handler.cc

/* Copyright (c) 2000, 2020, Oracle and/or its affiliates.

   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 */

/** @file sql/handler.cc

    @brief
    Implements functions in the handler interface that are shared between all
    storage engines.
*/

#include "sql/handler.h"

#include <ctype.h>
#include <errno.h>
#include <limits.h>
#include <stdio.h>
#include <stdlib.h>
#include <algorithm>
#include <atomic>
#include <cmath>
#include <list>
#include <random>  // std::uniform_real_distribution
#include <string>
#include <vector>

#include "keycache.h"
#include "libbinlogevents/include/binlog_event.h"
#include "m_ctype.h"
#include "m_string.h"
#include "my_bit.h"     // my_count_bits
#include "my_bitmap.h"  // MY_BITMAP
#include "my_check_opt.h"
#include "my_dbug.h"
#include "my_loglevel.h"
#include "my_macros.h"
#include "my_pointer_arithmetic.h"
#include "my_psi_config.h"
#include "my_sqlcommand.h"
#include "my_sys.h"  // MEM_DEFINED_IF_ADDRESSABLE()
#include "myisam.h"  // TT_FOR_UPGRADE
#include "mysql/components/services/log_builtins.h"
#include "mysql/components/services/log_shared.h"
#include "mysql/plugin.h"
#include "mysql/psi/mysql_file.h"
#include "mysql/psi/mysql_mutex.h"
#include "mysql/psi/mysql_table.h"
#include "mysql/psi/mysql_transaction.h"
#include "mysql/psi/psi_base.h"
#include "mysql/psi/psi_table.h"
#include "mysql/service_mysql_alloc.h"
#include "mysql_com.h"
#include "mysql_version.h"  // MYSQL_VERSION_ID
#include "mysqld_error.h"
#include "prealloced_array.h"
#include "sql/auth/auth_common.h"  // check_readonly() and SUPER_ACL
#include "sql/binlog.h"            // mysql_bin_log
#include "sql/check_stack.h"
#include "sql/clone_handler.h"
#include "sql/current_thd.h"
#include "sql/dd/cache/dictionary_client.h"  // dd::cache::Dictionary_client
#include "sql/dd/dd.h"                       // dd::get_dictionary
#include "sql/dd/dictionary.h"               // dd:acquire_shared_table_mdl
#include "sql/dd/types/table.h"              // dd::Table
#include "sql/dd_table_share.h"              // open_table_def
#include "sql/debug_sync.h"                  // DEBUG_SYNC
#include "sql/derror.h"                      // ER_DEFAULT
#include "sql/error_handler.h"               // Internal_error_handler
#include "sql/field.h"
#include "sql/item.h"
#include "sql/lock.h"  // MYSQL_LOCK
#include "sql/log.h"
#include "sql/log_event.h"  // Write_rows_log_event
#include "sql/mdl.h"
#include "sql/mysqld.h"                 // global_system_variables heap_hton ..
#include "sql/opt_costconstantcache.h"  // reload_optimizer_cost_constants
#include "sql/opt_costmodel.h"
#include "sql/opt_hints.h"
#include "sql/protocol.h"
#include "sql/psi_memory_key.h"
#include "sql/query_options.h"
#include "sql/record_buffer.h"  // Record_buffer
#include "sql/rpl_filter.h"
#include "sql/rpl_gtid.h"
#include "sql/rpl_handler.h"  // RUN_HOOK
#include "sql/rpl_rli.h"      // is_atomic_ddl_commit_on_slave
#include "sql/rpl_slave_commit_order_manager.h"  // Commit_order_manager
#include "sql/rpl_write_set_handler.h"           // add_pke
#include "sql/sdi_utils.h"                       // import_serialized_meta_data
#include "sql/session_tracker.h"
#include "sql/sql_base.h"  // free_io_cache
#include "sql/sql_bitmap.h"
#include "sql/sql_class.h"
#include "sql/sql_error.h"
#include "sql/sql_lex.h"
#include "sql/sql_parse.h"   // check_stack_overrun
#include "sql/sql_plugin.h"  // plugin_foreach
#include "sql/sql_select.h"  // actual_key_parts
#include "sql/sql_table.h"   // build_table_filename
#include "sql/strfunc.h"     // strnncmp_nopads
#include "sql/system_variables.h"
#include "sql/table.h"
#include "sql/tc_log.h"
#include "sql/thr_malloc.h"
#include "sql/transaction.h"  // trans_commit_implicit
#include "sql/transaction_info.h"
#include "sql/xa.h"
#include "sql_string.h"
#include "sql_tmp_table.h"  // free_tmp_table
#include "template_utils.h"
#include "uniques.h"  // Unique_on_insert
#include "varlen_sort.h"

/**
  @def MYSQL_TABLE_IO_WAIT
  Instrumentation helper for table io_waits.
  Note that this helper is intended to be used from
  within the handler class only, as it uses members
  from @c handler
  Performance schema events are instrumented as follows:
  - in non batch mode, one event is generated per call
  - in batch mode, the number of rows affected is saved
  in @c m_psi_numrows, so that @c end_psi_batch_mode()
  generates a single event for the batch.
  @param OP the table operation to be performed
  @param INDEX the table index used if any, or MAX_KEY.
  @param RESULT the result of the table operation performed
  @param PAYLOAD instrumented code to execute
  @sa handler::end_psi_batch_mode.
*/
#ifdef HAVE_PSI_TABLE_INTERFACE
#define MYSQL_TABLE_IO_WAIT(OP, INDEX, RESULT, PAYLOAD)                     \
  {                                                                         \
    if (m_psi != NULL) {                                                    \
      switch (m_psi_batch_mode) {                                           \
        case PSI_BATCH_MODE_NONE: {                                         \
          PSI_table_locker *sub_locker = NULL;                              \
          PSI_table_locker_state reentrant_safe_state;                      \
          sub_locker = PSI_TABLE_CALL(start_table_io_wait)(                 \
              &reentrant_safe_state, m_psi, OP, INDEX, __FILE__, __LINE__); \
          PAYLOAD                                                           \
          if (sub_locker != NULL) PSI_TABLE_CALL(end_table_io_wait)         \
          (sub_locker, 1);                                                  \
          break;                                                            \
        }                                                                   \
        case PSI_BATCH_MODE_STARTING: {                                     \
          m_psi_locker = PSI_TABLE_CALL(start_table_io_wait)(               \
              &m_psi_locker_state, m_psi, OP, INDEX, __FILE__, __LINE__);   \
          PAYLOAD                                                           \
          if (!RESULT) m_psi_numrows++;                                     \
          m_psi_batch_mode = PSI_BATCH_MODE_STARTED;                        \
          break;                                                            \
        }                                                                   \
        case PSI_BATCH_MODE_STARTED:                                        \
        default: {                                                          \
          DBUG_ASSERT(m_psi_batch_mode == PSI_BATCH_MODE_STARTED);          \
          PAYLOAD                                                           \
          if (!RESULT) m_psi_numrows++;                                     \
          break;                                                            \
        }                                                                   \
      }                                                                     \
    } else {                                                                \
      PAYLOAD                                                               \
    }                                                                       \
  }
#else
#define MYSQL_TABLE_IO_WAIT(OP, INDEX, RESULT, PAYLOAD) PAYLOAD
#endif

/**
  @def MYSQL_TABLE_LOCK_WAIT
  Instrumentation helper for table io_waits.
  @param OP the table operation to be performed
  @param FLAGS per table operation flags.
  @param PAYLOAD the code to instrument.
  @sa MYSQL_END_TABLE_WAIT.
*/
#ifdef HAVE_PSI_TABLE_INTERFACE
#define MYSQL_TABLE_LOCK_WAIT(OP, FLAGS, PAYLOAD)                              \
  {                                                                            \
    if (m_psi != NULL) {                                                       \
      PSI_table_locker *locker;                                                \
      PSI_table_locker_state state;                                            \
      locker = PSI_TABLE_CALL(start_table_lock_wait)(&state, m_psi, OP, FLAGS, \
                                                     __FILE__, __LINE__);      \
      PAYLOAD                                                                  \
      if (locker != NULL) PSI_TABLE_CALL(end_table_lock_wait)(locker);         \
    } else {                                                                   \
      PAYLOAD                                                                  \
    }                                                                          \
  }
#else
#define MYSQL_TABLE_LOCK_WAIT(OP, FLAGS, PAYLOAD) PAYLOAD
#endif

using std::list;
using std::log2;
using std::max;
using std::min;

/**
  While we have legacy_db_type, we have this array to
  check for dups and to find handlerton from legacy_db_type.
  Remove when legacy_db_type is finally gone
*/
static Prealloced_array<st_plugin_int *, PREALLOC_NUM_HA> se_plugin_array(
    PSI_NOT_INSTRUMENTED);

/**
  Array allowing to check if handlerton is builtin without
  acquiring LOCK_plugin.
*/
static Prealloced_array<bool, PREALLOC_NUM_HA> builtin_htons(
    PSI_NOT_INSTRUMENTED);

st_plugin_int *hton2plugin(uint slot) { return se_plugin_array[slot]; }

size_t num_hton2plugins() { return se_plugin_array.size(); }

st_plugin_int *insert_hton2plugin(uint slot, st_plugin_int *plugin) {
  if (se_plugin_array.assign_at(slot, plugin)) return nullptr;
  builtin_htons.assign_at(slot, true);
  return se_plugin_array[slot];
}

st_plugin_int *remove_hton2plugin(uint slot) {
  st_plugin_int *retval = se_plugin_array[slot];
  se_plugin_array[slot] = NULL;
  builtin_htons.assign_at(slot, false);
  return retval;
}

const char *ha_resolve_storage_engine_name(const handlerton *db_type) {
  return db_type == nullptr ? "UNKNOWN" : hton2plugin(db_type->slot)->name.str;
}

static handlerton *installed_htons[128];

/* number of storage engines (from installed_htons[]) that support 2pc */
ulong total_ha_2pc = 0;
/* size of savepoint storage area (see ha_init) */
ulong savepoint_alloc_size = 0;

namespace {
struct Storage_engine_identifier {
  const LEX_CSTRING canonical;
  const LEX_CSTRING legacy;
};
const Storage_engine_identifier se_names[] = {
    {{STRING_WITH_LEN("INNODB")}, {STRING_WITH_LEN("INNOBASE")}},
    {{STRING_WITH_LEN("NDBCLUSTER")}, {STRING_WITH_LEN("NDB")}},
    {{STRING_WITH_LEN("MEMORY")}, {STRING_WITH_LEN("HEAP")}},
    {{STRING_WITH_LEN("MRG_MYISAM")}, {STRING_WITH_LEN("MERGE")}}};
const auto se_names_end = std::end(se_names);
std::vector<std::string> disabled_se_names;
}  // namespace

const char *ha_row_type[] = {"",
                             "FIXED",
                             "DYNAMIC",
                             "COMPRESSED",
                             "REDUNDANT",
                             "COMPACT",
                             /* Reserved to be "PAGE" in future versions */ "?",
                             "?",
                             "?",
                             "?"};

const char *tx_isolation_names[] = {"READ-UNCOMMITTED", "READ-COMMITTED",
                                    "REPEATABLE-READ", "SERIALIZABLE", NullS};
TYPELIB tx_isolation_typelib = {array_elements(tx_isolation_names) - 1, "",
                                tx_isolation_names, nullptr};

// Called for each SE to check if given db.table_name is a system table.
static bool check_engine_system_table_handlerton(THD *unused, plugin_ref plugin,
                                                 void *arg);

static int ha_discover(THD *thd, const char *db, const char *name,
                       uchar **frmblob, size_t *frmlen);

/**
  Structure used by SE during check for system table.
  This structure is passed to each SE handlerton and the status (OUT param)
  is collected.
*/
struct st_sys_tbl_chk_params {
  const char *db;                  // IN param
  const char *table_name;          // IN param
  bool is_sql_layer_system_table;  // IN param
  legacy_db_type db_type;          // IN param

  enum enum_sys_tbl_chk_status {
    // db.table_name is not a supported system table.
    NOT_KNOWN_SYSTEM_TABLE,
    /*
      db.table_name is a system table,
      but may not be supported by SE.
    */
    KNOWN_SYSTEM_TABLE,
    /*
      db.table_name is a system table,
      and is supported by SE.
    */
    SUPPORTED_SYSTEM_TABLE
  } status;  // OUT param
};

static plugin_ref ha_default_plugin(THD *thd) {
  if (thd->variables.table_plugin) return thd->variables.table_plugin;
  return my_plugin_lock(thd, &global_system_variables.table_plugin);
}

/** @brief
  Return the default storage engine handlerton used for non-temp tables
  for thread

  SYNOPSIS
    ha_default_handlerton(thd)
    thd         current thread

  RETURN
    pointer to handlerton
*/
handlerton *ha_default_handlerton(THD *thd) {
  plugin_ref plugin = ha_default_plugin(thd);
  DBUG_ASSERT(plugin);
  handlerton *hton = plugin_data<handlerton *>(plugin);
  DBUG_ASSERT(hton);
  return hton;
}

static plugin_ref ha_default_temp_plugin(THD *thd) {
  if (thd->variables.temp_table_plugin) return thd->variables.temp_table_plugin;
  return my_plugin_lock(thd, &global_system_variables.temp_table_plugin);
}

/** @brief
  Return the default storage engine handlerton used for explicitly
  created temp tables for a thread

  SYNOPSIS
    ha_default_temp_handlerton(thd)
    thd         current thread

  RETURN
    pointer to handlerton
*/
handlerton *ha_default_temp_handlerton(THD *thd) {
  plugin_ref plugin = ha_default_temp_plugin(thd);
  DBUG_ASSERT(plugin);
  handlerton *hton = plugin_data<handlerton *>(plugin);
  DBUG_ASSERT(hton);
  return hton;
}

/**
  Resolve handlerton plugin by name, without checking for "DEFAULT" or
  HTON_NOT_USER_SELECTABLE.

  @param thd  Thread context.
  @param name Plugin name.

  @return plugin or NULL if not found.
*/
plugin_ref ha_resolve_by_name_raw(THD *thd, const LEX_CSTRING &name) {
  return plugin_lock_by_name(thd, name, MYSQL_STORAGE_ENGINE_PLUGIN);
}

static const CHARSET_INFO &hton_charset() { return *system_charset_info; }

/**
  Return the storage engine handlerton for the supplied name.

  @param thd           Current thread. May be nullptr, (e.g. during initialize).
  @param name          Name of storage engine.
  @param is_temp_table true if table is a temporary table.

  @return Pointer to storage engine plugin handle.
*/
plugin_ref ha_resolve_by_name(THD *thd, const LEX_CSTRING *name,
                              bool is_temp_table) {
  if (thd && 0 == strnncmp_nopads(hton_charset(), *name,
                                  {STRING_WITH_LEN("DEFAULT")})) {
    return is_temp_table ? ha_default_plugin(thd) : ha_default_temp_plugin(thd);
  }

  // Note that thd CAN be nullptr here - it is not actually needed by
  // ha_resolve_by_name_raw().
  plugin_ref plugin = ha_resolve_by_name_raw(thd, *name);
  if (plugin == nullptr) {
    // If we fail to resolve the name passed in, we try to see if it is a
    // historical alias.
    auto match = std::find_if(
        std::begin(se_names), se_names_end,
        [&](const Storage_engine_identifier &sei) {
          return (0 == strnncmp_nopads(hton_charset(), *name, sei.legacy));
        });
    if (match != se_names_end) {
      // if it is, we resolve using the new name
      plugin = ha_resolve_by_name_raw(thd, match->canonical);
    }
  }
  if (plugin != nullptr) {
    handlerton *hton = plugin_data<handlerton *>(plugin);
    if (hton && !(hton->flags & HTON_NOT_USER_SELECTABLE)) return plugin;

    /*
      unlocking plugin immediately after locking is relatively low cost.
    */
    plugin_unlock(thd, plugin);
  }
  return nullptr;
}

/**
  Read a comma-separated list of storage engine names. Look up each in the
  known list of canonical and legacy names. In case of a match; add both the
  canonical and the legacy name to disabled_se_names, which is a static vector
  of disabled storage engine names.
  If there is no match, the unmodified name is added to the vector.
*/
void set_externally_disabled_storage_engine_names(const char *disabled_list) {
  DBUG_ASSERT(disabled_list != nullptr);

  myu::Split(
      disabled_list, disabled_list + strlen(disabled_list), myu::IsComma,
      [](const char *f, const char *l) {
        auto tr = myu::FindTrimmedRange(f, l, myu::IsSpace);
        if (tr.first == tr.second) return;

        const LEX_CSTRING dse{tr.first,
                              static_cast<size_t>(tr.second - tr.first)};
        auto match = std::find_if(
            std::begin(se_names), se_names_end,
            [&](const Storage_engine_identifier &seid) {
              return (
                  (0 == strnncmp_nopads(hton_charset(), dse, seid.canonical)) ||
                  (0 == strnncmp_nopads(hton_charset(), dse, seid.legacy)));
            });
        if (match == se_names_end) {
          disabled_se_names.emplace_back(dse.str, dse.length);
          return;
        }
        disabled_se_names.emplace_back(match->canonical.str,
                                       match->canonical.length);
        disabled_se_names.emplace_back(match->legacy.str, match->legacy.length);
      });
}

static bool is_storage_engine_name_externally_disabled(const char *name) {
  const LEX_CSTRING n{name, strlen(name)};
  return std::any_of(
      disabled_se_names.begin(), disabled_se_names.end(),
      [&](const std::string &dse) {
        return (0 == strnncmp_nopads(hton_charset(), n,
                                     {dse.c_str(), dse.length()}));
      });
}

/**
  Returns true if the storage engine of the handlerton argument has
  been listed in the disabled_storage_engines system variable. @note
  that the SE may still be internally enabled, that is
  HaIsInternallyEnabled may return true.
 */
bool ha_is_externally_disabled(const handlerton &htnr) {
  const char *se_name = ha_resolve_storage_engine_name(&htnr);
  DBUG_ASSERT(se_name != nullptr);
  return is_storage_engine_name_externally_disabled(se_name);
}

// Check if storage engine is disabled for table/tablespace creation.
bool ha_is_storage_engine_disabled(handlerton *se_handle) {
  DBUG_ASSERT(se_handle != nullptr);
  return ha_is_externally_disabled(*se_handle);
}

plugin_ref ha_lock_engine(THD *thd, const handlerton *hton) {
  if (hton) {
    st_plugin_int **plugin = &se_plugin_array[hton->slot];

#ifdef DBUG_OFF
    /*
      Take a shortcut for builtin engines -- return pointer to plugin
      without acquiring LOCK_plugin mutex. This is safe safe since such
      plugins are not deleted until shutdown and we don't do reference
      counting in non-debug builds for them.

      Since we have reference to handlerton on our hands, this method
      can't be called concurrently to non-builtin handlerton initialization/
      deinitialization. So it is safe to access builtin_htons[] without
      additional locking.
     */
    if (builtin_htons[hton->slot]) return *plugin;

    return my_plugin_lock(thd, plugin);
#else
    /*
      We can't take shortcut in debug builds.
      At least assert that builtin_htons[slot] is set correctly.
    */
    DBUG_ASSERT(builtin_htons[hton->slot] == (plugin[0]->plugin_dl == nullptr));
    return my_plugin_lock(thd, &plugin);
#endif
  }
  return nullptr;
}

handlerton *ha_resolve_by_legacy_type(THD *thd, enum legacy_db_type db_type) {
  plugin_ref plugin;
  switch (db_type) {
    case DB_TYPE_DEFAULT:
      return ha_default_handlerton(thd);
    default:
      if (db_type > DB_TYPE_UNKNOWN && db_type < DB_TYPE_DEFAULT &&
          (plugin = ha_lock_engine(thd, installed_htons[db_type])))
        return plugin_data<handlerton *>(plugin);
      /* fall through */
    case DB_TYPE_UNKNOWN:
      return nullptr;
  }
}

/**
  Use other database handler if databasehandler is not compiled in.
*/
handlerton *ha_checktype(THD *thd, enum legacy_db_type database_type,
                         bool no_substitute, bool report_error) {
  DBUG_TRACE;
  handlerton *hton = ha_resolve_by_legacy_type(thd, database_type);
  if (ha_storage_engine_is_enabled(hton)) return hton;

  if (no_substitute) {
    if (report_error) {
      const char *engine_name = ha_resolve_storage_engine_name(hton);
      my_error(ER_FEATURE_DISABLED, MYF(0), engine_name, engine_name);
    }
    return nullptr;
  }

  (void)RUN_HOOK(transaction, after_rollback, (thd, false));

  switch (database_type) {
    case DB_TYPE_MRG_ISAM:
      return ha_resolve_by_legacy_type(thd, DB_TYPE_MRG_MYISAM);
    default:
      break;
  }

  return ha_default_handlerton(thd);
} /* ha_checktype */

/**
  Create handler object for the table in the storage engine.

  @param share        TABLE_SHARE for the table, can be NULL if caller
                      didn't perform full-blown open of table definition.
  @param partitioned  Indicates whether table is partitioned.
  @param alloc        Memory root to be used for allocating handler object.
  @param db_type      Table's storage engine.

  @note This function will try to use default storage engine if one which
        was specified through db_type parameter is not available.
*/
handler *get_new_handler(TABLE_SHARE *share, bool partitioned, MEM_ROOT *alloc,
                         handlerton *db_type) {
  handler *file;
  DBUG_TRACE;
  DBUG_PRINT("enter", ("alloc: %p", alloc));

  if (db_type && db_type->state == SHOW_OPTION_YES && db_type->create) {
    if ((file = db_type->create(db_type, share, partitioned, alloc)))
      file->init();
    return file;
  }
  /*
    Try the default table type
    Here the call to current_thd() is ok as we call this function a lot of
    times but we enter this branch very seldom.
  */
  return get_new_handler(share, partitioned, alloc,
                         ha_default_handlerton(current_thd));
}

static const char **handler_errmsgs;

static const char *get_handler_errmsg(int nr) {
  return handler_errmsgs[nr - HA_ERR_FIRST];
}

/**
  Register handler error messages for use with my_error().

  @retval
    0           OK
  @retval
    !=0         Error
*/

int ha_init_errors(void) {
#define SETMSG(nr, msg) handler_errmsgs[(nr)-HA_ERR_FIRST] = (msg)

  /* Allocate a pointer array for the error message strings. */
  /* Zerofill it to avoid uninitialized gaps. */
  if (!(handler_errmsgs = (const char **)my_malloc(
            key_memory_handler_errmsgs, HA_ERR_ERRORS * sizeof(char *),
            MYF(MY_WME | MY_ZEROFILL))))
    return 1;

  /* Set the dedicated error messages. */
  SETMSG(HA_ERR_KEY_NOT_FOUND, ER_DEFAULT(ER_KEY_NOT_FOUND));
  SETMSG(HA_ERR_FOUND_DUPP_KEY, ER_DEFAULT(ER_DUP_KEY));
  SETMSG(HA_ERR_RECORD_CHANGED, "Update wich is recoverable");
  SETMSG(HA_ERR_WRONG_INDEX, "Wrong index given to function");
  SETMSG(HA_ERR_CRASHED, ER_DEFAULT(ER_NOT_KEYFILE));
  SETMSG(HA_ERR_WRONG_IN_RECORD, ER_DEFAULT(ER_CRASHED_ON_USAGE));
  SETMSG(HA_ERR_OUT_OF_MEM, "Table handler out of memory");
  SETMSG(HA_ERR_NOT_A_TABLE, "Incorrect file format '%.64s'");
  SETMSG(HA_ERR_WRONG_COMMAND, "Command not supported");
  SETMSG(HA_ERR_OLD_FILE, ER_DEFAULT(ER_OLD_KEYFILE));
  SETMSG(HA_ERR_NO_ACTIVE_RECORD, "No record read in update");
  SETMSG(HA_ERR_RECORD_DELETED, "Intern record deleted");
  SETMSG(HA_ERR_RECORD_FILE_FULL, ER_DEFAULT(ER_RECORD_FILE_FULL));
  SETMSG(HA_ERR_INDEX_FILE_FULL, "No more room in index file '%.64s'");
  SETMSG(HA_ERR_END_OF_FILE, "End in next/prev/first/last");
  SETMSG(HA_ERR_UNSUPPORTED, ER_DEFAULT(ER_ILLEGAL_HA));
  SETMSG(HA_ERR_TOO_BIG_ROW, "Too big row");
  SETMSG(HA_WRONG_CREATE_OPTION, "Wrong create option");
  SETMSG(HA_ERR_FOUND_DUPP_UNIQUE, ER_DEFAULT(ER_DUP_UNIQUE));
  SETMSG(HA_ERR_UNKNOWN_CHARSET, "Can't open charset");
  SETMSG(HA_ERR_WRONG_MRG_TABLE_DEF, ER_DEFAULT(ER_WRONG_MRG_TABLE));
  SETMSG(HA_ERR_CRASHED_ON_REPAIR, ER_DEFAULT(ER_CRASHED_ON_REPAIR));
  SETMSG(HA_ERR_CRASHED_ON_USAGE, ER_DEFAULT(ER_CRASHED_ON_USAGE));
  SETMSG(HA_ERR_LOCK_WAIT_TIMEOUT, ER_DEFAULT(ER_LOCK_WAIT_TIMEOUT));
  SETMSG(HA_ERR_LOCK_TABLE_FULL, ER_DEFAULT(ER_LOCK_TABLE_FULL));
  SETMSG(HA_ERR_READ_ONLY_TRANSACTION, ER_DEFAULT(ER_READ_ONLY_TRANSACTION));
  SETMSG(HA_ERR_LOCK_DEADLOCK, ER_DEFAULT(ER_LOCK_DEADLOCK));
  SETMSG(HA_ERR_CANNOT_ADD_FOREIGN, ER_DEFAULT(ER_CANNOT_ADD_FOREIGN));
  SETMSG(HA_ERR_NO_REFERENCED_ROW, ER_DEFAULT(ER_NO_REFERENCED_ROW_2));
  SETMSG(HA_ERR_ROW_IS_REFERENCED, ER_DEFAULT(ER_ROW_IS_REFERENCED_2));
  SETMSG(HA_ERR_NO_SAVEPOINT, "No savepoint with that name");
  SETMSG(HA_ERR_NON_UNIQUE_BLOCK_SIZE, "Non unique key block size");
  SETMSG(HA_ERR_NO_SUCH_TABLE, "No such table: '%.64s'");
  SETMSG(HA_ERR_TABLE_EXIST, ER_DEFAULT(ER_TABLE_EXISTS_ERROR));
  SETMSG(HA_ERR_NO_CONNECTION, "Could not connect to storage engine");
  SETMSG(HA_ERR_TABLE_DEF_CHANGED, ER_DEFAULT(ER_TABLE_DEF_CHANGED));
  SETMSG(HA_ERR_FOREIGN_DUPLICATE_KEY,
         "FK constraint would lead to duplicate key");
  SETMSG(HA_ERR_TABLE_NEEDS_UPGRADE, ER_DEFAULT(ER_TABLE_NEEDS_UPGRADE));
  SETMSG(HA_ERR_TABLE_READONLY, ER_DEFAULT(ER_OPEN_AS_READONLY));
  SETMSG(HA_ERR_AUTOINC_READ_FAILED, ER_DEFAULT(ER_AUTOINC_READ_FAILED));
  SETMSG(HA_ERR_AUTOINC_ERANGE, ER_DEFAULT(ER_WARN_DATA_OUT_OF_RANGE));
  SETMSG(HA_ERR_TOO_MANY_CONCURRENT_TRXS,
         ER_DEFAULT(ER_TOO_MANY_CONCURRENT_TRXS));
  SETMSG(HA_ERR_INDEX_COL_TOO_LONG, ER_DEFAULT(ER_INDEX_COLUMN_TOO_LONG));
  SETMSG(HA_ERR_INDEX_CORRUPT, ER_DEFAULT(ER_INDEX_CORRUPT));
  SETMSG(HA_FTS_INVALID_DOCID, "Invalid InnoDB FTS Doc ID");
  SETMSG(HA_ERR_TABLE_IN_FK_CHECK, ER_DEFAULT(ER_TABLE_IN_FK_CHECK));
  SETMSG(HA_ERR_TABLESPACE_EXISTS, "Tablespace already exists");
  SETMSG(HA_ERR_TABLESPACE_MISSING, ER_DEFAULT(ER_TABLESPACE_MISSING));
  SETMSG(HA_ERR_FTS_EXCEED_RESULT_CACHE_LIMIT,
         "FTS query exceeds result cache limit");
  SETMSG(HA_ERR_TEMP_FILE_WRITE_FAILURE,
         ER_DEFAULT(ER_TEMP_FILE_WRITE_FAILURE));
  SETMSG(HA_ERR_INNODB_FORCED_RECOVERY, ER_DEFAULT(ER_INNODB_FORCED_RECOVERY));
  SETMSG(HA_ERR_FTS_TOO_MANY_WORDS_IN_PHRASE,
         "Too many words in a FTS phrase or proximity search");
  SETMSG(HA_ERR_TABLE_CORRUPT, ER_DEFAULT(ER_TABLE_CORRUPT));
  SETMSG(HA_ERR_TABLESPACE_MISSING, ER_DEFAULT(ER_TABLESPACE_MISSING));
  SETMSG(HA_ERR_TABLESPACE_IS_NOT_EMPTY,
         ER_DEFAULT(ER_TABLESPACE_IS_NOT_EMPTY));
  SETMSG(HA_ERR_WRONG_FILE_NAME, ER_DEFAULT(ER_WRONG_FILE_NAME));
  SETMSG(HA_ERR_NOT_ALLOWED_COMMAND, ER_DEFAULT(ER_NOT_ALLOWED_COMMAND));
  SETMSG(HA_ERR_COMPUTE_FAILED, "Compute virtual column value failed");
  SETMSG(HA_ERR_DISK_FULL_NOWAIT, ER_DEFAULT(ER_DISK_FULL_NOWAIT));
  SETMSG(HA_ERR_NO_SESSION_TEMP, ER_DEFAULT(ER_NO_SESSION_TEMP));
  SETMSG(HA_ERR_WRONG_TABLE_NAME, ER_DEFAULT(ER_WRONG_TABLE_NAME));
  SETMSG(HA_ERR_TOO_LONG_PATH, ER_DEFAULT(ER_TABLE_NAME_CAUSES_TOO_LONG_PATH));
  SETMSG(HA_ERR_FTS_TOO_MANY_NESTED_EXP,
         "Too many nested sub-expressions in a full-text search");
  /* Register the error messages for use with my_error(). */
  return my_error_register(get_handler_errmsg, HA_ERR_FIRST, HA_ERR_LAST);
}

int ha_finalize_handlerton(st_plugin_int *plugin) {
  handlerton *hton = (handlerton *)plugin->data;
  DBUG_TRACE;

  /* hton can be NULL here, if ha_initialize_handlerton() failed. */
  if (!hton) goto end;

  switch (hton->state) {
    case SHOW_OPTION_NO:
    case SHOW_OPTION_DISABLED:
      break;
    case SHOW_OPTION_YES:
      if (installed_htons[hton->db_type] == hton)
        installed_htons[hton->db_type] = nullptr;
      break;
  };

  if (hton->panic) hton->panic(hton, HA_PANIC_CLOSE);

  if (plugin->plugin->deinit) {
    /*
      Today we have no defined/special behavior for uninstalling
      engine plugins.
    */
    DBUG_PRINT("info", ("Deinitializing plugin: '%s'", plugin->name.str));
    if (plugin->plugin->deinit(nullptr)) {
      DBUG_PRINT("warning", ("Plugin '%s' deinit function returned error.",
                             plugin->name.str));
    }
  }

  /*
    In case a plugin is uninstalled and re-installed later, it should
    reuse an array slot. Otherwise the number of uninstall/install
    cycles would be limited.
  */
  if (hton->slot != HA_SLOT_UNDEF) {
    /* Make sure we are not unpluging another plugin */
    DBUG_ASSERT(se_plugin_array[hton->slot] == plugin);
    DBUG_ASSERT(hton->slot < se_plugin_array.size());
    se_plugin_array[hton->slot] = NULL;
    builtin_htons[hton->slot] = false; /* Extra correctness. */
  }

  my_free(hton);
  plugin->data = nullptr;
end:
  return 0;
}

int ha_initialize_handlerton(st_plugin_int *plugin) {
  handlerton *hton;
  DBUG_TRACE;
  DBUG_PRINT("plugin", ("initialize plugin: '%s'", plugin->name.str));

  hton = (handlerton *)my_malloc(key_memory_handlerton, sizeof(handlerton),
                                 MYF(MY_WME | MY_ZEROFILL));

  if (hton == nullptr) {
    LogErr(ERROR_LEVEL, ER_HANDLERTON_OOM, plugin->name.str);
    goto err_no_hton_memory;
  }

  hton->slot = HA_SLOT_UNDEF;
  /* Historical Requirement */
  plugin->data = hton;  // shortcut for the future
  if (plugin->plugin->init && plugin->plugin->init(hton)) {
    LogErr(ERROR_LEVEL, ER_PLUGIN_INIT_FAILED, plugin->name.str);
    goto err;
  }

  /*
    the switch below and hton->state should be removed when
    command-line options for plugins will be implemented
  */
  DBUG_PRINT("info", ("hton->state=%d", hton->state));
  switch (hton->state) {
    case SHOW_OPTION_NO:
      break;
    case SHOW_OPTION_YES: {
      uint tmp;
      ulong fslot;
      /* now check the db_type for conflict */
      if (hton->db_type <= DB_TYPE_UNKNOWN ||
          hton->db_type >= DB_TYPE_DEFAULT || installed_htons[hton->db_type]) {
        int idx = (int)DB_TYPE_FIRST_DYNAMIC;

        while (idx < (int)DB_TYPE_DEFAULT && installed_htons[idx]) idx++;

        if (idx == (int)DB_TYPE_DEFAULT) {
          LogErr(WARNING_LEVEL, ER_TOO_MANY_STORAGE_ENGINES);
          goto err_deinit;
        }
        if (hton->db_type != DB_TYPE_UNKNOWN)
          LogErr(WARNING_LEVEL, ER_SE_TYPECODE_CONFLICT, plugin->plugin->name,
                 idx);
        hton->db_type = (enum legacy_db_type)idx;
      }

      /*
        In case a plugin is uninstalled and re-installed later, it should
        reuse an array slot. Otherwise the number of uninstall/install
        cycles would be limited. So look for a free slot.
      */
      DBUG_PRINT("plugin",
                 ("total_ha: %lu", static_cast<ulong>(se_plugin_array.size())));
      for (fslot = 0; fslot < se_plugin_array.size(); fslot++) {
        if (!se_plugin_array[fslot]) break;
      }
      if (fslot < se_plugin_array.size())
        hton->slot = fslot;
      else {
        hton->slot = se_plugin_array.size();
      }
      if (se_plugin_array.assign_at(hton->slot, plugin) ||
          builtin_htons.assign_at(hton->slot, (plugin->plugin_dl == nullptr)))
        goto err_deinit;

      installed_htons[hton->db_type] = hton;
      tmp = hton->savepoint_offset;
      hton->savepoint_offset = savepoint_alloc_size;
      savepoint_alloc_size += tmp;
      if (hton->prepare) total_ha_2pc++;
      break;
    }
      /* fall through */
    default:
      hton->state = SHOW_OPTION_DISABLED;
      break;
  }

  /*
    This is entirely for legacy. We will create a new "disk based" hton and a
    "memory" hton which will be configurable longterm. We should be able to
    remove partition and myisammrg.
  */
  switch (hton->db_type) {
    case DB_TYPE_HEAP:
      heap_hton = hton;
      break;
    case DB_TYPE_TEMPTABLE:
      temptable_hton = hton;
      break;
    case DB_TYPE_MYISAM:
      myisam_hton = hton;
      break;
    case DB_TYPE_INNODB:
      innodb_hton = hton;
      break;
    default:
      break;
  };

  /*
    Re-load the optimizer cost constants since this storage engine can
    have non-default cost constants.
  */
  reload_optimizer_cost_constants();

  return 0;

err_deinit:
  /*
    Let plugin do its inner deinitialization as plugin->init()
    was successfully called before.
  */
  if (plugin->plugin->deinit) (void)plugin->plugin->deinit(nullptr);

err:
  my_free(hton);
err_no_hton_memory:
  plugin->data = nullptr;
  return 1;
}

int ha_init() {
  int error = 0;
  DBUG_TRACE;

  /*
    Check if there is a transaction-capable storage engine besides the
    binary log.
  */
  opt_using_transactions =
      se_plugin_array.size() > static_cast<ulong>(opt_bin_log);
  savepoint_alloc_size += sizeof(SAVEPOINT);

  return error;
}

void ha_end() {
  // Unregister handler error messages.
  my_error_unregister(HA_ERR_FIRST, HA_ERR_LAST);
  my_free(handler_errmsgs);
}

static bool dropdb_handlerton(THD *, plugin_ref plugin, void *path) {
  handlerton *hton = plugin_data<handlerton *>(plugin);
  if (hton->state == SHOW_OPTION_YES && hton->drop_database)
    hton->drop_database(hton, (char *)path);
  return false;
}

void ha_drop_database(char *path) {
  plugin_foreach(nullptr, dropdb_handlerton, MYSQL_STORAGE_ENGINE_PLUGIN, path);
}

static bool closecon_handlerton(THD *thd, plugin_ref plugin, void *) {
  handlerton *hton = plugin_data<handlerton *>(plugin);
  /*
    there's no need to rollback here as all transactions must
    be rolled back already
  */
  if (hton->state == SHOW_OPTION_YES && thd_get_ha_data(thd, hton)) {
    if (hton->close_connection) hton->close_connection(hton, thd);
    /* make sure ha_data is reset and ha_data_lock is released */
    thd_set_ha_data(thd, hton, nullptr);
  }
  return false;
}

/**
  @note
    don't bother to rollback here, it's done already
*/
void ha_close_connection(THD *thd) {
  plugin_foreach(thd, closecon_handlerton, MYSQL_STORAGE_ENGINE_PLUGIN,
                 nullptr);
}

static bool kill_handlerton(THD *thd, plugin_ref plugin, void *) {
  handlerton *hton = plugin_data<handlerton *>(plugin);

  if (hton->state == SHOW_OPTION_YES && hton->kill_connection) {
    if (thd_get_ha_data(thd, hton)) hton->kill_connection(hton, thd);
  }

  return false;
}

void ha_kill_connection(THD *thd) {
  plugin_foreach(thd, kill_handlerton, MYSQL_STORAGE_ENGINE_PLUGIN, nullptr);
}

/** Invoke handlerton::pre_dd_shutdown() on a plugin.
@param plugin	storage engine plugin
@retval false (always) */
static bool pre_dd_shutdown_handlerton(THD *, plugin_ref plugin, void *) {
  handlerton *hton = plugin_data<handlerton *>(plugin);
  if (hton->state == SHOW_OPTION_YES && hton->pre_dd_shutdown)
    hton->pre_dd_shutdown(hton);
  return false;
}

/** Invoke handlerton::pre_dd_shutdown() on every storage engine plugin. */
void ha_pre_dd_shutdown(void) {
  plugin_foreach(nullptr, pre_dd_shutdown_handlerton,
                 MYSQL_STORAGE_ENGINE_PLUGIN, nullptr);
}

/* ========================================================================
 ======================= TRANSACTIONS ===================================*/

/**
  Transaction handling in the server
  ==================================

  In each client connection, MySQL maintains two transactional
  states:
  - a statement transaction,
  - a standard, also called normal transaction.

  Historical note
  ---------------
  "Statement transaction" is a non-standard term that comes
  from the times when MySQL supported BerkeleyDB storage engine.

  First of all, it should be said that in BerkeleyDB auto-commit
  mode auto-commits operations that are atomic to the storage
  engine itself, such as a write of a record, and are too
  high-granular to be atomic from the application perspective
  (MySQL). One SQL statement could involve many BerkeleyDB
  auto-committed operations and thus BerkeleyDB auto-commit was of
  little use to MySQL.

  Secondly, instead of SQL standard savepoints, BerkeleyDB
  provided the concept of "nested transactions". In a nutshell,
  transactions could be arbitrarily nested, but when the parent
  transaction was committed or aborted, all its child (nested)
  transactions were handled committed or aborted as well.
  Commit of a nested transaction, in turn, made its changes
  visible, but not durable: it destroyed the nested transaction,
  all its changes would become available to the parent and
  currently active nested transactions of this parent.

  So the mechanism of nested transactions was employed to
  provide "all or nothing" guarantee of SQL statements
  required by the standard.
  A nested transaction would be created at start of each SQL
  statement, and destroyed (committed or aborted) at statement
  end. Such nested transaction was internally referred to as
  a "statement transaction" and gave birth to the term.

  (Historical note ends)

  Since then a statement transaction is started for each statement
  that accesses transactional tables or uses the binary log.  If
  the statement succeeds, the statement transaction is committed.
  If the statement fails, the transaction is rolled back. Commits
  of statement transactions are not durable -- each such
  transaction is nested in the normal transaction, and if the
  normal transaction is rolled back, the effects of all enclosed
  statement transactions are undone as well.  Technically,
  a statement transaction can be viewed as a savepoint which is
  maintained automatically in order to make effects of one
  statement atomic.

  The normal transaction is started by the user and is ended
  usually upon a user request as well. The normal transaction
  encloses transactions of all statements issued between
  its beginning and its end.
  In autocommit mode, the normal transaction is equivalent
  to the statement transaction.

  Since MySQL supports PSEA (pluggable storage engine
  architecture), more than one transactional engine can be
  active at a time. Hence transactions, from the server
  point of view, are always distributed. In particular,
  transactional state is maintained independently for each
  engine. In order to commit a transaction the two phase
  commit protocol is employed.

  Not all statements are executed in context of a transaction.
  Administrative and status information statements do not modify
  engine data, and thus do not start a statement transaction and
  also have no effect on the normal transaction. Examples of such
  statements are SHOW STATUS and RESET SLAVE.

  Similarly DDL statements are not transactional,
  and therefore a transaction is [almost] never started for a DDL
  statement. The difference between a DDL statement and a purely
  administrative statement though is that a DDL statement always
  commits the current transaction before proceeding, if there is
  any.

  At last, SQL statements that work with non-transactional
  engines also have no effect on the transaction state of the
  connection. Even though they are written to the binary log,
  and the binary log is, overall, transactional, the writes
  are done in "write-through" mode, directly to the binlog
  file, followed with a OS cache sync, in other words,
  bypassing the binlog undo log (translog).
  They do not commit the current normal transaction.
  A failure of a statement that uses non-transactional tables
  would cause a rollback of the statement transaction, but
  in case there no non-transactional tables are used,
  no statement transaction is started.

  Data layout
  -----------

  The server stores its transaction-related data in
  thd->transaction. This structure has two members of type
  THD_TRANS. These members correspond to the statement and
  normal transactions respectively:

  - thd->transaction.stmt contains a list of engines
  that are participating in the given statement
  - thd->transaction.all contains a list of engines that
  have participated in any of the statement transactions started
  within the context of the normal transaction.
  Each element of the list contains a pointer to the storage
  engine, engine-specific transactional data, and engine-specific
  transaction flags.

  In autocommit mode thd->transaction.all is empty.
  Instead, data of thd->transaction.stmt is
  used to commit/rollback the normal transaction.

  The list of registered engines has a few important properties:
  - no engine is registered in the list twice
  - engines are present in the list a reverse temporal order --
  new participants are always added to the beginning of the list.

  Transaction life cycle
  ----------------------

  When a new connection is established, thd->transaction
  members are initialized to an empty state.
  If a statement uses any tables, all affected engines
  are registered in the statement engine list. In
  non-autocommit mode, the same engines are registered in
  the normal transaction list.
  At the end of the statement, the server issues a commit
  or a roll back for all engines in the statement list.
  At this point transaction flags of an engine, if any, are
  propagated from the statement list to the list of the normal
  transaction.
  When commit/rollback is finished, the statement list is
  cleared. It will be filled in again by the next statement,
  and emptied again at the next statement's end.

  The normal transaction is committed in a similar way
  (by going over all engines in thd->transaction.all list)
  but at different times:
  - upon COMMIT SQL statement is issued by the user
  - implicitly, by the server, at the beginning of a DDL statement
  or SET AUTOCOMMIT={0|1} statement.

  The normal transaction can be rolled back as well:
  - if the user has requested so, by issuing ROLLBACK SQL
  statement
  - if one of the storage engines requested a rollback
  by setting thd->transaction_rollback_request. This may
  happen in case, e.g., when the transaction in the engine was
  chosen a victim of the internal deadlock resolution algorithm
  and rolled back internally. When such a situation happens, there
  is little the server can do and the only option is to rollback
  transactions in all other participating engines.  In this case
  the rollback is accompanied by an error sent to the user.

  As follows from the use cases above, the normal transaction
  is never committed when there is an outstanding statement
  transaction. In most cases there is no conflict, since
  commits of the normal transaction are issued by a stand-alone
  administrative or DDL statement, thus no outstanding statement
  transaction of the previous statement exists. Besides,
  all statements that manipulate with the normal transaction
  are prohibited in stored functions and triggers, therefore
  no conflicting situation can occur in a sub-statement either.
  The remaining rare cases when the server explicitly has
  to commit the statement transaction prior to committing the normal
  one cover error-handling scenarios (see for example
  SQLCOM_LOCK_TABLES).

  When committing a statement or a normal transaction, the server
  either uses the two-phase commit protocol, or issues a commit
  in each engine independently. The two-phase commit protocol
  is used only if:
  - all participating engines support two-phase commit (provide
    handlerton::prepare PSEA API call) and
  - transactions in at least two engines modify data (i.e. are
  not read-only).

  Note that the two phase commit is used for
  statement transactions, even though they are not durable anyway.
  This is done to ensure logical consistency of data in a multiple-
  engine transaction.
  For example, imagine that some day MySQL supports unique
  constraint checks deferred till the end of statement. In such
  case a commit in one of the engines may yield ER_DUP_KEY,
  and MySQL should be able to gracefully abort statement
  transactions of other participants.

  After the normal transaction has been committed,
  thd->transaction.all list is cleared.

  When a connection is closed, the current normal transaction, if
  any, is rolled back.

  Roles and responsibilities
  --------------------------

  The server has no way to know that an engine participates in
  the statement and a transaction has been started
  in it unless the engine says so. Thus, in order to be
  a part of a transaction, the engine must "register" itself.
  This is done by invoking trans_register_ha() server call.
  Normally the engine registers itself whenever handler::external_lock()
  is called. trans_register_ha() can be invoked many times: if
  an engine is already registered, the call does nothing.
  In case autocommit is not set, the engine must register itself
  twice -- both in the statement list and in the normal transaction
  list.
  In which list to register is a parameter of trans_register_ha().

  Note, that although the registration interface in itself is
  fairly clear, the current usage practice often leads to undesired
  effects. E.g. since a call to trans_register_ha() in most engines
  is embedded into implementation of handler::external_lock(), some
  DDL statements start a transaction (at least from the server
  point of view) even though they are not expected to. E.g.
  CREATE TABLE does not start a transaction, since
  handler::external_lock() is never called during CREATE TABLE. But
  CREATE TABLE ... SELECT does, since handler::external_lock() is
  called for the table that is being selected from. This has no
  practical effects currently, but must be kept in mind
  nevertheless.

  Once an engine is registered, the server will do the rest
  of the work.

  During statement execution, whenever any of data-modifying
  PSEA API methods is used, e.g. handler::write_row() or
  handler::update_row(), the read-write flag is raised in the
  statement transaction for the involved engine.
  Currently All PSEA calls are "traced", and the data can not be
  changed in a way other than issuing a PSEA call. Important:
  unless this invariant is preserved the server will not know that
  a transaction in a given engine is read-write and will not
  involve the two-phase commit protocol!

  At the end of a statement, server call trans_commit_stmt is
  invoked. This call in turn invokes handlerton::prepare()
  for every involved engine. Prepare is followed by a call
  to handlerton::commit_one_phase() If a one-phase commit
  will suffice, handlerton::prepare() is not invoked and
  the server only calls handlerton::commit_one_phase().
  At statement commit, the statement-related read-write
  engine flag is propagated to the corresponding flag in the
  normal transaction.  When the commit is complete, the list
  of registered engines is cleared.

  Rollback is handled in a similar fashion.

  Additional notes on DDL and the normal transaction.
  ---------------------------------------------------

  DDLs and operations with non-transactional engines
  do not "register" in thd->transaction lists, and thus do not
  modify the transaction state. Besides, each DDL in
  MySQL is prefixed with an implicit normal transaction commit
  (a call to trans_commit_implicit()), and thus leaves nothing
  to modify.
  However, as it has been pointed out with CREATE TABLE .. SELECT,
  some DDL statements can start a *new* transaction.

  Behaviour of the server in this case is currently badly
  defined.
  DDL statements use a form of "semantic" logging
  to maintain atomicity: if CREATE TABLE .. SELECT failed,
  the newly created table is deleted.
  In addition, some DDL statements issue interim transaction
  commits: e.g. ALTER TABLE issues a commit after data is copied
  from the original table to the internal temporary table. Other
  statements, e.g. CREATE TABLE ... SELECT do not always commit
  after itself.
  And finally there is a group of DDL statements such as
  RENAME/DROP TABLE that doesn't start a new transaction
  and doesn't commit.

  This diversity makes it hard to say what will happen if
  by chance a stored function is invoked during a DDL --
  whether any modifications it makes will be committed or not
  is not clear. Fortunately, SQL grammar of few DDLs allows
  invocation of a stored function.

  A consistent behaviour is perhaps to always commit the normal
  transaction after all DDLs, just like the statement transaction
  is always committed at the end of all statements.
*/

/**
  Register a storage engine for a transaction.

  Every storage engine MUST call this function when it starts
  a transaction or a statement (that is it must be called both for the
  "beginning of transaction" and "beginning of statement").
  Only storage engines registered for the transaction/statement
  will know when to commit/rollback it.

  @note
    trans_register_ha is idempotent - storage engine may register many
    times per transaction.

*/
void trans_register_ha(THD *thd, bool all, handlerton *ht_arg,
                       const ulonglong *trxid MY_ATTRIBUTE((unused))) {
  Ha_trx_info *ha_info;
  Transaction_ctx *trn_ctx = thd->get_transaction();
  Transaction_ctx::enum_trx_scope trx_scope =
      all ? Transaction_ctx::SESSION : Transaction_ctx::STMT;

  DBUG_TRACE;
  DBUG_PRINT("enter", ("%s", all ? "all" : "stmt"));

  if (all) {
    /*
      Ensure no active backup engine data exists, unless the current
      transaction is from replication and in active xa state.
    */
    DBUG_ASSERT(
        thd->get_ha_data(ht_arg->slot)->ha_ptr_backup == nullptr ||
        (thd->get_transaction()->xid_state()->has_state(XID_STATE::XA_ACTIVE)));
    DBUG_ASSERT(thd->get_ha_data(ht_arg->slot)->ha_ptr_backup == nullptr ||
                (thd->is_binlog_applier() || thd->slave_thread));

    thd->server_status |= SERVER_STATUS_IN_TRANS;
    if (thd->tx_read_only)
      thd->server_status |= SERVER_STATUS_IN_TRANS_READONLY;
    DBUG_PRINT("info", ("setting SERVER_STATUS_IN_TRANS"));
  }

  ha_info = thd->get_ha_data(ht_arg->slot)->ha_info + (all ? 1 : 0);

  if (ha_info->is_started()) {
    DBUG_ASSERT(trn_ctx->ha_trx_info(trx_scope));
    return; /* already registered, return */
  }

  trn_ctx->register_ha(trx_scope, ha_info, ht_arg);
  trn_ctx->set_ha_trx_info(trx_scope, ha_info);

  if (ht_arg->prepare == nullptr) trn_ctx->set_no_2pc(trx_scope, true);

  trn_ctx->xid_state()->set_query_id(thd->query_id);
/*
  Register transaction start in performance schema if not done already.
  By doing this, we handle cases when the transaction is started implicitly in
  autocommit=0 mode, and cases when we are in normal autocommit=1 mode and the
  executed statement is a single-statement transaction.

  Explicitly started transactions are handled in trans_begin().

  Do not register transactions in which binary log is the only participating
  transactional storage engine.
*/
#ifdef HAVE_PSI_TRANSACTION_INTERFACE
  if (thd->m_transaction_psi == nullptr && ht_arg->db_type != DB_TYPE_BINLOG &&
      !thd->is_attachable_transaction_active()) {
    const XID *xid = trn_ctx->xid_state()->get_xid();
    bool autocommit = !thd->in_multi_stmt_transaction_mode();
    thd->m_transaction_psi = MYSQL_START_TRANSACTION(
        &thd->m_transaction_state, xid, trxid, thd->tx_isolation,
        thd->tx_read_only, autocommit);
    DEBUG_SYNC(thd, "after_set_transaction_psi_before_set_transaction_gtid");
    gtid_set_performance_schema_values(thd);
  }
#endif
}

/** XA Prepare one SE.
@param[in]	thd	Session THD
@param[in]	ht	SE handlerton
@return 0 for success, 1 for error - entire transaction is rolled back. */
static int prepare_one_ht(THD *thd, handlerton *ht) {
  DBUG_TRACE;
  DBUG_ASSERT(!thd->status_var_aggregated);
  thd->status_var.ha_prepare_count++;
  if (ht->prepare) {
    DBUG_EXECUTE_IF("simulate_xa_failure_prepare", {
      ha_rollback_trans(thd, true);
      return 1;
    });
    if (ht->prepare(ht, thd, true)) {
      ha_rollback_trans(thd, true);
      return 1;
    }
  } else {
    push_warning_printf(thd, Sql_condition::SL_WARNING, ER_ILLEGAL_HA,
                        ER_THD(thd, ER_ILLEGAL_HA),
                        ha_resolve_storage_engine_name(ht));
  }
  return 0;
}

/**
  @retval
    0   ok
  @retval
    1   error, transaction was rolled back
*/
int ha_xa_prepare(THD *thd) {
  int error = 0;
  Transaction_ctx *trn_ctx = thd->get_transaction();
  DBUG_TRACE;

  if (trn_ctx->is_active(Transaction_ctx::SESSION)) {
    const Ha_trx_info *ha_info = trn_ctx->ha_trx_info(Transaction_ctx::SESSION);
    bool gtid_error = false;
    bool need_clear_owned_gtid = false;
    std::tie(gtid_error, need_clear_owned_gtid) = commit_owned_gtids(thd, true);
    if (gtid_error) {
      DBUG_ASSERT(need_clear_owned_gtid);

      ha_rollback_trans(thd, true);
      error = 1;
      goto err;
    }

    /*
      Ensure externalization order for applier threads.

      Note: the calls to Commit_order_manager::wait/wait_and_finish() will be
            no-op for threads other than replication applier threads.
    */
    if (Commit_order_manager::wait(thd)) {
      thd->commit_error = THD::CE_NONE;
      ha_rollback_trans(thd, true);
      error = 1;
      gtid_error = true;
      goto err;
    }

    /* Allow GTID to be read by SE for XA prepare. */
    {
      Clone_handler::XA_Operation xa_guard(thd);

      /* Prepare binlog SE first, if there. */
      while (ha_info != nullptr && error == 0) {
        auto ht = ha_info->ht();
        if (ht->db_type == DB_TYPE_BINLOG) {
          error = prepare_one_ht(thd, ht);
          break;
        }
        ha_info = ha_info->next();
      }
      /* Prepare all SE other than binlog. */
      ha_info = trn_ctx->ha_trx_info(Transaction_ctx::SESSION);
      while (ha_info != nullptr && error == 0) {
        auto ht = ha_info->ht();
        error = prepare_one_ht(thd, ht);
        if (error != 0) {
          break;
        }
        ha_info = ha_info->next();
      }
    }

    DBUG_ASSERT(error != 0 || thd->get_transaction()->xid_state()->has_state(
                                  XID_STATE::XA_IDLE));

  err:
    /*
      After ensuring externalization order for applier thread, remove it
      from waiting (Commit Order Queue) and allow next applier thread to
      be ordered.

      Note: the calls to Commit_order_manager::wait_and_finish() will be
            no-op for threads other than replication applier threads.
    */
    Commit_order_manager::wait_and_finish(thd, error);
    gtid_state_commit_or_rollback(thd, need_clear_owned_gtid, !gtid_error);
  }

  return error;
}

/**
  Check if we can skip the two-phase commit.

  A helper function to evaluate if two-phase commit is mandatory.
  As a side effect, propagates the read-only/read-write flags
  of the statement transaction to its enclosing normal transaction.

  If we have at least two engines with read-write changes we must
  run a two-phase commit. Otherwise we can run several independent
  commits as the only transactional engine has read-write changes
  and others are read-only.

  @retval   0   All engines are read-only.
  @retval   1   We have the only engine with read-write changes.
  @retval   >1  More than one engine have read-write changes.
                Note: return value might NOT be the exact number of
                engines with read-write changes.
*/

static uint ha_check_and_coalesce_trx_read_only(THD *thd, Ha_trx_info *ha_list,
                                                bool all) {
  /* The number of storage engines that have actual changes. */
  unsigned rw_ha_count = 0;
  Ha_trx_info *ha_info;

  for (ha_info = ha_list; ha_info; ha_info = ha_info->next()) {
    if (ha_info->is_trx_read_write()) ++rw_ha_count;

    if (!all) {
      Ha_trx_info *ha_info_all =
          &thd->get_ha_data(ha_info->ht()->slot)->ha_info[1];
      DBUG_ASSERT(ha_info != ha_info_all);
      /*
        Merge read-only/read-write information about statement
        transaction to its enclosing normal transaction. Do this
        only if in a real transaction -- that is, if we know
        that ha_info_all is registered in thd->transaction.all.
        Since otherwise we only clutter the normal transaction flags.
      */
      if (ha_info_all->is_started()) /* false if autocommit. */
        ha_info_all->coalesce_trx_with(ha_info);
    } else if (rw_ha_count > 1) {
      /*
        It is a normal transaction, so we don't need to merge read/write
        information up, and the need for two-phase commit has been
        already established. Break the loop prematurely.
      */
      break;
    }
  }
  return rw_ha_count;
}

/**
  The function computes condition to call gtid persistor wrapper,
  and executes it.
  It is invoked at committing a statement or transaction, including XA,
  and also at XA prepare handling.

  @param thd  Thread context.
  @param all  The execution scope, true for the transaction one, false
              for the statement one.

  @return   std::pair containing: Error and Owned GTID release status
   Error
            @retval  0    Ok
            @retval !0    Error

   Owned GTID release status
            @retval  true   remove the GTID owned by thread from owned GTIDs
            @retval  false  removal of the GTID owned by thread from owned GTIDs
                            is not required
*/

std::pair<int, bool> commit_owned_gtids(THD *thd, bool all) {
  DBUG_TRACE;
  int error = 0;
  bool need_clear_owned_gtid = false;

  /*
    If the binary log is disabled for this thread (either by
    log_bin=0 or sql_log_bin=0 or by log_slave_updates=0 for a
    slave thread), then the statement will not be written to
    the binary log. In this case, we should save its GTID into
    mysql.gtid_executed table and @@GLOBAL.GTID_EXECUTED as it
    did when binlog is enabled.

    We also skip saving GTID into mysql.gtid_executed table and
    @@GLOBAL.GTID_EXECUTED when slave-preserve-commit-order is enabled. We skip
    as GTID will be saved in
    Commit_order_manager::flush_engine_and_signal_threads (invoked from
    Commit_order_manager::wait_and_finish). In particular, there is the
    following call stack under ha_commit_low which save GTID in case its skipped
    here:

      ha_commit_low ->
      Commit_order_manager::wait_and_finish ->
      Commit_order_manager::finish ->
      Commit_order_manager::flush_engine_and_signal_threads ->
      Gtid_state::update_commit_group

    We also skip saving GTID for intermediate commits i.e. when
    thd->is_operating_substatement_implicitly is enabled.
  */
  if (thd->is_current_stmt_binlog_log_slave_updates_disabled() &&
      ending_trans(thd, all) && !thd->is_operating_gtid_table_implicitly &&
      !thd->is_operating_substatement_implicitly) {
    if (!has_commit_order_manager(thd) &&
        (thd->owned_gtid.sidno > 0 ||
         thd->owned_gtid.sidno == THD::OWNED_SIDNO_ANONYMOUS)) {
      need_clear_owned_gtid = true;
    }

    /*
      If GTID is not persisted by SE, write it to
      mysql.gtid_executed table.
    */
    if (thd->owned_gtid.sidno > 0 && !thd->se_persists_gtid()) {
      error = gtid_state->save(thd);
    }
  }

  return std::make_pair(error, need_clear_owned_gtid);
}

/**
  @param[in] thd                       Thread handle.
  @param[in] all                       Session transaction if true, statement
                                       otherwise.
  @param[in] ignore_global_read_lock   Allow commit to complete even if a
                                       global read lock is active. This can be
                                       used to allow changes to internal tables
                                       (e.g. slave status tables).

  @retval
    0   ok
  @retval
    1   transaction was rolled back
  @retval
    2   error during commit, data may be inconsistent

  @todo
    Since we don't support nested statement transactions in 5.0,
    we can't commit or rollback stmt transactions while we are inside
    stored functions or triggers. So we simply do nothing now.
    TODO: This should be fixed in later ( >= 5.1) releases.
*/

int ha_commit_trans(THD *thd, bool all, bool ignore_global_read_lock) {
  int error = 0;
  THD_STAGE_INFO(thd, stage_waiting_for_handler_commit);
  bool run_slave_post_commit = false;
  bool need_clear_owned_gtid = false;
  /*
    Save transaction owned gtid into table before transaction prepare
    if binlog is disabled, or binlog is enabled and log_slave_updates
    is disabled with slave SQL thread or slave worker thread.
  */
  std::tie(error, need_clear_owned_gtid) = commit_owned_gtids(thd, all);

  /*
    'all' means that this is either an explicit commit issued by
    user, or an implicit commit issued by a DDL.
  */
  Transaction_ctx *trn_ctx = thd->get_transaction();
  Transaction_ctx::enum_trx_scope trx_scope =
      all ? Transaction_ctx::SESSION : Transaction_ctx::STMT;

  /*
    "real" is a nick name for a transaction for which a commit will
    make persistent changes. E.g. a 'stmt' transaction inside a 'all'
    transation is not 'real': even though it's possible to commit it,
    the changes are not durable as they might be rolled back if the
    enclosing 'all' transaction is rolled back.
  */
  bool is_real_trans = all || !trn_ctx->is_active(Transaction_ctx::SESSION);

  Ha_trx_info *ha_info = trn_ctx->ha_trx_info(trx_scope);
  XID_STATE *xid_state = trn_ctx->xid_state();

  DBUG_TRACE;

  DBUG_PRINT("info", ("all=%d thd->in_sub_stmt=%d ha_info=%p is_real_trans=%d",
                      all, thd->in_sub_stmt, ha_info, is_real_trans));
  /*
    We must not commit the normal transaction if a statement
    transaction is pending. Otherwise statement transaction
    flags will not get propagated to its normal transaction's
    counterpart.
  */
  DBUG_ASSERT(!trn_ctx->is_active(Transaction_ctx::STMT) || !all);

  DBUG_EXECUTE_IF("pre_commit_error", {
    error = true;
    my_error(ER_UNKNOWN_ERROR, MYF(0));
  });

  /*
    When atomic DDL is executed on the slave, we would like to
    to update slave applier state as part of DDL's transaction.
    Call Relay_log_info::pre_commit() hook to do this before DDL
    gets committed in the following block.
    Failed atomic DDL statements should've been marked as executed/committed
    during statement rollback, though some like GRANT may continue until
    this point.
    When applying a DDL statement on a slave and the statement is filtered
    out by a table filter, we report an error "ER_SLAVE_IGNORED_TABLE" to
    warn slave applier thread. We need to save the DDL statement's gtid
    into mysql.gtid_executed system table if the binary log is disabled
    on the slave and gtids are enabled.
  */
  if (is_real_trans && is_atomic_ddl_commit_on_slave(thd) &&
      (!thd->is_error() ||
       (thd->is_operating_gtid_table_implicitly &&
        thd->get_stmt_da()->mysql_errno() == ER_SLAVE_IGNORED_TABLE))) {
    run_slave_post_commit = true;
    error = error || thd->rli_slave->pre_commit();

    DBUG_EXECUTE_IF("rli_pre_commit_error", {
      error = true;
      my_error(ER_UNKNOWN_ERROR, MYF(0));
    });
    DBUG_EXECUTE_IF("slave_crash_before_commit", {
      /* This pre-commit crash aims solely at atomic DDL */
      DBUG_SUICIDE();
    });
  }

  if (thd->in_sub_stmt) {
    DBUG_ASSERT(0);
    /*
      Since we don't support nested statement transactions in 5.0,
      we can't commit or rollback stmt transactions while we are inside
      stored functions or triggers. So we simply do nothing now.
      TODO: This should be fixed in later ( >= 5.1) releases.
    */
    if (!all) return 0;
    /*
      We assume that all statements which commit or rollback main transaction
      are prohibited inside of stored functions or triggers. So they should
      bail out with error even before ha_commit_trans() call. To be 100% safe
      let us throw error in non-debug builds.
    */
    my_error(ER_COMMIT_NOT_ALLOWED_IN_SF_OR_TRG, MYF(0));
    return 2;
  }

  MDL_request mdl_request;
  bool release_mdl = false;
  if (ha_info && !error) {
    uint rw_ha_count = 0;
    bool rw_trans;

    DBUG_EXECUTE_IF("crash_commit_before", DBUG_SUICIDE(););

    /*
     skip 2PC if the transaction is empty and it is not marked as started (which
     can happen when the slave's binlog is disabled)
    */
    if (ha_info->is_started())
      rw_ha_count = ha_check_and_coalesce_trx_read_only(thd, ha_info, all);
    trn_ctx->set_rw_ha_count(trx_scope, rw_ha_count);
    /* rw_trans is true when we in a transaction changing data */
    rw_trans = is_real_trans && (rw_ha_count > 0);

    DBUG_EXECUTE_IF("dbug.enabled_commit", {
      const char act[] = "now signal Reached wait_for signal.commit_continue";
      DBUG_ASSERT(!debug_sync_set_action(thd, STRING_WITH_LEN(act)));
    };);
    DEBUG_SYNC(thd, "ha_commit_trans_before_acquire_commit_lock");
    if (rw_trans && !ignore_global_read_lock) {
      /*
        Acquire a metadata lock which will ensure that COMMIT is blocked
        by an active FLUSH TABLES WITH READ LOCK (and vice versa:
        COMMIT in progress blocks FTWRL).

        We allow the owner of FTWRL to COMMIT; we assume that it knows
        what it does.
      */
      MDL_REQUEST_INIT(&mdl_request, MDL_key::COMMIT, "", "",
                       MDL_INTENTION_EXCLUSIVE, MDL_EXPLICIT);

      DBUG_PRINT("debug", ("Acquire MDL commit lock"));
      if (thd->mdl_context.acquire_lock(&mdl_request,
                                        thd->variables.lock_wait_timeout)) {
        ha_rollback_trans(thd, all);
        return 1;
      }
      release_mdl = true;

      DEBUG_SYNC(thd, "ha_commit_trans_after_acquire_commit_lock");
    }

    if (rw_trans && stmt_has_updated_trans_table(ha_info) &&
        check_readonly(thd, true)) {
      ha_rollback_trans(thd, all);
      error = 1;
      goto end;
    }

    if (!trn_ctx->no_2pc(trx_scope) && (trn_ctx->rw_ha_count(trx_scope) > 1))
      error = tc_log->prepare(thd, all);
  }
  /*
    The state of XA transaction is changed to Prepared, intermediately.
    It's going to change to the regular NOTR at the end.
    The fact of the Prepared state is of interest to binary logger.
  */
  if (!error && all && xid_state->has_state(XID_STATE::XA_IDLE)) {
    DBUG_ASSERT(
        thd->lex->sql_command == SQLCOM_XA_COMMIT &&
        static_cast<Sql_cmd_xa_commit *>(thd->lex->m_sql_cmd)->get_xa_opt() ==
            XA_ONE_PHASE);

    xid_state->set_state(XID_STATE::XA_PREPARED);
  }
  if (error || (error = tc_log->commit(thd, all))) {
    ha_rollback_trans(thd, all);
    error = 1;
    goto end;
  }
/*
  Mark multi-statement (any autocommit mode) or single-statement
  (autocommit=1) transaction as rolled back
*/
#ifdef HAVE_PSI_TRANSACTION_INTERFACE
  if (is_real_trans && thd->m_transaction_psi != nullptr) {
    MYSQL_COMMIT_TRANSACTION(thd->m_transaction_psi);
    thd->m_transaction_psi = nullptr;
  }
#endif
  DBUG_EXECUTE_IF("crash_commit_after",
                  if (!thd->is_operating_gtid_table_implicitly)
                      DBUG_SUICIDE(););
end:
  if (release_mdl && mdl_request.ticket) {
    /*
      We do not always immediately release transactional locks
      after ha_commit_trans() (see uses of ha_enable_transaction()),
      thus we release the commit blocker lock as soon as it's
      not needed.
    */
    DBUG_PRINT("debug", ("Releasing MDL commit lock"));
    thd->mdl_context.release_lock(mdl_request.ticket);
  }
  /* Free resources and perform other cleanup even for 'empty' transactions. */
  if (is_real_trans) {
    trn_ctx->cleanup();
    thd->tx_priority = 0;
  }

  if (need_clear_owned_gtid) {
    thd->server_status &= ~SERVER_STATUS_IN_TRANS;
    /*
      Release the owned GTID when binlog is disabled, or binlog is
      enabled and log_slave_updates is disabled with slave SQL thread
      or slave worker thread.
    */
    if (error)
      gtid_state->update_on_rollback(thd);
    else
      gtid_state->update_on_commit(thd);
  } else {
    if (has_commit_order_manager(thd) && error) {
      gtid_state->update_on_rollback(thd);
    }
  }
  if (run_slave_post_commit) {
    DBUG_EXECUTE_IF("slave_crash_after_commit", DBUG_SUICIDE(););

    thd->rli_slave->post_commit(error != 0);
    /*
      SERVER_STATUS_IN_TRANS may've been gained by pre_commit alone
      when the main DDL transaction is filtered out of execution.
      In such case the status has to be reset now.

      TODO: move/refactor this handling onto trans_commit/commit_implicit()
            the caller level.
    */
    thd->server_status &= ~SERVER_STATUS_IN_TRANS;
  } else {
    DBUG_EXECUTE_IF("slave_crash_after_commit", {
      if (thd->slave_thread && thd->rli_slave &&
          thd->rli_slave->current_event &&
          thd->rli_slave->current_event->get_type_code() ==
              binary_log::XID_EVENT &&
          !thd->is_operating_substatement_implicitly &&
          !thd->is_operating_gtid_table_implicitly)
        DBUG_SUICIDE();
    });
  }

  return error;
}

/**
  Commit the sessions outstanding transaction.

  @pre thd->transaction.flags.commit_low == true
  @post thd->transaction.flags.commit_low == false

  @note This function does not care about global read lock; the caller
  should.

  @param[in]  thd  Thread handle.
  @param[in]  all  Is set in case of explicit commit
                   (COMMIT statement), or implicit commit
                   issued by DDL. Is not set when called
                   at the end of statement, even if
                   autocommit=1.
  @param[in]  run_after_commit
                   True by default, otherwise, does not execute
                   the after_commit hook in the function.
*/

int ha_commit_low(THD *thd, bool all, bool run_after_commit) {
  int error = 0;
  Transaction_ctx *trn_ctx = thd->get_transaction();
  Transaction_ctx::enum_trx_scope trx_scope =
      all ? Transaction_ctx::SESSION : Transaction_ctx::STMT;
  Ha_trx_info *ha_info = trn_ctx->ha_trx_info(trx_scope), *ha_info_next;

  DBUG_TRACE;

  if (ha_info) {
    bool restore_backup_ha_data = false;
    /*
      At execution of XA COMMIT ONE PHASE binlog or slave applier
      reattaches the engine ha_data to THD, previously saved at XA START.
    */
    if (all && thd->rpl_unflag_detached_engine_ha_data()) {
      DBUG_PRINT("info", ("query='%s'", thd->query().str));
      DBUG_ASSERT(thd->lex->sql_command == SQLCOM_XA_COMMIT);
      DBUG_ASSERT(
          static_cast<Sql_cmd_xa_commit *>(thd->lex->m_sql_cmd)->get_xa_opt() ==
          XA_ONE_PHASE);
      restore_backup_ha_data = true;
    }

    bool is_applier_wait_enabled = false;

    /*
      Preserve externalization and persistence order for applier threads.

      The conditions should be understood as follows:

      - When the binlog is enabled, this will be done from
        MYSQL_BIN_LOG::ordered_commit and should not be done here.
        Therefore, we have the condition
        thd->is_current_stmt_binlog_disabled().

      - This function is usually called once per statement, with
        all=false.  We should not preserve the commit order when this
        function is called in that context.  Therefore, we have the
        condition ending_trans(thd, all).

      - Statements such as ANALYZE/OPTIMIZE/REPAIR TABLE will call
        ha_commit_low multiple times with all=true from within
        mysql_admin_table, mysql_recreate_table, and
        handle_histogram_command. After returing to
        mysql_execute_command, it will call ha_commit_low a final
        time.  It is only in this final call that we should preserve
        the commit order. Therefore, we set the flag
        thd->is_operating_substatement_implicitly while executing
        mysql_admin_table, mysql_recreate_table, and
        handle_histogram_command, clear it when returning from those
        functions, and check the flag here in ha_commit_low().

      - In all the above cases, we should make the current transaction
        fail early in case a previous transaction has rolled back.
        Therefore, we also invoke the commit order manager in case
        get_rollback_status returns true.

      Note: the calls to Commit_order_manager::wait/wait_and_finish() will be
            no-op for threads other than replication applier threads.
    */
    if ((!thd->is_operating_substatement_implicitly &&
         !thd->is_operating_gtid_table_implicitly &&
         thd->is_current_stmt_binlog_log_slave_updates_disabled() &&
         ending_trans(thd, all)) ||
        Commit_order_manager::get_rollback_status(thd)) {
      if (Commit_order_manager::wait(thd)) {
        error = 1;
        /*
          Remove applier thread from waiting in Commit Order Queue and
          allow next applier thread to be ordered.
        */
        Commit_order_manager::wait_and_finish(thd, error);
        goto err;
      }
      is_applier_wait_enabled = true;
    }

    for (; ha_info; ha_info = ha_info_next) {
      int err;
      handlerton *ht = ha_info->ht();
      if ((err = ht->commit(ht, thd, all))) {
        char errbuf[MYSQL_ERRMSG_SIZE];
        my_error(ER_ERROR_DURING_COMMIT, MYF(0), err,
                 my_strerror(errbuf, MYSQL_ERRMSG_SIZE, err));
        error = 1;
      }
      DBUG_ASSERT(!thd->status_var_aggregated);
      thd->status_var.ha_commit_count++;
      ha_info_next = ha_info->next();
      if (restore_backup_ha_data) reattach_engine_ha_data_to_thd(thd, ht);
      ha_info->reset(); /* keep it conveniently zero-filled */
    }
    trn_ctx->reset_scope(trx_scope);

    /*
      After ensuring externalization order for applier thread, remove it
      from waiting (Commit Order Queue) and allow next applier thread to
      be ordered.

      Note: the calls to Commit_order_manager::wait_and_finish() will be
            no-op for threads other than replication applier threads.
    */
    if (is_applier_wait_enabled) {
      Commit_order_manager::wait_and_finish(thd, error);
    }
  }

err:
  /* Free resources and perform other cleanup even for 'empty' transactions. */
  if (all) trn_ctx->cleanup();
  /*
    When the transaction has been committed, we clear the commit_low
    flag. This allow other parts of the system to check if commit_low
    was called.
  */
  trn_ctx->m_flags.commit_low = false;
  if (run_after_commit && thd->get_transaction()->m_flags.run_hooks) {
    /*
       If commit succeeded, we call the after_commit hook.

       TODO: Investigate if this can be refactored so that there is
             only one invocation of this hook in the code (in
             MYSQL_LOG_BIN::finish_commit).
    */
    if (!error) (void)RUN_HOOK(transaction, after_commit, (thd, all));
    trn_ctx->m_flags.run_hooks = false;
  }
  return error;
}

int ha_rollback_low(THD *thd, bool all) {
  Transaction_ctx *trn_ctx = thd->get_transaction();
  int error = 0;
  Transaction_ctx::enum_trx_scope trx_scope =
      all ? Transaction_ctx::SESSION : Transaction_ctx::STMT;
  Ha_trx_info *ha_info = trn_ctx->ha_trx_info(trx_scope), *ha_info_next;

  (void)RUN_HOOK(transaction, before_rollback, (thd, all));

  if (ha_info) {
    bool restore_backup_ha_data = false;
    /*
      Similarly to the commit case, the binlog or slave applier
      reattaches the engine ha_data to THD.
    */
    if (all && thd->rpl_unflag_detached_engine_ha_data()) {
      DBUG_ASSERT(trn_ctx->xid_state()->get_state() != XID_STATE::XA_NOTR ||
                  thd->killed == THD::KILL_CONNECTION);

      restore_backup_ha_data = true;
    }

    for (; ha_info; ha_info = ha_info_next) {
      int err;
      handlerton *ht = ha_info->ht();
      if ((err = ht->rollback(ht, thd, all))) {  // cannot happen
        char errbuf[MYSQL_ERRMSG_SIZE];
        my_error(ER_ERROR_DURING_ROLLBACK, MYF(0), err,
                 my_strerror(errbuf, MYSQL_ERRMSG_SIZE, err));
        error = 1;
      }
      DBUG_ASSERT(!thd->status_var_aggregated);
      thd->status_var.ha_rollback_count++;
      ha_info_next = ha_info->next();
      if (restore_backup_ha_data) reattach_engine_ha_data_to_thd(thd, ht);
      ha_info->reset(); /* keep it conveniently zero-filled */
    }
    trn_ctx->reset_scope(trx_scope);
  }

  /*
    Thanks to possibility of MDL deadlock rollback request can come even if
    transaction hasn't been started in any transactional storage engine.

    It is possible to have a call of ha_rollback_low() while handling
    failure from ha_xa_prepare() and an error in Daignostics_area still
    wasn't set. Therefore it is required to check that an error in
    Diagnostics_area is set before calling the method XID_STATE::set_error().

    If it wasn't done it would lead to failure of the assertion
      DBUG_ASSERT(m_status == DA_ERROR)
    in the method Diagnostics_area::mysql_errno().

    In case ha_xa_prepare is failed and an error wasn't set in Diagnostics_area
    the error ER_XA_RBROLLBACK is set in the Diagnostics_area from
    the method Sql_cmd_xa_prepare::trans_xa_prepare() when non-zero result code
    returned by ha_xa_prepare() is handled.
  */
  if (all && thd->transaction_rollback_request && thd->is_error())
    trn_ctx->xid_state()->set_error(thd);

  (void)RUN_HOOK(transaction, after_rollback, (thd, all));
  return error;
}

int ha_rollback_trans(THD *thd, bool all) {
  int error = 0;
  Transaction_ctx *trn_ctx = thd->get_transaction();
  bool is_xa_rollback = trn_ctx->xid_state()->has_state(XID_STATE::XA_PREPARED);

  /*
    "real" is a nick name for a transaction for which a commit will
    make persistent changes. E.g. a 'stmt' transaction inside a 'all'
    transaction is not 'real': even though it's possible to commit it,
    the changes are not durable as they might be rolled back if the
    enclosing 'all' transaction is rolled back.
    We establish the value of 'is_real_trans' by checking
    if it's an explicit COMMIT or BEGIN statement, or implicit
    commit issued by DDL (in these cases all == true),
    or if we're running in autocommit mode (it's only in the autocommit mode
    ha_commit_one_phase() is called with an empty
    transaction.all.ha_list, see why in trans_register_ha()).
  */
  bool is_real_trans = all || !trn_ctx->is_active(Transaction_ctx::SESSION);

  DBUG_TRACE;

  /*
    We must not rollback the normal transaction if a statement
    transaction is pending.
  */
  DBUG_ASSERT(!trn_ctx->is_active(Transaction_ctx::STMT) || !all);

  if (thd->in_sub_stmt) {
    DBUG_ASSERT(0);
    /*
      If we are inside stored function or trigger we should not commit or
      rollback current statement transaction. See comment in ha_commit_trans()
      call for more information.
    */
    if (!all) return 0;
    my_error(ER_COMMIT_NOT_ALLOWED_IN_SF_OR_TRG, MYF(0));
    return 1;
  }

  if (tc_log) error = tc_log->rollback(thd, all);
    /*
      Mark multi-statement (any autocommit mode) or single-statement
      (autocommit=1) transaction as rolled back
    */
#ifdef HAVE_PSI_TRANSACTION_INTERFACE
  if (all || !thd->in_active_multi_stmt_transaction()) {
    MYSQL_ROLLBACK_TRANSACTION(thd->m_transaction_psi);
    thd->m_transaction_psi = nullptr;
  }
#endif

  /* Always cleanup. Even if nht==0. There may be savepoints. */
  if (is_real_trans) {
    trn_ctx->cleanup();
    thd->tx_priority = 0;
  }

  if (all) thd->transaction_rollback_request = false;

  /*
    Only call gtid_rollback(THD*), which will purge thd->owned_gtid, if
    complete transaction is being rollback or autocommit=1.
    Notice, XA rollback has just invoked update_on_commit() through
    tc_log->*rollback* stack.
  */
  if (is_real_trans && !is_xa_rollback) gtid_state->update_on_rollback(thd);

  /*
    If the transaction cannot be rolled back safely, warn; don't warn if this
    is a slave thread (because when a slave thread executes a ROLLBACK, it has
    been read from the binary log, so it's 100% sure and normal to produce
    error ER_WARNING_NOT_COMPLETE_ROLLBACK. If we sent the warning to the
    slave SQL thread, it would not stop the thread but just be printed in
    the error log; but we don't want users to wonder why they have this
    message in the error log, so we don't send it.
  */
  if (is_real_trans &&
      trn_ctx->cannot_safely_rollback(Transaction_ctx::SESSION) &&
      !thd->slave_thread && thd->killed != THD::KILL_CONNECTION)
    trn_ctx->push_unsafe_rollback_warnings(thd);

  return error;
}

/**
  Commit the attachable transaction in storage engines.

  @note This is slimmed down version of ha_commit_trans()/ha_commit_low()
        which commits attachable transaction but skips code which is
        unnecessary and unsafe for them (like dealing with GTIDs).
        Since attachable transactions are read-only their commit only
        needs to release resources and cleanup state in SE.

  @param thd     Current thread

  @retval 0      - Success
  @retval non-0  - Failure
*/
int ha_commit_attachable(THD *thd) {
  int error = 0;
  Transaction_ctx *trn_ctx = thd->get_transaction();
  Ha_trx_info *ha_info = trn_ctx->ha_trx_info(Transaction_ctx::STMT);
  Ha_trx_info *ha_info_next;

  /* This function only handles attachable transactions. */
  DBUG_ASSERT(thd->is_attachable_ro_transaction_active());
  /*
    Since the attachable transaction is AUTOCOMMIT we only need
    to care about statement transaction.
  */
  DBUG_ASSERT(!trn_ctx->is_active(Transaction_ctx::SESSION));

  if (ha_info) {
    for (; ha_info; ha_info = ha_info_next) {
      /* Attachable transaction is not supposed to modify anything. */
      DBUG_ASSERT(!ha_info->is_trx_read_write());

      handlerton *ht = ha_info->ht();
      if (ht->commit(ht, thd, false)) {
        /*
          In theory this should not happen since attachable transactions
          are read only and therefore commit is supposed to only release
          resources/cleanup state. Even if this happens we will simply
          continue committing attachable transaction in other SEs.
        */
        DBUG_ASSERT(false);
        error = 1;
      }
      DBUG_ASSERT(!thd->status_var_aggregated);
      thd->status_var.ha_commit_count++;
      ha_info_next = ha_info->next();

      ha_info->reset(); /* keep it conveniently zero-filled */
    }
    trn_ctx->reset_scope(Transaction_ctx::STMT);
  }

  /*
    Mark transaction as commited in PSI.
  */
#ifdef HAVE_PSI_TRANSACTION_INTERFACE
  if (thd->m_transaction_psi != nullptr) {
    MYSQL_COMMIT_TRANSACTION(thd->m_transaction_psi);
    thd->m_transaction_psi = nullptr;
  }
#endif

  /* Free resources and perform other cleanup even for 'empty' transactions. */
  trn_ctx->cleanup();

  return (error);
}

/**
  Check if all storage engines used in transaction agree that after
  rollback to savepoint it is safe to release MDL locks acquired after
  savepoint creation.

  @param thd   The client thread that executes the transaction.

  @return true  - It is safe to release MDL locks.
          false - If it is not.
*/
bool ha_rollback_to_savepoint_can_release_mdl(THD *thd) {
  Ha_trx_info *ha_info;
  Transaction_ctx *trn_ctx = thd->get_transaction();
  Transaction_ctx::enum_trx_scope trx_scope =
      thd->in_sub_stmt ? Transaction_ctx::STMT : Transaction_ctx::SESSION;

  DBUG_TRACE;

  /**
    Checking whether it is safe to release metadata locks after rollback to
    savepoint in all the storage engines that are part of the transaction.
  */
  for (ha_info = trn_ctx->ha_trx_info(trx_scope); ha_info;
       ha_info = ha_info->next()) {
    handlerton *ht = ha_info->ht();
    DBUG_ASSERT(ht);

    if (ht->savepoint_rollback_can_release_mdl == nullptr ||
        ht->savepoint_rollback_can_release_mdl(ht, thd) == false)
      return false;
  }

  return true;
}

int ha_rollback_to_savepoint(THD *thd, SAVEPOINT *sv) {
  int error = 0;
  Transaction_ctx *trn_ctx = thd->get_transaction();
  Transaction_ctx::enum_trx_scope trx_scope =
      !thd->in_sub_stmt ? Transaction_ctx::SESSION : Transaction_ctx::STMT;

  Ha_trx_info *ha_info, *ha_info_next;

  DBUG_TRACE;

  trn_ctx->set_rw_ha_count(trx_scope, 0);
  trn_ctx->set_no_2pc(trx_scope, false);
  /*
    rolling back to savepoint in all storage engines that were part of the
    transaction when the savepoint was set
  */
  for (ha_info = sv->ha_list; ha_info; ha_info = ha_info->next()) {
    int err;
    handlerton *ht = ha_info->ht();
    DBUG_ASSERT(ht);
    DBUG_ASSERT(ht->savepoint_set != nullptr);
    if ((err = ht->savepoint_rollback(
             ht, thd,
             (uchar *)(sv + 1) + ht->savepoint_offset))) {  // cannot happen
      char errbuf[MYSQL_ERRMSG_SIZE];
      my_error(ER_ERROR_DURING_ROLLBACK, MYF(0), err,
               my_strerror(errbuf, MYSQL_ERRMSG_SIZE, err));
      error = 1;
    }
    DBUG_ASSERT(!thd->status_var_aggregated);
    thd->status_var.ha_savepoint_rollback_count++;
    if (ht->prepare == nullptr) trn_ctx->set_no_2pc(trx_scope, true);
  }

  /*
    rolling back the transaction in all storage engines that were not part of
    the transaction when the savepoint was set
  */
  for (ha_info = trn_ctx->ha_trx_info(trx_scope); ha_info != sv->ha_list;
       ha_info = ha_info_next) {
    int err;
    handlerton *ht = ha_info->ht();
    if ((err = ht->rollback(ht, thd, !thd->in_sub_stmt))) {  // cannot happen
      char errbuf[MYSQL_ERRMSG_SIZE];
      my_error(ER_ERROR_DURING_ROLLBACK, MYF(0), err,
               my_strerror(errbuf, MYSQL_ERRMSG_SIZE, err));
      error = 1;
    }
    DBUG_ASSERT(!thd->status_var_aggregated);
    thd->status_var.ha_rollback_count++;
    ha_info_next = ha_info->next();
    ha_info->reset(); /* keep it conveniently zero-filled */
  }
  trn_ctx->set_ha_trx_info(trx_scope, sv->ha_list);

#ifdef HAVE_PSI_TRANSACTION_INTERFACE
  if (thd->m_transaction_psi != nullptr)
    MYSQL_INC_TRANSACTION_ROLLBACK_TO_SAVEPOINT(thd->m_transaction_psi, 1);
#endif

  return error;
}

int ha_prepare_low(THD *thd, bool all) {
  int error = 0;
  Transaction_ctx::enum_trx_scope trx_scope =
      all ? Transaction_ctx::SESSION : Transaction_ctx::STMT;
  Ha_trx_info *ha_info = thd->get_transaction()->ha_trx_info(trx_scope);

  DBUG_TRACE;

  if (ha_info) {
    for (; ha_info && !error; ha_info = ha_info->next()) {
      int err = 0;
      handlerton *ht = ha_info->ht();
      /*
        Do not call two-phase commit if this particular
        transaction is read-only. This allows for simpler
        implementation in engines that are always read-only.
      */
      if (!ha_info->is_trx_read_write()) continue;
      if ((err = ht->prepare(ht, thd, all))) {
        char errbuf[MYSQL_ERRMSG_SIZE];
        my_error(ER_ERROR_DURING_COMMIT, MYF(0), err,
                 my_strerror(errbuf, MYSQL_ERRMSG_SIZE, err));
        error = 1;
      }
      DBUG_ASSERT(!thd->status_var_aggregated);
      thd->status_var.ha_prepare_count++;
    }
    DBUG_EXECUTE_IF("crash_commit_after_prepare", DBUG_SUICIDE(););
  }

  return error;
}

/**
  @note
  according to the sql standard (ISO/IEC 9075-2:2003)
  section "4.33.4 SQL-statements and transaction states",
  SAVEPOINT is *not* transaction-initiating SQL-statement
*/
int ha_savepoint(THD *thd, SAVEPOINT *sv) {
  int error = 0;
  Transaction_ctx::enum_trx_scope trx_scope =
      !thd->in_sub_stmt ? Transaction_ctx::SESSION : Transaction_ctx::STMT;
  Ha_trx_info *ha_info = thd->get_transaction()->ha_trx_info(trx_scope);
  Ha_trx_info *begin_ha_info = ha_info;

  DBUG_TRACE;

  for (; ha_info; ha_info = ha_info->next()) {
    int err;
    handlerton *ht = ha_info->ht();
    DBUG_ASSERT(ht);
    if (!ht->savepoint_set) {
      my_error(ER_CHECK_NOT_IMPLEMENTED, MYF(0), "SAVEPOINT");
      error = 1;
      break;
    }
    if ((err = ht->savepoint_set(
             ht, thd,
             (uchar *)(sv + 1) + ht->savepoint_offset))) {  // cannot happen
      char errbuf[MYSQL_ERRMSG_SIZE];
      my_error(ER_GET_ERRNO, MYF(0), err,
               my_strerror(errbuf, MYSQL_ERRMSG_SIZE, err));
      error = 1;
    }
    DBUG_ASSERT(!thd->status_var_aggregated);
    thd->status_var.ha_savepoint_count++;
  }
  /*
    Remember the list of registered storage engines. All new
    engines are prepended to the beginning of the list.
  */
  sv->ha_list = begin_ha_info;

#ifdef HAVE_PSI_TRANSACTION_INTERFACE
  if (!error && thd->m_transaction_psi != nullptr)
    MYSQL_INC_TRANSACTION_SAVEPOINTS(thd->m_transaction_psi, 1);
#endif

  return error;
}

int ha_release_savepoint(THD *thd, SAVEPOINT *sv) {
  int error = 0;
  Ha_trx_info *ha_info = sv->ha_list;
  DBUG_TRACE;

  for (; ha_info; ha_info = ha_info->next()) {
    int err;
    handlerton *ht = ha_info->ht();
    /* Savepoint life time is enclosed into transaction life time. */
    DBUG_ASSERT(ht);
    if (!ht->savepoint_release) continue;
    if ((err = ht->savepoint_release(
             ht, thd,
             (uchar *)(sv + 1) + ht->savepoint_offset))) {  // cannot happen
      char errbuf[MYSQL_ERRMSG_SIZE];
      my_error(ER_GET_ERRNO, MYF(0), err,
               my_strerror(errbuf, MYSQL_ERRMSG_SIZE, err));
      error = 1;
    }
  }
  DBUG_EXECUTE_IF("fail_ha_release_savepoint", {
    my_error(ER_UNKNOWN_ERROR, MYF(0));
    error = 1;
  });

#ifdef HAVE_PSI_TRANSACTION_INTERFACE
  if (thd->m_transaction_psi != nullptr)
    MYSQL_INC_TRANSACTION_RELEASE_SAVEPOINT(thd->m_transaction_psi, 1);
#endif
  return error;
}

static bool snapshot_handlerton(THD *thd, plugin_ref plugin, void *arg) {
  handlerton *hton = plugin_data<handlerton *>(plugin);
  if (hton->state == SHOW_OPTION_YES && hton->start_consistent_snapshot) {
    hton->start_consistent_snapshot(hton, thd);
    *((bool *)arg) = false;
  }
  return false;
}

int ha_start_consistent_snapshot(THD *thd) {
  bool warn = true;

  plugin_foreach(thd, snapshot_handlerton, MYSQL_STORAGE_ENGINE_PLUGIN, &warn);

  /*
    Same idea as when one wants to CREATE TABLE in one engine which does not
    exist:
  */
  if (warn)
    push_warning(thd, Sql_condition::SL_WARNING, ER_UNKNOWN_ERROR,
                 "This MySQL server does not support any "
                 "consistent-read capable storage engine");
  return 0;
}

static bool flush_handlerton(THD *, plugin_ref plugin, void *arg) {
  handlerton *hton = plugin_data<handlerton *>(plugin);
  if (hton->state == SHOW_OPTION_YES && hton->flush_logs &&
      hton->flush_logs(hton, *(static_cast<bool *>(arg))))
    return true;
  return false;
}

bool ha_flush_logs(bool binlog_group_flush) {
  if (plugin_foreach(nullptr, flush_handlerton, MYSQL_STORAGE_ENGINE_PLUGIN,
                     static_cast<void *>(&binlog_group_flush))) {
    return true;
  }
  return false;
}

/**
  @brief make canonical filename

  @param[in]  file     table handler
  @param[in]  path     original path
  @param[out] tmp_path buffer for canonized path

  @details Lower case db name and table name path parts for
           non file based tables when lower_case_table_names
           is 2 (store as is, compare in lower case).
           Filesystem path prefix (mysql_data_home or tmpdir)
           is left intact.

  @note tmp_path may be left intact if no conversion was
        performed.

  @retval canonized path

  @todo This may be done more efficiently when table path
        gets built. Convert this function to something like
        ASSERT_CANONICAL_FILENAME.
*/
const char *get_canonical_filename(handler *file, const char *path,
                                   char *tmp_path) {
  uint i;
  if (lower_case_table_names != 2 || (file->ha_table_flags() & HA_FILE_BASED))
    return path;

  for (i = 0; i <= mysql_tmpdir_list.max; i++) {
    if (is_prefix(path, mysql_tmpdir_list.list[i])) return path;
  }

  /* Ensure that table handler get path in lower case */
  if (tmp_path != path) my_stpcpy(tmp_path, path);

  /*
    we only should turn into lowercase database/table part
    so start the process after homedirectory
  */
  my_casedn_str(files_charset_info, tmp_path + mysql_data_home_len);
  return tmp_path;
}

class Ha_delete_table_error_handler : public Internal_error_handler {
 public:
  bool handle_condition(THD *, uint, const char *,
                        Sql_condition::enum_severity_level *level,
                        const char *) override {
    /* Downgrade errors to warnings. */
    if (*level == Sql_condition::SL_ERROR) *level = Sql_condition::SL_WARNING;
    return false;
  }
};

/**
  Delete table from the storage engine.

  @param thd                Thread context.
  @param table_type         Handlerton for table's SE.
  @param path               Path to table (without extension).
  @param db                 Table database.
  @param alias              Table name.
  @param table_def          dd::Table object describing the table.
  @param generate_warning   Indicates whether errors during deletion
                            should be reported as warnings.

  @return  0 - in case of success, non-0 in case of failure, ENOENT
           if the file doesn't exists.
*/
int ha_delete_table(THD *thd, handlerton *table_type, const char *path,
                    const char *db, const char *alias,
                    const dd::Table *table_def, bool generate_warning) {
  handler *file;
  char tmp_path[FN_REFLEN];
  int error;
  TABLE dummy_table;
  TABLE_SHARE dummy_share;
  DBUG_TRACE;

  dummy_table.s = &dummy_share;

  /* DB_TYPE_UNKNOWN is used in ALTER TABLE when renaming only .frm files */
  if (table_type == nullptr ||
      !(file =
            get_new_handler((TABLE_SHARE *)nullptr,
                            table_def->partition_type() != dd::Table::PT_NONE,
                            thd->mem_root, table_type))) {
    return ENOENT;
  }

  path = get_canonical_filename(file, path, tmp_path);

  if ((error = file->ha_delete_table(path, table_def)) && generate_warning) {
    /*
      Because file->print_error() use my_error() to generate the error message
      we use an internal error handler to intercept it and store the text
      in a temporary buffer. Later the message will be presented to user
      as a warning.
    */
    Ha_delete_table_error_handler ha_delete_table_error_handler;

    /* Fill up strucutures that print_error may need */
    dummy_share.path.str = const_cast<char *>(path);
    dummy_share.path.length = strlen(path);
    dummy_share.db.str = db;
    dummy_share.db.length = strlen(db);
    dummy_share.table_name.str = alias;
    dummy_share.table_name.length = strlen(alias);
    dummy_table.alias = alias;

    file->change_table_ptr(&dummy_table, &dummy_share);

    /*
      XXX: should we convert *all* errors to warnings here?
      What if the error is fatal?
    */
    thd->push_internal_handler(&ha_delete_table_error_handler);
    file->print_error(error, 0);

    thd->pop_internal_handler();
  }

  destroy(file);

#ifdef HAVE_PSI_TABLE_INTERFACE
  if (likely(error == 0)) {
    /* Table share not available, so check path for temp_table prefix. */
    bool temp_table = (strstr(path, tmp_file_prefix) != nullptr);
    PSI_TABLE_CALL(drop_table_share)
    (temp_table, db, strlen(db), alias, strlen(alias));
  }
#endif

  return error;
}

// Prepare HA_CREATE_INFO to be used by ALTER as well as upgrade code.
void HA_CREATE_INFO::init_create_options_from_share(const TABLE_SHARE *share,
                                                    uint used_fields) {
  if (!(used_fields & HA_CREATE_USED_MIN_ROWS)) min_rows = share->min_rows;

  if (!(used_fields & HA_CREATE_USED_MAX_ROWS)) max_rows = share->max_rows;

  if (!(used_fields & HA_CREATE_USED_AVG_ROW_LENGTH))
    avg_row_length = share->avg_row_length;

  if (!(used_fields & HA_CREATE_USED_DEFAULT_CHARSET))
    default_table_charset = share->table_charset;

  if (!(used_fields & HA_CREATE_USED_KEY_BLOCK_SIZE))
    key_block_size = share->key_block_size;

  if (!(used_fields & HA_CREATE_USED_STATS_SAMPLE_PAGES))
    stats_sample_pages = share->stats_sample_pages;

  if (!(used_fields & HA_CREATE_USED_STATS_AUTO_RECALC))
    stats_auto_recalc = share->stats_auto_recalc;

  if (!(used_fields & HA_CREATE_USED_TABLESPACE))
    tablespace = share->tablespace;

  if (storage_media == HA_SM_DEFAULT)
    storage_media = share->default_storage_media;

  /* Creation of federated table with LIKE clause needs connection string */
  if (!(used_fields & HA_CREATE_USED_CONNECTION))
    connect_string = share->connect_string;

  if (!(used_fields & HA_CREATE_USED_COMMENT)) {
    // Assert to check that used_fields flag and comment are in sync.
    DBUG_ASSERT(!comment.str);
    comment = share->comment;
  }

  if (!(used_fields & HA_CREATE_USED_COMPRESS)) {
    // Assert to check that used_fields flag and compress are in sync
    DBUG_ASSERT(!compress.str);
    compress = share->compress;
  }

  if (!(used_fields & (HA_CREATE_USED_ENCRYPT))) {
    // Assert to check that used_fields flag and encrypt_type are in sync
    DBUG_ASSERT(!encrypt_type.str);
    encrypt_type = share->encrypt_type;
  }

  if (!(used_fields & HA_CREATE_USED_SECONDARY_ENGINE)) {
    DBUG_ASSERT(secondary_engine.str == nullptr);
    secondary_engine = share->secondary_engine;
  }

  if (engine_attribute.str == nullptr)
    engine_attribute = share->engine_attribute;

  if (secondary_engine_attribute.str == nullptr)
    secondary_engine_attribute = share->secondary_engine_attribute;
}

/****************************************************************************
** General handler functions
****************************************************************************/
handler *handler::clone(const char *name, MEM_ROOT *mem_root) {
  DBUG_TRACE;

  handler *new_handler = get_new_handler(
      table->s, (table->s->m_part_info != nullptr), mem_root, ht);

  if (!new_handler) return nullptr;
  if (new_handler->set_ha_share_ref(ha_share)) goto err;

  /*
    Allocate handler->ref here because otherwise ha_open will allocate it
    on this->table->mem_root and we will not be able to reclaim that memory
    when the clone handler object is destroyed.
  */
  if (!(new_handler->ref =
            (uchar *)mem_root->Alloc(ALIGN_SIZE(ref_length) * 2)))
    goto err;
  /*
    TODO: Implement a more efficient way to have more than one index open for
    the same table instance. The ha_open call is not cachable for clone.
  */
  if (new_handler->ha_open(table, name, table->db_stat,
                           HA_OPEN_IGNORE_IF_LOCKED, nullptr))
    goto err;

  return new_handler;

err:
  destroy(new_handler);
  return nullptr;
}

void handler::ha_statistic_increment(
    ulonglong System_status_var::*offset) const {
  if (table && table->in_use) (table->in_use->status_var.*offset)++;
}

THD *handler::ha_thd(void) const {
  DBUG_ASSERT(!table || !table->in_use || table->in_use == current_thd);
  return (table && table->in_use) ? table->in_use : current_thd;
}

void handler::unbind_psi() {
#ifdef HAVE_PSI_TABLE_INTERFACE
  DBUG_ASSERT(m_lock_type == F_UNLCK);
  DBUG_ASSERT(inited == NONE);
  /*
    Notify the instrumentation that this table is not owned
    by this thread any more.
  */
  PSI_TABLE_CALL(unbind_table)(m_psi);
#endif
}

void handler::rebind_psi() {
#ifdef HAVE_PSI_TABLE_INTERFACE
  DBUG_ASSERT(m_lock_type == F_UNLCK);
  DBUG_ASSERT(inited == NONE);
  /*
    Notify the instrumentation that this table is now owned
    by this thread.
  */
  PSI_table_share *share_psi = ha_table_share_psi(table_share);
  m_psi = PSI_TABLE_CALL(rebind_table)(share_psi, this, m_psi);
#endif
}

void handler::start_psi_batch_mode() {
#ifdef HAVE_PSI_TABLE_INTERFACE
  DBUG_ASSERT(m_psi_batch_mode == PSI_BATCH_MODE_NONE);
  DBUG_ASSERT(m_psi_locker == nullptr);
  m_psi_batch_mode = PSI_BATCH_MODE_STARTING;
  m_psi_numrows = 0;
#endif
}

void handler::end_psi_batch_mode() {
#ifdef HAVE_PSI_TABLE_INTERFACE
  DBUG_ASSERT(m_psi_batch_mode != PSI_BATCH_MODE_NONE);
  if (m_psi_locker != nullptr) {
    DBUG_ASSERT(m_psi_batch_mode == PSI_BATCH_MODE_STARTED);
    PSI_TABLE_CALL(end_table_io_wait)(m_psi_locker, m_psi_numrows);
    m_psi_locker = nullptr;
  }
  m_psi_batch_mode = PSI_BATCH_MODE_NONE;
#endif
}

PSI_table_share *handler::ha_table_share_psi(const TABLE_SHARE *share) const {
  return share->m_psi;
}

/*
  Open database handler object.

  Used for opening tables. The name will be the name of the file.
  A table is opened when it needs to be opened. For instance
  when a request comes in for a select on the table (tables are not
  open and closed for each request, they are cached).

  The server opens all tables by calling ha_open() which then calls
  the handler specific open().

  Try O_RDONLY if cannot open as O_RDWR. Don't wait for locks if not
  HA_OPEN_WAIT_IF_LOCKED is set

  @param  [out] table_arg             Table structure.
  @param        name                  Full path of table name.
  @param        mode                  Open mode flags.
  @param        test_if_locked        ?
  @param        table_def             dd::Table object describing table
                                      being open. Can be NULL for temporary
                                      tables created by optimizer.

  @retval >0    Error.
  @retval  0    Success.
*/

int handler::ha_open(TABLE *table_arg, const char *name, int mode,
                     int test_if_locked, const dd::Table *table_def) {
  int error;
  DBUG_TRACE;
  DBUG_PRINT("enter",
             ("name: %s  db_type: %d  db_stat: %d  mode: %d  lock_test: %d",
              name, ht->db_type, table_arg->db_stat, mode, test_if_locked));

  table = table_arg;
  DBUG_ASSERT(table->s == table_share);
  DBUG_ASSERT(m_lock_type == F_UNLCK);
  DBUG_PRINT("info", ("old m_lock_type: %d F_UNLCK %d", m_lock_type, F_UNLCK));
  MEM_ROOT *mem_root = (test_if_locked & HA_OPEN_TMP_TABLE)
                           ? &table->s->mem_root
                           : &table->mem_root;
  DBUG_ASSERT(alloc_root_inited(mem_root));

  if ((error = open(name, mode, test_if_locked, table_def))) {
    if ((error == EACCES || error == EROFS) && mode == O_RDWR &&
        (table->db_stat & HA_TRY_READ_ONLY)) {
      table->db_stat |= HA_READ_ONLY;
      error = open(name, O_RDONLY, test_if_locked, table_def);
    }
  }
  if (error) {
    set_my_errno(error); /* Safeguard */
    DBUG_PRINT("error", ("error: %d  errno: %d", error, errno));
  } else {
    DBUG_ASSERT(m_psi == nullptr);
    DBUG_ASSERT(table_share != nullptr);
#ifdef HAVE_PSI_TABLE_INTERFACE
    PSI_table_share *share_psi = ha_table_share_psi(table_share);
    m_psi = PSI_TABLE_CALL(open_table)(share_psi, this);
#endif

    if (table->s->db_options_in_use & HA_OPTION_READ_ONLY_DATA)
      table->db_stat |= HA_READ_ONLY;
    (void)extra(HA_EXTRA_NO_READCHECK);  // Not needed in SQL

    /* ref is already allocated for us if we're called from handler::clone() */
    if (!ref && !(ref = (uchar *)mem_root->Alloc(ALIGN_SIZE(ref_length) * 2))) {
      ha_close();
      error = HA_ERR_OUT_OF_MEM;
    } else
      dup_ref = ref + ALIGN_SIZE(ref_length);

    // Give the table a defined starting cursor, even if it never actually seeks
    // or writes. This is important for things like weedout on const tables
    // (which is a nonsensical combination, but can happen).
    memset(ref, 0, ref_length);
    cached_table_flags = table_flags();
  }

  return error;
}

/**
  Close handler.

  Called from sql_base.cc, sql_select.cc, and table.cc.
  In sql_select.cc it is only used to close up temporary tables or during
  the process where a temporary table is converted over to being a
  myisam table.
  For sql_base.cc look at close_data_tables().

  @return Operation status
    @retval 0     Success
    @retval != 0  Error (error code returned)
*/

int handler::ha_close(void) {
  DBUG_TRACE;
#ifdef HAVE_PSI_TABLE_INTERFACE
  PSI_TABLE_CALL(close_table)(table_share, m_psi);
  m_psi = nullptr; /* instrumentation handle, invalid after close_table() */
  DBUG_ASSERT(m_psi_batch_mode == PSI_BATCH_MODE_NONE);
  DBUG_ASSERT(m_psi_locker == nullptr);
#endif
  // TODO: set table= NULL to mark the handler as closed?
  DBUG_ASSERT(m_psi == nullptr);
  DBUG_ASSERT(m_lock_type == F_UNLCK);
  DBUG_ASSERT(inited == NONE);
  if (m_unique) {
    // It's allocated on memroot and will be freed along with it
    m_unique->cleanup();
    m_unique = nullptr;
  }
  return close();
}

/**
  Initialize use of index.

  @param idx     Index to use
  @param sorted  Use sorted order

  @return Operation status
    @retval 0     Success
    @retval != 0  Error (error code returned)
*/

int handler::ha_index_init(uint idx, bool sorted) {
  DBUG_EXECUTE_IF("ha_index_init_fail", return HA_ERR_TABLE_DEF_CHANGED;);
  int result;
  DBUG_TRACE;
  DBUG_ASSERT(table_share->tmp_table != NO_TMP_TABLE || m_lock_type != F_UNLCK);
  DBUG_ASSERT(inited == NONE);
  if (!(result = index_init(idx, sorted))) inited = INDEX;
  end_range = nullptr;
  return result;
}

/**
  End use of index.

  @return Operation status
    @retval 0     Success
    @retval != 0  Error (error code returned)
*/

int handler::ha_index_end() {
  DBUG_TRACE;
  /* SQL HANDLER function can call this without having it locked. */
  DBUG_ASSERT(table->open_by_handler ||
              table_share->tmp_table != NO_TMP_TABLE || m_lock_type != F_UNLCK);
  DBUG_ASSERT(inited == INDEX);
  inited = NONE;
  end_range = nullptr;
  m_record_buffer = nullptr;
  if (m_unique) m_unique->reset(false);
  return index_end();
}

/**
  Initialize table for random read or scan.

  @param scan  if true: Initialize for random scans through rnd_next()
               if false: Initialize for random reads through rnd_pos()

  @return Operation status
    @retval 0     Success
    @retval != 0  Error (error code returned)
*/

int handler::ha_rnd_init(bool scan) {
  DBUG_EXECUTE_IF("ha_rnd_init_fail", return HA_ERR_TABLE_DEF_CHANGED;);
  int result;
  DBUG_TRACE;
  DBUG_ASSERT(table_share->tmp_table != NO_TMP_TABLE || m_lock_type != F_UNLCK);
  DBUG_ASSERT(inited == NONE || (inited == RND && scan));
  inited = (result = rnd_init(scan)) ? NONE : RND;
  end_range = nullptr;
  return result;
}

/**
  End use of random access.

  @return Operation status
    @retval 0     Success
    @retval != 0  Error (error code returned)
*/

int handler::ha_rnd_end() {
  DBUG_TRACE;
  /* SQL HANDLER function can call this without having it locked. */
  DBUG_ASSERT(table->open_by_handler ||
              table_share->tmp_table != NO_TMP_TABLE || m_lock_type != F_UNLCK);
  DBUG_ASSERT(inited == RND);
  inited = NONE;
  end_range = nullptr;
  m_record_buffer = nullptr;
  return rnd_end();
}

/**
  Read next row via random scan.

  @param buf  Buffer to read the row into

  @return Operation status
    @retval 0     Success
    @retval != 0  Error (error code returned)
*/

int handler::ha_rnd_next(uchar *buf) {
  int result;
  DBUG_EXECUTE_IF("ha_rnd_next_deadlock", return HA_ERR_LOCK_DEADLOCK;);
  DBUG_TRACE;
  DBUG_ASSERT(table_share->tmp_table != NO_TMP_TABLE || m_lock_type != F_UNLCK);
  DBUG_ASSERT(inited == RND);

  // Set status for the need to update generated fields
  m_update_generated_read_fields = table->has_gcol();

  MYSQL_TABLE_IO_WAIT(PSI_TABLE_FETCH_ROW, MAX_KEY, result,
                      { result = rnd_next(buf); })
  if (!result && m_update_generated_read_fields) {
    result = update_generated_read_fields(buf, table);
    m_update_generated_read_fields = false;
  }
  table->set_row_status_from_handler(result);
  return result;
}

/**
  Read row via random scan from position.

  @param[out] buf  Buffer to read the row into
  @param      pos  Position from position() call

  @return Operation status
    @retval 0     Success
    @retval != 0  Error (error code returned)
*/

int handler::ha_rnd_pos(uchar *buf, uchar *pos) {
  int result;
  DBUG_TRACE;
  DBUG_ASSERT(table_share->tmp_table != NO_TMP_TABLE || m_lock_type != F_UNLCK);
  /* TODO: Find out how to solve ha_rnd_pos when finding duplicate update. */
  /* DBUG_ASSERT(inited == RND); */

  // Set status for the need to update generated fields
  m_update_generated_read_fields = table->has_gcol();

  MYSQL_TABLE_IO_WAIT(PSI_TABLE_FETCH_ROW, MAX_KEY, result,
                      { result = rnd_pos(buf, pos); })
  if (!result && m_update_generated_read_fields) {
    result = update_generated_read_fields(buf, table);
    m_update_generated_read_fields = false;
  }
  table->set_row_status_from_handler(result);
  return result;
}

int handler::ha_ft_read(uchar *buf) {
  int result;
  DBUG_TRACE;

  // Set status for the need to update generated fields
  m_update_generated_read_fields = table->has_gcol();

  result = ft_read(buf);
  if (!result && m_update_generated_read_fields) {
    result = update_generated_read_fields(buf, table);
    m_update_generated_read_fields = false;
  }
  table->set_row_status_from_handler(result);
  return result;
}

int handler::ha_sample_init(void *&scan_ctx, double sampling_percentage,
                            int sampling_seed,
                            enum_sampling_method sampling_method) {
  DBUG_TRACE;
  DBUG_ASSERT(sampling_percentage >= 0.0);
  DBUG_ASSERT(sampling_percentage <= 100.0);
  DBUG_ASSERT(inited == NONE);

  // Initialise the random number generator.
  m_random_number_engine.seed(sampling_seed);
  m_sampling_percentage = sampling_percentage;

  int result = sample_init(scan_ctx, sampling_percentage, sampling_seed,
                           sampling_method);
  inited = (result != 0) ? NONE : SAMPLING;
  return result;
}

int handler::ha_sample_end(void *scan_ctx) {
  DBUG_TRACE;
  DBUG_ASSERT(inited == SAMPLING);
  inited = NONE;
  int result = sample_end(scan_ctx);
  return result;
}

int handler::ha_sample_next(void *scan_ctx, uchar *buf) {
  DBUG_TRACE;
  DBUG_ASSERT(inited == SAMPLING);

  if (m_sampling_percentage == 0.0) return HA_ERR_END_OF_FILE;

  m_update_generated_read_fields = table->has_gcol();

  int result;
  MYSQL_TABLE_IO_WAIT(PSI_TABLE_FETCH_ROW, MAX_KEY, result,
                      { result = sample_next(scan_ctx, buf); })

  if (result == 0 && m_update_generated_read_fields) {
    result = update_generated_read_fields(buf, table);
    m_update_generated_read_fields = false;
  }
  table->set_row_status_from_handler(result);

  return result;
}

int handler::sample_init(void *&scan_ctx MY_ATTRIBUTE((unused)), double, int,
                         enum_sampling_method) {
  return rnd_init(true);
}

int handler::sample_end(void *scan_ctx MY_ATTRIBUTE((unused))) {
  return rnd_end();
}

int handler::sample_next(void *scan_ctx MY_ATTRIBUTE((unused)), uchar *buf) {
  // Temporary set inited to RND, since we are calling rnd_next().
  int res = rnd_next(buf);

  std::uniform_real_distribution<double> rnd(0.0, 1.0);
  while (!res && rnd(m_random_number_engine) > (m_sampling_percentage / 100.0))
    res = rnd_next(buf);

  return res;
}

int handler::records(ha_rows *num_rows) {
  if (ha_table_flags() & HA_COUNT_ROWS_INSTANT) {
    *num_rows = stats.records;
    return 0;
  }

  int error = 0;
  ha_rows rows = 0;
  start_psi_batch_mode();

  if (!(error = ha_rnd_init(true))) {
    while (!table->in_use->killed) {
      DBUG_EXECUTE_IF("bug28079850", table->in_use->killed = THD::KILL_QUERY;);
      if ((error = ha_rnd_next(table->record[0]))) {
        if (error == HA_ERR_RECORD_DELETED)
          continue;
        else
          break;
      }
      ++rows;
    }
  }

  *num_rows = rows;
  end_psi_batch_mode();
  int ha_rnd_end_error = 0;
  if (error != HA_ERR_END_OF_FILE) *num_rows = HA_POS_ERROR;

  // Call ha_rnd_end() only if only if handler has been initialized.
  if (inited && (ha_rnd_end_error = ha_rnd_end())) *num_rows = HA_POS_ERROR;

  return (error != HA_ERR_END_OF_FILE) ? error : ha_rnd_end_error;
}

int handler::records_from_index(ha_rows *num_rows, uint index) {
  if (ha_table_flags() & HA_COUNT_ROWS_INSTANT) {
    *num_rows = stats.records;
    return 0;
  }

  int error = 0;
  ha_rows rows = 0;
  uchar *buf = table->record[0];
  start_psi_batch_mode();

  if (!(error = ha_index_init(index, false))) {
    if (!(error = ha_index_first(buf))) {
      rows = 1;

      while (!table->in_use->killed) {
        DBUG_EXECUTE_IF("bug28079850",
                        table->in_use->killed = THD::KILL_QUERY;);
        if ((error = ha_index_next(buf))) {
          if (error == HA_ERR_RECORD_DELETED)
            continue;
          else
            break;
        }
        ++rows;
      }
    }
  }

  *num_rows = rows;
  end_psi_batch_mode();
  int ha_index_end_error = 0;
  if (error != HA_ERR_END_OF_FILE) *num_rows = HA_POS_ERROR;

  // Call ha_index_end() only if handler has been initialized.
  if (inited && (ha_index_end_error = ha_index_end())) *num_rows = HA_POS_ERROR;

  return (error != HA_ERR_END_OF_FILE) ? error : ha_index_end_error;
}

int handler::handle_records_error(int error, ha_rows *num_rows) {
  // If query was killed set the error since not all storage engines do it.
  if (table->in_use->killed) {
    *num_rows = HA_POS_ERROR;
    if (error == 0) error = HA_ERR_QUERY_INTERRUPTED;
  }

  if (error != 0) DBUG_ASSERT(*num_rows == HA_POS_ERROR);
  if (*num_rows == HA_POS_ERROR) DBUG_ASSERT(error != 0);
  if (error != 0) {
    /*
      ha_innobase::records may have rolled back internally.
      In this case, thd_mark_transaction_to_rollback() will have been called.
      For the errors below, we need to abort right away.
    */
    switch (error) {
      case HA_ERR_LOCK_DEADLOCK:
      case HA_ERR_LOCK_TABLE_FULL:
      case HA_ERR_LOCK_WAIT_TIMEOUT:
      case HA_ERR_QUERY_INTERRUPTED:
        print_error(error, MYF(0));
        return error;
      default:
        return error;
    }
  }
  return 0;
}

/**
  Read [part of] row via [part of] index.
  @param[out] buf          buffer where store the data
  @param      key          Key to search for
  @param      keypart_map  Which part of key to use
  @param      find_flag    Direction/condition on key usage

  @returns Operation status
    @retval  0                   Success (found a record, and function has
                                 set table status to "has row")
    @retval  HA_ERR_END_OF_FILE  Row not found (function has set table status
                                 to "no row"). End of index passed.
    @retval  HA_ERR_KEY_NOT_FOUND Row not found (function has set table status
                                 to "no row"). Index cursor positioned.
    @retval  != 0                Error

  @note Positions an index cursor to the index specified in the handle.
  Fetches the row if available. If the key value is null,
  begin at the first key of the index.
  ha_index_read_map can be restarted without calling index_end on the previous
  index scan and without calling ha_index_init. In this case the
  ha_index_read_map is on the same index as the previous ha_index_scan.
  This is particularly used in conjunction with multi read ranges.
*/

int handler::ha_index_read_map(uchar *buf, const uchar *key,
                               key_part_map keypart_map,
                               enum ha_rkey_function find_flag) {
  int result;
  DBUG_TRACE;
  DBUG_ASSERT(table_share->tmp_table != NO_TMP_TABLE || m_lock_type != F_UNLCK);
  DBUG_ASSERT(inited == INDEX);
  DBUG_ASSERT(!pushed_idx_cond || buf == table->record[0]);

  // Set status for the need to update generated fields
  m_update_generated_read_fields = table->has_gcol();

  MYSQL_TABLE_IO_WAIT(PSI_TABLE_FETCH_ROW, active_index, result, {
    result = index_read_map(buf, key, keypart_map, find_flag);
  })
  if (!result && m_update_generated_read_fields) {
    result = update_generated_read_fields(buf, table, active_index);
    m_update_generated_read_fields = false;
  }
  table->set_row_status_from_handler(result);
  return result;
}

int handler::ha_index_read_last_map(uchar *buf, const uchar *key,
                                    key_part_map keypart_map) {
  int result;
  DBUG_TRACE;
  DBUG_ASSERT(table_share->tmp_table != NO_TMP_TABLE || m_lock_type != F_UNLCK);
  DBUG_ASSERT(inited == INDEX);
  DBUG_ASSERT(!pushed_idx_cond || buf == table->record[0]);

  // Set status for the need to update generated fields
  m_update_generated_read_fields = table->has_gcol();

  MYSQL_TABLE_IO_WAIT(PSI_TABLE_FETCH_ROW, active_index, result,
                      { result = index_read_last_map(buf, key, keypart_map); })
  if (!result && m_update_generated_read_fields) {
    result = update_generated_read_fields(buf, table, active_index);
    m_update_generated_read_fields = false;
  }
  table->set_row_status_from_handler(result);
  return result;
}

/**
  Initializes an index and read it.

  @see handler::ha_index_read_map.
*/

int handler::ha_index_read_idx_map(uchar *buf, uint index, const uchar *key,
                                   key_part_map keypart_map,
                                   enum ha_rkey_function find_flag) {
  int result;
  DBUG_TRACE;
  DBUG_ASSERT(table_share->tmp_table != NO_TMP_TABLE || m_lock_type != F_UNLCK);
  DBUG_ASSERT(end_range == nullptr);
  DBUG_ASSERT(!pushed_idx_cond || buf == table->record[0]);

  // Set status for the need to update generated fields
  m_update_generated_read_fields = table->has_gcol();

  MYSQL_TABLE_IO_WAIT(PSI_TABLE_FETCH_ROW, index, result, {
    result = index_read_idx_map(buf, index, key, keypart_map, find_flag);
  })
  if (!result && m_update_generated_read_fields) {
    result = update_generated_read_fields(buf, table, index);
    m_update_generated_read_fields = false;
  }
  table->set_row_status_from_handler(result);
  return result;
}

/**
  Reads the next row via index.

  @param[out] buf  Row data

  @return Operation status.
    @retval  0                   Success
    @retval  HA_ERR_END_OF_FILE  Row not found
    @retval  != 0                Error
*/

int handler::ha_index_next(uchar *buf) {
  int result;
  DBUG_TRACE;
  DBUG_ASSERT(table_share->tmp_table != NO_TMP_TABLE || m_lock_type != F_UNLCK);
  DBUG_ASSERT(inited == INDEX);
  DBUG_ASSERT(!pushed_idx_cond || buf == table->record[0]);

  // Set status for the need to update generated fields
  m_update_generated_read_fields = table->has_gcol();

  MYSQL_TABLE_IO_WAIT(PSI_TABLE_FETCH_ROW, active_index, result,
                      { result = index_next(buf); })
  if (!result && m_update_generated_read_fields) {
    result = update_generated_read_fields(buf, table, active_index);
    m_update_generated_read_fields = false;
  }
  table->set_row_status_from_handler(result);
  return result;
}

/**
  Reads the previous row via index.

  @param[out] buf  Row data

  @return Operation status.
    @retval  0                   Success
    @retval  HA_ERR_END_OF_FILE  Row not found
    @retval  != 0                Error
*/

int handler::ha_index_prev(uchar *buf) {
  int result;
  DBUG_TRACE;
  DBUG_ASSERT(table_share->tmp_table != NO_TMP_TABLE || m_lock_type != F_UNLCK);
  DBUG_ASSERT(inited == INDEX);
  DBUG_ASSERT(!pushed_idx_cond || buf == table->record[0]);

  // Set status for the need to update generated fields
  m_update_generated_read_fields = table->has_gcol();

  MYSQL_TABLE_IO_WAIT(PSI_TABLE_FETCH_ROW, active_index, result,
                      { result = index_prev(buf); })
  if (!result && m_update_generated_read_fields) {
    result = update_generated_read_fields(buf, table, active_index);
    m_update_generated_read_fields = false;
  }
  table->set_row_status_from_handler(result);
  return result;
}

/**
  Reads the first row via index.

  @param[out] buf  Row data

  @return Operation status.
    @retval  0                   Success
    @retval  HA_ERR_END_OF_FILE  Row not found
    @retval  != 0                Error
*/

int handler::ha_index_first(uchar *buf) {
  int result;
  DBUG_TRACE;
  DBUG_ASSERT(table_share->tmp_table != NO_TMP_TABLE || m_lock_type != F_UNLCK);
  DBUG_ASSERT(inited == INDEX);
  DBUG_ASSERT(!pushed_idx_cond || buf == table->record[0]);

  // Set status for the need to update generated fields
  m_update_generated_read_fields = table->has_gcol();

  MYSQL_TABLE_IO_WAIT(PSI_TABLE_FETCH_ROW, active_index, result,
                      { result = index_first(buf); })
  if (!result && m_update_generated_read_fields) {
    result = update_generated_read_fields(buf, table, active_index);
    m_update_generated_read_fields = false;
  }
  table->set_row_status_from_handler(result);
  return result;
}

/**
  Reads the last row via index.

  @param[out] buf  Row data

  @return Operation status.
    @retval  0                   Success
    @retval  HA_ERR_END_OF_FILE  Row not found
    @retval  != 0                Error
*/

int handler::ha_index_last(uchar *buf) {
  int result;
  DBUG_TRACE;
  DBUG_ASSERT(table_share->tmp_table != NO_TMP_TABLE || m_lock_type != F_UNLCK);
  DBUG_ASSERT(inited == INDEX);
  DBUG_ASSERT(!pushed_idx_cond || buf == table->record[0]);

  // Set status for the need to update generated fields
  m_update_generated_read_fields = table->has_gcol();

  MYSQL_TABLE_IO_WAIT(PSI_TABLE_FETCH_ROW, active_index, result,
                      { result = index_last(buf); })
  if (!result && m_update_generated_read_fields) {
    result = update_generated_read_fields(buf, table, active_index);
    m_update_generated_read_fields = false;
  }
  table->set_row_status_from_handler(result);
  return result;
}

/**
  Reads the next same row via index.

  @param[out] buf     Row data
  @param      key     Key to search for
  @param      keylen  Length of key

  @return Operation status.
    @retval  0                   Success
    @retval  HA_ERR_END_OF_FILE  Row not found
    @retval  != 0                Error
*/

int handler::ha_index_next_same(uchar *buf, const uchar *key, uint keylen) {
  int result;
  DBUG_TRACE;
  DBUG_ASSERT(table_share->tmp_table != NO_TMP_TABLE || m_lock_type != F_UNLCK);
  DBUG_ASSERT(inited == INDEX);
  DBUG_ASSERT(!pushed_idx_cond || buf == table->record[0]);

  // Set status for the need to update generated fields
  m_update_generated_read_fields = table->has_gcol();

  MYSQL_TABLE_IO_WAIT(PSI_TABLE_FETCH_ROW, active_index, result,
                      { result = index_next_same(buf, key, keylen); })
  if (!result && m_update_generated_read_fields) {
    result = update_generated_read_fields(buf, table, active_index);
    m_update_generated_read_fields = false;
  }
  table->set_row_status_from_handler(result);
  return result;
}

/**
  Read first row (only) from a table.