diff --git a/include/crm/cluster.h b/include/crm/cluster.h index a110587412..88b69a6959 100644 --- a/include/crm/cluster.h +++ b/include/crm/cluster.h @@ -1,229 +1,232 @@ /* * Copyright 2004-2021 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #ifndef PCMK__CRM_CLUSTER__H # define PCMK__CRM_CLUSTER__H # include // uint32_t, uint64_t # include // gboolean, GHashTable # include // xmlNode # include # include #ifdef __cplusplus extern "C" { #endif # if SUPPORT_COROSYNC # include # endif extern gboolean crm_have_quorum; extern GHashTable *crm_peer_cache; extern GHashTable *crm_remote_peer_cache; extern unsigned long long crm_peer_seq; #define CRM_NODE_LOST "lost" #define CRM_NODE_MEMBER "member" enum crm_join_phase { /* @COMPAT: crm_join_nack_quiet can be replaced by crm_node_t:user_data * at a compatibility break. */ //! Not allowed to join, but don't send a nack message crm_join_nack_quiet = -2, crm_join_nack = -1, crm_join_none = 0, crm_join_welcomed = 1, crm_join_integrated = 2, crm_join_finalized = 3, crm_join_confirmed = 4, }; enum crm_node_flags { /* node is not a cluster node and should not be considered for cluster membership */ crm_remote_node = 0x0001, /* node's cache entry is dirty */ crm_node_dirty = 0x0010, }; typedef struct crm_peer_node_s { char *uname; // Node name as known to cluster char *uuid; // Node UUID to ensure uniqueness char *state; // @TODO change to enum uint64_t flags; // Bitmask of crm_node_flags uint64_t last_seen; // Only needed by cluster nodes uint32_t processes; // @TODO most not needed, merge into flags /* @TODO When we can break public API compatibility, we can make the rest of * these members separate structs and use void *cluster_data and * void *user_data here instead, to abstract the cluster layer further. */ // Currently only needed by corosync stack uint32_t id; // Node ID time_t when_lost; // When CPG membership was last lost // Only used by controller enum crm_join_phase join; char *expected; } crm_node_t; void crm_peer_init(void); void crm_peer_destroy(void); typedef struct crm_cluster_s { char *uuid; char *uname; uint32_t nodeid; void (*destroy) (gpointer); # if SUPPORT_COROSYNC /* @TODO When we can break public API compatibility, make these members a * separate struct and use void *cluster_data here instead, to abstract the * cluster layer further. */ struct cpg_name group; cpg_callbacks_t cpg; cpg_handle_t cpg_handle; # endif } crm_cluster_t; gboolean crm_cluster_connect(crm_cluster_t *cluster); void crm_cluster_disconnect(crm_cluster_t *cluster); +crm_cluster_t *pcmk_cluster_new(void); +void pcmk_cluster_free(crm_cluster_t *cluster); + enum crm_ais_msg_class { crm_class_cluster = 0, }; enum crm_ais_msg_types { crm_msg_none = 0, crm_msg_ais = 1, crm_msg_lrmd = 2, crm_msg_cib = 3, crm_msg_crmd = 4, crm_msg_attrd = 5, crm_msg_stonithd = 6, crm_msg_te = 7, crm_msg_pe = 8, crm_msg_stonith_ng = 9, }; /* used with crm_get_peer_full */ enum crm_get_peer_flags { CRM_GET_PEER_CLUSTER = 0x0001, CRM_GET_PEER_REMOTE = 0x0002, CRM_GET_PEER_ANY = CRM_GET_PEER_CLUSTER|CRM_GET_PEER_REMOTE, }; gboolean send_cluster_message(crm_node_t *node, enum crm_ais_msg_types service, xmlNode *data, gboolean ordered); int crm_remote_peer_cache_size(void); /* Initialize and refresh the remote peer cache from a cib config */ void crm_remote_peer_cache_refresh(xmlNode *cib); crm_node_t *crm_remote_peer_get(const char *node_name); void crm_remote_peer_cache_remove(const char *node_name); /* allows filtering of remote and cluster nodes using crm_get_peer_flags */ crm_node_t *crm_get_peer_full(unsigned int id, const char *uname, int flags); /* only searches cluster nodes */ crm_node_t *crm_get_peer(unsigned int id, const char *uname); guint crm_active_peers(void); gboolean crm_is_peer_active(const crm_node_t * node); guint reap_crm_member(uint32_t id, const char *name); # if SUPPORT_COROSYNC uint32_t get_local_nodeid(cpg_handle_t handle); gboolean cluster_connect_cpg(crm_cluster_t *cluster); void cluster_disconnect_cpg(crm_cluster_t * cluster); void pcmk_cpg_membership(cpg_handle_t handle, const struct cpg_name *groupName, const struct cpg_address *member_list, size_t member_list_entries, const struct cpg_address *left_list, size_t left_list_entries, const struct cpg_address *joined_list, size_t joined_list_entries); gboolean crm_is_corosync_peer_active(const crm_node_t * node); gboolean send_cluster_text(enum crm_ais_msg_class msg_class, const char *data, gboolean local, crm_node_t * node, enum crm_ais_msg_types dest); char *pcmk_message_common_cs(cpg_handle_t handle, uint32_t nodeid, uint32_t pid, void *msg, uint32_t *kind, const char **from); # endif const char *crm_peer_uuid(crm_node_t *node); const char *crm_peer_uname(const char *uuid); void set_uuid(xmlNode *xml, const char *attr, crm_node_t *node); enum crm_status_type { crm_status_uname, crm_status_nstate, crm_status_processes, }; enum crm_ais_msg_types text2msg_type(const char *text); void crm_set_status_callback(void (*dispatch) (enum crm_status_type, crm_node_t *, const void *)); void crm_set_autoreap(gboolean autoreap); enum cluster_type_e { pcmk_cluster_unknown = 0x0001, pcmk_cluster_invalid = 0x0002, // 0x0004 was heartbeat // 0x0010 was corosync 1 with plugin pcmk_cluster_corosync = 0x0020, // 0x0040 was corosync 1 with CMAN }; enum cluster_type_e get_cluster_type(void); const char *name_for_cluster_type(enum cluster_type_e type); gboolean is_corosync_cluster(void); const char *get_local_node_name(void); char *get_node_name(uint32_t nodeid); /*! * \brief Get log-friendly string equivalent of a join phase * * \param[in] phase Join phase * * \return Log-friendly string equivalent of \p phase */ static inline const char * crm_join_phase_str(enum crm_join_phase phase) { switch (phase) { case crm_join_nack_quiet: return "nack_quiet"; case crm_join_nack: return "nack"; case crm_join_none: return "none"; case crm_join_welcomed: return "welcomed"; case crm_join_integrated: return "integrated"; case crm_join_finalized: return "finalized"; case crm_join_confirmed: return "confirmed"; default: return "invalid"; } } #if !defined(PCMK_ALLOW_DEPRECATED) || (PCMK_ALLOW_DEPRECATED == 1) #include #endif #ifdef __cplusplus } #endif #endif diff --git a/lib/cluster/cluster.c b/lib/cluster/cluster.c index 09192510d9..27117b8532 100644 --- a/lib/cluster/cluster.c +++ b/lib/cluster/cluster.c @@ -1,378 +1,410 @@ /* * Copyright 2004-2021 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include "crmcluster_private.h" CRM_TRACE_INIT_DATA(cluster); /*! * \brief Get (and set if needed) a node's UUID * * \param[in] peer Node to check * * \return Node UUID of \p peer, or NULL if unknown */ const char * crm_peer_uuid(crm_node_t *peer) { char *uuid = NULL; // Check simple cases first, to avoid any calls that might block if (peer == NULL) { return NULL; } if (peer->uuid != NULL) { return peer->uuid; } switch (get_cluster_type()) { case pcmk_cluster_corosync: #if SUPPORT_COROSYNC uuid = pcmk__corosync_uuid(peer); #endif break; case pcmk_cluster_unknown: case pcmk_cluster_invalid: crm_err("Unsupported cluster type"); break; } peer->uuid = uuid; return peer->uuid; } /*! * \brief Connect to the cluster layer * * \param[in] Initialized cluster object to connect * * \return TRUE on success, otherwise FALSE */ gboolean crm_cluster_connect(crm_cluster_t *cluster) { enum cluster_type_e type = get_cluster_type(); crm_notice("Connecting to %s cluster infrastructure", name_for_cluster_type(type)); switch (type) { case pcmk_cluster_corosync: #if SUPPORT_COROSYNC if (is_corosync_cluster()) { crm_peer_init(); return pcmk__corosync_connect(cluster); } #endif break; default: break; } return FALSE; } /*! * \brief Disconnect from the cluster layer * * \param[in] cluster Cluster object to disconnect */ void crm_cluster_disconnect(crm_cluster_t *cluster) { enum cluster_type_e type = get_cluster_type(); crm_info("Disconnecting from %s cluster infrastructure", name_for_cluster_type(type)); switch (type) { case pcmk_cluster_corosync: #if SUPPORT_COROSYNC if (is_corosync_cluster()) { crm_peer_destroy(); pcmk__corosync_disconnect(cluster); } #endif break; default: break; } } +/*! + * \brief Allocate a new \p crm_cluster_t object + * + * \return A newly allocated \p crm_cluster_t object (guaranteed not \p NULL) + * \note The caller is responsible for freeing the return value using + * \p pcmk_cluster_free(). + */ +crm_cluster_t * +pcmk_cluster_new(void) +{ + crm_cluster_t *cluster = calloc(1, sizeof(crm_cluster_t)); + + CRM_ASSERT(cluster != NULL); + return cluster; +} + +/*! + * \brief Free a \p crm_cluster_t object and its dynamically allocated members + * + * \param[in,out] cluster Cluster object to free + */ +void +pcmk_cluster_free(crm_cluster_t *cluster) +{ + if (cluster == NULL) { + return; + } + free(cluster->uuid); + free(cluster->uname); + free(cluster); +} + /*! * \brief Send an XML message via the cluster messaging layer * * \param[in] node Cluster node to send message to * \param[in] service Message type to use in message host info * \param[in] data XML message to send * \param[in] ordered Ignored for currently supported messaging layers * * \return TRUE on success, otherwise FALSE */ gboolean send_cluster_message(crm_node_t *node, enum crm_ais_msg_types service, xmlNode *data, gboolean ordered) { switch (get_cluster_type()) { case pcmk_cluster_corosync: #if SUPPORT_COROSYNC return pcmk__cpg_send_xml(data, node, service); #endif break; default: break; } return FALSE; } /*! * \brief Get the local node's name * * \return Local node's name * \note This will fatally exit if local node name cannot be known. */ const char * get_local_node_name(void) { static char *name = NULL; if (name == NULL) { name = get_node_name(0); } return name; } /*! * \brief Get the node name corresponding to a cluster node ID * * \param[in] nodeid Node ID to check (or 0 for local node) * * \return Node name corresponding to \p nodeid * \note This will fatally exit if \p nodeid is 0 and local node name cannot be * known. */ char * get_node_name(uint32_t nodeid) { char *name = NULL; enum cluster_type_e stack = get_cluster_type(); switch (stack) { # if SUPPORT_COROSYNC case pcmk_cluster_corosync: name = pcmk__corosync_name(0, nodeid); break; # endif default: crm_err("Unknown cluster type: %s (%d)", name_for_cluster_type(stack), stack); } if ((name == NULL) && (nodeid == 0)) { name = pcmk_hostname(); if (name == NULL) { // @TODO Maybe let the caller decide what to do crm_err("Could not obtain the local %s node name", name_for_cluster_type(stack)); crm_exit(CRM_EX_FATAL); } crm_notice("Defaulting to uname -n for the local %s node name", name_for_cluster_type(stack)); } if (name == NULL) { crm_notice("Could not obtain a node name for %s node with id %u", name_for_cluster_type(stack), nodeid); } return name; } /*! * \brief Get the node name corresponding to a node UUID * * \param[in] uuid UUID of desired node * * \return name of desired node * * \note This relies on the remote peer cache being populated with all * remote nodes in the cluster, so callers should maintain that cache. */ const char * crm_peer_uname(const char *uuid) { GHashTableIter iter; crm_node_t *node = NULL; CRM_CHECK(uuid != NULL, return NULL); /* remote nodes have the same uname and uuid */ if (g_hash_table_lookup(crm_remote_peer_cache, uuid)) { return uuid; } /* avoid blocking calls where possible */ g_hash_table_iter_init(&iter, crm_peer_cache); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &node)) { if (pcmk__str_eq(node->uuid, uuid, pcmk__str_casei)) { if (node->uname != NULL) { return node->uname; } break; } } node = NULL; #if SUPPORT_COROSYNC if (is_corosync_cluster()) { long long id; if ((pcmk__scan_ll(uuid, &id, 0LL) != pcmk_rc_ok) || (id < 1LL) || (id > UINT32_MAX)) { crm_err("Invalid Corosync node ID '%s'", uuid); return NULL; } node = pcmk__search_cluster_node_cache((uint32_t) id, NULL); if (node != NULL) { crm_info("Setting uuid for node %s[%u] to %s", node->uname, node->id, uuid); node->uuid = strdup(uuid); return node->uname; } return NULL; } #endif return NULL; } /*! * \brief Add a node's UUID as an XML attribute * * \param[in,out] xml XML element to add UUID to * \param[in] attr XML attribute name to set * \param[in] node Node whose UUID should be used as attribute value */ void set_uuid(xmlNode *xml, const char *attr, crm_node_t *node) { crm_xml_add(xml, attr, crm_peer_uuid(node)); } /*! * \brief Get a log-friendly string equivalent of a cluster type * * \param[in] type Cluster type * * \return Log-friendly string corresponding to \p type */ const char * name_for_cluster_type(enum cluster_type_e type) { switch (type) { case pcmk_cluster_corosync: return "corosync"; case pcmk_cluster_unknown: return "unknown"; case pcmk_cluster_invalid: return "invalid"; } crm_err("Invalid cluster type: %d", type); return "invalid"; } /*! * \brief Get (and validate) the local cluster type * * \return Local cluster type * \note This will fatally exit if the local cluster type is invalid. */ enum cluster_type_e get_cluster_type(void) { bool detected = false; const char *cluster = NULL; static enum cluster_type_e cluster_type = pcmk_cluster_unknown; /* Return the previous calculation, if any */ if (cluster_type != pcmk_cluster_unknown) { return cluster_type; } cluster = pcmk__env_option(PCMK__ENV_CLUSTER_TYPE); #if SUPPORT_COROSYNC /* If nothing is defined in the environment, try corosync (if supported) */ if (cluster == NULL) { crm_debug("Testing with Corosync"); cluster_type = pcmk__corosync_detect(); if (cluster_type != pcmk_cluster_unknown) { detected = true; goto done; } } #endif /* Something was defined in the environment, test it against what we support */ crm_info("Verifying cluster type: '%s'", ((cluster == NULL)? "-unspecified-" : cluster)); if (cluster == NULL) { #if SUPPORT_COROSYNC } else if (pcmk__str_eq(cluster, "corosync", pcmk__str_casei)) { cluster_type = pcmk_cluster_corosync; #endif } else { cluster_type = pcmk_cluster_invalid; goto done; /* Keep the compiler happy when no stacks are supported */ } done: if (cluster_type == pcmk_cluster_unknown) { crm_notice("Could not determine the current cluster type"); } else if (cluster_type == pcmk_cluster_invalid) { crm_notice("This installation does not support the '%s' cluster infrastructure: terminating.", cluster); crm_exit(CRM_EX_FATAL); } else { crm_info("%s an active '%s' cluster", (detected? "Detected" : "Assuming"), name_for_cluster_type(cluster_type)); } return cluster_type; } /*! * \brief Check whether the local cluster is a Corosync cluster * * \return TRUE if the local cluster is a Corosync cluster, otherwise FALSE */ gboolean is_corosync_cluster(void) { return get_cluster_type() == pcmk_cluster_corosync; }