diff --git a/lib/cib/cib_native.c b/lib/cib/cib_native.c index 98f7866c36..4f6f755e4a 100644 --- a/lib/cib/cib_native.c +++ b/lib/cib/cib_native.c @@ -1,511 +1,507 @@ /* * Copyright 2004 International Business Machines * Later changes 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 -#ifndef _GNU_SOURCE -# define _GNU_SOURCE -#endif - #include #include #include #include #include #include #include #include #include #include #include #include typedef struct cib_native_opaque_s { char *token; crm_ipc_t *ipc; void (*dnotify_fn) (gpointer user_data); mainloop_io_t *source; } cib_native_opaque_t; static int cib_native_perform_op_delegate(cib_t *cib, const char *op, const char *host, const char *section, xmlNode *data, xmlNode **output_data, int call_options, const char *user_name) { int rc = pcmk_ok; int reply_id = 0; enum crm_ipc_flags ipc_flags = crm_ipc_flags_none; xmlNode *op_msg = NULL; xmlNode *op_reply = NULL; cib_native_opaque_t *native = cib->variant_opaque; if (cib->state == cib_disconnected) { return -ENOTCONN; } if (output_data != NULL) { *output_data = NULL; } if (op == NULL) { crm_err("No operation specified"); return -EINVAL; } if (call_options & cib_sync_call) { pcmk__set_ipc_flags(ipc_flags, "client", crm_ipc_client_response); } rc = cib__create_op(cib, op, host, section, data, call_options, user_name, NULL, &op_msg); if (rc != pcmk_ok) { return rc; } if (pcmk_is_set(call_options, cib_transaction)) { rc = cib__extend_transaction(cib, op_msg); goto done; } crm_trace("Sending %s message to the CIB manager (timeout=%ds)", op, cib->call_timeout); rc = crm_ipc_send(native->ipc, op_msg, ipc_flags, cib->call_timeout * 1000, &op_reply); if (rc < 0) { crm_err("Couldn't perform %s operation (timeout=%ds): %s (%d)", op, cib->call_timeout, pcmk_strerror(rc), rc); rc = -ECOMM; goto done; } crm_log_xml_trace(op_reply, "Reply"); if (!(call_options & cib_sync_call)) { crm_trace("Async call, returning %d", cib->call_id); CRM_CHECK(cib->call_id != 0, rc = -ENOMSG; goto done); rc = cib->call_id; goto done; } rc = pcmk_ok; crm_element_value_int(op_reply, PCMK__XA_CIB_CALLID, &reply_id); if (reply_id == cib->call_id) { xmlNode *wrapper = pcmk__xe_first_child(op_reply, PCMK__XE_CIB_CALLDATA, NULL, NULL); xmlNode *tmp = pcmk__xe_first_child(wrapper, NULL, NULL, NULL); crm_trace("Synchronous reply %d received", reply_id); if (crm_element_value_int(op_reply, PCMK__XA_CIB_RC, &rc) != 0) { rc = -EPROTO; } if (output_data == NULL || (call_options & cib_discard_reply)) { crm_trace("Discarding reply"); } else { *output_data = pcmk__xml_copy(NULL, tmp); } } else if (reply_id <= 0) { crm_err("Received bad reply: No id set"); crm_log_xml_err(op_reply, "Bad reply"); rc = -ENOMSG; goto done; } else { crm_err("Received bad reply: %d (wanted %d)", reply_id, cib->call_id); crm_log_xml_err(op_reply, "Old reply"); rc = -ENOMSG; goto done; } if (op_reply == NULL && cib->state == cib_disconnected) { rc = -ENOTCONN; } else if (rc == pcmk_ok && op_reply == NULL) { rc = -ETIME; } switch (rc) { case pcmk_ok: case -EPERM: break; /* This is an internal value that clients do not and should not care about */ case -pcmk_err_diff_resync: rc = pcmk_ok; break; /* These indicate internal problems */ case -EPROTO: case -ENOMSG: crm_err("Call failed: %s", pcmk_strerror(rc)); if (op_reply) { crm_log_xml_err(op_reply, "Invalid reply"); } break; default: if (!pcmk__str_eq(op, PCMK__CIB_REQUEST_QUERY, pcmk__str_none)) { crm_warn("Call failed: %s", pcmk_strerror(rc)); } } done: if (!crm_ipc_connected(native->ipc)) { crm_err("The CIB manager disconnected"); cib->state = cib_disconnected; } pcmk__xml_free(op_msg); pcmk__xml_free(op_reply); return rc; } static int cib_native_dispatch_internal(const char *buffer, ssize_t length, gpointer userdata) { const char *type = NULL; xmlNode *msg = NULL; cib_t *cib = userdata; crm_trace("dispatching %p", userdata); if (cib == NULL) { crm_err("No CIB!"); return 0; } msg = pcmk__xml_parse(buffer); if (msg == NULL) { crm_warn("Received a NULL message from the CIB manager"); return 0; } /* do callbacks */ type = crm_element_value(msg, PCMK__XA_T); crm_trace("Activating %s callbacks...", type); crm_log_xml_explicit(msg, "cib-reply"); if (pcmk__str_eq(type, PCMK__VALUE_CIB, pcmk__str_none)) { cib_native_callback(cib, msg, 0, 0); } else if (pcmk__str_eq(type, PCMK__VALUE_CIB_NOTIFY, pcmk__str_none)) { g_list_foreach(cib->notify_list, cib_native_notify, msg); } else { crm_err("Unknown message type: %s", type); } pcmk__xml_free(msg); return 0; } static void cib_native_destroy(void *userdata) { cib_t *cib = userdata; cib_native_opaque_t *native = cib->variant_opaque; crm_trace("destroying %p", userdata); cib->state = cib_disconnected; native->source = NULL; native->ipc = NULL; if (native->dnotify_fn) { native->dnotify_fn(userdata); } } static int cib_native_signoff(cib_t *cib) { cib_native_opaque_t *native = cib->variant_opaque; crm_debug("Disconnecting from the CIB manager"); cib_free_notify(cib); remove_cib_op_callback(0, TRUE); 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); } cib->cmds->end_transaction(cib, false, cib_none); cib->state = cib_disconnected; cib->type = cib_no_connection; return pcmk_ok; } static int cib_native_signon_raw(cib_t *cib, const char *name, enum cib_conn_type type, int *async_fd) { int rc = pcmk_ok; const char *channel = NULL; cib_native_opaque_t *native = cib->variant_opaque; xmlNode *hello = NULL; struct ipc_client_callbacks cib_callbacks = { .dispatch = cib_native_dispatch_internal, .destroy = cib_native_destroy }; cib->call_timeout = PCMK__IPC_TIMEOUT; if (type == cib_command) { cib->state = cib_connected_command; channel = PCMK__SERVER_BASED_RW; } else if (type == cib_command_nonblocking) { cib->state = cib_connected_command; channel = PCMK__SERVER_BASED_SHM; } else if (type == cib_query) { cib->state = cib_connected_query; channel = PCMK__SERVER_BASED_RO; } else { return -ENOTCONN; } crm_trace("Connecting %s channel", channel); if (async_fd != NULL) { native->ipc = crm_ipc_new(channel, 0); if (native->ipc != NULL) { rc = pcmk__connect_generic_ipc(native->ipc); if (rc == pcmk_rc_ok) { rc = pcmk__ipc_fd(native->ipc, async_fd); if (rc != pcmk_rc_ok) { crm_info("Couldn't get file descriptor for %s IPC", channel); } } rc = pcmk_rc2legacy(rc); } } else { native->source = mainloop_add_ipc_client(channel, G_PRIORITY_HIGH, 512 * 1024 /* 512k */ , cib, &cib_callbacks); native->ipc = mainloop_get_ipc_client(native->source); } if (rc != pcmk_ok || native->ipc == NULL || !crm_ipc_connected(native->ipc)) { crm_info("Could not connect to CIB manager for %s", name); rc = -ENOTCONN; } if (rc == pcmk_ok) { rc = cib__create_op(cib, CRM_OP_REGISTER, NULL, NULL, NULL, cib_sync_call, NULL, name, &hello); } if (rc == pcmk_ok) { xmlNode *reply = NULL; if (crm_ipc_send(native->ipc, hello, crm_ipc_client_response, -1, &reply) > 0) { const char *msg_type = crm_element_value(reply, PCMK__XA_CIB_OP); crm_log_xml_trace(reply, "reg-reply"); if (!pcmk__str_eq(msg_type, CRM_OP_REGISTER, pcmk__str_casei)) { crm_info("Reply to CIB registration message has unknown type " "'%s'", msg_type); rc = -EPROTO; } else { native->token = crm_element_value_copy(reply, PCMK__XA_CIB_CLIENTID); if (native->token == NULL) { rc = -EPROTO; } } pcmk__xml_free(reply); } else { rc = -ECOMM; } pcmk__xml_free(hello); } if (rc == pcmk_ok) { crm_info("Successfully connected to CIB manager for %s", name); return pcmk_ok; } crm_info("Connection to CIB manager for %s failed: %s", name, pcmk_strerror(rc)); cib_native_signoff(cib); return rc; } static int cib_native_signon(cib_t *cib, const char *name, enum cib_conn_type type) { return cib_native_signon_raw(cib, name, type, NULL); } static int cib_native_free(cib_t *cib) { int rc = pcmk_ok; if (cib->state != cib_disconnected) { rc = cib_native_signoff(cib); } if (cib->state == cib_disconnected) { cib_native_opaque_t *native = cib->variant_opaque; free(native->token); free(cib->variant_opaque); free(cib->cmds); free(cib->user); free(cib); } return rc; } static int cib_native_register_notification(cib_t *cib, const char *callback, int enabled) { int rc = pcmk_ok; xmlNode *notify_msg = pcmk__xe_create(NULL, PCMK__XE_CIB_CALLBACK); cib_native_opaque_t *native = cib->variant_opaque; if (cib->state != cib_disconnected) { crm_xml_add(notify_msg, PCMK__XA_CIB_OP, PCMK__VALUE_CIB_NOTIFY); crm_xml_add(notify_msg, PCMK__XA_CIB_NOTIFY_TYPE, callback); crm_xml_add_int(notify_msg, PCMK__XA_CIB_NOTIFY_ACTIVATE, enabled); rc = crm_ipc_send(native->ipc, notify_msg, crm_ipc_client_response, 1000 * cib->call_timeout, NULL); if (rc <= 0) { crm_trace("Notification not registered: %d", rc); rc = -ECOMM; } } pcmk__xml_free(notify_msg); return rc; } static int cib_native_set_connection_dnotify(cib_t *cib, void (*dnotify) (gpointer user_data)) { cib_native_opaque_t *native = NULL; if (cib == NULL) { crm_err("No CIB!"); return FALSE; } native = cib->variant_opaque; native->dnotify_fn = dnotify; return pcmk_ok; } /*! * \internal * \brief Get the given CIB connection's unique client identifier * * 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 (specifically, \p pcmk_ok) * * \note This is the \p cib_native variant implementation of * \p cib_api_operations_t:client_id(). * \note For \p cib_native objects, \p async_id and \p sync_id are the same. * \note The client ID is assigned during CIB sign-on. */ static int cib_native_client_id(const cib_t *cib, const char **async_id, const char **sync_id) { cib_native_opaque_t *native = cib->variant_opaque; if (async_id != NULL) { *async_id = native->token; } if (sync_id != NULL) { *sync_id = native->token; } return pcmk_ok; } cib_t * cib_native_new(void) { cib_native_opaque_t *native = NULL; cib_t *cib = cib_new_variant(); if (cib == NULL) { return NULL; } native = calloc(1, sizeof(cib_native_opaque_t)); if (native == NULL) { free(cib); return NULL; } cib->variant = cib_native; cib->variant_opaque = native; native->ipc = NULL; native->source = NULL; native->dnotify_fn = NULL; /* assign variant specific ops */ cib->delegate_fn = cib_native_perform_op_delegate; cib->cmds->signon = cib_native_signon; cib->cmds->signon_raw = cib_native_signon_raw; cib->cmds->signoff = cib_native_signoff; cib->cmds->free = cib_native_free; cib->cmds->register_notification = cib_native_register_notification; cib->cmds->set_connection_dnotify = cib_native_set_connection_dnotify; cib->cmds->client_id = cib_native_client_id; return cib; } diff --git a/lib/cluster/membership.c b/lib/cluster/membership.c index cfc861367b..813941e6b4 100644 --- a/lib/cluster/membership.c +++ b/lib/cluster/membership.c @@ -1,1505 +1,1501 @@ /* * 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 -#ifndef _GNU_SOURCE -# define _GNU_SOURCE -#endif - #include // PRIu32 #include // bool #include #include #include #include #include #include #include #include #include #include #include #include "crmcluster_private.h" /* The peer cache remembers cluster nodes that have been seen. This is managed * mostly automatically by libcrmcluster, based on cluster membership events. * * Because cluster nodes can have conflicting names or UUIDs, the hash table key * is a uniquely generated ID. * * @TODO Move caches to pcmk_cluster_t */ GHashTable *pcmk__peer_cache = NULL; /* The remote peer cache tracks pacemaker_remote nodes. While the * value has the same type as the peer cache's, it is tracked separately for * three reasons: pacemaker_remote nodes can't have conflicting names or UUIDs, * so the name (which is also the UUID) is used as the hash table key; there * is no equivalent of membership events, so management is not automatic; and * most users of the peer cache need to exclude pacemaker_remote nodes. * * @TODO That said, using a single cache would be more logical and less * error-prone, so it would be a good idea to merge them one day. * * libcrmcluster provides two avenues for populating the cache: * pcmk__cluster_lookup_remote_node() and pcmk__cluster_forget_remote_node() * directly manage it, while refresh_remote_nodes() populates it via the CIB. * * @TODO Move caches to pcmk_cluster_t */ GHashTable *pcmk__remote_peer_cache = NULL; /* * The CIB cluster node cache tracks cluster nodes that have been seen in * the CIB. It is useful mainly when a caller needs to know about a node that * may no longer be in the membership, but doesn't want to add the node to the * main peer cache tables. */ static GHashTable *cluster_node_cib_cache = NULL; static bool autoreap = true; static bool has_quorum = false; // Flag setting and clearing for pcmk__node_status_t:flags #define set_peer_flags(peer, flags_to_set) do { \ (peer)->flags = pcmk__set_flags_as(__func__, __LINE__, LOG_TRACE, \ "Peer", (peer)->name, \ (peer)->flags, (flags_to_set), \ #flags_to_set); \ } while (0) #define clear_peer_flags(peer, flags_to_clear) do { \ (peer)->flags = pcmk__clear_flags_as(__func__, __LINE__, \ LOG_TRACE, \ "Peer", (peer)->name, \ (peer)->flags, (flags_to_clear), \ #flags_to_clear); \ } while (0) static void update_peer_uname(pcmk__node_status_t *node, const char *uname); static pcmk__node_status_t *find_cib_cluster_node(const char *id, const char *uname); /*! * \internal * \brief Check whether the cluster currently has quorum * * \return \c true if the cluster has quorum, or \c false otherwise */ bool pcmk__cluster_has_quorum(void) { return has_quorum; } /*! * \internal * \brief Set whether the cluster currently has quorum * * \param[in] quorate \c true if the cluster has quorum, or \c false otherwise */ void pcmk__cluster_set_quorum(bool quorate) { has_quorum = quorate; } /*! * \internal * \brief Get the number of Pacemaker Remote nodes that have been seen * * \return Number of cached Pacemaker Remote nodes */ unsigned int pcmk__cluster_num_remote_nodes(void) { if (pcmk__remote_peer_cache == NULL) { return 0U; } return g_hash_table_size(pcmk__remote_peer_cache); } /*! * \internal * \brief Get a remote node cache entry, creating it if necessary * * \param[in] node_name Name of remote node * * \return Cache entry for node on success, or \c NULL (and set \c errno) * otherwise * * \note When creating a new entry, this will leave the node state undetermined. * The caller should also call \c pcmk__update_peer_state() if the state * is known. * \note Because this can add and remove cache entries, callers should not * assume any previously obtained cache entry pointers remain valid. */ pcmk__node_status_t * pcmk__cluster_lookup_remote_node(const char *node_name) { pcmk__node_status_t *node = NULL; char *node_name_copy = NULL; if (node_name == NULL) { errno = EINVAL; return NULL; } /* It's theoretically possible that the node was added to the cluster peer * cache before it was known to be a Pacemaker Remote node. Remove that * entry unless it has a node ID, which means the name actually is * associated with a cluster node. (@TODO return an error in that case?) */ node = pcmk__search_node_caches(0, node_name, pcmk__node_search_cluster_member); if ((node != NULL) && (node->xml_id == NULL)) { /* node_name could be a pointer into the cache entry being removed, so * reassign it to a copy before the original gets freed */ node_name_copy = strdup(node_name); if (node_name_copy == NULL) { errno = ENOMEM; return NULL; } node_name = node_name_copy; pcmk__cluster_forget_cluster_node(0, node_name); } /* Return existing cache entry if one exists */ node = g_hash_table_lookup(pcmk__remote_peer_cache, node_name); if (node) { free(node_name_copy); return node; } /* Allocate a new entry */ node = calloc(1, sizeof(pcmk__node_status_t)); if (node == NULL) { free(node_name_copy); return NULL; } /* Populate the essential information */ set_peer_flags(node, pcmk__node_status_remote); node->xml_id = strdup(node_name); if (node->xml_id == NULL) { free(node); errno = ENOMEM; free(node_name_copy); return NULL; } /* Add the new entry to the cache */ g_hash_table_replace(pcmk__remote_peer_cache, node->xml_id, node); crm_trace("added %s to remote cache", node_name); /* Update the entry's uname, ensuring peer status callbacks are called */ update_peer_uname(node, node_name); free(node_name_copy); return node; } /*! * \internal * \brief Remove a node from the Pacemaker Remote node cache * * \param[in] node_name Name of node to remove from cache * * \note The caller must be careful not to use \p node_name after calling this * function if it might be a pointer into the cache entry being removed. */ void pcmk__cluster_forget_remote_node(const char *node_name) { /* Do a lookup first, because node_name could be a pointer within the entry * being removed -- we can't log it *after* removing it. */ if (g_hash_table_lookup(pcmk__remote_peer_cache, node_name) != NULL) { crm_trace("Removing %s from Pacemaker Remote node cache", node_name); g_hash_table_remove(pcmk__remote_peer_cache, node_name); } } /*! * \internal * \brief Return node status based on a CIB status entry * * \param[in] node_state XML of node state * * \return \c PCMK_VALUE_MEMBER if \c PCMK__XA_IN_CCM is true in * \c PCMK__XE_NODE_STATE, or \c PCMK__VALUE_LOST otherwise */ static const char * remote_state_from_cib(const xmlNode *node_state) { bool in_ccm = false; if ((pcmk__xe_get_bool_attr(node_state, PCMK__XA_IN_CCM, &in_ccm) == pcmk_rc_ok) && in_ccm) { return PCMK_VALUE_MEMBER; } return PCMK__VALUE_LOST; } /* user data for looping through remote node xpath searches */ struct refresh_data { const char *field; /* XML attribute to check for node name */ gboolean has_state; /* whether to update node state based on XML */ }; /*! * \internal * \brief Process one pacemaker_remote node xpath search result * * \param[in] result XML search result * \param[in] user_data what to look for in the XML */ static void remote_cache_refresh_helper(xmlNode *result, void *user_data) { const struct refresh_data *data = user_data; const char *remote = crm_element_value(result, data->field); const char *state = NULL; pcmk__node_status_t *node; CRM_CHECK(remote != NULL, return); /* Determine node's state, if the result has it */ if (data->has_state) { state = remote_state_from_cib(result); } /* Check whether cache already has entry for node */ node = g_hash_table_lookup(pcmk__remote_peer_cache, remote); if (node == NULL) { /* Node is not in cache, so add a new entry for it */ node = pcmk__cluster_lookup_remote_node(remote); CRM_ASSERT(node); if (state) { pcmk__update_peer_state(__func__, node, state, 0); } } else if (pcmk_is_set(node->flags, pcmk__node_status_dirty)) { /* Node is in cache and hasn't been updated already, so mark it clean */ clear_peer_flags(node, pcmk__node_status_dirty); if (state) { pcmk__update_peer_state(__func__, node, state, 0); } } } static void mark_dirty(gpointer key, gpointer value, gpointer user_data) { set_peer_flags((pcmk__node_status_t *) value, pcmk__node_status_dirty); } static gboolean is_dirty(gpointer key, gpointer value, gpointer user_data) { const pcmk__node_status_t *node = value; return pcmk_is_set(node->flags, pcmk__node_status_dirty); } /*! * \internal * \brief Repopulate the remote node cache based on CIB XML * * \param[in] cib CIB XML to parse */ static void refresh_remote_nodes(xmlNode *cib) { struct refresh_data data; pcmk__cluster_init_node_caches(); /* First, we mark all existing cache entries as dirty, * so that later we can remove any that weren't in the CIB. * We don't empty the cache, because we need to detect changes in state. */ g_hash_table_foreach(pcmk__remote_peer_cache, mark_dirty, NULL); /* Look for guest nodes and remote nodes in the status section */ data.field = PCMK_XA_ID; data.has_state = TRUE; crm_foreach_xpath_result(cib, PCMK__XP_REMOTE_NODE_STATUS, remote_cache_refresh_helper, &data); /* Look for guest nodes and remote nodes in the configuration section, * because they may have just been added and not have a status entry yet. * In that case, the cached node state will be left NULL, so that the * peer status callback isn't called until we're sure the node started * successfully. */ data.field = PCMK_XA_VALUE; data.has_state = FALSE; crm_foreach_xpath_result(cib, PCMK__XP_GUEST_NODE_CONFIG, remote_cache_refresh_helper, &data); data.field = PCMK_XA_ID; data.has_state = FALSE; crm_foreach_xpath_result(cib, PCMK__XP_REMOTE_NODE_CONFIG, remote_cache_refresh_helper, &data); /* Remove all old cache entries that weren't seen in the CIB */ g_hash_table_foreach_remove(pcmk__remote_peer_cache, is_dirty, NULL); } /*! * \internal * \brief Check whether a node is an active cluster node * * Remote nodes are never considered active. This guarantees that they can never * become DC. * * \param[in] node Node to check * * \return \c true if the node is an active cluster node, or \c false otherwise */ bool pcmk__cluster_is_node_active(const pcmk__node_status_t *node) { const enum pcmk_cluster_layer cluster_layer = pcmk_get_cluster_layer(); if ((node == NULL) || pcmk_is_set(node->flags, pcmk__node_status_remote)) { return false; } switch (cluster_layer) { case pcmk_cluster_layer_corosync: #if SUPPORT_COROSYNC return pcmk__corosync_is_peer_active(node); #else break; #endif // SUPPORT_COROSYNC default: break; } crm_err("Unhandled cluster layer: %s", pcmk_cluster_layer_text(cluster_layer)); return false; } /*! * \internal * \brief Check if a node's entry should be removed from the cluster node cache * * A node should be removed from the cache if it's inactive and matches another * \c pcmk__node_status_t (the search object). The node is considered a * mismatch if any of the following are true: * * The search object is \c NULL. * * The search object has an ID set and the cached node's ID does not match it. * * The search object does not have an ID set, and the cached node's name does * not match the search node's name. (If both names are \c NULL, it's a * match.) * * Otherwise, the node is considered a match. * * Note that if the search object has both an ID and a name set, the name is * ignored for matching purposes. * * \param[in] key Ignored * \param[in] value \c pcmk__node_status_t object from cluster node cache * \param[in] user_data \c pcmk__node_status_t object to match against (search * object) * * \return \c TRUE if the node entry should be removed from \c pcmk__peer_cache, * or \c FALSE otherwise */ static gboolean should_forget_cluster_node(gpointer key, gpointer value, gpointer user_data) { pcmk__node_status_t *node = value; pcmk__node_status_t *search = user_data; if (search == NULL) { return FALSE; } if ((search->cluster_layer_id != 0) && (node->cluster_layer_id != search->cluster_layer_id)) { return FALSE; } if ((search->cluster_layer_id == 0) && !pcmk__str_eq(node->name, search->name, pcmk__str_casei)) { // @TODO Consider name even if ID is set? return FALSE; } if (pcmk__cluster_is_node_active(value)) { return FALSE; } crm_info("Removing node with name %s and cluster layer ID " PRIu32 " from membership cache", pcmk__s(node->name, "(unknown)"), node->cluster_layer_id); return TRUE; } /*! * \internal * \brief Remove one or more inactive nodes from the cluster node cache * * All inactive nodes matching \p id and \p node_name as described in * \c should_forget_cluster_node documentation are removed from the cache. * * If \p id is 0 and \p node_name is \c NULL, all inactive nodes are removed * from the cache regardless of ID and name. This differs from clearing the * cache, in that entries for active nodes are preserved. * * \param[in] id ID of node to remove from cache (0 to ignore) * \param[in] node_name Name of node to remove from cache (ignored if \p id is * nonzero) * * \note \p node_name is not modified directly, but it will be freed if it's a * pointer into a cache entry that is removed. */ void pcmk__cluster_forget_cluster_node(uint32_t id, const char *node_name) { pcmk__node_status_t search = { 0, }; char *criterion = NULL; // For logging guint matches = 0; if (pcmk__peer_cache == NULL) { crm_trace("Membership cache not initialized, ignoring removal request"); return; } search.cluster_layer_id = id; search.name = pcmk__str_copy(node_name); // May log after original freed if (id > 0) { criterion = crm_strdup_printf("cluster layer ID %" PRIu32, id); } else if (node_name != NULL) { criterion = crm_strdup_printf("name %s", node_name); } matches = g_hash_table_foreach_remove(pcmk__peer_cache, should_forget_cluster_node, &search); if (matches > 0) { if (criterion != NULL) { crm_notice("Removed %u inactive node%s with %s from the membership " "cache", matches, pcmk__plural_s(matches), criterion); } else { crm_notice("Removed all (%u) inactive cluster nodes from the " "membership cache", matches); } } else { crm_info("No inactive cluster nodes%s%s to remove from the membership " "cache", ((criterion != NULL)? " with " : ""), pcmk__s(criterion, "")); } free(search.name); free(criterion); } static void count_peer(gpointer key, gpointer value, gpointer user_data) { unsigned int *count = user_data; pcmk__node_status_t *node = value; if (pcmk__cluster_is_node_active(node)) { *count = *count + 1; } } /*! * \internal * \brief Get the number of active cluster nodes that have been seen * * Remote nodes are never considered active. This guarantees that they can never * become DC. * * \return Number of active nodes in the cluster node cache */ unsigned int pcmk__cluster_num_active_nodes(void) { unsigned int count = 0; if (pcmk__peer_cache != NULL) { g_hash_table_foreach(pcmk__peer_cache, count_peer, &count); } return count; } static void destroy_crm_node(gpointer data) { pcmk__node_status_t *node = data; crm_trace("Destroying entry for node %" PRIu32 ": %s", node->cluster_layer_id, node->name); free(node->name); free(node->state); free(node->xml_id); free(node->user_data); free(node->expected); free(node->conn_host); free(node); } /*! * \internal * \brief Initialize node caches */ void pcmk__cluster_init_node_caches(void) { if (pcmk__peer_cache == NULL) { pcmk__peer_cache = pcmk__strikey_table(free, destroy_crm_node); } if (pcmk__remote_peer_cache == NULL) { pcmk__remote_peer_cache = pcmk__strikey_table(NULL, destroy_crm_node); } if (cluster_node_cib_cache == NULL) { cluster_node_cib_cache = pcmk__strikey_table(free, destroy_crm_node); } } /*! * \internal * \brief Initialize node caches */ void pcmk__cluster_destroy_node_caches(void) { if (pcmk__peer_cache != NULL) { crm_trace("Destroying peer cache with %d members", g_hash_table_size(pcmk__peer_cache)); g_hash_table_destroy(pcmk__peer_cache); pcmk__peer_cache = NULL; } if (pcmk__remote_peer_cache != NULL) { crm_trace("Destroying remote peer cache with %d members", pcmk__cluster_num_remote_nodes()); g_hash_table_destroy(pcmk__remote_peer_cache); pcmk__remote_peer_cache = NULL; } if (cluster_node_cib_cache != NULL) { crm_trace("Destroying configured cluster node cache with %d members", g_hash_table_size(cluster_node_cib_cache)); g_hash_table_destroy(cluster_node_cib_cache); cluster_node_cib_cache = NULL; } } static void (*peer_status_callback)(enum pcmk__node_update, pcmk__node_status_t *, const void *) = NULL; /*! * \internal * \brief Set a client function that will be called after peer status changes * * \param[in] dispatch Pointer to function to use as callback * * \note Client callbacks should do only client-specific handling. Callbacks * must not add or remove entries in the peer caches. */ void pcmk__cluster_set_status_callback(void (*dispatch)(enum pcmk__node_update, pcmk__node_status_t *, const void *)) { // @TODO Improve documentation of peer_status_callback peer_status_callback = dispatch; } /*! * \internal * \brief Tell the library whether to automatically reap lost nodes * * If \c true (the default), calling \c crm_update_peer_proc() will also update * the peer state to \c PCMK_VALUE_MEMBER or \c PCMK__VALUE_LOST, and updating * the peer state will reap peers whose state changes to anything other than * \c PCMK_VALUE_MEMBER. * * Callers should leave this enabled unless they plan to manage the cache * separately on their own. * * \param[in] enable \c true to enable automatic reaping, \c false to disable */ void pcmk__cluster_set_autoreap(bool enable) { autoreap = enable; } static void dump_peer_hash(int level, const char *caller) { GHashTableIter iter; const char *id = NULL; pcmk__node_status_t *node = NULL; g_hash_table_iter_init(&iter, pcmk__peer_cache); while (g_hash_table_iter_next(&iter, (gpointer *) &id, (gpointer *) &node)) { do_crm_log(level, "%s: Node %" PRIu32 "/%s = %p - %s", caller, node->cluster_layer_id, node->name, node, id); } } static gboolean hash_find_by_data(gpointer key, gpointer value, gpointer user_data) { return value == user_data; } /*! * \internal * \brief Search cluster member node cache * * \param[in] id If not 0, cluster node ID to search for * \param[in] uname If not NULL, node name to search for * \param[in] uuid If not NULL while id is 0, node UUID instead of cluster * node ID to search for * * \return Cluster node cache entry if found, otherwise NULL */ static pcmk__node_status_t * search_cluster_member_cache(unsigned int id, const char *uname, const char *uuid) { GHashTableIter iter; pcmk__node_status_t *node = NULL; pcmk__node_status_t *by_id = NULL; pcmk__node_status_t *by_name = NULL; CRM_ASSERT(id > 0 || uname != NULL); pcmk__cluster_init_node_caches(); if (uname != NULL) { g_hash_table_iter_init(&iter, pcmk__peer_cache); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &node)) { if (pcmk__str_eq(node->name, uname, pcmk__str_casei)) { crm_trace("Name match: %s", node->name); by_name = node; break; } } } if (id > 0) { g_hash_table_iter_init(&iter, pcmk__peer_cache); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &node)) { if (node->cluster_layer_id == id) { crm_trace("ID match: %" PRIu32, node->cluster_layer_id); by_id = node; break; } } } else if (uuid != NULL) { g_hash_table_iter_init(&iter, pcmk__peer_cache); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &node)) { if (pcmk__str_eq(node->xml_id, uuid, pcmk__str_casei)) { crm_trace("UUID match: %s", node->xml_id); by_id = node; break; } } } node = by_id; /* Good default */ if(by_id == by_name) { /* Nothing to do if they match (both NULL counts) */ crm_trace("Consistent: %p for %u/%s", by_id, id, uname); } else if(by_id == NULL && by_name) { crm_trace("Only one: %p for %u/%s", by_name, id, uname); if (id && by_name->cluster_layer_id) { dump_peer_hash(LOG_WARNING, __func__); crm_crit("Nodes %u and %" PRIu32 " share the same name '%s'", id, by_name->cluster_layer_id, uname); node = NULL; /* Create a new one */ } else { node = by_name; } } else if(by_name == NULL && by_id) { crm_trace("Only one: %p for %u/%s", by_id, id, uname); if ((uname != NULL) && (by_id->name != NULL)) { dump_peer_hash(LOG_WARNING, __func__); crm_crit("Nodes '%s' and '%s' share the same cluster nodeid %u: " "assuming '%s' is correct", uname, by_id->name, id, uname); } } else if ((uname != NULL) && (by_id->name != NULL)) { if (pcmk__str_eq(uname, by_id->name, pcmk__str_casei)) { crm_notice("Node '%s' has changed its cluster layer ID " "from %" PRIu32 " to %" PRIu32, by_id->name, by_name->cluster_layer_id, by_id->cluster_layer_id); g_hash_table_foreach_remove(pcmk__peer_cache, hash_find_by_data, by_name); } else { crm_warn("Nodes '%s' and '%s' share the same cluster nodeid: %u %s", by_id->name, by_name->name, id, uname); dump_peer_hash(LOG_INFO, __func__); crm_abort(__FILE__, __func__, __LINE__, "member weirdness", TRUE, TRUE); } } else if ((id > 0) && (by_name->cluster_layer_id > 0)) { crm_warn("Nodes %" PRIu32 " and %" PRIu32 " share the same name: '%s'", by_id->cluster_layer_id, by_name->cluster_layer_id, uname); } else { /* Simple merge */ /* Only corosync-based clusters use node IDs. The functions that call * pcmk__update_peer_state() and crm_update_peer_proc() only know * nodeid, so 'by_id' is authoritative when merging. */ dump_peer_hash(LOG_DEBUG, __func__); crm_info("Merging %p into %p", by_name, by_id); g_hash_table_foreach_remove(pcmk__peer_cache, hash_find_by_data, by_name); } return node; } /*! * \internal * \brief Search caches for a node (cluster or Pacemaker Remote) * * \param[in] id If not 0, cluster node ID to search for * \param[in] uname If not NULL, node name to search for * \param[in] flags Group of enum pcmk__node_search_flags * * \return Node cache entry if found, otherwise NULL */ pcmk__node_status_t * pcmk__search_node_caches(unsigned int id, const char *uname, uint32_t flags) { pcmk__node_status_t *node = NULL; CRM_ASSERT(id > 0 || uname != NULL); pcmk__cluster_init_node_caches(); if ((uname != NULL) && pcmk_is_set(flags, pcmk__node_search_remote)) { node = g_hash_table_lookup(pcmk__remote_peer_cache, uname); } if ((node == NULL) && pcmk_is_set(flags, pcmk__node_search_cluster_member)) { node = search_cluster_member_cache(id, uname, NULL); } if ((node == NULL) && pcmk_is_set(flags, pcmk__node_search_cluster_cib)) { char *id_str = (id == 0)? NULL : crm_strdup_printf("%u", id); node = find_cib_cluster_node(id_str, uname); free(id_str); } return node; } /*! * \internal * \brief Purge a node from cache (both cluster and Pacemaker Remote) * * \param[in] node_name If not NULL, purge only nodes with this name * \param[in] node_id If not 0, purge cluster nodes only if they have this ID * * \note If \p node_name is NULL and \p node_id is 0, no nodes will be purged. * If \p node_name is not NULL and \p node_id is not 0, Pacemaker Remote * nodes that match \p node_name will be purged, and cluster nodes that * match both \p node_name and \p node_id will be purged. * \note The caller must be careful not to use \p node_name after calling this * function if it might be a pointer into a cache entry being removed. */ void pcmk__purge_node_from_cache(const char *node_name, uint32_t node_id) { char *node_name_copy = NULL; if ((node_name == NULL) && (node_id == 0U)) { return; } // Purge from Pacemaker Remote node cache if ((node_name != NULL) && (g_hash_table_lookup(pcmk__remote_peer_cache, node_name) != NULL)) { /* node_name could be a pointer into the cache entry being purged, * so reassign it to a copy before the original gets freed */ node_name_copy = pcmk__str_copy(node_name); node_name = node_name_copy; crm_trace("Purging %s from Pacemaker Remote node cache", node_name); g_hash_table_remove(pcmk__remote_peer_cache, node_name); } pcmk__cluster_forget_cluster_node(node_id, node_name); free(node_name_copy); } #if SUPPORT_COROSYNC static guint remove_conflicting_peer(pcmk__node_status_t *node) { int matches = 0; GHashTableIter iter; pcmk__node_status_t *existing_node = NULL; if ((node->cluster_layer_id == 0) || (node->name == NULL)) { return 0; } if (!pcmk__corosync_has_nodelist()) { return 0; } g_hash_table_iter_init(&iter, pcmk__peer_cache); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &existing_node)) { if ((existing_node->cluster_layer_id > 0) && (existing_node->cluster_layer_id != node->cluster_layer_id) && pcmk__str_eq(existing_node->name, node->name, pcmk__str_casei)) { if (pcmk__cluster_is_node_active(existing_node)) { continue; } crm_warn("Removing cached offline node %" PRIu32 "/%s which has " "conflicting name with %" PRIu32, existing_node->cluster_layer_id, existing_node->name, node->cluster_layer_id); g_hash_table_iter_remove(&iter); matches++; } } return matches; } #endif /*! * \internal * \brief Get a cluster node cache entry, possibly creating one if not found * * If \c pcmk__node_search_cluster_member is set in \p flags, the return value * is guaranteed not to be \c NULL. A new cache entry is created if one does not * already exist. * * \param[in] id If not 0, cluster node ID to search for * \param[in] uname If not NULL, node name to search for * \param[in] uuid If not NULL while id is 0, node UUID instead of cluster * node ID to search for * \param[in] flags Group of enum pcmk__node_search_flags * * \return (Possibly newly created) cluster node cache entry */ /* coverity[-alloc] Memory is referenced in one or both hashtables */ pcmk__node_status_t * pcmk__get_node(unsigned int id, const char *uname, const char *uuid, uint32_t flags) { pcmk__node_status_t *node = NULL; char *uname_lookup = NULL; CRM_ASSERT(id > 0 || uname != NULL); pcmk__cluster_init_node_caches(); // Check the Pacemaker Remote node cache first if (pcmk_is_set(flags, pcmk__node_search_remote)) { node = g_hash_table_lookup(pcmk__remote_peer_cache, uname); if (node != NULL) { return node; } } if (!pcmk_is_set(flags, pcmk__node_search_cluster_member)) { return NULL; } node = search_cluster_member_cache(id, uname, uuid); /* if uname wasn't provided, and find_peer did not turn up a uname based on id. * we need to do a lookup of the node name using the id in the cluster membership. */ if ((uname == NULL) && ((node == NULL) || (node->name == NULL))) { uname_lookup = pcmk__cluster_node_name(id); } if (uname_lookup) { uname = uname_lookup; crm_trace("Inferred a name of '%s' for node %u", uname, id); /* try to turn up the node one more time now that we know the uname. */ if (node == NULL) { node = search_cluster_member_cache(id, uname, uuid); } } if (node == NULL) { char *uniqueid = crm_generate_uuid(); node = pcmk__assert_alloc(1, sizeof(pcmk__node_status_t)); crm_info("Created entry %s/%p for node %s/%u (%d total)", uniqueid, node, uname, id, 1 + g_hash_table_size(pcmk__peer_cache)); g_hash_table_replace(pcmk__peer_cache, uniqueid, node); } if ((id > 0) && (uname != NULL) && ((node->cluster_layer_id == 0) || (node->name == NULL))) { crm_info("Node %u is now known as %s", id, uname); } if ((id > 0) && (node->cluster_layer_id == 0)) { node->cluster_layer_id = id; } if ((uname != NULL) && (node->name == NULL)) { update_peer_uname(node, uname); } if (node->xml_id == NULL) { if (uuid == NULL) { uuid = pcmk__cluster_node_uuid(node); } if (uuid) { crm_info("Node %u has uuid %s", id, uuid); } else { crm_info("Cannot obtain a UUID for node %u/%s", id, node->name); } } free(uname_lookup); return node; } /*! * \internal * \brief Update a node's uname * * \param[in,out] node Node object to update * \param[in] uname New name to set * * \note This function should not be called within a peer cache iteration, * because in some cases it can remove conflicting cache entries, * which would invalidate the iterator. */ static void update_peer_uname(pcmk__node_status_t *node, const char *uname) { CRM_CHECK(uname != NULL, crm_err("Bug: can't update node name without name"); return); CRM_CHECK(node != NULL, crm_err("Bug: can't update node name to %s without node", uname); return); if (pcmk__str_eq(uname, node->name, pcmk__str_casei)) { crm_debug("Node name '%s' did not change", uname); return; } for (const char *c = uname; *c; ++c) { if ((*c >= 'A') && (*c <= 'Z')) { crm_warn("Node names with capitals are discouraged, consider changing '%s'", uname); break; } } pcmk__str_update(&node->name, uname); if (peer_status_callback != NULL) { peer_status_callback(pcmk__node_update_name, node, NULL); } #if SUPPORT_COROSYNC if ((pcmk_get_cluster_layer() == pcmk_cluster_layer_corosync) && !pcmk_is_set(node->flags, pcmk__node_status_remote)) { remove_conflicting_peer(node); } #endif } /*! * \internal * \brief Get log-friendly string equivalent of a process flag * * \param[in] proc Process flag * * \return Log-friendly string equivalent of \p proc */ static inline const char * proc2text(enum crm_proc_flag proc) { const char *text = "unknown"; switch (proc) { case crm_proc_none: text = "none"; break; case crm_proc_cpg: text = "corosync-cpg"; break; } return text; } /*! * \internal * \brief Update a node's process information (and potentially state) * * \param[in] source Caller's function name (for log messages) * \param[in,out] node Node object to update * \param[in] flag Bitmask of new process information * \param[in] status node status (online, offline, etc.) * * \return NULL if any node was reaped from peer caches, value of node otherwise * * \note If this function returns NULL, the supplied node object was likely * freed and should not be used again. This function should not be * called within a cache iteration if reaping is possible, otherwise * reaping could invalidate the iterator. */ pcmk__node_status_t * crm_update_peer_proc(const char *source, pcmk__node_status_t *node, uint32_t flag, const char *status) { uint32_t last = 0; gboolean changed = FALSE; CRM_CHECK(node != NULL, crm_err("%s: Could not set %s to %s for NULL", source, proc2text(flag), status); return NULL); /* Pacemaker doesn't spawn processes on remote nodes */ if (pcmk_is_set(node->flags, pcmk__node_status_remote)) { return node; } last = node->processes; if (status == NULL) { node->processes = flag; if (node->processes != last) { changed = TRUE; } } else if (pcmk__str_eq(status, PCMK_VALUE_ONLINE, pcmk__str_casei)) { if ((node->processes & flag) != flag) { node->processes = pcmk__set_flags_as(__func__, __LINE__, LOG_TRACE, "Peer process", node->name, node->processes, flag, "processes"); changed = TRUE; } } else if (node->processes & flag) { node->processes = pcmk__clear_flags_as(__func__, __LINE__, LOG_TRACE, "Peer process", node->name, node->processes, flag, "processes"); changed = TRUE; } if (changed) { if (status == NULL && flag <= crm_proc_none) { crm_info("%s: Node %s[%" PRIu32 "] - all processes are now offline", source, node->name, node->cluster_layer_id); } else { crm_info("%s: Node %s[%" PRIu32 "] - %s is now %s", source, node->name, node->cluster_layer_id, proc2text(flag), status); } if (pcmk_is_set(node->processes, crm_get_cluster_proc())) { node->when_online = time(NULL); } else { node->when_online = 0; } /* Call the client callback first, then update the peer state, * in case the node will be reaped */ if (peer_status_callback != NULL) { peer_status_callback(pcmk__node_update_processes, node, &last); } /* The client callback shouldn't touch the peer caches, * but as a safety net, bail if the peer cache was destroyed. */ if (pcmk__peer_cache == NULL) { return NULL; } if (autoreap) { const char *peer_state = NULL; if (pcmk_is_set(node->processes, crm_get_cluster_proc())) { peer_state = PCMK_VALUE_MEMBER; } else { peer_state = PCMK__VALUE_LOST; } node = pcmk__update_peer_state(__func__, node, peer_state, 0); } } else { crm_trace("%s: Node %s[%" PRIu32 "] - %s is unchanged (%s)", source, node->name, node->cluster_layer_id, proc2text(flag), status); } return node; } /*! * \internal * \brief Update a cluster node cache entry's expected join state * * \param[in] source Caller's function name (for logging) * \param[in,out] node Node to update * \param[in] expected Node's new join state */ void pcmk__update_peer_expected(const char *source, pcmk__node_status_t *node, const char *expected) { char *last = NULL; gboolean changed = FALSE; CRM_CHECK(node != NULL, crm_err("%s: Could not set 'expected' to %s", source, expected); return); /* Remote nodes don't participate in joins */ if (pcmk_is_set(node->flags, pcmk__node_status_remote)) { return; } last = node->expected; if (expected != NULL && !pcmk__str_eq(node->expected, expected, pcmk__str_casei)) { node->expected = strdup(expected); changed = TRUE; } if (changed) { crm_info("%s: Node %s[%" PRIu32 "] - expected state is now %s (was %s)", source, node->name, node->cluster_layer_id, expected, last); free(last); } else { crm_trace("%s: Node %s[%" PRIu32 "] - expected state is unchanged (%s)", source, node->name, node->cluster_layer_id, expected); } } /*! * \internal * \brief Update a node's state and membership information * * \param[in] source Caller's function name (for log messages) * \param[in,out] node Node object to update * \param[in] state Node's new state * \param[in] membership Node's new membership ID * \param[in,out] iter If not NULL, pointer to node's peer cache iterator * * \return NULL if any node was reaped, value of node otherwise * * \note If this function returns NULL, the supplied node object was likely * freed and should not be used again. This function may be called from * within a peer cache iteration if the iterator is supplied. */ static pcmk__node_status_t * update_peer_state_iter(const char *source, pcmk__node_status_t *node, const char *state, uint64_t membership, GHashTableIter *iter) { gboolean is_member; CRM_CHECK(node != NULL, crm_err("Could not set state for unknown host to %s" QB_XS " source=%s", state, source); return NULL); is_member = pcmk__str_eq(state, PCMK_VALUE_MEMBER, pcmk__str_none); if (is_member) { node->when_lost = 0; if (membership) { node->membership_id = membership; } } if (state && !pcmk__str_eq(node->state, state, pcmk__str_casei)) { char *last = node->state; if (is_member) { node->when_member = time(NULL); } else { node->when_member = 0; } node->state = strdup(state); crm_notice("Node %s state is now %s " QB_XS " nodeid=%" PRIu32 " previous=%s source=%s", node->name, state, node->cluster_layer_id, pcmk__s(last, "unknown"), source); if (peer_status_callback != NULL) { peer_status_callback(pcmk__node_update_state, node, last); } free(last); if (autoreap && !is_member && !pcmk_is_set(node->flags, pcmk__node_status_remote)) { /* We only autoreap from the peer cache, not the remote peer cache, * because the latter should be managed only by * refresh_remote_nodes(). */ if(iter) { crm_notice("Purged 1 peer with cluster layer ID=" PRIu32 "and/or name=%s from the membership cache", node->cluster_layer_id, node->name); g_hash_table_iter_remove(iter); } else { pcmk__cluster_forget_cluster_node(node->cluster_layer_id, node->name); } node = NULL; } } else { crm_trace("Node %s state is unchanged (%s) " QB_XS " nodeid=%" PRIu32 " source=%s", node->name, state, node->cluster_layer_id, source); } return node; } /*! * \brief Update a node's state and membership information * * \param[in] source Caller's function name (for log messages) * \param[in,out] node Node object to update * \param[in] state Node's new state * \param[in] membership Node's new membership ID * * \return NULL if any node was reaped, value of node otherwise * * \note If this function returns NULL, the supplied node object was likely * freed and should not be used again. This function should not be * called within a cache iteration if reaping is possible, * otherwise reaping could invalidate the iterator. */ pcmk__node_status_t * pcmk__update_peer_state(const char *source, pcmk__node_status_t *node, const char *state, uint64_t membership) { return update_peer_state_iter(source, node, state, membership, NULL); } /*! * \internal * \brief Reap all nodes from cache whose membership information does not match * * \param[in] membership Membership ID of nodes to keep */ void pcmk__reap_unseen_nodes(uint64_t membership) { GHashTableIter iter; pcmk__node_status_t *node = NULL; crm_trace("Reaping unseen nodes..."); g_hash_table_iter_init(&iter, pcmk__peer_cache); while (g_hash_table_iter_next(&iter, NULL, (gpointer *)&node)) { if (node->membership_id != membership) { if (node->state) { /* Calling update_peer_state_iter() allows us to remove the node * from pcmk__peer_cache without invalidating our iterator */ update_peer_state_iter(__func__, node, PCMK__VALUE_LOST, membership, &iter); } else { crm_info("State of node %s[%" PRIu32 "] is still unknown", node->name, node->cluster_layer_id); } } } } static pcmk__node_status_t * find_cib_cluster_node(const char *id, const char *uname) { GHashTableIter iter; pcmk__node_status_t *node = NULL; pcmk__node_status_t *by_id = NULL; pcmk__node_status_t *by_name = NULL; if (uname) { g_hash_table_iter_init(&iter, cluster_node_cib_cache); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &node)) { if (pcmk__str_eq(node->name, uname, pcmk__str_casei)) { crm_trace("Name match: %s = %p", node->name, node); by_name = node; break; } } } if (id) { g_hash_table_iter_init(&iter, cluster_node_cib_cache); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &node)) { if (pcmk__str_eq(node->xml_id, id, pcmk__str_casei)) { crm_trace("ID match: %s= %p", id, node); by_id = node; break; } } } node = by_id; /* Good default */ if (by_id == by_name) { /* Nothing to do if they match (both NULL counts) */ crm_trace("Consistent: %p for %s/%s", by_id, id, uname); } else if (by_id == NULL && by_name) { crm_trace("Only one: %p for %s/%s", by_name, id, uname); if (id) { node = NULL; } else { node = by_name; } } else if (by_name == NULL && by_id) { crm_trace("Only one: %p for %s/%s", by_id, id, uname); if (uname) { node = NULL; } } else if ((uname != NULL) && (by_id->name != NULL) && pcmk__str_eq(uname, by_id->name, pcmk__str_casei)) { /* Multiple nodes have the same uname in the CIB. * Return by_id. */ } else if ((id != NULL) && (by_name->xml_id != NULL) && pcmk__str_eq(id, by_name->xml_id, pcmk__str_casei)) { /* Multiple nodes have the same id in the CIB. * Return by_name. */ node = by_name; } else { node = NULL; } if (node == NULL) { crm_debug("Couldn't find node%s%s%s%s", id? " " : "", id? id : "", uname? " with name " : "", uname? uname : ""); } return node; } static void cluster_node_cib_cache_refresh_helper(xmlNode *xml_node, void *user_data) { const char *id = crm_element_value(xml_node, PCMK_XA_ID); const char *uname = crm_element_value(xml_node, PCMK_XA_UNAME); pcmk__node_status_t * node = NULL; CRM_CHECK(id != NULL && uname !=NULL, return); node = find_cib_cluster_node(id, uname); if (node == NULL) { char *uniqueid = crm_generate_uuid(); node = pcmk__assert_alloc(1, sizeof(pcmk__node_status_t)); node->name = pcmk__str_copy(uname); node->xml_id = pcmk__str_copy(id); g_hash_table_replace(cluster_node_cib_cache, uniqueid, node); } else if (pcmk_is_set(node->flags, pcmk__node_status_dirty)) { pcmk__str_update(&node->name, uname); /* Node is in cache and hasn't been updated already, so mark it clean */ clear_peer_flags(node, pcmk__node_status_dirty); } } static void refresh_cluster_node_cib_cache(xmlNode *cib) { pcmk__cluster_init_node_caches(); g_hash_table_foreach(cluster_node_cib_cache, mark_dirty, NULL); crm_foreach_xpath_result(cib, PCMK__XP_MEMBER_NODE_CONFIG, cluster_node_cib_cache_refresh_helper, NULL); // Remove all old cache entries that weren't seen in the CIB g_hash_table_foreach_remove(cluster_node_cib_cache, is_dirty, NULL); } void pcmk__refresh_node_caches_from_cib(xmlNode *cib) { refresh_remote_nodes(cib); refresh_cluster_node_cib_cache(cib); } // Deprecated functions kept only for backward API compatibility // LCOV_EXCL_START #include void crm_peer_init(void) { pcmk__cluster_init_node_caches(); } // LCOV_EXCL_STOP // End deprecated API diff --git a/lib/common/actions.c b/lib/common/actions.c index 815bc5b3ea..ecba4765b5 100644 --- a/lib/common/actions.c +++ b/lib/common/actions.c @@ -1,590 +1,586 @@ /* * 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 -#ifndef _GNU_SOURCE -# define _GNU_SOURCE -#endif - #include #include #include #include #include #include #include #include #include #include #include /*! * \internal * \brief Get string equivalent of an action type * * \param[in] action Action type * * \return Static string describing \p action */ const char * pcmk__action_text(enum pcmk__action_type action) { switch (action) { case pcmk__action_stop: return PCMK_ACTION_STOP; case pcmk__action_stopped: return PCMK_ACTION_STOPPED; case pcmk__action_start: return PCMK_ACTION_START; case pcmk__action_started: return PCMK_ACTION_RUNNING; case pcmk__action_shutdown: return PCMK_ACTION_DO_SHUTDOWN; case pcmk__action_fence: return PCMK_ACTION_STONITH; case pcmk__action_monitor: return PCMK_ACTION_MONITOR; case pcmk__action_notify: return PCMK_ACTION_NOTIFY; case pcmk__action_notified: return PCMK_ACTION_NOTIFIED; case pcmk__action_promote: return PCMK_ACTION_PROMOTE; case pcmk__action_promoted: return PCMK_ACTION_PROMOTED; case pcmk__action_demote: return PCMK_ACTION_DEMOTE; case pcmk__action_demoted: return PCMK_ACTION_DEMOTED; default: // pcmk__action_unspecified or invalid return "no_action"; } } /*! * \internal * \brief Parse an action type from an action name * * \param[in] action_name Action name * * \return Action type corresponding to \p action_name */ enum pcmk__action_type pcmk__parse_action(const char *action_name) { if (pcmk__str_eq(action_name, PCMK_ACTION_STOP, pcmk__str_none)) { return pcmk__action_stop; } else if (pcmk__str_eq(action_name, PCMK_ACTION_STOPPED, pcmk__str_none)) { return pcmk__action_stopped; } else if (pcmk__str_eq(action_name, PCMK_ACTION_START, pcmk__str_none)) { return pcmk__action_start; } else if (pcmk__str_eq(action_name, PCMK_ACTION_RUNNING, pcmk__str_none)) { return pcmk__action_started; } else if (pcmk__str_eq(action_name, PCMK_ACTION_DO_SHUTDOWN, pcmk__str_none)) { return pcmk__action_shutdown; } else if (pcmk__str_eq(action_name, PCMK_ACTION_STONITH, pcmk__str_none)) { return pcmk__action_fence; } else if (pcmk__str_eq(action_name, PCMK_ACTION_MONITOR, pcmk__str_none)) { return pcmk__action_monitor; } else if (pcmk__str_eq(action_name, PCMK_ACTION_NOTIFY, pcmk__str_none)) { return pcmk__action_notify; } else if (pcmk__str_eq(action_name, PCMK_ACTION_NOTIFIED, pcmk__str_none)) { return pcmk__action_notified; } else if (pcmk__str_eq(action_name, PCMK_ACTION_PROMOTE, pcmk__str_none)) { return pcmk__action_promote; } else if (pcmk__str_eq(action_name, PCMK_ACTION_DEMOTE, pcmk__str_none)) { return pcmk__action_demote; } else if (pcmk__str_eq(action_name, PCMK_ACTION_PROMOTED, pcmk__str_none)) { return pcmk__action_promoted; } else if (pcmk__str_eq(action_name, PCMK_ACTION_DEMOTED, pcmk__str_none)) { return pcmk__action_demoted; } return pcmk__action_unspecified; } /*! * \internal * \brief Get string equivalent of a failure handling type * * \param[in] on_fail Failure handling type * * \return Static string describing \p on_fail */ const char * pcmk__on_fail_text(enum pcmk__on_fail on_fail) { switch (on_fail) { case pcmk__on_fail_ignore: return "ignore"; case pcmk__on_fail_demote: return "demote"; case pcmk__on_fail_block: return "block"; case pcmk__on_fail_restart: return "recover"; case pcmk__on_fail_ban: return "migrate"; case pcmk__on_fail_stop: return "stop"; case pcmk__on_fail_fence_node: return "fence"; case pcmk__on_fail_standby_node: return "standby"; case pcmk__on_fail_restart_container: return "restart-container"; case pcmk__on_fail_reset_remote: return "reset-remote"; } return ""; } /*! * \brief Generate an operation key (RESOURCE_ACTION_INTERVAL) * * \param[in] rsc_id ID of resource being operated on * \param[in] op_type Operation name * \param[in] interval_ms Operation interval * * \return Newly allocated memory containing operation key as string * * \note This function asserts on errors, so it will never return NULL. * The caller is responsible for freeing the result with free(). */ char * pcmk__op_key(const char *rsc_id, const char *op_type, guint interval_ms) { CRM_ASSERT(rsc_id != NULL); CRM_ASSERT(op_type != NULL); return crm_strdup_printf(PCMK__OP_FMT, rsc_id, op_type, interval_ms); } static inline gboolean convert_interval(const char *s, guint *interval_ms) { unsigned long l; errno = 0; l = strtoul(s, NULL, 10); if (errno != 0) { return FALSE; } *interval_ms = (guint) l; return TRUE; } /*! * \internal * \brief Check for underbar-separated substring match * * \param[in] key Overall string being checked * \param[in] position Match before underbar at this \p key index * \param[in] matches Substrings to match (may contain underbars) * * \return \p key index of underbar before any matching substring, * or 0 if none */ static size_t match_before(const char *key, size_t position, const char **matches) { for (int i = 0; matches[i] != NULL; ++i) { const size_t match_len = strlen(matches[i]); // Must have at least X_MATCH before position if (position > (match_len + 1)) { const size_t possible = position - match_len - 1; if ((key[possible] == '_') && (strncmp(key + possible + 1, matches[i], match_len) == 0)) { return possible; } } } return 0; } gboolean parse_op_key(const char *key, char **rsc_id, char **op_type, guint *interval_ms) { guint local_interval_ms = 0; const size_t key_len = (key == NULL)? 0 : strlen(key); // Operation keys must be formatted as RSC_ACTION_INTERVAL size_t action_underbar = 0; // Index in key of underbar before ACTION size_t interval_underbar = 0; // Index in key of underbar before INTERVAL size_t possible = 0; /* Underbar was a poor choice of separator since both RSC and ACTION can * contain underbars. Here, list action names and name prefixes that can. */ const char *actions_with_underbars[] = { PCMK_ACTION_MIGRATE_FROM, PCMK_ACTION_MIGRATE_TO, NULL }; const char *action_prefixes_with_underbars[] = { "pre_" PCMK_ACTION_NOTIFY, "post_" PCMK_ACTION_NOTIFY, "confirmed-pre_" PCMK_ACTION_NOTIFY, "confirmed-post_" PCMK_ACTION_NOTIFY, NULL, }; // Initialize output variables in case of early return if (rsc_id) { *rsc_id = NULL; } if (op_type) { *op_type = NULL; } if (interval_ms) { *interval_ms = 0; } // RSC_ACTION_INTERVAL implies a minimum of 5 characters if (key_len < 5) { return FALSE; } // Find, parse, and validate interval interval_underbar = key_len - 2; while ((interval_underbar > 2) && (key[interval_underbar] != '_')) { --interval_underbar; } if ((interval_underbar == 2) || !convert_interval(key + interval_underbar + 1, &local_interval_ms)) { return FALSE; } // Find the base (OCF) action name, disregarding prefixes action_underbar = match_before(key, interval_underbar, actions_with_underbars); if (action_underbar == 0) { action_underbar = interval_underbar - 2; while ((action_underbar > 0) && (key[action_underbar] != '_')) { --action_underbar; } if (action_underbar == 0) { return FALSE; } } possible = match_before(key, action_underbar, action_prefixes_with_underbars); if (possible != 0) { action_underbar = possible; } // Set output variables if (rsc_id != NULL) { *rsc_id = strndup(key, action_underbar); pcmk__mem_assert(*rsc_id); } if (op_type != NULL) { *op_type = strndup(key + action_underbar + 1, interval_underbar - action_underbar - 1); pcmk__mem_assert(*op_type); } if (interval_ms != NULL) { *interval_ms = local_interval_ms; } return TRUE; } char * pcmk__notify_key(const char *rsc_id, const char *notify_type, const char *op_type) { CRM_CHECK(rsc_id != NULL, return NULL); CRM_CHECK(op_type != NULL, return NULL); CRM_CHECK(notify_type != NULL, return NULL); return crm_strdup_printf("%s_%s_notify_%s_0", rsc_id, notify_type, op_type); } /*! * \brief Parse a transition magic string into its constituent parts * * \param[in] magic Magic string to parse (must be non-NULL) * \param[out] uuid If non-NULL, where to store copy of parsed UUID * \param[out] transition_id If non-NULL, where to store parsed transition ID * \param[out] action_id If non-NULL, where to store parsed action ID * \param[out] op_status If non-NULL, where to store parsed result status * \param[out] op_rc If non-NULL, where to store parsed actual rc * \param[out] target_rc If non-NULL, where to stored parsed target rc * * \return TRUE if key was valid, FALSE otherwise * \note If uuid is supplied and this returns TRUE, the caller is responsible * for freeing the memory for *uuid using free(). */ gboolean decode_transition_magic(const char *magic, char **uuid, int *transition_id, int *action_id, int *op_status, int *op_rc, int *target_rc) { int res = 0; char *key = NULL; gboolean result = TRUE; int local_op_status = -1; int local_op_rc = -1; CRM_CHECK(magic != NULL, return FALSE); #ifdef HAVE_SSCANF_M res = sscanf(magic, "%d:%d;%ms", &local_op_status, &local_op_rc, &key); #else // magic must have >=4 other characters key = pcmk__assert_alloc(1, strlen(magic) - 3); res = sscanf(magic, "%d:%d;%s", &local_op_status, &local_op_rc, key); #endif if (res == EOF) { crm_err("Could not decode transition information '%s': %s", magic, pcmk_rc_str(errno)); result = FALSE; } else if (res < 3) { crm_warn("Transition information '%s' incomplete (%d of 3 expected items)", magic, res); result = FALSE; } else { if (op_status) { *op_status = local_op_status; } if (op_rc) { *op_rc = local_op_rc; } result = decode_transition_key(key, uuid, transition_id, action_id, target_rc); } free(key); return result; } char * pcmk__transition_key(int transition_id, int action_id, int target_rc, const char *node) { CRM_CHECK(node != NULL, return NULL); return crm_strdup_printf("%d:%d:%d:%-*s", action_id, transition_id, target_rc, 36, node); } /*! * \brief Parse a transition key into its constituent parts * * \param[in] key Transition key to parse (must be non-NULL) * \param[out] uuid If non-NULL, where to store copy of parsed UUID * \param[out] transition_id If non-NULL, where to store parsed transition ID * \param[out] action_id If non-NULL, where to store parsed action ID * \param[out] target_rc If non-NULL, where to stored parsed target rc * * \return TRUE if key was valid, FALSE otherwise * \note If uuid is supplied and this returns TRUE, the caller is responsible * for freeing the memory for *uuid using free(). */ gboolean decode_transition_key(const char *key, char **uuid, int *transition_id, int *action_id, int *target_rc) { int local_transition_id = -1; int local_action_id = -1; int local_target_rc = -1; char local_uuid[37] = { '\0' }; // Initialize any supplied output arguments if (uuid) { *uuid = NULL; } if (transition_id) { *transition_id = -1; } if (action_id) { *action_id = -1; } if (target_rc) { *target_rc = -1; } CRM_CHECK(key != NULL, return FALSE); if (sscanf(key, "%d:%d:%d:%36s", &local_action_id, &local_transition_id, &local_target_rc, local_uuid) != 4) { crm_err("Invalid transition key '%s'", key); return FALSE; } if (strlen(local_uuid) != 36) { crm_warn("Invalid UUID '%s' in transition key '%s'", local_uuid, key); } if (uuid) { *uuid = pcmk__str_copy(local_uuid); } if (transition_id) { *transition_id = local_transition_id; } if (action_id) { *action_id = local_action_id; } if (target_rc) { *target_rc = local_target_rc; } return TRUE; } int rsc_op_expected_rc(const lrmd_event_data_t *op) { int rc = 0; if (op && op->user_data) { decode_transition_key(op->user_data, NULL, NULL, NULL, &rc); } return rc; } gboolean did_rsc_op_fail(lrmd_event_data_t * op, int target_rc) { switch (op->op_status) { case PCMK_EXEC_CANCELLED: case PCMK_EXEC_PENDING: return FALSE; case PCMK_EXEC_NOT_SUPPORTED: case PCMK_EXEC_TIMEOUT: case PCMK_EXEC_ERROR: case PCMK_EXEC_NOT_CONNECTED: case PCMK_EXEC_NO_FENCE_DEVICE: case PCMK_EXEC_NO_SECRETS: case PCMK_EXEC_INVALID: return TRUE; default: if (target_rc != op->rc) { return TRUE; } } return FALSE; } /*! * \brief Create a CIB XML element for an operation * * \param[in,out] parent If not NULL, make new XML node a child of this * \param[in] prefix Generate an ID using this prefix * \param[in] task Operation task to set * \param[in] interval_spec Operation interval to set * \param[in] timeout If not NULL, operation timeout to set * * \return New XML object on success, NULL otherwise */ xmlNode * crm_create_op_xml(xmlNode *parent, const char *prefix, const char *task, const char *interval_spec, const char *timeout) { xmlNode *xml_op; CRM_CHECK(prefix && task && interval_spec, return NULL); xml_op = pcmk__xe_create(parent, PCMK_XE_OP); pcmk__xe_set_id(xml_op, "%s-%s-%s", prefix, task, interval_spec); crm_xml_add(xml_op, PCMK_META_INTERVAL, interval_spec); crm_xml_add(xml_op, PCMK_XA_NAME, task); if (timeout) { crm_xml_add(xml_op, PCMK_META_TIMEOUT, timeout); } return xml_op; } /*! * \brief Check whether an operation requires resource agent meta-data * * \param[in] rsc_class Resource agent class (or NULL to skip class check) * \param[in] op Operation action (or NULL to skip op check) * * \return true if operation needs meta-data, false otherwise * \note At least one of rsc_class and op must be specified. */ bool crm_op_needs_metadata(const char *rsc_class, const char *op) { /* Agent metadata is used to determine whether an agent reload is possible, * so if this op is not relevant to that feature, we don't need metadata. */ CRM_CHECK((rsc_class != NULL) || (op != NULL), return false); if ((rsc_class != NULL) && !pcmk_is_set(pcmk_get_ra_caps(rsc_class), pcmk_ra_cap_params)) { // Metadata is needed only for resource classes that use parameters return false; } if (op == NULL) { return true; } // Metadata is needed only for these actions return pcmk__str_any_of(op, PCMK_ACTION_START, PCMK_ACTION_MONITOR, PCMK_ACTION_PROMOTE, PCMK_ACTION_DEMOTE, PCMK_ACTION_RELOAD, PCMK_ACTION_RELOAD_AGENT, PCMK_ACTION_MIGRATE_TO, PCMK_ACTION_MIGRATE_FROM, PCMK_ACTION_NOTIFY, NULL); } /*! * \internal * \brief Check whether an action name is for a fencing action * * \param[in] action Action name to check * * \return \c true if \p action is \c PCMK_ACTION_OFF, \c PCMK_ACTION_REBOOT, * or \c PCMK__ACTION_POWEROFF, otherwise \c false */ bool pcmk__is_fencing_action(const char *action) { return pcmk__str_any_of(action, PCMK_ACTION_OFF, PCMK_ACTION_REBOOT, PCMK__ACTION_POWEROFF, NULL); } diff --git a/lib/common/agents.c b/lib/common/agents.c index 16e064fb24..9a47619959 100644 --- a/lib/common/agents.c +++ b/lib/common/agents.c @@ -1,200 +1,196 @@ /* * 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 -#ifndef _GNU_SOURCE -# define _GNU_SOURCE -#endif - #include #include #include #include #include /*! * \brief Get capabilities of a resource agent standard * * \param[in] standard Standard name * * \return Bitmask of enum pcmk_ra_caps values */ uint32_t pcmk_get_ra_caps(const char *standard) { /* @COMPAT This should probably be case-sensitive, but isn't, * for backward compatibility. */ if (standard == NULL) { return pcmk_ra_cap_none; } else if (!strcasecmp(standard, PCMK_RESOURCE_CLASS_OCF)) { return pcmk_ra_cap_provider | pcmk_ra_cap_params | pcmk_ra_cap_unique | pcmk_ra_cap_promotable; } else if (!strcasecmp(standard, PCMK_RESOURCE_CLASS_STONITH)) { /* @COMPAT Stonith resources can't really be unique clones, but we've * allowed it in the past and have it in some scheduler regression tests * (which were likely never used as real configurations). * * @TODO Remove pcmk_ra_cap_unique at the next major schema version * bump, with a transform to remove PCMK_META_GLOBALLY_UNIQUE from the * config. */ return pcmk_ra_cap_params | pcmk_ra_cap_unique | pcmk_ra_cap_stdin | pcmk_ra_cap_fence_params; } else if (!strcasecmp(standard, PCMK_RESOURCE_CLASS_SYSTEMD) || !strcasecmp(standard, PCMK_RESOURCE_CLASS_SERVICE) || !strcasecmp(standard, PCMK_RESOURCE_CLASS_LSB) || !strcasecmp(standard, PCMK_RESOURCE_CLASS_UPSTART)) { /* Since service can map to LSB, systemd, or upstart, these should * have identical capabilities */ return pcmk_ra_cap_status; } else if (!strcasecmp(standard, PCMK_RESOURCE_CLASS_NAGIOS)) { return pcmk_ra_cap_params; } return pcmk_ra_cap_none; } int pcmk__effective_rc(int rc) { int remapped_rc = rc; switch (rc) { case PCMK_OCF_DEGRADED: remapped_rc = PCMK_OCF_OK; break; case PCMK_OCF_DEGRADED_PROMOTED: remapped_rc = PCMK_OCF_RUNNING_PROMOTED; break; default: break; } return remapped_rc; } char * crm_generate_ra_key(const char *standard, const char *provider, const char *type) { bool std_empty = pcmk__str_empty(standard); bool prov_empty = pcmk__str_empty(provider); bool ty_empty = pcmk__str_empty(type); if (std_empty || ty_empty) { return NULL; } return crm_strdup_printf("%s%s%s:%s", standard, (prov_empty ? "" : ":"), (prov_empty ? "" : provider), type); } /*! * \brief Parse a "standard[:provider]:type" agent specification * * \param[in] spec Agent specification * \param[out] standard Newly allocated memory containing agent standard (or NULL) * \param[out] provider Newly allocated memory containing agent provider (or NULL) * \param[put] type Newly allocated memory containing agent type (or NULL) * * \return pcmk_ok if the string could be parsed, -EINVAL otherwise * * \note It is acceptable for the type to contain a ':' if the standard supports * that. For example, systemd supports the form "systemd:UNIT@A:B". * \note It is the caller's responsibility to free the returned values. */ int crm_parse_agent_spec(const char *spec, char **standard, char **provider, char **type) { char *colon; CRM_CHECK(spec && standard && provider && type, return -EINVAL); *standard = NULL; *provider = NULL; *type = NULL; colon = strchr(spec, ':'); if ((colon == NULL) || (colon == spec)) { return -EINVAL; } *standard = strndup(spec, colon - spec); spec = colon + 1; if (pcmk_is_set(pcmk_get_ra_caps(*standard), pcmk_ra_cap_provider)) { colon = strchr(spec, ':'); if ((colon == NULL) || (colon == spec)) { free(*standard); return -EINVAL; } *provider = strndup(spec, colon - spec); spec = colon + 1; } if (*spec == '\0') { free(*standard); free(*provider); return -EINVAL; } *type = strdup(spec); return pcmk_ok; } /*! * \brief Check whether a given stonith parameter is handled by Pacemaker * * Return true if a given string is the name of one of the special resource * instance attributes interpreted directly by Pacemaker for stonith-class * resources. * * \param[in] param Parameter name to check * * \return true if \p param is a special fencing parameter */ bool pcmk_stonith_param(const char *param) { if (param == NULL) { return false; } if (pcmk__str_any_of(param, PCMK_STONITH_PROVIDES, PCMK_STONITH_STONITH_TIMEOUT, NULL)) { return true; } if (!pcmk__starts_with(param, "pcmk_")) { // Short-circuit common case return false; } if (pcmk__str_any_of(param, PCMK_STONITH_ACTION_LIMIT, PCMK_STONITH_DELAY_BASE, PCMK_STONITH_DELAY_MAX, PCMK_STONITH_HOST_ARGUMENT, PCMK_STONITH_HOST_CHECK, PCMK_STONITH_HOST_LIST, PCMK_STONITH_HOST_MAP, NULL)) { return true; } param = strchr(param + 5, '_'); // Skip past "pcmk_ACTION" return pcmk__str_any_of(param, "_action", "_timeout", "_retries", NULL); } diff --git a/lib/common/attrs.c b/lib/common/attrs.c index cc8f42df30..7d313d1e1b 100644 --- a/lib/common/attrs.c +++ b/lib/common/attrs.c @@ -1,183 +1,179 @@ /* * Copyright 2011-2024 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 _GNU_SOURCE -# define _GNU_SOURCE -#endif - #include #include #include #include #include #define OCF_RESKEY_PREFIX "OCF_RESKEY_" #define LRM_TARGET_ENV OCF_RESKEY_PREFIX CRM_META "_" PCMK__META_ON_NODE /*! * \internal * \brief Get the node name that should be used to set node attributes * * If given NULL, "auto", or "localhost" as an argument, check the environment * to detect the node name that should be used to set node attributes. (The * caller might not know the correct name, for example if the target is part of * a bundle with \c PCMK_META_CONTAINER_ATTRIBUTE_TARGET set to * \c PCMK_VALUE_HOST.) * * \param[in] name NULL, "auto" or "localhost" to check environment variables, * or anything else to return NULL * * \return Node name that should be used for node attributes based on the * environment if known, otherwise NULL */ const char * pcmk__node_attr_target(const char *name) { if (name == NULL || pcmk__strcase_any_of(name, "auto", "localhost", NULL)) { char buf[128] = OCF_RESKEY_PREFIX; size_t offset = sizeof(OCF_RESKEY_PREFIX) - 1; char *target_var = crm_meta_name(PCMK_META_CONTAINER_ATTRIBUTE_TARGET); char *phys_var = crm_meta_name(PCMK__META_PHYSICAL_HOST); const char *target = NULL; const char *host_physical = NULL; snprintf(buf + offset, sizeof(buf) - offset, "%s", target_var); target = getenv(buf); snprintf(buf + offset, sizeof(buf) - offset, "%s", phys_var); host_physical = getenv(buf); // It is important to use the name by which the scheduler knows us if (host_physical && pcmk__str_eq(target, PCMK_VALUE_HOST, pcmk__str_casei)) { name = host_physical; } else { const char *host_pcmk = getenv(LRM_TARGET_ENV); if (host_pcmk) { name = host_pcmk; } } free(target_var); free(phys_var); // TODO? Call pcmk__cluster_local_node_name() if name == NULL // (currently would require linkage against libcrmcluster) return name; } else { return NULL; } } /*! * \brief Return the name of the node attribute used as a promotion score * * \param[in] rsc_id Resource ID that promotion score is for (or NULL to * check the OCF_RESOURCE_INSTANCE environment variable) * * \return Newly allocated string with the node attribute name (or NULL on * error, including no ID or environment variable specified) * \note It is the caller's responsibility to free() the result. */ char * pcmk_promotion_score_name(const char *rsc_id) { if (pcmk__str_empty(rsc_id)) { rsc_id = getenv("OCF_RESOURCE_INSTANCE"); if (pcmk__str_empty(rsc_id)) { return NULL; } } return crm_strdup_printf("master-%s", rsc_id); } /*! * \internal * \brief Get the value of a node attribute * * \param[in] node Node to get attribute for * \param[in] name Name of node attribute to get * \param[in] target If this is \c PCMK_VALUE_HOST and \p node is a guest * (bundle) node, get the value from the guest's host, * otherwise get the value from \p node itself * \param[in] node_type If getting the value from \p node's host, this * indicates whether to check the current or assigned host * * \return Value of \p name attribute for \p node */ const char * pcmk__node_attr(const pcmk_node_t *node, const char *name, const char *target, enum pcmk__rsc_node node_type) { const char *value = NULL; // Attribute value to return const char *node_type_s = NULL; // Readable equivalent of node_type const pcmk_node_t *host = NULL; const pcmk_resource_t *container = NULL; if ((node == NULL) || (name == NULL)) { return NULL; } /* Check the node's own attributes unless this is a guest (bundle) node with * the container host as the attribute target. */ if (!pcmk__is_guest_or_bundle_node(node) || !pcmk__str_eq(target, PCMK_VALUE_HOST, pcmk__str_casei)) { value = g_hash_table_lookup(node->priv->attrs, name); crm_trace("%s='%s' on %s", name, pcmk__s(value, ""), pcmk__node_name(node)); return value; } /* This resource needs attributes set for the container's host instead of * for the container itself (useful when the container uses the host's * storage). */ container = node->priv->remote->priv->launcher; switch (node_type) { case pcmk__rsc_node_assigned: host = container->priv->assigned_node; if (host == NULL) { crm_trace("Skipping %s lookup for %s because " "its container %s is unassigned", name, pcmk__node_name(node), container->id); return NULL; } node_type_s = "assigned"; break; case pcmk__rsc_node_current: if (container->priv->active_nodes != NULL) { host = container->priv->active_nodes->data; } if (host == NULL) { crm_trace("Skipping %s lookup for %s because " "its container %s is inactive", name, pcmk__node_name(node), container->id); return NULL; } node_type_s = "current"; break; default: // Add support for other enum pcmk__rsc_node values if needed CRM_ASSERT(false); break; } value = g_hash_table_lookup(host->priv->attrs, name); crm_trace("%s='%s' for %s on %s container host %s", name, pcmk__s(value, ""), pcmk__node_name(node), node_type_s, pcmk__node_name(host)); return value; } diff --git a/lib/common/io.c b/lib/common/io.c index 4c034bf439..0e556378a1 100644 --- a/lib/common/io.c +++ b/lib/common/io.c @@ -1,643 +1,639 @@ /* * 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 -#ifndef _GNU_SOURCE -# define _GNU_SOURCE -#endif - #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include /*! * \internal * \brief Create a directory, including any parent directories needed * * \param[in] path_c Pathname of the directory to create * \param[in] mode Permissions to be used (with current umask) when creating * * \return Standard Pacemaker return code */ int pcmk__build_path(const char *path_c, mode_t mode) { int offset = 1, len = 0; int rc = pcmk_rc_ok; char *path = strdup(path_c); // cppcheck seems not to understand the abort logic in CRM_CHECK // cppcheck-suppress memleak CRM_CHECK(path != NULL, return -ENOMEM); for (len = strlen(path); offset < len; offset++) { if (path[offset] == '/') { path[offset] = 0; if ((mkdir(path, mode) < 0) && (errno != EEXIST)) { rc = errno; goto done; } path[offset] = '/'; } } if ((mkdir(path, mode) < 0) && (errno != EEXIST)) { rc = errno; } done: free(path); return rc; } /*! * \internal * \brief Return canonicalized form of a path name * * \param[in] path Pathname to canonicalize * \param[out] resolved_path Where to store canonicalized pathname * * \return Standard Pacemaker return code * \note The caller is responsible for freeing \p resolved_path on success. * \note This function exists because not all C library versions of * realpath(path, resolved_path) support a NULL resolved_path. */ int pcmk__real_path(const char *path, char **resolved_path) { CRM_CHECK((path != NULL) && (resolved_path != NULL), return EINVAL); #if _POSIX_VERSION >= 200809L /* Recent C libraries can dynamically allocate memory as needed */ *resolved_path = realpath(path, NULL); return (*resolved_path == NULL)? errno : pcmk_rc_ok; #elif defined(PATH_MAX) /* Older implementations require pre-allocated memory */ /* (this is less desirable because PATH_MAX may be huge or not defined) */ *resolved_path = malloc(PATH_MAX); if ((*resolved_path == NULL) || (realpath(path, *resolved_path) == NULL)) { return errno; } return pcmk_rc_ok; #else *resolved_path = NULL; return ENOTSUP; #endif } /*! * \internal * \brief Create a file name using a sequence number * * \param[in] directory Directory that contains the file series * \param[in] series Start of file name * \param[in] sequence Sequence number * \param[in] bzip Whether to use ".bz2" instead of ".raw" as extension * * \return Newly allocated file path (asserts on error, so always non-NULL) * \note The caller is responsible for freeing the return value. */ char * pcmk__series_filename(const char *directory, const char *series, int sequence, bool bzip) { CRM_ASSERT((directory != NULL) && (series != NULL)); return crm_strdup_printf("%s/%s-%d.%s", directory, series, sequence, (bzip? "bz2" : "raw")); } /*! * \internal * \brief Read sequence number stored in a file series' .last file * * \param[in] directory Directory that contains the file series * \param[in] series Start of file name * \param[out] seq Where to store the sequence number * * \return Standard Pacemaker return code */ int pcmk__read_series_sequence(const char *directory, const char *series, unsigned int *seq) { int rc; FILE *fp = NULL; char *series_file = NULL; if ((directory == NULL) || (series == NULL) || (seq == NULL)) { return EINVAL; } series_file = crm_strdup_printf("%s/%s.last", directory, series); fp = fopen(series_file, "r"); if (fp == NULL) { rc = errno; crm_debug("Could not open series file %s: %s", series_file, strerror(rc)); free(series_file); return rc; } errno = 0; if (fscanf(fp, "%u", seq) != 1) { rc = (errno == 0)? ENODATA : errno; crm_debug("Could not read sequence number from series file %s: %s", series_file, pcmk_rc_str(rc)); fclose(fp); return rc; } fclose(fp); crm_trace("Found last sequence number %u in series file %s", *seq, series_file); free(series_file); return pcmk_rc_ok; } /*! * \internal * \brief Write sequence number to a file series' .last file * * \param[in] directory Directory that contains the file series * \param[in] series Start of file name * \param[in] sequence Sequence number to write * \param[in] max Maximum sequence value, after which it is reset to 0 * * \note This function logs some errors but does not return any to the caller */ void pcmk__write_series_sequence(const char *directory, const char *series, unsigned int sequence, int max) { int rc = 0; FILE *file_strm = NULL; char *series_file = NULL; CRM_CHECK(directory != NULL, return); CRM_CHECK(series != NULL, return); if (max == 0) { return; } if (max > 0 && sequence >= max) { sequence = 0; } series_file = crm_strdup_printf("%s/%s.last", directory, series); file_strm = fopen(series_file, "w"); if (file_strm != NULL) { rc = fprintf(file_strm, "%u", sequence); if (rc < 0) { crm_perror(LOG_ERR, "Cannot write to series file %s", series_file); } } else { crm_err("Cannot open series file %s for writing", series_file); } if (file_strm != NULL) { fflush(file_strm); fclose(file_strm); } crm_trace("Wrote %d to %s", sequence, series_file); free(series_file); } /*! * \internal * \brief Change the owner and group of a file series' .last file * * \param[in] directory Directory that contains series * \param[in] series Series to change * \param[in] uid User ID of desired file owner * \param[in] gid Group ID of desired file group * * \return Standard Pacemaker return code * \note The caller must have the appropriate privileges. */ int pcmk__chown_series_sequence(const char *directory, const char *series, uid_t uid, gid_t gid) { char *series_file = NULL; int rc = pcmk_rc_ok; if ((directory == NULL) || (series == NULL)) { return EINVAL; } series_file = crm_strdup_printf("%s/%s.last", directory, series); if (chown(series_file, uid, gid) < 0) { rc = errno; } free(series_file); return rc; } static bool pcmk__daemon_user_can_write(const char *target_name, struct stat *target_stat) { struct passwd *sys_user = NULL; errno = 0; sys_user = getpwnam(CRM_DAEMON_USER); if (sys_user == NULL) { crm_notice("Could not find user %s: %s", CRM_DAEMON_USER, pcmk_rc_str(errno)); return FALSE; } if (target_stat->st_uid != sys_user->pw_uid) { crm_notice("%s is not owned by user %s " QB_XS " uid %d != %d", target_name, CRM_DAEMON_USER, sys_user->pw_uid, target_stat->st_uid); return FALSE; } if ((target_stat->st_mode & (S_IRUSR | S_IWUSR)) == 0) { crm_notice("%s is not readable and writable by user %s " QB_XS " st_mode=0%lo", target_name, CRM_DAEMON_USER, (unsigned long) target_stat->st_mode); return FALSE; } return TRUE; } static bool pcmk__daemon_group_can_write(const char *target_name, struct stat *target_stat) { struct group *sys_grp = NULL; errno = 0; sys_grp = getgrnam(CRM_DAEMON_GROUP); if (sys_grp == NULL) { crm_notice("Could not find group %s: %s", CRM_DAEMON_GROUP, pcmk_rc_str(errno)); return FALSE; } if (target_stat->st_gid != sys_grp->gr_gid) { crm_notice("%s is not owned by group %s " QB_XS " uid %d != %d", target_name, CRM_DAEMON_GROUP, sys_grp->gr_gid, target_stat->st_gid); return FALSE; } if ((target_stat->st_mode & (S_IRGRP | S_IWGRP)) == 0) { crm_notice("%s is not readable and writable by group %s " QB_XS " st_mode=0%lo", target_name, CRM_DAEMON_GROUP, (unsigned long) target_stat->st_mode); return FALSE; } return TRUE; } /*! * \internal * \brief Check whether a directory or file is writable by the cluster daemon * * Return true if either the cluster daemon user or cluster daemon group has * write permission on a specified file or directory. * * \param[in] dir Directory to check (this argument must be specified, and * the directory must exist) * \param[in] file File to check (only the directory will be checked if this * argument is not specified or the file does not exist) * * \return true if target is writable by cluster daemon, false otherwise */ bool pcmk__daemon_can_write(const char *dir, const char *file) { int s_res = 0; struct stat buf; char *full_file = NULL; const char *target = NULL; // Caller must supply directory CRM_ASSERT(dir != NULL); // If file is given, check whether it exists as a regular file if (file != NULL) { full_file = crm_strdup_printf("%s/%s", dir, file); target = full_file; s_res = stat(full_file, &buf); if (s_res < 0) { crm_notice("%s not found: %s", target, pcmk_rc_str(errno)); free(full_file); full_file = NULL; target = NULL; } else if (S_ISREG(buf.st_mode) == FALSE) { crm_err("%s must be a regular file " QB_XS " st_mode=0%lo", target, (unsigned long) buf.st_mode); free(full_file); return false; } } // If file is not given, ensure dir exists as directory if (target == NULL) { target = dir; s_res = stat(dir, &buf); if (s_res < 0) { crm_err("%s not found: %s", dir, pcmk_rc_str(errno)); return false; } else if (S_ISDIR(buf.st_mode) == FALSE) { crm_err("%s must be a directory " QB_XS " st_mode=0%lo", dir, (unsigned long) buf.st_mode); return false; } } if (!pcmk__daemon_user_can_write(target, &buf) && !pcmk__daemon_group_can_write(target, &buf)) { crm_err("%s must be owned and writable by either user %s or group %s " QB_XS " st_mode=0%lo", target, CRM_DAEMON_USER, CRM_DAEMON_GROUP, (unsigned long) buf.st_mode); free(full_file); return false; } free(full_file); return true; } /*! * \internal * \brief Flush and sync a directory to disk * * \param[in] name Directory to flush and sync * \note This function logs errors but does not return them to the caller */ void pcmk__sync_directory(const char *name) { int fd; DIR *directory; directory = opendir(name); if (directory == NULL) { crm_perror(LOG_ERR, "Could not open %s for syncing", name); return; } fd = dirfd(directory); if (fd < 0) { crm_perror(LOG_ERR, "Could not obtain file descriptor for %s", name); return; } if (fsync(fd) < 0) { crm_perror(LOG_ERR, "Could not sync %s", name); } if (closedir(directory) < 0) { crm_perror(LOG_ERR, "Could not close %s after fsync", name); } } /*! * \internal * \brief Read the contents of a file * * \param[in] filename Name of file to read * \param[out] contents Where to store file contents * * \return Standard Pacemaker return code * \note On success, the caller is responsible for freeing contents. */ int pcmk__file_contents(const char *filename, char **contents) { FILE *fp; int length, read_len; int rc = pcmk_rc_ok; if ((filename == NULL) || (contents == NULL)) { return EINVAL; } fp = fopen(filename, "r"); if ((fp == NULL) || (fseek(fp, 0L, SEEK_END) < 0)) { rc = errno; goto bail; } length = ftell(fp); if (length < 0) { rc = errno; goto bail; } if (length == 0) { *contents = NULL; } else { *contents = calloc(length + 1, sizeof(char)); if (*contents == NULL) { rc = errno; goto bail; } rewind(fp); read_len = fread(*contents, 1, length, fp); if (read_len != length) { free(*contents); *contents = NULL; rc = EIO; } else { /* Coverity thinks *contents isn't null-terminated. It doesn't * understand calloc(). */ (*contents)[length] = '\0'; } } bail: if (fp != NULL) { fclose(fp); } return rc; } /*! * \internal * \brief Write text to a file, flush and sync it to disk, then close the file * * \param[in] fd File descriptor opened for writing * \param[in] contents String to write to file * * \return Standard Pacemaker return code */ int pcmk__write_sync(int fd, const char *contents) { int rc = 0; FILE *fp = fdopen(fd, "w"); if (fp == NULL) { return errno; } if ((contents != NULL) && (fprintf(fp, "%s", contents) < 0)) { rc = EIO; } if (fflush(fp) != 0) { rc = errno; } if (fsync(fileno(fp)) < 0) { rc = errno; } fclose(fp); return rc; } /*! * \internal * \brief Set a file descriptor to non-blocking * * \param[in] fd File descriptor to use * * \return Standard Pacemaker return code */ int pcmk__set_nonblocking(int fd) { int flag = fcntl(fd, F_GETFL); if (flag < 0) { return errno; } if (fcntl(fd, F_SETFL, flag | O_NONBLOCK) < 0) { return errno; } return pcmk_rc_ok; } /*! * \internal * \brief Get directory name for temporary files * * Return the value of the TMPDIR environment variable if it is set to a * full path, otherwise return "/tmp". * * \return Name of directory to be used for temporary files */ const char * pcmk__get_tmpdir(void) { const char *dir = getenv("TMPDIR"); return (dir && (*dir == '/'))? dir : "/tmp"; } /*! * \internal * \brief Close open file descriptors * * Close all file descriptors (except optionally stdin, stdout, and stderr), * which is a best practice for a new child process forked for the purpose of * executing an external program. * * \param[in] bool If true, close stdin, stdout, and stderr as well */ void pcmk__close_fds_in_child(bool all) { DIR *dir; struct rlimit rlim; rlim_t max_fd; int min_fd = (all? 0 : (STDERR_FILENO + 1)); /* Find the current process's (soft) limit for open files. getrlimit() * should always work, but have a fallback just in case. */ if (getrlimit(RLIMIT_NOFILE, &rlim) == 0) { max_fd = rlim.rlim_cur - 1; } else { long conf_max = sysconf(_SC_OPEN_MAX); max_fd = (conf_max > 0)? conf_max : 1024; } /* /proc/self/fd (on Linux) or /dev/fd (on most OSes) contains symlinks to * all open files for the current process, named as the file descriptor. * Use this if available, because it's more efficient than a shotgun * approach to closing descriptors. */ #if HAVE_LINUX_PROCFS dir = opendir("/proc/self/fd"); if (dir == NULL) { dir = opendir("/dev/fd"); } #else dir = opendir("/dev/fd"); #endif // HAVE_LINUX_PROCFS if (dir != NULL) { struct dirent *entry; int dir_fd = dirfd(dir); while ((entry = readdir(dir)) != NULL) { int lpc = atoi(entry->d_name); /* How could one of these entries be higher than max_fd, you ask? * It isn't possible in normal operation, but when run under * valgrind, valgrind can open high-numbered file descriptors for * its own use that are higher than the process's soft limit. * These will show up in the fd directory but aren't closable. */ if ((lpc >= min_fd) && (lpc <= max_fd) && (lpc != dir_fd)) { close(lpc); } } closedir(dir); return; } /* If no fd directory is available, iterate over all possible descriptors. * This is less efficient due to the overhead of many system calls. */ for (int lpc = max_fd; lpc >= min_fd; lpc--) { close(lpc); } } /*! * \brief Duplicate a file path, inserting a prefix if not absolute * * \param[in] filename File path to duplicate * \param[in] dirname If filename is not absolute, prefix to add * * \return Newly allocated memory with full path (guaranteed non-NULL) */ char * pcmk__full_path(const char *filename, const char *dirname) { CRM_ASSERT(filename != NULL); if (filename[0] == '/') { return pcmk__str_copy(filename); } CRM_ASSERT(dirname != NULL); return crm_strdup_printf("%s/%s", dirname, filename); } diff --git a/lib/common/ipc_attrd.c b/lib/common/ipc_attrd.c index da54440d34..6e91819c19 100644 --- a/lib/common/ipc_attrd.c +++ b/lib/common/ipc_attrd.c @@ -1,485 +1,481 @@ /* * Copyright 2011-2024 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 _GNU_SOURCE -# define _GNU_SOURCE -#endif - #include #include #include #include #include #include #include #include "crmcommon_private.h" static void set_pairs_data(pcmk__attrd_api_reply_t *data, xmlNode *msg_data) { const char *name = NULL; pcmk__attrd_query_pair_t *pair; name = crm_element_value(msg_data, PCMK__XA_ATTR_NAME); for (xmlNode *node = pcmk__xe_first_child(msg_data, PCMK_XE_NODE, NULL, NULL); node != NULL; node = pcmk__xe_next_same(node)) { pair = pcmk__assert_alloc(1, sizeof(pcmk__attrd_query_pair_t)); pair->node = crm_element_value(node, PCMK__XA_ATTR_HOST); pair->name = name; pair->value = crm_element_value(node, PCMK__XA_ATTR_VALUE); data->data.pairs = g_list_prepend(data->data.pairs, pair); } } static bool reply_expected(pcmk_ipc_api_t *api, const xmlNode *request) { const char *command = crm_element_value(request, PCMK_XA_TASK); return pcmk__str_any_of(command, PCMK__ATTRD_CMD_CLEAR_FAILURE, PCMK__ATTRD_CMD_QUERY, PCMK__ATTRD_CMD_REFRESH, PCMK__ATTRD_CMD_UPDATE, PCMK__ATTRD_CMD_UPDATE_BOTH, PCMK__ATTRD_CMD_UPDATE_DELAY, NULL); } static bool dispatch(pcmk_ipc_api_t *api, xmlNode *reply) { const char *value = NULL; crm_exit_t status = CRM_EX_OK; pcmk__attrd_api_reply_t reply_data = { pcmk__attrd_reply_unknown }; if (pcmk__xe_is(reply, PCMK__XE_ACK)) { return false; } /* Do some basic validation of the reply */ value = crm_element_value(reply, PCMK__XA_T); if (pcmk__str_empty(value) || !pcmk__str_eq(value, PCMK__VALUE_ATTRD, pcmk__str_none)) { crm_info("Unrecognizable message from attribute manager: " "message type '%s' not '" PCMK__VALUE_ATTRD "'", pcmk__s(value, "")); status = CRM_EX_PROTOCOL; goto done; } value = crm_element_value(reply, PCMK__XA_SUBT); /* Only the query command gets a reply for now. NULL counts as query for * backward compatibility with attribute managers <2.1.3 that didn't set it. */ if (pcmk__str_eq(value, PCMK__ATTRD_CMD_QUERY, pcmk__str_null_matches)) { if (!xmlHasProp(reply, (pcmkXmlStr) PCMK__XA_ATTR_NAME)) { status = ENXIO; // Most likely, the attribute doesn't exist goto done; } reply_data.reply_type = pcmk__attrd_reply_query; set_pairs_data(&reply_data, reply); } else { crm_info("Unrecognizable message from attribute manager: " "message subtype '%s' unknown", pcmk__s(value, "")); status = CRM_EX_PROTOCOL; goto done; } done: pcmk__call_ipc_callback(api, pcmk_ipc_event_reply, status, &reply_data); /* Free any reply data that was allocated */ if (reply_data.data.pairs) { g_list_free_full(reply_data.data.pairs, free); } return false; } pcmk__ipc_methods_t * pcmk__attrd_api_methods(void) { pcmk__ipc_methods_t *cmds = calloc(1, sizeof(pcmk__ipc_methods_t)); if (cmds != NULL) { cmds->new_data = NULL; cmds->free_data = NULL; cmds->post_connect = NULL; cmds->reply_expected = reply_expected; cmds->dispatch = dispatch; } return cmds; } /*! * \internal * \brief Create a generic pacemaker-attrd operation * * \param[in] user_name If not NULL, ACL user to set for operation * * \return XML of pacemaker-attrd operation */ static xmlNode * create_attrd_op(const char *user_name) { xmlNode *attrd_op = pcmk__xe_create(NULL, __func__); crm_xml_add(attrd_op, PCMK__XA_T, PCMK__VALUE_ATTRD); crm_xml_add(attrd_op, PCMK__XA_SRC, pcmk__s(crm_system_name, "unknown")); crm_xml_add(attrd_op, PCMK__XA_ATTR_USER, user_name); return attrd_op; } static int connect_and_send_attrd_request(pcmk_ipc_api_t *api, const xmlNode *request) { int rc = pcmk_rc_ok; bool created_api = false; if (api == NULL) { rc = pcmk_new_ipc_api(&api, pcmk_ipc_attrd); if (rc != pcmk_rc_ok) { return rc; } created_api = true; } rc = pcmk__connect_ipc(api, pcmk_ipc_dispatch_sync, 5); if (rc == pcmk_rc_ok) { rc = pcmk__send_ipc_request(api, request); } if (created_api) { pcmk_free_ipc_api(api); } return rc; } int pcmk__attrd_api_clear_failures(pcmk_ipc_api_t *api, const char *node, const char *resource, const char *operation, const char *interval_spec, const char *user_name, uint32_t options) { int rc = pcmk_rc_ok; xmlNode *request = create_attrd_op(user_name); const char *interval_desc = NULL; const char *op_desc = NULL; const char *target = pcmk__node_attr_target(node); if (target != NULL) { node = target; } if (operation) { interval_desc = pcmk__s(interval_spec, "nonrecurring"); op_desc = operation; } else { interval_desc = "all"; op_desc = "operations"; } crm_debug("Asking %s to clear failure of %s %s for %s on %s", pcmk_ipc_name(api, true), interval_desc, op_desc, pcmk__s(resource, "all resources"), pcmk__s(node, "all nodes")); crm_xml_add(request, PCMK_XA_TASK, PCMK__ATTRD_CMD_CLEAR_FAILURE); pcmk__xe_add_node(request, node, 0); crm_xml_add(request, PCMK__XA_ATTR_RESOURCE, resource); crm_xml_add(request, PCMK__XA_ATTR_CLEAR_OPERATION, operation); crm_xml_add(request, PCMK__XA_ATTR_CLEAR_INTERVAL, interval_spec); crm_xml_add_int(request, PCMK__XA_ATTR_IS_REMOTE, pcmk_is_set(options, pcmk__node_attr_remote)); rc = connect_and_send_attrd_request(api, request); pcmk__xml_free(request); return rc; } int pcmk__attrd_api_delete(pcmk_ipc_api_t *api, const char *node, const char *name, uint32_t options) { const char *target = NULL; if (name == NULL) { return EINVAL; } target = pcmk__node_attr_target(node); if (target != NULL) { node = target; } /* Make sure the right update option is set. */ options &= ~pcmk__node_attr_delay; options |= pcmk__node_attr_value; return pcmk__attrd_api_update(api, node, name, NULL, NULL, NULL, NULL, options); } int pcmk__attrd_api_purge(pcmk_ipc_api_t *api, const char *node, bool reap) { int rc = pcmk_rc_ok; xmlNode *request = NULL; const char *target = pcmk__node_attr_target(node); if (target != NULL) { node = target; } crm_debug("Asking %s to purge transient attributes%s for %s", pcmk_ipc_name(api, true), (reap? " and node cache entries" : ""), pcmk__s(node, "local node")); request = create_attrd_op(NULL); crm_xml_add(request, PCMK_XA_TASK, PCMK__ATTRD_CMD_PEER_REMOVE); pcmk__xe_set_bool_attr(request, PCMK__XA_REAP, reap); pcmk__xe_add_node(request, node, 0); rc = connect_and_send_attrd_request(api, request); pcmk__xml_free(request); return rc; } int pcmk__attrd_api_query(pcmk_ipc_api_t *api, const char *node, const char *name, uint32_t options) { int rc = pcmk_rc_ok; xmlNode *request = NULL; const char *target = NULL; if (name == NULL) { return EINVAL; } if (pcmk_is_set(options, pcmk__node_attr_query_all)) { node = NULL; } else { target = pcmk__node_attr_target(node); if (target != NULL) { node = target; } } crm_debug("Querying %s for value of '%s'%s%s", pcmk_ipc_name(api, true), name, ((node == NULL)? "" : " on "), pcmk__s(node, "")); request = create_attrd_op(NULL); crm_xml_add(request, PCMK__XA_ATTR_NAME, name); crm_xml_add(request, PCMK_XA_TASK, PCMK__ATTRD_CMD_QUERY); pcmk__xe_add_node(request, node, 0); rc = connect_and_send_attrd_request(api, request); pcmk__xml_free(request); return rc; } int pcmk__attrd_api_refresh(pcmk_ipc_api_t *api, const char *node) { int rc = pcmk_rc_ok; xmlNode *request = NULL; const char *target = pcmk__node_attr_target(node); if (target != NULL) { node = target; } crm_debug("Asking %s to write all transient attributes for %s to CIB", pcmk_ipc_name(api, true), pcmk__s(node, "local node")); request = create_attrd_op(NULL); crm_xml_add(request, PCMK_XA_TASK, PCMK__ATTRD_CMD_REFRESH); pcmk__xe_add_node(request, node, 0); rc = connect_and_send_attrd_request(api, request); pcmk__xml_free(request); return rc; } static void add_op_attr(xmlNode *op, uint32_t options) { if (pcmk_all_flags_set(options, pcmk__node_attr_value | pcmk__node_attr_delay)) { crm_xml_add(op, PCMK_XA_TASK, PCMK__ATTRD_CMD_UPDATE_BOTH); } else if (pcmk_is_set(options, pcmk__node_attr_value)) { crm_xml_add(op, PCMK_XA_TASK, PCMK__ATTRD_CMD_UPDATE); } else if (pcmk_is_set(options, pcmk__node_attr_delay)) { crm_xml_add(op, PCMK_XA_TASK, PCMK__ATTRD_CMD_UPDATE_DELAY); } } static void populate_update_op(xmlNode *op, const char *node, const char *name, const char *value, const char *dampen, const char *set, uint32_t options) { if (pcmk_is_set(options, pcmk__node_attr_pattern)) { crm_xml_add(op, PCMK__XA_ATTR_REGEX, name); } else { crm_xml_add(op, PCMK__XA_ATTR_NAME, name); } if (pcmk_is_set(options, pcmk__node_attr_utilization)) { crm_xml_add(op, PCMK__XA_ATTR_SET_TYPE, PCMK_XE_UTILIZATION); } else { crm_xml_add(op, PCMK__XA_ATTR_SET_TYPE, PCMK_XE_INSTANCE_ATTRIBUTES); } add_op_attr(op, options); crm_xml_add(op, PCMK__XA_ATTR_VALUE, value); crm_xml_add(op, PCMK__XA_ATTR_DAMPENING, dampen); pcmk__xe_add_node(op, node, 0); crm_xml_add(op, PCMK__XA_ATTR_SET, set); crm_xml_add_int(op, PCMK__XA_ATTR_IS_REMOTE, pcmk_is_set(options, pcmk__node_attr_remote)); crm_xml_add_int(op, PCMK__XA_ATTR_IS_PRIVATE, pcmk_is_set(options, pcmk__node_attr_private)); if (pcmk_is_set(options, pcmk__node_attr_sync_local)) { crm_xml_add(op, PCMK__XA_ATTR_SYNC_POINT, PCMK__VALUE_LOCAL); } else if (pcmk_is_set(options, pcmk__node_attr_sync_cluster)) { crm_xml_add(op, PCMK__XA_ATTR_SYNC_POINT, PCMK__VALUE_CLUSTER); } } int pcmk__attrd_api_update(pcmk_ipc_api_t *api, const char *node, const char *name, const char *value, const char *dampen, const char *set, const char *user_name, uint32_t options) { int rc = pcmk_rc_ok; xmlNode *request = NULL; const char *target = NULL; if (name == NULL) { return EINVAL; } target = pcmk__node_attr_target(node); if (target != NULL) { node = target; } crm_debug("Asking %s to update '%s' to '%s' for %s", pcmk_ipc_name(api, true), name, pcmk__s(value, "(null)"), pcmk__s(node, "local node")); request = create_attrd_op(user_name); populate_update_op(request, node, name, value, dampen, set, options); rc = connect_and_send_attrd_request(api, request); pcmk__xml_free(request); return rc; } int pcmk__attrd_api_update_list(pcmk_ipc_api_t *api, GList *attrs, const char *dampen, const char *set, const char *user_name, uint32_t options) { int rc = pcmk_rc_ok; xmlNode *request = NULL; if (attrs == NULL) { return EINVAL; } /* There are two different ways of handling a list of attributes: * * (1) For messages originating from some command line tool, we have to send * them one at a time. In this loop, we just call pcmk__attrd_api_update * for each, letting it deal with creating the API object if it doesn't * already exist. * * The reason we can't use a single message in this case is that we can't * trust that the server supports it. Remote nodes could be involved * here, and there's no guarantee that a newer client running on a remote * node is talking to (or proxied through) a cluster node with a newer * attrd. We also can't just try sending a single message and then falling * back on multiple. There's no handshake with the attrd server to * determine its version. And then we would need to do that fallback in the * dispatch function for this to work for all connection types (mainloop in * particular), and at that point we won't know what the original message * was in order to break it apart and resend as individual messages. * * (2) For messages between daemons, we can be assured that the local attrd * will support the new message and that it can send to the other attrds * as one request or split up according to the minimum supported version. */ for (GList *iter = attrs; iter != NULL; iter = iter->next) { pcmk__attrd_query_pair_t *pair = (pcmk__attrd_query_pair_t *) iter->data; if (pcmk__is_daemon) { const char *target = NULL; xmlNode *child = NULL; /* First time through this loop - create the basic request. */ if (request == NULL) { request = create_attrd_op(user_name); add_op_attr(request, options); } /* Add a child node for this operation. We add the task to the top * level XML node so attrd_ipc_dispatch doesn't need changes. And * then we also add the task to each child node in populate_update_op * so attrd_client_update knows what form of update is taking place. */ child = pcmk__xe_create(request, PCMK_XE_OP); target = pcmk__node_attr_target(pair->node); if (target != NULL) { pair->node = target; } populate_update_op(child, pair->node, pair->name, pair->value, dampen, set, options); } else { rc = pcmk__attrd_api_update(api, pair->node, pair->name, pair->value, dampen, set, user_name, options); } } /* If we were doing multiple attributes at once, we still need to send the * request. Do that now, creating and destroying the API object if needed. */ if (pcmk__is_daemon) { rc = connect_and_send_attrd_request(api, request); pcmk__xml_free(request); } return rc; } diff --git a/lib/common/ipc_client.c b/lib/common/ipc_client.c index a0a78ff09a..3fd1e69b37 100644 --- a/lib/common/ipc_client.c +++ b/lib/common/ipc_client.c @@ -1,1694 +1,1689 @@ /* * 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 #if defined(HAVE_UCRED) || defined(HAVE_SOCKPEERCRED) -# ifdef HAVE_UCRED -# ifndef _GNU_SOURCE -# define _GNU_SOURCE -# endif -# endif -# include +#include #elif defined(HAVE_GETPEERUCRED) -# include +#include #endif #include #include #include #include #include /* indirectly: pcmk_err_generic */ #include #include #include #include "crmcommon_private.h" static int is_ipc_provider_expected(qb_ipcc_connection_t *qb_ipc, int sock, uid_t refuid, gid_t refgid, pid_t *gotpid, uid_t *gotuid, gid_t *gotgid); /*! * \brief Create a new object for using Pacemaker daemon IPC * * \param[out] api Where to store new IPC object * \param[in] server Which Pacemaker daemon the object is for * * \return Standard Pacemaker result code * * \note The caller is responsible for freeing *api using pcmk_free_ipc_api(). * \note This is intended to supersede crm_ipc_new() but currently only * supports the controller, pacemakerd, and schedulerd IPC API. */ int pcmk_new_ipc_api(pcmk_ipc_api_t **api, enum pcmk_ipc_server server) { if (api == NULL) { return EINVAL; } *api = calloc(1, sizeof(pcmk_ipc_api_t)); if (*api == NULL) { return errno; } (*api)->server = server; if (pcmk_ipc_name(*api, false) == NULL) { pcmk_free_ipc_api(*api); *api = NULL; return EOPNOTSUPP; } (*api)->ipc_size_max = 0; // Set server methods and max_size (if not default) switch (server) { case pcmk_ipc_attrd: (*api)->cmds = pcmk__attrd_api_methods(); break; case pcmk_ipc_based: (*api)->ipc_size_max = 512 * 1024; // 512KB break; case pcmk_ipc_controld: (*api)->cmds = pcmk__controld_api_methods(); break; case pcmk_ipc_execd: break; case pcmk_ipc_fenced: break; case pcmk_ipc_pacemakerd: (*api)->cmds = pcmk__pacemakerd_api_methods(); break; case pcmk_ipc_schedulerd: (*api)->cmds = pcmk__schedulerd_api_methods(); // @TODO max_size could vary by client, maybe take as argument? (*api)->ipc_size_max = 5 * 1024 * 1024; // 5MB break; } if ((*api)->cmds == NULL) { pcmk_free_ipc_api(*api); *api = NULL; return ENOMEM; } (*api)->ipc = crm_ipc_new(pcmk_ipc_name(*api, false), (*api)->ipc_size_max); if ((*api)->ipc == NULL) { pcmk_free_ipc_api(*api); *api = NULL; return ENOMEM; } // If daemon API has its own data to track, allocate it if ((*api)->cmds->new_data != NULL) { if ((*api)->cmds->new_data(*api) != pcmk_rc_ok) { pcmk_free_ipc_api(*api); *api = NULL; return ENOMEM; } } crm_trace("Created %s API IPC object", pcmk_ipc_name(*api, true)); return pcmk_rc_ok; } static void free_daemon_specific_data(pcmk_ipc_api_t *api) { if ((api != NULL) && (api->cmds != NULL)) { if ((api->cmds->free_data != NULL) && (api->api_data != NULL)) { api->cmds->free_data(api->api_data); api->api_data = NULL; } free(api->cmds); api->cmds = NULL; } } /*! * \internal * \brief Call an IPC API event callback, if one is registed * * \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 */ void pcmk__call_ipc_callback(pcmk_ipc_api_t *api, enum pcmk_ipc_event event_type, crm_exit_t status, void *event_data) { if ((api != NULL) && (api->cb != NULL)) { api->cb(api, event_type, status, event_data, api->user_data); } } /*! * \internal * \brief Clean up after an IPC disconnect * * \param[in,out] user_data IPC API connection that disconnected * * \note This function can be used as a main loop IPC destroy callback. */ static void ipc_post_disconnect(gpointer user_data) { pcmk_ipc_api_t *api = user_data; crm_info("Disconnected from %s", pcmk_ipc_name(api, true)); // Perform any daemon-specific handling needed if ((api->cmds != NULL) && (api->cmds->post_disconnect != NULL)) { api->cmds->post_disconnect(api); } // Call client's registered event callback pcmk__call_ipc_callback(api, pcmk_ipc_event_disconnect, CRM_EX_DISCONNECT, NULL); /* If this is being called from a running main loop, mainloop_gio_destroy() * will free ipc and mainloop_io immediately after calling this function. * If this is called from a stopped main loop, these will leak, so the best * practice is to close the connection before stopping the main loop. */ api->ipc = NULL; api->mainloop_io = NULL; if (api->free_on_disconnect) { /* pcmk_free_ipc_api() has already been called, but did not free api * or api->cmds because this function needed them. Do that now. */ free_daemon_specific_data(api); crm_trace("Freeing IPC API object after disconnect"); free(api); } } /*! * \brief Free the contents of an IPC API object * * \param[in,out] api IPC API object to free */ void pcmk_free_ipc_api(pcmk_ipc_api_t *api) { bool free_on_disconnect = false; if (api == NULL) { return; } crm_debug("Releasing %s IPC API", pcmk_ipc_name(api, true)); if (api->ipc != NULL) { if (api->mainloop_io != NULL) { /* We need to keep the api pointer itself around, because it is the * user data for the IPC client destroy callback. That will be * triggered by the pcmk_disconnect_ipc() call below, but it might * happen later in the main loop (if still running). * * This flag tells the destroy callback to free the object. It can't * do that unconditionally, because the application might call this * function after a disconnect that happened by other means. */ free_on_disconnect = api->free_on_disconnect = true; } pcmk_disconnect_ipc(api); // Frees api if free_on_disconnect is true } if (!free_on_disconnect) { free_daemon_specific_data(api); crm_trace("Freeing IPC API object"); free(api); } } /*! * \brief Get the IPC name used with an IPC API connection * * \param[in] api IPC API connection * \param[in] for_log If true, return human-friendly name instead of IPC name * * \return IPC API's human-friendly or connection name, or if none is available, * "Pacemaker" if for_log is true and NULL if for_log is false */ const char * pcmk_ipc_name(const pcmk_ipc_api_t *api, bool for_log) { if (api == NULL) { return for_log? "Pacemaker" : NULL; } switch (api->server) { case pcmk_ipc_attrd: return for_log? "attribute manager" : PCMK__VALUE_ATTRD; case pcmk_ipc_based: return for_log? "CIB manager" : NULL /* PCMK__SERVER_BASED_RW */; case pcmk_ipc_controld: return for_log? "controller" : CRM_SYSTEM_CRMD; case pcmk_ipc_execd: return for_log? "executor" : NULL /* CRM_SYSTEM_LRMD */; case pcmk_ipc_fenced: return for_log? "fencer" : NULL /* "stonith-ng" */; case pcmk_ipc_pacemakerd: return for_log? "launcher" : CRM_SYSTEM_MCP; case pcmk_ipc_schedulerd: return for_log? "scheduler" : CRM_SYSTEM_PENGINE; default: return for_log? "Pacemaker" : NULL; } } /*! * \brief Check whether an IPC API connection is active * * \param[in,out] api IPC API connection * * \return true if IPC is connected, false otherwise */ bool pcmk_ipc_is_connected(pcmk_ipc_api_t *api) { return (api != NULL) && crm_ipc_connected(api->ipc); } /*! * \internal * \brief Call the daemon-specific API's dispatch function * * Perform daemon-specific handling of IPC reply dispatch. It is the daemon * method's responsibility to call the client's registered event callback, as * well as allocate and free any event data. * * \param[in,out] api IPC API connection * \param[in,out] message IPC reply XML to dispatch */ static bool call_api_dispatch(pcmk_ipc_api_t *api, xmlNode *message) { crm_log_xml_trace(message, "ipc-received"); if ((api->cmds != NULL) && (api->cmds->dispatch != NULL)) { return api->cmds->dispatch(api, message); } return false; } /*! * \internal * \brief Dispatch previously read IPC data * * \param[in] buffer Data read from IPC * \param[in,out] api IPC object * * \return Standard Pacemaker return code. In particular: * * pcmk_rc_ok: There are no more messages expected from the server. Quit * reading. * EINPROGRESS: There are more messages expected from the server. Keep reading. * * All other values indicate an error. */ static int dispatch_ipc_data(const char *buffer, pcmk_ipc_api_t *api) { bool more = false; xmlNode *msg; if (buffer == NULL) { crm_warn("Empty message received from %s IPC", pcmk_ipc_name(api, true)); return ENOMSG; } msg = pcmk__xml_parse(buffer); if (msg == NULL) { crm_warn("Malformed message received from %s IPC", pcmk_ipc_name(api, true)); return EPROTO; } more = call_api_dispatch(api, msg); pcmk__xml_free(msg); if (more) { return EINPROGRESS; } else { return pcmk_rc_ok; } } /*! * \internal * \brief Dispatch data read from IPC source * * \param[in] buffer Data read from IPC * \param[in] length Number of bytes of data in buffer (ignored) * \param[in,out] user_data IPC object * * \return Always 0 (meaning connection is still required) * * \note This function can be used as a main loop IPC dispatch callback. */ static int dispatch_ipc_source_data(const char *buffer, ssize_t length, gpointer user_data) { pcmk_ipc_api_t *api = user_data; CRM_CHECK(api != NULL, return 0); dispatch_ipc_data(buffer, api); return 0; } /*! * \brief Check whether an IPC connection has data available (without main loop) * * \param[in] api IPC API connection * \param[in] timeout_ms If less than 0, poll indefinitely; if 0, poll once * and return immediately; otherwise, poll for up to * this many milliseconds * * \return Standard Pacemaker return code * * \note Callers of pcmk_connect_ipc() using pcmk_ipc_dispatch_poll should call * this function to check whether IPC data is available. Return values of * interest include pcmk_rc_ok meaning data is available, and EAGAIN * meaning no data is available; all other values indicate errors. * \todo This does not allow the caller to poll multiple file descriptors at * once. If there is demand for that, we could add a wrapper for * pcmk__ipc_fd(api->ipc), so the caller can call poll() themselves. */ int pcmk_poll_ipc(const pcmk_ipc_api_t *api, int timeout_ms) { int rc; struct pollfd pollfd = { 0, }; if ((api == NULL) || (api->dispatch_type != pcmk_ipc_dispatch_poll)) { return EINVAL; } rc = pcmk__ipc_fd(api->ipc, &(pollfd.fd)); if (rc != pcmk_rc_ok) { crm_debug("Could not obtain file descriptor for %s IPC: %s", pcmk_ipc_name(api, true), pcmk_rc_str(rc)); return rc; } pollfd.events = POLLIN; rc = poll(&pollfd, 1, timeout_ms); if (rc < 0) { /* Some UNIX systems return negative and set EAGAIN for failure to * allocate memory; standardize the return code in that case */ return (errno == EAGAIN)? ENOMEM : errno; } else if (rc == 0) { return EAGAIN; } return pcmk_rc_ok; } /*! * \brief Dispatch available messages on an IPC connection (without main loop) * * \param[in,out] api IPC API connection * * \return Standard Pacemaker return code * * \note Callers of pcmk_connect_ipc() using pcmk_ipc_dispatch_poll should call * this function when IPC data is available. */ void pcmk_dispatch_ipc(pcmk_ipc_api_t *api) { if (api == NULL) { return; } while (crm_ipc_ready(api->ipc) > 0) { if (crm_ipc_read(api->ipc) > 0) { dispatch_ipc_data(crm_ipc_buffer(api->ipc), api); } } } // \return Standard Pacemaker return code static int connect_with_main_loop(pcmk_ipc_api_t *api) { int rc; struct ipc_client_callbacks callbacks = { .dispatch = dispatch_ipc_source_data, .destroy = ipc_post_disconnect, }; rc = pcmk__add_mainloop_ipc(api->ipc, G_PRIORITY_DEFAULT, api, &callbacks, &(api->mainloop_io)); if (rc != pcmk_rc_ok) { return rc; } crm_debug("Connected to %s IPC (attached to main loop)", pcmk_ipc_name(api, true)); /* After this point, api->mainloop_io owns api->ipc, so api->ipc * should not be explicitly freed. */ return pcmk_rc_ok; } // \return Standard Pacemaker return code static int connect_without_main_loop(pcmk_ipc_api_t *api) { int rc = pcmk__connect_generic_ipc(api->ipc); if (rc != pcmk_rc_ok) { crm_ipc_close(api->ipc); } else { crm_debug("Connected to %s IPC (without main loop)", pcmk_ipc_name(api, true)); } return rc; } /*! * \internal * \brief Connect to a Pacemaker daemon via IPC (retrying after soft errors) * * \param[in,out] api IPC API instance * \param[in] dispatch_type How IPC replies should be dispatched * \param[in] attempts How many times to try (in case of soft error) * * \return Standard Pacemaker return code */ int pcmk__connect_ipc(pcmk_ipc_api_t *api, enum pcmk_ipc_dispatch dispatch_type, int attempts) { int rc = pcmk_rc_ok; if ((api == NULL) || (attempts < 1)) { return EINVAL; } if (api->ipc == NULL) { api->ipc = crm_ipc_new(pcmk_ipc_name(api, false), api->ipc_size_max); if (api->ipc == NULL) { return ENOMEM; } } if (crm_ipc_connected(api->ipc)) { crm_trace("Already connected to %s", pcmk_ipc_name(api, true)); return pcmk_rc_ok; } api->dispatch_type = dispatch_type; crm_debug("Attempting connection to %s (up to %d time%s)", pcmk_ipc_name(api, true), attempts, pcmk__plural_s(attempts)); for (int remaining = attempts - 1; remaining >= 0; --remaining) { switch (dispatch_type) { case pcmk_ipc_dispatch_main: rc = connect_with_main_loop(api); break; case pcmk_ipc_dispatch_sync: case pcmk_ipc_dispatch_poll: rc = connect_without_main_loop(api); break; } if ((remaining == 0) || ((rc != EAGAIN) && (rc != EALREADY))) { break; // Result is final } // Retry after soft error (interrupted by signal, etc.) pcmk__sleep_ms((attempts - remaining) * 500); crm_debug("Re-attempting connection to %s (%d attempt%s remaining)", pcmk_ipc_name(api, true), remaining, pcmk__plural_s(remaining)); } if (rc != pcmk_rc_ok) { return rc; } if ((api->cmds != NULL) && (api->cmds->post_connect != NULL)) { rc = api->cmds->post_connect(api); if (rc != pcmk_rc_ok) { crm_ipc_close(api->ipc); } } return rc; } /*! * \brief Connect to a Pacemaker daemon via IPC * * \param[in,out] api IPC API instance * \param[in] dispatch_type How IPC replies should be dispatched * * \return Standard Pacemaker return code */ int pcmk_connect_ipc(pcmk_ipc_api_t *api, enum pcmk_ipc_dispatch dispatch_type) { int rc = pcmk__connect_ipc(api, dispatch_type, 2); if (rc != pcmk_rc_ok) { crm_err("Connection to %s failed: %s", pcmk_ipc_name(api, true), pcmk_rc_str(rc)); } return rc; } /*! * \brief Disconnect an IPC API instance * * \param[in,out] api IPC API connection * * \return Standard Pacemaker return code * * \note If the connection is attached to a main loop, this function should be * called before quitting the main loop, to ensure that all memory is * freed. */ void pcmk_disconnect_ipc(pcmk_ipc_api_t *api) { if ((api == NULL) || (api->ipc == NULL)) { return; } switch (api->dispatch_type) { case pcmk_ipc_dispatch_main: { mainloop_io_t *mainloop_io = api->mainloop_io; // Make sure no code with access to api can use these again api->mainloop_io = NULL; api->ipc = NULL; mainloop_del_ipc_client(mainloop_io); // After this point api might have already been freed } break; case pcmk_ipc_dispatch_poll: case pcmk_ipc_dispatch_sync: { crm_ipc_t *ipc = api->ipc; // Make sure no code with access to api can use ipc again api->ipc = NULL; // This should always be the case already, but to be safe api->free_on_disconnect = false; crm_ipc_close(ipc); crm_ipc_destroy(ipc); ipc_post_disconnect(api); } break; } } /*! * \brief Register a callback for IPC API events * * \param[in,out] api IPC API connection * \param[in] callback Callback to register * \param[in] userdata Caller data to pass to callback * * \note This function may be called multiple times to update the callback * and/or user data. The caller remains responsible for freeing * userdata in any case (after the IPC is disconnected, if the * user data is still registered with the IPC). */ void pcmk_register_ipc_callback(pcmk_ipc_api_t *api, pcmk_ipc_callback_t cb, void *user_data) { if (api == NULL) { return; } api->cb = cb; api->user_data = user_data; } /*! * \internal * \brief Send an XML request across an IPC API connection * * \param[in,out] api IPC API connection * \param[in] request XML request to send * * \return Standard Pacemaker return code * * \note Daemon-specific IPC API functions should call this function to send * requests, because it handles different dispatch types appropriately. */ int pcmk__send_ipc_request(pcmk_ipc_api_t *api, const xmlNode *request) { int rc; xmlNode *reply = NULL; enum crm_ipc_flags flags = crm_ipc_flags_none; if ((api == NULL) || (api->ipc == NULL) || (request == NULL)) { return EINVAL; } crm_log_xml_trace(request, "ipc-sent"); // Synchronous dispatch requires waiting for a reply if ((api->dispatch_type == pcmk_ipc_dispatch_sync) && (api->cmds != NULL) && (api->cmds->reply_expected != NULL) && (api->cmds->reply_expected(api, request))) { flags = crm_ipc_client_response; } // The 0 here means a default timeout of 5 seconds rc = crm_ipc_send(api->ipc, request, flags, 0, &reply); if (rc < 0) { return pcmk_legacy2rc(rc); } else if (rc == 0) { return ENODATA; } // With synchronous dispatch, we dispatch any reply now if (reply != NULL) { bool more = call_api_dispatch(api, reply); pcmk__xml_free(reply); while (more) { rc = crm_ipc_read(api->ipc); if (rc == -EAGAIN) { continue; } else if (rc == -ENOMSG || rc == pcmk_ok) { return pcmk_rc_ok; } else if (rc < 0) { return -rc; } rc = dispatch_ipc_data(crm_ipc_buffer(api->ipc), api); if (rc == pcmk_rc_ok) { more = false; } else if (rc == EINPROGRESS) { more = true; } else { continue; } } } return pcmk_rc_ok; } /*! * \internal * \brief Create the XML for an IPC request to purge a node from the peer cache * * \param[in] api IPC API connection * \param[in] node_name If not NULL, name of node to purge * \param[in] nodeid If not 0, node ID of node to purge * * \return Newly allocated IPC request XML * * \note The controller, fencer, and pacemakerd use the same request syntax, but * the attribute manager uses a different one. The CIB manager doesn't * have any syntax for it. The executor and scheduler don't connect to the * cluster layer and thus don't have or need any syntax for it. * * \todo Modify the attribute manager to accept the common syntax (as well * as its current one, for compatibility with older clients). Modify * the CIB manager to accept and honor the common syntax. Modify the * executor and scheduler to accept the syntax (immediately returning * success), just for consistency. Modify this function to use the * common syntax with all daemons if their version supports it. */ static xmlNode * create_purge_node_request(const pcmk_ipc_api_t *api, const char *node_name, uint32_t nodeid) { xmlNode *request = NULL; const char *client = crm_system_name? crm_system_name : "client"; switch (api->server) { case pcmk_ipc_attrd: request = pcmk__xe_create(NULL, __func__); crm_xml_add(request, PCMK__XA_T, PCMK__VALUE_ATTRD); crm_xml_add(request, PCMK__XA_SRC, crm_system_name); crm_xml_add(request, PCMK_XA_TASK, PCMK__ATTRD_CMD_PEER_REMOVE); pcmk__xe_set_bool_attr(request, PCMK__XA_REAP, true); pcmk__xe_add_node(request, node_name, nodeid); break; case pcmk_ipc_controld: case pcmk_ipc_fenced: case pcmk_ipc_pacemakerd: request = create_request(CRM_OP_RM_NODE_CACHE, NULL, NULL, pcmk_ipc_name(api, false), client, NULL); if (nodeid > 0) { pcmk__xe_set_id(request, "%lu", (unsigned long) nodeid); } crm_xml_add(request, PCMK_XA_UNAME, node_name); break; case pcmk_ipc_based: case pcmk_ipc_execd: case pcmk_ipc_schedulerd: break; } return request; } /*! * \brief Ask a Pacemaker daemon to purge a node from its peer cache * * \param[in,out] api IPC API connection * \param[in] node_name If not NULL, name of node to purge * \param[in] nodeid If not 0, node ID of node to purge * * \return Standard Pacemaker return code * * \note At least one of node_name or nodeid must be specified. */ int pcmk_ipc_purge_node(pcmk_ipc_api_t *api, const char *node_name, uint32_t nodeid) { int rc = 0; xmlNode *request = NULL; if (api == NULL) { return EINVAL; } if ((node_name == NULL) && (nodeid == 0)) { return EINVAL; } request = create_purge_node_request(api, node_name, nodeid); if (request == NULL) { return EOPNOTSUPP; } rc = pcmk__send_ipc_request(api, request); pcmk__xml_free(request); crm_debug("%s peer cache purge of node %s[%lu]: rc=%d", pcmk_ipc_name(api, true), node_name, (unsigned long) nodeid, rc); return rc; } /* * Generic IPC API (to eventually be deprecated as public API and made internal) */ struct crm_ipc_s { struct pollfd pfd; unsigned int max_buf_size; // maximum bytes we can send or receive over IPC unsigned int buf_size; // size of allocated buffer int msg_size; int need_reply; char *buffer; char *server_name; // server IPC name being connected to qb_ipcc_connection_t *ipc; }; /*! * \brief Create a new (legacy) object for using Pacemaker daemon IPC * * \param[in] name IPC system name to connect to * \param[in] max_size Use a maximum IPC buffer size of at least this size * * \return Newly allocated IPC object on success, NULL otherwise * * \note The caller is responsible for freeing the result using * crm_ipc_destroy(). * \note This should be considered deprecated for use with daemons supported by * pcmk_new_ipc_api(). */ crm_ipc_t * crm_ipc_new(const char *name, size_t max_size) { crm_ipc_t *client = NULL; client = calloc(1, sizeof(crm_ipc_t)); if (client == NULL) { crm_err("Could not create IPC connection: %s", strerror(errno)); return NULL; } client->server_name = strdup(name); if (client->server_name == NULL) { crm_err("Could not create %s IPC connection: %s", name, strerror(errno)); free(client); return NULL; } client->buf_size = pcmk__ipc_buffer_size(max_size); client->buffer = malloc(client->buf_size); if (client->buffer == NULL) { crm_err("Could not create %s IPC connection: %s", name, strerror(errno)); free(client->server_name); free(client); return NULL; } /* Clients initiating connection pick the max buf size */ client->max_buf_size = client->buf_size; client->pfd.fd = -1; client->pfd.events = POLLIN; client->pfd.revents = 0; return client; } /*! * \internal * \brief Connect a generic (not daemon-specific) IPC object * * \param[in,out] ipc Generic IPC object to connect * * \return Standard Pacemaker return code */ int pcmk__connect_generic_ipc(crm_ipc_t *ipc) { uid_t cl_uid = 0; gid_t cl_gid = 0; pid_t found_pid = 0; uid_t found_uid = 0; gid_t found_gid = 0; int rc = pcmk_rc_ok; if (ipc == NULL) { return EINVAL; } ipc->need_reply = FALSE; ipc->ipc = qb_ipcc_connect(ipc->server_name, ipc->buf_size); if (ipc->ipc == NULL) { return errno; } rc = qb_ipcc_fd_get(ipc->ipc, &ipc->pfd.fd); if (rc < 0) { // -errno crm_ipc_close(ipc); return -rc; } rc = pcmk_daemon_user(&cl_uid, &cl_gid); rc = pcmk_legacy2rc(rc); if (rc != pcmk_rc_ok) { crm_ipc_close(ipc); return rc; } rc = is_ipc_provider_expected(ipc->ipc, ipc->pfd.fd, cl_uid, cl_gid, &found_pid, &found_uid, &found_gid); if (rc != pcmk_rc_ok) { if (rc == pcmk_rc_ipc_unauthorized) { crm_info("%s IPC provider authentication failed: process %lld has " "uid %lld (expected %lld) and gid %lld (expected %lld)", ipc->server_name, (long long) PCMK__SPECIAL_PID_AS_0(found_pid), (long long) found_uid, (long long) cl_uid, (long long) found_gid, (long long) cl_gid); } crm_ipc_close(ipc); return rc; } ipc->max_buf_size = qb_ipcc_get_buffer_size(ipc->ipc); if (ipc->max_buf_size > ipc->buf_size) { free(ipc->buffer); ipc->buffer = calloc(ipc->max_buf_size, sizeof(char)); if (ipc->buffer == NULL) { rc = errno; crm_ipc_close(ipc); return rc; } ipc->buf_size = ipc->max_buf_size; } return pcmk_rc_ok; } /*! * \brief Establish an IPC connection to a Pacemaker component * * \param[in,out] client Connection instance obtained from crm_ipc_new() * * \return true on success, false otherwise (in which case errno will be set; * specifically, in case of discovering the remote side is not * authentic, its value is set to ECONNABORTED). */ bool crm_ipc_connect(crm_ipc_t *client) { int rc = pcmk__connect_generic_ipc(client); if (rc == pcmk_rc_ok) { return true; } if ((client != NULL) && (client->ipc == NULL)) { errno = (rc > 0)? rc : ENOTCONN; crm_debug("Could not establish %s IPC connection: %s (%d)", client->server_name, pcmk_rc_str(errno), errno); } else if (rc == pcmk_rc_ipc_unauthorized) { crm_err("%s IPC provider authentication failed", (client == NULL)? "Pacemaker" : client->server_name); errno = ECONNABORTED; } else { crm_perror(LOG_ERR, "Could not verify authenticity of %s IPC provider", (client == NULL)? "Pacemaker" : client->server_name); errno = ENOTCONN; } return false; } void crm_ipc_close(crm_ipc_t * client) { if (client) { if (client->ipc) { qb_ipcc_connection_t *ipc = client->ipc; client->ipc = NULL; qb_ipcc_disconnect(ipc); } } } void crm_ipc_destroy(crm_ipc_t * client) { if (client) { if (client->ipc && qb_ipcc_is_connected(client->ipc)) { crm_notice("Destroying active %s IPC connection", client->server_name); /* The next line is basically unsafe * * If this connection was attached to mainloop and mainloop is active, * the 'disconnected' callback will end up back here and we'll end * up free'ing the memory twice - something that can still happen * even without this if we destroy a connection and it closes before * we call exit */ /* crm_ipc_close(client); */ } else { crm_trace("Destroying inactive %s IPC connection", client->server_name); } free(client->buffer); free(client->server_name); free(client); } } /*! * \internal * \brief Get the file descriptor for a generic IPC object * * \param[in,out] ipc Generic IPC object to get file descriptor for * \param[out] fd Where to store file descriptor * * \return Standard Pacemaker return code */ int pcmk__ipc_fd(crm_ipc_t *ipc, int *fd) { if ((ipc == NULL) || (fd == NULL)) { return EINVAL; } if ((ipc->ipc == NULL) || (ipc->pfd.fd < 0)) { return ENOTCONN; } *fd = ipc->pfd.fd; return pcmk_rc_ok; } int crm_ipc_get_fd(crm_ipc_t * client) { int fd = -1; if (pcmk__ipc_fd(client, &fd) != pcmk_rc_ok) { crm_err("Could not obtain file descriptor for %s IPC", ((client == NULL)? "unspecified" : client->server_name)); errno = EINVAL; return -EINVAL; } return fd; } bool crm_ipc_connected(crm_ipc_t * client) { bool rc = FALSE; if (client == NULL) { crm_trace("No client"); return FALSE; } else if (client->ipc == NULL) { crm_trace("No connection"); return FALSE; } else if (client->pfd.fd < 0) { crm_trace("Bad descriptor"); return FALSE; } rc = qb_ipcc_is_connected(client->ipc); if (rc == FALSE) { client->pfd.fd = -EINVAL; } return rc; } /*! * \brief Check whether an IPC connection is ready to be read * * \param[in,out] client Connection to check * * \return Positive value if ready to be read, 0 if not ready, -errno on error */ int crm_ipc_ready(crm_ipc_t *client) { int rc; CRM_ASSERT(client != NULL); if (!crm_ipc_connected(client)) { return -ENOTCONN; } client->pfd.revents = 0; rc = poll(&(client->pfd), 1, 0); return (rc < 0)? -errno : rc; } // \return Standard Pacemaker return code static int crm_ipc_decompress(crm_ipc_t * client) { pcmk__ipc_header_t *header = (pcmk__ipc_header_t *)(void*)client->buffer; if (header->size_compressed) { int rc = 0; unsigned int size_u = 1 + header->size_uncompressed; /* never let buf size fall below our max size required for ipc reads. */ unsigned int new_buf_size = QB_MAX((sizeof(pcmk__ipc_header_t) + size_u), client->max_buf_size); char *uncompressed = pcmk__assert_alloc(1, new_buf_size); crm_trace("Decompressing message data %u bytes into %u bytes", header->size_compressed, size_u); rc = BZ2_bzBuffToBuffDecompress(uncompressed + sizeof(pcmk__ipc_header_t), &size_u, client->buffer + sizeof(pcmk__ipc_header_t), header->size_compressed, 1, 0); rc = pcmk__bzlib2rc(rc); if (rc != pcmk_rc_ok) { crm_err("Decompression failed: %s " QB_XS " rc=%d", pcmk_rc_str(rc), rc); free(uncompressed); return rc; } /* * This assert no longer holds true. For an identical msg, some clients may * require compression, and others may not. If that same msg (event) is sent * to multiple clients, it could result in some clients receiving a compressed * msg even though compression was not explicitly required for them. * * CRM_ASSERT((header->size_uncompressed + sizeof(pcmk__ipc_header_t)) >= ipc_buffer_max); */ CRM_ASSERT(size_u == header->size_uncompressed); memcpy(uncompressed, client->buffer, sizeof(pcmk__ipc_header_t)); /* Preserve the header */ header = (pcmk__ipc_header_t *)(void*)uncompressed; free(client->buffer); client->buf_size = new_buf_size; client->buffer = uncompressed; } CRM_ASSERT(client->buffer[sizeof(pcmk__ipc_header_t) + header->size_uncompressed - 1] == 0); return pcmk_rc_ok; } long crm_ipc_read(crm_ipc_t * client) { pcmk__ipc_header_t *header = NULL; CRM_ASSERT(client != NULL); CRM_ASSERT(client->ipc != NULL); CRM_ASSERT(client->buffer != NULL); client->buffer[0] = 0; client->msg_size = qb_ipcc_event_recv(client->ipc, client->buffer, client->buf_size, 0); if (client->msg_size >= 0) { int rc = crm_ipc_decompress(client); if (rc != pcmk_rc_ok) { return pcmk_rc2legacy(rc); } header = (pcmk__ipc_header_t *)(void*)client->buffer; if (!pcmk__valid_ipc_header(header)) { return -EBADMSG; } crm_trace("Received %s IPC event %d size=%u rc=%d text='%.100s'", client->server_name, header->qb.id, header->qb.size, client->msg_size, client->buffer + sizeof(pcmk__ipc_header_t)); } else { crm_trace("No message received from %s IPC: %s", client->server_name, pcmk_strerror(client->msg_size)); if (client->msg_size == -EAGAIN) { return -EAGAIN; } } if (!crm_ipc_connected(client) || client->msg_size == -ENOTCONN) { crm_err("Connection to %s IPC failed", client->server_name); } if (header) { /* Data excluding the header */ return header->size_uncompressed; } return -ENOMSG; } const char * crm_ipc_buffer(crm_ipc_t * client) { CRM_ASSERT(client != NULL); return client->buffer + sizeof(pcmk__ipc_header_t); } uint32_t crm_ipc_buffer_flags(crm_ipc_t * client) { pcmk__ipc_header_t *header = NULL; CRM_ASSERT(client != NULL); if (client->buffer == NULL) { return 0; } header = (pcmk__ipc_header_t *)(void*)client->buffer; return header->flags; } const char * crm_ipc_name(crm_ipc_t * client) { CRM_ASSERT(client != NULL); return client->server_name; } // \return Standard Pacemaker return code static int internal_ipc_get_reply(crm_ipc_t *client, int request_id, int ms_timeout, ssize_t *bytes) { time_t timeout = time(NULL) + 1 + (ms_timeout / 1000); int rc = pcmk_rc_ok; /* get the reply */ crm_trace("Waiting on reply to %s IPC message %d", client->server_name, request_id); do { *bytes = qb_ipcc_recv(client->ipc, client->buffer, client->buf_size, 1000); if (*bytes > 0) { pcmk__ipc_header_t *hdr = NULL; rc = crm_ipc_decompress(client); if (rc != pcmk_rc_ok) { return rc; } hdr = (pcmk__ipc_header_t *)(void*)client->buffer; if (hdr->qb.id == request_id) { /* Got it */ break; } else if (hdr->qb.id < request_id) { xmlNode *bad = pcmk__xml_parse(crm_ipc_buffer(client)); crm_err("Discarding old reply %d (need %d)", hdr->qb.id, request_id); crm_log_xml_notice(bad, "OldIpcReply"); } else { xmlNode *bad = pcmk__xml_parse(crm_ipc_buffer(client)); crm_err("Discarding newer reply %d (need %d)", hdr->qb.id, request_id); crm_log_xml_notice(bad, "ImpossibleReply"); CRM_ASSERT(hdr->qb.id <= request_id); } } else if (!crm_ipc_connected(client)) { crm_err("%s IPC provider disconnected while waiting for message %d", client->server_name, request_id); break; } } while (time(NULL) < timeout); if (*bytes < 0) { rc = (int) -*bytes; // System errno } return rc; } /*! * \brief Send an IPC XML message * * \param[in,out] client Connection to IPC server * \param[in] message XML message to send * \param[in] flags Bitmask of crm_ipc_flags * \param[in] ms_timeout Give up if not sent within this much time * (5 seconds if 0, or no timeout if negative) * \param[out] reply Reply from server (or NULL if none) * * \return Negative errno on error, otherwise size of reply received in bytes * if reply was needed, otherwise number of bytes sent */ int crm_ipc_send(crm_ipc_t *client, const xmlNode *message, enum crm_ipc_flags flags, int32_t ms_timeout, xmlNode **reply) { int rc = 0; ssize_t qb_rc = 0; ssize_t bytes = 0; struct iovec *iov; static uint32_t id = 0; static int factor = 8; pcmk__ipc_header_t *header; if (client == NULL) { crm_notice("Can't send IPC request without connection (bug?): %.100s", message); return -ENOTCONN; } else if (!crm_ipc_connected(client)) { /* Don't even bother */ crm_notice("Can't send %s IPC requests: Connection closed", client->server_name); return -ENOTCONN; } if (ms_timeout == 0) { ms_timeout = 5000; } if (client->need_reply) { qb_rc = qb_ipcc_recv(client->ipc, client->buffer, client->buf_size, ms_timeout); if (qb_rc < 0) { crm_warn("Sending %s IPC disabled until pending reply received", client->server_name); return -EALREADY; } else { crm_notice("Sending %s IPC re-enabled after pending reply received", client->server_name); client->need_reply = FALSE; } } id++; CRM_LOG_ASSERT(id != 0); /* Crude wrap-around detection */ rc = pcmk__ipc_prepare_iov(id, message, client->max_buf_size, &iov, &bytes); if (rc != pcmk_rc_ok) { crm_warn("Couldn't prepare %s IPC request: %s " QB_XS " rc=%d", client->server_name, pcmk_rc_str(rc), rc); return pcmk_rc2legacy(rc); } header = iov[0].iov_base; pcmk__set_ipc_flags(header->flags, client->server_name, flags); if (pcmk_is_set(flags, crm_ipc_proxied)) { /* Don't look for a synchronous response */ pcmk__clear_ipc_flags(flags, "client", crm_ipc_client_response); } if(header->size_compressed) { if(factor < 10 && (client->max_buf_size / 10) < (bytes / factor)) { crm_notice("Compressed message exceeds %d0%% of configured IPC " "limit (%u bytes); consider setting PCMK_ipc_buffer to " "%u or higher", factor, client->max_buf_size, 2 * client->max_buf_size); factor++; } } crm_trace("Sending %s IPC request %d of %u bytes using %dms timeout", client->server_name, header->qb.id, header->qb.size, ms_timeout); if ((ms_timeout > 0) || !pcmk_is_set(flags, crm_ipc_client_response)) { time_t timeout = time(NULL) + 1 + (ms_timeout / 1000); do { /* @TODO Is this check really needed? Won't qb_ipcc_sendv() return * an error if it's not connected? */ if (!crm_ipc_connected(client)) { goto send_cleanup; } qb_rc = qb_ipcc_sendv(client->ipc, iov, 2); } while ((qb_rc == -EAGAIN) && (time(NULL) < timeout)); rc = (int) qb_rc; // Negative of system errno, or bytes sent if (qb_rc <= 0) { goto send_cleanup; } else if (!pcmk_is_set(flags, crm_ipc_client_response)) { crm_trace("Not waiting for reply to %s IPC request %d", client->server_name, header->qb.id); goto send_cleanup; } rc = internal_ipc_get_reply(client, header->qb.id, ms_timeout, &bytes); if (rc != pcmk_rc_ok) { /* We didn't get the reply in time, so disable future sends for now. * The only alternative would be to close the connection since we * don't know how to detect and discard out-of-sequence replies. * * @TODO Implement out-of-sequence detection */ client->need_reply = TRUE; } rc = (int) bytes; // Negative system errno, or size of reply received } else { // No timeout, and client response needed do { qb_rc = qb_ipcc_sendv_recv(client->ipc, iov, 2, client->buffer, client->buf_size, -1); } while ((qb_rc == -EAGAIN) && crm_ipc_connected(client)); rc = (int) qb_rc; // Negative system errno, or size of reply received } if (rc > 0) { pcmk__ipc_header_t *hdr = (pcmk__ipc_header_t *)(void*)client->buffer; crm_trace("Received %d-byte reply %d to %s IPC %d: %.100s", rc, hdr->qb.id, client->server_name, header->qb.id, crm_ipc_buffer(client)); if (reply) { *reply = pcmk__xml_parse(crm_ipc_buffer(client)); } } else { crm_trace("No reply to %s IPC %d: rc=%d", client->server_name, header->qb.id, rc); } send_cleanup: if (!crm_ipc_connected(client)) { crm_notice("Couldn't send %s IPC request %d: Connection closed " QB_XS " rc=%d", client->server_name, header->qb.id, rc); } else if (rc == -ETIMEDOUT) { crm_warn("%s IPC request %d failed: %s after %dms " QB_XS " rc=%d", client->server_name, header->qb.id, pcmk_strerror(rc), ms_timeout, rc); crm_write_blackbox(0, NULL); } else if (rc <= 0) { crm_warn("%s IPC request %d failed: %s " QB_XS " rc=%d", client->server_name, header->qb.id, ((rc == 0)? "No bytes sent" : pcmk_strerror(rc)), rc); } pcmk_free_ipc_event(iov); return rc; } /*! * \brief Ensure an IPC provider has expected user or group * * \param[in] qb_ipc libqb client connection if available * \param[in] sock Connected Unix socket for IPC * \param[in] refuid Expected user ID * \param[in] refgid Expected group ID * \param[out] gotpid If not NULL, where to store provider's actual process ID * (or 1 on platforms where ID is not available) * \param[out] gotuid If not NULL, where to store provider's actual user ID * \param[out] gotgid If not NULL, where to store provider's actual group ID * * \return Standard Pacemaker return code * \note An actual user ID of 0 (root) will always be considered authorized, * regardless of the expected values provided. The caller can use the * output arguments to be stricter than this function. */ static int is_ipc_provider_expected(qb_ipcc_connection_t *qb_ipc, int sock, uid_t refuid, gid_t refgid, pid_t *gotpid, uid_t *gotuid, gid_t *gotgid) { int rc = EOPNOTSUPP; pid_t found_pid = 0; uid_t found_uid = 0; gid_t found_gid = 0; #ifdef HAVE_QB_IPCC_AUTH_GET if (qb_ipc != NULL) { rc = qb_ipcc_auth_get(qb_ipc, &found_pid, &found_uid, &found_gid); rc = -rc; // libqb returns 0 or -errno if (rc == pcmk_rc_ok) { goto found; } } #endif #ifdef HAVE_UCRED { struct ucred ucred; socklen_t ucred_len = sizeof(ucred); if (getsockopt(sock, SOL_SOCKET, SO_PEERCRED, &ucred, &ucred_len) < 0) { rc = errno; } else if (ucred_len != sizeof(ucred)) { rc = EOPNOTSUPP; } else { found_pid = ucred.pid; found_uid = ucred.uid; found_gid = ucred.gid; goto found; } } #endif #ifdef HAVE_SOCKPEERCRED { struct sockpeercred sockpeercred; socklen_t sockpeercred_len = sizeof(sockpeercred); if (getsockopt(sock, SOL_SOCKET, SO_PEERCRED, &sockpeercred, &sockpeercred_len) < 0) { rc = errno; } else if (sockpeercred_len != sizeof(sockpeercred)) { rc = EOPNOTSUPP; } else { found_pid = sockpeercred.pid; found_uid = sockpeercred.uid; found_gid = sockpeercred.gid; goto found; } } #endif #ifdef HAVE_GETPEEREID // For example, FreeBSD if (getpeereid(sock, &found_uid, &found_gid) < 0) { rc = errno; } else { found_pid = PCMK__SPECIAL_PID; goto found; } #endif #ifdef HAVE_GETPEERUCRED { ucred_t *ucred = NULL; if (getpeerucred(sock, &ucred) < 0) { rc = errno; } else { found_pid = ucred_getpid(ucred); found_uid = ucred_geteuid(ucred); found_gid = ucred_getegid(ucred); ucred_free(ucred); goto found; } } #endif return rc; // If we get here, nothing succeeded found: if (gotpid != NULL) { *gotpid = found_pid; } if (gotuid != NULL) { *gotuid = found_uid; } if (gotgid != NULL) { *gotgid = found_gid; } if ((found_uid != 0) && (found_uid != refuid) && (found_gid != refgid)) { return pcmk_rc_ipc_unauthorized; } return pcmk_rc_ok; } int crm_ipc_is_authentic_process(int sock, uid_t refuid, gid_t refgid, pid_t *gotpid, uid_t *gotuid, gid_t *gotgid) { int ret = is_ipc_provider_expected(NULL, sock, refuid, refgid, gotpid, gotuid, gotgid); /* The old function had some very odd return codes*/ if (ret == 0) { return 1; } else if (ret == pcmk_rc_ipc_unauthorized) { return 0; } else { return pcmk_rc2legacy(ret); } } int pcmk__ipc_is_authentic_process_active(const char *name, uid_t refuid, gid_t refgid, pid_t *gotpid) { static char last_asked_name[PATH_MAX / 2] = ""; /* log spam prevention */ int fd; int rc = pcmk_rc_ipc_unresponsive; int auth_rc = 0; int32_t qb_rc; pid_t found_pid = 0; uid_t found_uid = 0; gid_t found_gid = 0; qb_ipcc_connection_t *c; #ifdef HAVE_QB_IPCC_CONNECT_ASYNC struct pollfd pollfd = { 0, }; int poll_rc; c = qb_ipcc_connect_async(name, 0, &(pollfd.fd)); #else c = qb_ipcc_connect(name, 0); #endif if (c == NULL) { crm_info("Could not connect to %s IPC: %s", name, strerror(errno)); rc = pcmk_rc_ipc_unresponsive; goto bail; } #ifdef HAVE_QB_IPCC_CONNECT_ASYNC pollfd.events = POLLIN; do { poll_rc = poll(&pollfd, 1, 2000); } while ((poll_rc == -1) && (errno == EINTR)); /* If poll() failed, given that disconnect function is not registered yet, * qb_ipcc_disconnect() won't clean up the socket. In any case, call * qb_ipcc_connect_continue() here so that it may fail and do the cleanup * for us. */ if (qb_ipcc_connect_continue(c) != 0) { crm_info("Could not connect to %s IPC: %s", name, (poll_rc == 0)?"timeout":strerror(errno)); rc = pcmk_rc_ipc_unresponsive; c = NULL; // qb_ipcc_connect_continue cleaned up for us goto bail; } #endif qb_rc = qb_ipcc_fd_get(c, &fd); if (qb_rc != 0) { rc = (int) -qb_rc; // System errno crm_err("Could not get fd from %s IPC: %s " QB_XS " rc=%d", name, pcmk_rc_str(rc), rc); goto bail; } auth_rc = is_ipc_provider_expected(c, fd, refuid, refgid, &found_pid, &found_uid, &found_gid); if (auth_rc == pcmk_rc_ipc_unauthorized) { crm_err("Daemon (IPC %s) effectively blocked with unauthorized" " process %lld (uid: %lld, gid: %lld)", name, (long long) PCMK__SPECIAL_PID_AS_0(found_pid), (long long) found_uid, (long long) found_gid); rc = pcmk_rc_ipc_unauthorized; goto bail; } if (auth_rc != pcmk_rc_ok) { rc = auth_rc; crm_err("Could not get peer credentials from %s IPC: %s " QB_XS " rc=%d", name, pcmk_rc_str(rc), rc); goto bail; } if (gotpid != NULL) { *gotpid = found_pid; } rc = pcmk_rc_ok; if ((found_uid != refuid || found_gid != refgid) && strncmp(last_asked_name, name, sizeof(last_asked_name))) { if ((found_uid == 0) && (refuid != 0)) { crm_warn("Daemon (IPC %s) runs as root, whereas the expected" " credentials are %lld:%lld, hazard of violating" " the least privilege principle", name, (long long) refuid, (long long) refgid); } else { crm_notice("Daemon (IPC %s) runs as %lld:%lld, whereas the" " expected credentials are %lld:%lld, which may" " mean a different set of privileges than expected", name, (long long) found_uid, (long long) found_gid, (long long) refuid, (long long) refgid); } memccpy(last_asked_name, name, '\0', sizeof(last_asked_name)); } bail: if (c != NULL) { qb_ipcc_disconnect(c); } return rc; } diff --git a/lib/common/mainloop.c b/lib/common/mainloop.c index b06a31366e..e53b0b1c1b 100644 --- a/lib/common/mainloop.c +++ b/lib/common/mainloop.c @@ -1,1467 +1,1463 @@ /* * 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 -#ifndef _GNU_SOURCE -# define _GNU_SOURCE -#endif - #include #include #include #include #include #include #include #include #include #include struct mainloop_child_s { pid_t pid; char *desc; unsigned timerid; gboolean timeout; void *privatedata; enum mainloop_child_flags flags; /* Called when a process dies */ void (*callback) (mainloop_child_t * p, pid_t pid, int core, int signo, int exitcode); }; struct trigger_s { GSource source; gboolean running; gboolean trigger; void *user_data; guint id; }; struct mainloop_timer_s { guint id; guint period_ms; bool repeat; char *name; GSourceFunc cb; void *userdata; }; static gboolean crm_trigger_prepare(GSource * source, gint * timeout) { crm_trigger_t *trig = (crm_trigger_t *) source; /* cluster-glue's FD and IPC related sources make use of * g_source_add_poll() but do not set a timeout in their prepare * functions * * This means mainloop's poll() will block until an event for one * of these sources occurs - any /other/ type of source, such as * this one or g_idle_*, that doesn't use g_source_add_poll() is * S-O-L and won't be processed until there is something fd-based * happens. * * Luckily the timeout we can set here affects all sources and * puts an upper limit on how long poll() can take. * * So unconditionally set a small-ish timeout, not too small that * we're in constant motion, which will act as an upper bound on * how long the signal handling might be delayed for. */ *timeout = 500; /* Timeout in ms */ return trig->trigger; } static gboolean crm_trigger_check(GSource * source) { crm_trigger_t *trig = (crm_trigger_t *) source; return trig->trigger; } /*! * \internal * \brief GSource dispatch function for crm_trigger_t * * \param[in] source crm_trigger_t being dispatched * \param[in] callback Callback passed at source creation * \param[in,out] userdata User data passed at source creation * * \return G_SOURCE_REMOVE to remove source, G_SOURCE_CONTINUE to keep it */ static gboolean crm_trigger_dispatch(GSource *source, GSourceFunc callback, gpointer userdata) { gboolean rc = G_SOURCE_CONTINUE; crm_trigger_t *trig = (crm_trigger_t *) source; if (trig->running) { /* Wait until the existing job is complete before starting the next one */ return G_SOURCE_CONTINUE; } trig->trigger = FALSE; if (callback) { int callback_rc = callback(trig->user_data); if (callback_rc < 0) { crm_trace("Trigger handler %p not yet complete", trig); trig->running = TRUE; } else if (callback_rc == 0) { rc = G_SOURCE_REMOVE; } } return rc; } static void crm_trigger_finalize(GSource * source) { crm_trace("Trigger %p destroyed", source); } static GSourceFuncs crm_trigger_funcs = { crm_trigger_prepare, crm_trigger_check, crm_trigger_dispatch, crm_trigger_finalize, }; static crm_trigger_t * mainloop_setup_trigger(GSource * source, int priority, int (*dispatch) (gpointer user_data), gpointer userdata) { crm_trigger_t *trigger = NULL; trigger = (crm_trigger_t *) source; trigger->id = 0; trigger->trigger = FALSE; trigger->user_data = userdata; if (dispatch) { g_source_set_callback(source, dispatch, trigger, NULL); } g_source_set_priority(source, priority); g_source_set_can_recurse(source, FALSE); trigger->id = g_source_attach(source, NULL); return trigger; } void mainloop_trigger_complete(crm_trigger_t * trig) { crm_trace("Trigger handler %p complete", trig); trig->running = FALSE; } /*! * \brief Create a trigger to be used as a mainloop source * * \param[in] priority Relative priority of source (lower number is higher priority) * \param[in] dispatch Trigger dispatch function (should return 0 to remove the * trigger from the mainloop, -1 if the trigger should be * kept but the job is still running and not complete, and * 1 if the trigger should be kept and the job is complete) * \param[in] userdata Pointer to pass to \p dispatch * * \return Newly allocated mainloop source for trigger */ crm_trigger_t * mainloop_add_trigger(int priority, int (*dispatch) (gpointer user_data), gpointer userdata) { GSource *source = NULL; CRM_ASSERT(sizeof(crm_trigger_t) > sizeof(GSource)); source = g_source_new(&crm_trigger_funcs, sizeof(crm_trigger_t)); return mainloop_setup_trigger(source, priority, dispatch, userdata); } void mainloop_set_trigger(crm_trigger_t * source) { if(source) { source->trigger = TRUE; } } gboolean mainloop_destroy_trigger(crm_trigger_t * source) { GSource *gs = NULL; if(source == NULL) { return TRUE; } gs = (GSource *)source; g_source_destroy(gs); /* Remove from mainloop, ref_count-- */ g_source_unref(gs); /* The caller no longer carries a reference to source * * At this point the source should be free'd, * unless we're currently processing said * source, in which case mainloop holds an * additional reference and it will be free'd * once our processing completes */ return TRUE; } // Define a custom glib source for signal handling // Data structure for custom glib source typedef struct signal_s { crm_trigger_t trigger; // trigger that invoked source (must be first) void (*handler) (int sig); // signal handler int signal; // signal that was received } crm_signal_t; // Table to associate signal handlers with signal numbers static crm_signal_t *crm_signals[NSIG]; /*! * \internal * \brief Dispatch an event from custom glib source for signals * * Given an signal event, clear the event trigger and call any registered * signal handler. * * \param[in] source glib source that triggered this dispatch * \param[in] callback (ignored) * \param[in] userdata (ignored) */ static gboolean crm_signal_dispatch(GSource *source, GSourceFunc callback, gpointer userdata) { crm_signal_t *sig = (crm_signal_t *) source; if(sig->signal != SIGCHLD) { crm_notice("Caught '%s' signal " QB_XS " %d (%s handler)", strsignal(sig->signal), sig->signal, (sig->handler? "invoking" : "no")); } sig->trigger.trigger = FALSE; if (sig->handler) { sig->handler(sig->signal); } return TRUE; } /*! * \internal * \brief Handle a signal by setting a trigger for signal source * * \param[in] sig Signal number that was received * * \note This is the true signal handler for the mainloop signal source, and * must be async-safe. */ static void mainloop_signal_handler(int sig) { if (sig > 0 && sig < NSIG && crm_signals[sig] != NULL) { mainloop_set_trigger((crm_trigger_t *) crm_signals[sig]); } } // Functions implementing our custom glib source for signal handling static GSourceFuncs crm_signal_funcs = { crm_trigger_prepare, crm_trigger_check, crm_signal_dispatch, crm_trigger_finalize, }; /*! * \internal * \brief Set a true signal handler * * signal()-like interface to sigaction() * * \param[in] sig Signal number to register handler for * \param[in] dispatch Signal handler * * \return The previous value of the signal handler, or SIG_ERR on error * \note The dispatch function must be async-safe. */ sighandler_t crm_signal_handler(int sig, sighandler_t dispatch) { sigset_t mask; struct sigaction sa; struct sigaction old; if (sigemptyset(&mask) < 0) { crm_err("Could not set handler for signal %d: %s", sig, pcmk_rc_str(errno)); return SIG_ERR; } memset(&sa, 0, sizeof(struct sigaction)); sa.sa_handler = dispatch; sa.sa_flags = SA_RESTART; sa.sa_mask = mask; if (sigaction(sig, &sa, &old) < 0) { crm_err("Could not set handler for signal %d: %s", sig, pcmk_rc_str(errno)); return SIG_ERR; } return old.sa_handler; } static void mainloop_destroy_signal_entry(int sig) { crm_signal_t *tmp = crm_signals[sig]; crm_signals[sig] = NULL; crm_trace("Destroying signal %d", sig); mainloop_destroy_trigger((crm_trigger_t *) tmp); } /*! * \internal * \brief Add a signal handler to a mainloop * * \param[in] sig Signal number to handle * \param[in] dispatch Signal handler function * * \note The true signal handler merely sets a mainloop trigger to call this * dispatch function via the mainloop. Therefore, the dispatch function * does not need to be async-safe. */ gboolean mainloop_add_signal(int sig, void (*dispatch) (int sig)) { GSource *source = NULL; int priority = G_PRIORITY_HIGH - 1; if (sig == SIGTERM) { /* TERM is higher priority than other signals, * signals are higher priority than other ipc. * Yes, minus: smaller is "higher" */ priority--; } if (sig >= NSIG || sig < 0) { crm_err("Signal %d is out of range", sig); return FALSE; } else if (crm_signals[sig] != NULL && crm_signals[sig]->handler == dispatch) { crm_trace("Signal handler for %d is already installed", sig); return TRUE; } else if (crm_signals[sig] != NULL) { crm_err("Different signal handler for %d is already installed", sig); return FALSE; } CRM_ASSERT(sizeof(crm_signal_t) > sizeof(GSource)); source = g_source_new(&crm_signal_funcs, sizeof(crm_signal_t)); crm_signals[sig] = (crm_signal_t *) mainloop_setup_trigger(source, priority, NULL, NULL); CRM_ASSERT(crm_signals[sig] != NULL); crm_signals[sig]->handler = dispatch; crm_signals[sig]->signal = sig; if (crm_signal_handler(sig, mainloop_signal_handler) == SIG_ERR) { mainloop_destroy_signal_entry(sig); return FALSE; } return TRUE; } gboolean mainloop_destroy_signal(int sig) { if (sig >= NSIG || sig < 0) { crm_err("Signal %d is out of range", sig); return FALSE; } else if (crm_signal_handler(sig, NULL) == SIG_ERR) { crm_perror(LOG_ERR, "Could not uninstall signal handler for signal %d", sig); return FALSE; } else if (crm_signals[sig] == NULL) { return TRUE; } mainloop_destroy_signal_entry(sig); return TRUE; } static qb_array_t *gio_map = NULL; void mainloop_cleanup(void) { if (gio_map) { qb_array_free(gio_map); } for (int sig = 0; sig < NSIG; ++sig) { mainloop_destroy_signal_entry(sig); } } /* * libqb... */ struct gio_to_qb_poll { int32_t is_used; guint source; int32_t events; void *data; qb_ipcs_dispatch_fn_t fn; enum qb_loop_priority p; }; static gboolean gio_read_socket(GIOChannel * gio, GIOCondition condition, gpointer data) { struct gio_to_qb_poll *adaptor = (struct gio_to_qb_poll *)data; gint fd = g_io_channel_unix_get_fd(gio); crm_trace("%p.%d %d", data, fd, condition); /* if this assert get's hit, then there is a race condition between * when we destroy a fd and when mainloop actually gives it up */ CRM_ASSERT(adaptor->is_used > 0); return (adaptor->fn(fd, condition, adaptor->data) == 0); } static void gio_poll_destroy(gpointer data) { struct gio_to_qb_poll *adaptor = (struct gio_to_qb_poll *)data; adaptor->is_used--; CRM_ASSERT(adaptor->is_used >= 0); if (adaptor->is_used == 0) { crm_trace("Marking adaptor %p unused", adaptor); adaptor->source = 0; } } /*! * \internal * \brief Convert libqb's poll priority into GLib's one * * \param[in] prio libqb's poll priority (#QB_LOOP_MED assumed as fallback) * * \return best matching GLib's priority */ static gint conv_prio_libqb2glib(enum qb_loop_priority prio) { switch (prio) { case QB_LOOP_LOW: return G_PRIORITY_LOW; case QB_LOOP_HIGH: return G_PRIORITY_HIGH; default: return G_PRIORITY_DEFAULT; // QB_LOOP_MED } } /*! * \internal * \brief Convert libqb's poll priority to rate limiting spec * * \param[in] prio libqb's poll priority (#QB_LOOP_MED assumed as fallback) * * \return best matching rate limiting spec * \note This is the inverse of libqb's qb_ipcs_request_rate_limit(). */ static enum qb_ipcs_rate_limit conv_libqb_prio2ratelimit(enum qb_loop_priority prio) { switch (prio) { case QB_LOOP_LOW: return QB_IPCS_RATE_SLOW; case QB_LOOP_HIGH: return QB_IPCS_RATE_FAST; default: return QB_IPCS_RATE_NORMAL; // QB_LOOP_MED } } static int32_t gio_poll_dispatch_update(enum qb_loop_priority p, int32_t fd, int32_t evts, void *data, qb_ipcs_dispatch_fn_t fn, int32_t add) { struct gio_to_qb_poll *adaptor; GIOChannel *channel; int32_t res = 0; res = qb_array_index(gio_map, fd, (void **)&adaptor); if (res < 0) { crm_err("Array lookup failed for fd=%d: %d", fd, res); return res; } crm_trace("Adding fd=%d to mainloop as adaptor %p", fd, adaptor); if (add && adaptor->source) { crm_err("Adaptor for descriptor %d is still in-use", fd); return -EEXIST; } if (!add && !adaptor->is_used) { crm_err("Adaptor for descriptor %d is not in-use", fd); return -ENOENT; } /* channel is created with ref_count = 1 */ channel = g_io_channel_unix_new(fd); if (!channel) { crm_err("No memory left to add fd=%d", fd); return -ENOMEM; } if (adaptor->source) { g_source_remove(adaptor->source); adaptor->source = 0; } /* Because unlike the poll() API, glib doesn't tell us about HUPs by default */ evts |= (G_IO_HUP | G_IO_NVAL | G_IO_ERR); adaptor->fn = fn; adaptor->events = evts; adaptor->data = data; adaptor->p = p; adaptor->is_used++; adaptor->source = g_io_add_watch_full(channel, conv_prio_libqb2glib(p), evts, gio_read_socket, adaptor, gio_poll_destroy); /* Now that mainloop now holds a reference to channel, * thanks to g_io_add_watch_full(), drop ours from g_io_channel_unix_new(). * * This means that channel will be free'd by: * g_main_context_dispatch() * -> g_source_destroy_internal() * -> g_source_callback_unref() * shortly after gio_poll_destroy() completes */ g_io_channel_unref(channel); crm_trace("Added to mainloop with gsource id=%d", adaptor->source); if (adaptor->source > 0) { return 0; } return -EINVAL; } static int32_t gio_poll_dispatch_add(enum qb_loop_priority p, int32_t fd, int32_t evts, void *data, qb_ipcs_dispatch_fn_t fn) { return gio_poll_dispatch_update(p, fd, evts, data, fn, QB_TRUE); } static int32_t gio_poll_dispatch_mod(enum qb_loop_priority p, int32_t fd, int32_t evts, void *data, qb_ipcs_dispatch_fn_t fn) { return gio_poll_dispatch_update(p, fd, evts, data, fn, QB_FALSE); } static int32_t gio_poll_dispatch_del(int32_t fd) { struct gio_to_qb_poll *adaptor; crm_trace("Looking for fd=%d", fd); if (qb_array_index(gio_map, fd, (void **)&adaptor) == 0) { if (adaptor->source) { g_source_remove(adaptor->source); adaptor->source = 0; } } return 0; } struct qb_ipcs_poll_handlers gio_poll_funcs = { .job_add = NULL, .dispatch_add = gio_poll_dispatch_add, .dispatch_mod = gio_poll_dispatch_mod, .dispatch_del = gio_poll_dispatch_del, }; static enum qb_ipc_type pick_ipc_type(enum qb_ipc_type requested) { const char *env = pcmk__env_option(PCMK__ENV_IPC_TYPE); if (env && strcmp("shared-mem", env) == 0) { return QB_IPC_SHM; } else if (env && strcmp("socket", env) == 0) { return QB_IPC_SOCKET; } else if (env && strcmp("posix", env) == 0) { return QB_IPC_POSIX_MQ; } else if (env && strcmp("sysv", env) == 0) { return QB_IPC_SYSV_MQ; } else if (requested == QB_IPC_NATIVE) { /* We prefer shared memory because the server never blocks on * send. If part of a message fits into the socket, libqb * needs to block until the remainder can be sent also. * Otherwise the client will wait forever for the remaining * bytes. */ return QB_IPC_SHM; } return requested; } qb_ipcs_service_t * mainloop_add_ipc_server(const char *name, enum qb_ipc_type type, struct qb_ipcs_service_handlers *callbacks) { return mainloop_add_ipc_server_with_prio(name, type, callbacks, QB_LOOP_MED); } qb_ipcs_service_t * mainloop_add_ipc_server_with_prio(const char *name, enum qb_ipc_type type, struct qb_ipcs_service_handlers *callbacks, enum qb_loop_priority prio) { int rc = 0; qb_ipcs_service_t *server = NULL; if (gio_map == NULL) { gio_map = qb_array_create_2(64, sizeof(struct gio_to_qb_poll), 1); } server = qb_ipcs_create(name, 0, pick_ipc_type(type), callbacks); if (server == NULL) { crm_err("Could not create %s IPC server: %s (%d)", name, pcmk_rc_str(errno), errno); return NULL; } if (prio != QB_LOOP_MED) { qb_ipcs_request_rate_limit(server, conv_libqb_prio2ratelimit(prio)); } /* All clients should use at least ipc_buffer_max as their buffer size */ qb_ipcs_enforce_buffer_size(server, crm_ipc_default_buffer_size()); qb_ipcs_poll_handlers_set(server, &gio_poll_funcs); rc = qb_ipcs_run(server); if (rc < 0) { crm_err("Could not start %s IPC server: %s (%d)", name, pcmk_strerror(rc), rc); return NULL; // qb_ipcs_run() destroys server on failure } return server; } void mainloop_del_ipc_server(qb_ipcs_service_t * server) { if (server) { qb_ipcs_destroy(server); } } struct mainloop_io_s { char *name; void *userdata; int fd; guint source; crm_ipc_t *ipc; GIOChannel *channel; int (*dispatch_fn_ipc) (const char *buffer, ssize_t length, gpointer userdata); int (*dispatch_fn_io) (gpointer userdata); void (*destroy_fn) (gpointer userdata); }; /*! * \internal * \brief I/O watch callback function (GIOFunc) * * \param[in] gio I/O channel being watched * \param[in] condition I/O condition satisfied * \param[in] data User data passed when source was created * * \return G_SOURCE_REMOVE to remove source, G_SOURCE_CONTINUE to keep it */ static gboolean mainloop_gio_callback(GIOChannel *gio, GIOCondition condition, gpointer data) { gboolean rc = G_SOURCE_CONTINUE; mainloop_io_t *client = data; CRM_ASSERT(client->fd == g_io_channel_unix_get_fd(gio)); if (condition & G_IO_IN) { if (client->ipc) { long read_rc = 0L; int max = 10; do { read_rc = crm_ipc_read(client->ipc); if (read_rc <= 0) { crm_trace("Could not read IPC message from %s: %s (%ld)", client->name, pcmk_strerror(read_rc), read_rc); } else if (client->dispatch_fn_ipc) { const char *buffer = crm_ipc_buffer(client->ipc); crm_trace("New %ld-byte IPC message from %s " "after I/O condition %d", read_rc, client->name, (int) condition); if (client->dispatch_fn_ipc(buffer, read_rc, client->userdata) < 0) { crm_trace("Connection to %s no longer required", client->name); rc = G_SOURCE_REMOVE; } } } while ((rc == G_SOURCE_CONTINUE) && (read_rc > 0) && --max > 0); } else { crm_trace("New I/O event for %s after I/O condition %d", client->name, (int) condition); if (client->dispatch_fn_io) { if (client->dispatch_fn_io(client->userdata) < 0) { crm_trace("Connection to %s no longer required", client->name); rc = G_SOURCE_REMOVE; } } } } if (client->ipc && !crm_ipc_connected(client->ipc)) { crm_err("Connection to %s closed " QB_XS "client=%p condition=%d", client->name, client, condition); rc = G_SOURCE_REMOVE; } else if (condition & (G_IO_HUP | G_IO_NVAL | G_IO_ERR)) { crm_trace("The connection %s[%p] has been closed (I/O condition=%d)", client->name, client, condition); rc = G_SOURCE_REMOVE; } else if ((condition & G_IO_IN) == 0) { /* #define GLIB_SYSDEF_POLLIN =1 #define GLIB_SYSDEF_POLLPRI =2 #define GLIB_SYSDEF_POLLOUT =4 #define GLIB_SYSDEF_POLLERR =8 #define GLIB_SYSDEF_POLLHUP =16 #define GLIB_SYSDEF_POLLNVAL =32 typedef enum { G_IO_IN GLIB_SYSDEF_POLLIN, G_IO_OUT GLIB_SYSDEF_POLLOUT, G_IO_PRI GLIB_SYSDEF_POLLPRI, G_IO_ERR GLIB_SYSDEF_POLLERR, G_IO_HUP GLIB_SYSDEF_POLLHUP, G_IO_NVAL GLIB_SYSDEF_POLLNVAL } GIOCondition; A bitwise combination representing a condition to watch for on an event source. G_IO_IN There is data to read. G_IO_OUT Data can be written (without blocking). G_IO_PRI There is urgent data to read. G_IO_ERR Error condition. G_IO_HUP Hung up (the connection has been broken, usually for pipes and sockets). G_IO_NVAL Invalid request. The file descriptor is not open. */ crm_err("Strange condition: %d", condition); } /* G_SOURCE_REMOVE results in mainloop_gio_destroy() being called * just before the source is removed from mainloop */ return rc; } static void mainloop_gio_destroy(gpointer c) { mainloop_io_t *client = c; char *c_name = strdup(client->name); /* client->source is valid but about to be destroyed (ref_count == 0) in gmain.c * client->channel will still have ref_count > 0... should be == 1 */ crm_trace("Destroying client %s[%p]", c_name, c); if (client->ipc) { crm_ipc_close(client->ipc); } if (client->destroy_fn) { void (*destroy_fn) (gpointer userdata) = client->destroy_fn; client->destroy_fn = NULL; destroy_fn(client->userdata); } if (client->ipc) { crm_ipc_t *ipc = client->ipc; client->ipc = NULL; crm_ipc_destroy(ipc); } crm_trace("Destroyed client %s[%p]", c_name, c); free(client->name); client->name = NULL; free(client); free(c_name); } /*! * \brief Connect to IPC and add it as a main loop source * * \param[in,out] ipc IPC connection to add * \param[in] priority Event source priority to use for connection * \param[in] userdata Data to register with callbacks * \param[in] callbacks Dispatch and destroy callbacks for connection * \param[out] source Newly allocated event source * * \return Standard Pacemaker return code * * \note On failure, the caller is still responsible for ipc. On success, the * caller should call mainloop_del_ipc_client() when source is no longer * needed, which will lead to the disconnection of the IPC later in the * main loop if it is connected. However the IPC disconnects, * mainloop_gio_destroy() will free ipc and source after calling the * destroy callback. */ int pcmk__add_mainloop_ipc(crm_ipc_t *ipc, int priority, void *userdata, const struct ipc_client_callbacks *callbacks, mainloop_io_t **source) { int rc = pcmk_rc_ok; int fd = -1; const char *ipc_name = NULL; CRM_CHECK((ipc != NULL) && (callbacks != NULL), return EINVAL); ipc_name = pcmk__s(crm_ipc_name(ipc), "Pacemaker"); rc = pcmk__connect_generic_ipc(ipc); if (rc != pcmk_rc_ok) { crm_debug("Connection to %s failed: %s", ipc_name, pcmk_rc_str(rc)); return rc; } rc = pcmk__ipc_fd(ipc, &fd); if (rc != pcmk_rc_ok) { crm_debug("Could not obtain file descriptor for %s IPC: %s", ipc_name, pcmk_rc_str(rc)); crm_ipc_close(ipc); return rc; } *source = mainloop_add_fd(ipc_name, priority, fd, userdata, NULL); if (*source == NULL) { rc = errno; crm_ipc_close(ipc); return rc; } (*source)->ipc = ipc; (*source)->destroy_fn = callbacks->destroy; (*source)->dispatch_fn_ipc = callbacks->dispatch; return pcmk_rc_ok; } /*! * \brief Get period for mainloop timer * * \param[in] timer Timer * * \return Period in ms */ guint pcmk__mainloop_timer_get_period(const mainloop_timer_t *timer) { if (timer) { return timer->period_ms; } return 0; } mainloop_io_t * mainloop_add_ipc_client(const char *name, int priority, size_t max_size, void *userdata, struct ipc_client_callbacks *callbacks) { crm_ipc_t *ipc = crm_ipc_new(name, max_size); mainloop_io_t *source = NULL; int rc = pcmk__add_mainloop_ipc(ipc, priority, userdata, callbacks, &source); if (rc != pcmk_rc_ok) { if (crm_log_level == LOG_STDOUT) { fprintf(stderr, "Connection to %s failed: %s", name, pcmk_rc_str(rc)); } crm_ipc_destroy(ipc); if (rc > 0) { errno = rc; } else { errno = ENOTCONN; } return NULL; } return source; } void mainloop_del_ipc_client(mainloop_io_t * client) { mainloop_del_fd(client); } crm_ipc_t * mainloop_get_ipc_client(mainloop_io_t * client) { if (client) { return client->ipc; } return NULL; } mainloop_io_t * mainloop_add_fd(const char *name, int priority, int fd, void *userdata, struct mainloop_fd_callbacks * callbacks) { mainloop_io_t *client = NULL; if (fd >= 0) { client = calloc(1, sizeof(mainloop_io_t)); if (client == NULL) { return NULL; } client->name = strdup(name); client->userdata = userdata; if (callbacks) { client->destroy_fn = callbacks->destroy; client->dispatch_fn_io = callbacks->dispatch; } client->fd = fd; client->channel = g_io_channel_unix_new(fd); client->source = g_io_add_watch_full(client->channel, priority, (G_IO_IN | G_IO_HUP | G_IO_NVAL | G_IO_ERR), mainloop_gio_callback, client, mainloop_gio_destroy); /* Now that mainloop now holds a reference to channel, * thanks to g_io_add_watch_full(), drop ours from g_io_channel_unix_new(). * * This means that channel will be free'd by: * g_main_context_dispatch() or g_source_remove() * -> g_source_destroy_internal() * -> g_source_callback_unref() * shortly after mainloop_gio_destroy() completes */ g_io_channel_unref(client->channel); crm_trace("Added connection %d for %s[%p].%d", client->source, client->name, client, fd); } else { errno = EINVAL; } return client; } void mainloop_del_fd(mainloop_io_t * client) { if (client != NULL) { crm_trace("Removing client %s[%p]", client->name, client); if (client->source) { /* Results in mainloop_gio_destroy() being called just * before the source is removed from mainloop */ g_source_remove(client->source); } } } static GList *child_list = NULL; pid_t mainloop_child_pid(mainloop_child_t * child) { return child->pid; } const char * mainloop_child_name(mainloop_child_t * child) { return child->desc; } int mainloop_child_timeout(mainloop_child_t * child) { return child->timeout; } void * mainloop_child_userdata(mainloop_child_t * child) { return child->privatedata; } void mainloop_clear_child_userdata(mainloop_child_t * child) { child->privatedata = NULL; } /* good function name */ static void child_free(mainloop_child_t *child) { if (child->timerid != 0) { crm_trace("Removing timer %d", child->timerid); g_source_remove(child->timerid); child->timerid = 0; } free(child->desc); free(child); } /* terrible function name */ static int child_kill_helper(mainloop_child_t *child) { int rc; if (child->flags & mainloop_leave_pid_group) { crm_debug("Kill pid %d only. leave group intact.", child->pid); rc = kill(child->pid, SIGKILL); } else { crm_debug("Kill pid %d's group", child->pid); rc = kill(-child->pid, SIGKILL); } if (rc < 0) { if (errno != ESRCH) { crm_perror(LOG_ERR, "kill(%d, KILL) failed", child->pid); } return -errno; } return 0; } static gboolean child_timeout_callback(gpointer p) { mainloop_child_t *child = p; int rc = 0; child->timerid = 0; if (child->timeout) { crm_warn("%s process (PID %d) will not die!", child->desc, (int)child->pid); return FALSE; } rc = child_kill_helper(child); if (rc == -ESRCH) { /* Nothing left to do. pid doesn't exist */ return FALSE; } child->timeout = TRUE; crm_debug("%s process (PID %d) timed out", child->desc, (int)child->pid); child->timerid = g_timeout_add(5000, child_timeout_callback, child); return FALSE; } static bool child_waitpid(mainloop_child_t *child, int flags) { int rc = 0; int core = 0; int signo = 0; int status = 0; int exitcode = 0; bool callback_needed = true; rc = waitpid(child->pid, &status, flags); if (rc == 0) { // WNOHANG in flags, and child status is not available crm_trace("Child process %d (%s) still active", child->pid, child->desc); callback_needed = false; } else if (rc != child->pid) { /* According to POSIX, possible conditions: * - child->pid was non-positive (process group or any child), * and rc is specific child * - errno ECHILD (pid does not exist or is not child) * - errno EINVAL (invalid flags) * - errno EINTR (caller interrupted by signal) * * @TODO Handle these cases more specifically. */ signo = SIGCHLD; exitcode = 1; crm_notice("Wait for child process %d (%s) interrupted: %s", child->pid, child->desc, pcmk_rc_str(errno)); } else if (WIFEXITED(status)) { exitcode = WEXITSTATUS(status); crm_trace("Child process %d (%s) exited with status %d", child->pid, child->desc, exitcode); } else if (WIFSIGNALED(status)) { signo = WTERMSIG(status); crm_trace("Child process %d (%s) exited with signal %d (%s)", child->pid, child->desc, signo, strsignal(signo)); #ifdef WCOREDUMP // AIX, SunOS, maybe others } else if (WCOREDUMP(status)) { core = 1; crm_err("Child process %d (%s) dumped core", child->pid, child->desc); #endif } else { // flags must contain WUNTRACED and/or WCONTINUED to reach this crm_trace("Child process %d (%s) stopped or continued", child->pid, child->desc); callback_needed = false; } if (callback_needed && child->callback) { child->callback(child, child->pid, core, signo, exitcode); } return callback_needed; } static void child_death_dispatch(int signal) { for (GList *iter = child_list; iter; ) { GList *saved = iter; mainloop_child_t *child = iter->data; iter = iter->next; if (child_waitpid(child, WNOHANG)) { crm_trace("Removing completed process %d from child list", child->pid); child_list = g_list_remove_link(child_list, saved); g_list_free(saved); child_free(child); } } } static gboolean child_signal_init(gpointer p) { crm_trace("Installed SIGCHLD handler"); /* Do NOT use g_child_watch_add() and friends, they rely on pthreads */ mainloop_add_signal(SIGCHLD, child_death_dispatch); /* In case they terminated before the signal handler was installed */ child_death_dispatch(SIGCHLD); return FALSE; } gboolean mainloop_child_kill(pid_t pid) { GList *iter; mainloop_child_t *child = NULL; mainloop_child_t *match = NULL; /* It is impossible to block SIGKILL, this allows us to * call waitpid without WNOHANG flag.*/ int waitflags = 0, rc = 0; for (iter = child_list; iter != NULL && match == NULL; iter = iter->next) { child = iter->data; if (pid == child->pid) { match = child; } } if (match == NULL) { return FALSE; } rc = child_kill_helper(match); if(rc == -ESRCH) { /* It's gone, but hasn't shown up in waitpid() yet. Wait until we get * SIGCHLD and let handler clean it up as normal (so we get the correct * return code/status). The blocking alternative would be to call * child_waitpid(match, 0). */ crm_trace("Waiting for signal that child process %d completed", match->pid); return TRUE; } else if(rc != 0) { /* If KILL for some other reason set the WNOHANG flag since we * can't be certain what happened. */ waitflags = WNOHANG; } if (!child_waitpid(match, waitflags)) { /* not much we can do if this occurs */ return FALSE; } child_list = g_list_remove(child_list, match); child_free(match); return TRUE; } /* Create/Log a new tracked process * To track a process group, use -pid * * @TODO Using a non-positive pid (i.e. any child, or process group) would * likely not be useful since we will free the child after the first * completed process. */ void mainloop_child_add_with_flags(pid_t pid, int timeout, const char *desc, void *privatedata, enum mainloop_child_flags flags, void (*callback) (mainloop_child_t * p, pid_t pid, int core, int signo, int exitcode)) { static bool need_init = TRUE; mainloop_child_t *child = pcmk__assert_alloc(1, sizeof(mainloop_child_t)); child->pid = pid; child->timerid = 0; child->timeout = FALSE; child->privatedata = privatedata; child->callback = callback; child->flags = flags; child->desc = pcmk__str_copy(desc); if (timeout) { child->timerid = g_timeout_add(timeout, child_timeout_callback, child); } child_list = g_list_append(child_list, child); if(need_init) { need_init = FALSE; /* SIGCHLD processing has to be invoked from mainloop. * We do not want it to be possible to both add a child pid * to mainloop, and have the pid's exit callback invoked within * the same callstack. */ g_timeout_add(1, child_signal_init, NULL); } } void mainloop_child_add(pid_t pid, int timeout, const char *desc, void *privatedata, void (*callback) (mainloop_child_t * p, pid_t pid, int core, int signo, int exitcode)) { mainloop_child_add_with_flags(pid, timeout, desc, privatedata, 0, callback); } static gboolean mainloop_timer_cb(gpointer user_data) { int id = 0; bool repeat = FALSE; struct mainloop_timer_s *t = user_data; CRM_ASSERT(t != NULL); id = t->id; t->id = 0; /* Ensure it's unset during callbacks so that * mainloop_timer_running() works as expected */ if(t->cb) { crm_trace("Invoking callbacks for timer %s", t->name); repeat = t->repeat; if(t->cb(t->userdata) == FALSE) { crm_trace("Timer %s complete", t->name); repeat = FALSE; } } if(repeat) { /* Restore if repeating */ t->id = id; } return repeat; } bool mainloop_timer_running(mainloop_timer_t *t) { if(t && t->id != 0) { return TRUE; } return FALSE; } void mainloop_timer_start(mainloop_timer_t *t) { mainloop_timer_stop(t); if(t && t->period_ms > 0) { crm_trace("Starting timer %s", t->name); t->id = g_timeout_add(t->period_ms, mainloop_timer_cb, t); } } void mainloop_timer_stop(mainloop_timer_t *t) { if(t && t->id != 0) { crm_trace("Stopping timer %s", t->name); g_source_remove(t->id); t->id = 0; } } guint mainloop_timer_set_period(mainloop_timer_t *t, guint period_ms) { guint last = 0; if(t) { last = t->period_ms; t->period_ms = period_ms; } if(t && t->id != 0 && last != t->period_ms) { mainloop_timer_start(t); } return last; } mainloop_timer_t * mainloop_timer_add(const char *name, guint period_ms, bool repeat, GSourceFunc cb, void *userdata) { mainloop_timer_t *t = pcmk__assert_alloc(1, sizeof(mainloop_timer_t)); if (name != NULL) { t->name = crm_strdup_printf("%s-%u-%d", name, period_ms, repeat); } else { t->name = crm_strdup_printf("%p-%u-%d", t, period_ms, repeat); } t->id = 0; t->period_ms = period_ms; t->repeat = repeat; t->cb = cb; t->userdata = userdata; crm_trace("Created timer %s with %p %p", t->name, userdata, t->userdata); return t; } void mainloop_timer_del(mainloop_timer_t *t) { if(t) { crm_trace("Destroying timer %s", t->name); mainloop_timer_stop(t); free(t->name); free(t); } } /* * Helpers to make sure certain events aren't lost at shutdown */ static gboolean drain_timeout_cb(gpointer user_data) { bool *timeout_popped = (bool*) user_data; *timeout_popped = TRUE; return FALSE; } /*! * \brief Drain some remaining main loop events then quit it * * \param[in,out] mloop Main loop to drain and quit * \param[in] n Drain up to this many pending events */ void pcmk_quit_main_loop(GMainLoop *mloop, unsigned int n) { if ((mloop != NULL) && g_main_loop_is_running(mloop)) { GMainContext *ctx = g_main_loop_get_context(mloop); /* Drain up to n events in case some memory clean-up is pending * (helpful to reduce noise in valgrind output). */ for (int i = 0; (i < n) && g_main_context_pending(ctx); ++i) { g_main_context_dispatch(ctx); } g_main_loop_quit(mloop); } } /*! * \brief Process main loop events while a certain condition is met * * \param[in,out] mloop Main loop to process * \param[in] timer_ms Don't process longer than this amount of time * \param[in] check Function that returns true if events should be * processed * * \note This function is intended to be called at shutdown if certain important * events should not be missed. The caller would likely quit the main loop * or exit after calling this function. The check() function will be * passed the remaining timeout in milliseconds. */ void pcmk_drain_main_loop(GMainLoop *mloop, guint timer_ms, bool (*check)(guint)) { bool timeout_popped = FALSE; guint timer = 0; GMainContext *ctx = NULL; CRM_CHECK(mloop && check, return); ctx = g_main_loop_get_context(mloop); if (ctx) { time_t start_time = time(NULL); timer = g_timeout_add(timer_ms, drain_timeout_cb, &timeout_popped); while (!timeout_popped && check(timer_ms - (time(NULL) - start_time) * 1000)) { g_main_context_iteration(ctx, TRUE); } } if (!timeout_popped && (timer > 0)) { g_source_remove(timer); } } diff --git a/lib/common/options.c b/lib/common/options.c index ee1486f8a8..d26e7807e1 100644 --- a/lib/common/options.c +++ b/lib/common/options.c @@ -1,1565 +1,1561 @@ /* * 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 _GNU_SOURCE -# define _GNU_SOURCE -#endif - #include #include #include #include #include #include #include #include void pcmk__cli_help(char cmd) { if (cmd == 'v' || cmd == '$') { printf("Pacemaker %s\n", PACEMAKER_VERSION); printf("Written by Andrew Beekhof and " "the Pacemaker project contributors\n"); } else if (cmd == '!') { printf("Pacemaker %s (Build: %s): %s\n", PACEMAKER_VERSION, BUILD_VERSION, CRM_FEATURES); } crm_exit(CRM_EX_OK); while(1); // above does not return } /* * Option metadata */ static const pcmk__cluster_option_t cluster_options[] = { /* name, old name, type, allowed values, * default value, validator, * flags, * short description, * long description */ { PCMK_OPT_DC_VERSION, NULL, PCMK_VALUE_VERSION, NULL, NULL, NULL, pcmk__opt_controld|pcmk__opt_generated, N_("Pacemaker version on cluster node elected Designated Controller " "(DC)"), N_("Includes a hash which identifies the exact revision the code was " "built from. Used for diagnostic purposes."), }, { PCMK_OPT_CLUSTER_INFRASTRUCTURE, NULL, PCMK_VALUE_STRING, NULL, NULL, NULL, pcmk__opt_controld|pcmk__opt_generated, N_("The messaging layer on which Pacemaker is currently running"), N_("Used for informational and diagnostic purposes."), }, { PCMK_OPT_CLUSTER_NAME, NULL, PCMK_VALUE_STRING, NULL, NULL, NULL, pcmk__opt_controld, N_("An arbitrary name for the cluster"), N_("This optional value is mostly for users' convenience as desired " "in administration, but may also be used in Pacemaker " "configuration rules via the #cluster-name node attribute, and " "by higher-level tools and resource agents."), }, { PCMK_OPT_DC_DEADTIME, NULL, PCMK_VALUE_DURATION, NULL, "20s", pcmk__valid_interval_spec, pcmk__opt_controld, N_("How long to wait for a response from other nodes during start-up"), N_("The optimal value will depend on the speed and load of your " "network and the type of switches used."), }, { PCMK_OPT_CLUSTER_RECHECK_INTERVAL, NULL, PCMK_VALUE_DURATION, NULL, "15min", pcmk__valid_interval_spec, pcmk__opt_controld, N_("Polling interval to recheck cluster state and evaluate rules " "with date specifications"), N_("Pacemaker is primarily event-driven, and looks ahead to know when " "to recheck cluster state for failure-timeout settings and most " "time-based rules. However, it will also recheck the cluster after " "this amount of inactivity, to evaluate rules with date " "specifications and serve as a fail-safe for certain types of " "scheduler bugs. A value of 0 disables polling. A positive value " "sets an interval in seconds, unless other units are specified " "(for example, \"5min\")."), }, { PCMK_OPT_FENCE_REACTION, NULL, PCMK_VALUE_SELECT, PCMK_VALUE_STOP ", " PCMK_VALUE_PANIC, PCMK_VALUE_STOP, NULL, pcmk__opt_controld, N_("How a cluster node should react if notified of its own fencing"), N_("A cluster node may receive notification of a \"succeeded\" " "fencing that targeted it if fencing is misconfigured, or if " "fabric fencing is in use that doesn't cut cluster communication. " "Use \"stop\" to attempt to immediately stop Pacemaker and stay " "stopped, or \"panic\" to attempt to immediately reboot the local " "node, falling back to stop on failure."), }, { PCMK_OPT_ELECTION_TIMEOUT, NULL, PCMK_VALUE_DURATION, NULL, "2min", pcmk__valid_interval_spec, pcmk__opt_controld|pcmk__opt_advanced, N_("Declare an election failed if it is not decided within this much " "time. If you need to adjust this value, it probably indicates " "the presence of a bug."), NULL, }, { PCMK_OPT_SHUTDOWN_ESCALATION, NULL, PCMK_VALUE_DURATION, NULL, "20min", pcmk__valid_interval_spec, pcmk__opt_controld|pcmk__opt_advanced, N_("Exit immediately if shutdown does not complete within this much " "time. If you need to adjust this value, it probably indicates " "the presence of a bug."), NULL, }, { PCMK_OPT_JOIN_INTEGRATION_TIMEOUT, "crmd-integration-timeout", PCMK_VALUE_DURATION, NULL, "3min", pcmk__valid_interval_spec, pcmk__opt_controld|pcmk__opt_advanced, N_("If you need to adjust this value, it probably indicates " "the presence of a bug."), NULL, }, { PCMK_OPT_JOIN_FINALIZATION_TIMEOUT, "crmd-finalization-timeout", PCMK_VALUE_DURATION, NULL, "30min", pcmk__valid_interval_spec, pcmk__opt_controld|pcmk__opt_advanced, N_("If you need to adjust this value, it probably indicates " "the presence of a bug."), NULL, }, { PCMK_OPT_TRANSITION_DELAY, "crmd-transition-delay", PCMK_VALUE_DURATION, NULL, "0s", pcmk__valid_interval_spec, pcmk__opt_controld|pcmk__opt_advanced, N_("Enabling this option will slow down cluster recovery under all " "conditions"), N_("Delay cluster recovery for this much time to allow for additional " "events to occur. Useful if your configuration is sensitive to " "the order in which ping updates arrive."), }, { PCMK_OPT_NO_QUORUM_POLICY, NULL, PCMK_VALUE_SELECT, PCMK_VALUE_STOP ", " PCMK_VALUE_FREEZE ", " PCMK_VALUE_IGNORE ", " PCMK_VALUE_DEMOTE ", " PCMK_VALUE_FENCE_LEGACY, PCMK_VALUE_STOP, pcmk__valid_no_quorum_policy, pcmk__opt_schedulerd, N_("What to do when the cluster does not have quorum"), NULL, }, { PCMK_OPT_SHUTDOWN_LOCK, NULL, PCMK_VALUE_BOOLEAN, NULL, PCMK_VALUE_FALSE, pcmk__valid_boolean, pcmk__opt_schedulerd, N_("Whether to lock resources to a cleanly shut down node"), N_("When true, resources active on a node when it is cleanly shut down " "are kept \"locked\" to that node (not allowed to run elsewhere) " "until they start again on that node after it rejoins (or for at " "most shutdown-lock-limit, if set). Stonith resources and " "Pacemaker Remote connections are never locked. Clone and bundle " "instances and the promoted role of promotable clones are " "currently never locked, though support could be added in a future " "release."), }, { PCMK_OPT_SHUTDOWN_LOCK_LIMIT, NULL, PCMK_VALUE_DURATION, NULL, "0", pcmk__valid_interval_spec, pcmk__opt_schedulerd, N_("Do not lock resources to a cleanly shut down node longer than " "this"), N_("If shutdown-lock is true and this is set to a nonzero time " "duration, shutdown locks will expire after this much time has " "passed since the shutdown was initiated, even if the node has not " "rejoined."), }, { PCMK_OPT_ENABLE_ACL, NULL, PCMK_VALUE_BOOLEAN, NULL, PCMK_VALUE_FALSE, pcmk__valid_boolean, pcmk__opt_based, N_("Enable Access Control Lists (ACLs) for the CIB"), NULL, }, { PCMK_OPT_SYMMETRIC_CLUSTER, NULL, PCMK_VALUE_BOOLEAN, NULL, PCMK_VALUE_TRUE, pcmk__valid_boolean, pcmk__opt_schedulerd, N_("Whether resources can run on any node by default"), NULL, }, { PCMK_OPT_MAINTENANCE_MODE, NULL, PCMK_VALUE_BOOLEAN, NULL, PCMK_VALUE_FALSE, pcmk__valid_boolean, pcmk__opt_schedulerd, N_("Whether the cluster should refrain from monitoring, starting, and " "stopping resources"), NULL, }, { PCMK_OPT_START_FAILURE_IS_FATAL, NULL, PCMK_VALUE_BOOLEAN, NULL, PCMK_VALUE_TRUE, pcmk__valid_boolean, pcmk__opt_schedulerd, N_("Whether a start failure should prevent a resource from being " "recovered on the same node"), N_("When true, the cluster will immediately ban a resource from a node " "if it fails to start there. When false, the cluster will instead " "check the resource's fail count against its migration-threshold.") }, { PCMK_OPT_ENABLE_STARTUP_PROBES, NULL, PCMK_VALUE_BOOLEAN, NULL, PCMK_VALUE_TRUE, pcmk__valid_boolean, pcmk__opt_schedulerd, N_("Whether the cluster should check for active resources during " "start-up"), NULL, }, // Fencing-related options { PCMK_OPT_STONITH_ENABLED, NULL, PCMK_VALUE_BOOLEAN, NULL, PCMK_VALUE_TRUE, pcmk__valid_boolean, pcmk__opt_schedulerd|pcmk__opt_advanced, N_("Whether nodes may be fenced as part of recovery"), N_("If false, unresponsive nodes are immediately assumed to be " "harmless, and resources that were active on them may be recovered " "elsewhere. This can result in a \"split-brain\" situation, " "potentially leading to data loss and/or service unavailability."), }, { PCMK_OPT_STONITH_ACTION, NULL, PCMK_VALUE_SELECT, PCMK_ACTION_REBOOT ", " PCMK_ACTION_OFF ", " PCMK__ACTION_POWEROFF, PCMK_ACTION_REBOOT, pcmk__is_fencing_action, pcmk__opt_schedulerd, N_("Action to send to fence device when a node needs to be fenced " "(\"poweroff\" is a deprecated alias for \"off\")"), NULL, }, { PCMK_OPT_STONITH_TIMEOUT, NULL, PCMK_VALUE_DURATION, NULL, "60s", pcmk__valid_interval_spec, pcmk__opt_schedulerd, N_("How long to wait for on, off, and reboot fence actions to complete " "by default"), NULL, }, { PCMK_OPT_HAVE_WATCHDOG, NULL, PCMK_VALUE_BOOLEAN, NULL, PCMK_VALUE_FALSE, pcmk__valid_boolean, pcmk__opt_schedulerd|pcmk__opt_generated, N_("Whether watchdog integration is enabled"), N_("This is set automatically by the cluster according to whether SBD " "is detected to be in use. User-configured values are ignored. " "The value `true` is meaningful if diskless SBD is used and " "`stonith-watchdog-timeout` is nonzero. In that case, if fencing " "is required, watchdog-based self-fencing will be performed via " "SBD without requiring a fencing resource explicitly configured."), }, { /* @COMPAT Currently, unparsable values default to -1 (auto-calculate), * while missing values default to 0 (disable). All values are accepted * (unless the controller finds that the value conflicts with the * SBD_WATCHDOG_TIMEOUT). * * At a compatibility break: properly validate as a timeout, let * either negative values or a particular string like "auto" mean auto- * calculate, and use 0 as the single default for when the option either * is unset or fails to validate. */ PCMK_OPT_STONITH_WATCHDOG_TIMEOUT, NULL, PCMK_VALUE_TIMEOUT, NULL, "0", NULL, pcmk__opt_controld, N_("How long before nodes can be assumed to be safely down when " "watchdog-based self-fencing via SBD is in use"), N_("If this is set to a positive value, lost nodes are assumed to " "achieve self-fencing using watchdog-based SBD within this much " "time. This does not require a fencing resource to be explicitly " "configured, though a fence_watchdog resource can be configured, to " "limit use to specific nodes. If this is set to 0 (the default), " "the cluster will never assume watchdog-based self-fencing. If this " "is set to a negative value, the cluster will use twice the local " "value of the `SBD_WATCHDOG_TIMEOUT` environment variable if that " "is positive, or otherwise treat this as 0. WARNING: When used, " "this timeout must be larger than `SBD_WATCHDOG_TIMEOUT` on all " "nodes that use watchdog-based SBD, and Pacemaker will refuse to " "start on any of those nodes where this is not true for the local " "value or SBD is not active. When this is set to a negative value, " "`SBD_WATCHDOG_TIMEOUT` must be set to the same value on all nodes " "that use SBD, otherwise data corruption or loss could occur."), }, { PCMK_OPT_STONITH_MAX_ATTEMPTS, NULL, PCMK_VALUE_SCORE, NULL, "10", pcmk__valid_positive_int, pcmk__opt_controld, N_("How many times fencing can fail before it will no longer be " "immediately re-attempted on a target"), NULL, }, { PCMK_OPT_CONCURRENT_FENCING, NULL, PCMK_VALUE_BOOLEAN, NULL, PCMK__CONCURRENT_FENCING_DEFAULT, pcmk__valid_boolean, pcmk__opt_schedulerd, N_("Allow performing fencing operations in parallel"), NULL, }, { PCMK_OPT_STARTUP_FENCING, NULL, PCMK_VALUE_BOOLEAN, NULL, PCMK_VALUE_TRUE, pcmk__valid_boolean, pcmk__opt_schedulerd|pcmk__opt_advanced, N_("Whether to fence unseen nodes at start-up"), N_("Setting this to false may lead to a \"split-brain\" situation, " "potentially leading to data loss and/or service unavailability."), }, { PCMK_OPT_PRIORITY_FENCING_DELAY, NULL, PCMK_VALUE_DURATION, NULL, "0", pcmk__valid_interval_spec, pcmk__opt_schedulerd, N_("Apply fencing delay targeting the lost nodes with the highest " "total resource priority"), N_("Apply specified delay for the fencings that are targeting the lost " "nodes with the highest total resource priority in case we don't " "have the majority of the nodes in our cluster partition, so that " "the more significant nodes potentially win any fencing match, " "which is especially meaningful under split-brain of 2-node " "cluster. A promoted resource instance takes the base priority + 1 " "on calculation if the base priority is not 0. Any static/random " "delays that are introduced by `pcmk_delay_base/max` configured " "for the corresponding fencing resources will be added to this " "delay. This delay should be significantly greater than, safely " "twice, the maximum `pcmk_delay_base/max`. By default, priority " "fencing delay is disabled."), }, { PCMK_OPT_NODE_PENDING_TIMEOUT, NULL, PCMK_VALUE_DURATION, NULL, "0", pcmk__valid_interval_spec, pcmk__opt_schedulerd, N_("How long to wait for a node that has joined the cluster to join " "the controller process group"), N_("Fence nodes that do not join the controller process group within " "this much time after joining the cluster, to allow the cluster " "to continue managing resources. A value of 0 means never fence " "pending nodes. Setting the value to 2h means fence nodes after " "2 hours."), }, { PCMK_OPT_CLUSTER_DELAY, NULL, PCMK_VALUE_DURATION, NULL, "60s", pcmk__valid_interval_spec, pcmk__opt_schedulerd, N_("Maximum time for node-to-node communication"), N_("The node elected Designated Controller (DC) will consider an action " "failed if it does not get a response from the node executing the " "action within this time (after considering the action's own " "timeout). The \"correct\" value will depend on the speed and " "load of your network and cluster nodes.") }, // Limits { PCMK_OPT_LOAD_THRESHOLD, NULL, PCMK_VALUE_PERCENTAGE, NULL, "80%", pcmk__valid_percentage, pcmk__opt_controld, N_("Maximum amount of system load that should be used by cluster " "nodes"), N_("The cluster will slow down its recovery process when the amount of " "system resources used (currently CPU) approaches this limit"), }, { PCMK_OPT_NODE_ACTION_LIMIT, NULL, PCMK_VALUE_INTEGER, NULL, "0", pcmk__valid_int, pcmk__opt_controld, N_("Maximum number of jobs that can be scheduled per node (defaults to " "2x cores)"), NULL, }, { PCMK_OPT_BATCH_LIMIT, NULL, PCMK_VALUE_INTEGER, NULL, "0", pcmk__valid_int, pcmk__opt_schedulerd, N_("Maximum number of jobs that the cluster may execute in parallel " "across all nodes"), N_("The \"correct\" value will depend on the speed and load of your " "network and cluster nodes. If set to 0, the cluster will " "impose a dynamically calculated limit when any node has a " "high load."), }, { PCMK_OPT_MIGRATION_LIMIT, NULL, PCMK_VALUE_INTEGER, NULL, "-1", pcmk__valid_int, pcmk__opt_schedulerd, N_("The number of live migration actions that the cluster is allowed " "to execute in parallel on a node (-1 means no limit)"), NULL, }, { /* @TODO This is actually ignored if not strictly positive. We should * overhaul value types in Pacemaker Explained. There are lots of * inaccurate ranges (assumptions of 32-bit width, "nonnegative" when * positive is required, etc.). * * Maybe a single integer type with the allowed range specified would be * better. * * Drop the PCMK_VALUE_NONNEGATIVE_INTEGER constant if we do this before * a release. */ PCMK_OPT_CLUSTER_IPC_LIMIT, NULL, PCMK_VALUE_NONNEGATIVE_INTEGER, NULL, "500", pcmk__valid_positive_int, pcmk__opt_based, N_("Maximum IPC message backlog before disconnecting a cluster daemon"), N_("Raise this if log has \"Evicting client\" messages for cluster " "daemon PIDs (a good value is the number of resources in the " "cluster multiplied by the number of nodes)."), }, // Orphans and stopping { PCMK_OPT_STOP_ALL_RESOURCES, NULL, PCMK_VALUE_BOOLEAN, NULL, PCMK_VALUE_FALSE, pcmk__valid_boolean, pcmk__opt_schedulerd, N_("Whether the cluster should stop all active resources"), NULL, }, { PCMK_OPT_STOP_ORPHAN_RESOURCES, NULL, PCMK_VALUE_BOOLEAN, NULL, PCMK_VALUE_TRUE, pcmk__valid_boolean, pcmk__opt_schedulerd, N_("Whether to stop resources that were removed from the " "configuration"), NULL, }, { PCMK_OPT_STOP_ORPHAN_ACTIONS, NULL, PCMK_VALUE_BOOLEAN, NULL, PCMK_VALUE_TRUE, pcmk__valid_boolean, pcmk__opt_schedulerd, N_("Whether to cancel recurring actions removed from the " "configuration"), NULL, }, { PCMK__OPT_REMOVE_AFTER_STOP, NULL, PCMK_VALUE_BOOLEAN, NULL, PCMK_VALUE_FALSE, pcmk__valid_boolean, pcmk__opt_schedulerd|pcmk__opt_deprecated, N_("Whether to remove stopped resources from the executor"), N_("Values other than default are poorly tested and potentially " "dangerous."), }, // Storing inputs { PCMK_OPT_PE_ERROR_SERIES_MAX, NULL, PCMK_VALUE_INTEGER, NULL, "-1", pcmk__valid_int, pcmk__opt_schedulerd, N_("The number of scheduler inputs resulting in errors to save"), N_("Zero to disable, -1 to store unlimited."), }, { PCMK_OPT_PE_WARN_SERIES_MAX, NULL, PCMK_VALUE_INTEGER, NULL, "5000", pcmk__valid_int, pcmk__opt_schedulerd, N_("The number of scheduler inputs resulting in warnings to save"), N_("Zero to disable, -1 to store unlimited."), }, { PCMK_OPT_PE_INPUT_SERIES_MAX, NULL, PCMK_VALUE_INTEGER, NULL, "4000", pcmk__valid_int, pcmk__opt_schedulerd, N_("The number of scheduler inputs without errors or warnings to save"), N_("Zero to disable, -1 to store unlimited."), }, // Node health { PCMK_OPT_NODE_HEALTH_STRATEGY, NULL, PCMK_VALUE_SELECT, PCMK_VALUE_NONE ", " PCMK_VALUE_MIGRATE_ON_RED ", " PCMK_VALUE_ONLY_GREEN ", " PCMK_VALUE_PROGRESSIVE ", " PCMK_VALUE_CUSTOM, PCMK_VALUE_NONE, pcmk__validate_health_strategy, pcmk__opt_schedulerd, N_("How cluster should react to node health attributes"), N_("Requires external entities to create node attributes (named with " "the prefix \"#health\") with values \"red\", \"yellow\", or " "\"green\".") }, { PCMK_OPT_NODE_HEALTH_BASE, NULL, PCMK_VALUE_SCORE, NULL, "0", pcmk__valid_int, pcmk__opt_schedulerd, N_("Base health score assigned to a node"), N_("Only used when \"node-health-strategy\" is set to " "\"progressive\"."), }, { PCMK_OPT_NODE_HEALTH_GREEN, NULL, PCMK_VALUE_SCORE, NULL, "0", pcmk__valid_int, pcmk__opt_schedulerd, N_("The score to use for a node health attribute whose value is " "\"green\""), N_("Only used when \"node-health-strategy\" is set to \"custom\" or " "\"progressive\"."), }, { PCMK_OPT_NODE_HEALTH_YELLOW, NULL, PCMK_VALUE_SCORE, NULL, "0", pcmk__valid_int, pcmk__opt_schedulerd, N_("The score to use for a node health attribute whose value is " "\"yellow\""), N_("Only used when \"node-health-strategy\" is set to \"custom\" or " "\"progressive\"."), }, { PCMK_OPT_NODE_HEALTH_RED, NULL, PCMK_VALUE_SCORE, NULL, "-INFINITY", pcmk__valid_int, pcmk__opt_schedulerd, N_("The score to use for a node health attribute whose value is " "\"red\""), N_("Only used when \"node-health-strategy\" is set to \"custom\" or " "\"progressive\".") }, // Placement strategy { PCMK_OPT_PLACEMENT_STRATEGY, NULL, PCMK_VALUE_SELECT, PCMK_VALUE_DEFAULT ", " PCMK_VALUE_UTILIZATION ", " PCMK_VALUE_MINIMAL ", " PCMK_VALUE_BALANCED, PCMK_VALUE_DEFAULT, pcmk__valid_placement_strategy, pcmk__opt_schedulerd, N_("How the cluster should allocate resources to nodes"), NULL, }, { NULL, }, }; static const pcmk__cluster_option_t fencing_params[] = { /* name, old name, type, allowed values, * default value, validator, * flags, * short description, * long description */ { PCMK_STONITH_HOST_ARGUMENT, NULL, PCMK_VALUE_STRING, NULL, "port", NULL, pcmk__opt_advanced, N_("An alternate parameter to supply instead of 'port'"), N_("Some devices do not support the standard 'port' parameter or may " "provide additional ones. Use this to specify an alternate, device-" "specific, parameter that should indicate the machine to be " "fenced. A value of \"none\" can be used to tell the cluster not " "to supply any additional parameters."), }, { PCMK_STONITH_HOST_MAP, NULL, PCMK_VALUE_STRING, NULL, NULL, NULL, pcmk__opt_none, N_("A mapping of node names to port numbers for devices that do not " "support node names."), N_("For example, \"node1:1;node2:2,3\" would tell the cluster to use " "port 1 for node1 and ports 2 and 3 for node2."), }, { PCMK_STONITH_HOST_LIST, NULL, PCMK_VALUE_STRING, NULL, NULL, NULL, pcmk__opt_none, N_("Nodes targeted by this device"), N_("Comma-separated list of nodes that can be targeted by this device " "(for example, \"node1,node2,node3\"). If pcmk_host_check is " "\"static-list\", either this or pcmk_host_map must be set."), }, { PCMK_STONITH_HOST_CHECK, NULL, PCMK_VALUE_SELECT, PCMK_VALUE_DYNAMIC_LIST ", " PCMK_VALUE_STATIC_LIST ", " PCMK_VALUE_STATUS ", " PCMK_VALUE_NONE, NULL, NULL, pcmk__opt_none, N_("How to determine which nodes can be targeted by the device"), N_("Use \"dynamic-list\" to query the device via the 'list' command; " "\"static-list\" to check the pcmk_host_list attribute; " "\"status\" to query the device via the 'status' command; or " "\"none\" to assume every device can fence every node. " "The default value is \"static-list\" if pcmk_host_map or " "pcmk_host_list is set; otherwise \"dynamic-list\" if the device " "supports the list operation; otherwise \"status\" if the device " "supports the status operation; otherwise \"none\""), }, { PCMK_STONITH_DELAY_MAX, NULL, PCMK_VALUE_DURATION, NULL, "0s", NULL, pcmk__opt_none, N_("Enable a delay of no more than the time specified before executing " "fencing actions."), N_("Enable a delay of no more than the time specified before executing " "fencing actions. Pacemaker derives the overall delay by taking " "the value of pcmk_delay_base and adding a random delay value such " "that the sum is kept below this maximum."), }, { PCMK_STONITH_DELAY_BASE, NULL, PCMK_VALUE_STRING, NULL, "0s", NULL, pcmk__opt_none, N_("Enable a base delay for fencing actions and specify base delay " "value."), N_("This enables a static delay for fencing actions, which can help " "avoid \"death matches\" where two nodes try to fence each other " "at the same time. If pcmk_delay_max is also used, a random delay " "will be added such that the total delay is kept below that value. " "This can be set to a single time value to apply to any node " "targeted by this device (useful if a separate device is " "configured for each target), or to a node map (for example, " "\"node1:1s;node2:5\") to set a different value for each target."), }, { PCMK_STONITH_ACTION_LIMIT, NULL, PCMK_VALUE_INTEGER, NULL, "1", NULL, pcmk__opt_none, N_("The maximum number of actions can be performed in parallel on this " "device"), N_("Cluster property concurrent-fencing=\"true\" needs to be " "configured first. Then use this to specify the maximum number of " "actions can be performed in parallel on this device. A value of " "-1 means an unlimited number of actions can be performed in " "parallel."), }, { "pcmk_reboot_action", NULL, PCMK_VALUE_STRING, NULL, PCMK_ACTION_REBOOT, NULL, pcmk__opt_advanced, N_("An alternate command to run instead of 'reboot'"), N_("Some devices do not support the standard commands or may provide " "additional ones. Use this to specify an alternate, device-" "specific, command that implements the 'reboot' action."), }, { "pcmk_reboot_timeout", NULL, PCMK_VALUE_TIMEOUT, NULL, "60s", NULL, pcmk__opt_advanced, N_("Specify an alternate timeout to use for 'reboot' actions instead " "of stonith-timeout"), N_("Some devices need much more/less time to complete than normal. " "Use this to specify an alternate, device-specific, timeout for " "'reboot' actions."), }, { "pcmk_reboot_retries", NULL, PCMK_VALUE_INTEGER, NULL, "2", NULL, pcmk__opt_advanced, N_("The maximum number of times to try the 'reboot' command within the " "timeout period"), N_("Some devices do not support multiple connections. Operations may " "\"fail\" if the device is busy with another task. In that case, " "Pacemaker will automatically retry the operation if there is time " "remaining. Use this option to alter the number of times Pacemaker " "tries a 'reboot' action before giving up."), }, { "pcmk_off_action", NULL, PCMK_VALUE_STRING, NULL, PCMK_ACTION_OFF, NULL, pcmk__opt_advanced, N_("An alternate command to run instead of 'off'"), N_("Some devices do not support the standard commands or may provide " "additional ones. Use this to specify an alternate, device-" "specific, command that implements the 'off' action."), }, { "pcmk_off_timeout", NULL, PCMK_VALUE_TIMEOUT, NULL, "60s", NULL, pcmk__opt_advanced, N_("Specify an alternate timeout to use for 'off' actions instead of " "stonith-timeout"), N_("Some devices need much more/less time to complete than normal. " "Use this to specify an alternate, device-specific, timeout for " "'off' actions."), }, { "pcmk_off_retries", NULL, PCMK_VALUE_INTEGER, NULL, "2", NULL, pcmk__opt_advanced, N_("The maximum number of times to try the 'off' command within the " "timeout period"), N_("Some devices do not support multiple connections. Operations may " "\"fail\" if the device is busy with another task. In that case, " "Pacemaker will automatically retry the operation if there is time " "remaining. Use this option to alter the number of times Pacemaker " "tries a 'off' action before giving up."), }, { "pcmk_on_action", NULL, PCMK_VALUE_STRING, NULL, PCMK_ACTION_ON, NULL, pcmk__opt_advanced, N_("An alternate command to run instead of 'on'"), N_("Some devices do not support the standard commands or may provide " "additional ones. Use this to specify an alternate, device-" "specific, command that implements the 'on' action."), }, { "pcmk_on_timeout", NULL, PCMK_VALUE_TIMEOUT, NULL, "60s", NULL, pcmk__opt_advanced, N_("Specify an alternate timeout to use for 'on' actions instead of " "stonith-timeout"), N_("Some devices need much more/less time to complete than normal. " "Use this to specify an alternate, device-specific, timeout for " "'on' actions."), }, { "pcmk_on_retries", NULL, PCMK_VALUE_INTEGER, NULL, "2", NULL, pcmk__opt_advanced, N_("The maximum number of times to try the 'on' command within the " "timeout period"), N_("Some devices do not support multiple connections. Operations may " "\"fail\" if the device is busy with another task. In that case, " "Pacemaker will automatically retry the operation if there is time " "remaining. Use this option to alter the number of times Pacemaker " "tries a 'on' action before giving up."), }, { "pcmk_list_action", NULL, PCMK_VALUE_STRING, NULL, PCMK_ACTION_LIST, NULL, pcmk__opt_advanced, N_("An alternate command to run instead of 'list'"), N_("Some devices do not support the standard commands or may provide " "additional ones. Use this to specify an alternate, device-" "specific, command that implements the 'list' action."), }, { "pcmk_list_timeout", NULL, PCMK_VALUE_TIMEOUT, NULL, "60s", NULL, pcmk__opt_advanced, N_("Specify an alternate timeout to use for 'list' actions instead of " "stonith-timeout"), N_("Some devices need much more/less time to complete than normal. " "Use this to specify an alternate, device-specific, timeout for " "'list' actions."), }, { "pcmk_list_retries", NULL, PCMK_VALUE_INTEGER, NULL, "2", NULL, pcmk__opt_advanced, N_("The maximum number of times to try the 'list' command within the " "timeout period"), N_("Some devices do not support multiple connections. Operations may " "\"fail\" if the device is busy with another task. In that case, " "Pacemaker will automatically retry the operation if there is time " "remaining. Use this option to alter the number of times Pacemaker " "tries a 'list' action before giving up."), }, { "pcmk_monitor_action", NULL, PCMK_VALUE_STRING, NULL, PCMK_ACTION_MONITOR, NULL, pcmk__opt_advanced, N_("An alternate command to run instead of 'monitor'"), N_("Some devices do not support the standard commands or may provide " "additional ones. Use this to specify an alternate, device-" "specific, command that implements the 'monitor' action."), }, { "pcmk_monitor_timeout", NULL, PCMK_VALUE_TIMEOUT, NULL, "60s", NULL, pcmk__opt_advanced, N_("Specify an alternate timeout to use for 'monitor' actions instead " "of stonith-timeout"), N_("Some devices need much more/less time to complete than normal. " "Use this to specify an alternate, device-specific, timeout for " "'monitor' actions."), }, { "pcmk_monitor_retries", NULL, PCMK_VALUE_INTEGER, NULL, "2", NULL, pcmk__opt_advanced, N_("The maximum number of times to try the 'monitor' command within " "the timeout period"), N_("Some devices do not support multiple connections. Operations may " "\"fail\" if the device is busy with another task. In that case, " "Pacemaker will automatically retry the operation if there is time " "remaining. Use this option to alter the number of times Pacemaker " "tries a 'monitor' action before giving up."), }, { "pcmk_status_action", NULL, PCMK_VALUE_STRING, NULL, PCMK_ACTION_STATUS, NULL, pcmk__opt_advanced, N_("An alternate command to run instead of 'status'"), N_("Some devices do not support the standard commands or may provide " "additional ones. Use this to specify an alternate, device-" "specific, command that implements the 'status' action."), }, { "pcmk_status_timeout", NULL, PCMK_VALUE_TIMEOUT, NULL, "60s", NULL, pcmk__opt_advanced, N_("Specify an alternate timeout to use for 'status' actions instead " "of stonith-timeout"), N_("Some devices need much more/less time to complete than normal. " "Use this to specify an alternate, device-specific, timeout for " "'status' actions."), }, { "pcmk_status_retries", NULL, PCMK_VALUE_INTEGER, NULL, "2", NULL, pcmk__opt_advanced, N_("The maximum number of times to try the 'status' command within " "the timeout period"), N_("Some devices do not support multiple connections. Operations may " "\"fail\" if the device is busy with another task. In that case, " "Pacemaker will automatically retry the operation if there is time " "remaining. Use this option to alter the number of times Pacemaker " "tries a 'status' action before giving up."), }, { NULL, }, }; static const pcmk__cluster_option_t primitive_meta[] = { /* name, old name, type, allowed values, * default value, validator, * flags, * short description, * long description */ { PCMK_META_PRIORITY, NULL, PCMK_VALUE_SCORE, NULL, "0", NULL, pcmk__opt_none, N_("Resource assignment priority"), N_("If not all resources can be active, the cluster will stop " "lower-priority resources in order to keep higher-priority ones " "active."), }, { PCMK_META_CRITICAL, NULL, PCMK_VALUE_BOOLEAN, NULL, PCMK_VALUE_TRUE, NULL, pcmk__opt_none, N_("Default value for influence in colocation constraints"), N_("Use this value as the default for influence in all colocation " "constraints involving this resource, as well as in the implicit " "colocation constraints created if this resource is in a group."), }, { PCMK_META_TARGET_ROLE, NULL, PCMK_VALUE_SELECT, PCMK_ROLE_STOPPED ", " PCMK_ROLE_STARTED ", " PCMK_ROLE_UNPROMOTED ", " PCMK_ROLE_PROMOTED, PCMK_ROLE_STARTED, NULL, pcmk__opt_none, N_("State the cluster should attempt to keep this resource in"), N_("\"Stopped\" forces the resource to be stopped. " "\"Started\" allows the resource to be started (and in the case of " "promotable clone resources, promoted if appropriate). " "\"Unpromoted\" allows the resource to be started, but only in the " "unpromoted role if the resource is promotable. " "\"Promoted\" is equivalent to \"Started\"."), }, { PCMK_META_IS_MANAGED, NULL, PCMK_VALUE_BOOLEAN, NULL, PCMK_VALUE_TRUE, NULL, pcmk__opt_none, N_("Whether the cluster is allowed to actively change the resource's " "state"), N_("If false, the cluster will not start, stop, promote, or demote the " "resource on any node. Recurring actions for the resource are " "unaffected. If true, a true value for the maintenance-mode " "cluster option, the maintenance node attribute, or the " "maintenance resource meta-attribute overrides this."), }, { PCMK_META_MAINTENANCE, NULL, PCMK_VALUE_BOOLEAN, NULL, PCMK_VALUE_FALSE, NULL, pcmk__opt_none, N_("If true, the cluster will not schedule any actions involving the " "resource"), N_("If true, the cluster will not start, stop, promote, or demote the " "resource on any node, and will pause any recurring monitors " "(except those specifying role as \"Stopped\"). If false, a true " "value for the maintenance-mode cluster option or maintenance node " "attribute overrides this."), }, { PCMK_META_RESOURCE_STICKINESS, NULL, PCMK_VALUE_SCORE, NULL, NULL, NULL, pcmk__opt_none, N_("Score to add to the current node when a resource is already " "active"), N_("Score to add to the current node when a resource is already " "active. This allows running resources to stay where they are, " "even if they would be placed elsewhere if they were being started " "from a stopped state. " "The default is 1 for individual clone instances, and 0 for all " "other resources."), }, { PCMK_META_REQUIRES, NULL, PCMK_VALUE_SELECT, PCMK_VALUE_NOTHING ", " PCMK_VALUE_QUORUM ", " PCMK_VALUE_FENCING ", " PCMK_VALUE_UNFENCING, NULL, NULL, pcmk__opt_none, N_("Conditions under which the resource can be started"), N_("Conditions under which the resource can be started. " "\"nothing\" means the cluster can always start this resource. " "\"quorum\" means the cluster can start this resource only if a " "majority of the configured nodes are active. " "\"fencing\" means the cluster can start this resource only if a " "majority of the configured nodes are active and any failed or " "unknown nodes have been fenced. " "\"unfencing\" means the cluster can start this resource only if " "a majority of the configured nodes are active and any failed or " "unknown nodes have been fenced, and only on nodes that have been " "unfenced. " "The default is \"quorum\" for resources with a class of stonith; " "otherwise, \"unfencing\" if unfencing is active in the cluster; " "otherwise, \"fencing\" if the stonith-enabled cluster option is " "true; " "otherwise, \"quorum\"."), }, { PCMK_META_MIGRATION_THRESHOLD, NULL, PCMK_VALUE_SCORE, NULL, PCMK_VALUE_INFINITY, NULL, pcmk__opt_none, N_("Number of failures on a node before the resource becomes " "ineligible to run there."), N_("Number of failures that may occur for this resource on a node, " "before that node is marked ineligible to host this resource. A " "value of 0 indicates that this feature is disabled (the node will " "never be marked ineligible). By contrast, the cluster treats " "\"INFINITY\" (the default) as a very large but finite number. " "This option has an effect only if the failed operation specifies " "its on-fail attribute as \"restart\" (the default), and " "additionally for failed start operations, if the " "start-failure-is-fatal cluster property is set to false."), }, { PCMK_META_FAILURE_TIMEOUT, NULL, PCMK_VALUE_DURATION, NULL, "0", NULL, pcmk__opt_none, N_("Number of seconds before acting as if a failure had not occurred"), N_("Number of seconds after a failed action for this resource before " "acting as if the failure had not occurred, and potentially " "allowing the resource back to the node on which it failed. " "A value of 0 indicates that this feature is disabled."), }, { PCMK_META_MULTIPLE_ACTIVE, NULL, PCMK_VALUE_SELECT, PCMK_VALUE_BLOCK ", " PCMK_VALUE_STOP_ONLY ", " PCMK_VALUE_STOP_START ", " PCMK_VALUE_STOP_UNEXPECTED, PCMK_VALUE_STOP_START, NULL, pcmk__opt_none, N_("What to do if the cluster finds the resource active on more than " "one node"), N_("What to do if the cluster finds the resource active on more than " "one node. " "\"block\" means to mark the resource as unmanaged. " "\"stop_only\" means to stop all active instances of this resource " "and leave them stopped. " "\"stop_start\" means to stop all active instances of this " "resource and start the resource in one location only. " "\"stop_unexpected\" means to stop all active instances of this " "resource except where the resource should be active. (This should " "be used only when extra instances are not expected to disrupt " "existing instances, and the resource agent's monitor of an " "existing instance is capable of detecting any problems that could " "be caused. Note that any resources ordered after this one will " "still need to be restarted.)"), }, { PCMK_META_ALLOW_MIGRATE, NULL, PCMK_VALUE_BOOLEAN, NULL, NULL, NULL, pcmk__opt_none, N_("Whether the cluster should try to \"live migrate\" this resource " "when it needs to be moved"), N_("Whether the cluster should try to \"live migrate\" this resource " "when it needs to be moved. " "The default is true for ocf:pacemaker:remote resources, and false " "otherwise."), }, { PCMK_META_ALLOW_UNHEALTHY_NODES, NULL, PCMK_VALUE_BOOLEAN, NULL, PCMK_VALUE_FALSE, NULL, pcmk__opt_none, N_("Whether the resource should be allowed to run on a node even if " "the node's health score would otherwise prevent it"), NULL, }, { PCMK_META_CONTAINER_ATTRIBUTE_TARGET, NULL, PCMK_VALUE_STRING, NULL, NULL, NULL, pcmk__opt_none, N_("Where to check user-defined node attributes"), N_("Whether to check user-defined node attributes on the physical host " "where a container is running or on the local node. This is " "usually set for a bundle resource and inherited by the bundle's " "primitive resource. " "A value of \"host\" means to check user-defined node attributes " "on the underlying physical host. Any other value means to check " "user-defined node attributes on the local node (for a bundled " "primitive resource, this is the bundle node)."), }, { PCMK_META_REMOTE_NODE, NULL, PCMK_VALUE_STRING, NULL, NULL, NULL, pcmk__opt_none, N_("Name of the Pacemaker Remote guest node this resource is " "associated with, if any"), N_("Name of the Pacemaker Remote guest node this resource is " "associated with, if any. If specified, this both enables the " "resource as a guest node and defines the unique name used to " "identify the guest node. The guest must be configured to run the " "Pacemaker Remote daemon when it is started. " "WARNING: This value cannot overlap with any resource or node " "IDs."), }, { PCMK_META_REMOTE_ADDR, NULL, PCMK_VALUE_STRING, NULL, NULL, NULL, pcmk__opt_none, N_("If remote-node is specified, the IP address or hostname used to " "connect to the guest via Pacemaker Remote"), N_("If remote-node is specified, the IP address or hostname used to " "connect to the guest via Pacemaker Remote. The Pacemaker Remote " "daemon on the guest must be configured to accept connections on " "this address. " "The default is the value of the remote-node meta-attribute."), }, { PCMK_META_REMOTE_PORT, NULL, PCMK_VALUE_PORT, NULL, "3121", NULL, pcmk__opt_none, N_("If remote-node is specified, port on the guest used for its " "Pacemaker Remote connection"), N_("If remote-node is specified, the port on the guest used for its " "Pacemaker Remote connection. The Pacemaker Remote daemon on the " "guest must be configured to listen on this port."), }, { PCMK_META_REMOTE_CONNECT_TIMEOUT, NULL, PCMK_VALUE_TIMEOUT, NULL, "60s", NULL, pcmk__opt_none, N_("If remote-node is specified, how long before a pending Pacemaker " "Remote guest connection times out."), NULL, }, { PCMK_META_REMOTE_ALLOW_MIGRATE, NULL, PCMK_VALUE_BOOLEAN, NULL, PCMK_VALUE_TRUE, NULL, pcmk__opt_none, N_("If remote-node is specified, this acts as the allow-migrate " "meta-attribute for the implicit remote connection resource " "(ocf:pacemaker:remote)."), NULL, }, { NULL, }, }; /* * Environment variable option handling */ /*! * \internal * \brief Get the value of a Pacemaker environment variable option * * If an environment variable option is set, with either a PCMK_ or (for * backward compatibility) HA_ prefix, log and return the value. * * \param[in] option Environment variable name (without prefix) * * \return Value of environment variable option, or NULL in case of * option name too long or value not found */ const char * pcmk__env_option(const char *option) { const char *const prefixes[] = {"PCMK_", "HA_"}; char env_name[NAME_MAX]; const char *value = NULL; CRM_CHECK(!pcmk__str_empty(option), return NULL); for (int i = 0; i < PCMK__NELEM(prefixes); i++) { int rv = snprintf(env_name, NAME_MAX, "%s%s", prefixes[i], option); if (rv < 0) { crm_err("Failed to write %s%s to buffer: %s", prefixes[i], option, strerror(errno)); return NULL; } if (rv >= sizeof(env_name)) { crm_trace("\"%s%s\" is too long", prefixes[i], option); continue; } value = getenv(env_name); if (value != NULL) { crm_trace("Found %s = %s", env_name, value); return value; } } crm_trace("Nothing found for %s", option); return NULL; } /*! * \brief Set or unset a Pacemaker environment variable option * * Set an environment variable option with a \c "PCMK_" prefix and optionally * an \c "HA_" prefix for backward compatibility. * * \param[in] option Environment variable name (without prefix) * \param[in] value New value (or NULL to unset) * \param[in] compat If false and \p value is not \c NULL, set only * \c "PCMK_