X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fptlrpc%2Fnodemap_handler.c;h=59b17349259b351e66e55a0829fe4b015070f74f;hp=5a634b28c82a64cd3f25e7950f32812724d99e35;hb=be7efb4ce06996444c08914305a73833a7123eeb;hpb=d8278c699434fd7975609e121a1a75820595a601 diff --git a/lustre/ptlrpc/nodemap_handler.c b/lustre/ptlrpc/nodemap_handler.c index 5a634b2..59b1734 100644 --- a/lustre/ptlrpc/nodemap_handler.c +++ b/lustre/ptlrpc/nodemap_handler.c @@ -44,29 +44,15 @@ /* nodemap proc root proc directory under fs/lustre */ struct proc_dir_entry *proc_lustre_nodemap_root; -/* Highest numerical lu_nodemap.nm_id defined */ -static atomic_t nodemap_highest_id; - -/* Simple flag to determine if nodemaps are active */ +/* Copy of config active flag to avoid locking in mapping functions */ bool nodemap_active; -/** - * pointer to default nodemap kept to keep from - * lookup it up in the hash since it is needed - * more often - */ -static struct lu_nodemap *default_nodemap; - -/** - * Lock required to access the range tree. - */ -rwlock_t nm_range_tree_lock; - -/** - * Hash keyed on nodemap name containing all - * nodemaps +/* Lock protecting the active config, useful primarily when proc and + * nodemap_hash might be replaced when loading a new config + * Any time the active config is referenced, the lock should be held. */ -static struct cfs_hash *nodemap_hash; +DEFINE_MUTEX(active_config_lock); +struct nodemap_config *active_config; /** * Nodemap destructor @@ -75,44 +61,61 @@ static struct cfs_hash *nodemap_hash; */ static void nodemap_destroy(struct lu_nodemap *nodemap) { - struct lu_nid_range *range; - struct lu_nid_range *range_temp; + ENTRY; - write_lock(&nm_range_tree_lock); - list_for_each_entry_safe(range, range_temp, &nodemap->nm_ranges, - rn_list) { - range_delete(range); - } - write_unlock(&nm_range_tree_lock); + if (nodemap->nm_pde_data != NULL) + lprocfs_nodemap_remove(nodemap->nm_pde_data); + + mutex_lock(&active_config_lock); + down_read(&active_config->nmc_range_tree_lock); + nm_member_reclassify_nodemap(nodemap); + up_read(&active_config->nmc_range_tree_lock); write_lock(&nodemap->nm_idmap_lock); idmap_delete_tree(nodemap); write_unlock(&nodemap->nm_idmap_lock); - nm_member_reclassify_nodemap(nodemap); - if (!cfs_hash_is_empty(nodemap->nm_member_hash)) + + mutex_unlock(&active_config_lock); + + if (!list_empty(&nodemap->nm_member_list)) CWARN("nodemap_destroy failed to reclassify all members\n"); - nm_member_delete_hash(nodemap); + nm_member_delete_list(nodemap); OBD_FREE_PTR(nodemap); + + EXIT; } /** * Functions used for the cfs_hash */ -static void nodemap_getref(struct lu_nodemap *nodemap) +void nodemap_getref(struct lu_nodemap *nodemap) { atomic_inc(&nodemap->nm_refcount); + CDEBUG(D_INFO, "GETting nodemap %s(p=%p) : new refcount %d\n", + nodemap->nm_name, nodemap, atomic_read(&nodemap->nm_refcount)); } +/** + * Destroy nodemap if last reference is put. Should be called outside + * active_config_lock + */ void nodemap_putref(struct lu_nodemap *nodemap) { - LASSERT(nodemap != NULL); + if (!nodemap) + return; + LASSERT(atomic_read(&nodemap->nm_refcount) > 0); + CDEBUG(D_INFO, "PUTting nodemap %s(p=%p) : new refcount %d\n", + nodemap->nm_name, nodemap, + atomic_read(&nodemap->nm_refcount) - 1); + if (atomic_dec_and_test(&nodemap->nm_refcount)) nodemap_destroy(nodemap); } +EXPORT_SYMBOL(nodemap_putref); static __u32 nodemap_hashfn(struct cfs_hash *hash_body, const void *key, unsigned mask) @@ -173,50 +176,23 @@ static struct cfs_hash_ops nodemap_hash_operations = { /* end of cfs_hash functions */ /** - * Helper iterator to clean up nodemap on module exit. - * - * \param hs hash structure - * \param bd bucket descriptor - * \param hnode hash node - * \param data not used here - */ -static int nodemap_cleanup_iter_cb(struct cfs_hash *hs, struct cfs_hash_bd *bd, - struct hlist_node *hnode, void *data) -{ - struct lu_nodemap *nodemap; - - nodemap = hlist_entry(hnode, struct lu_nodemap, nm_hash); - nodemap_putref(nodemap); - - return 0; -} - -/** - * Walk the nodemap_hash and remove all nodemaps. - */ -void nodemap_cleanup_all(void) -{ - cfs_hash_for_each_safe(nodemap_hash, nodemap_cleanup_iter_cb, NULL); - cfs_hash_putref(nodemap_hash); -} - -/** * Initialize nodemap_hash * * \retval 0 success * \retval -ENOMEM cannot create hash */ -static int nodemap_init_hash(void) +static int nodemap_init_hash(struct nodemap_config *nmc) { - nodemap_hash = cfs_hash_create("NODEMAP", HASH_NODEMAP_CUR_BITS, - HASH_NODEMAP_MAX_BITS, - HASH_NODEMAP_BKT_BITS, 0, - CFS_HASH_MIN_THETA, - CFS_HASH_MAX_THETA, - &nodemap_hash_operations, - CFS_HASH_DEFAULT); - - if (nodemap_hash == NULL) { + nmc->nmc_nodemap_hash = cfs_hash_create("NODEMAP", + HASH_NODEMAP_CUR_BITS, + HASH_NODEMAP_MAX_BITS, + HASH_NODEMAP_BKT_BITS, 0, + CFS_HASH_MIN_THETA, + CFS_HASH_MAX_THETA, + &nodemap_hash_operations, + CFS_HASH_DEFAULT); + + if (nmc->nmc_nodemap_hash == NULL) { CERROR("cannot create nodemap_hash table\n"); return -ENOMEM; } @@ -248,48 +224,71 @@ static bool nodemap_name_is_valid(const char *name) /** * Nodemap lookup * - * Look nodemap up in the nodemap hash + * Look nodemap up in the active_config nodemap hash. Caller should hold the + * active_config_lock. * * \param name name of nodemap - * \param nodemap found nodemap or NULL - * \retval lu_nodemap named nodemap - * \retval NULL nodemap doesn't exist + * \retval nodemap pointer set to found nodemap + * \retval -EINVAL name is not valid + * \retval -ENOENT nodemap not found */ -static int nodemap_lookup(const char *name, struct lu_nodemap **nodemap) +struct lu_nodemap *nodemap_lookup(const char *name) { - int rc = 0; - - *nodemap = NULL; + struct lu_nodemap *nodemap = NULL; if (!nodemap_name_is_valid(name)) - GOTO(out, rc = -EINVAL); + return ERR_PTR(-EINVAL); - *nodemap = cfs_hash_lookup(nodemap_hash, name); - if (*nodemap == NULL) - rc = -ENOENT; + nodemap = cfs_hash_lookup(active_config->nmc_nodemap_hash, name); + if (nodemap == NULL) + return ERR_PTR(-ENOENT); -out: - return rc; + return nodemap; } /** - * classify the nid into the proper nodemap + * Classify the nid into the proper nodemap. Caller must hold active config and + * nm_range_tree_lock, and call nodemap_putref when done with nodemap. * * \param nid nid to classify * \retval nodemap nodemap containing the nid * \retval default_nodemap default nodemap + * \retval -EINVAL LO nid given without other local nid */ struct lu_nodemap *nodemap_classify_nid(lnet_nid_t nid) { struct lu_nid_range *range; + struct lu_nodemap *nodemap; + int rc; + + ENTRY; + + /* don't use 0@lo, use the first non-lo local NID instead */ + if (LNET_NETTYP(LNET_NIDNET(nid)) == LOLND) { + lnet_process_id_t id; + int i = 0; + + do { + rc = LNetGetId(i++, &id); + if (rc < 0) + RETURN(ERR_PTR(-EINVAL)); + } while (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND); + + nid = id.nid; + CDEBUG(D_INFO, "found nid %s\n", libcfs_nid2str(nid)); + } - range = range_search(nid); + range = range_search(&active_config->nmc_range_tree, nid); if (range != NULL) - return range->rn_nodemap; + nodemap = range->rn_nodemap; + else + nodemap = active_config->nmc_default_nodemap; + + LASSERT(nodemap != NULL); + nodemap_getref(nodemap); - return default_nodemap; + RETURN(nodemap); } -EXPORT_SYMBOL(nodemap_classify_nid); /** * simple check for default nodemap @@ -376,19 +375,35 @@ EXPORT_SYMBOL(nodemap_parse_idmap); * \param nid nid to add to the members * \param exp obd_export structure for the connection * that is being added - * \retval -EINVAL export is NULL + * \retval -EINVAL export is NULL, or has invalid NID * \retval -EEXIST export is already member of a nodemap */ int nodemap_add_member(lnet_nid_t nid, struct obd_export *exp) { - struct lu_nodemap *nodemap; - int rc; + struct lu_nodemap *nodemap; + int rc = 0; + ENTRY; + + mutex_lock(&active_config_lock); + down_read(&active_config->nmc_range_tree_lock); - read_lock(&nm_range_tree_lock); nodemap = nodemap_classify_nid(nid); - rc = nm_member_add(nodemap, exp); - read_unlock(&nm_range_tree_lock); - return rc; + + if (IS_ERR(nodemap)) { + CWARN("%s: error adding to nodemap, no valid NIDs found\n", + exp->exp_obd->obd_name); + rc = -EINVAL; + } else { + rc = nm_member_add(nodemap, exp); + } + + up_read(&active_config->nmc_range_tree_lock); + mutex_unlock(&active_config_lock); + + if (!IS_ERR(nodemap)) + nodemap_putref(nodemap); + + RETURN(rc); } EXPORT_SYMBOL(nodemap_add_member); @@ -399,10 +414,34 @@ EXPORT_SYMBOL(nodemap_add_member); */ void nodemap_del_member(struct obd_export *exp) { - struct lu_nodemap *nodemap = exp->exp_target_data.ted_nodemap; + struct lu_nodemap *nodemap; + + ENTRY; + + /* using ac lock to prevent nodemap reclassification while deleting */ + mutex_lock(&active_config_lock); + + /* use of ted_nodemap is protected by active_config_lock. we take an + * extra reference to make sure nodemap isn't destroyed under + * active_config_lock + */ + nodemap = exp->exp_target_data.ted_nodemap; + if (nodemap == NULL) + goto out; + else + nodemap_getref(nodemap); + + mutex_lock(&nodemap->nm_member_list_lock); + nm_member_del(nodemap, exp); + mutex_unlock(&nodemap->nm_member_list_lock); - if (nodemap != NULL) - nm_member_del(nodemap, exp); +out: + mutex_unlock(&active_config_lock); + + if (nodemap) + nodemap_putref(nodemap); + + EXIT; } EXPORT_SYMBOL(nodemap_del_member); @@ -417,28 +456,49 @@ EXPORT_SYMBOL(nodemap_del_member); * * \retval 0 on success */ -int nodemap_add_idmap(const char *name, enum nodemap_id_type id_type, - const __u32 map[2]) +int nodemap_add_idmap_helper(struct lu_nodemap *nodemap, + enum nodemap_id_type id_type, + const __u32 map[2]) { - struct lu_nodemap *nodemap = NULL; struct lu_idmap *idmap; int rc = 0; - rc = nodemap_lookup(name, &nodemap); - if (nodemap == NULL || is_default_nodemap(nodemap)) - GOTO(out, rc = -EINVAL); - idmap = idmap_create(map[0], map[1]); if (idmap == NULL) - GOTO(out_putref, rc = -ENOMEM); + GOTO(out, rc = -ENOMEM); write_lock(&nodemap->nm_idmap_lock); idmap_insert(id_type, idmap, nodemap); write_unlock(&nodemap->nm_idmap_lock); nm_member_revoke_locks(nodemap); -out_putref: +out: + return rc; +} + +int nodemap_add_idmap(const char *name, enum nodemap_id_type id_type, + const __u32 map[2]) +{ + struct lu_nodemap *nodemap = NULL; + int rc; + + mutex_lock(&active_config_lock); + nodemap = nodemap_lookup(name); + if (IS_ERR(nodemap)) { + mutex_unlock(&active_config_lock); + GOTO(out, rc = PTR_ERR(nodemap)); + } + + if (is_default_nodemap(nodemap)) { + rc = -EINVAL; + } else { + rc = nodemap_add_idmap_helper(nodemap, id_type, map); + if (rc == 0) + rc = nodemap_idx_idmap_add(nodemap, id_type, map); + } + mutex_unlock(&active_config_lock); nodemap_putref(nodemap); + out: return rc; } @@ -462,30 +522,83 @@ int nodemap_del_idmap(const char *name, enum nodemap_id_type id_type, struct lu_idmap *idmap = NULL; int rc = 0; - rc = nodemap_lookup(name, &nodemap); - if (nodemap == NULL || is_default_nodemap(nodemap)) - GOTO(out, rc = -EINVAL); + mutex_lock(&active_config_lock); + nodemap = nodemap_lookup(name); + if (IS_ERR(nodemap)) { + mutex_unlock(&active_config_lock); + GOTO(out, rc = PTR_ERR(nodemap)); + } + + if (is_default_nodemap(nodemap)) + GOTO(out_putref, rc = -EINVAL); write_lock(&nodemap->nm_idmap_lock); idmap = idmap_search(nodemap, NODEMAP_CLIENT_TO_FS, id_type, map[0]); if (idmap == NULL) { - write_unlock(&nodemap->nm_idmap_lock); - GOTO(out_putref, rc = -EINVAL); + rc = -EINVAL; + } else { + idmap_delete(id_type, idmap, nodemap); + rc = nodemap_idx_idmap_del(nodemap, id_type, map); } - - idmap_delete(id_type, idmap, nodemap); write_unlock(&nodemap->nm_idmap_lock); - nm_member_revoke_locks(nodemap); out_putref: + mutex_unlock(&active_config_lock); + if (rc == 0) + nm_member_revoke_locks(nodemap); nodemap_putref(nodemap); + out: return rc; } EXPORT_SYMBOL(nodemap_del_idmap); /** + * Get nodemap assigned to given export. Takes a reference on the nodemap. + * + * Note that this function may return either NULL, or an ERR_PTR() + * or a valid nodemap pointer. All of the functions accessing the + * returned nodemap can check IS_ERR(nodemap) to see if an error is + * returned. NULL is not considered an error, which is OK since this + * is a valid case if nodemap are not in use. All nodemap handling + * functions must check for nodemap == NULL and do nothing, and the + * nodemap returned from this function should not be dereferenced. + * + * \param export export to get nodemap for + * + * \retval pointer to nodemap on success + * \retval NULL nodemap subsystem disabled + * \retval -EACCES export does not have nodemap assigned + */ +struct lu_nodemap *nodemap_get_from_exp(struct obd_export *exp) +{ + struct lu_nodemap *nodemap; + + ENTRY; + + if (!nodemap_active) + RETURN(NULL); + + spin_lock(&exp->exp_target_data.ted_nodemap_lock); + nodemap = exp->exp_target_data.ted_nodemap; + if (nodemap) + nodemap_getref(nodemap); + spin_unlock(&exp->exp_target_data.ted_nodemap_lock); + + if (!nodemap) { + CDEBUG(D_INFO, "%s: nodemap null on export %s (at %s)\n", + exp->exp_obd->obd_name, + obd_uuid2str(&exp->exp_client_uuid), + obd_export_nid2str(exp)); + RETURN(ERR_PTR(-EACCES)); + } + + RETURN(nodemap); +} +EXPORT_SYMBOL(nodemap_get_from_exp); + +/** * mapping function for nodemap idmaps * * \param nodemap lu_nodemap structure defining nodemap @@ -517,6 +630,8 @@ __u32 nodemap_map_id(struct lu_nodemap *nodemap, struct lu_idmap *idmap = NULL; __u32 found_id; + ENTRY; + if (!nodemap_active) goto out; @@ -548,15 +663,15 @@ __u32 nodemap_map_id(struct lu_nodemap *nodemap, else found_id = idmap->id_fs; read_unlock(&nodemap->nm_idmap_lock); - return found_id; + RETURN(found_id); squash: if (id_type == NODEMAP_UID) - return nodemap->nm_squash_uid; + RETURN(nodemap->nm_squash_uid); else - return nodemap->nm_squash_gid; + RETURN(nodemap->nm_squash_gid); out: - return id; + RETURN(id); } EXPORT_SYMBOL(nodemap_map_id); @@ -625,47 +740,79 @@ ssize_t nodemap_map_acl(struct lu_nodemap *nodemap, void *buf, size_t size, EXPORT_SYMBOL(nodemap_map_acl); /* - * add nid range to nodemap - * \param name nodemap name - * \param range_st string containing nid range - * \retval 0 on success + * Add nid range to given nodemap + * + * \param config nodemap config to work on + * \param nodemap nodemap to add range to + * \param nid nid range to add + * \param range_id should be 0 unless loading from disk + * \retval 0 success + * \retval -ENOMEM * - * add an range to the global range tree and attached the - * range to the named nodemap. */ -int nodemap_add_range(const char *name, const lnet_nid_t nid[2]) +int nodemap_add_range_helper(struct nodemap_config *config, + struct lu_nodemap *nodemap, + const lnet_nid_t nid[2], + unsigned int range_id) { - struct lu_nodemap *nodemap = NULL; struct lu_nid_range *range; int rc; - rc = nodemap_lookup(name, &nodemap); - if (nodemap == NULL || is_default_nodemap(nodemap)) - GOTO(out, rc = -EINVAL); - - range = range_create(nid[0], nid[1], nodemap); - if (range == NULL) - GOTO(out_putref, rc = -ENOMEM); + down_write(&config->nmc_range_tree_lock); + range = range_create(&config->nmc_range_tree, nid[0], nid[1], + nodemap, range_id); + if (range == NULL) { + up_write(&config->nmc_range_tree_lock); + GOTO(out, rc = -ENOMEM); + } - write_lock(&nm_range_tree_lock); - rc = range_insert(range); + rc = range_insert(&config->nmc_range_tree, range); if (rc != 0) { CERROR("cannot insert nodemap range into '%s': rc = %d\n", nodemap->nm_name, rc); - write_unlock(&nm_range_tree_lock); + up_write(&config->nmc_range_tree_lock); list_del(&range->rn_list); range_destroy(range); - GOTO(out_putref, rc = -ENOMEM); + GOTO(out, rc = -ENOMEM); } list_add(&range->rn_list, &nodemap->nm_ranges); - write_unlock(&nm_range_tree_lock); - nm_member_reclassify_nodemap(default_nodemap); - nm_member_revoke_locks(default_nodemap); - nm_member_revoke_locks(nodemap); + /* nodemaps have no members if they aren't on the active config */ + if (config == active_config) + nm_member_reclassify_nodemap(config->nmc_default_nodemap); -out_putref: + up_write(&config->nmc_range_tree_lock); + + /* if range_id is non-zero, we are loading from disk */ + if (range_id == 0) + rc = nodemap_idx_range_add(range, nid); + + if (config == active_config) { + nm_member_revoke_locks(config->nmc_default_nodemap); + nm_member_revoke_locks(nodemap); + } + +out: + return rc; +} +int nodemap_add_range(const char *name, const lnet_nid_t nid[2]) +{ + struct lu_nodemap *nodemap = NULL; + int rc; + + mutex_lock(&active_config_lock); + nodemap = nodemap_lookup(name); + if (IS_ERR(nodemap)) { + mutex_unlock(&active_config_lock); + GOTO(out, rc = PTR_ERR(nodemap)); + } + + if (is_default_nodemap(nodemap)) + rc = -EINVAL; + else + rc = nodemap_add_range_helper(active_config, nodemap, nid, 0); + mutex_unlock(&active_config_lock); nodemap_putref(nodemap); out: return rc; @@ -687,24 +834,32 @@ int nodemap_del_range(const char *name, const lnet_nid_t nid[2]) struct lu_nid_range *range; int rc = 0; - rc = nodemap_lookup(name, &nodemap); - if (nodemap == NULL || is_default_nodemap(nodemap)) - GOTO(out, rc = -EINVAL); + mutex_lock(&active_config_lock); + nodemap = nodemap_lookup(name); + if (IS_ERR(nodemap)) { + mutex_unlock(&active_config_lock); + GOTO(out, rc = PTR_ERR(nodemap)); + } + + if (is_default_nodemap(nodemap)) + GOTO(out_putref, rc = -EINVAL); - write_lock(&nm_range_tree_lock); - range = range_find(nid[0], nid[1]); + down_write(&active_config->nmc_range_tree_lock); + range = range_find(&active_config->nmc_range_tree, nid[0], nid[1]); if (range == NULL) { - write_unlock(&nm_range_tree_lock); + up_write(&active_config->nmc_range_tree_lock); GOTO(out_putref, rc = -EINVAL); } - - range_delete(range); - write_unlock(&nm_range_tree_lock); + rc = nodemap_idx_range_del(range); + range_delete(&active_config->nmc_range_tree, range); nm_member_reclassify_nodemap(nodemap); - nm_member_revoke_locks(default_nodemap); + up_write(&active_config->nmc_range_tree_lock); + + nm_member_revoke_locks(active_config->nmc_default_nodemap); nm_member_revoke_locks(nodemap); out_putref: + mutex_unlock(&active_config_lock); nodemap_putref(nodemap); out: return rc; @@ -712,6 +867,72 @@ out: EXPORT_SYMBOL(nodemap_del_range); /** + * set fileset on nodemap + * \param name nodemap to set fileset on + * \param fileset string containing fileset + * \retval 0 on success + * + * set a fileset on the named nodemap + */ +static int nodemap_set_fileset_helper(struct nodemap_config *config, + struct lu_nodemap *nodemap, + const char *fileset) +{ + int rc = 0; + + /* we allow fileset = "" which means clear fileset info */ + if (fileset == NULL || (fileset[0] != 0 && fileset[0] != '/')) + rc = -EINVAL; + else if (strlcpy(nodemap->nm_fileset, fileset, + sizeof(nodemap->nm_fileset)) >= + sizeof(nodemap->nm_fileset)) + rc = -ENAMETOOLONG; + + return rc; +} + +int nodemap_set_fileset(const char *name, const char *fileset) +{ + struct lu_nodemap *nodemap = NULL; + int rc = 0; + + mutex_lock(&active_config_lock); + nodemap = nodemap_lookup(name); + if (IS_ERR(nodemap)) { + mutex_unlock(&active_config_lock); + GOTO(out, rc = PTR_ERR(nodemap)); + } + + if (is_default_nodemap(nodemap)) + rc = -EINVAL; + else + rc = nodemap_set_fileset_helper(active_config, nodemap, + fileset); + mutex_unlock(&active_config_lock); + + nodemap_putref(nodemap); +out: + return rc; +} +EXPORT_SYMBOL(nodemap_set_fileset); + +/** + * get fileset defined on nodemap + * \param nodemap nodemap to get fileset from + * \retval fileset name, or NULL if not defined or not activated + * + * get the fileset defined on the nodemap + */ +char *nodemap_get_fileset(const struct lu_nodemap *nodemap) +{ + if (!nodemap_active || is_default_nodemap(nodemap)) + return NULL; + else + return (char *)nodemap->nm_fileset; +} +EXPORT_SYMBOL(nodemap_get_fileset); + +/** * Nodemap constructor * * Creates an lu_nodemap structure and assigns sane default @@ -720,21 +941,34 @@ EXPORT_SYMBOL(nodemap_del_range); * the default flags should be inherited from the default nodemap. * The adds nodemap to nodemap_hash. * + * Requires that the caller take the active_config_lock + * * \param name name of nodemap * \param is_default true if default nodemap - * \retval 0 success + * \retval nodemap success * \retval -EINVAL invalid nodemap name * \retval -EEXIST nodemap already exists * \retval -ENOMEM cannot allocate memory for nodemap */ -static int nodemap_create(const char *name, bool is_default) +struct lu_nodemap *nodemap_create(const char *name, + struct nodemap_config *config, + bool is_default) { struct lu_nodemap *nodemap = NULL; - int rc = 0; + struct lu_nodemap *default_nodemap; + struct cfs_hash *hash = config->nmc_nodemap_hash; + int rc = 0; + + default_nodemap = config->nmc_default_nodemap; if (!nodemap_name_is_valid(name)) GOTO(out, rc = -EINVAL); + if (hash == NULL) { + CERROR("Config nodemap hash is NULL, unable to add %s\n", name); + GOTO(out, rc = -EINVAL); + } + OBD_ALLOC_PTR(nodemap); if (nodemap == NULL) { CERROR("cannot allocate memory (%zu bytes)" @@ -745,25 +979,21 @@ static int nodemap_create(const char *name, bool is_default) /* * take an extra reference to prevent nodemap from being destroyed - * while its being created. + * while it's being created. */ atomic_set(&nodemap->nm_refcount, 2); snprintf(nodemap->nm_name, sizeof(nodemap->nm_name), "%s", name); - rc = cfs_hash_add_unique(nodemap_hash, name, &nodemap->nm_hash); + rc = cfs_hash_add_unique(hash, name, &nodemap->nm_hash); if (rc != 0) { OBD_FREE_PTR(nodemap); GOTO(out, rc = -EEXIST); } - - rc = nm_member_init_hash(nodemap); - if (rc != 0) { - OBD_FREE_PTR(nodemap); - goto out; - } - INIT_LIST_HEAD(&nodemap->nm_ranges); + INIT_LIST_HEAD(&nodemap->nm_list); + INIT_LIST_HEAD(&nodemap->nm_member_list); + mutex_init(&nodemap->nm_member_list_lock); rwlock_init(&nodemap->nm_idmap_lock); nodemap->nm_fs_to_client_uidmap = RB_ROOT; nodemap->nm_client_to_fs_uidmap = RB_ROOT; @@ -772,61 +1002,91 @@ static int nodemap_create(const char *name, bool is_default) if (is_default) { nodemap->nm_id = LUSTRE_NODEMAP_DEFAULT_ID; + config->nmc_default_nodemap = nodemap; + } else { + config->nmc_nodemap_highest_id++; + nodemap->nm_id = config->nmc_nodemap_highest_id; + } + + if (is_default || default_nodemap == NULL) { nodemap->nmf_trust_client_ids = 0; nodemap->nmf_allow_root_access = 0; - nodemap->nmf_block_lookups = 0; + nodemap->nmf_deny_unknown = 0; nodemap->nm_squash_uid = NODEMAP_NOBODY_UID; nodemap->nm_squash_gid = NODEMAP_NOBODY_GID; - - lprocfs_nodemap_register(name, is_default, nodemap); - - default_nodemap = nodemap; + if (!is_default) + CWARN("adding nodemap '%s' to config without" + " default nodemap\n", nodemap->nm_name); } else { - nodemap->nm_id = atomic_inc_return(&nodemap_highest_id); nodemap->nmf_trust_client_ids = default_nodemap->nmf_trust_client_ids; nodemap->nmf_allow_root_access = default_nodemap->nmf_allow_root_access; - nodemap->nmf_block_lookups = - default_nodemap->nmf_block_lookups; + nodemap->nmf_deny_unknown = + default_nodemap->nmf_deny_unknown; nodemap->nm_squash_uid = default_nodemap->nm_squash_uid; nodemap->nm_squash_gid = default_nodemap->nm_squash_gid; - - lprocfs_nodemap_register(name, is_default, nodemap); + nodemap->nm_fileset[0] = 0; } - if (rc == 0) { - nodemap_putref(nodemap); - goto out; - } + return nodemap; +out: CERROR("cannot add nodemap: '%s': rc = %d\n", name, rc); - nodemap_destroy(nodemap); + return ERR_PTR(rc); +} + +/** + * Set the nmf_deny_unknown flag to true or false. + * \param name nodemap name + * \param deny_unknown if true, squashed users will get EACCES + * \retval 0 on success + * + */ +int nodemap_set_deny_unknown(const char *name, bool deny_unknown) +{ + struct lu_nodemap *nodemap = NULL; + int rc = 0; + + mutex_lock(&active_config_lock); + nodemap = nodemap_lookup(name); + mutex_unlock(&active_config_lock); + if (IS_ERR(nodemap)) + GOTO(out, rc = PTR_ERR(nodemap)); + + nodemap->nmf_deny_unknown = deny_unknown; + rc = nodemap_idx_nodemap_update(nodemap); + nm_member_revoke_locks(nodemap); + nodemap_putref(nodemap); out: return rc; } +EXPORT_SYMBOL(nodemap_set_deny_unknown); /** - * update flag to turn on or off nodemap functions + * Set the nmf_allow_root_access flag to true or false. * \param name nodemap name - * \param admin_string string containing updated value + * \param allow_root if true, nodemap will not squash the root user * \retval 0 on success * - * Update admin flag to turn on or off nodemap functions. */ int nodemap_set_allow_root(const char *name, bool allow_root) { struct lu_nodemap *nodemap = NULL; int rc = 0; - rc = nodemap_lookup(name, &nodemap); - if (nodemap == NULL) - GOTO(out, rc = -ENOENT); + mutex_lock(&active_config_lock); + nodemap = nodemap_lookup(name); + mutex_unlock(&active_config_lock); + if (IS_ERR(nodemap)) + GOTO(out, rc = PTR_ERR(nodemap)); nodemap->nmf_allow_root_access = allow_root; + rc = nodemap_idx_nodemap_update(nodemap); + nm_member_revoke_locks(nodemap); nodemap_putref(nodemap); out: @@ -835,24 +1095,27 @@ out: EXPORT_SYMBOL(nodemap_set_allow_root); /** - * updated trust_client_ids flag for nodemap + * Set the nmf_trust_client_ids flag to true or false. * - * \param name nodemap name - * \param trust_string new value for trust flag + * \param name nodemap name + * \param trust_client_ids if true, nodemap will not map its IDs * \retval 0 on success * - * Update the trust_client_ids flag for a nodemap. */ int nodemap_set_trust_client_ids(const char *name, bool trust_client_ids) { struct lu_nodemap *nodemap = NULL; int rc = 0; - rc = nodemap_lookup(name, &nodemap); - if (nodemap == NULL) - GOTO(out, rc = -ENOENT); + mutex_lock(&active_config_lock); + nodemap = nodemap_lookup(name); + mutex_unlock(&active_config_lock); + if (IS_ERR(nodemap)) + GOTO(out, rc = PTR_ERR(nodemap)); nodemap->nmf_trust_client_ids = trust_client_ids; + rc = nodemap_idx_nodemap_update(nodemap); + nm_member_revoke_locks(nodemap); nodemap_putref(nodemap); out: @@ -861,10 +1124,10 @@ out: EXPORT_SYMBOL(nodemap_set_trust_client_ids); /** - * update the squash_uid for a nodemap + * Update the squash_uid for a nodemap. * * \param name nodemap name - * \param uid_string string containing new squash_uid value + * \param uid the new uid to squash unknown users to * \retval 0 on success * * Update the squash_uid for a nodemap. The squash_uid is the uid @@ -877,11 +1140,15 @@ int nodemap_set_squash_uid(const char *name, uid_t uid) struct lu_nodemap *nodemap = NULL; int rc = 0; - rc = nodemap_lookup(name, &nodemap); - if (nodemap == NULL) - GOTO(out, rc = -ENOENT); + mutex_lock(&active_config_lock); + nodemap = nodemap_lookup(name); + mutex_unlock(&active_config_lock); + if (IS_ERR(nodemap)) + GOTO(out, rc = PTR_ERR(nodemap)); nodemap->nm_squash_uid = uid; + rc = nodemap_idx_nodemap_update(nodemap); + nm_member_revoke_locks(nodemap); nodemap_putref(nodemap); out: @@ -893,7 +1160,7 @@ EXPORT_SYMBOL(nodemap_set_squash_uid); * Update the squash_gid for a nodemap. * * \param name nodemap name - * \param gid_string string containing new squash_gid value + * \param gid the new gid to squash unknown gids to * \retval 0 on success * * Update the squash_gid for a nodemap. The squash_uid is the gid @@ -906,11 +1173,15 @@ int nodemap_set_squash_gid(const char *name, gid_t gid) struct lu_nodemap *nodemap = NULL; int rc = 0; - rc = nodemap_lookup(name, &nodemap); - if (nodemap == NULL) - GOTO(out, rc = -ENOENT); + mutex_lock(&active_config_lock); + nodemap = nodemap_lookup(name); + mutex_unlock(&active_config_lock); + if (IS_ERR(nodemap)) + GOTO(out, rc = PTR_ERR(nodemap)); nodemap->nm_squash_gid = gid; + rc = nodemap_idx_nodemap_update(nodemap); + nm_member_revoke_locks(nodemap); nodemap_putref(nodemap); out: @@ -926,7 +1197,7 @@ EXPORT_SYMBOL(nodemap_set_squash_gid); */ bool nodemap_can_setquota(const struct lu_nodemap *nodemap) { - return !nodemap_active || nodemap->nmf_allow_root_access; + return !nodemap_active || (nodemap && nodemap->nmf_allow_root_access); } EXPORT_SYMBOL(nodemap_can_setquota); @@ -941,7 +1212,24 @@ EXPORT_SYMBOL(nodemap_can_setquota); */ int nodemap_add(const char *nodemap_name) { - return nodemap_create(nodemap_name, 0); + struct lu_nodemap *nodemap; + int rc; + + mutex_lock(&active_config_lock); + nodemap = nodemap_create(nodemap_name, active_config, 0); + if (IS_ERR(nodemap)) { + mutex_unlock(&active_config_lock); + return PTR_ERR(nodemap); + } + + rc = nodemap_idx_nodemap_add(nodemap); + if (rc == 0) + rc = lprocfs_nodemap_register(nodemap, 0); + + mutex_unlock(&active_config_lock); + nodemap_putref(nodemap); + + return rc; } EXPORT_SYMBOL(nodemap_add); @@ -955,22 +1243,58 @@ EXPORT_SYMBOL(nodemap_add); */ int nodemap_del(const char *nodemap_name) { - struct lu_nodemap *nodemap; - int rc = 0; + struct lu_nodemap *nodemap; + struct lu_nid_range *range; + struct lu_nid_range *range_temp; + int rc = 0; + int rc2 = 0; if (strcmp(nodemap_name, DEFAULT_NODEMAP) == 0) - GOTO(out, rc = -EINVAL); + RETURN(-EINVAL); - nodemap = cfs_hash_del_key(nodemap_hash, nodemap_name); - if (nodemap == NULL) + mutex_lock(&active_config_lock); + nodemap = cfs_hash_del_key(active_config->nmc_nodemap_hash, + nodemap_name); + if (nodemap == NULL) { + mutex_unlock(&active_config_lock); GOTO(out, rc = -ENOENT); + } + + /* erase nodemap from active ranges to prevent client assignment */ + down_write(&active_config->nmc_range_tree_lock); + list_for_each_entry_safe(range, range_temp, &nodemap->nm_ranges, + rn_list) { + rc2 = nodemap_idx_range_del(range); + if (rc2 < 0) + rc = rc2; + + range_delete(&active_config->nmc_range_tree, range); + } + up_write(&active_config->nmc_range_tree_lock); + + rc2 = nodemap_idx_nodemap_del(nodemap); + if (rc2 < 0) + rc = rc2; /* * remove procfs here in case nodemap_create called with same name - * before nodemap_destory is run. + * before nodemap_destroy is run. */ - lprocfs_remove(&nodemap->nm_proc_entry); + lprocfs_nodemap_remove(nodemap->nm_pde_data); + nodemap->nm_pde_data = NULL; + + /* reclassify all member exports from nodemap, so they put their refs */ + down_read(&active_config->nmc_range_tree_lock); + nm_member_reclassify_nodemap(nodemap); + up_read(&active_config->nmc_range_tree_lock); + + if (!list_empty(&nodemap->nm_member_list)) + CWARN("nodemap_del failed to reclassify all members\n"); + + mutex_unlock(&active_config_lock); + nodemap_putref(nodemap); + out: return rc; } @@ -983,18 +1307,182 @@ EXPORT_SYMBOL(nodemap_del); */ void nodemap_activate(const bool value) { + mutex_lock(&active_config_lock); + active_config->nmc_nodemap_is_active = value; + + /* copy active value to global to avoid locking in map functions */ nodemap_active = value; + nodemap_idx_nodemap_activate(value); + mutex_unlock(&active_config_lock); nm_member_revoke_all(); } EXPORT_SYMBOL(nodemap_activate); /** + * Helper iterator to convert nodemap hash to list. + * + * \param hs hash structure + * \param bd bucket descriptor + * \param hnode hash node + * \param nodemap_list_head list head for list of nodemaps in hash + */ +static int nodemap_cleanup_iter_cb(struct cfs_hash *hs, struct cfs_hash_bd *bd, + struct hlist_node *hnode, + void *nodemap_list_head) +{ + struct lu_nodemap *nodemap; + + nodemap = hlist_entry(hnode, struct lu_nodemap, nm_hash); + list_add(&nodemap->nm_list, nodemap_list_head); + + cfs_hash_bd_del_locked(hs, bd, hnode); + + return 0; +} + +struct nodemap_config *nodemap_config_alloc(void) +{ + struct nodemap_config *config; + int rc = 0; + + OBD_ALLOC_PTR(config); + if (config == NULL) + return ERR_PTR(-ENOMEM); + + rc = nodemap_init_hash(config); + if (rc != 0) { + OBD_FREE_PTR(config); + return ERR_PTR(rc); + } + + init_rwsem(&config->nmc_range_tree_lock); + + return config; +} +EXPORT_SYMBOL(nodemap_config_alloc); + +/** + * Walk the nodemap_hash and remove all nodemaps. + */ +void nodemap_config_dealloc(struct nodemap_config *config) +{ + struct lu_nodemap *nodemap = NULL; + struct lu_nodemap *nodemap_temp; + struct lu_nid_range *range; + struct lu_nid_range *range_temp; + LIST_HEAD(nodemap_list_head); + + cfs_hash_for_each_safe(config->nmc_nodemap_hash, + nodemap_cleanup_iter_cb, &nodemap_list_head); + cfs_hash_putref(config->nmc_nodemap_hash); + + /* Because nodemap_destroy might sleep, we can't destroy them + * in cfs_hash_for_each, so we build a list there and destroy here + */ + list_for_each_entry_safe(nodemap, nodemap_temp, &nodemap_list_head, + nm_list) { + mutex_lock(&active_config_lock); + down_write(&config->nmc_range_tree_lock); + + /* move members to new config, requires ac lock */ + nm_member_reclassify_nodemap(nodemap); + list_for_each_entry_safe(range, range_temp, &nodemap->nm_ranges, + rn_list) + range_delete(&config->nmc_range_tree, range); + up_write(&config->nmc_range_tree_lock); + mutex_unlock(&active_config_lock); + + /* putref must be outside of ac lock if nm could be destroyed */ + nodemap_putref(nodemap); + } + OBD_FREE_PTR(config); +} +EXPORT_SYMBOL(nodemap_config_dealloc); + +/* + * callback for cfs_hash_for_each_safe used to convert a nodemap hash to a + * nodemap list, generally for locking purposes as a hash cb can't sleep. + */ +int nm_hash_list_cb(struct cfs_hash *hs, struct cfs_hash_bd *bd, + struct hlist_node *hnode, + void *nodemap_list_head) +{ + struct lu_nodemap *nodemap; + + nodemap = hlist_entry(hnode, struct lu_nodemap, nm_hash); + list_add(&nodemap->nm_list, nodemap_list_head); + return 0; +} + +void nodemap_config_set_active(struct nodemap_config *config) +{ + struct nodemap_config *old_config = active_config; + struct lu_nodemap *nodemap; + struct lu_nodemap *tmp; + bool revoke_locks; + LIST_HEAD(nodemap_list_head); + + ENTRY; + + LASSERT(active_config != config); + LASSERT(config->nmc_default_nodemap); + + mutex_lock(&active_config_lock); + + /* move proc entries from already existing nms, create for new nms */ + cfs_hash_for_each_safe(config->nmc_nodemap_hash, + nm_hash_list_cb, &nodemap_list_head); + list_for_each_entry_safe(nodemap, tmp, &nodemap_list_head, nm_list) { + struct lu_nodemap *old_nm = NULL; + + if (active_config != NULL) + old_nm = cfs_hash_lookup( + active_config->nmc_nodemap_hash, + nodemap->nm_name); + if (old_nm != NULL) { + nodemap->nm_pde_data = old_nm->nm_pde_data; + old_nm->nm_pde_data = NULL; + nodemap_putref(old_nm); + } else { + bool is_def = (nodemap == config->nmc_default_nodemap); + + lprocfs_nodemap_register(nodemap, is_def); + } + } + + /* + * We only need to revoke locks if old nodemap was active, and new + * config is now nodemap inactive. nodemap_config_dealloc will + * reclassify exports, triggering a lock revoke if and only if new + * nodemap is active. + */ + revoke_locks = !config->nmc_nodemap_is_active && nodemap_active; + + /* if new config is inactive, deactivate live config before switching */ + if (!config->nmc_nodemap_is_active) + nodemap_active = false; + active_config = config; + if (config->nmc_nodemap_is_active) + nodemap_active = true; + + mutex_unlock(&active_config_lock); + + if (old_config != NULL) + nodemap_config_dealloc(old_config); + + if (revoke_locks) + nm_member_revoke_all(); + + EXIT; +} + +/** * Cleanup nodemap module on exit */ void nodemap_mod_exit(void) { - nodemap_cleanup_all(); - lprocfs_remove(&proc_lustre_nodemap_root); + nodemap_config_dealloc(active_config); + nodemap_procfs_exit(); } /** @@ -1002,38 +1490,113 @@ void nodemap_mod_exit(void) */ int nodemap_mod_init(void) { - int rc = 0; + struct nodemap_config *new_config; + struct lu_nodemap *nodemap; + int rc = 0; - rc = nodemap_init_hash(); + rc = nodemap_procfs_init(); if (rc != 0) - goto cleanup; + return rc; - rwlock_init(&nm_range_tree_lock); - nodemap_procfs_init(); - rc = nodemap_create(DEFAULT_NODEMAP, 1); + new_config = nodemap_config_alloc(); + if (IS_ERR(new_config)) { + nodemap_procfs_exit(); + GOTO(out, rc = PTR_ERR(new_config)); + } -cleanup: - if (rc != 0) - nodemap_mod_exit(); + nodemap = nodemap_create(DEFAULT_NODEMAP, new_config, 1); + if (IS_ERR(nodemap)) { + nodemap_config_dealloc(new_config); + nodemap_procfs_exit(); + GOTO(out, rc = PTR_ERR(nodemap)); + } + + nodemap_config_set_active(new_config); + nodemap_putref(nodemap); +out: return rc; } -static int nm_member_revoke_all_cb(struct cfs_hash *hs, struct cfs_hash_bd *bd, - struct hlist_node *hnode, void *data) +/** + * Revoke locks for all nodemaps. + */ +void nm_member_revoke_all(void) { struct lu_nodemap *nodemap; + struct lu_nodemap *tmp; + LIST_HEAD(nodemap_list_head); - nodemap = hlist_entry(hnode, struct lu_nodemap, nm_hash); - nm_member_revoke_locks(nodemap); - return 0; + mutex_lock(&active_config_lock); + cfs_hash_for_each_safe(active_config->nmc_nodemap_hash, + nm_hash_list_cb, &nodemap_list_head); + + /* revoke_locks sleeps, so can't call in cfs hash cb */ + list_for_each_entry_safe(nodemap, tmp, &nodemap_list_head, nm_list) + nm_member_revoke_locks_always(nodemap); + mutex_unlock(&active_config_lock); } /** - * Revoke locks for all nodemaps. + * Returns the nodemap classification for a given nid into an ioctl buffer. + * Useful for testing the nodemap configuration to make sure it is working as + * expected. + * + * \param nid nid to classify + * \param[out] name_buf buffer to write the nodemap name to + * \param name_len length of buffer */ -void nm_member_revoke_all() +void nodemap_test_nid(lnet_nid_t nid, char *name_buf, size_t name_len) { - cfs_hash_for_each_safe(nodemap_hash, nm_member_revoke_all_cb, NULL); + struct lu_nodemap *nodemap; + + mutex_lock(&active_config_lock); + down_read(&active_config->nmc_range_tree_lock); + nodemap = nodemap_classify_nid(nid); + up_read(&active_config->nmc_range_tree_lock); + mutex_unlock(&active_config_lock); + + if (IS_ERR(nodemap)) + return; + + strncpy(name_buf, nodemap->nm_name, name_len); + if (name_len > 0) + name_buf[name_len - 1] = '\0'; + + nodemap_putref(nodemap); } +EXPORT_SYMBOL(nodemap_test_nid); +/** + * Passes back the id mapping for a given nid/id pair. Useful for testing the + * nodemap configuration to make sure it is working as expected. + * + * \param nid nid to classify + * \param idtype uid or gid + * \param client_id id to map to fs + * \param fs_id_buf pointer to save mapped fs_id to + * + * \retval 0 success + * \retval -EINVAL invalid NID + */ +int nodemap_test_id(lnet_nid_t nid, enum nodemap_id_type idtype, + __u32 client_id, __u32 *fs_id) +{ + struct lu_nodemap *nodemap; + + mutex_lock(&active_config_lock); + down_read(&active_config->nmc_range_tree_lock); + nodemap = nodemap_classify_nid(nid); + up_read(&active_config->nmc_range_tree_lock); + mutex_unlock(&active_config_lock); + + if (IS_ERR(nodemap)) + return PTR_ERR(nodemap); + + *fs_id = nodemap_map_id(nodemap, idtype, NODEMAP_CLIENT_TO_FS, + client_id); + nodemap_putref(nodemap); + + return 0; +} +EXPORT_SYMBOL(nodemap_test_id);