percona-server-5.7.36-39/sql/xa.cc

/* Copyright (c) 2013, 2021, 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,
   51 Franklin Street, Suite 500, Boston, MA 02110-1335 USA */

#include "xa.h"

#include "hash.h"               // HASH
#include "sql_class.h"          // THD
#include "transaction.h"        // trans_begin, trans_rollback
#include "debug_sync.h"         // DEBUG_SYNC
#include "log.h"                // tc_log
#include "sql_plugin.h"         // plugin_foreach
#include <pfs_transaction_provider.h>
#include <mysql/psi/mysql_transaction.h>
#include "binlog.h"

const char *XID_STATE::xa_state_names[]={
  "NON-EXISTING", "ACTIVE", "IDLE", "PREPARED", "ROLLBACK ONLY"
};

/* for recover() handlerton call */
static const int MIN_XID_LIST_SIZE= 128;
static const int MAX_XID_LIST_SIZE= 1024*128;

static mysql_mutex_t LOCK_transaction_cache;
static HASH transaction_cache;

static my_bool xacommit_handlerton(THD *unused1, plugin_ref plugin,
                                   void *arg)
{
  handlerton *hton= plugin_data<handlerton*>(plugin);
  if (hton->state == SHOW_OPTION_YES && hton->recover)
    hton->commit_by_xid(hton, (XID *)arg);

  return FALSE;
}


static my_bool xarollback_handlerton(THD *unused1, plugin_ref plugin,
                                     void *arg)
{
  handlerton *hton= plugin_data<handlerton*>(plugin);
  if (hton->state == SHOW_OPTION_YES && hton->recover)
    hton->rollback_by_xid(hton, (XID *)arg);

  return FALSE;
}


static void ha_commit_or_rollback_by_xid(THD *thd, XID *xid, bool commit)
{
  plugin_foreach(NULL, commit ? xacommit_handlerton : xarollback_handlerton,
                 MYSQL_STORAGE_ENGINE_PLUGIN, xid);
}


struct xarecover_st
{
  int len, found_foreign_xids, found_my_xids;
  XID *list;
  HASH *commit_list;
  bool dry_run;
};


static my_bool xarecover_handlerton(THD *unused, plugin_ref plugin,
                                    void *arg)
{
  handlerton *hton= plugin_data<handlerton*>(plugin);
  struct xarecover_st *info= (struct xarecover_st *) arg;
  int got;

  if (hton->state == SHOW_OPTION_YES && hton->recover)
  {
    while ((got= hton->recover(hton, info->list, info->len)) > 0)
    {
      sql_print_information("Found %d prepared transaction(s) in %s",
                            got, ha_resolve_storage_engine_name(hton));
      for (int i= 0; i < got; i++)
      {
        my_xid x= info->list[i].get_my_xid();
        if (!x) // not "mine" - that is generated by external TM
        {
#ifndef NDEBUG
          char buf[XIDDATASIZE * 4 + 6]; // see xid_to_str
          XID *xid= info->list + i;
          sql_print_information("ignore xid %s", xid->xid_to_str(buf));
#endif
          transaction_cache_insert_recovery(info->list + i);
          info->found_foreign_xids++;
          continue;
        }
        if (info->dry_run)
        {
          info->found_my_xids++;
          continue;
        }
        // recovery mode
        if (info->commit_list ?
            my_hash_search(info->commit_list, (uchar *)&x, sizeof(x)) != 0 :
            tc_heuristic_recover == TC_HEURISTIC_RECOVER_COMMIT)
        {
#ifndef NDEBUG
          char buf[XIDDATASIZE * 4 + 6]; // see xid_to_str
          XID *xid= info->list + i;
          sql_print_information("commit xid %s", xid->xid_to_str(buf));
#endif
          hton->commit_by_xid(hton, info->list + i);
        }
        else
        {
#ifndef NDEBUG
          char buf[XIDDATASIZE * 4 + 6]; // see xid_to_str
          XID *xid= info->list + i;
          sql_print_information("rollback xid %s", xid->xid_to_str(buf));
#endif
          hton->rollback_by_xid(hton, info->list + i);
        }
      }
      if (got < info->len)
        break;
    }
  }
  return false;
}


int ha_recover(HASH *commit_list)
{
  struct xarecover_st info;
  DBUG_ENTER("ha_recover");
  info.found_foreign_xids= info.found_my_xids= 0;
  info.commit_list= commit_list;
  info.dry_run= (info.commit_list == 0 &&
                 tc_heuristic_recover == TC_HEURISTIC_NOT_USED);
  info.list= NULL;

  /* commit_list and tc_heuristic_recover cannot be set both */
  assert(info.commit_list == 0 ||
         tc_heuristic_recover == TC_HEURISTIC_NOT_USED);
  /* if either is set, total_ha_2pc must be set too */
  assert(info.dry_run || total_ha_2pc>(ulong)opt_bin_log);

  if (total_ha_2pc <= (ulong)opt_bin_log)
    DBUG_RETURN(0);

  if (info.commit_list)
    sql_print_information("Starting crash recovery...");

  if (total_ha_2pc > (ulong)opt_bin_log + 1)
  {
    if (tc_heuristic_recover == TC_HEURISTIC_RECOVER_ROLLBACK)
    {
      sql_print_error("--tc-heuristic-recover rollback strategy is not safe "
                      "on systems with more than one 2-phase-commit-capable "
                      "storage engine. Aborting crash recovery.");
      DBUG_RETURN(1);
    }
  }
  else
  {
    /*
      If there is only one 2pc capable storage engine it is always safe
      to rollback. This setting will be ignored if we are in automatic
      recovery mode.
    */
    tc_heuristic_recover= TC_HEURISTIC_RECOVER_ROLLBACK; // forcing ROLLBACK
    info.dry_run= false;
  }

  for (info.len= MAX_XID_LIST_SIZE ;
       info.list == 0 && info.len > MIN_XID_LIST_SIZE; info.len/= 2)
  {
    info.list= (XID *)my_malloc(key_memory_XID,
                                info.len * sizeof(XID), MYF(0));
  }
  if (!info.list)
  {
    sql_print_error(ER(ER_OUTOFMEMORY),
                    static_cast<int>(info.len * sizeof(XID)));
    DBUG_RETURN(1);
  }

  plugin_foreach(NULL, xarecover_handlerton,
                 MYSQL_STORAGE_ENGINE_PLUGIN, &info);

  my_free(info.list);
  if (info.found_foreign_xids)
    sql_print_warning("Found %d prepared XA transactions",
                      info.found_foreign_xids);
  if (info.dry_run && info.found_my_xids)
  {
    sql_print_error("Found %d prepared transactions! It means that mysqld was "
                    "not shut down properly last time and critical recovery "
                    "information (last binlog or %s file) was manually deleted"
                    " after a crash. You have to start mysqld with "
                    "--tc-heuristic-recover switch to commit or rollback "
                    "pending transactions.",
                    info.found_my_xids, opt_tc_log_file);
    DBUG_RETURN(1);
  }
  if (info.commit_list)
    sql_print_information("Crash recovery finished.");
  DBUG_RETURN(0);
}


bool xa_trans_force_rollback(THD *thd)
{
  /*
    We must reset rm_error before calling ha_rollback(),
    so thd->transaction.xid structure gets reset
    by ha_rollback()/THD::transaction::cleanup().
  */
  thd->get_transaction()->xid_state()->reset_error();
  if (ha_rollback_trans(thd, true))
  {
    my_error(ER_XAER_RMERR, MYF(0));
    return true;
  }
  return false;
}


void cleanup_trans_state(THD *thd)
{
  thd->variables.option_bits&= ~OPTION_BEGIN;
  thd->server_status&=
    ~(SERVER_STATUS_IN_TRANS | SERVER_STATUS_IN_TRANS_READONLY);
  thd->get_transaction()->reset_unsafe_rollback_flags(Transaction_ctx::SESSION);
  DBUG_PRINT("info", ("clearing SERVER_STATUS_IN_TRANS"));
  transaction_cache_delete(thd->get_transaction());
}


/**
  Commit and terminate a XA transaction.

  @param thd    Current thread

  @retval false  Success
  @retval true   Failure
*/

bool Sql_cmd_xa_commit::trans_xa_commit(THD *thd)
{
  bool res= true;
  XID_STATE *xid_state= thd->get_transaction()->xid_state();
  bool gtid_error= false, need_clear_owned_gtid= false;

  DBUG_ENTER("trans_xa_commit");

  assert(!thd->slave_thread || xid_state->get_xid()->is_null() ||
         m_xa_opt == XA_ONE_PHASE);

  if (!xid_state->has_same_xid(m_xid))
  {
    if (!xid_state->has_state(XID_STATE::XA_NOTR))
    {
      my_error(ER_XAER_RMFAIL, MYF(0), xid_state->state_name());

      DBUG_RETURN(true);
    }
    /*
      Note, that there is no race condition here between
      transaction_cache_search and transaction_cache_delete,
      since we always delete our own XID
      (m_xid == thd->transaction().xid_state().m_xid).
      The only case when m_xid != thd->transaction.xid_state.m_xid
      and xid_state->in_thd == 0 is in the function
      transaction_cache_insert_recovery(XID), which is called before starting
      client connections, and thus is always single-threaded.
    */
    Transaction_ctx *transaction= transaction_cache_search(m_xid);
    XID_STATE *xs= (transaction ? transaction->xid_state() : NULL);
    res= !xs || !xs->is_in_recovery();
    if (res)  // todo: fix transaction cleanup, BUG#20451386
    {
      my_error(ER_XAER_NOTA, MYF(0));
      DBUG_RETURN(true);
    }
    else if (thd->in_multi_stmt_transaction_mode())
    {
      my_error(ER_XAER_RMFAIL, MYF(0), xid_state->state_name());
      DBUG_RETURN(true);
    }

    assert(xs->is_in_recovery());
    /*
      Resumed transaction XA-commit.
      The case deals with the "external" XA-commit by either a slave applier
      or a different than XA-prepared transaction session.
    */
    res= xs->xa_trans_rolled_back();

#ifdef HAVE_PSI_TRANSACTION_INTERFACE
    /*
      If the original transaction is not rolled back then initiate a new PSI
      transaction to update performance schema related information.
     */
    if (!res)
    {
      thd->m_transaction_psi= MYSQL_START_TRANSACTION(&thd->m_transaction_state,
                                                      NULL, NULL, thd->tx_isolation,
                                                      thd->tx_read_only, false);
      gtid_set_performance_schema_values(thd);
      MYSQL_SET_TRANSACTION_XID(thd->m_transaction_psi,
                                (const void *)xs->get_xid(),
                                (int)xs->get_state());
    }
#endif
    /*
      xs' is_binlogged() is passed through xid_state's member to low-level
      logging routines for deciding how to log.  The same applies to
      Rollback case.
    */
    if (xs->is_binlogged())
      xid_state->set_binlogged();
    else
      xid_state->unset_binlogged();

    /*
      Acquire metadata lock which will ensure that COMMIT is blocked
      by active FLUSH TABLES WITH READ LOCK (and vice versa COMMIT in
      progress blocks FTWRL).

      We allow FLUSHer to COMMIT; we assume FLUSHer knows what it does.
    */
    MDL_request mdl_request;
    MDL_REQUEST_INIT(&mdl_request,
                     MDL_key::COMMIT, "", "", MDL_INTENTION_EXCLUSIVE,
                     MDL_STATEMENT);
    if (thd->mdl_context.acquire_lock(&mdl_request,
                                      thd->variables.lock_wait_timeout))
    {
      /*
        We can't rollback an XA transaction on lock failure due to
        Innodb redo log and bin log update is involved in rollback.
        Return error to user for a retry.
      */
      my_error(ER_XA_RETRY, MYF(0));
      DBUG_RETURN(true);
    }

    /* Do not execute gtid wrapper whenever 'res' is true (rm error) */
    gtid_error= MY_TEST(commit_owned_gtids(thd,
                                           true, &need_clear_owned_gtid));
    if (gtid_error)
      my_error(ER_XA_RBROLLBACK, MYF(0));
    res= res || gtid_error;

    // todo xa framework: return an error
    ha_commit_or_rollback_by_xid(thd, m_xid, !res);
    xid_state->unset_binlogged();

#ifdef HAVE_PSI_TRANSACTION_INTERFACE
    if (!res)
    {
      if (thd->m_transaction_psi)
      {
        /*
          Set the COMMITTED state in PSI context at the end of committing the
          XA transaction.
        */
        MYSQL_COMMIT_TRANSACTION(thd->m_transaction_psi);
        thd->m_transaction_psi= NULL;
      }
    }
#endif

    transaction_cache_delete(transaction);
    gtid_state_commit_or_rollback(thd, need_clear_owned_gtid, !gtid_error);
    DBUG_RETURN(res);
  }

  if (xid_state->xa_trans_rolled_back())
  {
    xa_trans_force_rollback(thd);
    res= thd->is_error();
  }
  else if (xid_state->has_state(XID_STATE::XA_IDLE) &&
           m_xa_opt == XA_ONE_PHASE)
  {
    int r= ha_commit_trans(thd, true);
    if ((res= MY_TEST(r)))
      my_error(r == 1 ? ER_XA_RBROLLBACK : ER_XAER_RMERR, MYF(0));
  }
  else if (xid_state->has_state(XID_STATE::XA_PREPARED) &&
           m_xa_opt == XA_NONE)
  {
    MDL_request mdl_request;

    /*
      Acquire metadata lock which will ensure that COMMIT is blocked
      by active FLUSH TABLES WITH READ LOCK (and vice versa COMMIT in
      progress blocks FTWRL).

      We allow FLUSHer to COMMIT; we assume FLUSHer knows what it does.
    */
    MDL_REQUEST_INIT(&mdl_request,
                     MDL_key::COMMIT, "", "", MDL_INTENTION_EXCLUSIVE,
                     MDL_STATEMENT);
    if (thd->mdl_context.acquire_lock(&mdl_request,
                                      thd->variables.lock_wait_timeout))
    {
      /*
        We can't rollback an XA transaction on lock failure due to
        Innodb redo log and bin log update are involved in rollback.
        Return error to user for a retry.
      */
      my_error(ER_XA_RETRY, MYF(0));
      DBUG_RETURN(true);
    }

    gtid_error= MY_TEST(commit_owned_gtids(thd, true, &need_clear_owned_gtid));
    if (gtid_error)
    {
      res= true;
      /*
        Failure to store gtid is regarded as a unilateral one of the
        resource manager therefore the transaction is to be rolled back.
        The specified error is the same as @c xa_trans_force_rollback.
        The prepared XA will be rolled back along and so will do Gtid state,
        see ha_rollback_trans().

        Todo/fixme: fix binlogging, "XA rollback" event could be missed out.
        Todo/fixme: as to XAER_RMERR, should not it be XA_RBROLLBACK?
                    Rationale: there's no consistency concern after rollback,
                    unlike what XAER_RMERR suggests.
      */
      ha_rollback_trans(thd, true);
      my_error(ER_XAER_RMERR, MYF(0));
    }
    else
    {
      DBUG_EXECUTE_IF("simulate_crash_on_commit_xa_trx", DBUG_SUICIDE(););
      DEBUG_SYNC(thd, "trans_xa_commit_after_acquire_commit_lock");

      if (tc_log)
        res= MY_TEST(tc_log->commit(thd, /* all */ true));
      else
        res= MY_TEST(ha_commit_low(thd, /* all */ true));

      DBUG_EXECUTE_IF("simulate_xa_commit_log_failure", { res= true; });

      if (res)
        my_error(ER_XAER_RMERR, MYF(0)); // todo/fixme: consider to rollback it
#ifdef HAVE_PSI_TRANSACTION_INTERFACE
      else
      {
        /*
          Since we don't call ha_commit_trans() for prepared transactions,
          we need to explicitly mark the transaction as committed.
        */
        MYSQL_COMMIT_TRANSACTION(thd->m_transaction_psi);
      }

      thd->m_transaction_psi= NULL;
#endif
    }
  }
  else
  {
    assert(!need_clear_owned_gtid);

    my_error(ER_XAER_RMFAIL, MYF(0), xid_state->state_name());
    DBUG_RETURN(true);
  }
  gtid_state_commit_or_rollback(thd, need_clear_owned_gtid, !gtid_error);
  cleanup_trans_state(thd);

  xid_state->set_state(XID_STATE::XA_NOTR);
  xid_state->unset_binlogged();
  trans_track_end_trx(thd);
  /* The transaction should be marked as complete in P_S. */
  assert(thd->m_transaction_psi == NULL || res);
  DBUG_RETURN(res);
}


bool Sql_cmd_xa_commit::execute(THD *thd)
{
  bool st= trans_xa_commit(thd);

  if (!st)
  {
    thd->mdl_context.release_transactional_locks();
    /*
        We've just done a commit, reset transaction
        isolation level and access mode to the session default.
    */
    trans_reset_one_shot_chistics(thd);

    my_ok(thd);
  }
  return st;
}


/**
  Roll back and terminate a XA transaction.

  @param thd    Current thread

  @retval false  Success
  @retval true   Failure
*/

bool Sql_cmd_xa_rollback::trans_xa_rollback(THD *thd)
{
  XID_STATE *xid_state= thd->get_transaction()->xid_state();
  bool need_clear_owned_gtid= false;

  DBUG_ENTER("trans_xa_rollback");

  if (!xid_state->has_same_xid(m_xid))
  {
    if (!xid_state->has_state(XID_STATE::XA_NOTR))
    {
      my_error(ER_XAER_RMFAIL, MYF(0), xid_state->state_name());

      DBUG_RETURN(true);
    }

    Transaction_ctx *transaction= transaction_cache_search(m_xid);
    XID_STATE *xs= (transaction ? transaction->xid_state() : NULL);
    if (!xs || !xs->is_in_recovery())
    {
      my_error(ER_XAER_NOTA, MYF(0));
      DBUG_RETURN(true);
    }
    else if (thd->in_multi_stmt_transaction_mode())
    {
      my_error(ER_XAER_RMFAIL, MYF(0), xid_state->state_name());
      DBUG_RETURN(true);
    }

    bool gtid_error= false;

    assert(xs->is_in_recovery());

    /*
      Acquire metadata lock which will ensure that XA ROLLBACK is blocked
      by active FLUSH TABLES WITH READ LOCK (and vice versa ROLLBACK in
      progress blocks FTWRL). This is to avoid binlog and redo entries
      while a backup is in progress.
    */
    MDL_request mdl_request;
    MDL_REQUEST_INIT(&mdl_request,
                     MDL_key::COMMIT, "", "", MDL_INTENTION_EXCLUSIVE,
                     MDL_STATEMENT);
    if (thd->mdl_context.acquire_lock(&mdl_request,
                                      thd->variables.lock_wait_timeout))
    {
      /*
        We can't rollback an XA transaction on lock failure due to
        Innodb redo log and bin log update is involved in rollback.
        Return error to user for a retry.
      */
      my_error(ER_XAER_RMERR, MYF(0));
      DBUG_RETURN(true);
    }

    /*
      Like in the commit case a failure to store gtid is regarded
      as the resource manager issue.
    */
    if ((gtid_error= commit_owned_gtids(thd, true, &need_clear_owned_gtid)))
      my_error(ER_XA_RBROLLBACK, MYF(0));
    xs->xa_trans_rolled_back();
    if (xs->is_binlogged())
      xid_state->set_binlogged();
    else
      xid_state->unset_binlogged();
    ha_commit_or_rollback_by_xid(thd, m_xid, false);
    xid_state->unset_binlogged();
    transaction_cache_delete(transaction);
    gtid_state_commit_or_rollback(thd, need_clear_owned_gtid, !gtid_error);
    DBUG_RETURN(thd->is_error());
  }

  if (xid_state->has_state(XID_STATE::XA_NOTR) ||
      xid_state->has_state(XID_STATE::XA_ACTIVE))
  {
    my_error(ER_XAER_RMFAIL, MYF(0), xid_state->state_name());
    DBUG_RETURN(true);
  }

  /*
    Acquire metadata lock which will ensure that XA ROLLBACK is blocked
    by active FLUSH TABLES WITH READ LOCK (and vice versa ROLLBACK in
    progress blocks FTWRL). This is to avoid binlog and redo entries
    while a backup is in progress.
  */
  MDL_request mdl_request;
  MDL_REQUEST_INIT(&mdl_request,
                   MDL_key::COMMIT, "", "", MDL_INTENTION_EXCLUSIVE,
                   MDL_STATEMENT);
  if (thd->mdl_context.acquire_lock(&mdl_request,
                                    thd->variables.lock_wait_timeout))
  {
    /*
      We can't rollback an XA transaction on lock failure due to
      Innodb redo log and bin log update is involved in rollback.
      Return error to user for a retry.
    */
    my_error(ER_XAER_RMERR, MYF(0));
    DBUG_RETURN(true);
  }

  bool gtid_error= MY_TEST(commit_owned_gtids(thd, true,
                                              &need_clear_owned_gtid));
  bool res= xa_trans_force_rollback(thd) || gtid_error;
  gtid_state_commit_or_rollback(thd, need_clear_owned_gtid, !gtid_error);
  // todo: report a bug in that the raised rm_error in this branch
  //       is masked unlike the "external" rollback branch above.
  DBUG_EXECUTE_IF("simulate_xa_rm_error",
                  {
                    my_error(ER_XA_RBROLLBACK, MYF(0));
                    res= true;
                  });

  cleanup_trans_state(thd);

  xid_state->set_state(XID_STATE::XA_NOTR);
  xid_state->unset_binlogged();
  trans_track_end_trx(thd);
  /* The transaction should be marked as complete in P_S. */
  assert(thd->m_transaction_psi == NULL);
  DBUG_RETURN(res);
}


bool Sql_cmd_xa_rollback::execute(THD *thd)
{
  bool st= trans_xa_rollback(thd);

  if (!st)
  {
    thd->mdl_context.release_transactional_locks();
    /*
      We've just done a rollback, reset transaction
      isolation level and access mode to the session default.
    */
    trans_reset_one_shot_chistics(thd);
    my_ok(thd);
  }

  DBUG_EXECUTE_IF("crash_after_xa_rollback", DBUG_SUICIDE(););

  return st;
}


/**
  Start a XA transaction with the given xid value.

  @param thd    Current thread

  @retval false  Success
  @retval true   Failure
*/

bool Sql_cmd_xa_start::trans_xa_start(THD *thd)
{
  XID_STATE *xid_state= thd->get_transaction()->xid_state();
  DBUG_ENTER("trans_xa_start");

  if (xid_state->has_state(XID_STATE::XA_IDLE) &&
      m_xa_opt == XA_RESUME)
  {
    bool not_equal= !xid_state->has_same_xid(m_xid);
    if (not_equal)
      my_error(ER_XAER_NOTA, MYF(0));
    else
    {
      xid_state->set_state(XID_STATE::XA_ACTIVE);
      MYSQL_SET_TRANSACTION_XA_STATE(thd->m_transaction_psi,
                                  (int)thd->get_transaction()->xid_state()->get_state());
    }
    DBUG_RETURN(not_equal);
  }

  /* TODO: JOIN is not supported yet. */
  if (m_xa_opt != XA_NONE)
    my_error(ER_XAER_INVAL, MYF(0));
  else if (!xid_state->has_state(XID_STATE::XA_NOTR))
    my_error(ER_XAER_RMFAIL, MYF(0), xid_state->state_name());
  else if (thd->locked_tables_mode || thd->in_active_multi_stmt_transaction())
    my_error(ER_XAER_OUTSIDE, MYF(0));
  else if (!trans_begin(thd))
  {
    xid_state->start_normal_xa(m_xid);
    MYSQL_SET_TRANSACTION_XID(thd->m_transaction_psi,
                              (const void *)xid_state->get_xid(),
                              (int)xid_state->get_state());
    if (transaction_cache_insert(m_xid, thd->get_transaction()))
    {
      xid_state->reset();
      trans_rollback(thd);
    }
  }

  DBUG_RETURN(thd->is_error() ||
              !xid_state->has_state(XID_STATE::XA_ACTIVE));
}


bool Sql_cmd_xa_start::execute(THD *thd)
{
  bool st= trans_xa_start(thd);

  if (!st)
  {
    thd->rpl_detach_engine_ha_data();
    my_ok(thd);
  }

  return st;
}


/**
  Put a XA transaction in the IDLE state.

  @param thd    Current thread

  @retval false  Success
  @retval true   Failure
*/

bool Sql_cmd_xa_end::trans_xa_end(THD *thd)
{
  XID_STATE *xid_state= thd->get_transaction()->xid_state();
  DBUG_ENTER("trans_xa_end");

  /* TODO: SUSPEND and FOR MIGRATE are not supported yet. */
  if (m_xa_opt != XA_NONE)
    my_error(ER_XAER_INVAL, MYF(0));
  else if (!xid_state->has_state(XID_STATE::XA_ACTIVE) &&
           !xid_state->has_state(XID_STATE::XA_ROLLBACK_ONLY))
    my_error(ER_XAER_RMFAIL, MYF(0), xid_state->state_name());
  else if (!xid_state->has_same_xid(m_xid))
    my_error(ER_XAER_NOTA, MYF(0));
  else if (!xid_state->xa_trans_rolled_back())
  {
    xid_state->set_state(XID_STATE::XA_IDLE);
    MYSQL_SET_TRANSACTION_XA_STATE(thd->m_transaction_psi,
                                   (int)xid_state->get_state());
  }
  else
  {
    MYSQL_SET_TRANSACTION_XA_STATE(thd->m_transaction_psi,
                                   (int)xid_state->get_state());
  }

  DBUG_RETURN(thd->is_error() ||
              !xid_state->has_state(XID_STATE::XA_IDLE));
}


bool Sql_cmd_xa_end::execute(THD *thd)
{
  bool st= trans_xa_end(thd);

  if (!st)
    my_ok(thd);

  return st;
}


/**
  Put a XA transaction in the PREPARED state.

  @param thd    Current thread

  @retval false  Success
  @retval true   Failure
*/

bool Sql_cmd_xa_prepare::trans_xa_prepare(THD *thd)
{
  XID_STATE *xid_state= thd->get_transaction()->xid_state();
  DBUG_ENTER("trans_xa_prepare");

  if (!xid_state->has_state(XID_STATE::XA_IDLE))
    my_error(ER_XAER_RMFAIL, MYF(0), xid_state->state_name());
  else if (!xid_state->has_same_xid(m_xid))
    my_error(ER_XAER_NOTA, MYF(0));
  else if (thd->slave_thread &&
           is_transaction_empty(
               thd)) // No changes in none of the storage engine
                     // means, filtered statements in the slave
    my_error(ER_XA_REPLICATION_FILTERS,
             MYF(0)); // Empty XA transactions not allowed
  else {
    /*
      Acquire metadata lock which will ensure that XA PREPARE is blocked
      by active FLUSH TABLES WITH READ LOCK (and vice versa PREPARE in
      progress blocks FTWRL). This is to avoid binlog and redo entries
      while a backup is in progress.
    */
    MDL_request mdl_request;
    MDL_REQUEST_INIT(&mdl_request,
                     MDL_key::COMMIT, "", "", MDL_INTENTION_EXCLUSIVE,
                     MDL_STATEMENT);
    if (thd->mdl_context.acquire_lock(&mdl_request,
                                      thd->variables.lock_wait_timeout) ||
        ha_prepare(thd))
    {
      /*
        Rollback the transaction if lock failed. For ha_prepare() failure
        scenarios, transaction is already rolled back by ha_prepare().
      */
      if (!mdl_request.ticket)
        ha_rollback_trans(thd, true);

#ifdef HAVE_PSI_TRANSACTION_INTERFACE
      assert(thd->m_transaction_psi == NULL);
#endif

      /*
        Reset rm_error in case ha_prepare() returned error,
        so thd->transaction.xid structure gets reset
        by THD::transaction::cleanup().
      */
      thd->get_transaction()->xid_state()->reset_error();
      cleanup_trans_state(thd);
      xid_state->set_state(XID_STATE::XA_NOTR);
      thd->get_transaction()->cleanup();
      my_error(ER_XA_RBROLLBACK, MYF(0));
    }
    else
    {
      xid_state->set_state(XID_STATE::XA_PREPARED);
      MYSQL_SET_TRANSACTION_XA_STATE(thd->m_transaction_psi,
                                     (int)xid_state->get_state());
      if (thd->rpl_thd_ctx.session_gtids_ctx().notify_after_xa_prepare(thd))
        sql_print_warning("Failed to collect GTID to send in the response packet!");
    }
  }

  DBUG_RETURN(thd->is_error() ||
              !xid_state->has_state(XID_STATE::XA_PREPARED));
}


bool Sql_cmd_xa_prepare::execute(THD *thd)
{
  bool st= trans_xa_prepare(thd);

  if (!st)
  {
    if (!thd->rpl_unflag_detached_engine_ha_data() ||
        !(st= applier_reset_xa_trans(thd)))
      my_ok(thd);
  }

  return st;
}


/**
  Return the list of XID's to a client, the same way SHOW commands do.

  @param thd    Current thread

  @retval false  Success
  @retval true   Failure

  @note
    I didn't find in XA specs that an RM cannot return the same XID twice,
    so trans_xa_recover does not filter XID's to ensure uniqueness.
    It can be easily fixed later, if necessary.
*/

bool Sql_cmd_xa_recover::trans_xa_recover(THD *thd)
{
  List<Item> field_list;
  Protocol *protocol= thd->get_protocol();
  int i= 0;
  Transaction_ctx *transaction;

  DBUG_ENTER("trans_xa_recover");

  field_list.push_back(new Item_int(NAME_STRING("formatID"), 0,
                                    MY_INT32_NUM_DECIMAL_DIGITS));
  field_list.push_back(new Item_int(NAME_STRING("gtrid_length"), 0,
                                    MY_INT32_NUM_DECIMAL_DIGITS));
  field_list.push_back(new Item_int(NAME_STRING("bqual_length"), 0,
                                    MY_INT32_NUM_DECIMAL_DIGITS));
  field_list.push_back(new Item_empty_string("data", XIDDATASIZE*2+2));

  if (thd->send_result_metadata(&field_list,
                                Protocol::SEND_NUM_ROWS | Protocol::SEND_EOF))
    DBUG_RETURN(true);

  mysql_mutex_lock(&LOCK_transaction_cache);

  while ((transaction= (Transaction_ctx*) my_hash_element(&transaction_cache,
                                                         i++)))
  {
    XID_STATE *xs= transaction->xid_state();
    if (xs->has_state(XID_STATE::XA_PREPARED))
    {
      protocol->start_row();
      xs->store_xid_info(protocol, m_print_xid_as_hex);

      if (protocol->end_row())
      {
        mysql_mutex_unlock(&LOCK_transaction_cache);
        DBUG_RETURN(true);
      }
    }
  }

  mysql_mutex_unlock(&LOCK_transaction_cache);
  my_eof(thd);
  DBUG_RETURN(false);
}


bool Sql_cmd_xa_recover::execute(THD *thd)
{
  bool st= trans_xa_recover(thd);
  DBUG_EXECUTE_IF("crash_after_xa_recover", {DBUG_SUICIDE();});

  return st;
}


bool XID_STATE::xa_trans_rolled_back()
{
  DBUG_EXECUTE_IF("simulate_xa_rm_error", rm_error= true;);
  if (rm_error)
  {
    switch (rm_error)
    {
    case ER_LOCK_WAIT_TIMEOUT:
      my_error(ER_XA_RBTIMEOUT, MYF(0));
      break;
    case ER_LOCK_DEADLOCK:
      my_error(ER_XA_RBDEADLOCK, MYF(0));
      break;
    default:
      my_error(ER_XA_RBROLLBACK, MYF(0));
    }
    xa_state= XID_STATE::XA_ROLLBACK_ONLY;
  }

  return (xa_state == XID_STATE::XA_ROLLBACK_ONLY);
}


bool XID_STATE::check_xa_idle_or_prepared(bool report_error) const
{
  if (xa_state == XA_IDLE ||
      xa_state == XA_PREPARED)
  {
    if (report_error)
      my_error(ER_XAER_RMFAIL, MYF(0), xa_state_names[xa_state]);

    return true;
  }

  return false;
}


bool XID_STATE::check_has_uncommitted_xa() const
{
  if (xa_state == XA_IDLE ||
      xa_state == XA_PREPARED ||
      xa_state == XA_ROLLBACK_ONLY)
  {
    my_error(ER_XAER_RMFAIL, MYF(0), xa_state_names[xa_state]);
    return true;
  }

  return false;
}


bool XID_STATE::check_in_xa(bool report_error) const
{
  if (xa_state != XA_NOTR)
  {
    if (report_error)
      my_error(ER_XAER_RMFAIL, MYF(0), xa_state_names[xa_state]);
    return true;
  }

  return false;
}


void XID_STATE::set_error(THD *thd)
{
  if (xa_state != XA_NOTR)
    rm_error= thd->get_stmt_da()->mysql_errno();
}


void XID_STATE::store_xid_info(Protocol *protocol, bool print_xid_as_hex) const
{
  protocol->store_longlong(static_cast<longlong>(m_xid.formatID), false);
  protocol->store_longlong(static_cast<longlong>(m_xid.gtrid_length), false);
  protocol->store_longlong(static_cast<longlong>(m_xid.bqual_length), false);

  if (print_xid_as_hex)
  {
    /*
      xid_buf contains enough space for 0x followed by HEX representation
      of the binary XID data and one null termination character.
    */
    char xid_buf[XIDDATASIZE * 2 + 2 + 1];

    xid_buf[0]= '0';
    xid_buf[1]= 'x';

    size_t xid_str_len= bin_to_hex_str(xid_buf + 2, sizeof(xid_buf) - 2,
                                       const_cast<char*>(m_xid.data),
                                       m_xid.gtrid_length +
                                       m_xid.bqual_length) + 2;
    protocol->store(xid_buf, xid_str_len, &my_charset_bin);
  }
  else
  {
    protocol->store(m_xid.data, m_xid.gtrid_length + m_xid.bqual_length,
                    &my_charset_bin);
  }


}


#ifndef NDEBUG
char* XID::xid_to_str(char *buf) const
{
  char *s= buf;
  *s++= '\'';

  for (int i= 0; i < gtrid_length + bqual_length; i++)
  {
    /* is_next_dig is set if next character is a number */
    bool is_next_dig= false;
    if (i < XIDDATASIZE)
    {
      char ch= data[i+1];
      is_next_dig= (ch >= '0' && ch <='9');
    }
    if (i == gtrid_length)
    {
      *s++= '\'';
      if (bqual_length)
      {
        *s++= '.';
        *s++= '\'';
      }
    }
    uchar c= static_cast<uchar>(data[i]);
    if (c < 32 || c > 126)
    {
      *s++= '\\';
      /*
        If next character is a number, write current character with
        3 octal numbers to ensure that the next number is not seen
        as part of the octal number
      */
      if (c > 077 || is_next_dig)
        *s++= _dig_vec_lower[c >> 6];
      if (c > 007 || is_next_dig)
        *s++=_dig_vec_lower[(c >> 3) & 7];
      *s++= _dig_vec_lower[c & 7];
    }
    else
    {
      if (c == '\'' || c == '\\')
        *s++= '\\';
      *s++= c;
    }
  }
  *s++= '\'';
  *s= 0;
  return buf;
}
#endif


extern "C" uchar *transaction_get_hash_key(const uchar *, size_t *, my_bool);
extern "C" void transaction_free_hash(void *);


/**
  Callback that is called to get the key for a hash.

  @param ptr  pointer to the record
  @param length  length of the record

  @return  pointer to a record stored in cache
*/

extern "C" uchar *transaction_get_hash_key(const uchar *ptr, size_t *length,
                                           my_bool not_used MY_ATTRIBUTE((unused)))
{
  *length= ((Transaction_ctx*)ptr)->xid_state()->get_xid()->key_length();
  return ((Transaction_ctx*)ptr)->xid_state()->get_xid()->key();
}


/**
  Callback that is called to do cleanup.

  @param ptr  pointer to free
*/

void transaction_free_hash(void *ptr)
{
  Transaction_ctx *transaction= (Transaction_ctx*)ptr;
  // Only time it's allocated is during recovery process.
  if (transaction->xid_state()->is_in_recovery())
    delete transaction;
}


#ifdef HAVE_PSI_INTERFACE
static PSI_mutex_key key_LOCK_transaction_cache;

static PSI_mutex_info transaction_cache_mutexes[]=
{
  { &key_LOCK_transaction_cache, "LOCK_transaction_cache", PSI_FLAG_GLOBAL}
};

static void init_transaction_cache_psi_keys(void)
{
  const char* category= "sql";
  int count;

  count= array_elements(transaction_cache_mutexes);
  mysql_mutex_register(category, transaction_cache_mutexes, count);
}
#endif /* HAVE_PSI_INTERFACE */


bool transaction_cache_init()
{
#ifdef HAVE_PSI_INTERFACE
  init_transaction_cache_psi_keys();
#endif

  mysql_mutex_init(key_LOCK_transaction_cache, &LOCK_transaction_cache,
                   MY_MUTEX_INIT_FAST);
  return my_hash_init(&transaction_cache, &my_charset_bin, 100, 0, 0,
                      transaction_get_hash_key, transaction_free_hash, 0,
                      key_memory_XID) != 0;
}

void transaction_cache_free()
{
  if (my_hash_inited(&transaction_cache))
  {
    my_hash_free(&transaction_cache);
    mysql_mutex_destroy(&LOCK_transaction_cache);
  }
}


Transaction_ctx *transaction_cache_search(XID *xid)
{
  mysql_mutex_lock(&LOCK_transaction_cache);

  Transaction_ctx *res=
      (Transaction_ctx *)my_hash_search(&transaction_cache,
                                        xid->key(),
                                        xid->key_length());
  mysql_mutex_unlock(&LOCK_transaction_cache);
  return res;
}


bool transaction_cache_insert(XID *xid, Transaction_ctx *transaction)
{
  mysql_mutex_lock(&LOCK_transaction_cache);
  if (my_hash_search(&transaction_cache, xid->key(),
                     xid->key_length()))
  {
    mysql_mutex_unlock(&LOCK_transaction_cache);
    my_error(ER_XAER_DUPID, MYF(0));
    return true;
  }
  bool res= my_hash_insert(&transaction_cache, (uchar*)transaction);
  mysql_mutex_unlock(&LOCK_transaction_cache);
  return res;
}


inline bool create_and_insert_new_transaction(XID *xid, bool is_binlogged_arg)
{
  Transaction_ctx *transaction= new (std::nothrow) Transaction_ctx();
  XID_STATE *xs;

  if (!transaction)
  {
    my_error(ER_OUTOFMEMORY, MYF(ME_FATALERROR), sizeof(Transaction_ctx));
    return true;
  }
  xs= transaction->xid_state();
  xs->start_recovery_xa(xid, is_binlogged_arg);

  return my_hash_insert(&transaction_cache, (uchar*)transaction);
}


bool transaction_cache_detach(Transaction_ctx *transaction)
{
  bool res= false;
  XID_STATE *xs= transaction->xid_state();
  XID xid= *(xs->get_xid());
  bool was_logged= xs->is_binlogged();


  assert(xs->has_state(XID_STATE::XA_PREPARED));

  mysql_mutex_lock(&LOCK_transaction_cache);

  assert(my_hash_search(&transaction_cache, xid.key(),
                        xid.key_length()));

  my_hash_delete(&transaction_cache, (uchar *)transaction);
  res= create_and_insert_new_transaction(&xid, was_logged);

  mysql_mutex_unlock(&LOCK_transaction_cache);

  return res;
}


bool transaction_cache_insert_recovery(XID *xid)
{
  mysql_mutex_lock(&LOCK_transaction_cache);

  if (my_hash_search(&transaction_cache, xid->key(),
                     xid->key_length()))
  {
    mysql_mutex_unlock(&LOCK_transaction_cache);
    return false;
  }

  /*
    It's assumed that XA transaction was binlogged before the server
    shutdown. If --log-bin has changed since that from OFF to ON, XA
    COMMIT or XA ROLLBACK of this transaction may be logged alone into
    the binary log.
  */
  bool res= create_and_insert_new_transaction(xid, true);

  mysql_mutex_unlock(&LOCK_transaction_cache);

  return res;
}


void transaction_cache_delete(Transaction_ctx *transaction)
{
  mysql_mutex_lock(&LOCK_transaction_cache);
  my_hash_delete(&transaction_cache, (uchar *)transaction);
  mysql_mutex_unlock(&LOCK_transaction_cache);
}


/**
  The function restores previously saved storage engine transaction context.

  @param     thd     Thread context
*/
static void attach_native_trx(THD *thd)
{
  Ha_trx_info *ha_info=
    thd->get_transaction()->ha_trx_info(Transaction_ctx::SESSION);
  Ha_trx_info *ha_info_next;

  if (ha_info)
  {
    for (; ha_info; ha_info= ha_info_next)
    {
      handlerton *hton= ha_info->ht();
      reattach_engine_ha_data_to_thd(thd, hton);
      ha_info_next= ha_info->next();
      ha_info->reset();
    }
  }
  else
  {
    /*
      Although the current `Ha_trx_info` object is null, we need to make sure
      that the data engine plugins have the oportunity to attach their internal
      transactions and clean up the session.
     */
    thd->rpl_reattach_engine_ha_data();
  }
}


/**
  This is a specific to "slave" applier collection of standard cleanup
  actions to reset XA transaction states at the end of XA prepare rather than
  to do it at the transaction commit, see @c ha_commit_one_phase.
  THD of the slave applier is dissociated from a transaction object in engine
  that continues to exist there.

  @param  THD current thread
  @return the value of is_error()
*/

bool applier_reset_xa_trans(THD *thd)
{
  Transaction_ctx *trn_ctx= thd->get_transaction();
  XID_STATE *xid_state= trn_ctx->xid_state();
  /*
    In the following the server transaction state gets reset for
    a slave applier thread similarly to xa_commit logics
    except commit does not run.
  */
  thd->variables.option_bits&= ~OPTION_BEGIN;
  trn_ctx->reset_unsafe_rollback_flags(Transaction_ctx::STMT);
  thd->server_status&= ~SERVER_STATUS_IN_TRANS;
  /* Server transaction ctx is detached from THD */
  transaction_cache_detach(trn_ctx);
  xid_state->reset();
  /*
     The current engine transactions is detached from THD, and
     previously saved is restored.
  */
  attach_native_trx(thd);
  trn_ctx->set_ha_trx_info(Transaction_ctx::SESSION, NULL);
  trn_ctx->set_no_2pc(Transaction_ctx::SESSION, false);
  trn_ctx->cleanup();
#ifdef HAVE_PSI_TRANSACTION_INTERFACE
  thd->m_transaction_psi= NULL;
#endif
  thd->mdl_context.release_transactional_locks();
  /*
    On client sessions a XA PREPARE will always be followed by a XA COMMIT
    or a XA ROLLBACK, and both statements will reset the tx isolation level
    and access mode when the statement is finishing a transaction.

    For replicated workload it is possible to have other transactions between
    the XA PREPARE and the XA [COMMIT|ROLLBACK].

    So, if the slave applier changed the current transaction isolation level,
    it needs to be restored to the session default value after having the
    XA transaction prepared.
  */
  trans_reset_one_shot_chistics(thd);

  return thd->is_error();
}


/**
  The function detaches existing storage engines transaction
  context from thd. Backup area to save it is provided to low level
  storage engine function.

  is invoked by plugin_foreach() after
  trans_xa_start() for each storage engine.

  @param[in,out]     thd     Thread context
  @param             plugin  Reference to handlerton

  @return    FALSE   on success, TRUE otherwise.
*/

my_bool detach_native_trx(THD *thd, plugin_ref plugin, void *unused)
{
  handlerton *hton= plugin_data<handlerton *>(plugin);

  if (hton->replace_native_transaction_in_thd)
  {
    /* Ensure any active backup engine ha_data won't be overwritten */
    assert(!thd->ha_data[hton->slot].ha_ptr_backup);

    hton->replace_native_transaction_in_thd(thd, NULL,
                                            thd_ha_data_backup(thd, hton));
  }

  return FALSE;
}

my_bool reattach_native_trx(THD *thd, plugin_ref plugin, void *)
{
  DBUG_ENTER("reattach_native_trx");
  handlerton *hton = plugin_data<handlerton *>(plugin);

  if (hton->replace_native_transaction_in_thd)
  {
    /* restore the saved original engine transaction's link with thd */
    void **trx_backup = &thd->ha_data[hton->slot].ha_ptr_backup;

    hton->replace_native_transaction_in_thd(thd, *trx_backup, NULL);
    *trx_backup = NULL;
  }
  DBUG_RETURN(FALSE);
}