| /** |
| * @file tree_data_helpers.c |
| * @author Radek Krejci <rkrejci@cesnet.cz> |
| * @brief Parsing and validation helper functions for data trees |
| * |
| * Copyright (c) 2015 - 2018 CESNET, z.s.p.o. |
| * |
| * This source code is licensed under BSD 3-Clause License (the "License"). |
| * You may not use this file except in compliance with the License. |
| * You may obtain a copy of the License at |
| * |
| * https://opensource.org/licenses/BSD-3-Clause |
| */ |
| #define _XOPEN_SOURCE 500 /* strdup */ |
| #define _POSIX_C_SOURCE 200809L /*strndup */ |
| |
| #include <assert.h> |
| #include <stdint.h> |
| #include <stdlib.h> |
| #include <string.h> |
| |
| #include "common.h" |
| #include "compat.h" |
| #include "context.h" |
| #include "dict.h" |
| #include "hash_table.h" |
| #include "log.h" |
| #include "lyb.h" |
| #include "parser_data.h" |
| #include "set.h" |
| #include "tree.h" |
| #include "tree_data.h" |
| #include "tree_data_internal.h" |
| #include "tree_schema.h" |
| |
| struct lyd_node * |
| lys_getnext_data(const struct lyd_node *last, const struct lyd_node *sibling, const struct lysc_node **slast, |
| const struct lysc_node *parent, const struct lysc_module *module) |
| { |
| const struct lysc_node *siter = NULL; |
| struct lyd_node *match = NULL; |
| |
| assert(parent || module); |
| assert(!last || (slast && *slast)); |
| |
| if (slast) { |
| siter = *slast; |
| } |
| |
| if (last && last->next && (last->next->schema == siter)) { |
| /* return next data instance */ |
| return last->next; |
| } |
| |
| /* find next schema node data instance */ |
| while ((siter = lys_getnext(siter, parent, module, 0))) { |
| if (!lyd_find_sibling_val(sibling, siter, NULL, 0, &match)) { |
| break; |
| } |
| } |
| |
| if (slast) { |
| *slast = siter; |
| } |
| return match; |
| } |
| |
| struct lyd_node ** |
| lyd_node_children_p(struct lyd_node *node) |
| { |
| assert(node); |
| |
| if (!node->schema) { |
| return &((struct lyd_node_opaq *)node)->child; |
| } else { |
| switch (node->schema->nodetype) { |
| case LYS_CONTAINER: |
| case LYS_LIST: |
| case LYS_RPC: |
| case LYS_ACTION: |
| case LYS_NOTIF: |
| return &((struct lyd_node_inner *)node)->child; |
| default: |
| return NULL; |
| } |
| } |
| } |
| |
| API struct lyd_node * |
| lyd_parent(const struct lyd_node *node) |
| { |
| if (!node) { |
| return NULL; |
| } |
| |
| return (struct lyd_node *)(node)->parent; |
| } |
| |
| API struct lyd_node * |
| lyd_child(const struct lyd_node *node) |
| { |
| struct lyd_node **children; |
| |
| if (!node) { |
| return NULL; |
| } |
| |
| if (!node->schema) { |
| /* opaq node */ |
| return ((struct lyd_node_opaq *)(node))->child; |
| } |
| |
| children = lyd_node_children_p((struct lyd_node *)node); |
| if (children) { |
| return *children; |
| } else { |
| return NULL; |
| } |
| } |
| |
| API struct lyd_node * |
| lyd_child_no_keys(const struct lyd_node *node) |
| { |
| struct lyd_node **children; |
| |
| if (!node) { |
| return NULL; |
| } |
| |
| if (!node->schema) { |
| /* opaq node */ |
| return ((struct lyd_node_opaq *)(node))->child; |
| } |
| |
| children = lyd_node_children_p((struct lyd_node *)node); |
| if (children) { |
| struct lyd_node *child = *children; |
| while (child && child->schema && (child->schema->flags & LYS_KEY)) { |
| child = child->next; |
| } |
| return child; |
| } else { |
| return NULL; |
| } |
| } |
| |
| API const struct lys_module * |
| lyd_owner_module(const struct lyd_node *node) |
| { |
| const struct lysc_node *schema; |
| |
| if (!node || !node->schema) { |
| return NULL; |
| } |
| |
| for (schema = node->schema; schema->parent; schema = schema->parent) {} |
| return schema->module; |
| } |
| |
| API const struct lysc_when * |
| lyd_has_when(const struct lyd_node *node) |
| { |
| const struct lysc_node *schema; |
| |
| if (!node || !node->schema) { |
| return NULL; |
| } |
| |
| schema = node->schema; |
| do { |
| if (schema->when) { |
| return *schema->when; |
| } |
| schema = schema->parent; |
| } while (schema && (schema->nodetype & (LYS_CASE | LYS_CHOICE))); |
| |
| return NULL; |
| } |
| |
| const struct lys_module * |
| lyd_mod_next_module(struct lyd_node *tree, const struct lys_module *module, const struct ly_ctx *ctx, uint32_t *i, |
| struct lyd_node **first) |
| { |
| struct lyd_node *iter; |
| const struct lys_module *mod; |
| |
| /* get the next module */ |
| if (module) { |
| if (*i) { |
| mod = NULL; |
| } else { |
| mod = module; |
| ++(*i); |
| } |
| } else { |
| do { |
| mod = ly_ctx_get_module_iter(ctx, i); |
| } while (mod && !mod->implemented); |
| } |
| |
| /* find its data */ |
| *first = NULL; |
| if (mod) { |
| LY_LIST_FOR(tree, iter) { |
| if (lyd_owner_module(iter) == mod) { |
| *first = iter; |
| break; |
| } |
| } |
| } |
| |
| return mod; |
| } |
| |
| const struct lys_module * |
| lyd_data_next_module(struct lyd_node **next, struct lyd_node **first) |
| { |
| const struct lys_module *mod; |
| |
| if (!*next) { |
| /* all data traversed */ |
| *first = NULL; |
| return NULL; |
| } |
| |
| *first = *next; |
| |
| /* prepare next */ |
| mod = lyd_owner_module(*next); |
| LY_LIST_FOR(*next, *next) { |
| if (lyd_owner_module(*next) != mod) { |
| break; |
| } |
| } |
| |
| return mod; |
| } |
| |
| LY_ERR |
| lyd_parse_check_keys(struct lyd_node *node) |
| { |
| const struct lysc_node *skey = NULL; |
| const struct lyd_node *key; |
| |
| assert(node->schema->nodetype == LYS_LIST); |
| |
| key = lyd_child(node); |
| while ((skey = lys_getnext(skey, node->schema, NULL, 0)) && (skey->flags & LYS_KEY)) { |
| if (!key || (key->schema != skey)) { |
| LOGVAL(LYD_CTX(node), LY_VLOG_LYD, node, LY_VCODE_NOKEY, skey->name); |
| return LY_EVALID; |
| } |
| |
| key = key->next; |
| } |
| |
| return LY_SUCCESS; |
| } |
| |
| void |
| lyd_parse_set_data_flags(struct lyd_node *node, struct ly_set *when_check, struct lyd_meta **meta, uint32_t options) |
| { |
| struct lyd_meta *meta2, *prev_meta = NULL; |
| |
| if (!(node->schema->nodetype & (LYS_RPC | LYS_ACTION | LYS_NOTIF)) && node->schema->when) { |
| if (!(options & LYD_PARSE_ONLY)) { |
| /* remember we need to evaluate this node's when */ |
| LY_CHECK_RET(ly_set_add(when_check, node, 1, NULL), ); |
| } |
| } |
| |
| LY_LIST_FOR(*meta, meta2) { |
| if (!strcmp(meta2->name, "default") && !strcmp(meta2->annotation->module->name, "ietf-netconf-with-defaults") && |
| meta2->value.boolean) { |
| /* node is default according to the metadata */ |
| node->flags |= LYD_DEFAULT; |
| |
| /* delete the metadata */ |
| if (prev_meta) { |
| prev_meta->next = meta2->next; |
| } else { |
| *meta = (*meta)->next; |
| } |
| lyd_free_meta_single(meta2); |
| break; |
| } |
| |
| prev_meta = meta2; |
| } |
| } |
| |
| API LY_ERR |
| lyd_any_copy_value(struct lyd_node *trg, const union lyd_any_value *value, LYD_ANYDATA_VALUETYPE value_type) |
| { |
| struct lyd_node_any *t; |
| |
| assert(trg->schema->nodetype & LYS_ANYDATA); |
| |
| t = (struct lyd_node_any *)trg; |
| |
| /* free trg */ |
| switch (t->value_type) { |
| case LYD_ANYDATA_DATATREE: |
| lyd_free_all(t->value.tree); |
| break; |
| case LYD_ANYDATA_STRING: |
| case LYD_ANYDATA_XML: |
| case LYD_ANYDATA_JSON: |
| FREE_STRING(LYD_CTX(trg), t->value.str); |
| break; |
| case LYD_ANYDATA_LYB: |
| free(t->value.mem); |
| break; |
| } |
| t->value.str = NULL; |
| |
| if (!value) { |
| /* only free value in this case */ |
| return LY_SUCCESS; |
| } |
| |
| /* copy src */ |
| t->value_type = value_type; |
| switch (value_type) { |
| case LYD_ANYDATA_DATATREE: |
| if (value->tree) { |
| LY_CHECK_RET(lyd_dup_siblings(value->tree, NULL, LYD_DUP_RECURSIVE, &t->value.tree)); |
| } |
| break; |
| case LYD_ANYDATA_STRING: |
| case LYD_ANYDATA_XML: |
| case LYD_ANYDATA_JSON: |
| if (value->str) { |
| LY_CHECK_RET(lydict_insert(LYD_CTX(trg), value->str, 0, &t->value.str)); |
| } |
| break; |
| case LYD_ANYDATA_LYB: |
| if (value->mem) { |
| int len = lyd_lyb_data_length(value->mem); |
| LY_CHECK_RET(len == -1, LY_EINVAL); |
| t->value.mem = malloc(len); |
| LY_CHECK_ERR_RET(!t->value.mem, LOGMEM(LYD_CTX(trg)), LY_EMEM); |
| memcpy(t->value.mem, value->mem, len); |
| } |
| break; |
| } |
| |
| return LY_SUCCESS; |
| } |
| |
| LYB_HASH |
| lyb_hash(struct lysc_node *sibling, uint8_t collision_id) |
| { |
| const struct lys_module *mod; |
| uint32_t full_hash; |
| LYB_HASH hash; |
| |
| if ((collision_id < LYS_NODE_HASH_COUNT) && sibling->hash[collision_id]) { |
| return sibling->hash[collision_id]; |
| } |
| |
| mod = sibling->module; |
| |
| full_hash = dict_hash_multi(0, mod->name, strlen(mod->name)); |
| full_hash = dict_hash_multi(full_hash, sibling->name, strlen(sibling->name)); |
| if (collision_id) { |
| size_t ext_len; |
| |
| if (collision_id > strlen(mod->name)) { |
| /* fine, we will not hash more bytes, just use more bits from the hash than previously */ |
| ext_len = strlen(mod->name); |
| } else { |
| /* use one more byte from the module name than before */ |
| ext_len = collision_id; |
| } |
| full_hash = dict_hash_multi(full_hash, mod->name, ext_len); |
| } |
| full_hash = dict_hash_multi(full_hash, NULL, 0); |
| |
| /* use the shortened hash */ |
| hash = full_hash & (LYB_HASH_MASK >> collision_id); |
| /* add colision identificator */ |
| hash |= LYB_HASH_COLLISION_ID >> collision_id; |
| |
| /* save this hash */ |
| if (collision_id < LYS_NODE_HASH_COUNT) { |
| sibling->hash[collision_id] = hash; |
| } |
| |
| return hash; |
| } |
| |
| ly_bool |
| lyb_has_schema_model(const struct lysc_node *sibling, const struct lys_module **models) |
| { |
| LY_ARRAY_COUNT_TYPE u; |
| |
| LY_ARRAY_FOR(models, u) { |
| if (sibling->module == models[u]) { |
| return 1; |
| } |
| } |
| |
| return 0; |
| } |
| |
| void |
| ly_free_prefix_data(LY_PREFIX_FORMAT format, void *prefix_data) |
| { |
| struct ly_set *ns_list; |
| struct lysc_prefix *prefixes; |
| uint32_t i; |
| LY_ARRAY_COUNT_TYPE u; |
| |
| if (!prefix_data) { |
| return; |
| } |
| |
| switch (format) { |
| case LY_PREF_XML: |
| ns_list = prefix_data; |
| for (i = 0; i < ns_list->count; ++i) { |
| free(((struct lyxml_ns *)ns_list->objs[i])->prefix); |
| free(((struct lyxml_ns *)ns_list->objs[i])->uri); |
| } |
| ly_set_free(ns_list, free); |
| break; |
| case LY_PREF_SCHEMA_RESOLVED: |
| prefixes = prefix_data; |
| LY_ARRAY_FOR(prefixes, u) { |
| free(prefixes[u].prefix); |
| } |
| LY_ARRAY_FREE(prefixes); |
| break; |
| case LY_PREF_SCHEMA: |
| case LY_PREF_JSON: |
| break; |
| } |
| } |
| |
| LY_ERR |
| ly_dup_prefix_data(const struct ly_ctx *ctx, LY_PREFIX_FORMAT format, const void *prefix_data, |
| void **prefix_data_p) |
| { |
| LY_ERR ret = LY_SUCCESS; |
| struct lyxml_ns *ns; |
| struct lysc_prefix *prefixes = NULL, *orig_pref; |
| struct ly_set *ns_list, *orig_ns; |
| uint32_t i; |
| LY_ARRAY_COUNT_TYPE u; |
| |
| assert(!*prefix_data_p); |
| |
| switch (format) { |
| case LY_PREF_SCHEMA: |
| *prefix_data_p = (void *)prefix_data; |
| break; |
| case LY_PREF_SCHEMA_RESOLVED: |
| /* copy all the value prefixes */ |
| orig_pref = (struct lysc_prefix *)prefix_data; |
| LY_ARRAY_CREATE_GOTO(ctx, prefixes, LY_ARRAY_COUNT(orig_pref), ret, cleanup); |
| *prefix_data_p = prefixes; |
| |
| LY_ARRAY_FOR(orig_pref, u) { |
| if (orig_pref[u].prefix) { |
| prefixes[u].prefix = strdup(orig_pref[u].prefix); |
| LY_CHECK_ERR_GOTO(!prefixes[u].prefix, LOGMEM(ctx); ret = LY_EMEM, cleanup); |
| } |
| prefixes[u].mod = orig_pref[u].mod; |
| LY_ARRAY_INCREMENT(prefixes); |
| } |
| break; |
| case LY_PREF_XML: |
| /* copy all the namespaces */ |
| LY_CHECK_GOTO(ret = ly_set_new(&ns_list), cleanup); |
| *prefix_data_p = ns_list; |
| |
| orig_ns = (struct ly_set *)prefix_data; |
| for (i = 0; i < orig_ns->count; ++i) { |
| ns = calloc(1, sizeof *ns); |
| LY_CHECK_ERR_GOTO(!ns, LOGMEM(ctx); ret = LY_EMEM, cleanup); |
| LY_CHECK_GOTO(ret = ly_set_add(ns_list, ns, 1, NULL), cleanup); |
| |
| if (((struct lyxml_ns *)orig_ns->objs[i])->prefix) { |
| ns->prefix = strdup(((struct lyxml_ns *)orig_ns->objs[i])->prefix); |
| LY_CHECK_ERR_GOTO(!ns->prefix, LOGMEM(ctx); ret = LY_EMEM, cleanup); |
| } |
| ns->uri = strdup(((struct lyxml_ns *)orig_ns->objs[i])->uri); |
| LY_CHECK_ERR_GOTO(!ns->uri, LOGMEM(ctx); ret = LY_EMEM, cleanup); |
| } |
| break; |
| case LY_PREF_JSON: |
| assert(!prefix_data); |
| *prefix_data_p = NULL; |
| break; |
| } |
| |
| cleanup: |
| if (ret) { |
| ly_free_prefix_data(format, *prefix_data_p); |
| *prefix_data_p = NULL; |
| } |
| return ret; |
| } |
| |
| LY_ERR |
| ly_store_prefix_data(const struct ly_ctx *ctx, const char *value, size_t value_len, LY_PREFIX_FORMAT format, |
| void *prefix_data, LY_PREFIX_FORMAT *format_p, void **prefix_data_p) |
| { |
| LY_ERR ret = LY_SUCCESS; |
| const char *start, *stop; |
| const struct lys_module *mod; |
| struct lyxml_ns *ns; |
| struct ly_set *ns_list; |
| struct lysc_prefix *prefixes = NULL, *val_pref; |
| |
| switch (format) { |
| case LY_PREF_SCHEMA: |
| case LY_PREF_XML: |
| /* copy all referenced modules... */ |
| if (format == LY_PREF_XML) { |
| /* ...as prefix - namespace pairs */ |
| LY_CHECK_GOTO(ret = ly_set_new(&ns_list), cleanup); |
| *format_p = LY_PREF_XML; |
| *prefix_data_p = ns_list; |
| } else { |
| /* ...as prefix - module pairs */ |
| assert(format == LY_PREF_SCHEMA); |
| LY_ARRAY_CREATE_GOTO(ctx, prefixes, 0, ret, cleanup); |
| *format_p = LY_PREF_SCHEMA_RESOLVED; |
| *prefix_data_p = prefixes; |
| } |
| |
| /* add all used prefixes */ |
| for (stop = start = value; (size_t)(stop - value) < value_len; start = stop) { |
| size_t bytes; |
| uint32_t c; |
| |
| ly_getutf8(&stop, &c, &bytes); |
| if (is_xmlqnamestartchar(c)) { |
| for (ly_getutf8(&stop, &c, &bytes); |
| is_xmlqnamechar(c) && (size_t)(stop - value) < value_len; |
| ly_getutf8(&stop, &c, &bytes)) {} |
| stop = stop - bytes; |
| if (*stop == ':') { |
| /* we have a possible prefix */ |
| size_t len = stop - start; |
| |
| /* do we already have the prefix? */ |
| mod = ly_type_store_resolve_prefix(ctx, start, len, *format_p, *prefix_data_p); |
| if (!mod) { |
| mod = ly_type_store_resolve_prefix(ctx, start, len, format, prefix_data); |
| if (mod) { |
| if (*format_p == LY_PREF_XML) { |
| /* store a new prefix - namespace pair */ |
| ns = calloc(1, sizeof *ns); |
| LY_CHECK_ERR_GOTO(!ns, LOGMEM(ctx); ret = LY_EMEM, cleanup); |
| LY_CHECK_GOTO(ret = ly_set_add(ns_list, ns, 1, NULL), cleanup); |
| |
| ns->prefix = strndup(start, len); |
| LY_CHECK_ERR_GOTO(!ns->prefix, LOGMEM(ctx); ret = LY_EMEM, cleanup); |
| ns->uri = strdup(mod->ns); |
| LY_CHECK_ERR_GOTO(!ns->uri, LOGMEM(ctx); ret = LY_EMEM, cleanup); |
| } else { |
| assert(*format_p == LY_PREF_SCHEMA_RESOLVED); |
| /* store a new prefix - module pair */ |
| LY_ARRAY_NEW_GOTO(ctx, prefixes, val_pref, ret, cleanup); |
| *prefix_data_p = prefixes; |
| |
| val_pref->prefix = strndup(start, len); |
| LY_CHECK_ERR_GOTO(!val_pref->prefix, LOGMEM(ctx); ret = LY_EMEM, cleanup); |
| val_pref->mod = mod; |
| } |
| } /* else it is not even defined */ |
| } /* else the prefix is already present */ |
| } |
| stop = stop + bytes; |
| } |
| } |
| break; |
| case LY_PREF_SCHEMA_RESOLVED: |
| case LY_PREF_JSON: |
| /* simply copy all the prefix data */ |
| *format_p = format; |
| LY_CHECK_GOTO(ret = ly_dup_prefix_data(ctx, format, prefix_data, prefix_data_p), cleanup); |
| break; |
| } |
| |
| cleanup: |
| if (ret) { |
| ly_free_prefix_data(*format_p, *prefix_data_p); |
| *prefix_data_p = NULL; |
| } |
| return ret; |
| } |