/* Copyright (c) 2014, 2019, Oracle and/or its affiliates. All rights reserved. This program is free software; you can redistribute it and/or modify it under the terms of the GNU General Public License, version 2.0, as published by the Free Software Foundation. This program is also distributed with certain software (including but not limited to OpenSSL) that is licensed under separate terms, as designated in a particular file or component or in included license documentation. The authors of MySQL hereby grant you an additional permission to link the program and your derivative works with the separately licensed software that they have included with MySQL. This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License, version 2.0, for more details. You should have received a copy of the GNU General Public License along with this program; if not, write to the Free Software Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA */ #include "sql/dd/impl/bootstrap/bootstrapper.h" #include #include #include #include #include #include #include #include #include "m_ctype.h" #include "my_dbug.h" #include "my_loglevel.h" #include "my_sys.h" #include "mysql/components/services/log_builtins.h" #include "mysql_version.h" // MYSQL_VERSION_ID #include "mysqld_error.h" #include "sql/auth/sql_security_ctx.h" #include "sql/dd/cache/dictionary_client.h" // dd::cache::Dictionary_client #include "sql/dd/dd.h" // dd::create_object #include "sql/dd/impl/bootstrap/bootstrap_ctx.h" // DD_bootstrap_ctx #include "sql/dd/impl/cache/shared_dictionary_cache.h" // Shared_dictionary_cache #include "sql/dd/impl/cache/storage_adapter.h" // Storage_adapter #include "sql/dd/impl/dictionary_impl.h" // dd::Dictionary_impl #include "sql/dd/impl/raw/object_keys.h" #include "sql/dd/impl/sdi.h" // dd::sdi::store #include "sql/dd/impl/tables/character_sets.h" // dd::tables::Character_sets #include "sql/dd/impl/tables/collations.h" // dd::tables::Collations #include "sql/dd/impl/tables/dd_properties.h" // dd::tables::DD_properties #include "sql/dd/impl/tables/tables.h" // dd::tables::Tables #include "sql/dd/impl/types/schema_impl.h" // dd::Schema_impl #include "sql/dd/impl/types/table_impl.h" // dd::Table_impl #include "sql/dd/impl/types/tablespace_impl.h" // dd::Table_impl #include "sql/dd/impl/upgrade/dd.h" // dd::upgrade::upgrade_tables #include "sql/dd/impl/upgrade/server.h" // dd::upgrade::do_server_upgrade_checks #include "sql/dd/impl/utils.h" // dd::execute_query #include "sql/dd/object_id.h" #include "sql/dd/types/abstract_table.h" #include "sql/dd/types/object_table.h" // dd::Object_table #include "sql/dd/types/object_table_definition.h" // dd::Object_table_definition #include "sql/dd/types/table.h" #include "sql/dd/types/tablespace.h" #include "sql/dd/types/tablespace_file.h" // dd::Tablespace_file #include "sql/dd/upgrade/server.h" // UPGRADE_NONE #include "sql/dd/upgrade_57/upgrade.h" // upgrade_57::Upgrade_status #include "sql/handler.h" // dict_init_mode_t #include "sql/mdl.h" #include "sql/mysqld.h" #include "sql/sd_notify.h" // sysd::notify #include "sql/thd_raii.h" using namespace dd; /////////////////////////////////////////////////////////////////////////// namespace { // Initialize recovery in the DDSE. bool DDSE_dict_recover(THD *thd, dict_recovery_mode_t dict_recovery_mode, uint version) { handlerton *ddse = ha_resolve_by_legacy_type(thd, DB_TYPE_INNODB); if (ddse->dict_recover == nullptr) return true; bool error = ddse->dict_recover(dict_recovery_mode, version); /* Commit when tablespaces have been initialized, since in that case, tablespace meta data is added. */ if (dict_recovery_mode == DICT_RECOVERY_INITIALIZE_TABLESPACES) return dd::end_transaction(thd, error); return error; } /* Update the System_tables registry with meta data from 'dd_properties'. Iterate over the tables in the DD_properties. If this is minor downgrade, add new tables that were added in the newer version to the System_tables registry. If this is not minor downgrade, assert that all tables in the DD_properties indeed have a corresponding entry in the System_tables registry. */ bool update_system_tables(THD *thd) { std::unique_ptr system_tables_props; bool exists = false; if (dd::tables::DD_properties::instance().get( thd, "SYSTEM_TABLES", &system_tables_props, &exists) || !exists) { my_error(ER_DD_INIT_FAILED, MYF(0)); return true; } /* We would normally use a range based loop here, but the developerstudio compiler on Solaris does not handle this when the base collection has pure virtual begin() and end() functions. */ for (Properties::const_iterator it = system_tables_props->begin(); it != system_tables_props->end(); ++it) { // Check if this is a CORE, INERT, SECOND or DDSE table. if (!dd::get_dictionary()->is_dd_table_name(MYSQL_SCHEMA_NAME.str, it->first)) { if (bootstrap::DD_bootstrap_ctx::instance().is_minor_downgrade()) { /* Add tables as type CORE regardless of the actual type, which is irrelevant in this case. */ System_tables::instance()->add(MYSQL_SCHEMA_NAME.str, it->first, System_tables::Types::CORE, nullptr); } else { my_error(ER_DD_METADATA_NOT_FOUND, MYF(0), it->first.c_str()); return true; } } else { /* The table is a known DD table. Then, we get its definition and add it to the Object_table instance. The definition might not exist if the table was added after the version that we are upgrading from. */ String_type tbl_prop_str; if (!system_tables_props->exists(it->first) || system_tables_props->get(it->first, &tbl_prop_str)) continue; const Object_table *table_def = System_tables::instance()->find_table( MYSQL_SCHEMA_NAME.str, it->first); DBUG_ASSERT(table_def); std::unique_ptr tbl_props( Properties::parse_properties(tbl_prop_str)); String_type def; if (tbl_props->get(dd::tables::DD_properties::dd_key( dd::tables::DD_properties::DD_property::DEF), &def)) { my_error(ER_DD_METADATA_NOT_FOUND, MYF(0), it->first.c_str()); return true; } std::unique_ptr table_def_properties( Properties::parse_properties(def)); table_def->set_actual_table_definition(*table_def_properties); } } return false; } // Create a DD table using the target table definition. bool create_target_table(THD *thd, const Object_table *object_table) { DBUG_ASSERT(object_table != nullptr); /* The target table definition may not be present if the table is abandoned. That's ok, not an error. */ if (object_table->is_abandoned()) return false; String_type target_ddl_statement(""); const Object_table_definition *target_table_def = object_table->target_table_definition(); DBUG_ASSERT(target_table_def != nullptr); target_ddl_statement = target_table_def->get_ddl(); DBUG_ASSERT(!target_ddl_statement.empty()); return dd::execute_query(thd, target_ddl_statement); } // Create a DD table using the actual table definition. /* purecov: begin inspected */ bool create_actual_table(THD *thd, const Object_table *object_table) { /* For minor downgrade, tables might have been added in the upgraded server that we do not have any Object_table instance for. In that case, we just skip them. */ if (object_table == nullptr) { DBUG_ASSERT(bootstrap::DD_bootstrap_ctx::instance().is_minor_downgrade()); return false; } String_type actual_ddl_statement(""); const Object_table_definition *actual_table_def = object_table->actual_table_definition(); /* The actual definition may not be present. This will happen during upgrade if the new DD version adds a new DD table which was not present in the DD we are upgrading from. This is OK, not an error. */ if (actual_table_def == nullptr) return false; actual_ddl_statement = actual_table_def->get_ddl(); DBUG_ASSERT(!actual_ddl_statement.empty()); return dd::execute_query(thd, actual_ddl_statement); } /* purecov: end */ /* Acquire exclusive meta data locks for the DD schema, tablespace and table names. */ bool acquire_exclusive_mdl(THD *thd) { // All MDL requests. MDL_request_list mdl_requests; // Prepare MDL request for the schema name. MDL_request schema_request; MDL_REQUEST_INIT(&schema_request, MDL_key::SCHEMA, MYSQL_SCHEMA_NAME.str, "", MDL_EXCLUSIVE, MDL_TRANSACTION); mdl_requests.push_front(&schema_request); // Prepare MDL request for the tablespace name. MDL_request tablespace_request; MDL_REQUEST_INIT(&tablespace_request, MDL_key::TABLESPACE, "", MYSQL_TABLESPACE_NAME.str, MDL_EXCLUSIVE, MDL_TRANSACTION); mdl_requests.push_front(&tablespace_request); // Prepare MDL requests for all tables names. for (System_tables::Const_iterator it = System_tables::instance()->begin(); it != System_tables::instance()->end(); ++it) { // Skip extraneous tables during minor downgrade. if ((*it)->entity() == nullptr) continue; MDL_request *table_request = new (thd->mem_root) MDL_request; if (table_request == NULL) return true; MDL_REQUEST_INIT(table_request, MDL_key::TABLE, MYSQL_SCHEMA_NAME.str, (*it)->entity()->name().c_str(), MDL_EXCLUSIVE, MDL_TRANSACTION); mdl_requests.push_front(table_request); } // Finally, acquire all the MDL locks. return (thd->mdl_context.acquire_locks(&mdl_requests, thd->variables.lock_wait_timeout)); } /* Acquire the DD schema, tablespace and table objects. Clone the objects, reset ID, store persistently, and update the storage adapter. */ bool flush_meta_data(THD *thd) { // Acquire exclusive meta data locks for the relevant DD objects. if (acquire_exclusive_mdl(thd)) return true; { /* Use a scoped auto releaser to make sure the cached objects are released before the shared cache is reset. */ dd::cache::Dictionary_client::Auto_releaser releaser(thd->dd_client()); /* First, we acquire the DD schema and tablespace and keep them in local variables. We also clone them, the clones will be used for updating the ids. We also acquire all the DD table objects to make sure the shared cache is populated, and we keep the original objects as well as clones in a vector. The auto releaser will make sure the objects are not evicted. This must be ensured since we need to make sure the ids stay consistent across all objects in the shared cache. */ const Schema *dd_schema = nullptr; const Tablespace *dd_tspace = nullptr; std::vector dd_tables; // Owned by the shared cache. std::vector> dd_table_clones; if (thd->dd_client()->acquire(dd::String_type(MYSQL_SCHEMA_NAME.str), &dd_schema) || thd->dd_client()->acquire(dd::String_type(MYSQL_TABLESPACE_NAME.str), &dd_tspace)) return dd::end_transaction(thd, true); std::unique_ptr dd_schema_clone( dynamic_cast(dd_schema->clone())); std::unique_ptr dd_tspace_clone( dynamic_cast(dd_tspace->clone())); for (System_tables::Const_iterator it = System_tables::instance()->begin(); it != System_tables::instance()->end(); ++it) { /* We add nullptr to the dd_tables vector for abandoned tables and system tables to have the same number of objects in the System_tables list, the dd_tables vector and the dd_table_clones vector. */ const dd::Table *dd_table = nullptr; if ((*it)->property() != System_tables::Types::SYSTEM && thd->dd_client()->acquire(MYSQL_SCHEMA_NAME.str, (*it)->entity()->name(), &dd_table)) return dd::end_transaction(thd, true); dd_tables.push_back(dynamic_cast(dd_table)); /* If this is an abandoned table, we can't clone it. Thus, we push back a nullptr to make sure we have the same number of elements in the dd_table_clones as in the System_tables. */ std::unique_ptr dd_table_clone; if (dd_table != nullptr) { dd_table_clones.push_back(std::unique_ptr( dynamic_cast(dd_table->clone()))); } else { dd_table_clones.push_back(nullptr); } } /* We have now populated the shared cache with the core objects, and kept clones of all DD objects. The scoped auto releaser makes sure we will not evict the objects from the shared cache until the auto releaser exits scope. Thus, within the scope of the auto releaser, we can modify the contents of the core registry in the storage adapter without risking that this will interfere with the contents of the shared cache, because the DD transactions will acquire the core objects from the shared cache. */ /* First, we modify and store the DD schema without changing the cached copy. We cannot use acquire_for_modification() here, because that would make the DD sub-transactions (e.g. when calling store()) see a partially modified set of core objects, where e.g. the mysql schema object has got its new, real id (from the auto-inc column in the dd.schemata table), whereas the core DD table objects still refer to the id that was allocated when creating the scaffolding. So we first store all the objects persistently, and make sure that the on-disk data will have correct and consistent ids. When all objects are stored, we update the contents of the core registry in the storage adapter to reflect the persisted data. Finally, the shared cache is reset so that on next acquisition, the DD objects will be fetched from the core registry in the storage adapter. */ /* We must set the ID to INVALID to make the object get a fresh ID from the auto inc ID column. */ dd_schema_clone->set_id(INVALID_OBJECT_ID); dd_tspace_clone->set_id(INVALID_OBJECT_ID); if (dd::cache::Storage_adapter::instance()->store( thd, static_cast(dd_schema_clone.get())) || dd::cache::Storage_adapter::instance()->store( thd, static_cast(dd_tspace_clone.get()))) return dd::end_transaction(thd, true); /* Now, the DD schema and DD tablespace are stored persistently. We will not update the core registry until after we have stored all DD tables. At that point, we can update all the core registry objects in one go and avoid using a partially update core registry for e.g. object acquisition. */ std::vector>::iterator clone_it = dd_table_clones.begin(); for (System_tables::Const_iterator it = System_tables::instance()->begin(); it != System_tables::instance()->end() && clone_it != dd_table_clones.end(); ++it, ++clone_it) { // Skip abandoned tables and system tables. if ((*clone_it) == nullptr || (*it)->property() == System_tables::Types::SYSTEM) continue; DBUG_ASSERT((*it)->entity()->name() == (*clone_it)->name()); // We must set the ID to INVALID to let the object get an auto inc ID. (*clone_it)->set_id(INVALID_OBJECT_ID); /* Change the schema and tablespace id to match the ids of the persisted objects. Note that this means the persisted DD table objects will have consistent IDs, but the IDs in the objects in the core registry will not be updated yet. */ (*clone_it)->set_schema_id(dd_schema_clone->id()); (*clone_it)->set_tablespace_id(dd_tspace_clone->id()); if (dd::cache::Storage_adapter::instance()->store( thd, static_cast((*clone_it).get()))) return dd::end_transaction(thd, true); } /* Update and store the predefined tablespace objects. The DD tablespace has already been stored above, so we iterate only over the tablespaces of type PREDEFINED_DDSE. */ for (System_tablespaces::Const_iterator it = System_tablespaces::instance()->begin( System_tablespaces::Types::PREDEFINED_DDSE); it != System_tablespaces::instance()->end(); it = System_tablespaces::instance()->next( it, System_tablespaces::Types::PREDEFINED_DDSE)) { const dd::Tablespace *tspace = nullptr; if (thd->dd_client()->acquire((*it)->key().second, &tspace)) return dd::end_transaction(thd, true); std::unique_ptr tspace_clone( dynamic_cast(tspace->clone())); // We must set the ID to INVALID to enable storing the object. tspace_clone->set_id(INVALID_OBJECT_ID); if (dd::cache::Storage_adapter::instance()->store( thd, static_cast(tspace_clone.get()))) return dd::end_transaction(thd, true); /* Only the DD tablespace is needed to handle cache misses, so we can just drop the predefined tablespaces from the core registry now that it has been persisted. */ dd::cache::Storage_adapter::instance()->core_drop(thd, tspace); } /* Now, the DD schema and tablespace as well as the DD tables have been persisted. The last thing we do before resetting the shared cache is to update the contents of the core registry to match the persisted objects. First, we update the core registry with the persisted DD schema and tablespace. */ dd::cache::Storage_adapter::instance()->core_drop(thd, dd_schema); dd::cache::Storage_adapter::instance()->core_store( thd, static_cast(dd_schema_clone.get())); dd::cache::Storage_adapter::instance()->core_drop(thd, dd_tspace); dd::cache::Storage_adapter::instance()->core_store( thd, static_cast(dd_tspace_clone.get())); // Make sure the IDs after storing are as expected. DBUG_ASSERT(dd_schema_clone->id() == 1); DBUG_ASSERT(dd_tspace_clone->id() == 1); /* Finally, we update the core registry of the DD tables. This must be done in two loops to avoid issues related to overlapping ID sequences. */ std::vector::const_iterator table_it = dd_tables.begin(); for (System_tables::Const_iterator it = System_tables::instance()->begin(); it != System_tables::instance()->end() && table_it != dd_tables.end(); ++it, ++table_it) { // Skip abandoned tables and system tables. if ((*table_it) == nullptr || (*it)->property() == System_tables::Types::SYSTEM) continue; DBUG_ASSERT((*it)->entity()->name() == (*table_it)->name()); dd::cache::Storage_adapter::instance()->core_drop( thd, static_cast(*table_it)); } clone_it = dd_table_clones.begin(); for (System_tables::Const_iterator it = System_tables::instance()->begin(); it != System_tables::instance()->end() && clone_it != dd_table_clones.end(); ++it, ++clone_it) { // Skip abandoned tables and system tables. if ((*clone_it) == nullptr || (*it)->property() == System_tables::Types::SYSTEM) continue; if ((*it)->property() == System_tables::Types::CORE) { DBUG_ASSERT((*it)->entity()->name() == (*clone_it)->name()); dd::cache::Storage_adapter::instance()->core_store( thd, static_cast
((*clone_it).get())); } } } /* Now, the auto releaser has released the objects, and we can go ahead and reset the shared cache. */ dd::cache::Shared_dictionary_cache::instance()->reset(true); if (dd::end_transaction(thd, false)) { return true; } /* Use a scoped auto releaser to make sure the objects cached for SDI writing, FK parent information reload, and DD property storage are released. */ dd::cache::Dictionary_client::Auto_releaser releaser(thd->dd_client()); // Acquire the DD tablespace and write SDI const Tablespace *dd_tspace = nullptr; if (thd->dd_client()->acquire(dd::String_type(MYSQL_TABLESPACE_NAME.str), &dd_tspace) || dd::sdi::store(thd, dd_tspace)) { return dd::end_transaction(thd, true); } // Acquire the DD schema and write SDI const Schema *dd_schema = nullptr; if (thd->dd_client()->acquire(dd::String_type(MYSQL_SCHEMA_NAME.str), &dd_schema) || dd::sdi::store(thd, dd_schema)) { return dd::end_transaction(thd, true); } /* Acquire the DD table objects and write SDI for them. Also sync from the DD tables in order to get the FK parent information reloaded. */ for (System_tables::Const_iterator it = System_tables::instance()->begin(); it != System_tables::instance()->end(); ++it) { // Skip system tables. if ((*it)->property() == System_tables::Types::SYSTEM) continue; const dd::Table *dd_table = nullptr; if (thd->dd_client()->acquire(MYSQL_SCHEMA_NAME.str, (*it)->entity()->name(), &dd_table)) { return dd::end_transaction(thd, true); } // Skip abandoned tables. if (dd_table == nullptr) continue; /* Make sure the registry of the core DD objects is updated with an object read from the DD tables, with updated FK parent information. Store the object to make sure SDI is written. */ Abstract_table::Name_key table_key; Abstract_table::update_name_key(&table_key, dd_schema->id(), dd_table->name()); const dd::Abstract_table *persisted_dd_table = nullptr; if (dd::cache::Storage_adapter::instance()->get( thd, table_key, ISO_READ_COMMITTED, true, &persisted_dd_table) || persisted_dd_table == nullptr || dd::sdi::store(thd, dynamic_cast(persisted_dd_table))) { if (persisted_dd_table != nullptr) delete persisted_dd_table; return dd::end_transaction(thd, true); } if ((*it)->property() == System_tables::Types::CORE) { dd::cache::Storage_adapter::instance()->core_drop(thd, dd_table); dd::cache::Storage_adapter::instance()->core_store( thd, dynamic_cast
( const_cast(persisted_dd_table))); } if (persisted_dd_table != nullptr) delete persisted_dd_table; } bootstrap::DD_bootstrap_ctx::instance().set_stage(bootstrap::Stage::SYNCED); return dd::end_transaction(thd, false); } // Insert additional data into the DD tables. bool populate_tables(THD *thd) { // Iterate over DD tables, populate tables. for (System_tables::Const_iterator it = System_tables::instance()->begin(); it != System_tables::instance()->end(); ++it) { // Skip system tables. if ((*it)->property() == System_tables::Types::SYSTEM) continue; // Retrieve list of SQL statements to execute. const Object_table_definition *table_def = (*it)->entity()->target_table_definition(); // Skip abandoned tables. if (table_def == nullptr) continue; bool error = false; std::vector stmt = table_def->get_dml(); for (std::vector::iterator stmt_it = stmt.begin(); stmt_it != stmt.end() && !error; ++stmt_it) error = dd::execute_query(thd, *stmt_it); // Commit the statement based population. if (dd::end_transaction(thd, error)) return true; // If no error, call the low level table population method, and commit it. error = (*it)->entity()->populate(thd); if (dd::end_transaction(thd, error)) return true; } bootstrap::DD_bootstrap_ctx::instance().set_stage( bootstrap::Stage::POPULATED); return false; } // Re-populate character sets and collations upon normal restart. bool repopulate_charsets_and_collations(THD *thd) { /* We must check if the DDSE is started in a way that makes the DD read only. For now, we only support InnoDB as SE for the DD. The call to retrieve the handlerton for the DDSE should be replaced by a more generic mechanism. */ handlerton *ddse = ha_resolve_by_legacy_type(thd, DB_TYPE_INNODB); if (ddse->is_dict_readonly && ddse->is_dict_readonly()) { LogErr(WARNING_LEVEL, ER_DD_NO_WRITES_NO_REPOPULATION, "InnoDB", " "); return false; } /* Otherwise, turn off FK checks, re-populate and commit. The FK checks must be turned off since the collations and character sets reference each other. */ bool error = dd::execute_query(thd, "SET FOREIGN_KEY_CHECKS= 0") || tables::Collations::instance().populate(thd) || tables::Character_sets::instance().populate(thd); /* We must commit the re-population before executing a new query, which expects the transaction to be empty, and finally, turn FK checks back on. */ error |= dd::end_transaction(thd, error); error |= dd::execute_query(thd, "SET FOREIGN_KEY_CHECKS= 1"); bootstrap::DD_bootstrap_ctx::instance().set_stage( bootstrap::Stage::POPULATED); return error; } /* Verify that the storage adapter contains the core DD objects and nothing else. */ bool verify_contents(THD *thd) { // Verify that the DD schema is present, and that its id == 1. Schema::Name_key schema_key; Schema::update_name_key(&schema_key, MYSQL_SCHEMA_NAME.str); Object_id dd_schema_id = cache::Storage_adapter::instance()->core_get_id(schema_key); DBUG_ASSERT(dd_schema_id == MYSQL_SCHEMA_DD_ID); if (dd_schema_id == INVALID_OBJECT_ID) { LogErr(ERROR_LEVEL, ER_DD_SCHEMA_NOT_FOUND, MYSQL_SCHEMA_NAME.str); return dd::end_transaction(thd, true); } DBUG_ASSERT(cache::Storage_adapter::instance()->core_size() == 1); // Verify that the core DD tables are present. #ifndef DBUG_OFF size_t n_core_tables = 0; #endif for (System_tables::Const_iterator it = System_tables::instance()->begin(System_tables::Types::CORE); it != System_tables::instance()->end(); it = System_tables::instance()->next(it, System_tables::Types::CORE)) { // Skip extraneous tables for minor downgrade. if ((*it)->entity() == nullptr) continue; #ifndef DBUG_OFF n_core_tables++; #endif Table::Name_key table_key; Table::update_name_key(&table_key, dd_schema_id, (*it)->entity()->name()); Object_id dd_table_id = cache::Storage_adapter::instance()->core_get_id
(table_key); DBUG_ASSERT(dd_table_id != INVALID_OBJECT_ID); if (dd_table_id == INVALID_OBJECT_ID) { LogErr(ERROR_LEVEL, ER_DD_TABLE_NOT_FOUND, (*it)->entity()->name().c_str()); return dd::end_transaction(thd, true); } } DBUG_ASSERT(cache::Storage_adapter::instance()->core_size() == n_core_tables); // Verify that the dictionary tablespace is present and that its id == 1. Tablespace::Name_key tspace_key; Tablespace::update_name_key(&tspace_key, MYSQL_TABLESPACE_NAME.str); Object_id dd_tspace_id = cache::Storage_adapter::instance()->core_get_id(tspace_key); DBUG_ASSERT(dd_tspace_id == MYSQL_TABLESPACE_DD_ID); if (dd_tspace_id == INVALID_OBJECT_ID) { LogErr(ERROR_LEVEL, ER_DD_TABLESPACE_NOT_FOUND, MYSQL_TABLESPACE_NAME.str); return dd::end_transaction(thd, true); } DBUG_ASSERT(cache::Storage_adapter::instance()->core_size() == 1); return dd::end_transaction(thd, false); } } // namespace /////////////////////////////////////////////////////////////////////////// namespace dd { namespace bootstrap { /* Do the necessary DD-related initialization in the DDSE, and get the predefined tables and tablespaces. */ bool DDSE_dict_init(THD *thd, dict_init_mode_t dict_init_mode, uint version) { handlerton *ddse = ha_resolve_by_legacy_type(thd, DB_TYPE_INNODB); /* The lists with element wrappers are mem root allocated. The wrapped instances are allocated dynamically in the DDSE. These instances will be owned by the System_tables registry by the end of this function. */ List ddse_tables; List ddse_tablespaces; if (ddse->ddse_dict_init == nullptr || ddse->ddse_dict_init(dict_init_mode, version, &ddse_tables, &ddse_tablespaces)) return true; /* Iterate over the table definitions and add them to the System_tables registry. The Object_table instances will later be used to execute CREATE TABLE statements to actually create the tables. If Object_table::is_hidden(), then we add the tables as type DDSE_PRIVATE (not available neither for DDL nor DML), otherwise, we add them as type DDSE_PROTECTED (available for DML, not for DDL). */ List_iterator table_it(ddse_tables); const Object_table *ddse_table = nullptr; while ((ddse_table = table_it++)) { System_tables::Types table_type = System_tables::Types::DDSE_PROTECTED; if (ddse_table->is_hidden()) { table_type = System_tables::Types::DDSE_PRIVATE; } System_tables::instance()->add(MYSQL_SCHEMA_NAME.str, ddse_table->name(), table_type, ddse_table); } /* Get the server version number from the DD tablespace header and verify that we are allowed to upgrade from that version. The error handling is done after adding the ddse tables into the system registry to avoid memory leaks. */ if (!opt_initialize) { uint server_version = 0; if (ddse->dict_get_server_version == nullptr || ddse->dict_get_server_version(&server_version)) { LogErr(ERROR_LEVEL, ER_CANNOT_GET_SERVER_VERSION_FROM_TABLESPACE_HEADER); return true; } if (server_version != MYSQL_VERSION_ID) { if (opt_upgrade_mode == UPGRADE_NONE) { LogErr(ERROR_LEVEL, ER_SERVER_UPGRADE_OFF); return true; } if (!DD_bootstrap_ctx::instance().supported_server_version( server_version)) { LogErr(ERROR_LEVEL, ER_SERVER_UPGRADE_VERSION_NOT_SUPPORTED, server_version); return true; } } } /* At this point, the Systen_tables registry contains the INERT DD tables, and the DDSE tables. Before we continue, we must add the remaining DD tables. */ System_tables::instance()->add_remaining_dd_tables(); /* Iterate over the tablespace definitions, add the names and the tablespace meta data to the System_tablespaces registry. The meta data will be used later to create dd::Tablespace objects. The Plugin_tablespace instances are owned by the DDSE. */ List_iterator tablespace_it(ddse_tablespaces); const Plugin_tablespace *tablespace = nullptr; while ((tablespace = tablespace_it++)) { // Add the name and the object instance to the registry with the // appropriate property. if (my_strcasecmp(system_charset_info, MYSQL_TABLESPACE_NAME.str, tablespace->get_name()) == 0) System_tablespaces::instance()->add( tablespace->get_name(), System_tablespaces::Types::DD, tablespace); else System_tablespaces::instance()->add( tablespace->get_name(), System_tablespaces::Types::PREDEFINED_DDSE, tablespace); } return false; } // Initialize the data dictionary. bool initialize_dictionary(THD *thd, bool is_dd_upgrade_57, Dictionary_impl *d) { if (is_dd_upgrade_57) bootstrap::DD_bootstrap_ctx::instance().set_stage( bootstrap::Stage::STARTED); store_predefined_tablespace_metadata(thd); if (create_dd_schema(thd) || initialize_dd_properties(thd) || create_tables(thd, nullptr)) return true; if (is_dd_upgrade_57) { // Add status to mark creation of dictionary in InnoDB. // Till this step, no new undo log is created by InnoDB. if (upgrade_57::Upgrade_status().update( upgrade_57::Upgrade_status::enum_stage::DICT_TABLES_CREATED)) return true; } DBUG_EXECUTE_IF( "dd_upgrade_stage_2", if (is_dd_upgrade_57) { /* Server will crash will upgrading 5.7 data directory. This will leave server is an inconsistent state. File tracking upgrade will have Stage 2 written in it. Next restart of server on same data directory should revert all changes done by upgrade and data directory should be reusable by 5.7 server. */ DBUG_SUICIDE(); }); if (DDSE_dict_recover(thd, DICT_RECOVERY_INITIALIZE_SERVER, d->get_target_dd_version()) || flush_meta_data(thd) || DDSE_dict_recover(thd, DICT_RECOVERY_INITIALIZE_TABLESPACES, d->get_target_dd_version()) || populate_tables(thd) || update_properties(thd, nullptr, nullptr, String_type(MYSQL_SCHEMA_NAME.str)) || verify_contents(thd) | update_versions(thd, is_dd_upgrade_57)) return true; bootstrap::DD_bootstrap_ctx::instance().set_stage(bootstrap::Stage::FINISHED); return false; } // First time server start and initialization of the data dictionary. bool initialize(THD *thd) { bootstrap::DD_bootstrap_ctx::instance().set_stage(bootstrap::Stage::STARTED); /* Set tx_read_only to false to allow installing DD tables even if the server is started with --transaction-read-only=true. */ thd->variables.transaction_read_only = false; thd->tx_read_only = false; Disable_autocommit_guard autocommit_guard(thd); Dictionary_impl *d = dd::Dictionary_impl::instance(); DBUG_ASSERT(d); cache::Dictionary_client::Auto_releaser releaser(thd->dd_client()); /* Each step in the install process below is committed independently, either implicitly (for e.g. "CREATE TABLE") or explicitly (for the operations in the "populate()" methods). Thus, there is no need to commit explicitly here. */ if (DDSE_dict_init(thd, DICT_INIT_CREATE_FILES, d->get_target_dd_version()) || initialize_dictionary(thd, false, d)) return true; DBUG_ASSERT(d->get_target_dd_version() == d->get_actual_dd_version(thd)); LogErr(INFORMATION_LEVEL, ER_DD_VERSION_INSTALLED, d->get_target_dd_version()); return false; } // Normal server restart. bool restart(THD *thd) { bootstrap::DD_bootstrap_ctx::instance().set_stage(bootstrap::Stage::STARTED); /* Set tx_read_only to false to allow installing DD tables even if the server is started with --transaction-read-only=true. */ thd->variables.transaction_read_only = false; thd->tx_read_only = false; // Set explicit_defaults_for_timestamp variable for dictionary creation thd->variables.explicit_defaults_for_timestamp = true; Disable_autocommit_guard autocommit_guard(thd); Dictionary_impl *d = dd::Dictionary_impl::instance(); DBUG_ASSERT(d); cache::Dictionary_client::Auto_releaser releaser(thd->dd_client()); store_predefined_tablespace_metadata(thd); if (create_dd_schema(thd) || initialize_dd_properties(thd) || create_tables(thd, nullptr) || sync_meta_data(thd) || DDSE_dict_recover(thd, DICT_RECOVERY_RESTART_SERVER, d->get_actual_dd_version(thd)) || upgrade::do_server_upgrade_checks(thd) || upgrade::upgrade_tables(thd) || repopulate_charsets_and_collations(thd) || verify_contents(thd) || update_versions(thd, false)) { return true; } bootstrap::DD_bootstrap_ctx::instance().set_stage(bootstrap::Stage::FINISHED); LogErr(INFORMATION_LEVEL, ER_DD_VERSION_FOUND, d->get_actual_dd_version(thd)); return false; } // Initialize dictionary in case of server restart. void recover_innodb_upon_upgrade(THD *thd) { Dictionary_impl *d = dd::Dictionary_impl::instance(); store_predefined_tablespace_metadata(thd); // RAII to handle error in execution of CREATE TABLE. Key_length_error_handler key_error_handler; /* Ignore ER_TOO_LONG_KEY for dictionary tables during restart. Do not print the error in error log as we are creating only the cached objects and not physical tables. TODO: Workaround due to bug#20629014. Remove when the bug is fixed. */ thd->push_internal_handler(&key_error_handler); if (create_dd_schema(thd) || initialize_dd_properties(thd) || create_tables(thd, nullptr) || DDSE_dict_recover(thd, DICT_RECOVERY_RESTART_SERVER, d->get_actual_dd_version(thd))) { // Error is not be handled in this case as we are on cleanup code path. LogErr(WARNING_LEVEL, ER_DD_INIT_UPGRADE_FAILED); } thd->pop_internal_handler(); return; } bool setup_dd_objects_and_collations(THD *thd) { // Continue with server startup. bootstrap::DD_bootstrap_ctx::instance().set_stage( bootstrap::Stage::CREATED_TABLES); /* Set tx_read_only to false to allow installing DD tables even if the server is started with --transaction-read-only=true. */ thd->variables.transaction_read_only = false; thd->tx_read_only = false; Disable_autocommit_guard autocommit_guard(thd); Dictionary_impl *d = dd::Dictionary_impl::instance(); DBUG_ASSERT(d); DBUG_ASSERT(d->get_target_dd_version() == d->get_actual_dd_version(thd)); /* In this context, we initialize the target tables directly since this is a restart based on a pre-transactional-DD server, so ordinary upgrade does not need to be considered. */ if (sync_meta_data(thd) || repopulate_charsets_and_collations(thd) || verify_contents(thd) || update_versions(thd, false)) { return true; } bootstrap::DD_bootstrap_ctx::instance().set_stage(bootstrap::Stage::FINISHED); LogErr(INFORMATION_LEVEL, ER_DD_VERSION_FOUND, d->get_actual_dd_version(thd)); return false; } } // namespace bootstrap void store_predefined_tablespace_metadata(THD *thd) { /* Create dd::Tablespace objects and store them (which will add their meta data to the storage adapter registry of DD entities). The tablespaces are already created physically in the DDSE, so we only need to create the corresponding meta data. */ for (System_tablespaces::Const_iterator it = System_tablespaces::instance()->begin(); it != System_tablespaces::instance()->end(); ++it) { const Plugin_tablespace *tablespace_def = (*it)->entity(); // Create the dd::Tablespace object. std::unique_ptr tablespace(dd::create_object()); tablespace->set_name(tablespace_def->get_name()); tablespace->set_options(tablespace_def->get_options()); tablespace->set_se_private_data(tablespace_def->get_se_private_data()); tablespace->set_engine(tablespace_def->get_engine()); // Loop over the tablespace files, create dd::Tablespace_file objects. List files = tablespace_def->get_files(); List_iterator file_it( files); const Plugin_tablespace::Plugin_tablespace_file *file = NULL; while ((file = file_it++)) { Tablespace_file *space_file = tablespace->add_file(); space_file->set_filename(file->get_name()); space_file->set_se_private_data(file->get_se_private_data()); } // All the predefined tablespace are unencrypted (atleast for now). tablespace->options().set("encryption", "N"); /* Here, we just want to populate the core registry in the storage adapter. We do not want to have the object registered in the uncommitted registry, this will only add complexity to the DD cache usage during bootstrap. Thus, we call the storage adapter directly instead of going through the dictionary client. */ dd::cache::Storage_adapter::instance()->store(thd, tablespace.get()); } bootstrap::DD_bootstrap_ctx::instance().set_stage( bootstrap::Stage::CREATED_TABLESPACES); } bool create_dd_schema(THD *thd) { return dd::execute_query(thd, dd::String_type("CREATE SCHEMA ") + dd::String_type(MYSQL_SCHEMA_NAME.str) + dd::String_type(" DEFAULT COLLATE ") + dd::String_type(default_charset_info->name)) || dd::execute_query(thd, dd::String_type("USE ") + dd::String_type(MYSQL_SCHEMA_NAME.str)); } bool initialize_dd_properties(THD *thd) { // Create the dd_properties table. const Object_table_definition *dd_properties_def = dd::tables::DD_properties::instance().target_table_definition(); if (dd::execute_query(thd, dd_properties_def->get_ddl())) return true; /* We can now decide which version number we will use for the DD, and initialize the DD_bootstrap_ctx with the relevant version number. */ uint actual_version = dd::DD_VERSION; uint actual_server_version = MYSQL_VERSION_ID; uint upgraded_server_version = MYSQL_VERSION_ID; bootstrap::DD_bootstrap_ctx::instance().set_actual_dd_version(actual_version); bootstrap::DD_bootstrap_ctx::instance().set_upgraded_server_version( actual_server_version); Minor_upgrade_ctx::instance()->set_extra_mvu_version( Minor_upgrade_ctx::instance()->get_target_extra_mvu_version()); if (!opt_initialize) { bool exists = false; bool exists_server = false; bool exists_upgraded_version = false; // Check 'DD_version' too in order to catch an upgrade from 8.0.3. if (dd::tables::DD_properties::instance().get(thd, "DD_VERSION", &actual_version, &exists) || !exists) { LogErr(ERROR_LEVEL, ER_DD_NO_VERSION_FOUND); return true; } /* purecov: begin inspected */ if (actual_version != dd::DD_VERSION) { bootstrap::DD_bootstrap_ctx::instance().set_actual_dd_version( actual_version); if (opt_no_dd_upgrade) { push_deprecated_warn(thd, "--no-dd-upgrade", "--upgrade=NONE"); LogErr(ERROR_LEVEL, ER_DD_UPGRADE_OFF); return true; } if (!bootstrap::DD_bootstrap_ctx::instance().supported_dd_version()) { /* If we are attempting on minor downgrade, make sure this is supported. */ if (!bootstrap::DD_bootstrap_ctx::instance().is_minor_downgrade()) { LogErr(ERROR_LEVEL, ER_DD_UPGRADE_VERSION_NOT_SUPPORTED, actual_version); return true; } uint minor_downgrade_threshold = 0; if (dd::tables::DD_properties::instance().get( thd, "MINOR_DOWNGRADE_THRESHOLD", &minor_downgrade_threshold, &exists) || !exists || minor_downgrade_threshold > dd::DD_VERSION) { LogErr(ERROR_LEVEL, ER_DD_MINOR_DOWNGRADE_VERSION_NOT_SUPPORTED, actual_version); return true; } } } /* purecov: end */ if (dd::tables::DD_properties::instance().get( thd, "MYSQLD_VERSION", &actual_server_version, &exists_server) || !exists_server) return true; if (dd::tables::DD_properties::instance().get( thd, "MYSQLD_VERSION_UPGRADED", &upgraded_server_version, &exists_upgraded_version) || !exists_upgraded_version) upgraded_server_version = actual_server_version; bootstrap::DD_bootstrap_ctx::instance().set_upgraded_server_version( upgraded_server_version); Minor_upgrade_ctx::instance()->retrieve_and_set_extra_mvu_version(thd); if (DBUG_EVALUATE_IF("simulate_mysql_upgrade_skip_pending", true, actual_server_version != upgraded_server_version && actual_server_version != MYSQL_VERSION_ID)) { LogErr(ERROR_LEVEL, ER_SERVER_UPGRADE_PENDING, MYSQL_VERSION_ID, upgraded_server_version); return true; } if (upgraded_server_version != MYSQL_VERSION_ID) { /* This check is also done in DDSE_dict_init() based on the version number from the DD tablespace header. Here, we repeat the check, this time based on the server version number stored in the DD table 'dd_properties'. The two checks should give the same result, so this check should never fail; hence, the debug assert. */ if (!bootstrap::DD_bootstrap_ctx::instance().supported_server_version()) { LogErr(ERROR_LEVEL, ER_SERVER_UPGRADE_VERSION_NOT_SUPPORTED, actual_server_version); DBUG_ASSERT(false); return true; } } /* Reject restarting with a changed LCTN setting, since the collation for LCTN-dependent columns is decided during server initialization. */ uint actual_lctn = 0; exists = false; if (dd::tables::DD_properties::instance().get(thd, "LCTN", &actual_lctn, &exists) || !exists) { LogErr(WARNING_LEVEL, ER_LCTN_NOT_FOUND, lower_case_table_names); } else if (actual_lctn != lower_case_table_names) { LogErr(ERROR_LEVEL, ER_LCTN_CHANGED, lower_case_table_names, actual_lctn); return true; } } if (bootstrap::DD_bootstrap_ctx::instance().is_initialize()) LogErr(INFORMATION_LEVEL, ER_DD_INITIALIZE, dd::DD_VERSION); else if (bootstrap::DD_bootstrap_ctx::instance().is_restart()) LogErr(INFORMATION_LEVEL, ER_DD_RESTART, dd::DD_VERSION); else if (bootstrap::DD_bootstrap_ctx::instance().is_minor_downgrade()) LogErr(INFORMATION_LEVEL, ER_DD_MINOR_DOWNGRADE, actual_version, dd::DD_VERSION); else { /* If none of the above, then this must be DD upgrade or server upgrade, or both. */ if (bootstrap::DD_bootstrap_ctx::instance().is_dd_upgrade()) { LogErr(SYSTEM_LEVEL, ER_DD_UPGRADE, actual_version, dd::DD_VERSION); log_sink_buffer_check_timeout(); sysd::notify("STATUS=Data Dictionary upgrade in progress\n"); } if (bootstrap::DD_bootstrap_ctx::instance().is_server_upgrade()) { // This condition is hit only if upgrade has been skipped before if (opt_upgrade_mode == UPGRADE_NONE) { LogErr(ERROR_LEVEL, ER_SERVER_UPGRADE_OFF); return true; } LogErr(INFORMATION_LEVEL, ER_SERVER_UPGRADE_FROM_VERSION, upgraded_server_version, MYSQL_VERSION_ID); LogErr(INFORMATION_LEVEL, ER_SERVER_UPGRADE_FROM_VERSION, Minor_upgrade_ctx::instance()->get_extra_mvu_version(), Minor_upgrade_ctx::instance()->get_target_extra_mvu_version()); } DBUG_ASSERT(bootstrap::DD_bootstrap_ctx::instance().is_dd_upgrade() || bootstrap::DD_bootstrap_ctx::instance().is_server_upgrade()); } /* Unless this is initialization or restart, we must update the System_tables registry with the information from the 'dd_properties' regarding the actual DD tables. */ if (!bootstrap::DD_bootstrap_ctx::instance().is_initialize() && bootstrap::DD_bootstrap_ctx::instance().is_dd_upgrade() && update_system_tables(thd)) { return true; } bootstrap::DD_bootstrap_ctx::instance().set_stage( bootstrap::Stage::FETCHED_PROPERTIES); return false; } bool is_non_inert_dd_or_ddse_table(System_tables::Types table_type) { return table_type == System_tables::Types::CORE || table_type == System_tables::Types::SECOND || table_type == System_tables::Types::DDSE_PRIVATE || table_type == System_tables::Types::DDSE_PROTECTED; } bool create_tables(THD *thd, const std::set *create_set) { // Turn off FK checks, this is needed since we have cyclic FKs. if (dd::execute_query(thd, "SET FOREIGN_KEY_CHECKS= 0")) return true; /* Decide whether we should create actual or target tables. For plain restart and initialize, we create the target tables. For the second table creation stage during upgrade, we also create target tables. So we create the actual tables only during the first table creation stage for upgrade, and for minor downgrade. */ bool create_target_tables = true; if (bootstrap::DD_bootstrap_ctx::instance().get_stage() == bootstrap::Stage::FETCHED_PROPERTIES && (bootstrap::DD_bootstrap_ctx::instance().is_dd_upgrade() || bootstrap::DD_bootstrap_ctx::instance().is_minor_downgrade())) create_target_tables = false; /* Iterate over DD tables and create the tables. Note that we do not iterate over INERT tables here, there is currently only one INERT table (the 'dd_properties'), and it is created in 'initialize_dd_properties' in order to get hold of e.g. version information. */ bool error = false; for (System_tables::Const_iterator it = System_tables::instance()->begin(); it != System_tables::instance()->end() && !error; ++it) { if (is_non_inert_dd_or_ddse_table((*it)->property())) { /* If a create set is submitted, create only the target tables that are in the create set. */ if (create_set == nullptr || create_set->find((*it)->entity()->name()) != create_set->end()) { /* Use the actual or target definition to create the table depending on the context. */ if (create_target_tables) error = create_target_table(thd, (*it)->entity()); else error = create_actual_table(thd, (*it)->entity()); } } } // Turn FK checks back on. if (error || dd::execute_query(thd, "SET FOREIGN_KEY_CHECKS= 1")) return true; bootstrap::DD_bootstrap_ctx::instance().set_stage( bootstrap::Stage::CREATED_TABLES); return false; } bool sync_meta_data(THD *thd) { // Acquire exclusive meta data locks for the relevant DD objects. if (acquire_exclusive_mdl(thd)) return true; { /* Use a scoped auto releaser to make sure the cached objects are released before the shared cache is reset. */ dd::cache::Dictionary_client::Auto_releaser releaser(thd->dd_client()); /* First, we acquire the DD schema and tablespace and keep them in local variables. The DD table objects are acquired and put into a vector. We also get hold of the corresponding persisted objects. In this way, we make sure the shared cache is populated. The auto releaser will make sure the objects are not evicted. This must be ensured since we need to make sure the ids stay consistent across all objects in the shared cache. */ const Schema *dd_schema = nullptr; const Tablespace *dd_tspace = nullptr; if (thd->dd_client()->acquire(dd::String_type(MYSQL_SCHEMA_NAME.str), &dd_schema) || thd->dd_client()->acquire(dd::String_type(MYSQL_TABLESPACE_NAME.str), &dd_tspace)) return dd::end_transaction(thd, true); std::vector
dd_tables; // Owned by the shared cache. for (System_tables::Const_iterator it = System_tables::instance()->begin(); it != System_tables::instance()->end(); ++it) { // Skip extraneous tables during minor downgrade. if ((*it)->entity() == nullptr) continue; const dd::Table *dd_table = nullptr; if (thd->dd_client()->acquire(MYSQL_SCHEMA_NAME.str, (*it)->entity()->name(), &dd_table)) return dd::end_transaction(thd, true); dd_tables.push_back(const_cast
(dd_table)); } // Get the persisted DD schema and tablespace. Schema::Name_key schema_key; dd_schema->update_name_key(&schema_key); const Schema *tmp_schema = nullptr; Tablespace::Name_key tspace_key; dd_tspace->update_name_key(&tspace_key); const Tablespace *tmp_tspace = nullptr; if (dd::cache::Storage_adapter::instance()->get( thd, schema_key, ISO_READ_COMMITTED, true, &tmp_schema) || dd::cache::Storage_adapter::instance()->get( thd, tspace_key, ISO_READ_COMMITTED, true, &tmp_tspace)) return dd::end_transaction(thd, true); DBUG_ASSERT(tmp_schema != nullptr && tmp_tspace != nullptr); std::unique_ptr persisted_dd_schema( const_cast(tmp_schema)); std::unique_ptr persisted_dd_tspace( const_cast(tmp_tspace)); // If the persisted meta data indicates that the DD tablespace is // encrypted, then we record this fact to make sure the DDL statements // that are genereated during e.g. upgrade will have the correct // encryption option. String_type encryption(""); Object_table_definition_impl::set_dd_tablespace_encrypted( persisted_dd_tspace->options().exists("encryption") && !persisted_dd_tspace->options().get("encryption", &encryption) && encryption == "Y"); // Get the persisted DD table objects into a vector. std::vector> persisted_dd_tables; for (System_tables::Const_iterator it = System_tables::instance()->begin(); it != System_tables::instance()->end(); ++it) { // Skip extraneous tables during minor downgrade. if ((*it)->entity() == nullptr) continue; const dd::Abstract_table *dd_table = nullptr; dd::Abstract_table::Name_key table_key; Abstract_table::update_name_key(&table_key, persisted_dd_schema->id(), (*it)->entity()->name()); if (dd::cache::Storage_adapter::instance()->get( thd, table_key, ISO_READ_COMMITTED, true, &dd_table)) return dd::end_transaction(thd, true); std::unique_ptr persisted_dd_table( dynamic_cast(const_cast(dd_table))); persisted_dd_tables.push_back(std::move(persisted_dd_table)); } // Drop the tablespaces with type PREDEFINED_DDSE from the storage adapter. for (System_tablespaces::Const_iterator it = System_tablespaces::instance()->begin( System_tablespaces::Types::PREDEFINED_DDSE); it != System_tablespaces::instance()->end(); it = System_tablespaces::instance()->next( it, System_tablespaces::Types::PREDEFINED_DDSE)) { const Tablespace *tspace = nullptr; if (thd->dd_client()->acquire((*it)->entity()->get_name(), &tspace)) return dd::end_transaction(thd, true); dd::cache::Storage_adapter::instance()->core_drop(thd, tspace); } /* We have now populated the shared cache with the core objects. The scoped auto releaser makes sure we will not evict the objects from the shared cache until the auto releaser exits scope. Thus, within the scope of the auto releaser, we can modify the contents of the core registry in the storage adapter without risking that this will interfere with the contents of the shared cache, because the DD transactions will acquire the core objects from the shared cache. */ /* We have also read the DD schema and tablespace as well as the DD tables from persistent storage. The last thing we do before resetting the shared cache is to update the contents of the core registry to match the persisted objects. First, we update the core registry with the persisted DD schema and tablespace. */ dd::cache::Storage_adapter::instance()->core_drop(thd, dd_schema); dd::cache::Storage_adapter::instance()->core_store( thd, persisted_dd_schema.get()); dd::cache::Storage_adapter::instance()->core_drop(thd, dd_tspace); dd::cache::Storage_adapter::instance()->core_store( thd, persisted_dd_tspace.get()); // Make sure the IDs after storing are as expected. DBUG_ASSERT(persisted_dd_schema->id() == 1); DBUG_ASSERT(persisted_dd_tspace->id() == 1); /* Finally, we update the core registry of the DD tables. This must be done in two loops to avoid issues related to overlapping ID sequences. */ std::vector
::const_iterator table_it = dd_tables.begin(); for (System_tables::Const_iterator it = System_tables::instance()->begin(); it != System_tables::instance()->end() && table_it != dd_tables.end(); ++it, ++table_it) { /* If we are in the process of upgrading, there may not be an entry in the dd_tables for new tables that have been added after the version we are upgrading from. */ if ((*table_it) != nullptr) { DBUG_ASSERT((*it)->entity()->name() == (*table_it)->name()); dd::cache::Storage_adapter::instance()->core_drop(thd, *table_it); } } std::vector>::const_iterator persisted_it = persisted_dd_tables.begin(); for (System_tables::Const_iterator it = System_tables::instance()->begin(); it != System_tables::instance()->end() && persisted_it != persisted_dd_tables.end(); ++it, ++persisted_it) { /* If we are in the process of upgrading, there may not be an entry in the persisted_dd_tables for new tables that have been added after the version we are upgrading from. */ if ((*persisted_it) == nullptr) continue; if ((*it)->property() == System_tables::Types::CORE) { dd::cache::Storage_adapter::instance()->core_store( thd, static_cast
((*persisted_it).get())); } } } /* Now, the auto releaser has released the objects, and we can go ahead and reset the shared cache. */ dd::cache::Shared_dictionary_cache::instance()->reset(true); bootstrap::DD_bootstrap_ctx::instance().set_stage(bootstrap::Stage::SYNCED); // Commit and flush tables to force re-opening using the refreshed meta data. if (dd::end_transaction(thd, false) || dd::execute_query(thd, "FLUSH TABLES")) return true; // Get hold of the temporary actual and target schema names. String_type target_schema_name; bool target_schema_exists = false; if (dd::tables::DD_properties::instance().get(thd, "UPGRADE_TARGET_SCHEMA", &target_schema_name, &target_schema_exists)) return true; String_type actual_schema_name; bool actual_schema_exists = false; if (dd::tables::DD_properties::instance().get(thd, "UPGRADE_ACTUAL_SCHEMA", &actual_schema_name, &actual_schema_exists)) return true; // Reset the DDSE local dictionary cache. handlerton *ddse = ha_resolve_by_legacy_type(thd, DB_TYPE_INNODB); if (ddse->dict_cache_reset == nullptr) return true; for (System_tables::Const_iterator it = System_tables::instance()->begin(); it != System_tables::instance()->end(); ++it) { // Skip extraneous tables during minor downgrade. if ((*it)->entity() == nullptr) continue; if ((*it)->property() == System_tables::Types::CORE || (*it)->property() == System_tables::Types::SECOND) { ddse->dict_cache_reset(MYSQL_SCHEMA_NAME.str, (*it)->entity()->name().c_str()); if (target_schema_exists && !target_schema_name.empty()) ddse->dict_cache_reset(target_schema_name.c_str(), (*it)->entity()->name().c_str()); if (actual_schema_exists && !actual_schema_name.empty()) ddse->dict_cache_reset(actual_schema_name.c_str(), (*it)->entity()->name().c_str()); } } /* At this point, we're to a large extent open for business. If there are leftover schema names from upgrade, delete them and remove the names from the DD properties. */ if (target_schema_exists && !target_schema_name.empty()) { std::stringstream ss; ss << "DROP SCHEMA IF EXISTS " << target_schema_name; if (dd::execute_query(thd, ss.str().c_str())) return true; } if (actual_schema_exists && !actual_schema_name.empty()) { std::stringstream ss; ss << "DROP SCHEMA IF EXISTS " << actual_schema_name; if (dd::execute_query(thd, ss.str().c_str())) return true; } /* The statements above are auto committed, so there is nothing uncommitted at this stage. Go ahead and remove the schema keys. */ if (actual_schema_exists) (void)dd::tables::DD_properties::instance().remove(thd, "UPGRADE_ACTUAL_SCHEMA"); if (target_schema_exists) (void)dd::tables::DD_properties::instance().remove(thd, "UPGRADE_TARGET_SCHEMA"); if (actual_schema_exists || target_schema_exists) return end_transaction(thd, false); return false; } bool update_properties(THD *thd, const std::set *create_set, const std::set *remove_set, const String_type &target_table_schema_name) { /* Populate the dd properties with the SQL DDL and SE private data. Store meta data of non-inert tables only. */ std::unique_ptr system_tables_props( dd::Properties::parse_properties("")); dd::cache::Dictionary_client::Auto_releaser releaser(thd->dd_client()); for (System_tables::Const_iterator it = System_tables::instance()->begin(); it != System_tables::instance()->end(); ++it) { if (is_non_inert_dd_or_ddse_table((*it)->property())) { /* This will not be called for minor downgrade, so all tables will have a corresponding Object_table. */ DBUG_ASSERT((*it)->entity() != nullptr); const Object_table_definition *table_def = (*it)->entity()->target_table_definition(); // May be null for abandoned tables, which should be skipped. if (table_def == nullptr) { continue; } /* Tables that are in the remove_set, but not in the create_set, should not be reflected in the DD properties. */ if (remove_set != nullptr && create_set != nullptr && remove_set->find((*it)->entity()->name()) != remove_set->end() && create_set->find((*it)->entity()->name()) == create_set->end()) { continue; } /* If a create set is submitted, use this to decide whether we should get the meta data from the table in the 'mysql' schema or the temporary target schema. */ String_type table_schema_name{MYSQL_SCHEMA_NAME.str}; if (create_set != nullptr && create_set->find((*it)->entity()->name()) != create_set->end()) { table_schema_name = target_table_schema_name; } /* Acquire the table object to get hold of the se private data etc. Note that we must acquire it from the appropriate schema. */ const dd::Table *dd_table = nullptr; if (thd->dd_client()->acquire(table_schema_name, (*it)->entity()->name(), &dd_table)) return dd::end_transaction(thd, true); // All non-abandoned tables should have a table object present. DBUG_ASSERT(dd_table != nullptr); std::unique_ptr tbl_props( dd::Properties::parse_properties("")); using dd::tables::DD_properties; tbl_props->set(DD_properties::dd_key(DD_properties::DD_property::ID), dd_table->se_private_id()); tbl_props->set(DD_properties::dd_key(DD_properties::DD_property::DATA), dd_table->se_private_data().raw_string()); tbl_props->set( DD_properties::dd_key(DD_properties::DD_property::SPACE_ID), dd_table->tablespace_id()); // Store the structured representation of the table definition. std::unique_ptr definition(Properties::parse_properties("")); table_def->store_into_properties(definition.get()); tbl_props->set(DD_properties::dd_key(DD_properties::DD_property::DEF), definition->raw_string()); // Store the se private data for each index. dd::Table::Index_collection::const_iterator idx( dd_table->indexes().begin()); for (int count = 0; idx != dd_table->indexes().end(); ++idx, ++count) { std::stringstream ss; ss << DD_properties::dd_key(DD_properties::DD_property::IDX) << count; tbl_props->set(ss.str().c_str(), (*idx)->se_private_data().raw_string()); } // Store the se private data for each column. dd::Table::Column_collection::const_iterator col( dd_table->columns().begin()); for (int count = 0; col != dd_table->columns().end(); ++col, ++count) { std::stringstream ss; ss << DD_properties::dd_key(DD_properties::DD_property::COL) << count; tbl_props->set(ss.str().c_str(), (*col)->se_private_data().raw_string()); } // All tables should be reflected in the System tables list. system_tables_props->set(dd_table->name(), tbl_props->raw_string()); } } if (dd::tables::DD_properties::instance().set(thd, "SYSTEM_TABLES", *system_tables_props.get())) return dd::end_transaction(thd, true); bootstrap::DD_bootstrap_ctx::instance().set_stage( bootstrap::Stage::STORED_DD_META_DATA); // Delay commit. return false; } bool update_versions(THD *thd, bool is_dd_upgrade_57) { /* During initialize, store the DD version number, the LCTN used, and the mysqld server version. */ if (opt_initialize) { if (dd::tables::DD_properties::instance().set(thd, "DD_VERSION", dd::DD_VERSION) || dd::tables::DD_properties::instance().set( thd, "MINOR_DOWNGRADE_THRESHOLD", dd::DD_VERSION_MINOR_DOWNGRADE_THRESHOLD) || dd::tables::DD_properties::instance().set(thd, "SDI_VERSION", dd::SDI_VERSION) || dd::tables::DD_properties::instance().set(thd, "LCTN", lower_case_table_names) || dd::tables::DD_properties::instance().set(thd, "MYSQLD_VERSION_LO", MYSQL_VERSION_ID) || dd::tables::DD_properties::instance().set(thd, "MYSQLD_VERSION_HI", MYSQL_VERSION_ID) || dd::tables::DD_properties::instance().set(thd, "MYSQLD_VERSION", MYSQL_VERSION_ID)) return dd::end_transaction(thd, true); if (is_dd_upgrade_57) { if (dd::tables::DD_properties::instance().set( thd, "MYSQLD_VERSION_UPGRADED", bootstrap::SERVER_VERSION_50700)) return true; bootstrap::DD_bootstrap_ctx::instance().set_upgraded_server_version( bootstrap::SERVER_VERSION_50700); } else { if (dd::tables::DD_properties::instance().set( thd, "MYSQLD_VERSION_UPGRADED", MYSQL_VERSION_ID)) return true; bootstrap::DD_bootstrap_ctx::instance().set_upgraded_server_version( MYSQL_VERSION_ID); if (Minor_upgrade_ctx::instance()->save_and_set_extra_mvu_version( thd, Minor_upgrade_ctx::instance()->get_target_extra_mvu_version())) return true; } } else { uint mysqld_version_lo = 0; uint mysqld_version_hi = 0; uint mysqld_version = 0; uint upgraded_server_version = 0; bool exists_lo = false; bool exists_hi = false; bool exists = false; bool exists_upgraded_version = false; if ((dd::tables::DD_properties::instance().get( thd, "MYSQLD_VERSION_LO", &mysqld_version_lo, &exists_lo) || !exists_lo) || (dd::tables::DD_properties::instance().get( thd, "MYSQLD_VERSION_HI", &mysqld_version_hi, &exists_hi) || !exists_hi) || (dd::tables::DD_properties::instance().get(thd, "MYSQLD_VERSION", &mysqld_version, &exists) || !exists)) return dd::end_transaction(thd, true); if (dd::tables::DD_properties::instance().get( thd, "MYSQLD_VERSION_UPGRADED", &upgraded_server_version, &exists_upgraded_version) || !exists_upgraded_version) { if (dd::tables::DD_properties::instance().set( thd, "MYSQLD_VERSION_UPGRADED", mysqld_version)) return true; upgraded_server_version = mysqld_version; } bootstrap::DD_bootstrap_ctx::instance().set_upgraded_server_version( upgraded_server_version); Minor_upgrade_ctx::instance()->retrieve_and_set_extra_mvu_version(thd); if ((mysqld_version_lo > MYSQL_VERSION_ID && dd::tables::DD_properties::instance().set(thd, "MYSQLD_VERSION_LO", MYSQL_VERSION_ID)) || (mysqld_version_hi < MYSQL_VERSION_ID && dd::tables::DD_properties::instance().set(thd, "MYSQLD_VERSION_HI", MYSQL_VERSION_ID)) || (mysqld_version != MYSQL_VERSION_ID && dd::tables::DD_properties::instance().set(thd, "MYSQLD_VERSION", MYSQL_VERSION_ID))) return dd::end_transaction(thd, true); /* Update the SDI version number in case of upgrade. Note that on downgrade, we keep the old SDI version. */ uint stored_sdi_version = 0; bool exists_sdi = false; if ((dd::tables::DD_properties::instance().get( thd, "SDI_VERSION", &stored_sdi_version, &exists_sdi) || !exists_sdi) || (stored_sdi_version < dd::SDI_VERSION && dd::tables::DD_properties::instance().set(thd, "SDI_VERSION", dd::SDI_VERSION))) return dd::end_transaction(thd, true); /* Update the DD version number in case of upgrade. Note that on downgrade, we keep the old DD version. */ uint dd_version = 0; bool exists_dd = false; if ((dd::tables::DD_properties::instance().get(thd, "DD_VERSION", &dd_version, &exists_dd) || !exists_dd) || (dd_version < dd::DD_VERSION && dd::tables::DD_properties::instance().set(thd, "DD_VERSION", dd::DD_VERSION))) return dd::end_transaction(thd, true); /* Update the minor downgrade threshold in case of upgrade. Note that on downgrade, we keep the threshold version which is already present. */ if (dd_version < dd::DD_VERSION && dd::tables::DD_properties::instance().set( thd, "MINOR_DOWNGRADE_THRESHOLD", dd::DD_VERSION_MINOR_DOWNGRADE_THRESHOLD)) return dd::end_transaction(thd, true); } /* Update the server version number in the bootstrap ctx and the DD tablespace header if we have been doing a server upgrade. Note that the update of the tablespace header is not rolled back in case of an abort, so this better be the last step we do before committing. */ handlerton *ddse = ha_resolve_by_legacy_type(thd, DB_TYPE_INNODB); if (bootstrap::DD_bootstrap_ctx::instance().is_server_upgrade()) { if (ddse->dict_set_server_version == nullptr || ddse->dict_set_server_version()) { LogErr(ERROR_LEVEL, ER_CANNOT_SET_SERVER_VERSION_IN_TABLESPACE_HEADER); return dd::end_transaction(thd, true); } } #ifndef DBUG_OFF /* Debug code to make sure that after updating version numbers, regardless of the type of initialization, restart or upgrade, the server version number in the DD tablespace header is indeed the same as this server's version number. */ uint version = 0; DBUG_ASSERT(ddse->dict_get_server_version != nullptr); DBUG_ASSERT(!ddse->dict_get_server_version(&version)); DBUG_ASSERT(version == MYSQL_VERSION_ID); #endif bootstrap::DD_bootstrap_ctx::instance().set_stage( bootstrap::Stage::VERSION_UPDATED); /* During upgrade, this will commit the swap of the old and new DD tables. */ return dd::end_transaction(thd, false); } } // namespace dd