Backport "InnoDB shutdown hangs if innodb_force_recovery >= 3 skipped any rollback" from 5.7

[Laurynas Biveinis]

On 5.6 trunk, innodb.innodb_bug-13628249 fails intermittently with

innodb.innodb_bug-13628249 w4 [ fail ]
        Test ended at 2016-12-02 15:20:31

CURRENT_TEST: innodb.innodb_bug-13628249
mysqltest: In included file "./include/wait_until_connected_again.inc":
included from ./include/wait_until_connected_again.inc at line 25:
At line 25: Server failed to restart

The result from queries just before the failure was:
call mtr.add_suppression('InnoDB: Failed to find tablespace for table \'"mysql"."slave_master_info"\' in the cache');
call mtr.add_suppression('InnoDB: Failed to find tablespace for table \'"mysql"."slave_relay_log_info"\' in the cache');
call mtr.add_suppression('InnoDB: Failed to find tablespace for table \'"mysql"."slave_worker_info"\' in the cache');
call mtr.add_suppression('InnoDB: Failed to find tablespace for table \'"test"."t1"\' in the cache');
call mtr.add_suppression('InnoDB: Allocated tablespace [0-9]+, old maximum was [0-9]+');
# Stop server
# Restart server.
# Restart server.
CREATE TABLE t1(c1 INT PRIMARY KEY) ENGINE=InnoDB STATS_PERSISTENT=0;
BEGIN;
INSERT INTO t1 VALUES(1), (2), (3), (4);
SET SESSION debug="+d,crash_commit_before";
COMMIT;
ERROR HY000: Lost connection to MySQL server during query
SELECT COUNT(*) IN (0,4) yes FROM t1;
yes
1
safe_process[30106]: Child process: 30107, exit: 1
...
2016-12-02 18:12:07 30206 [Note] InnoDB: Starting shutdown...
2016-12-02 18:13:07 30206 [Note] InnoDB: Waiting for 1 active transactions to finish
2016-12-02 18:14:07 30206 [Note] InnoDB: Waiting for 1 active transactions to finish
2016-12-02 18:15:12 30206 [Note] InnoDB: Waiting for 1 active transactions to finish
2016-12-02 18:16:13 30206 [Note] InnoDB: Waiting for 1 active transactions to finish
2016-12-02 18:17:13 30206 [Note] InnoDB: Waiting for 1 active transactions to finish
2016-12-02 18:18:13 30206 [Note] InnoDB: Waiting for 1 active transactions to finish
2016-12-02 18:19:13 30206 [Note] InnoDB: Waiting for 1 active transactions to finish
2016-12-02 18:20:13 30206 [Note] InnoDB: Waiting for 1 active transactions to finish

The fix appears to backport

commit b7a0c7f5527647e8e3bdae9c7bbef922f8a1b5e4
Author: Marko M<C3><A4>kel<C3><A4> <email address hidden>
Date: Tue Apr 14 09:25:23 2015 +0300

    Bug#20874411 INNODB SHUTDOWN HANGS IF INNODB_FORCE_RECOVERY>=3
    SKIPPED ANY ROLLBACK

    trx_sys_any_active_transactions(): If any prepared ACTIVE transactions
    exist, and their rollback was prevented by innodb_force_recovery,
    convert these transactions to XA PREPARE state in the main-memory
    data structures, so that shutdown will proceed normally. These transactions
    will again recover as ACTIVE on the next restart, and they will be rolled
    back unless innodb_force_recovery>=3 again.

    trx_undo_fake_prepared(): An auxiliary function to set an undo log to
    PREPARED state.

    RB: 8610
    Reviewed-by: Jimmy Yang <email address hidden>

[Laurynas Biveinis (laurynas-biveinis)]

     https://github.com/percona/percona-server/pull/1227
     https://github.com/percona/percona-server/pull/1228

[Shahriyar Rzayev (rzayev-sehriyar)]

Percona now uses JIRA for bug reports so this bug report is migrated to: https://jira.percona.com/browse/PS-3615