pcmk_rc_no_output; } // Resource information snprintf(ra_name, LINE_MAX, "%s%s%s:%s", class, ((prov == NULL)? "" : ":"), ((prov == NULL)? "" : prov), crm_element_value(rsc->private->xml, PCMK_XA_TYPE)); target_role = g_hash_table_lookup(rsc->private->meta, PCMK_META_TARGET_ROLE); nodes_running_on = pcmk__itoa(g_list_length(rsc->private->active_nodes)); if (rsc->private->lock_node != NULL) { locked_to = rsc->private->lock_node->private->name; } rc = pe__name_and_nvpairs_xml(out, true, PCMK_XE_RESOURCE, PCMK_XA_ID, rsc_printable_id(rsc), PCMK_XA_RESOURCE_AGENT, ra_name, PCMK_XA_ROLE, rsc_state, PCMK_XA_TARGET_ROLE, target_role, PCMK_XA_ACTIVE, active, PCMK_XA_ORPHANED, orphaned, PCMK_XA_BLOCKED, blocked, PCMK_XA_MAINTENANCE, maintenance, PCMK_XA_MANAGED, managed, PCMK_XA_FAILED, failed, PCMK_XA_FAILURE_IGNORED, ignored, PCMK_XA_NODES_RUNNING_ON, nodes_running_on, PCMK_XA_PENDING, pending, PCMK_XA_LOCKED_TO, locked_to, PCMK_XA_DESCRIPTION, desc, NULL); free(nodes_running_on); CRM_ASSERT(rc == pcmk_rc_ok); for (GList *gIter = rsc->private->active_nodes; gIter != NULL; gIter = gIter->next) { pcmk_node_t *node = (pcmk_node_t *) gIter->data; const char *cached = pcmk__btoa(node->details->online); rc = pe__name_and_nvpairs_xml(out, false, PCMK_XE_NODE, PCMK_XA_NAME, node->private->name, PCMK_XA_ID, node->private->id, PCMK_XA_CACHED, cached, NULL); CRM_ASSERT(rc == pcmk_rc_ok); } pcmk__output_xml_pop_parent(out); return rc; } PCMK__OUTPUT_ARGS("primitive", "uint32_t", "pcmk_resource_t *", "GList *", "GList *") int pe__resource_html(pcmk__output_t *out, va_list args) { uint32_t show_opts = va_arg(args, uint32_t); pcmk_resource_t *rsc = va_arg(args, pcmk_resource_t *); GList *only_node G_GNUC_UNUSED = va_arg(args, GList *); GList *only_rsc = va_arg(args, GList *); const pcmk_node_t *node = pcmk__current_node(rsc); if (rsc->private->fns->is_filtered(rsc, only_rsc, TRUE)) { return pcmk_rc_no_output; } CRM_ASSERT(pcmk__is_primitive(rsc)); if (node == NULL) { // This is set only if a non-probe action is pending on this node node = rsc->private->pending_node; } return pe__common_output_html(out, rsc, rsc_printable_id(rsc), node, show_opts); } PCMK__OUTPUT_ARGS("primitive", "uint32_t", "pcmk_resource_t *", "GList *", "GList *") int pe__resource_text(pcmk__output_t *out, va_list args) { uint32_t show_opts = va_arg(args, uint32_t); pcmk_resource_t *rsc = va_arg(args, pcmk_resource_t *); GList *only_node G_GNUC_UNUSED = va_arg(args, GList *); GList *only_rsc = va_arg(args, GList *); const pcmk_node_t *node = pcmk__current_node(rsc); CRM_ASSERT(pcmk__is_primitive(rsc)); if (rsc->private->fns->is_filtered(rsc, only_rsc, TRUE)) { return pcmk_rc_no_output; } if (node == NULL) { // This is set only if a non-probe action is pending on this node node = rsc->private->pending_node; } return pe__common_output_text(out, rsc, rsc_printable_id(rsc), node, show_opts); } void native_free(pcmk_resource_t * rsc) { pcmk__rsc_trace(rsc, "Freeing resource action list (not the data)"); common_free(rsc); } enum rsc_role_e native_resource_state(const pcmk_resource_t * rsc, gboolean current) { enum rsc_role_e role = rsc->private->next_role; if (current) { role = rsc->private->orig_role; } pcmk__rsc_trace(rsc, "%s state: %s", rsc->id, pcmk_role_text(role)); return role; } /*! * \internal * \brief List nodes where a resource (or any of its children) is * * \param[in] rsc Resource to check * \param[out] list List to add result to * \param[in] current 0 = where allocated, 1 = where running, * 2 = where running or pending * * \return If list contains only one node, that node, or NULL otherwise */ pcmk_node_t * native_location(const pcmk_resource_t *rsc, GList **list, int current) { // @COMPAT: Accept a pcmk__rsc_node argument instead of int current pcmk_node_t *one = NULL; GList *result = NULL; if (rsc->private->children != NULL) { for (GList *gIter = rsc->private->children; gIter != NULL; gIter = gIter->next) { pcmk_resource_t *child = (pcmk_resource_t *) gIter->data; child->private->fns->location(child, &result, current); } } else if (current) { result = g_list_copy(rsc->private->active_nodes); if ((current == 2) && (rsc->private->pending_node != NULL) && !pe_find_node_id(result, rsc->private->pending_node->private->id)) { result = g_list_append(result, (gpointer) rsc->private->pending_node); } } else if (!current && (rsc->private->assigned_node != NULL)) { result = g_list_append(NULL, rsc->private->assigned_node); } if (result && (result->next == NULL)) { one = result->data; } if (list) { GList *gIter = result; for (; gIter != NULL; gIter = gIter->next) { pcmk_node_t *node = (pcmk_node_t *) gIter->data; if ((*list == NULL) || (pe_find_node_id(*list, node->private->id) == NULL)) { *list = g_list_append(*list, node); } } } g_list_free(result); return one; } static void get_rscs_brief(GList *rsc_list, GHashTable * rsc_table, GHashTable * active_table) { GList *gIter = rsc_list; for (; gIter != NULL; gIter = gIter->next) { pcmk_resource_t *rsc = (pcmk_resource_t *) gIter->data; const char *class = crm_element_value(rsc->private->xml, PCMK_XA_CLASS); const char *kind = crm_element_value(rsc->private->xml, PCMK_XA_TYPE); int offset = 0; char buffer[LINE_MAX]; int *rsc_counter = NULL; int *active_counter = NULL; if (!pcmk__is_primitive(rsc)) { continue; } offset += snprintf(buffer + offset, LINE_MAX - offset, "%s", class); if (pcmk_is_set(pcmk_get_ra_caps(class), pcmk_ra_cap_provider)) { const char *prov = crm_element_value(rsc->private->xml, PCMK_XA_PROVIDER); if (prov != NULL) { offset += snprintf(buffer + offset, LINE_MAX - offset, ":%s", prov); } } offset += snprintf(buffer + offset, LINE_MAX - offset, ":%s", kind); CRM_LOG_ASSERT(offset > 0); if (rsc_table) { rsc_counter = g_hash_table_lookup(rsc_table, buffer); if (rsc_counter == NULL) { rsc_counter = pcmk__assert_alloc(1, sizeof(int)); *rsc_counter = 0; g_hash_table_insert(rsc_table, strdup(buffer), rsc_counter); } (*rsc_counter)++; } if (active_table) { for (GList *gIter2 = rsc->private->active_nodes; gIter2 != NULL; gIter2 = gIter2->next) { pcmk_node_t *node = (pcmk_node_t *) gIter2->data; GHashTable *node_table = NULL; if (node->details->unclean == FALSE && node->details->online == FALSE && pcmk_is_set(rsc->flags, pcmk__rsc_managed)) { continue; } node_table = g_hash_table_lookup(active_table, node->private->name); if (node_table == NULL) { node_table = pcmk__strkey_table(free, free); g_hash_table_insert(active_table, strdup(node->private->name), node_table); } active_counter = g_hash_table_lookup(node_table, buffer); if (active_counter == NULL) { active_counter = pcmk__assert_alloc(1, sizeof(int)); *active_counter = 0; g_hash_table_insert(node_table, strdup(buffer), active_counter); } (*active_counter)++; } } } } static void destroy_node_table(gpointer data) { GHashTable *node_table = data; if (node_table) { g_hash_table_destroy(node_table); } } int pe__rscs_brief_output(pcmk__output_t *out, GList *rsc_list, uint32_t show_opts) { GHashTable *rsc_table = pcmk__strkey_table(free, free); GHashTable *active_table = pcmk__strkey_table(free, destroy_node_table); GList *sorted_rscs; int rc = pcmk_rc_no_output; get_rscs_brief(rsc_list, rsc_table, active_table); /* Make a list of the rsc_table keys so that it can be sorted. This is to make sure * output order stays consistent between systems. */ sorted_rscs = g_hash_table_get_keys(rsc_table); sorted_rscs = g_list_sort(sorted_rscs, (GCompareFunc) strcmp); for (GList *gIter = sorted_rscs; gIter; gIter = gIter->next) { char *type = (char *) gIter->data; int *rsc_counter = g_hash_table_lookup(rsc_table, type); GList *sorted_nodes = NULL; int active_counter_all = 0; /* Also make a list of the active_table keys so it can be sorted. If there's * more than one instance of a type of resource running, we need the nodes to * be sorted to make sure output order stays consistent between systems. */ sorted_nodes = g_hash_table_get_keys(active_table); sorted_nodes = g_list_sort(sorted_nodes, (GCompareFunc) pcmk__numeric_strcasecmp); for (GList *gIter2 = sorted_nodes; gIter2; gIter2 = gIter2->next) { char *node_name = (char *) gIter2->data; GHashTable *node_table = g_hash_table_lookup(active_table, node_name); int *active_counter = NULL; if (node_table == NULL) { continue; } active_counter = g_hash_table_lookup(node_table, type); if (active_counter == NULL || *active_counter == 0) { continue; } else { active_counter_all += *active_counter; } if (pcmk_is_set(show_opts, pcmk_show_rsc_only)) { node_name = NULL; } if (pcmk_is_set(show_opts, pcmk_show_inactive_rscs)) { out->list_item(out, NULL, "%d/%d\t(%s):\tActive %s", *active_counter, rsc_counter ? *rsc_counter : 0, type, (*active_counter > 0) && node_name ? node_name : ""); } else { out->list_item(out, NULL, "%d\t(%s):\tActive %s", *active_counter, type, (*active_counter > 0) && node_name ? node_name : ""); } rc = pcmk_rc_ok; } if (pcmk_is_set(show_opts, pcmk_show_inactive_rscs) && active_counter_all == 0) { out->list_item(out, NULL, "%d/%d\t(%s):\tActive", active_counter_all, rsc_counter ? *rsc_counter : 0, type); rc = pcmk_rc_ok; } if (sorted_nodes) { g_list_free(sorted_nodes); } } if (rsc_table) { g_hash_table_destroy(rsc_table); rsc_table = NULL; } if (active_table) { g_hash_table_destroy(active_table); active_table = NULL; } if (sorted_rscs) { g_list_free(sorted_rscs); } return rc; } gboolean pe__native_is_filtered(const pcmk_resource_t *rsc, GList *only_rsc, gboolean check_parent) { if (pcmk__str_in_list(rsc_printable_id(rsc), only_rsc, pcmk__str_star_matches) || pcmk__str_in_list(rsc->id, only_rsc, pcmk__str_star_matches)) { return FALSE; } else if (check_parent && (rsc->private->parent != NULL)) { const pcmk_resource_t *up = pe__const_top_resource(rsc, true); return up->private->fns->is_filtered(up, only_rsc, FALSE); } return TRUE; } /*! * \internal * \brief Get maximum primitive resource instances per node * * \param[in] rsc Primitive resource to check * * \return Maximum number of \p rsc instances that can be active on one node */ unsigned int pe__primitive_max_per_node(const pcmk_resource_t *rsc) { CRM_ASSERT(pcmk__is_primitive(rsc)); return 1U; } diff --git a/lib/pengine/unpack.c b/lib/pengine/unpack.c index 9a6818b810..5136ca26a6 100644 --- a/lib/pengine/unpack.c +++ b/lib/pengine/unpack.c @@ -1,5094 +1,5094 @@ /* * Copyright 2004-2024 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #include #include #include #include #include CRM_TRACE_INIT_DATA(pe_status); // A (parsed) resource action history entry struct action_history { pcmk_resource_t *rsc; // Resource that history is for pcmk_node_t *node; // Node that history is for xmlNode *xml; // History entry XML // Parsed from entry XML const char *id; // XML ID of history entry const char *key; // Operation key of action const char *task; // Action name const char *exit_reason; // Exit reason given for result guint interval_ms; // Action interval int call_id; // Call ID of action int expected_exit_status; // Expected exit status of action int exit_status; // Actual exit status of action int execution_status; // Execution status of action }; /* This uses pcmk__set_flags_as()/pcmk__clear_flags_as() directly rather than * use pcmk__set_scheduler_flags()/pcmk__clear_scheduler_flags() so that the * flag is stringified more readably in log messages. */ #define set_config_flag(scheduler, option, flag) do { \ GHashTable *config_hash = (scheduler)->config_hash; \ const char *scf_value = pcmk__cluster_option(config_hash, (option)); \ \ if (scf_value != NULL) { \ if (crm_is_true(scf_value)) { \ (scheduler)->flags = pcmk__set_flags_as(__func__, __LINE__, \ LOG_TRACE, "Scheduler", \ crm_system_name, (scheduler)->flags, \ (flag), #flag); \ } else { \ (scheduler)->flags = pcmk__clear_flags_as(__func__, __LINE__, \ LOG_TRACE, "Scheduler", \ crm_system_name, (scheduler)->flags, \ (flag), #flag); \ } \ } \ } while(0) static void unpack_rsc_op(pcmk_resource_t *rsc, pcmk_node_t *node, xmlNode *xml_op, xmlNode **last_failure, enum pcmk__on_fail *failed); static void determine_remote_online_status(pcmk_scheduler_t *scheduler, pcmk_node_t *this_node); static void add_node_attrs(const xmlNode *xml_obj, pcmk_node_t *node, bool overwrite, pcmk_scheduler_t *scheduler); static void determine_online_status(const xmlNode *node_state, pcmk_node_t *this_node, pcmk_scheduler_t *scheduler); static void unpack_node_lrm(pcmk_node_t *node, const xmlNode *xml, pcmk_scheduler_t *scheduler); /*! * \internal * \brief Check whether a node is a dangling guest node * * \param[in] node Node to check * * \return true if \p node had a Pacemaker Remote connection resource with a * launcher that was removed from the CIB, otherwise false. */ static bool is_dangling_guest_node(pcmk_node_t *node) { return pcmk__is_pacemaker_remote_node(node) && (node->private->remote != NULL) && (node->private->remote->private->launcher == NULL) && pcmk_is_set(node->private->remote->flags, pcmk__rsc_removed_launched); } /*! * \brief Schedule a fence action for a node * * \param[in,out] scheduler Scheduler data * \param[in,out] node Node to fence * \param[in] reason Text description of why fencing is needed * \param[in] priority_delay Whether to consider * \c PCMK_OPT_PRIORITY_FENCING_DELAY */ void pe_fence_node(pcmk_scheduler_t *scheduler, pcmk_node_t *node, const char *reason, bool priority_delay) { CRM_CHECK(node, return); if (pcmk__is_guest_or_bundle_node(node)) { // Fence a guest or bundle node by marking its launcher as failed pcmk_resource_t *rsc = node->private->remote->private->launcher; if (!pcmk_is_set(rsc->flags, pcmk__rsc_failed)) { if (!pcmk_is_set(rsc->flags, pcmk__rsc_managed)) { crm_notice("Not fencing guest node %s " "(otherwise would because %s): " "its guest resource %s is unmanaged", pcmk__node_name(node), reason, rsc->id); } else { pcmk__sched_warn("Guest node %s will be fenced " "(by recovering its guest resource %s): %s", pcmk__node_name(node), rsc->id, reason); /* We don't mark the node as unclean because that would prevent the * node from running resources. We want to allow it to run resources * in this transition if the recovery succeeds. */ pcmk__set_node_flags(node, pcmk__node_remote_reset); pcmk__set_rsc_flags(rsc, pcmk__rsc_failed|pcmk__rsc_stop_if_failed); } } } else if (is_dangling_guest_node(node)) { crm_info("Cleaning up dangling connection for guest node %s: " "fencing was already done because %s, " "and guest resource no longer exists", pcmk__node_name(node), reason); pcmk__set_rsc_flags(node->private->remote, pcmk__rsc_failed|pcmk__rsc_stop_if_failed); } else if (pcmk__is_remote_node(node)) { pcmk_resource_t *rsc = node->private->remote; if ((rsc != NULL) && !pcmk_is_set(rsc->flags, pcmk__rsc_managed)) { crm_notice("Not fencing remote node %s " "(otherwise would because %s): connection is unmanaged", pcmk__node_name(node), reason); } else if (!pcmk_is_set(node->private->flags, pcmk__node_remote_reset)) { pcmk__set_node_flags(node, pcmk__node_remote_reset); pcmk__sched_warn("Remote node %s %s: %s", pcmk__node_name(node), pe_can_fence(scheduler, node)? "will be fenced" : "is unclean", reason); } node->details->unclean = TRUE; // No need to apply PCMK_OPT_PRIORITY_FENCING_DELAY for remote nodes pe_fence_op(node, NULL, TRUE, reason, FALSE, scheduler); } else if (node->details->unclean) { crm_trace("Cluster node %s %s because %s", pcmk__node_name(node), pe_can_fence(scheduler, node)? "would also be fenced" : "also is unclean", reason); } else { pcmk__sched_warn("Cluster node %s %s: %s", pcmk__node_name(node), pe_can_fence(scheduler, node)? "will be fenced" : "is unclean", reason); node->details->unclean = TRUE; pe_fence_op(node, NULL, TRUE, reason, priority_delay, scheduler); } } // @TODO xpaths can't handle templates, rules, or id-refs // nvpair with provides or requires set to unfencing #define XPATH_UNFENCING_NVPAIR PCMK_XE_NVPAIR \ "[(@" PCMK_XA_NAME "='" PCMK_STONITH_PROVIDES "'" \ "or @" PCMK_XA_NAME "='" PCMK_META_REQUIRES "') " \ "and @" PCMK_XA_VALUE "='" PCMK_VALUE_UNFENCING "']" // unfencing in rsc_defaults or any resource #define XPATH_ENABLE_UNFENCING \ "/" PCMK_XE_CIB "/" PCMK_XE_CONFIGURATION "/" PCMK_XE_RESOURCES \ "//" PCMK_XE_META_ATTRIBUTES "/" XPATH_UNFENCING_NVPAIR \ "|/" PCMK_XE_CIB "/" PCMK_XE_CONFIGURATION "/" PCMK_XE_RSC_DEFAULTS \ "/" PCMK_XE_META_ATTRIBUTES "/" XPATH_UNFENCING_NVPAIR static void set_if_xpath(uint64_t flag, const char *xpath, pcmk_scheduler_t *scheduler) { xmlXPathObjectPtr result = NULL; if (!pcmk_is_set(scheduler->flags, flag)) { result = xpath_search(scheduler->input, xpath); if (result && (numXpathResults(result) > 0)) { pcmk__set_scheduler_flags(scheduler, flag); } freeXpathObject(result); } } gboolean unpack_config(xmlNode *config, pcmk_scheduler_t *scheduler) { const char *value = NULL; guint interval_ms = 0U; GHashTable *config_hash = pcmk__strkey_table(free, free); pe_rule_eval_data_t rule_data = { .node_hash = NULL, .now = scheduler->now, .match_data = NULL, .rsc_data = NULL, .op_data = NULL }; scheduler->config_hash = config_hash; pe__unpack_dataset_nvpairs(config, PCMK_XE_CLUSTER_PROPERTY_SET, &rule_data, config_hash, PCMK_VALUE_CIB_BOOTSTRAP_OPTIONS, FALSE, scheduler); pcmk__validate_cluster_options(config_hash); set_config_flag(scheduler, PCMK_OPT_ENABLE_STARTUP_PROBES, pcmk_sched_probe_resources); if (!pcmk_is_set(scheduler->flags, pcmk_sched_probe_resources)) { crm_info("Startup probes: disabled (dangerous)"); } value = pcmk__cluster_option(config_hash, PCMK_OPT_HAVE_WATCHDOG); if (value && crm_is_true(value)) { crm_info("Watchdog-based self-fencing will be performed via SBD if " "fencing is required and " PCMK_OPT_STONITH_WATCHDOG_TIMEOUT " is nonzero"); pcmk__set_scheduler_flags(scheduler, pcmk_sched_have_fencing); } /* Set certain flags via xpath here, so they can be used before the relevant * configuration sections are unpacked. */ set_if_xpath(pcmk_sched_enable_unfencing, XPATH_ENABLE_UNFENCING, scheduler); value = pcmk__cluster_option(config_hash, PCMK_OPT_STONITH_TIMEOUT); pcmk_parse_interval_spec(value, &interval_ms); if (interval_ms >= INT_MAX) { scheduler->stonith_timeout = INT_MAX; } else { scheduler->stonith_timeout = (int) interval_ms; } crm_debug("STONITH timeout: %d", scheduler->stonith_timeout); set_config_flag(scheduler, PCMK_OPT_STONITH_ENABLED, pcmk_sched_fencing_enabled); if (pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled)) { crm_debug("STONITH of failed nodes is enabled"); } else { crm_debug("STONITH of failed nodes is disabled"); } scheduler->stonith_action = pcmk__cluster_option(config_hash, PCMK_OPT_STONITH_ACTION); if (!strcmp(scheduler->stonith_action, PCMK__ACTION_POWEROFF)) { pcmk__warn_once(pcmk__wo_poweroff, "Support for " PCMK_OPT_STONITH_ACTION " of " "'" PCMK__ACTION_POWEROFF "' is deprecated and will be " "removed in a future release " "(use '" PCMK_ACTION_OFF "' instead)"); scheduler->stonith_action = PCMK_ACTION_OFF; } crm_trace("STONITH will %s nodes", scheduler->stonith_action); set_config_flag(scheduler, PCMK_OPT_CONCURRENT_FENCING, pcmk_sched_concurrent_fencing); if (pcmk_is_set(scheduler->flags, pcmk_sched_concurrent_fencing)) { crm_debug("Concurrent fencing is enabled"); } else { crm_debug("Concurrent fencing is disabled"); } value = pcmk__cluster_option(config_hash, PCMK_OPT_PRIORITY_FENCING_DELAY); if (value) { pcmk_parse_interval_spec(value, &interval_ms); scheduler->priority_fencing_delay = (int) (interval_ms / 1000); crm_trace("Priority fencing delay is %ds", scheduler->priority_fencing_delay); } set_config_flag(scheduler, PCMK_OPT_STOP_ALL_RESOURCES, pcmk_sched_stop_all); crm_debug("Stop all active resources: %s", pcmk__flag_text(scheduler->flags, pcmk_sched_stop_all)); set_config_flag(scheduler, PCMK_OPT_SYMMETRIC_CLUSTER, pcmk_sched_symmetric_cluster); if (pcmk_is_set(scheduler->flags, pcmk_sched_symmetric_cluster)) { crm_debug("Cluster is symmetric" " - resources can run anywhere by default"); } value = pcmk__cluster_option(config_hash, PCMK_OPT_NO_QUORUM_POLICY); if (pcmk__str_eq(value, PCMK_VALUE_IGNORE, pcmk__str_casei)) { scheduler->no_quorum_policy = pcmk_no_quorum_ignore; } else if (pcmk__str_eq(value, PCMK_VALUE_FREEZE, pcmk__str_casei)) { scheduler->no_quorum_policy = pcmk_no_quorum_freeze; } else if (pcmk__str_eq(value, PCMK_VALUE_DEMOTE, pcmk__str_casei)) { scheduler->no_quorum_policy = pcmk_no_quorum_demote; } else if (pcmk__str_eq(value, PCMK_VALUE_FENCE_LEGACY, pcmk__str_casei)) { if (pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled)) { int do_panic = 0; crm_element_value_int(scheduler->input, PCMK_XA_NO_QUORUM_PANIC, &do_panic); if (do_panic || pcmk_is_set(scheduler->flags, pcmk_sched_quorate)) { scheduler->no_quorum_policy = pcmk_no_quorum_fence; } else { crm_notice("Resetting " PCMK_OPT_NO_QUORUM_POLICY " to 'stop': cluster has never had quorum"); scheduler->no_quorum_policy = pcmk_no_quorum_stop; } } else { pcmk__config_err("Resetting " PCMK_OPT_NO_QUORUM_POLICY " to 'stop' because fencing is disabled"); scheduler->no_quorum_policy = pcmk_no_quorum_stop; } } else { scheduler->no_quorum_policy = pcmk_no_quorum_stop; } switch (scheduler->no_quorum_policy) { case pcmk_no_quorum_freeze: crm_debug("On loss of quorum: Freeze resources"); break; case pcmk_no_quorum_stop: crm_debug("On loss of quorum: Stop ALL resources"); break; case pcmk_no_quorum_demote: crm_debug("On loss of quorum: " "Demote promotable resources and stop other resources"); break; case pcmk_no_quorum_fence: crm_notice("On loss of quorum: Fence all remaining nodes"); break; case pcmk_no_quorum_ignore: crm_notice("On loss of quorum: Ignore"); break; } set_config_flag(scheduler, PCMK_OPT_STOP_ORPHAN_RESOURCES, pcmk_sched_stop_removed_resources); if (pcmk_is_set(scheduler->flags, pcmk_sched_stop_removed_resources)) { crm_trace("Orphan resources are stopped"); } else { crm_trace("Orphan resources are ignored"); } set_config_flag(scheduler, PCMK_OPT_STOP_ORPHAN_ACTIONS, pcmk_sched_cancel_removed_actions); if (pcmk_is_set(scheduler->flags, pcmk_sched_cancel_removed_actions)) { crm_trace("Orphan resource actions are stopped"); } else { crm_trace("Orphan resource actions are ignored"); } value = pcmk__cluster_option(config_hash, PCMK__OPT_REMOVE_AFTER_STOP); if (value != NULL) { if (crm_is_true(value)) { pcmk__set_scheduler_flags(scheduler, pcmk_sched_remove_after_stop); pcmk__warn_once(pcmk__wo_remove_after, "Support for the " PCMK__OPT_REMOVE_AFTER_STOP " cluster property is deprecated and will be " "removed in a future release"); } else { pcmk__clear_scheduler_flags(scheduler, pcmk_sched_remove_after_stop); } } set_config_flag(scheduler, PCMK_OPT_MAINTENANCE_MODE, pcmk_sched_in_maintenance); crm_trace("Maintenance mode: %s", pcmk__flag_text(scheduler->flags, pcmk_sched_in_maintenance)); set_config_flag(scheduler, PCMK_OPT_START_FAILURE_IS_FATAL, pcmk_sched_start_failure_fatal); if (pcmk_is_set(scheduler->flags, pcmk_sched_start_failure_fatal)) { crm_trace("Start failures are always fatal"); } else { crm_trace("Start failures are handled by failcount"); } if (pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled)) { set_config_flag(scheduler, PCMK_OPT_STARTUP_FENCING, pcmk_sched_startup_fencing); } if (pcmk_is_set(scheduler->flags, pcmk_sched_startup_fencing)) { crm_trace("Unseen nodes will be fenced"); } else { pcmk__warn_once(pcmk__wo_blind, "Blind faith: not fencing unseen nodes"); } pe__unpack_node_health_scores(scheduler); scheduler->placement_strategy = pcmk__cluster_option(config_hash, PCMK_OPT_PLACEMENT_STRATEGY); crm_trace("Placement strategy: %s", scheduler->placement_strategy); set_config_flag(scheduler, PCMK_OPT_SHUTDOWN_LOCK, pcmk_sched_shutdown_lock); if (pcmk_is_set(scheduler->flags, pcmk_sched_shutdown_lock)) { value = pcmk__cluster_option(config_hash, PCMK_OPT_SHUTDOWN_LOCK_LIMIT); pcmk_parse_interval_spec(value, &(scheduler->shutdown_lock)); scheduler->shutdown_lock /= 1000; crm_trace("Resources will be locked to nodes that were cleanly " "shut down (locks expire after %s)", pcmk__readable_interval(scheduler->shutdown_lock)); } else { crm_trace("Resources will not be locked to nodes that were cleanly " "shut down"); } value = pcmk__cluster_option(config_hash, PCMK_OPT_NODE_PENDING_TIMEOUT); pcmk_parse_interval_spec(value, &(scheduler->node_pending_timeout)); scheduler->node_pending_timeout /= 1000; if (scheduler->node_pending_timeout == 0) { crm_trace("Do not fence pending nodes"); } else { crm_trace("Fence pending nodes after %s", pcmk__readable_interval(scheduler->node_pending_timeout * 1000)); } return TRUE; } pcmk_node_t * pe_create_node(const char *id, const char *uname, const char *type, const char *score, pcmk_scheduler_t *scheduler) { pcmk_node_t *new_node = NULL; if (pcmk_find_node(scheduler, uname) != NULL) { pcmk__config_warn("More than one node entry has name '%s'", uname); } new_node = calloc(1, sizeof(pcmk_node_t)); if (new_node == NULL) { pcmk__sched_err("Could not allocate memory for node %s", uname); return NULL; } new_node->assign = calloc(1, sizeof(struct pcmk__node_assignment)); - new_node->details = calloc(1, sizeof(struct pe_node_shared_s)); + new_node->details = calloc(1, sizeof(struct pcmk__node_details)); new_node->private = calloc(1, sizeof(pcmk__node_private_t)); if ((new_node->assign == NULL) || (new_node->details == NULL) || (new_node->private == NULL)) { free(new_node->assign); free(new_node->details); free(new_node->private); free(new_node); pcmk__sched_err("Could not allocate memory for node %s", uname); return NULL; } crm_trace("Creating node for entry %s/%s", uname, id); new_node->assign->score = char2score(score); new_node->private->id = id; new_node->private->name = uname; new_node->private->flags = pcmk__node_probes_allowed; new_node->details->online = FALSE; new_node->details->shutdown = FALSE; new_node->details->running_rsc = NULL; new_node->private->scheduler = scheduler; if (pcmk__str_eq(type, PCMK_VALUE_MEMBER, pcmk__str_null_matches|pcmk__str_casei)) { new_node->private->variant = pcmk__node_variant_cluster; } else if (pcmk__str_eq(type, PCMK_VALUE_REMOTE, pcmk__str_casei)) { new_node->private->variant = pcmk__node_variant_remote; pcmk__set_scheduler_flags(scheduler, pcmk_sched_have_remote_nodes); } else { /* @COMPAT 'ping' is the default for backward compatibility, but it * should be changed to 'member' at a compatibility break */ if (!pcmk__str_eq(type, PCMK__VALUE_PING, pcmk__str_casei)) { pcmk__config_warn("Node %s has unrecognized type '%s', " "assuming '" PCMK__VALUE_PING "'", pcmk__s(uname, "without name"), type); } pcmk__warn_once(pcmk__wo_ping_node, "Support for nodes of type '" PCMK__VALUE_PING "' " "(such as %s) is deprecated and will be removed in a " "future release", pcmk__s(uname, "unnamed node")); new_node->private->variant = pcmk__node_variant_ping; } new_node->private->attrs = pcmk__strkey_table(free, free); if (pcmk__is_pacemaker_remote_node(new_node)) { pcmk__insert_dup(new_node->private->attrs, CRM_ATTR_KIND, "remote"); } else { pcmk__insert_dup(new_node->private->attrs, CRM_ATTR_KIND, "cluster"); } new_node->private->utilization = pcmk__strkey_table(free, free); new_node->private->digest_cache = pcmk__strkey_table(free, pe__free_digests); scheduler->nodes = g_list_insert_sorted(scheduler->nodes, new_node, pe__cmp_node_name); return new_node; } static const char * expand_remote_rsc_meta(xmlNode *xml_obj, xmlNode *parent, pcmk_scheduler_t *data) { xmlNode *attr_set = NULL; xmlNode *attr = NULL; const char *container_id = pcmk__xe_id(xml_obj); const char *remote_name = NULL; const char *remote_server = NULL; const char *remote_port = NULL; const char *connect_timeout = "60s"; const char *remote_allow_migrate=NULL; const char *is_managed = NULL; for (attr_set = pcmk__xe_first_child(xml_obj, NULL, NULL, NULL); attr_set != NULL; attr_set = pcmk__xe_next(attr_set)) { if (!pcmk__xe_is(attr_set, PCMK_XE_META_ATTRIBUTES)) { continue; } for (attr = pcmk__xe_first_child(attr_set, NULL, NULL, NULL); attr != NULL; attr = pcmk__xe_next(attr)) { const char *value = crm_element_value(attr, PCMK_XA_VALUE); const char *name = crm_element_value(attr, PCMK_XA_NAME); if (name == NULL) { // Sanity continue; } if (strcmp(name, PCMK_META_REMOTE_NODE) == 0) { remote_name = value; } else if (strcmp(name, PCMK_META_REMOTE_ADDR) == 0) { remote_server = value; } else if (strcmp(name, PCMK_META_REMOTE_PORT) == 0) { remote_port = value; } else if (strcmp(name, PCMK_META_REMOTE_CONNECT_TIMEOUT) == 0) { connect_timeout = value; } else if (strcmp(name, PCMK_META_REMOTE_ALLOW_MIGRATE) == 0) { remote_allow_migrate = value; } else if (strcmp(name, PCMK_META_IS_MANAGED) == 0) { is_managed = value; } } } if (remote_name == NULL) { return NULL; } if (pe_find_resource(data->resources, remote_name) != NULL) { return NULL; } pe_create_remote_xml(parent, remote_name, container_id, remote_allow_migrate, is_managed, connect_timeout, remote_server, remote_port); return remote_name; } static void handle_startup_fencing(pcmk_scheduler_t *scheduler, pcmk_node_t *new_node) { if ((new_node->private->variant == pcmk__node_variant_remote) && (new_node->private->remote == NULL)) { /* Ignore fencing for remote nodes that don't have a connection resource * associated with them. This happens when remote node entries get left * in the nodes section after the connection resource is removed. */ return; } if (pcmk_is_set(scheduler->flags, pcmk_sched_startup_fencing)) { // All nodes are unclean until we've seen their status entry new_node->details->unclean = TRUE; } else { // Blind faith ... new_node->details->unclean = FALSE; } } gboolean unpack_nodes(xmlNode *xml_nodes, pcmk_scheduler_t *scheduler) { xmlNode *xml_obj = NULL; pcmk_node_t *new_node = NULL; const char *id = NULL; const char *uname = NULL; const char *type = NULL; const char *score = NULL; for (xml_obj = pcmk__xe_first_child(xml_nodes, NULL, NULL, NULL); xml_obj != NULL; xml_obj = pcmk__xe_next(xml_obj)) { if (pcmk__xe_is(xml_obj, PCMK_XE_NODE)) { new_node = NULL; id = crm_element_value(xml_obj, PCMK_XA_ID); uname = crm_element_value(xml_obj, PCMK_XA_UNAME); type = crm_element_value(xml_obj, PCMK_XA_TYPE); score = crm_element_value(xml_obj, PCMK_XA_SCORE); crm_trace("Processing node %s/%s", uname, id); if (id == NULL) { pcmk__config_err("Ignoring <" PCMK_XE_NODE "> entry in configuration without id"); continue; } new_node = pe_create_node(id, uname, type, score, scheduler); if (new_node == NULL) { return FALSE; } handle_startup_fencing(scheduler, new_node); add_node_attrs(xml_obj, new_node, FALSE, scheduler); crm_trace("Done with node %s", crm_element_value(xml_obj, PCMK_XA_UNAME)); } } if (scheduler->localhost && (pcmk_find_node(scheduler, scheduler->localhost) == NULL)) { crm_info("Creating a fake local node"); pe_create_node(scheduler->localhost, scheduler->localhost, NULL, 0, scheduler); } return TRUE; } static void unpack_launcher(pcmk_resource_t *rsc, pcmk_scheduler_t *scheduler) { const char *launcher_id = NULL; if (rsc->private->children != NULL) { g_list_foreach(rsc->private->children, (GFunc) unpack_launcher, scheduler); return; } launcher_id = g_hash_table_lookup(rsc->private->meta, PCMK__META_CONTAINER); if ((launcher_id != NULL) && !pcmk__str_eq(launcher_id, rsc->id, pcmk__str_none)) { pcmk_resource_t *launcher = pe_find_resource(scheduler->resources, launcher_id); if (launcher != NULL) { rsc->private->launcher = launcher; launcher->private->launched = g_list_append(launcher->private->launched, rsc); pcmk__rsc_trace(rsc, "Resource %s's launcher is %s", rsc->id, launcher_id); } else { pcmk__config_err("Resource %s: Unknown " PCMK__META_CONTAINER " %s", rsc->id, launcher_id); } } } gboolean unpack_remote_nodes(xmlNode *xml_resources, pcmk_scheduler_t *scheduler) { xmlNode *xml_obj = NULL; /* Create remote nodes and guest nodes from the resource configuration * before unpacking resources. */ for (xml_obj = pcmk__xe_first_child(xml_resources, NULL, NULL, NULL); xml_obj != NULL; xml_obj = pcmk__xe_next(xml_obj)) { const char *new_node_id = NULL; /* Check for remote nodes, which are defined by ocf:pacemaker:remote * primitives. */ if (xml_contains_remote_node(xml_obj)) { new_node_id = pcmk__xe_id(xml_obj); /* The pcmk_find_node() check ensures we don't iterate over an * expanded node that has already been added to the node list */ if (new_node_id && (pcmk_find_node(scheduler, new_node_id) == NULL)) { crm_trace("Found remote node %s defined by resource %s", new_node_id, pcmk__xe_id(xml_obj)); pe_create_node(new_node_id, new_node_id, PCMK_VALUE_REMOTE, NULL, scheduler); } continue; } /* Check for guest nodes, which are defined by special meta-attributes * of a primitive of any type (for example, VirtualDomain or Xen). */ if (pcmk__xe_is(xml_obj, PCMK_XE_PRIMITIVE)) { /* This will add an ocf:pacemaker:remote primitive to the * configuration for the guest node's connection, to be unpacked * later. */ new_node_id = expand_remote_rsc_meta(xml_obj, xml_resources, scheduler); if (new_node_id && (pcmk_find_node(scheduler, new_node_id) == NULL)) { crm_trace("Found guest node %s in resource %s", new_node_id, pcmk__xe_id(xml_obj)); pe_create_node(new_node_id, new_node_id, PCMK_VALUE_REMOTE, NULL, scheduler); } continue; } /* Check for guest nodes inside a group. Clones are currently not * supported as guest nodes. */ if (pcmk__xe_is(xml_obj, PCMK_XE_GROUP)) { xmlNode *xml_obj2 = NULL; for (xml_obj2 = pcmk__xe_first_child(xml_obj, NULL, NULL, NULL); xml_obj2 != NULL; xml_obj2 = pcmk__xe_next(xml_obj2)) { new_node_id = expand_remote_rsc_meta(xml_obj2, xml_resources, scheduler); if (new_node_id && (pcmk_find_node(scheduler, new_node_id) == NULL)) { crm_trace("Found guest node %s in resource %s inside group %s", new_node_id, pcmk__xe_id(xml_obj2), pcmk__xe_id(xml_obj)); pe_create_node(new_node_id, new_node_id, PCMK_VALUE_REMOTE, NULL, scheduler); } } } } return TRUE; } /* Call this after all the nodes and resources have been * unpacked, but before the status section is read. * * A remote node's online status is reflected by the state * of the remote node's connection resource. We need to link * the remote node to this connection resource so we can have * easy access to the connection resource during the scheduler calculations. */ static void link_rsc2remotenode(pcmk_scheduler_t *scheduler, pcmk_resource_t *new_rsc) { pcmk_node_t *remote_node = NULL; if (!pcmk_is_set(new_rsc->flags, pcmk__rsc_is_remote_connection)) { return; } if (pcmk_is_set(scheduler->flags, pcmk_sched_location_only)) { /* remote_nodes and remote_resources are not linked in quick location calculations */ return; } remote_node = pcmk_find_node(scheduler, new_rsc->id); CRM_CHECK(remote_node != NULL, return); pcmk__rsc_trace(new_rsc, "Linking remote connection resource %s to %s", new_rsc->id, pcmk__node_name(remote_node)); remote_node->private->remote = new_rsc; if (new_rsc->private->launcher == NULL) { /* Handle start-up fencing for remote nodes (as opposed to guest nodes) * the same as is done for cluster nodes. */ handle_startup_fencing(scheduler, remote_node); } else { /* pe_create_node() marks the new node as "remote" or "cluster"; now * that we know the node is a guest node, update it correctly. */ pcmk__insert_dup(remote_node->private->attrs, CRM_ATTR_KIND, "container"); } } static void destroy_tag(gpointer data) { pcmk_tag_t *tag = data; if (tag) { free(tag->id); g_list_free_full(tag->refs, free); free(tag); } } /*! * \internal * \brief Parse configuration XML for resource information * * \param[in] xml_resources Top of resource configuration XML * \param[in,out] scheduler Scheduler data * * \return TRUE * * \note unpack_remote_nodes() MUST be called before this, so that the nodes can * be used when pe__unpack_resource() calls resource_location() */ gboolean unpack_resources(const xmlNode *xml_resources, pcmk_scheduler_t *scheduler) { xmlNode *xml_obj = NULL; GList *gIter = NULL; scheduler->template_rsc_sets = pcmk__strkey_table(free, destroy_tag); for (xml_obj = pcmk__xe_first_child(xml_resources, NULL, NULL, NULL); xml_obj != NULL; xml_obj = pcmk__xe_next(xml_obj)) { pcmk_resource_t *new_rsc = NULL; const char *id = pcmk__xe_id(xml_obj); if (pcmk__str_empty(id)) { pcmk__config_err("Ignoring <%s> resource without ID", xml_obj->name); continue; } if (pcmk__xe_is(xml_obj, PCMK_XE_TEMPLATE)) { if (g_hash_table_lookup_extended(scheduler->template_rsc_sets, id, NULL, NULL) == FALSE) { /* Record the template's ID for the knowledge of its existence anyway. */ pcmk__insert_dup(scheduler->template_rsc_sets, id, NULL); } continue; } crm_trace("Unpacking <%s " PCMK_XA_ID "='%s'>", xml_obj->name, id); if (pe__unpack_resource(xml_obj, &new_rsc, NULL, scheduler) == pcmk_rc_ok) { scheduler->resources = g_list_append(scheduler->resources, new_rsc); pcmk__rsc_trace(new_rsc, "Added resource %s", new_rsc->id); } else { pcmk__config_err("Ignoring <%s> resource '%s' " "because configuration is invalid", xml_obj->name, id); } } for (gIter = scheduler->resources; gIter != NULL; gIter = gIter->next) { pcmk_resource_t *rsc = (pcmk_resource_t *) gIter->data; unpack_launcher(rsc, scheduler); link_rsc2remotenode(scheduler, rsc); } scheduler->resources = g_list_sort(scheduler->resources, pe__cmp_rsc_priority); if (pcmk_is_set(scheduler->flags, pcmk_sched_location_only)) { /* Ignore */ } else if (pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled) && !pcmk_is_set(scheduler->flags, pcmk_sched_have_fencing)) { pcmk__config_err("Resource start-up disabled since no STONITH resources have been defined"); pcmk__config_err("Either configure some or disable STONITH with the " PCMK_OPT_STONITH_ENABLED " option"); pcmk__config_err("NOTE: Clusters with shared data need STONITH to ensure data integrity"); } return TRUE; } gboolean unpack_tags(xmlNode *xml_tags, pcmk_scheduler_t *scheduler) { xmlNode *xml_tag = NULL; scheduler->tags = pcmk__strkey_table(free, destroy_tag); for (xml_tag = pcmk__xe_first_child(xml_tags, NULL, NULL, NULL); xml_tag != NULL; xml_tag = pcmk__xe_next(xml_tag)) { xmlNode *xml_obj_ref = NULL; const char *tag_id = pcmk__xe_id(xml_tag); if (!pcmk__xe_is(xml_tag, PCMK_XE_TAG)) { continue; } if (tag_id == NULL) { pcmk__config_err("Ignoring <%s> without " PCMK_XA_ID, (const char *) xml_tag->name); continue; } for (xml_obj_ref = pcmk__xe_first_child(xml_tag, NULL, NULL, NULL); xml_obj_ref != NULL; xml_obj_ref = pcmk__xe_next(xml_obj_ref)) { const char *obj_ref = pcmk__xe_id(xml_obj_ref); if (!pcmk__xe_is(xml_obj_ref, PCMK_XE_OBJ_REF)) { continue; } if (obj_ref == NULL) { pcmk__config_err("Ignoring <%s> for tag '%s' without " PCMK_XA_ID, xml_obj_ref->name, tag_id); continue; } if (add_tag_ref(scheduler->tags, tag_id, obj_ref) == FALSE) { return FALSE; } } } return TRUE; } /* The ticket state section: * "/cib/status/tickets/ticket_state" */ static gboolean unpack_ticket_state(xmlNode *xml_ticket, pcmk_scheduler_t *scheduler) { const char *ticket_id = NULL; const char *granted = NULL; const char *last_granted = NULL; const char *standby = NULL; xmlAttrPtr xIter = NULL; pcmk_ticket_t *ticket = NULL; ticket_id = pcmk__xe_id(xml_ticket); if (pcmk__str_empty(ticket_id)) { return FALSE; } crm_trace("Processing ticket state for %s", ticket_id); ticket = g_hash_table_lookup(scheduler->tickets, ticket_id); if (ticket == NULL) { ticket = ticket_new(ticket_id, scheduler); if (ticket == NULL) { return FALSE; } } for (xIter = xml_ticket->properties; xIter; xIter = xIter->next) { const char *prop_name = (const char *)xIter->name; const char *prop_value = pcmk__xml_attr_value(xIter); if (pcmk__str_eq(prop_name, PCMK_XA_ID, pcmk__str_none)) { continue; } pcmk__insert_dup(ticket->state, prop_name, prop_value); } granted = g_hash_table_lookup(ticket->state, PCMK__XA_GRANTED); if (granted && crm_is_true(granted)) { ticket->granted = TRUE; crm_info("We have ticket '%s'", ticket->id); } else { ticket->granted = FALSE; crm_info("We do not have ticket '%s'", ticket->id); } last_granted = g_hash_table_lookup(ticket->state, PCMK_XA_LAST_GRANTED); if (last_granted) { long long last_granted_ll; pcmk__scan_ll(last_granted, &last_granted_ll, 0LL); ticket->last_granted = (time_t) last_granted_ll; } standby = g_hash_table_lookup(ticket->state, PCMK_XA_STANDBY); if (standby && crm_is_true(standby)) { ticket->standby = TRUE; if (ticket->granted) { crm_info("Granted ticket '%s' is in standby-mode", ticket->id); } } else { ticket->standby = FALSE; } crm_trace("Done with ticket state for %s", ticket_id); return TRUE; } static gboolean unpack_tickets_state(xmlNode *xml_tickets, pcmk_scheduler_t *scheduler) { xmlNode *xml_obj = NULL; for (xml_obj = pcmk__xe_first_child(xml_tickets, NULL, NULL, NULL); xml_obj != NULL; xml_obj = pcmk__xe_next(xml_obj)) { if (!pcmk__xe_is(xml_obj, PCMK__XE_TICKET_STATE)) { continue; } unpack_ticket_state(xml_obj, scheduler); } return TRUE; } static void unpack_handle_remote_attrs(pcmk_node_t *this_node, const xmlNode *state, pcmk_scheduler_t *scheduler) { const char *discovery = NULL; const xmlNode *attrs = NULL; pcmk_resource_t *rsc = NULL; int maint = 0; if (!pcmk__xe_is(state, PCMK__XE_NODE_STATE)) { return; } if ((this_node == NULL) || !pcmk__is_pacemaker_remote_node(this_node)) { return; } crm_trace("Processing Pacemaker Remote node %s", pcmk__node_name(this_node)); pcmk__scan_min_int(crm_element_value(state, PCMK__XA_NODE_IN_MAINTENANCE), &maint, 0); if (maint) { pcmk__set_node_flags(this_node, pcmk__node_remote_maint); } else { pcmk__clear_node_flags(this_node, pcmk__node_remote_maint); } rsc = this_node->private->remote; if (!pcmk_is_set(this_node->private->flags, pcmk__node_remote_reset)) { this_node->details->unclean = FALSE; pcmk__set_node_flags(this_node, pcmk__node_seen); } attrs = pcmk__xe_first_child(state, PCMK__XE_TRANSIENT_ATTRIBUTES, NULL, NULL); add_node_attrs(attrs, this_node, TRUE, scheduler); if (pe__shutdown_requested(this_node)) { crm_info("%s is shutting down", pcmk__node_name(this_node)); this_node->details->shutdown = TRUE; } if (crm_is_true(pcmk__node_attr(this_node, PCMK_NODE_ATTR_STANDBY, NULL, pcmk__rsc_node_current))) { crm_info("%s is in standby mode", pcmk__node_name(this_node)); pcmk__set_node_flags(this_node, pcmk__node_standby); } if (crm_is_true(pcmk__node_attr(this_node, PCMK_NODE_ATTR_MAINTENANCE, NULL, pcmk__rsc_node_current)) || ((rsc != NULL) && !pcmk_is_set(rsc->flags, pcmk__rsc_managed))) { crm_info("%s is in maintenance mode", pcmk__node_name(this_node)); this_node->details->maintenance = TRUE; } discovery = pcmk__node_attr(this_node, PCMK__NODE_ATTR_RESOURCE_DISCOVERY_ENABLED, NULL, pcmk__rsc_node_current); if ((discovery != NULL) && !crm_is_true(discovery)) { pcmk__warn_once(pcmk__wo_rdisc_enabled, "Support for the " PCMK__NODE_ATTR_RESOURCE_DISCOVERY_ENABLED " node attribute is deprecated and will be removed" " (and behave as 'true') in a future release."); if (pcmk__is_remote_node(this_node) && !pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled)) { pcmk__config_warn("Ignoring " PCMK__NODE_ATTR_RESOURCE_DISCOVERY_ENABLED " attribute on Pacemaker Remote node %s" " because fencing is disabled", pcmk__node_name(this_node)); } else { /* This is either a remote node with fencing enabled, or a guest * node. We don't care whether fencing is enabled when fencing guest * nodes, because they are "fenced" by recovering their containing * resource. */ crm_info("%s has resource discovery disabled", pcmk__node_name(this_node)); pcmk__clear_node_flags(this_node, pcmk__node_probes_allowed); } } } /*! * \internal * \brief Unpack a cluster node's transient attributes * * \param[in] state CIB node state XML * \param[in,out] node Cluster node whose attributes are being unpacked * \param[in,out] scheduler Scheduler data */ static void unpack_transient_attributes(const xmlNode *state, pcmk_node_t *node, pcmk_scheduler_t *scheduler) { const char *discovery = NULL; const xmlNode *attrs = pcmk__xe_first_child(state, PCMK__XE_TRANSIENT_ATTRIBUTES, NULL, NULL); add_node_attrs(attrs, node, TRUE, scheduler); if (crm_is_true(pcmk__node_attr(node, PCMK_NODE_ATTR_STANDBY, NULL, pcmk__rsc_node_current))) { crm_info("%s is in standby mode", pcmk__node_name(node)); pcmk__set_node_flags(node, pcmk__node_standby); } if (crm_is_true(pcmk__node_attr(node, PCMK_NODE_ATTR_MAINTENANCE, NULL, pcmk__rsc_node_current))) { crm_info("%s is in maintenance mode", pcmk__node_name(node)); node->details->maintenance = TRUE; } discovery = pcmk__node_attr(node, PCMK__NODE_ATTR_RESOURCE_DISCOVERY_ENABLED, NULL, pcmk__rsc_node_current); if ((discovery != NULL) && !crm_is_true(discovery)) { pcmk__config_warn("Ignoring " PCMK__NODE_ATTR_RESOURCE_DISCOVERY_ENABLED " attribute for %s because disabling resource" " discovery is not allowed for cluster nodes", pcmk__node_name(node)); } } /*! * \internal * \brief Unpack a node state entry (first pass) * * Unpack one node state entry from status. This unpacks information from the * \C PCMK__XE_NODE_STATE element itself and node attributes inside it, but not * the resource history inside it. Multiple passes through the status are needed * to fully unpack everything. * * \param[in] state CIB node state XML * \param[in,out] scheduler Scheduler data */ static void unpack_node_state(const xmlNode *state, pcmk_scheduler_t *scheduler) { const char *id = NULL; const char *uname = NULL; pcmk_node_t *this_node = NULL; id = crm_element_value(state, PCMK_XA_ID); if (id == NULL) { pcmk__config_err("Ignoring invalid " PCMK__XE_NODE_STATE " entry without " PCMK_XA_ID); crm_log_xml_info(state, "missing-id"); return; } uname = crm_element_value(state, PCMK_XA_UNAME); if (uname == NULL) { /* If a joining peer makes the cluster acquire the quorum from corosync * meanwhile it has not joined CPG membership of pacemaker-controld yet, * it's possible that the created PCMK__XE_NODE_STATE entry doesn't have * a PCMK_XA_UNAME yet. We should recognize the node as `pending` and * wait for it to join CPG. */ crm_trace("Handling " PCMK__XE_NODE_STATE " entry with id=\"%s\" " "without " PCMK_XA_UNAME, id); } this_node = pe_find_node_any(scheduler->nodes, id, uname); if (this_node == NULL) { crm_notice("Ignoring recorded state for removed node with name %s and " PCMK_XA_ID " %s", pcmk__s(uname, "unknown"), id); return; } if (pcmk__is_pacemaker_remote_node(this_node)) { int remote_fenced = 0; /* We can't determine the online status of Pacemaker Remote nodes until * after all resource history has been unpacked. In this first pass, we * do need to mark whether the node has been fenced, as this plays a * role during unpacking cluster node resource state. */ pcmk__scan_min_int(crm_element_value(state, PCMK__XA_NODE_FENCED), &remote_fenced, 0); if (remote_fenced) { pcmk__set_node_flags(this_node, pcmk__node_remote_fenced); } else { pcmk__clear_node_flags(this_node, pcmk__node_remote_fenced); } return; } unpack_transient_attributes(state, this_node, scheduler); /* Provisionally mark this cluster node as clean. We have at least seen it * in the current cluster's lifetime. */ this_node->details->unclean = FALSE; pcmk__set_node_flags(this_node, pcmk__node_seen); crm_trace("Determining online status of cluster node %s (id %s)", pcmk__node_name(this_node), id); determine_online_status(state, this_node, scheduler); if (!pcmk_is_set(scheduler->flags, pcmk_sched_quorate) && this_node->details->online && (scheduler->no_quorum_policy == pcmk_no_quorum_fence)) { /* Everything else should flow from this automatically * (at least until the scheduler becomes able to migrate off * healthy resources) */ pe_fence_node(scheduler, this_node, "cluster does not have quorum", FALSE); } } /*! * \internal * \brief Unpack nodes' resource history as much as possible * * Unpack as many nodes' resource history as possible in one pass through the * status. We need to process Pacemaker Remote nodes' connections/containers * before unpacking their history; the connection/container history will be * in another node's history, so it might take multiple passes to unpack * everything. * * \param[in] status CIB XML status section * \param[in] fence If true, treat any not-yet-unpacked nodes as unseen * \param[in,out] scheduler Scheduler data * * \return Standard Pacemaker return code (specifically pcmk_rc_ok if done, * or EAGAIN if more unpacking remains to be done) */ static int unpack_node_history(const xmlNode *status, bool fence, pcmk_scheduler_t *scheduler) { int rc = pcmk_rc_ok; // Loop through all PCMK__XE_NODE_STATE entries in CIB status for (const xmlNode *state = pcmk__xe_first_child(status, PCMK__XE_NODE_STATE, NULL, NULL); state != NULL; state = pcmk__xe_next_same(state)) { const char *id = pcmk__xe_id(state); const char *uname = crm_element_value(state, PCMK_XA_UNAME); pcmk_node_t *this_node = NULL; if ((id == NULL) || (uname == NULL)) { // Warning already logged in first pass through status section crm_trace("Not unpacking resource history from malformed " PCMK__XE_NODE_STATE " without id and/or uname"); continue; } this_node = pe_find_node_any(scheduler->nodes, id, uname); if (this_node == NULL) { // Warning already logged in first pass through status section crm_trace("Not unpacking resource history for node %s because " "no longer in configuration", id); continue; } if (pcmk_is_set(this_node->private->flags, pcmk__node_unpacked)) { crm_trace("Not unpacking resource history for node %s because " "already unpacked", id); continue; } if (fence) { // We're processing all remaining nodes } else if (pcmk__is_guest_or_bundle_node(this_node)) { /* We can unpack a guest node's history only after we've unpacked * other resource history to the point that we know that the node's * connection and containing resource are both up. */ const pcmk_resource_t *remote = this_node->private->remote; const pcmk_resource_t *launcher = remote->private->launcher; if ((remote->private->orig_role != pcmk_role_started) || (launcher->private->orig_role != pcmk_role_started)) { crm_trace("Not unpacking resource history for guest node %s " "because launcher and connection are not known to " "be up", id); continue; } } else if (pcmk__is_remote_node(this_node)) { /* We can unpack a remote node's history only after we've unpacked * other resource history to the point that we know that the node's * connection is up, with the exception of when shutdown locks are * in use. */ pcmk_resource_t *rsc = this_node->private->remote; if ((rsc == NULL) || (!pcmk_is_set(scheduler->flags, pcmk_sched_shutdown_lock) && (rsc->private->orig_role != pcmk_role_started))) { crm_trace("Not unpacking resource history for remote node %s " "because connection is not known to be up", id); continue; } /* If fencing and shutdown locks are disabled and we're not processing * unseen nodes, then we don't want to unpack offline nodes until online * nodes have been unpacked. This allows us to number active clone * instances first. */ } else if (!pcmk_any_flags_set(scheduler->flags, pcmk_sched_fencing_enabled |pcmk_sched_shutdown_lock) && !this_node->details->online) { crm_trace("Not unpacking resource history for offline " "cluster node %s", id); continue; } if (pcmk__is_pacemaker_remote_node(this_node)) { determine_remote_online_status(scheduler, this_node); unpack_handle_remote_attrs(this_node, state, scheduler); } crm_trace("Unpacking resource history for %snode %s", (fence? "unseen " : ""), id); pcmk__set_node_flags(this_node, pcmk__node_unpacked); unpack_node_lrm(this_node, state, scheduler); rc = EAGAIN; // Other node histories might depend on this one } return rc; } /* remove nodes that are down, stopping */ /* create positive rsc_to_node constraints between resources and the nodes they are running on */ /* anything else? */ gboolean unpack_status(xmlNode *status, pcmk_scheduler_t *scheduler) { xmlNode *state = NULL; crm_trace("Beginning unpack"); if (scheduler->tickets == NULL) { scheduler->tickets = pcmk__strkey_table(free, destroy_ticket); } for (state = pcmk__xe_first_child(status, NULL, NULL, NULL); state != NULL; state = pcmk__xe_next(state)) { if (pcmk__xe_is(state, PCMK_XE_TICKETS)) { unpack_tickets_state((xmlNode *) state, scheduler); } else if (pcmk__xe_is(state, PCMK__XE_NODE_STATE)) { unpack_node_state(state, scheduler); } } while (unpack_node_history(status, FALSE, scheduler) == EAGAIN) { crm_trace("Another pass through node resource histories is needed"); } // Now catch any nodes we didn't see unpack_node_history(status, pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled), scheduler); /* Now that we know where resources are, we can schedule stops of containers * with failed bundle connections */ if (scheduler->stop_needed != NULL) { for (GList *item = scheduler->stop_needed; item; item = item->next) { pcmk_resource_t *container = item->data; pcmk_node_t *node = pcmk__current_node(container); if (node) { stop_action(container, node, FALSE); } } g_list_free(scheduler->stop_needed); scheduler->stop_needed = NULL; } /* Now that we know status of all Pacemaker Remote connections and nodes, * we can stop connections for node shutdowns, and check the online status * of remote/guest nodes that didn't have any node history to unpack. */ for (GList *gIter = scheduler->nodes; gIter != NULL; gIter = gIter->next) { pcmk_node_t *this_node = gIter->data; if (!pcmk__is_pacemaker_remote_node(this_node)) { continue; } if (this_node->details->shutdown && (this_node->private->remote != NULL)) { pe__set_next_role(this_node->private->remote, pcmk_role_stopped, "remote shutdown"); } if (!pcmk_is_set(this_node->private->flags, pcmk__node_unpacked)) { determine_remote_online_status(scheduler, this_node); } } return TRUE; } /*! * \internal * \brief Unpack node's time when it became a member at the cluster layer * * \param[in] node_state Node's \c PCMK__XE_NODE_STATE entry * \param[in,out] scheduler Scheduler data * * \return Epoch time when node became a cluster member * (or scheduler effective time for legacy entries) if a member, * 0 if not a member, or -1 if no valid information available */ static long long unpack_node_member(const xmlNode *node_state, pcmk_scheduler_t *scheduler) { const char *member_time = crm_element_value(node_state, PCMK__XA_IN_CCM); int member = 0; if (member_time == NULL) { return -1LL; } else if (crm_str_to_boolean(member_time, &member) == 1) { /* If in_ccm=0, we'll return 0 here. If in_ccm=1, either the entry was * recorded as a boolean for a DC < 2.1.7, or the node is pending * shutdown and has left the CPG, in which case it was set to 1 to avoid * fencing for PCMK_OPT_NODE_PENDING_TIMEOUT. * * We return the effective time for in_ccm=1 because what's important to * avoid fencing is that effective time minus this value is less than * the pending node timeout. */ return member? (long long) get_effective_time(scheduler) : 0LL; } else { long long when_member = 0LL; if ((pcmk__scan_ll(member_time, &when_member, 0LL) != pcmk_rc_ok) || (when_member < 0LL)) { crm_warn("Unrecognized value '%s' for " PCMK__XA_IN_CCM " in " PCMK__XE_NODE_STATE " entry", member_time); return -1LL; } return when_member; } } /*! * \internal * \brief Unpack node's time when it became online in process group * * \param[in] node_state Node's \c PCMK__XE_NODE_STATE entry * * \return Epoch time when node became online in process group (or 0 if not * online, or 1 for legacy online entries) */ static long long unpack_node_online(const xmlNode *node_state) { const char *peer_time = crm_element_value(node_state, PCMK_XA_CRMD); // @COMPAT Entries recorded for DCs < 2.1.7 have "online" or "offline" if (pcmk__str_eq(peer_time, PCMK_VALUE_OFFLINE, pcmk__str_casei|pcmk__str_null_matches)) { return 0LL; } else if (pcmk__str_eq(peer_time, PCMK_VALUE_ONLINE, pcmk__str_casei)) { return 1LL; } else { long long when_online = 0LL; if ((pcmk__scan_ll(peer_time, &when_online, 0LL) != pcmk_rc_ok) || (when_online < 0)) { crm_warn("Unrecognized value '%s' for " PCMK_XA_CRMD " in " PCMK__XE_NODE_STATE " entry, assuming offline", peer_time); return 0LL; } return when_online; } } /*! * \internal * \brief Unpack node attribute for user-requested fencing * * \param[in] node Node to check * \param[in] node_state Node's \c PCMK__XE_NODE_STATE entry in CIB status * * \return \c true if fencing has been requested for \p node, otherwise \c false */ static bool unpack_node_terminate(const pcmk_node_t *node, const xmlNode *node_state) { long long value = 0LL; int value_i = 0; const char *value_s = pcmk__node_attr(node, PCMK_NODE_ATTR_TERMINATE, NULL, pcmk__rsc_node_current); // Value may be boolean or an epoch time if (crm_str_to_boolean(value_s, &value_i) == 1) { return (value_i != 0); } if (pcmk__scan_ll(value_s, &value, 0LL) == pcmk_rc_ok) { return (value > 0); } crm_warn("Ignoring unrecognized value '%s' for " PCMK_NODE_ATTR_TERMINATE "node attribute for %s", value_s, pcmk__node_name(node)); return false; } static gboolean determine_online_status_no_fencing(pcmk_scheduler_t *scheduler, const xmlNode *node_state, pcmk_node_t *this_node) { gboolean online = FALSE; const char *join = crm_element_value(node_state, PCMK__XA_JOIN); const char *exp_state = crm_element_value(node_state, PCMK_XA_EXPECTED); long long when_member = unpack_node_member(node_state, scheduler); long long when_online = unpack_node_online(node_state); if (when_member <= 0) { crm_trace("Node %s is %sdown", pcmk__node_name(this_node), ((when_member < 0)? "presumed " : "")); } else if (when_online > 0) { if (pcmk__str_eq(join, CRMD_JOINSTATE_MEMBER, pcmk__str_casei)) { online = TRUE; } else { crm_debug("Node %s is not ready to run resources: %s", pcmk__node_name(this_node), join); } } else if (!pcmk_is_set(this_node->private->flags, pcmk__node_expected_up)) { crm_trace("Node %s controller is down: " "member@%lld online@%lld join=%s expected=%s", pcmk__node_name(this_node), when_member, when_online, pcmk__s(join, ""), pcmk__s(exp_state, "")); } else { /* mark it unclean */ pe_fence_node(scheduler, this_node, "peer is unexpectedly down", FALSE); crm_info("Node %s member@%lld online@%lld join=%s expected=%s", pcmk__node_name(this_node), when_member, when_online, pcmk__s(join, ""), pcmk__s(exp_state, "")); } return online; } /*! * \internal * \brief Check whether a node has taken too long to join controller group * * \param[in,out] scheduler Scheduler data * \param[in] node Node to check * \param[in] when_member Epoch time when node became a cluster member * \param[in] when_online Epoch time when node joined controller group * * \return true if node has been pending (on the way up) longer than * \c PCMK_OPT_NODE_PENDING_TIMEOUT, otherwise false * \note This will also update the cluster's recheck time if appropriate. */ static inline bool pending_too_long(pcmk_scheduler_t *scheduler, const pcmk_node_t *node, long long when_member, long long when_online) { if ((scheduler->node_pending_timeout > 0) && (when_member > 0) && (when_online <= 0)) { // There is a timeout on pending nodes, and node is pending time_t timeout = when_member + scheduler->node_pending_timeout; if (get_effective_time(node->private->scheduler) >= timeout) { return true; // Node has timed out } // Node is pending, but still has time pe__update_recheck_time(timeout, scheduler, "pending node timeout"); } return false; } static bool determine_online_status_fencing(pcmk_scheduler_t *scheduler, const xmlNode *node_state, pcmk_node_t *this_node) { bool termination_requested = unpack_node_terminate(this_node, node_state); const char *join = crm_element_value(node_state, PCMK__XA_JOIN); const char *exp_state = crm_element_value(node_state, PCMK_XA_EXPECTED); long long when_member = unpack_node_member(node_state, scheduler); long long when_online = unpack_node_online(node_state); /* - PCMK__XA_JOIN ::= member|down|pending|banned - PCMK_XA_EXPECTED ::= member|down @COMPAT with entries recorded for DCs < 2.1.7 - PCMK__XA_IN_CCM ::= true|false - PCMK_XA_CRMD ::= online|offline Since crm_feature_set 3.18.0 (pacemaker-2.1.7): - PCMK__XA_IN_CCM ::= |0 Since when node has been a cluster member. A value 0 of means the node is not a cluster member. - PCMK_XA_CRMD ::= |0 Since when peer has been online in CPG. A value 0 means the peer is offline in CPG. */ crm_trace("Node %s member@%lld online@%lld join=%s expected=%s%s", pcmk__node_name(this_node), when_member, when_online, pcmk__s(join, ""), pcmk__s(exp_state, ""), (termination_requested? " (termination requested)" : "")); if (this_node->details->shutdown) { crm_debug("%s is shutting down", pcmk__node_name(this_node)); /* Slightly different criteria since we can't shut down a dead peer */ return (when_online > 0); } if (when_member < 0) { pe_fence_node(scheduler, this_node, "peer has not been seen by the cluster", FALSE); return false; } if (pcmk__str_eq(join, CRMD_JOINSTATE_NACK, pcmk__str_none)) { pe_fence_node(scheduler, this_node, "peer failed Pacemaker membership criteria", FALSE); } else if (termination_requested) { if ((when_member <= 0) && (when_online <= 0) && pcmk__str_eq(join, CRMD_JOINSTATE_DOWN, pcmk__str_none)) { crm_info("%s was fenced as requested", pcmk__node_name(this_node)); return false; } pe_fence_node(scheduler, this_node, "fencing was requested", false); } else if (pcmk__str_eq(exp_state, CRMD_JOINSTATE_DOWN, pcmk__str_null_matches)) { if (pending_too_long(scheduler, this_node, when_member, when_online)) { pe_fence_node(scheduler, this_node, "peer pending timed out on joining the process group", FALSE); } else if ((when_member > 0) || (when_online > 0)) { crm_info("- %s is not ready to run resources", pcmk__node_name(this_node)); pcmk__set_node_flags(this_node, pcmk__node_standby); this_node->details->pending = TRUE; } else { crm_trace("%s is down or still coming up", pcmk__node_name(this_node)); } } else if (when_member <= 0) { // Consider PCMK_OPT_PRIORITY_FENCING_DELAY for lost nodes pe_fence_node(scheduler, this_node, "peer is no longer part of the cluster", TRUE); } else if (when_online <= 0) { pe_fence_node(scheduler, this_node, "peer process is no longer available", FALSE); /* Everything is running at this point, now check join state */ } else if (pcmk__str_eq(join, CRMD_JOINSTATE_MEMBER, pcmk__str_none)) { crm_info("%s is active", pcmk__node_name(this_node)); } else if (pcmk__str_any_of(join, CRMD_JOINSTATE_PENDING, CRMD_JOINSTATE_DOWN, NULL)) { crm_info("%s is not ready to run resources", pcmk__node_name(this_node)); pcmk__set_node_flags(this_node, pcmk__node_standby); this_node->details->pending = TRUE; } else { pe_fence_node(scheduler, this_node, "peer was in an unknown state", FALSE); } return (when_member > 0); } static void determine_remote_online_status(pcmk_scheduler_t *scheduler, pcmk_node_t *this_node) { pcmk_resource_t *rsc = this_node->private->remote; pcmk_resource_t *launcher = NULL; pcmk_node_t *host = NULL; const char *node_type = "Remote"; if (rsc == NULL) { /* This is a leftover node state entry for a former Pacemaker Remote * node whose connection resource was removed. Consider it offline. */ crm_trace("Pacemaker Remote node %s is considered OFFLINE because " "its connection resource has been removed from the CIB", this_node->private->id); this_node->details->online = FALSE; return; } launcher = rsc->private->launcher; if (launcher != NULL) { node_type = "Guest"; if (pcmk__list_of_1(rsc->private->active_nodes)) { host = rsc->private->active_nodes->data; } } /* If the resource is currently started, mark it online. */ if (rsc->private->orig_role == pcmk_role_started) { this_node->details->online = TRUE; } /* consider this node shutting down if transitioning start->stop */ if ((rsc->private->orig_role == pcmk_role_started) && (rsc->private->next_role == pcmk_role_stopped)) { crm_trace("%s node %s shutting down because connection resource is stopping", node_type, this_node->private->id); this_node->details->shutdown = TRUE; } /* Now check all the failure conditions. */ if ((launcher != NULL) && pcmk_is_set(launcher->flags, pcmk__rsc_failed)) { crm_trace("Guest node %s UNCLEAN because guest resource failed", this_node->private->id); this_node->details->online = FALSE; pcmk__set_node_flags(this_node, pcmk__node_remote_reset); } else if (pcmk_is_set(rsc->flags, pcmk__rsc_failed)) { crm_trace("%s node %s OFFLINE because connection resource failed", node_type, this_node->private->id); this_node->details->online = FALSE; } else if ((rsc->private->orig_role == pcmk_role_stopped) || ((launcher != NULL) && (launcher->private->orig_role == pcmk_role_stopped))) { crm_trace("%s node %s OFFLINE because its resource is stopped", node_type, this_node->private->id); this_node->details->online = FALSE; pcmk__clear_node_flags(this_node, pcmk__node_remote_reset); } else if (host && (host->details->online == FALSE) && host->details->unclean) { crm_trace("Guest node %s UNCLEAN because host is unclean", this_node->private->id); this_node->details->online = FALSE; pcmk__set_node_flags(this_node, pcmk__node_remote_reset); } else { crm_trace("%s node %s is %s", node_type, this_node->private->id, this_node->details->online? "ONLINE" : "OFFLINE"); } } static void determine_online_status(const xmlNode *node_state, pcmk_node_t *this_node, pcmk_scheduler_t *scheduler) { gboolean online = FALSE; const char *exp_state = crm_element_value(node_state, PCMK_XA_EXPECTED); CRM_CHECK(this_node != NULL, return); this_node->details->shutdown = FALSE; if (pe__shutdown_requested(this_node)) { this_node->details->shutdown = TRUE; } else if (pcmk__str_eq(exp_state, CRMD_JOINSTATE_MEMBER, pcmk__str_casei)) { pcmk__set_node_flags(this_node, pcmk__node_expected_up); } if (this_node->private->variant == pcmk__node_variant_ping) { this_node->details->unclean = FALSE; online = FALSE; /* As far as resource management is concerned, * the node is safely offline. * Anyone caught abusing this logic will be shot */ } else if (!pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled)) { online = determine_online_status_no_fencing(scheduler, node_state, this_node); } else { online = determine_online_status_fencing(scheduler, node_state, this_node); } if (online) { this_node->details->online = TRUE; } else { /* remove node from contention */ this_node->assign->score = -PCMK_SCORE_INFINITY; } if (online && this_node->details->shutdown) { /* don't run resources here */ this_node->assign->score = -PCMK_SCORE_INFINITY; } if (this_node->private->variant == pcmk__node_variant_ping) { crm_info("%s is not a Pacemaker node", pcmk__node_name(this_node)); } else if (this_node->details->unclean) { pcmk__sched_warn("%s is unclean", pcmk__node_name(this_node)); } else if (!this_node->details->online) { crm_trace("%s is offline", pcmk__node_name(this_node)); } else if (this_node->details->shutdown) { crm_info("%s is shutting down", pcmk__node_name(this_node)); } else if (this_node->details->pending) { crm_info("%s is pending", pcmk__node_name(this_node)); } else if (pcmk_is_set(this_node->private->flags, pcmk__node_standby)) { crm_info("%s is in standby", pcmk__node_name(this_node)); } else if (this_node->details->maintenance) { crm_info("%s is in maintenance", pcmk__node_name(this_node)); } else { crm_info("%s is online", pcmk__node_name(this_node)); } } /*! * \internal * \brief Find the end of a resource's name, excluding any clone suffix * * \param[in] id Resource ID to check * * \return Pointer to last character of resource's base name */ const char * pe_base_name_end(const char *id) { if (!pcmk__str_empty(id)) { const char *end = id + strlen(id) - 1; for (const char *s = end; s > id; --s) { switch (*s) { case '0': case '1': case '2': case '3': case '4': case '5': case '6': case '7': case '8': case '9': break; case ':': return (s == end)? s : (s - 1); default: return end; } } return end; } return NULL; } /*! * \internal * \brief Get a resource name excluding any clone suffix * * \param[in] last_rsc_id Resource ID to check * * \return Pointer to newly allocated string with resource's base name * \note It is the caller's responsibility to free() the result. * This asserts on error, so callers can assume result is not NULL. */ char * clone_strip(const char *last_rsc_id) { const char *end = pe_base_name_end(last_rsc_id); char *basename = NULL; CRM_ASSERT(end); basename = strndup(last_rsc_id, end - last_rsc_id + 1); CRM_ASSERT(basename); return basename; } /*! * \internal * \brief Get the name of the first instance of a cloned resource * * \param[in] last_rsc_id Resource ID to check * * \return Pointer to newly allocated string with resource's base name plus :0 * \note It is the caller's responsibility to free() the result. * This asserts on error, so callers can assume result is not NULL. */ char * clone_zero(const char *last_rsc_id) { const char *end = pe_base_name_end(last_rsc_id); size_t base_name_len = end - last_rsc_id + 1; char *zero = NULL; CRM_ASSERT(end); zero = pcmk__assert_alloc(base_name_len + 3, sizeof(char)); memcpy(zero, last_rsc_id, base_name_len); zero[base_name_len] = ':'; zero[base_name_len + 1] = '0'; return zero; } static pcmk_resource_t * create_fake_resource(const char *rsc_id, const xmlNode *rsc_entry, pcmk_scheduler_t *scheduler) { pcmk_resource_t *rsc = NULL; xmlNode *xml_rsc = pcmk__xe_create(NULL, PCMK_XE_PRIMITIVE); pcmk__xe_copy_attrs(xml_rsc, rsc_entry, pcmk__xaf_none); crm_xml_add(xml_rsc, PCMK_XA_ID, rsc_id); crm_log_xml_debug(xml_rsc, "Orphan resource"); if (pe__unpack_resource(xml_rsc, &rsc, NULL, scheduler) != pcmk_rc_ok) { return NULL; } if (xml_contains_remote_node(xml_rsc)) { pcmk_node_t *node; crm_debug("Detected orphaned remote node %s", rsc_id); node = pcmk_find_node(scheduler, rsc_id); if (node == NULL) { node = pe_create_node(rsc_id, rsc_id, PCMK_VALUE_REMOTE, NULL, scheduler); } link_rsc2remotenode(scheduler, rsc); if (node) { crm_trace("Setting node %s as shutting down due to orphaned connection resource", rsc_id); node->details->shutdown = TRUE; } } if (crm_element_value(rsc_entry, PCMK__META_CONTAINER)) { // This removed resource needs to be mapped to a launcher crm_trace("Launched resource %s was removed from the configuration", rsc_id); pcmk__set_rsc_flags(rsc, pcmk__rsc_removed_launched); } pcmk__set_rsc_flags(rsc, pcmk__rsc_removed); scheduler->resources = g_list_append(scheduler->resources, rsc); return rsc; } /*! * \internal * \brief Create orphan instance for anonymous clone resource history * * \param[in,out] parent Clone resource that orphan will be added to * \param[in] rsc_id Orphan's resource ID * \param[in] node Where orphan is active (for logging only) * \param[in,out] scheduler Scheduler data * * \return Newly added orphaned instance of \p parent */ static pcmk_resource_t * create_anonymous_orphan(pcmk_resource_t *parent, const char *rsc_id, const pcmk_node_t *node, pcmk_scheduler_t *scheduler) { pcmk_resource_t *top = pe__create_clone_child(parent, scheduler); pcmk_resource_t *orphan = NULL; // find_rsc() because we might be a cloned group orphan = top->private->fns->find_rsc(top, rsc_id, NULL, pcmk_rsc_match_clone_only); pcmk__rsc_debug(parent, "Created orphan %s for %s: %s on %s", top->id, parent->id, rsc_id, pcmk__node_name(node)); return orphan; } /*! * \internal * \brief Check a node for an instance of an anonymous clone * * Return a child instance of the specified anonymous clone, in order of * preference: (1) the instance running on the specified node, if any; * (2) an inactive instance (i.e. within the total of \c PCMK_META_CLONE_MAX * instances); (3) a newly created orphan (that is, \c PCMK_META_CLONE_MAX * instances are already active). * * \param[in,out] scheduler Scheduler data * \param[in] node Node on which to check for instance * \param[in,out] parent Clone to check * \param[in] rsc_id Name of cloned resource in history (no instance) */ static pcmk_resource_t * find_anonymous_clone(pcmk_scheduler_t *scheduler, const pcmk_node_t *node, pcmk_resource_t *parent, const char *rsc_id) { GList *rIter = NULL; pcmk_resource_t *rsc = NULL; pcmk_resource_t *inactive_instance = NULL; gboolean skip_inactive = FALSE; CRM_ASSERT(pcmk__is_anonymous_clone(parent)); // Check for active (or partially active, for cloned groups) instance pcmk__rsc_trace(parent, "Looking for %s on %s in %s", rsc_id, pcmk__node_name(node), parent->id); for (rIter = parent->private->children; (rIter != NULL) && (rsc == NULL); rIter = rIter->next) { GList *locations = NULL; pcmk_resource_t *child = rIter->data; /* Check whether this instance is already known to be active or pending * anywhere, at this stage of unpacking. Because this function is called * for a resource before the resource's individual operation history * entries are unpacked, locations will generally not contain the * desired node. * * However, there are three exceptions: * (1) when child is a cloned group and we have already unpacked the * history of another member of the group on the same node; * (2) when we've already unpacked the history of another numbered * instance on the same node (which can happen if * PCMK_META_GLOBALLY_UNIQUE was flipped from true to false); and * (3) when we re-run calculations on the same scheduler data as part of * a simulation. */ child->private->fns->location(child, &locations, 2); if (locations) { /* We should never associate the same numbered anonymous clone * instance with multiple nodes, and clone instances can't migrate, * so there must be only one location, regardless of history. */ CRM_LOG_ASSERT(locations->next == NULL); if (pcmk__same_node((pcmk_node_t *) locations->data, node)) { /* This child instance is active on the requested node, so check * for a corresponding configured resource. We use find_rsc() * instead of child because child may be a cloned group, and we * need the particular member corresponding to rsc_id. * * If the history entry is orphaned, rsc will be NULL. */ rsc = parent->private->fns->find_rsc(child, rsc_id, NULL, pcmk_rsc_match_clone_only); if (rsc) { /* If there are multiple instance history entries for an * anonymous clone in a single node's history (which can * happen if PCMK_META_GLOBALLY_UNIQUE is switched from true * to false), we want to consider the instances beyond the * first as orphans, even if there are inactive instance * numbers available. */ if (rsc->private->active_nodes != NULL) { crm_notice("Active (now-)anonymous clone %s has " "multiple (orphan) instance histories on %s", parent->id, pcmk__node_name(node)); skip_inactive = TRUE; rsc = NULL; } else { pcmk__rsc_trace(parent, "Resource %s, active", rsc->id); } } } g_list_free(locations); } else { pcmk__rsc_trace(parent, "Resource %s, skip inactive", child->id); if (!skip_inactive && !inactive_instance && !pcmk_is_set(child->flags, pcmk__rsc_blocked)) { // Remember one inactive instance in case we don't find active inactive_instance = parent->private->fns->find_rsc(child, rsc_id, NULL, pcmk_rsc_match_clone_only); /* ... but don't use it if it was already associated with a * pending action on another node */ if (inactive_instance != NULL) { const pcmk_node_t *pending_node = NULL; pending_node = inactive_instance->private->pending_node; if ((pending_node != NULL) && !pcmk__same_node(pending_node, node)) { inactive_instance = NULL; } } } } } if ((rsc == NULL) && !skip_inactive && (inactive_instance != NULL)) { pcmk__rsc_trace(parent, "Resource %s, empty slot", inactive_instance->id); rsc = inactive_instance; } /* If the resource has PCMK_META_REQUIRES set to PCMK_VALUE_QUORUM or * PCMK_VALUE_NOTHING, and we don't have a clone instance for every node, we * don't want to consume a valid instance number for unclean nodes. Such * instances may appear to be active according to the history, but should be * considered inactive, so we can start an instance elsewhere. Treat such * instances as orphans. * * An exception is instances running on guest nodes -- since guest node * "fencing" is actually just a resource stop, requires shouldn't apply. * * @TODO Ideally, we'd use an inactive instance number if it is not needed * for any clean instances. However, we don't know that at this point. */ if ((rsc != NULL) && !pcmk_is_set(rsc->flags, pcmk__rsc_needs_fencing) && (!node->details->online || node->details->unclean) && !pcmk__is_guest_or_bundle_node(node) && !pe__is_universal_clone(parent, scheduler)) { rsc = NULL; } if (rsc == NULL) { rsc = create_anonymous_orphan(parent, rsc_id, node, scheduler); pcmk__rsc_trace(parent, "Resource %s, orphan", rsc->id); } return rsc; } static pcmk_resource_t * unpack_find_resource(pcmk_scheduler_t *scheduler, const pcmk_node_t *node, const char *rsc_id) { pcmk_resource_t *rsc = NULL; pcmk_resource_t *parent = NULL; crm_trace("looking for %s", rsc_id); rsc = pe_find_resource(scheduler->resources, rsc_id); if (rsc == NULL) { /* If we didn't find the resource by its name in the operation history, * check it again as a clone instance. Even when PCMK_META_CLONE_MAX=0, * we create a single :0 orphan to match against here. */ char *clone0_id = clone_zero(rsc_id); pcmk_resource_t *clone0 = pe_find_resource(scheduler->resources, clone0_id); if (clone0 && !pcmk_is_set(clone0->flags, pcmk__rsc_unique)) { rsc = clone0; parent = uber_parent(clone0); crm_trace("%s found as %s (%s)", rsc_id, clone0_id, parent->id); } else { crm_trace("%s is not known as %s either (orphan)", rsc_id, clone0_id); } free(clone0_id); } else if (rsc->private->variant > pcmk__rsc_variant_primitive) { crm_trace("Resource history for %s is orphaned " "because it is no longer primitive", rsc_id); return NULL; } else { parent = uber_parent(rsc); } if (pcmk__is_anonymous_clone(parent)) { if (pcmk__is_bundled(parent)) { rsc = pe__find_bundle_replica(parent->private->parent, node); } else { char *base = clone_strip(rsc_id); rsc = find_anonymous_clone(scheduler, node, parent, base); free(base); CRM_ASSERT(rsc != NULL); } } if (rsc && !pcmk__str_eq(rsc_id, rsc->id, pcmk__str_none) && !pcmk__str_eq(rsc_id, rsc->private->history_id, pcmk__str_none)) { pcmk__str_update(&(rsc->private->history_id), rsc_id); pcmk__rsc_debug(rsc, "Internally renamed %s on %s to %s%s", rsc_id, pcmk__node_name(node), rsc->id, pcmk_is_set(rsc->flags, pcmk__rsc_removed)? " (ORPHAN)" : ""); } return rsc; } static pcmk_resource_t * process_orphan_resource(const xmlNode *rsc_entry, const pcmk_node_t *node, pcmk_scheduler_t *scheduler) { pcmk_resource_t *rsc = NULL; const char *rsc_id = crm_element_value(rsc_entry, PCMK_XA_ID); crm_debug("Detected orphan resource %s on %s", rsc_id, pcmk__node_name(node)); rsc = create_fake_resource(rsc_id, rsc_entry, scheduler); if (rsc == NULL) { return NULL; } if (!pcmk_is_set(scheduler->flags, pcmk_sched_stop_removed_resources)) { pcmk__clear_rsc_flags(rsc, pcmk__rsc_managed); } else { CRM_CHECK(rsc != NULL, return NULL); pcmk__rsc_trace(rsc, "Added orphan %s", rsc->id); resource_location(rsc, NULL, -PCMK_SCORE_INFINITY, "__orphan_do_not_run__", scheduler); } return rsc; } static void process_rsc_state(pcmk_resource_t *rsc, pcmk_node_t *node, enum pcmk__on_fail on_fail) { pcmk_node_t *tmpnode = NULL; char *reason = NULL; enum pcmk__on_fail save_on_fail = pcmk__on_fail_ignore; pcmk_scheduler_t *scheduler = NULL; bool known_active = false; CRM_ASSERT(rsc); scheduler = rsc->private->scheduler; known_active = (rsc->private->orig_role > pcmk_role_stopped); pcmk__rsc_trace(rsc, "Resource %s is %s on %s: on_fail=%s", rsc->id, pcmk_role_text(rsc->private->orig_role), pcmk__node_name(node), pcmk__on_fail_text(on_fail)); /* process current state */ if (rsc->private->orig_role != pcmk_role_unknown) { pcmk_resource_t *iter = rsc; while (iter) { if (g_hash_table_lookup(iter->private->probed_nodes, node->private->id) == NULL) { pcmk_node_t *n = pe__copy_node(node); pcmk__rsc_trace(rsc, "%s (%s in history) known on %s", rsc->id, pcmk__s(rsc->private->history_id, "the same"), pcmk__node_name(n)); g_hash_table_insert(iter->private->probed_nodes, (gpointer) n->private->id, n); } if (pcmk_is_set(iter->flags, pcmk__rsc_unique)) { break; } iter = iter->private->parent; } } /* If a managed resource is believed to be running, but node is down ... */ if (known_active && !node->details->online && !node->details->maintenance && pcmk_is_set(rsc->flags, pcmk__rsc_managed)) { gboolean should_fence = FALSE; /* If this is a guest node, fence it (regardless of whether fencing is * enabled, because guest node fencing is done by recovery of the * container resource rather than by the fencer). Mark the resource * we're processing as failed. When the guest comes back up, its * operation history in the CIB will be cleared, freeing the affected * resource to run again once we are sure we know its state. */ if (pcmk__is_guest_or_bundle_node(node)) { pcmk__set_rsc_flags(rsc, pcmk__rsc_failed|pcmk__rsc_stop_if_failed); should_fence = TRUE; } else if (pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled)) { if (pcmk__is_remote_node(node) && (node->private->remote != NULL) && !pcmk_is_set(node->private->remote->flags, pcmk__rsc_failed)) { /* Setting unseen means that fencing of the remote node will * occur only if the connection resource is not going to start * somewhere. This allows connection resources on a failed * cluster node to move to another node without requiring the * remote nodes to be fenced as well. */ pcmk__clear_node_flags(node, pcmk__node_seen); reason = crm_strdup_printf("%s is active there (fencing will be" " revoked if remote connection can " "be re-established elsewhere)", rsc->id); } should_fence = TRUE; } if (should_fence) { if (reason == NULL) { reason = crm_strdup_printf("%s is thought to be active there", rsc->id); } pe_fence_node(scheduler, node, reason, FALSE); } free(reason); } /* In order to calculate priority_fencing_delay correctly, save the failure information and pass it to native_add_running(). */ save_on_fail = on_fail; if (node->details->unclean) { /* No extra processing needed * Also allows resources to be started again after a node is shot */ on_fail = pcmk__on_fail_ignore; } switch (on_fail) { case pcmk__on_fail_ignore: /* nothing to do */ break; case pcmk__on_fail_demote: pcmk__set_rsc_flags(rsc, pcmk__rsc_failed); demote_action(rsc, node, FALSE); break; case pcmk__on_fail_fence_node: /* treat it as if it is still running * but also mark the node as unclean */ reason = crm_strdup_printf("%s failed there", rsc->id); pe_fence_node(scheduler, node, reason, FALSE); free(reason); break; case pcmk__on_fail_standby_node: pcmk__set_node_flags(node, pcmk__node_standby|pcmk__node_fail_standby); break; case pcmk__on_fail_block: /* is_managed == FALSE will prevent any * actions being sent for the resource */ pcmk__clear_rsc_flags(rsc, pcmk__rsc_managed); pcmk__set_rsc_flags(rsc, pcmk__rsc_blocked); break; case pcmk__on_fail_ban: /* make sure it comes up somewhere else * or not at all */ resource_location(rsc, node, -PCMK_SCORE_INFINITY, "__action_migration_auto__", scheduler); break; case pcmk__on_fail_stop: pe__set_next_role(rsc, pcmk_role_stopped, PCMK_META_ON_FAIL "=" PCMK_VALUE_STOP); break; case pcmk__on_fail_restart: if (known_active) { pcmk__set_rsc_flags(rsc, pcmk__rsc_failed|pcmk__rsc_stop_if_failed); stop_action(rsc, node, FALSE); } break; case pcmk__on_fail_restart_container: pcmk__set_rsc_flags(rsc, pcmk__rsc_failed|pcmk__rsc_stop_if_failed); if ((rsc->private->launcher != NULL) && pcmk__is_bundled(rsc)) { /* A bundle's remote connection can run on a different node than * the bundle's container. We don't necessarily know where the * container is running yet, so remember it and add a stop * action for it later. */ scheduler->stop_needed = g_list_prepend(scheduler->stop_needed, rsc->private->launcher); } else if (rsc->private->launcher != NULL) { stop_action(rsc->private->launcher, node, FALSE); } else if (known_active) { stop_action(rsc, node, FALSE); } break; case pcmk__on_fail_reset_remote: pcmk__set_rsc_flags(rsc, pcmk__rsc_failed|pcmk__rsc_stop_if_failed); if (pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled)) { tmpnode = NULL; if (pcmk_is_set(rsc->flags, pcmk__rsc_is_remote_connection)) { tmpnode = pcmk_find_node(scheduler, rsc->id); } if (pcmk__is_remote_node(tmpnode) && !pcmk_is_set(tmpnode->private->flags, pcmk__node_remote_fenced)) { /* The remote connection resource failed in a way that * should result in fencing the remote node. */ pe_fence_node(scheduler, tmpnode, "remote connection is unrecoverable", FALSE); } } /* require the stop action regardless if fencing is occurring or not. */ if (known_active) { stop_action(rsc, node, FALSE); } /* if reconnect delay is in use, prevent the connection from exiting the * "STOPPED" role until the failure is cleared by the delay timeout. */ if (rsc->private->remote_reconnect_ms > 0U) { pe__set_next_role(rsc, pcmk_role_stopped, "remote reset"); } break; } /* Ensure a remote connection failure forces an unclean Pacemaker Remote * node to be fenced. By marking the node as seen, the failure will result * in a fencing operation regardless if we're going to attempt to reconnect * in this transition. */ if (pcmk_all_flags_set(rsc->flags, pcmk__rsc_failed|pcmk__rsc_is_remote_connection)) { tmpnode = pcmk_find_node(scheduler, rsc->id); if (tmpnode && tmpnode->details->unclean) { pcmk__set_node_flags(tmpnode, pcmk__node_seen); } } if (known_active) { if (pcmk_is_set(rsc->flags, pcmk__rsc_removed)) { if (pcmk_is_set(rsc->flags, pcmk__rsc_managed)) { crm_notice("Removed resource %s is active on %s and will be " "stopped when possible", rsc->id, pcmk__node_name(node)); } else { crm_notice("Removed resource %s must be stopped manually on %s " "because " PCMK_OPT_STOP_ORPHAN_RESOURCES " is set to false", rsc->id, pcmk__node_name(node)); } } native_add_running(rsc, node, scheduler, (save_on_fail != pcmk__on_fail_ignore)); switch (on_fail) { case pcmk__on_fail_ignore: break; case pcmk__on_fail_demote: case pcmk__on_fail_block: pcmk__set_rsc_flags(rsc, pcmk__rsc_failed); break; default: pcmk__set_rsc_flags(rsc, pcmk__rsc_failed|pcmk__rsc_stop_if_failed); break; } } else if ((rsc->private->history_id != NULL) && (strchr(rsc->private->history_id, ':') != NULL)) { /* Only do this for older status sections that included instance numbers * Otherwise stopped instances will appear as orphans */ pcmk__rsc_trace(rsc, "Clearing history ID %s for %s (stopped)", rsc->private->history_id, rsc->id); free(rsc->private->history_id); rsc->private->history_id = NULL; } else { GList *possible_matches = pe__resource_actions(rsc, node, PCMK_ACTION_STOP, FALSE); GList *gIter = possible_matches; for (; gIter != NULL; gIter = gIter->next) { pcmk_action_t *stop = (pcmk_action_t *) gIter->data; pcmk__set_action_flags(stop, pcmk__action_optional); } g_list_free(possible_matches); } /* A successful stop after migrate_to on the migration source doesn't make * the partially migrated resource stopped on the migration target. */ if ((rsc->private->orig_role == pcmk_role_stopped) && (rsc->private->active_nodes != NULL) && (rsc->private->partial_migration_target != NULL) && pcmk__same_node(rsc->private->partial_migration_source, node)) { rsc->private->orig_role = pcmk_role_started; } } /* create active recurring operations as optional */ static void process_recurring(pcmk_node_t *node, pcmk_resource_t *rsc, int start_index, int stop_index, GList *sorted_op_list, pcmk_scheduler_t *scheduler) { int counter = -1; const char *task = NULL; const char *status = NULL; GList *gIter = sorted_op_list; CRM_ASSERT(rsc); pcmk__rsc_trace(rsc, "%s: Start index %d, stop index = %d", rsc->id, start_index, stop_index); for (; gIter != NULL; gIter = gIter->next) { xmlNode *rsc_op = (xmlNode *) gIter->data; guint interval_ms = 0; char *key = NULL; const char *id = pcmk__xe_id(rsc_op); counter++; if (node->details->online == FALSE) { pcmk__rsc_trace(rsc, "Skipping %s on %s: node is offline", rsc->id, pcmk__node_name(node)); break; /* Need to check if there's a monitor for role="Stopped" */ } else if (start_index < stop_index && counter <= stop_index) { pcmk__rsc_trace(rsc, "Skipping %s on %s: resource is not active", id, pcmk__node_name(node)); continue; } else if (counter < start_index) { pcmk__rsc_trace(rsc, "Skipping %s on %s: old %d", id, pcmk__node_name(node), counter); continue; } crm_element_value_ms(rsc_op, PCMK_META_INTERVAL, &interval_ms); if (interval_ms == 0) { pcmk__rsc_trace(rsc, "Skipping %s on %s: non-recurring", id, pcmk__node_name(node)); continue; } status = crm_element_value(rsc_op, PCMK__XA_OP_STATUS); if (pcmk__str_eq(status, "-1", pcmk__str_casei)) { pcmk__rsc_trace(rsc, "Skipping %s on %s: status", id, pcmk__node_name(node)); continue; } task = crm_element_value(rsc_op, PCMK_XA_OPERATION); /* create the action */ key = pcmk__op_key(rsc->id, task, interval_ms); pcmk__rsc_trace(rsc, "Creating %s on %s", key, pcmk__node_name(node)); custom_action(rsc, key, task, node, TRUE, scheduler); } } void calculate_active_ops(const GList *sorted_op_list, int *start_index, int *stop_index) { int counter = -1; int implied_monitor_start = -1; int implied_clone_start = -1; const char *task = NULL; const char *status = NULL; *stop_index = -1; *start_index = -1; for (const GList *iter = sorted_op_list; iter != NULL; iter = iter->next) { const xmlNode *rsc_op = (const xmlNode *) iter->data; counter++; task = crm_element_value(rsc_op, PCMK_XA_OPERATION); status = crm_element_value(rsc_op, PCMK__XA_OP_STATUS); if (pcmk__str_eq(task, PCMK_ACTION_STOP, pcmk__str_casei) && pcmk__str_eq(status, "0", pcmk__str_casei)) { *stop_index = counter; } else if (pcmk__strcase_any_of(task, PCMK_ACTION_START, PCMK_ACTION_MIGRATE_FROM, NULL)) { *start_index = counter; } else if ((implied_monitor_start <= *stop_index) && pcmk__str_eq(task, PCMK_ACTION_MONITOR, pcmk__str_casei)) { const char *rc = crm_element_value(rsc_op, PCMK__XA_RC_CODE); if (pcmk__strcase_any_of(rc, "0", "8", NULL)) { implied_monitor_start = counter; } } else if (pcmk__strcase_any_of(task, PCMK_ACTION_PROMOTE, PCMK_ACTION_DEMOTE, NULL)) { implied_clone_start = counter; } } if (*start_index == -1) { if (implied_clone_start != -1) { *start_index = implied_clone_start; } else if (implied_monitor_start != -1) { *start_index = implied_monitor_start; } } } // If resource history entry has shutdown lock, remember lock node and time static void unpack_shutdown_lock(const xmlNode *rsc_entry, pcmk_resource_t *rsc, const pcmk_node_t *node, pcmk_scheduler_t *scheduler) { time_t lock_time = 0; // When lock started (i.e. node shutdown time) if ((crm_element_value_epoch(rsc_entry, PCMK_OPT_SHUTDOWN_LOCK, &lock_time) == pcmk_ok) && (lock_time != 0)) { if ((scheduler->shutdown_lock > 0) && (get_effective_time(scheduler) > (lock_time + scheduler->shutdown_lock))) { pcmk__rsc_info(rsc, "Shutdown lock for %s on %s expired", rsc->id, pcmk__node_name(node)); pe__clear_resource_history(rsc, node); } else { rsc->private->lock_node = node; rsc->private->lock_time = lock_time; } } } /*! * \internal * \brief Unpack one \c PCMK__XE_LRM_RESOURCE entry from a node's CIB status * * \param[in,out] node Node whose status is being unpacked * \param[in] rsc_entry \c PCMK__XE_LRM_RESOURCE XML being unpacked * \param[in,out] scheduler Scheduler data * * \return Resource corresponding to the entry, or NULL if no operation history */ static pcmk_resource_t * unpack_lrm_resource(pcmk_node_t *node, const xmlNode *lrm_resource, pcmk_scheduler_t *scheduler) { GList *gIter = NULL; int stop_index = -1; int start_index = -1; enum rsc_role_e req_role = pcmk_role_unknown; const char *rsc_id = pcmk__xe_id(lrm_resource); pcmk_resource_t *rsc = NULL; GList *op_list = NULL; GList *sorted_op_list = NULL; xmlNode *rsc_op = NULL; xmlNode *last_failure = NULL; enum pcmk__on_fail on_fail = pcmk__on_fail_ignore; enum rsc_role_e saved_role = pcmk_role_unknown; if (rsc_id == NULL) { pcmk__config_err("Ignoring invalid " PCMK__XE_LRM_RESOURCE " entry: No " PCMK_XA_ID); crm_log_xml_info(lrm_resource, "missing-id"); return NULL; } crm_trace("Unpacking " PCMK__XE_LRM_RESOURCE " for %s on %s", rsc_id, pcmk__node_name(node)); /* Build a list of individual PCMK__XE_LRM_RSC_OP entries, so we can sort * them */ for (rsc_op = pcmk__xe_first_child(lrm_resource, PCMK__XE_LRM_RSC_OP, NULL, NULL); rsc_op != NULL; rsc_op = pcmk__xe_next_same(rsc_op)) { op_list = g_list_prepend(op_list, rsc_op); } if (!pcmk_is_set(scheduler->flags, pcmk_sched_shutdown_lock)) { if (op_list == NULL) { // If there are no operations, there is nothing to do return NULL; } } /* find the resource */ rsc = unpack_find_resource(scheduler, node, rsc_id); if (rsc == NULL) { if (op_list == NULL) { // If there are no operations, there is nothing to do return NULL; } else { rsc = process_orphan_resource(lrm_resource, node, scheduler); } } CRM_ASSERT(rsc != NULL); // Check whether the resource is "shutdown-locked" to this node if (pcmk_is_set(scheduler->flags, pcmk_sched_shutdown_lock)) { unpack_shutdown_lock(lrm_resource, rsc, node, scheduler); } /* process operations */ saved_role = rsc->private->orig_role; rsc->private->orig_role = pcmk_role_unknown; sorted_op_list = g_list_sort(op_list, sort_op_by_callid); for (gIter = sorted_op_list; gIter != NULL; gIter = gIter->next) { xmlNode *rsc_op = (xmlNode *) gIter->data; unpack_rsc_op(rsc, node, rsc_op, &last_failure, &on_fail); } /* create active recurring operations as optional */ calculate_active_ops(sorted_op_list, &start_index, &stop_index); process_recurring(node, rsc, start_index, stop_index, sorted_op_list, scheduler); /* no need to free the contents */ g_list_free(sorted_op_list); process_rsc_state(rsc, node, on_fail); if (get_target_role(rsc, &req_role)) { if ((rsc->private->next_role == pcmk_role_unknown) || (req_role < rsc->private->next_role)) { pe__set_next_role(rsc, req_role, PCMK_META_TARGET_ROLE); } else if (req_role > rsc->private->next_role) { pcmk__rsc_info(rsc, "%s: Not overwriting calculated next role %s" " with requested next role %s", rsc->id, pcmk_role_text(rsc->private->next_role), pcmk_role_text(req_role)); } } if (saved_role > rsc->private->orig_role) { rsc->private->orig_role = saved_role; } return rsc; } static void handle_removed_launched_resources(const xmlNode *lrm_rsc_list, pcmk_scheduler_t *scheduler) { for (const xmlNode *rsc_entry = pcmk__xe_first_child(lrm_rsc_list, NULL, NULL, NULL); rsc_entry != NULL; rsc_entry = pcmk__xe_next(rsc_entry)) { pcmk_resource_t *rsc; pcmk_resource_t *launcher = NULL; const char *rsc_id; const char *launcher_id = NULL; if (!pcmk__xe_is(rsc_entry, PCMK__XE_LRM_RESOURCE)) { continue; } launcher_id = crm_element_value(rsc_entry, PCMK__META_CONTAINER); rsc_id = crm_element_value(rsc_entry, PCMK_XA_ID); if ((launcher_id == NULL) || (rsc_id == NULL)) { continue; } launcher = pe_find_resource(scheduler->resources, launcher_id); if (launcher == NULL) { continue; } rsc = pe_find_resource(scheduler->resources, rsc_id); if ((rsc == NULL) || (rsc->private->launcher != NULL) || !pcmk_is_set(rsc->flags, pcmk__rsc_removed_launched)) { continue; } pcmk__rsc_trace(rsc, "Mapped launcher of removed resource %s to %s", rsc->id, launcher_id); rsc->private->launcher = launcher; launcher->private->launched = g_list_append(launcher->private->launched, rsc); } } /*! * \internal * \brief Unpack one node's lrm status section * * \param[in,out] node Node whose status is being unpacked * \param[in] xml CIB node state XML * \param[in,out] scheduler Scheduler data */ static void unpack_node_lrm(pcmk_node_t *node, const xmlNode *xml, pcmk_scheduler_t *scheduler) { bool found_removed_launched_resource = false; // Drill down to PCMK__XE_LRM_RESOURCES section xml = pcmk__xe_first_child(xml, PCMK__XE_LRM, NULL, NULL); if (xml == NULL) { return; } xml = pcmk__xe_first_child(xml, PCMK__XE_LRM_RESOURCES, NULL, NULL); if (xml == NULL) { return; } // Unpack each PCMK__XE_LRM_RESOURCE entry for (const xmlNode *rsc_entry = pcmk__xe_first_child(xml, PCMK__XE_LRM_RESOURCE, NULL, NULL); rsc_entry != NULL; rsc_entry = pcmk__xe_next_same(rsc_entry)) { pcmk_resource_t *rsc = unpack_lrm_resource(node, rsc_entry, scheduler); if ((rsc != NULL) && pcmk_is_set(rsc->flags, pcmk__rsc_removed_launched)) { found_removed_launched_resource = true; } } /* Now that all resource state has been unpacked for this node, map any * removed launched resources to their launchers. */ if (found_removed_launched_resource) { handle_removed_launched_resources(xml, scheduler); } } static void set_active(pcmk_resource_t *rsc) { const pcmk_resource_t *top = pe__const_top_resource(rsc, false); if (top && pcmk_is_set(top->flags, pcmk__rsc_promotable)) { rsc->private->orig_role = pcmk_role_unpromoted; } else { rsc->private->orig_role = pcmk_role_started; } } static void set_node_score(gpointer key, gpointer value, gpointer user_data) { pcmk_node_t *node = value; int *score = user_data; node->assign->score = *score; } #define XPATH_NODE_STATE "/" PCMK_XE_CIB "/" PCMK_XE_STATUS \ "/" PCMK__XE_NODE_STATE #define SUB_XPATH_LRM_RESOURCE "/" PCMK__XE_LRM \ "/" PCMK__XE_LRM_RESOURCES \ "/" PCMK__XE_LRM_RESOURCE #define SUB_XPATH_LRM_RSC_OP "/" PCMK__XE_LRM_RSC_OP static xmlNode * find_lrm_op(const char *resource, const char *op, const char *node, const char *source, int target_rc, pcmk_scheduler_t *scheduler) { GString *xpath = NULL; xmlNode *xml = NULL; CRM_CHECK((resource != NULL) && (op != NULL) && (node != NULL), return NULL); xpath = g_string_sized_new(256); pcmk__g_strcat(xpath, XPATH_NODE_STATE "[@" PCMK_XA_UNAME "='", node, "']" SUB_XPATH_LRM_RESOURCE "[@" PCMK_XA_ID "='", resource, "']" SUB_XPATH_LRM_RSC_OP "[@" PCMK_XA_OPERATION "='", op, "'", NULL); /* Need to check against transition_magic too? */ if ((source != NULL) && (strcmp(op, PCMK_ACTION_MIGRATE_TO) == 0)) { pcmk__g_strcat(xpath, " and @" PCMK__META_MIGRATE_TARGET "='", source, "']", NULL); } else if ((source != NULL) && (strcmp(op, PCMK_ACTION_MIGRATE_FROM) == 0)) { pcmk__g_strcat(xpath, " and @" PCMK__META_MIGRATE_SOURCE "='", source, "']", NULL); } else { g_string_append_c(xpath, ']'); } xml = get_xpath_object((const char *) xpath->str, scheduler->input, LOG_DEBUG); g_string_free(xpath, TRUE); if (xml && target_rc >= 0) { int rc = PCMK_OCF_UNKNOWN_ERROR; int status = PCMK_EXEC_ERROR; crm_element_value_int(xml, PCMK__XA_RC_CODE, &rc); crm_element_value_int(xml, PCMK__XA_OP_STATUS, &status); if ((rc != target_rc) || (status != PCMK_EXEC_DONE)) { return NULL; } } return xml; } static xmlNode * find_lrm_resource(const char *rsc_id, const char *node_name, pcmk_scheduler_t *scheduler) { GString *xpath = NULL; xmlNode *xml = NULL; CRM_CHECK((rsc_id != NULL) && (node_name != NULL), return NULL); xpath = g_string_sized_new(256); pcmk__g_strcat(xpath, XPATH_NODE_STATE "[@" PCMK_XA_UNAME "='", node_name, "']" SUB_XPATH_LRM_RESOURCE "[@" PCMK_XA_ID "='", rsc_id, "']", NULL); xml = get_xpath_object((const char *) xpath->str, scheduler->input, LOG_DEBUG); g_string_free(xpath, TRUE); return xml; } /*! * \internal * \brief Check whether a resource has no completed action history on a node * * \param[in,out] rsc Resource to check * \param[in] node_name Node to check * * \return true if \p rsc_id is unknown on \p node_name, otherwise false */ static bool unknown_on_node(pcmk_resource_t *rsc, const char *node_name) { bool result = false; xmlXPathObjectPtr search; char *xpath = NULL; xpath = crm_strdup_printf(XPATH_NODE_STATE "[@" PCMK_XA_UNAME "='%s']" SUB_XPATH_LRM_RESOURCE "[@" PCMK_XA_ID "='%s']" SUB_XPATH_LRM_RSC_OP "[@" PCMK__XA_RC_CODE "!='%d']", node_name, rsc->id, PCMK_OCF_UNKNOWN); search = xpath_search(rsc->private->scheduler->input, xpath); result = (numXpathResults(search) == 0); freeXpathObject(search); free(xpath); return result; } /*! * \internal * \brief Check whether a probe/monitor indicating the resource was not running * on a node happened after some event * * \param[in] rsc_id Resource being checked * \param[in] node_name Node being checked * \param[in] xml_op Event that monitor is being compared to * \param[in,out] scheduler Scheduler data * * \return true if such a monitor happened after event, false otherwise */ static bool monitor_not_running_after(const char *rsc_id, const char *node_name, const xmlNode *xml_op, pcmk_scheduler_t *scheduler) { /* Any probe/monitor operation on the node indicating it was not running * there */ xmlNode *monitor = find_lrm_op(rsc_id, PCMK_ACTION_MONITOR, node_name, NULL, PCMK_OCF_NOT_RUNNING, scheduler); return (monitor != NULL) && (pe__is_newer_op(monitor, xml_op) > 0); } /*! * \internal * \brief Check whether any non-monitor operation on a node happened after some * event * * \param[in] rsc_id Resource being checked * \param[in] node_name Node being checked * \param[in] xml_op Event that non-monitor is being compared to * \param[in,out] scheduler Scheduler data * * \return true if such a operation happened after event, false otherwise */ static bool non_monitor_after(const char *rsc_id, const char *node_name, const xmlNode *xml_op, pcmk_scheduler_t *scheduler) { xmlNode *lrm_resource = NULL; lrm_resource = find_lrm_resource(rsc_id, node_name, scheduler); if (lrm_resource == NULL) { return false; } for (xmlNode *op = pcmk__xe_first_child(lrm_resource, PCMK__XE_LRM_RSC_OP, NULL, NULL); op != NULL; op = pcmk__xe_next_same(op)) { const char * task = NULL; if (op == xml_op) { continue; } task = crm_element_value(op, PCMK_XA_OPERATION); if (pcmk__str_any_of(task, PCMK_ACTION_START, PCMK_ACTION_STOP, PCMK_ACTION_MIGRATE_TO, PCMK_ACTION_MIGRATE_FROM, NULL) && pe__is_newer_op(op, xml_op) > 0) { return true; } } return false; } /*! * \internal * \brief Check whether the resource has newer state on a node after a migration * attempt * * \param[in] rsc_id Resource being checked * \param[in] node_name Node being checked * \param[in] migrate_to Any migrate_to event that is being compared to * \param[in] migrate_from Any migrate_from event that is being compared to * \param[in,out] scheduler Scheduler data * * \return true if such a operation happened after event, false otherwise */ static bool newer_state_after_migrate(const char *rsc_id, const char *node_name, const xmlNode *migrate_to, const xmlNode *migrate_from, pcmk_scheduler_t *scheduler) { const xmlNode *xml_op = (migrate_from != NULL)? migrate_from : migrate_to; const char *source = crm_element_value(xml_op, PCMK__META_MIGRATE_SOURCE); /* It's preferred to compare to the migrate event on the same node if * existing, since call ids are more reliable. */ if ((xml_op != migrate_to) && (migrate_to != NULL) && pcmk__str_eq(node_name, source, pcmk__str_casei)) { xml_op = migrate_to; } /* If there's any newer non-monitor operation on the node, or any newer * probe/monitor operation on the node indicating it was not running there, * the migration events potentially no longer matter for the node. */ return non_monitor_after(rsc_id, node_name, xml_op, scheduler) || monitor_not_running_after(rsc_id, node_name, xml_op, scheduler); } /*! * \internal * \brief Parse migration source and target node names from history entry * * \param[in] entry Resource history entry for a migration action * \param[in] source_node If not NULL, source must match this node * \param[in] target_node If not NULL, target must match this node * \param[out] source_name Where to store migration source node name * \param[out] target_name Where to store migration target node name * * \return Standard Pacemaker return code */ static int get_migration_node_names(const xmlNode *entry, const pcmk_node_t *source_node, const pcmk_node_t *target_node, const char **source_name, const char **target_name) { *source_name = crm_element_value(entry, PCMK__META_MIGRATE_SOURCE); *target_name = crm_element_value(entry, PCMK__META_MIGRATE_TARGET); if ((*source_name == NULL) || (*target_name == NULL)) { pcmk__config_err("Ignoring resource history entry %s without " PCMK__META_MIGRATE_SOURCE " and " PCMK__META_MIGRATE_TARGET, pcmk__xe_id(entry)); return pcmk_rc_unpack_error; } if ((source_node != NULL) && !pcmk__str_eq(*source_name, source_node->private->name, pcmk__str_casei|pcmk__str_null_matches)) { pcmk__config_err("Ignoring resource history entry %s because " PCMK__META_MIGRATE_SOURCE "='%s' does not match %s", pcmk__xe_id(entry), *source_name, pcmk__node_name(source_node)); return pcmk_rc_unpack_error; } if ((target_node != NULL) && !pcmk__str_eq(*target_name, target_node->private->name, pcmk__str_casei|pcmk__str_null_matches)) { pcmk__config_err("Ignoring resource history entry %s because " PCMK__META_MIGRATE_TARGET "='%s' does not match %s", pcmk__xe_id(entry), *target_name, pcmk__node_name(target_node)); return pcmk_rc_unpack_error; } return pcmk_rc_ok; } /* * \internal * \brief Add a migration source to a resource's list of dangling migrations * * If the migrate_to and migrate_from actions in a live migration both * succeeded, but there is no stop on the source, the migration is considered * "dangling." Add the source to the resource's dangling migration list, which * will be used to schedule a stop on the source without affecting the target. * * \param[in,out] rsc Resource involved in migration * \param[in] node Migration source */ static void add_dangling_migration(pcmk_resource_t *rsc, const pcmk_node_t *node) { pcmk__rsc_trace(rsc, "Dangling migration of %s requires stop on %s", rsc->id, pcmk__node_name(node)); rsc->private->orig_role = pcmk_role_stopped; rsc->private->dangling_migration_sources = g_list_prepend(rsc->private->dangling_migration_sources, (gpointer) node); } /*! * \internal * \brief Update resource role etc. after a successful migrate_to action * * \param[in,out] history Parsed action result history */ static void unpack_migrate_to_success(struct action_history *history) { /* A complete migration sequence is: * 1. migrate_to on source node (which succeeded if we get to this function) * 2. migrate_from on target node * 3. stop on source node * * If no migrate_from has happened, the migration is considered to be * "partial". If the migrate_from succeeded but no stop has happened, the * migration is considered to be "dangling". * * If a successful migrate_to and stop have happened on the source node, we * still need to check for a partial migration, due to scenarios (easier to * produce with batch-limit=1) like: * * - A resource is migrating from node1 to node2, and a migrate_to is * initiated for it on node1. * * - node2 goes into standby mode while the migrate_to is pending, which * aborts the transition. * * - Upon completion of the migrate_to, a new transition schedules a stop * on both nodes and a start on node1. * * - If the new transition is aborted for any reason while the resource is * stopping on node1, the transition after that stop completes will see * the migrate_to and stop on the source, but it's still a partial * migration, and the resource must be stopped on node2 because it is * potentially active there due to the migrate_to. * * We also need to take into account that either node's history may be * cleared at any point in the migration process. */ int from_rc = PCMK_OCF_OK; int from_status = PCMK_EXEC_PENDING; pcmk_node_t *target_node = NULL; xmlNode *migrate_from = NULL; const char *source = NULL; const char *target = NULL; bool source_newer_op = false; bool target_newer_state = false; bool active_on_target = false; pcmk_scheduler_t *scheduler = history->rsc->private->scheduler; // Get source and target node names from XML if (get_migration_node_names(history->xml, history->node, NULL, &source, &target) != pcmk_rc_ok) { return; } // Check for newer state on the source source_newer_op = non_monitor_after(history->rsc->id, source, history->xml, scheduler); // Check for a migrate_from action from this source on the target migrate_from = find_lrm_op(history->rsc->id, PCMK_ACTION_MIGRATE_FROM, target, source, -1, scheduler); if (migrate_from != NULL) { if (source_newer_op) { /* There's a newer non-monitor operation on the source and a * migrate_from on the target, so this migrate_to is irrelevant to * the resource's state. */ return; } crm_element_value_int(migrate_from, PCMK__XA_RC_CODE, &from_rc); crm_element_value_int(migrate_from, PCMK__XA_OP_STATUS, &from_status); } /* If the resource has newer state on both the source and target after the * migration events, this migrate_to is irrelevant to the resource's state. */ target_newer_state = newer_state_after_migrate(history->rsc->id, target, history->xml, migrate_from, scheduler); if (source_newer_op && target_newer_state) { return; } /* Check for dangling migration (migrate_from succeeded but stop not done). * We know there's no stop because we already returned if the target has a * migrate_from and the source has any newer non-monitor operation. */ if ((from_rc == PCMK_OCF_OK) && (from_status == PCMK_EXEC_DONE)) { add_dangling_migration(history->rsc, history->node); return; } /* Without newer state, this migrate_to implies the resource is active. * (Clones are not allowed to migrate, so role can't be promoted.) */ history->rsc->private->orig_role = pcmk_role_started; target_node = pcmk_find_node(scheduler, target); active_on_target = !target_newer_state && (target_node != NULL) && target_node->details->online; if (from_status != PCMK_EXEC_PENDING) { // migrate_from failed on target if (active_on_target) { native_add_running(history->rsc, target_node, scheduler, TRUE); } else { // Mark resource as failed, require recovery, and prevent migration pcmk__set_rsc_flags(history->rsc, pcmk__rsc_failed|pcmk__rsc_stop_if_failed); pcmk__clear_rsc_flags(history->rsc, pcmk__rsc_migratable); } return; } // The migrate_from is pending, complete but erased, or to be scheduled /* If there is no history at all for the resource on an online target, then * it was likely cleaned. Just return, and we'll schedule a probe. Once we * have the probe result, it will be reflected in target_newer_state. */ if ((target_node != NULL) && target_node->details->online && unknown_on_node(history->rsc, target)) { return; } if (active_on_target) { pcmk_node_t *source_node = pcmk_find_node(scheduler, source); native_add_running(history->rsc, target_node, scheduler, FALSE); if ((source_node != NULL) && source_node->details->online) { /* This is a partial migration: the migrate_to completed * successfully on the source, but the migrate_from has not * completed. Remember the source and target; if the newly * chosen target remains the same when we schedule actions * later, we may continue with the migration. */ history->rsc->private->partial_migration_target = target_node; history->rsc->private->partial_migration_source = source_node; } } else if (!source_newer_op) { // Mark resource as failed, require recovery, and prevent migration pcmk__set_rsc_flags(history->rsc, pcmk__rsc_failed|pcmk__rsc_stop_if_failed); pcmk__clear_rsc_flags(history->rsc, pcmk__rsc_migratable); } } /*! * \internal * \brief Update resource role etc. after a failed migrate_to action * * \param[in,out] history Parsed action result history */ static void unpack_migrate_to_failure(struct action_history *history) { xmlNode *target_migrate_from = NULL; const char *source = NULL; const char *target = NULL; pcmk_scheduler_t *scheduler = history->rsc->private->scheduler; // Get source and target node names from XML if (get_migration_node_names(history->xml, history->node, NULL, &source, &target) != pcmk_rc_ok) { return; } /* If a migration failed, we have to assume the resource is active. Clones * are not allowed to migrate, so role can't be promoted. */ history->rsc->private->orig_role = pcmk_role_started; // Check for migrate_from on the target target_migrate_from = find_lrm_op(history->rsc->id, PCMK_ACTION_MIGRATE_FROM, target, source, PCMK_OCF_OK, scheduler); if (/* If the resource state is unknown on the target, it will likely be * probed there. * Don't just consider it running there. We will get back here anyway in * case the probe detects it's running there. */ !unknown_on_node(history->rsc, target) /* If the resource has newer state on the target after the migration * events, this migrate_to no longer matters for the target. */ && !newer_state_after_migrate(history->rsc->id, target, history->xml, target_migrate_from, scheduler)) { /* The resource has no newer state on the target, so assume it's still * active there. * (if it is up). */ pcmk_node_t *target_node = pcmk_find_node(scheduler, target); if (target_node && target_node->details->online) { native_add_running(history->rsc, target_node, scheduler, FALSE); } } else if (!non_monitor_after(history->rsc->id, source, history->xml, scheduler)) { /* We know the resource has newer state on the target, but this * migrate_to still matters for the source as long as there's no newer * non-monitor operation there. */ // Mark node as having dangling migration so we can force a stop later history->rsc->private->dangling_migration_sources = g_list_prepend(history->rsc->private->dangling_migration_sources, (gpointer) history->node); } } /*! * \internal * \brief Update resource role etc. after a failed migrate_from action * * \param[in,out] history Parsed action result history */ static void unpack_migrate_from_failure(struct action_history *history) { xmlNode *source_migrate_to = NULL; const char *source = NULL; const char *target = NULL; pcmk_scheduler_t *scheduler = history->rsc->private->scheduler; // Get source and target node names from XML if (get_migration_node_names(history->xml, NULL, history->node, &source, &target) != pcmk_rc_ok) { return; } /* If a migration failed, we have to assume the resource is active. Clones * are not allowed to migrate, so role can't be promoted. */ history->rsc->private->orig_role = pcmk_role_started; // Check for a migrate_to on the source source_migrate_to = find_lrm_op(history->rsc->id, PCMK_ACTION_MIGRATE_TO, source, target, PCMK_OCF_OK, scheduler); if (/* If the resource state is unknown on the source, it will likely be * probed there. * Don't just consider it running there. We will get back here anyway in * case the probe detects it's running there. */ !unknown_on_node(history->rsc, source) /* If the resource has newer state on the source after the migration * events, this migrate_from no longer matters for the source. */ && !newer_state_after_migrate(history->rsc->id, source, source_migrate_to, history->xml, scheduler)) { /* The resource has no newer state on the source, so assume it's still * active there (if it is up). */ pcmk_node_t *source_node = pcmk_find_node(scheduler, source); if (source_node && source_node->details->online) { native_add_running(history->rsc, source_node, scheduler, TRUE); } } } /*! * \internal * \brief Add an action to cluster's list of failed actions * * \param[in,out] history Parsed action result history */ static void record_failed_op(struct action_history *history) { const pcmk_scheduler_t *scheduler = history->rsc->private->scheduler; if (!(history->node->details->online)) { return; } for (const xmlNode *xIter = scheduler->failed->children; xIter != NULL; xIter = xIter->next) { const char *key = pcmk__xe_history_key(xIter); const char *uname = crm_element_value(xIter, PCMK_XA_UNAME); if (pcmk__str_eq(history->key, key, pcmk__str_none) && pcmk__str_eq(uname, history->node->private->name, pcmk__str_casei)) { crm_trace("Skipping duplicate entry %s on %s", history->key, pcmk__node_name(history->node)); return; } } crm_trace("Adding entry for %s on %s to failed action list", history->key, pcmk__node_name(history->node)); crm_xml_add(history->xml, PCMK_XA_UNAME, history->node->private->name); crm_xml_add(history->xml, PCMK__XA_RSC_ID, history->rsc->id); pcmk__xml_copy(scheduler->failed, history->xml); } static char * last_change_str(const xmlNode *xml_op) { time_t when; char *result = NULL; if (crm_element_value_epoch(xml_op, PCMK_XA_LAST_RC_CHANGE, &when) == pcmk_ok) { char *when_s = pcmk__epoch2str(&when, 0); const char *p = strchr(when_s, ' '); // Skip day of week to make message shorter if ((p != NULL) && (*(++p) != '\0')) { result = pcmk__str_copy(p); } free(when_s); } if (result == NULL) { result = pcmk__str_copy("unknown_time"); } return result; } /*! * \internal * \brief Ban a resource (or its clone if an anonymous instance) from all nodes * * \param[in,out] rsc Resource to ban */ static void ban_from_all_nodes(pcmk_resource_t *rsc) { int score = -PCMK_SCORE_INFINITY; const pcmk_scheduler_t *scheduler = rsc->private->scheduler; if (rsc->private->parent != NULL) { pcmk_resource_t *parent = uber_parent(rsc); if (pcmk__is_anonymous_clone(parent)) { /* For anonymous clones, if an operation with * PCMK_META_ON_FAIL=PCMK_VALUE_STOP fails for any instance, the * entire clone must stop. */ rsc = parent; } } // Ban the resource from all nodes crm_notice("%s will not be started under current conditions", rsc->id); if (rsc->private->allowed_nodes != NULL) { g_hash_table_destroy(rsc->private->allowed_nodes); } rsc->private->allowed_nodes = pe__node_list2table(scheduler->nodes); g_hash_table_foreach(rsc->private->allowed_nodes, set_node_score, &score); } /*! * \internal * \brief Get configured failure handling and role after failure for an action * * \param[in,out] history Unpacked action history entry * \param[out] on_fail Where to set configured failure handling * \param[out] fail_role Where to set to role after failure */ static void unpack_failure_handling(struct action_history *history, enum pcmk__on_fail *on_fail, enum rsc_role_e *fail_role) { xmlNode *config = pcmk__find_action_config(history->rsc, history->task, history->interval_ms, true); GHashTable *meta = pcmk__unpack_action_meta(history->rsc, history->node, history->task, history->interval_ms, config); const char *on_fail_str = g_hash_table_lookup(meta, PCMK_META_ON_FAIL); *on_fail = pcmk__parse_on_fail(history->rsc, history->task, history->interval_ms, on_fail_str); *fail_role = pcmk__role_after_failure(history->rsc, history->task, *on_fail, meta); g_hash_table_destroy(meta); } /*! * \internal * \brief Update resource role, failure handling, etc., after a failed action * * \param[in,out] history Parsed action result history * \param[in] config_on_fail Action failure handling from configuration * \param[in] fail_role Resource's role after failure of this action * \param[out] last_failure This will be set to the history XML * \param[in,out] on_fail Actual handling of action result */ static void unpack_rsc_op_failure(struct action_history *history, enum pcmk__on_fail config_on_fail, enum rsc_role_e fail_role, xmlNode **last_failure, enum pcmk__on_fail *on_fail) { bool is_probe = false; char *last_change_s = NULL; pcmk_scheduler_t *scheduler = history->rsc->private->scheduler; *last_failure = history->xml; is_probe = pcmk_xe_is_probe(history->xml); last_change_s = last_change_str(history->xml); if (!pcmk_is_set(scheduler->flags, pcmk_sched_symmetric_cluster) && (history->exit_status == PCMK_OCF_NOT_INSTALLED)) { crm_trace("Unexpected result (%s%s%s) was recorded for " "%s of %s on %s at %s " QB_XS " exit-status=%d id=%s", services_ocf_exitcode_str(history->exit_status), (pcmk__str_empty(history->exit_reason)? "" : ": "), pcmk__s(history->exit_reason, ""), (is_probe? "probe" : history->task), history->rsc->id, pcmk__node_name(history->node), last_change_s, history->exit_status, history->id); } else { pcmk__sched_warn("Unexpected result (%s%s%s) was recorded for %s of " "%s on %s at %s " QB_XS " exit-status=%d id=%s", services_ocf_exitcode_str(history->exit_status), (pcmk__str_empty(history->exit_reason)? "" : ": "), pcmk__s(history->exit_reason, ""), (is_probe? "probe" : history->task), history->rsc->id, pcmk__node_name(history->node), last_change_s, history->exit_status, history->id); if (is_probe && (history->exit_status != PCMK_OCF_OK) && (history->exit_status != PCMK_OCF_NOT_RUNNING) && (history->exit_status != PCMK_OCF_RUNNING_PROMOTED)) { /* A failed (not just unexpected) probe result could mean the user * didn't know resources will be probed even where they can't run. */ crm_notice("If it is not possible for %s to run on %s, see " "the " PCMK_XA_RESOURCE_DISCOVERY " option for location " "constraints", history->rsc->id, pcmk__node_name(history->node)); } record_failed_op(history); } free(last_change_s); if (*on_fail < config_on_fail) { pcmk__rsc_trace(history->rsc, "on-fail %s -> %s for %s", pcmk__on_fail_text(*on_fail), pcmk__on_fail_text(config_on_fail), history->key); *on_fail = config_on_fail; } if (strcmp(history->task, PCMK_ACTION_STOP) == 0) { resource_location(history->rsc, history->node, -PCMK_SCORE_INFINITY, "__stop_fail__", scheduler); } else if (strcmp(history->task, PCMK_ACTION_MIGRATE_TO) == 0) { unpack_migrate_to_failure(history); } else if (strcmp(history->task, PCMK_ACTION_MIGRATE_FROM) == 0) { unpack_migrate_from_failure(history); } else if (strcmp(history->task, PCMK_ACTION_PROMOTE) == 0) { history->rsc->private->orig_role = pcmk_role_promoted; } else if (strcmp(history->task, PCMK_ACTION_DEMOTE) == 0) { if (config_on_fail == pcmk__on_fail_block) { history->rsc->private->orig_role = pcmk_role_promoted; pe__set_next_role(history->rsc, pcmk_role_stopped, "demote with " PCMK_META_ON_FAIL "=block"); } else if (history->exit_status == PCMK_OCF_NOT_RUNNING) { history->rsc->private->orig_role = pcmk_role_stopped; } else { /* Staying in the promoted role would put the scheduler and * controller into a loop. Setting the role to unpromoted is not * dangerous because the resource will be stopped as part of * recovery, and any promotion will be ordered after that stop. */ history->rsc->private->orig_role = pcmk_role_unpromoted; } } if (is_probe && (history->exit_status == PCMK_OCF_NOT_INSTALLED)) { /* leave stopped */ pcmk__rsc_trace(history->rsc, "Leaving %s stopped", history->rsc->id); history->rsc->private->orig_role = pcmk_role_stopped; } else if (history->rsc->private->orig_role < pcmk_role_started) { pcmk__rsc_trace(history->rsc, "Setting %s active", history->rsc->id); set_active(history->rsc); } pcmk__rsc_trace(history->rsc, "Resource %s: role=%s unclean=%s on_fail=%s fail_role=%s", history->rsc->id, pcmk_role_text(history->rsc->private->orig_role), pcmk__btoa(history->node->details->unclean), pcmk__on_fail_text(config_on_fail), pcmk_role_text(fail_role)); if ((fail_role != pcmk_role_started) && (history->rsc->private->next_role < fail_role)) { pe__set_next_role(history->rsc, fail_role, "failure"); } if (fail_role == pcmk_role_stopped) { ban_from_all_nodes(history->rsc); } } /*! * \internal * \brief Block a resource with a failed action if it cannot be recovered * * If resource action is a failed stop and fencing is not possible, mark the * resource as unmanaged and blocked, since recovery cannot be done. * * \param[in,out] history Parsed action history entry */ static void block_if_unrecoverable(struct action_history *history) { char *last_change_s = NULL; if (strcmp(history->task, PCMK_ACTION_STOP) != 0) { return; // All actions besides stop are always recoverable } if (pe_can_fence(history->node->private->scheduler, history->node)) { return; // Failed stops are recoverable via fencing } last_change_s = last_change_str(history->xml); pcmk__sched_err("No further recovery can be attempted for %s " "because %s on %s failed (%s%s%s) at %s " QB_XS " rc=%d id=%s", history->rsc->id, history->task, pcmk__node_name(history->node), services_ocf_exitcode_str(history->exit_status), (pcmk__str_empty(history->exit_reason)? "" : ": "), pcmk__s(history->exit_reason, ""), last_change_s, history->exit_status, history->id); free(last_change_s); pcmk__clear_rsc_flags(history->rsc, pcmk__rsc_managed); pcmk__set_rsc_flags(history->rsc, pcmk__rsc_blocked); } /*! * \internal * \brief Update action history's execution status and why * * \param[in,out] history Parsed action history entry * \param[out] why Where to store reason for update * \param[in] value New value * \param[in] reason Description of why value was changed */ static inline void remap_because(struct action_history *history, const char **why, int value, const char *reason) { if (history->execution_status != value) { history->execution_status = value; *why = reason; } } /*! * \internal * \brief Remap informational monitor results and operation status * * For the monitor results, certain OCF codes are for providing extended information * to the user about services that aren't yet failed but not entirely healthy either. * These must be treated as the "normal" result by Pacemaker. * * For operation status, the action result can be used to determine an appropriate * status for the purposes of responding to the action. The status provided by the * executor is not directly usable since the executor does not know what was expected. * * \param[in,out] history Parsed action history entry * \param[in,out] on_fail What should be done about the result * \param[in] expired Whether result is expired * * \note If the result is remapped and the node is not shutting down or failed, * the operation will be recorded in the scheduler data's list of failed * operations to highlight it for the user. * * \note This may update the resource's current and next role. */ static void remap_operation(struct action_history *history, enum pcmk__on_fail *on_fail, bool expired) { bool is_probe = false; int orig_exit_status = history->exit_status; int orig_exec_status = history->execution_status; const char *why = NULL; const char *task = history->task; // Remap degraded results to their successful counterparts history->exit_status = pcmk__effective_rc(history->exit_status); if (history->exit_status != orig_exit_status) { why = "degraded result"; if (!expired && (!history->node->details->shutdown || history->node->details->online)) { record_failed_op(history); } } if (!pcmk__is_bundled(history->rsc) && pcmk_xe_mask_probe_failure(history->xml) && ((history->execution_status != PCMK_EXEC_DONE) || (history->exit_status != PCMK_OCF_NOT_RUNNING))) { history->execution_status = PCMK_EXEC_DONE; history->exit_status = PCMK_OCF_NOT_RUNNING; why = "equivalent probe result"; } /* If the executor reported an execution status of anything but done or * error, consider that final. But for done or error, we know better whether * it should be treated as a failure or not, because we know the expected * result. */ switch (history->execution_status) { case PCMK_EXEC_DONE: case PCMK_EXEC_ERROR: break; // These should be treated as node-fatal case PCMK_EXEC_NO_FENCE_DEVICE: case PCMK_EXEC_NO_SECRETS: remap_because(history, &why, PCMK_EXEC_ERROR_HARD, "node-fatal error"); goto remap_done; default: goto remap_done; } is_probe = pcmk_xe_is_probe(history->xml); if (is_probe) { task = "probe"; } if (history->expected_exit_status < 0) { /* Pre-1.0 Pacemaker versions, and Pacemaker 1.1.6 or earlier with * Heartbeat 2.0.7 or earlier as the cluster layer, did not include the * expected exit status in the transition key, which (along with the * similar case of a corrupted transition key in the CIB) will be * reported to this function as -1. Pacemaker 2.0+ does not support * rolling upgrades from those versions or processing of saved CIB files * from those versions, so we do not need to care much about this case. */ remap_because(history, &why, PCMK_EXEC_ERROR, "obsolete history format"); pcmk__config_warn("Expected result not found for %s on %s " "(corrupt or obsolete CIB?)", history->key, pcmk__node_name(history->node)); } else if (history->exit_status == history->expected_exit_status) { remap_because(history, &why, PCMK_EXEC_DONE, "expected result"); } else { remap_because(history, &why, PCMK_EXEC_ERROR, "unexpected result"); pcmk__rsc_debug(history->rsc, "%s on %s: expected %d (%s), got %d (%s%s%s)", history->key, pcmk__node_name(history->node), history->expected_exit_status, services_ocf_exitcode_str(history->expected_exit_status), history->exit_status, services_ocf_exitcode_str(history->exit_status), (pcmk__str_empty(history->exit_reason)? "" : ": "), pcmk__s(history->exit_reason, "")); } switch (history->exit_status) { case PCMK_OCF_OK: if (is_probe && (history->expected_exit_status == PCMK_OCF_NOT_RUNNING)) { char *last_change_s = last_change_str(history->xml); remap_because(history, &why, PCMK_EXEC_DONE, "probe"); pcmk__rsc_info(history->rsc, "Probe found %s active on %s at %s", history->rsc->id, pcmk__node_name(history->node), last_change_s); free(last_change_s); } break; case PCMK_OCF_NOT_RUNNING: if (is_probe || (history->expected_exit_status == history->exit_status) || !pcmk_is_set(history->rsc->flags, pcmk__rsc_managed)) { /* For probes, recurring monitors for the Stopped role, and * unmanaged resources, "not running" is not considered a * failure. */ remap_because(history, &why, PCMK_EXEC_DONE, "exit status"); history->rsc->private->orig_role = pcmk_role_stopped; *on_fail = pcmk__on_fail_ignore; pe__set_next_role(history->rsc, pcmk_role_unknown, "not running"); } break; case PCMK_OCF_RUNNING_PROMOTED: if (is_probe && (history->exit_status != history->expected_exit_status)) { char *last_change_s = last_change_str(history->xml); remap_because(history, &why, PCMK_EXEC_DONE, "probe"); pcmk__rsc_info(history->rsc, "Probe found %s active and promoted on %s at %s", history->rsc->id, pcmk__node_name(history->node), last_change_s); free(last_change_s); } if (!expired || (history->exit_status == history->expected_exit_status)) { history->rsc->private->orig_role = pcmk_role_promoted; } break; case PCMK_OCF_FAILED_PROMOTED: if (!expired) { history->rsc->private->orig_role = pcmk_role_promoted; } remap_because(history, &why, PCMK_EXEC_ERROR, "exit status"); break; case PCMK_OCF_NOT_CONFIGURED: remap_because(history, &why, PCMK_EXEC_ERROR_FATAL, "exit status"); break; case PCMK_OCF_UNIMPLEMENT_FEATURE: { guint interval_ms = 0; crm_element_value_ms(history->xml, PCMK_META_INTERVAL, &interval_ms); if (interval_ms == 0) { if (!expired) { block_if_unrecoverable(history); } remap_because(history, &why, PCMK_EXEC_ERROR_HARD, "exit status"); } else { remap_because(history, &why, PCMK_EXEC_NOT_SUPPORTED, "exit status"); } } break; case PCMK_OCF_NOT_INSTALLED: case PCMK_OCF_INVALID_PARAM: case PCMK_OCF_INSUFFICIENT_PRIV: if (!expired) { block_if_unrecoverable(history); } remap_because(history, &why, PCMK_EXEC_ERROR_HARD, "exit status"); break; default: if (history->execution_status == PCMK_EXEC_DONE) { char *last_change_s = last_change_str(history->xml); crm_info("Treating unknown exit status %d from %s of %s " "on %s at %s as failure", history->exit_status, task, history->rsc->id, pcmk__node_name(history->node), last_change_s); remap_because(history, &why, PCMK_EXEC_ERROR, "unknown exit status"); free(last_change_s); } break; } remap_done: if (why != NULL) { pcmk__rsc_trace(history->rsc, "Remapped %s result from [%s: %s] to [%s: %s] " "because of %s", history->key, pcmk_exec_status_str(orig_exec_status), crm_exit_str(orig_exit_status), pcmk_exec_status_str(history->execution_status), crm_exit_str(history->exit_status), why); } } // return TRUE if start or monitor last failure but parameters changed static bool should_clear_for_param_change(const xmlNode *xml_op, const char *task, pcmk_resource_t *rsc, pcmk_node_t *node) { if (pcmk__str_any_of(task, PCMK_ACTION_START, PCMK_ACTION_MONITOR, NULL)) { if (pe__bundle_needs_remote_name(rsc)) { /* We haven't allocated resources yet, so we can't reliably * substitute addr parameters for the REMOTE_CONTAINER_HACK. * When that's needed, defer the check until later. */ pe__add_param_check(xml_op, rsc, node, pcmk__check_last_failure, rsc->private->scheduler); } else { pcmk__op_digest_t *digest_data = NULL; digest_data = rsc_action_digest_cmp(rsc, xml_op, node, rsc->private->scheduler); switch (digest_data->rc) { case pcmk__digest_unknown: crm_trace("Resource %s history entry %s on %s" " has no digest to compare", rsc->id, pcmk__xe_history_key(xml_op), node->private->id); break; case pcmk__digest_match: break; default: return TRUE; } } } return FALSE; } // Order action after fencing of remote node, given connection rsc static void order_after_remote_fencing(pcmk_action_t *action, pcmk_resource_t *remote_conn, pcmk_scheduler_t *scheduler) { pcmk_node_t *remote_node = pcmk_find_node(scheduler, remote_conn->id); if (remote_node) { pcmk_action_t *fence = pe_fence_op(remote_node, NULL, TRUE, NULL, FALSE, scheduler); order_actions(fence, action, pcmk__ar_first_implies_then); } } static bool should_ignore_failure_timeout(const pcmk_resource_t *rsc, const char *task, guint interval_ms, bool is_last_failure) { /* Clearing failures of recurring monitors has special concerns. The * executor reports only changes in the monitor result, so if the * monitor is still active and still getting the same failure result, * that will go undetected after the failure is cleared. * * Also, the operation history will have the time when the recurring * monitor result changed to the given code, not the time when the * result last happened. * * @TODO We probably should clear such failures only when the failure * timeout has passed since the last occurrence of the failed result. * However we don't record that information. We could maybe approximate * that by clearing only if there is a more recent successful monitor or * stop result, but we don't even have that information at this point * since we are still unpacking the resource's operation history. * * This is especially important for remote connection resources with a * reconnect interval, so in that case, we skip clearing failures * if the remote node hasn't been fenced. */ if ((rsc->private->remote_reconnect_ms > 0U) && pcmk_is_set(rsc->private->scheduler->flags, pcmk_sched_fencing_enabled) && (interval_ms != 0) && pcmk__str_eq(task, PCMK_ACTION_MONITOR, pcmk__str_casei)) { pcmk_node_t *remote_node = pcmk_find_node(rsc->private->scheduler, rsc->id); if (remote_node && !pcmk_is_set(remote_node->private->flags, pcmk__node_remote_fenced)) { if (is_last_failure) { crm_info("Waiting to clear monitor failure for remote node %s" " until fencing has occurred", rsc->id); } return TRUE; } } return FALSE; } /*! * \internal * \brief Check operation age and schedule failure clearing when appropriate * * This function has two distinct purposes. The first is to check whether an * operation history entry is expired (i.e. the resource has a failure timeout, * the entry is older than the timeout, and the resource either has no fail * count or its fail count is entirely older than the timeout). The second is to * schedule fail count clearing when appropriate (i.e. the operation is expired * and either the resource has an expired fail count or the operation is a * last_failure for a remote connection resource with a reconnect interval, * or the operation is a last_failure for a start or monitor operation and the * resource's parameters have changed since the operation). * * \param[in,out] history Parsed action result history * * \return true if operation history entry is expired, otherwise false */ static bool check_operation_expiry(struct action_history *history) { bool expired = false; bool is_last_failure = pcmk__ends_with(history->id, "_last_failure_0"); time_t last_run = 0; int unexpired_fail_count = 0; const char *clear_reason = NULL; const guint expiration_sec = history->rsc->private->failure_expiration_ms / 1000; pcmk_scheduler_t *scheduler = history->rsc->private->scheduler; if (history->execution_status == PCMK_EXEC_NOT_INSTALLED) { pcmk__rsc_trace(history->rsc, "Resource history entry %s on %s is not expired: " "Not Installed does not expire", history->id, pcmk__node_name(history->node)); return false; // "Not installed" must always be cleared manually } if ((expiration_sec > 0) && (crm_element_value_epoch(history->xml, PCMK_XA_LAST_RC_CHANGE, &last_run) == 0)) { /* Resource has a PCMK_META_FAILURE_TIMEOUT and history entry has a * timestamp */ time_t now = get_effective_time(scheduler); time_t last_failure = 0; // Is this particular operation history older than the failure timeout? if ((now >= (last_run + expiration_sec)) && !should_ignore_failure_timeout(history->rsc, history->task, history->interval_ms, is_last_failure)) { expired = true; } // Does the resource as a whole have an unexpired fail count? unexpired_fail_count = pe_get_failcount(history->node, history->rsc, &last_failure, pcmk__fc_effective, history->xml); // Update scheduler recheck time according to *last* failure crm_trace("%s@%lld is %sexpired @%lld with unexpired_failures=%d " "expiration=%s last-failure@%lld", history->id, (long long) last_run, (expired? "" : "not "), (long long) now, unexpired_fail_count, pcmk__readable_interval(expiration_sec * 1000), (long long) last_failure); last_failure += expiration_sec + 1; if (unexpired_fail_count && (now < last_failure)) { pe__update_recheck_time(last_failure, scheduler, "fail count expiration"); } } if (expired) { if (pe_get_failcount(history->node, history->rsc, NULL, pcmk__fc_default, history->xml)) { // There is a fail count ignoring timeout if (unexpired_fail_count == 0) { // There is no fail count considering timeout clear_reason = "it expired"; } else { /* This operation is old, but there is an unexpired fail count. * In a properly functioning cluster, this should only be * possible if this operation is not a failure (otherwise the * fail count should be expired too), so this is really just a * failsafe. */ pcmk__rsc_trace(history->rsc, "Resource history entry %s on %s is not " "expired: Unexpired fail count", history->id, pcmk__node_name(history->node)); expired = false; } } else if (is_last_failure && (history->rsc->private->remote_reconnect_ms > 0U)) { /* Clear any expired last failure when reconnect interval is set, * even if there is no fail count. */ clear_reason = "reconnect interval is set"; } } if (!expired && is_last_failure && should_clear_for_param_change(history->xml, history->task, history->rsc, history->node)) { clear_reason = "resource parameters have changed"; } if (clear_reason != NULL) { pcmk_action_t *clear_op = NULL; // Schedule clearing of the fail count clear_op = pe__clear_failcount(history->rsc, history->node, clear_reason, scheduler); if (pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled) && (history->rsc->private->remote_reconnect_ms > 0)) { /* If we're clearing a remote connection due to a reconnect * interval, we want to wait until any scheduled fencing * completes. * * We could limit this to remote_node->details->unclean, but at * this point, that's always true (it won't be reliable until * after unpack_node_history() is done). */ crm_info("Clearing %s failure will wait until any scheduled " "fencing of %s completes", history->task, history->rsc->id); order_after_remote_fencing(clear_op, history->rsc, scheduler); } } if (expired && (history->interval_ms == 0) && pcmk__str_eq(history->task, PCMK_ACTION_MONITOR, pcmk__str_none)) { switch (history->exit_status) { case PCMK_OCF_OK: case PCMK_OCF_NOT_RUNNING: case PCMK_OCF_RUNNING_PROMOTED: case PCMK_OCF_DEGRADED: case PCMK_OCF_DEGRADED_PROMOTED: // Don't expire probes that return these values pcmk__rsc_trace(history->rsc, "Resource history entry %s on %s is not " "expired: Probe result", history->id, pcmk__node_name(history->node)); expired = false; break; } } return expired; } int pe__target_rc_from_xml(const xmlNode *xml_op) { int target_rc = 0; const char *key = crm_element_value(xml_op, PCMK__XA_TRANSITION_KEY); if (key == NULL) { return -1; } decode_transition_key(key, NULL, NULL, NULL, &target_rc); return target_rc; } /*! * \internal * \brief Update a resource's state for an action result * * \param[in,out] history Parsed action history entry * \param[in] exit_status Exit status to base new state on * \param[in] last_failure Resource's last_failure entry, if known * \param[in,out] on_fail Resource's current failure handling */ static void update_resource_state(struct action_history *history, int exit_status, const xmlNode *last_failure, enum pcmk__on_fail *on_fail) { bool clear_past_failure = false; if ((exit_status == PCMK_OCF_NOT_INSTALLED) || (!pcmk__is_bundled(history->rsc) && pcmk_xe_mask_probe_failure(history->xml))) { history->rsc->private->orig_role = pcmk_role_stopped; } else if (exit_status == PCMK_OCF_NOT_RUNNING) { clear_past_failure = true; } else if (pcmk__str_eq(history->task, PCMK_ACTION_MONITOR, pcmk__str_none)) { if ((last_failure != NULL) && pcmk__str_eq(history->key, pcmk__xe_history_key(last_failure), pcmk__str_none)) { clear_past_failure = true; } if (history->rsc->private->orig_role < pcmk_role_started) { set_active(history->rsc); } } else if (pcmk__str_eq(history->task, PCMK_ACTION_START, pcmk__str_none)) { history->rsc->private->orig_role = pcmk_role_started; clear_past_failure = true; } else if (pcmk__str_eq(history->task, PCMK_ACTION_STOP, pcmk__str_none)) { history->rsc->private->orig_role = pcmk_role_stopped; clear_past_failure = true; } else if (pcmk__str_eq(history->task, PCMK_ACTION_PROMOTE, pcmk__str_none)) { history->rsc->private->orig_role = pcmk_role_promoted; clear_past_failure = true; } else if (pcmk__str_eq(history->task, PCMK_ACTION_DEMOTE, pcmk__str_none)) { if (*on_fail == pcmk__on_fail_demote) { /* Demote clears an error only if * PCMK_META_ON_FAIL=PCMK_VALUE_DEMOTE */ clear_past_failure = true; } history->rsc->private->orig_role = pcmk_role_unpromoted; } else if (pcmk__str_eq(history->task, PCMK_ACTION_MIGRATE_FROM, pcmk__str_none)) { history->rsc->private->orig_role = pcmk_role_started; clear_past_failure = true; } else if (pcmk__str_eq(history->task, PCMK_ACTION_MIGRATE_TO, pcmk__str_none)) { unpack_migrate_to_success(history); } else if (history->rsc->private->orig_role < pcmk_role_started) { pcmk__rsc_trace(history->rsc, "%s active on %s", history->rsc->id, pcmk__node_name(history->node)); set_active(history->rsc); } if (!clear_past_failure) { return; } switch (*on_fail) { case pcmk__on_fail_stop: case pcmk__on_fail_ban: case pcmk__on_fail_standby_node: case pcmk__on_fail_fence_node: pcmk__rsc_trace(history->rsc, "%s (%s) is not cleared by a completed %s", history->rsc->id, pcmk__on_fail_text(*on_fail), history->task); break; case pcmk__on_fail_block: case pcmk__on_fail_ignore: case pcmk__on_fail_demote: case pcmk__on_fail_restart: case pcmk__on_fail_restart_container: *on_fail = pcmk__on_fail_ignore; pe__set_next_role(history->rsc, pcmk_role_unknown, "clear past failures"); break; case pcmk__on_fail_reset_remote: if (history->rsc->private->remote_reconnect_ms == 0U) { /* With no reconnect interval, the connection is allowed to * start again after the remote node is fenced and * completely stopped. (With a reconnect interval, we wait * for the failure to be cleared entirely before attempting * to reconnect.) */ *on_fail = pcmk__on_fail_ignore; pe__set_next_role(history->rsc, pcmk_role_unknown, "clear past failures and reset remote"); } break; } } /*! * \internal * \brief Check whether a given history entry matters for resource state * * \param[in] history Parsed action history entry * * \return true if action can affect resource state, otherwise false */ static inline bool can_affect_state(struct action_history *history) { #if 0 /* @COMPAT It might be better to parse only actions we know we're interested * in, rather than exclude a couple we don't. However that would be a * behavioral change that should be done at a major or minor series release. * Currently, unknown operations can affect whether a resource is considered * active and/or failed. */ return pcmk__str_any_of(history->task, PCMK_ACTION_MONITOR, PCMK_ACTION_START, PCMK_ACTION_STOP, PCMK_ACTION_PROMOTE, PCMK_ACTION_DEMOTE, PCMK_ACTION_MIGRATE_TO, PCMK_ACTION_MIGRATE_FROM, "asyncmon", NULL); #else return !pcmk__str_any_of(history->task, PCMK_ACTION_NOTIFY, PCMK_ACTION_META_DATA, NULL); #endif } /*! * \internal * \brief Unpack execution/exit status and exit reason from a history entry * * \param[in,out] history Action history entry to unpack * * \return Standard Pacemaker return code */ static int unpack_action_result(struct action_history *history) { if ((crm_element_value_int(history->xml, PCMK__XA_OP_STATUS, &(history->execution_status)) < 0) || (history->execution_status < PCMK_EXEC_PENDING) || (history->execution_status > PCMK_EXEC_MAX) || (history->execution_status == PCMK_EXEC_CANCELLED)) { pcmk__config_err("Ignoring resource history entry %s for %s on %s " "with invalid " PCMK__XA_OP_STATUS " '%s'", history->id, history->rsc->id, pcmk__node_name(history->node), pcmk__s(crm_element_value(history->xml, PCMK__XA_OP_STATUS), "")); return pcmk_rc_unpack_error; } if ((crm_element_value_int(history->xml, PCMK__XA_RC_CODE, &(history->exit_status)) < 0) || (history->exit_status < 0) || (history->exit_status > CRM_EX_MAX)) { #if 0 /* @COMPAT We should ignore malformed entries, but since that would * change behavior, it should be done at a major or minor series * release. */ pcmk__config_err("Ignoring resource history entry %s for %s on %s " "with invalid " PCMK__XA_RC_CODE " '%s'", history->id, history->rsc->id, pcmk__node_name(history->node), pcmk__s(crm_element_value(history->xml, PCMK__XA_RC_CODE), "")); return pcmk_rc_unpack_error; #else history->exit_status = CRM_EX_ERROR; #endif } history->exit_reason = crm_element_value(history->xml, PCMK_XA_EXIT_REASON); return pcmk_rc_ok; } /*! * \internal * \brief Process an action history entry whose result expired * * \param[in,out] history Parsed action history entry * \param[in] orig_exit_status Action exit status before remapping * * \return Standard Pacemaker return code (in particular, pcmk_rc_ok means the * entry needs no further processing) */ static int process_expired_result(struct action_history *history, int orig_exit_status) { if (!pcmk__is_bundled(history->rsc) && pcmk_xe_mask_probe_failure(history->xml) && (orig_exit_status != history->expected_exit_status)) { if (history->rsc->private->orig_role <= pcmk_role_stopped) { history->rsc->private->orig_role = pcmk_role_unknown; } crm_trace("Ignoring resource history entry %s for probe of %s on %s: " "Masked failure expired", history->id, history->rsc->id, pcmk__node_name(history->node)); return pcmk_rc_ok; } if (history->exit_status == history->expected_exit_status) { return pcmk_rc_undetermined; // Only failures expire } if (history->interval_ms == 0) { crm_notice("Ignoring resource history entry %s for %s of %s on %s: " "Expired failure", history->id, history->task, history->rsc->id, pcmk__node_name(history->node)); return pcmk_rc_ok; } if (history->node->details->online && !history->node->details->unclean) { /* Reschedule the recurring action. schedule_cancel() won't work at * this stage, so as a hacky workaround, forcibly change the restart * digest so pcmk__check_action_config() does what we want later. * * @TODO We should skip this if there is a newer successful monitor. * Also, this causes rescheduling only if the history entry * has a PCMK__XA_OP_DIGEST (which the expire-non-blocked-failure * scheduler regression test doesn't, but that may not be a * realistic scenario in production). */ crm_notice("Rescheduling %s-interval %s of %s on %s " "after failure expired", pcmk__readable_interval(history->interval_ms), history->task, history->rsc->id, pcmk__node_name(history->node)); crm_xml_add(history->xml, PCMK__XA_OP_RESTART_DIGEST, "calculated-failure-timeout"); return pcmk_rc_ok; } return pcmk_rc_undetermined; } /*! * \internal * \brief Process a masked probe failure * * \param[in,out] history Parsed action history entry * \param[in] orig_exit_status Action exit status before remapping * \param[in] last_failure Resource's last_failure entry, if known * \param[in,out] on_fail Resource's current failure handling */ static void mask_probe_failure(struct action_history *history, int orig_exit_status, const xmlNode *last_failure, enum pcmk__on_fail *on_fail) { pcmk_resource_t *ban_rsc = history->rsc; if (!pcmk_is_set(history->rsc->flags, pcmk__rsc_unique)) { ban_rsc = uber_parent(history->rsc); } crm_notice("Treating probe result '%s' for %s on %s as 'not running'", services_ocf_exitcode_str(orig_exit_status), history->rsc->id, pcmk__node_name(history->node)); update_resource_state(history, history->expected_exit_status, last_failure, on_fail); crm_xml_add(history->xml, PCMK_XA_UNAME, history->node->private->name); record_failed_op(history); resource_location(ban_rsc, history->node, -PCMK_SCORE_INFINITY, "masked-probe-failure", ban_rsc->private->scheduler); } /*! * \internal Check whether a given failure is for a given pending action * * \param[in] history Parsed history entry for pending action * \param[in] last_failure Resource's last_failure entry, if known * * \return true if \p last_failure is failure of pending action in \p history, * otherwise false * \note Both \p history and \p last_failure must come from the same * \c PCMK__XE_LRM_RESOURCE block, as node and resource are assumed to be * the same. */ static bool failure_is_newer(const struct action_history *history, const xmlNode *last_failure) { guint failure_interval_ms = 0U; long long failure_change = 0LL; long long this_change = 0LL; if (last_failure == NULL) { return false; // Resource has no last_failure entry } if (!pcmk__str_eq(history->task, crm_element_value(last_failure, PCMK_XA_OPERATION), pcmk__str_none)) { return false; // last_failure is for different action } if ((crm_element_value_ms(last_failure, PCMK_META_INTERVAL, &failure_interval_ms) != pcmk_ok) || (history->interval_ms != failure_interval_ms)) { return false; // last_failure is for action with different interval } if ((pcmk__scan_ll(crm_element_value(history->xml, PCMK_XA_LAST_RC_CHANGE), &this_change, 0LL) != pcmk_rc_ok) || (pcmk__scan_ll(crm_element_value(last_failure, PCMK_XA_LAST_RC_CHANGE), &failure_change, 0LL) != pcmk_rc_ok) || (failure_change < this_change)) { return false; // Failure is not known to be newer } return true; } /*! * \internal * \brief Update a resource's role etc. for a pending action * * \param[in,out] history Parsed history entry for pending action * \param[in] last_failure Resource's last_failure entry, if known */ static void process_pending_action(struct action_history *history, const xmlNode *last_failure) { /* For recurring monitors, a failure is recorded only in RSC_last_failure_0, * and there might be a RSC_monitor_INTERVAL entry with the last successful * or pending result. * * If last_failure contains the failure of the pending recurring monitor * we're processing here, and is newer, the action is no longer pending. * (Pending results have call ID -1, which sorts last, so the last failure * if any should be known.) */ if (failure_is_newer(history, last_failure)) { return; } if (strcmp(history->task, PCMK_ACTION_START) == 0) { pcmk__set_rsc_flags(history->rsc, pcmk__rsc_start_pending); set_active(history->rsc); } else if (strcmp(history->task, PCMK_ACTION_PROMOTE) == 0) { history->rsc->private->orig_role = pcmk_role_promoted; } else if ((strcmp(history->task, PCMK_ACTION_MIGRATE_TO) == 0) && history->node->details->unclean) { /* A migrate_to action is pending on a unclean source, so force a stop * on the target. */ const char *migrate_target = NULL; pcmk_node_t *target = NULL; migrate_target = crm_element_value(history->xml, PCMK__META_MIGRATE_TARGET); target = pcmk_find_node(history->rsc->private->scheduler, migrate_target); if (target != NULL) { stop_action(history->rsc, target, FALSE); } } if (history->rsc->private->pending_action != NULL) { /* There should never be multiple pending actions, but as a failsafe, * just remember the first one processed for display purposes. */ return; } if (pcmk_is_probe(history->task, history->interval_ms)) { /* Pending probes are currently never displayed, even if pending * operations are requested. If we ever want to change that, * enable the below and the corresponding part of * native.c:native_pending_action(). */ #if 0 history->rsc->private->pending_action = strdup("probe"); history->rsc->private->pending_node = history->node; #endif } else { history->rsc->private->pending_action = strdup(history->task); history->rsc->private->pending_node = history->node; } } static void unpack_rsc_op(pcmk_resource_t *rsc, pcmk_node_t *node, xmlNode *xml_op, xmlNode **last_failure, enum pcmk__on_fail *on_fail) { int old_rc = 0; bool expired = false; pcmk_resource_t *parent = rsc; enum rsc_role_e fail_role = pcmk_role_unknown; enum pcmk__on_fail failure_strategy = pcmk__on_fail_restart; struct action_history history = { .rsc = rsc, .node = node, .xml = xml_op, .execution_status = PCMK_EXEC_UNKNOWN, }; CRM_CHECK(rsc && node && xml_op, return); history.id = pcmk__xe_id(xml_op); if (history.id == NULL) { pcmk__config_err("Ignoring resource history entry for %s on %s " "without ID", rsc->id, pcmk__node_name(node)); return; } // Task and interval history.task = crm_element_value(xml_op, PCMK_XA_OPERATION); if (history.task == NULL) { pcmk__config_err("Ignoring resource history entry %s for %s on %s " "without " PCMK_XA_OPERATION, history.id, rsc->id, pcmk__node_name(node)); return; } crm_element_value_ms(xml_op, PCMK_META_INTERVAL, &(history.interval_ms)); if (!can_affect_state(&history)) { pcmk__rsc_trace(rsc, "Ignoring resource history entry %s for %s on %s " "with irrelevant action '%s'", history.id, rsc->id, pcmk__node_name(node), history.task); return; } if (unpack_action_result(&history) != pcmk_rc_ok) { return; // Error already logged } history.expected_exit_status = pe__target_rc_from_xml(xml_op); history.key = pcmk__xe_history_key(xml_op); crm_element_value_int(xml_op, PCMK__XA_CALL_ID, &(history.call_id)); pcmk__rsc_trace(rsc, "Unpacking %s (%s call %d on %s): %s (%s)", history.id, history.task, history.call_id, pcmk__node_name(node), pcmk_exec_status_str(history.execution_status), crm_exit_str(history.exit_status)); if (node->details->unclean) { pcmk__rsc_trace(rsc, "%s is running on %s, which is unclean (further action " "depends on value of stop's on-fail attribute)", rsc->id, pcmk__node_name(node)); } expired = check_operation_expiry(&history); old_rc = history.exit_status; remap_operation(&history, on_fail, expired); if (expired && (process_expired_result(&history, old_rc) == pcmk_rc_ok)) { goto done; } if (!pcmk__is_bundled(rsc) && pcmk_xe_mask_probe_failure(xml_op)) { mask_probe_failure(&history, old_rc, *last_failure, on_fail); goto done; } if (!pcmk_is_set(rsc->flags, pcmk__rsc_unique)) { parent = uber_parent(rsc); } switch (history.execution_status) { case PCMK_EXEC_PENDING: process_pending_action(&history, *last_failure); goto done; case PCMK_EXEC_DONE: update_resource_state(&history, history.exit_status, *last_failure, on_fail); goto done; case PCMK_EXEC_NOT_INSTALLED: unpack_failure_handling(&history, &failure_strategy, &fail_role); if (failure_strategy == pcmk__on_fail_ignore) { crm_warn("Cannot ignore failed %s of %s on %s: " "Resource agent doesn't exist " QB_XS " status=%d rc=%d id=%s", history.task, rsc->id, pcmk__node_name(node), history.execution_status, history.exit_status, history.id); /* Also for printing it as "FAILED" by marking it as * pcmk__rsc_failed later */ *on_fail = pcmk__on_fail_ban; } resource_location(parent, node, -PCMK_SCORE_INFINITY, "hard-error", rsc->private->scheduler); unpack_rsc_op_failure(&history, failure_strategy, fail_role, last_failure, on_fail); goto done; case PCMK_EXEC_NOT_CONNECTED: if (pcmk__is_pacemaker_remote_node(node) && pcmk_is_set(node->private->remote->flags, pcmk__rsc_managed)) { /* We should never get into a situation where a managed remote * connection resource is considered OK but a resource action * behind the connection gets a "not connected" status. But as a * fail-safe in case a bug or unusual circumstances do lead to * that, ensure the remote connection is considered failed. */ pcmk__set_rsc_flags(node->private->remote, pcmk__rsc_failed|pcmk__rsc_stop_if_failed); } break; // Not done, do error handling case PCMK_EXEC_ERROR: case PCMK_EXEC_ERROR_HARD: case PCMK_EXEC_ERROR_FATAL: case PCMK_EXEC_TIMEOUT: case PCMK_EXEC_NOT_SUPPORTED: case PCMK_EXEC_INVALID: break; // Not done, do error handling default: // No other value should be possible at this point break; } unpack_failure_handling(&history, &failure_strategy, &fail_role); if ((failure_strategy == pcmk__on_fail_ignore) || ((failure_strategy == pcmk__on_fail_restart_container) && (strcmp(history.task, PCMK_ACTION_STOP) == 0))) { char *last_change_s = last_change_str(xml_op); crm_warn("Pretending failed %s (%s%s%s) of %s on %s at %s succeeded " QB_XS " %s", history.task, services_ocf_exitcode_str(history.exit_status), (pcmk__str_empty(history.exit_reason)? "" : ": "), pcmk__s(history.exit_reason, ""), rsc->id, pcmk__node_name(node), last_change_s, history.id); free(last_change_s); update_resource_state(&history, history.expected_exit_status, *last_failure, on_fail); crm_xml_add(xml_op, PCMK_XA_UNAME, node->private->name); pcmk__set_rsc_flags(rsc, pcmk__rsc_ignore_failure); record_failed_op(&history); if ((failure_strategy == pcmk__on_fail_restart_container) && (*on_fail <= pcmk__on_fail_restart)) { *on_fail = failure_strategy; } } else { unpack_rsc_op_failure(&history, failure_strategy, fail_role, last_failure, on_fail); if (history.execution_status == PCMK_EXEC_ERROR_HARD) { uint8_t log_level = LOG_ERR; if (history.exit_status == PCMK_OCF_NOT_INSTALLED) { log_level = LOG_NOTICE; } do_crm_log(log_level, "Preventing %s from restarting on %s because " "of hard failure (%s%s%s) " QB_XS " %s", parent->id, pcmk__node_name(node), services_ocf_exitcode_str(history.exit_status), (pcmk__str_empty(history.exit_reason)? "" : ": "), pcmk__s(history.exit_reason, ""), history.id); resource_location(parent, node, -PCMK_SCORE_INFINITY, "hard-error", rsc->private->scheduler); } else if (history.execution_status == PCMK_EXEC_ERROR_FATAL) { pcmk__sched_err("Preventing %s from restarting anywhere because " "of fatal failure (%s%s%s) " QB_XS " %s", parent->id, services_ocf_exitcode_str(history.exit_status), (pcmk__str_empty(history.exit_reason)? "" : ": "), pcmk__s(history.exit_reason, ""), history.id); resource_location(parent, NULL, -PCMK_SCORE_INFINITY, "fatal-error", rsc->private->scheduler); } } done: pcmk__rsc_trace(rsc, "%s role on %s after %s is %s (next %s)", rsc->id, pcmk__node_name(node), history.id, pcmk_role_text(rsc->private->orig_role), pcmk_role_text(rsc->private->next_role)); } static void add_node_attrs(const xmlNode *xml_obj, pcmk_node_t *node, bool overwrite, pcmk_scheduler_t *scheduler) { const char *cluster_name = NULL; pe_rule_eval_data_t rule_data = { .node_hash = NULL, .now = scheduler->now, .match_data = NULL, .rsc_data = NULL, .op_data = NULL }; pcmk__insert_dup(node->private->attrs, CRM_ATTR_UNAME, node->private->name); pcmk__insert_dup(node->private->attrs, CRM_ATTR_ID, node->private->id); if (pcmk__str_eq(node->private->id, scheduler->dc_uuid, pcmk__str_casei)) { scheduler->dc_node = node; pcmk__insert_dup(node->private->attrs, CRM_ATTR_IS_DC, PCMK_VALUE_TRUE); } else { pcmk__insert_dup(node->private->attrs, CRM_ATTR_IS_DC, PCMK_VALUE_FALSE); } cluster_name = g_hash_table_lookup(scheduler->config_hash, PCMK_OPT_CLUSTER_NAME); if (cluster_name) { pcmk__insert_dup(node->private->attrs, CRM_ATTR_CLUSTER_NAME, cluster_name); } pe__unpack_dataset_nvpairs(xml_obj, PCMK_XE_INSTANCE_ATTRIBUTES, &rule_data, node->private->attrs, NULL, overwrite, scheduler); pe__unpack_dataset_nvpairs(xml_obj, PCMK_XE_UTILIZATION, &rule_data, node->private->utilization, NULL, FALSE, scheduler); if (pcmk__node_attr(node, CRM_ATTR_SITE_NAME, NULL, pcmk__rsc_node_current) == NULL) { const char *site_name = pcmk__node_attr(node, "site-name", NULL, pcmk__rsc_node_current); if (site_name) { pcmk__insert_dup(node->private->attrs, CRM_ATTR_SITE_NAME, site_name); } else if (cluster_name) { /* Default to cluster-name if unset */ pcmk__insert_dup(node->private->attrs, CRM_ATTR_SITE_NAME, cluster_name); } } } static GList * extract_operations(const char *node, const char *rsc, xmlNode * rsc_entry, gboolean active_filter) { int counter = -1; int stop_index = -1; int start_index = -1; xmlNode *rsc_op = NULL; GList *gIter = NULL; GList *op_list = NULL; GList *sorted_op_list = NULL; /* extract operations */ op_list = NULL; sorted_op_list = NULL; for (rsc_op = pcmk__xe_first_child(rsc_entry, NULL, NULL, NULL); rsc_op != NULL; rsc_op = pcmk__xe_next(rsc_op)) { if (pcmk__xe_is(rsc_op, PCMK__XE_LRM_RSC_OP)) { crm_xml_add(rsc_op, PCMK_XA_RESOURCE, rsc); crm_xml_add(rsc_op, PCMK_XA_UNAME, node); op_list = g_list_prepend(op_list, rsc_op); } } if (op_list == NULL) { /* if there are no operations, there is nothing to do */ return NULL; } sorted_op_list = g_list_sort(op_list, sort_op_by_callid); /* create active recurring operations as optional */ if (active_filter == FALSE) { return sorted_op_list; } op_list = NULL; calculate_active_ops(sorted_op_list, &start_index, &stop_index); for (gIter = sorted_op_list; gIter != NULL; gIter = gIter->next) { xmlNode *rsc_op = (xmlNode *) gIter->data; counter++; if (start_index < stop_index) { crm_trace("Skipping %s: not active", pcmk__xe_id(rsc_entry)); break; } else if (counter < start_index) { crm_trace("Skipping %s: old", pcmk__xe_id(rsc_op)); continue; } op_list = g_list_append(op_list, rsc_op); } g_list_free(sorted_op_list); return op_list; } GList * find_operations(const char *rsc, const char *node, gboolean active_filter, pcmk_scheduler_t *scheduler) { GList *output = NULL; GList *intermediate = NULL; xmlNode *tmp = NULL; xmlNode *status = pcmk__xe_first_child(scheduler->input, PCMK_XE_STATUS, NULL, NULL); pcmk_node_t *this_node = NULL; xmlNode *node_state = NULL; CRM_CHECK(status != NULL, return NULL); for (node_state = pcmk__xe_first_child(status, NULL, NULL, NULL); node_state != NULL; node_state = pcmk__xe_next(node_state)) { if (pcmk__xe_is(node_state, PCMK__XE_NODE_STATE)) { const char *uname = crm_element_value(node_state, PCMK_XA_UNAME); if (node != NULL && !pcmk__str_eq(uname, node, pcmk__str_casei)) { continue; } this_node = pcmk_find_node(scheduler, uname); if(this_node == NULL) { CRM_LOG_ASSERT(this_node != NULL); continue; } else if (pcmk__is_pacemaker_remote_node(this_node)) { determine_remote_online_status(scheduler, this_node); } else { determine_online_status(node_state, this_node, scheduler); } if (this_node->details->online || pcmk_is_set(scheduler->flags, pcmk_sched_fencing_enabled)) { /* offline nodes run no resources... * unless stonith is enabled in which case we need to * make sure rsc start events happen after the stonith */ xmlNode *lrm_rsc = NULL; tmp = pcmk__xe_first_child(node_state, PCMK__XE_LRM, NULL, NULL); tmp = pcmk__xe_first_child(tmp, PCMK__XE_LRM_RESOURCES, NULL, NULL); for (lrm_rsc = pcmk__xe_first_child(tmp, NULL, NULL, NULL); lrm_rsc != NULL; lrm_rsc = pcmk__xe_next(lrm_rsc)) { if (pcmk__xe_is(lrm_rsc, PCMK__XE_LRM_RESOURCE)) { const char *rsc_id = crm_element_value(lrm_rsc, PCMK_XA_ID); if (rsc != NULL && !pcmk__str_eq(rsc_id, rsc, pcmk__str_casei)) { continue; } intermediate = extract_operations(uname, rsc_id, lrm_rsc, active_filter); output = g_list_concat(output, intermediate); } } } } } return output; }