Merge 10.3 into 10.4

mysql-test/suite/galera/disabled.def

@@ -14,7 +14,6 @@ GCF-1081 : MDEV-18283 Galera test failure on galera.GCF-1081
 GCF-939 : MDEV-21520 galera.GCF-939
 MW-328A : MDEV-22666 galera.MW-328A MTR failed: "Semaphore wait has lasted > 600 seconds" and do not release port 16002
 MW-328B : MDEV-22666 galera.MW-328A MTR failed: "Semaphore wait has lasted > 600 seconds" and do not release port 16002
-MW-328D : MDEV-27550 ER_LOCK_DEADLOCK is gone after MDEV-27025
 MW-329 : MDEV-19962 Galera test failure on MW-329
 galera_as_slave_replication_bundle : MDEV-15785 OPTION_GTID_BEGIN is set in Gtid_log_event::do_apply_event()
 galera_bf_abort_group_commit : MDEV-18282 Galera test failure on galera.galera_bf_abort_group_commit

mysql-test/suite/innodb/r/lock_delete_updated.result

+CREATE TABLE t(a INT PRIMARY KEY) ENGINE=InnoDB;
+INSERT INTO t VALUES (3);
+BEGIN;
+connection default;
+UPDATE t SET a = 2;
+connect con1,localhost,root;
+DELETE FROM t;
+connection default;
+UPDATE t SET a = 1;
+COMMIT;
+connection con1;
+ERROR 40001: Deadlock found when trying to get lock; try restarting transaction
+disconnect con1;
+connection default;
+# The above DELETE must delete all the rows in the table, so the
+# following SELECT must show 0 rows.
+SELECT count(*) FROM t;
+count(*)
+1
+DROP TABLE t;

mysql-test/suite/innodb/r/lock_wait_conflict.result

-#
-# MDEV-27025 insert-intention lock conflicts with waiting ORDINARY lock
-#
-CREATE TABLE t (a INT PRIMARY KEY, b INT NOT NULL UNIQUE) ENGINE=InnoDB;
-connect prevent_purge,localhost,root,,;
-start transaction with consistent snapshot;
-connection default;
-INSERT INTO t VALUES (20,20);
-DELETE FROM t WHERE b = 20;
-connect con_ins,localhost,root,,;
-SET DEBUG_SYNC = 'row_ins_sec_index_entry_dup_locks_created SIGNAL ins_set_locks WAIT_FOR ins_cont';
-INSERT INTO t VALUES(10, 20);
-connect con_del,localhost,root,,;
-SET DEBUG_SYNC = 'now WAIT_FOR ins_set_locks';
-SET DEBUG_SYNC = 'lock_wait_suspend_thread_enter SIGNAL del_locked';
-DELETE FROM t WHERE b = 20;
-connection default;
-SET DEBUG_SYNC = 'now WAIT_FOR del_locked';
-SET DEBUG_SYNC = 'now SIGNAL ins_cont';
-connection con_ins;
-disconnect con_ins;
-connection con_del;
-disconnect con_del;
-disconnect prevent_purge;
-connection default;
-SET DEBUG_SYNC = 'RESET';
-DROP TABLE t;

mysql-test/suite/innodb/t/lock_delete_updated.test

+--source include/have_innodb.inc
+--source include/count_sessions.inc
+
+CREATE TABLE t(a INT PRIMARY KEY) ENGINE=InnoDB;
+INSERT INTO t VALUES (3);
+
+BEGIN;
+
+connection default;
+UPDATE t SET a = 2;
+
+connect con1,localhost,root;
+send DELETE FROM t;
+
+connection default;
+let $wait_condition=
+    select count(*) = 1 from information_schema.processlist
+    where state = "Updating" and info = "DELETE FROM t";
+--source include/wait_condition.inc
+
+UPDATE t SET a = 1;
+COMMIT;
+
+connection con1;
+error ER_LOCK_DEADLOCK;
+reap;
+disconnect con1;
+
+connection default;
+--echo # The above DELETE must delete all the rows in the table, so the
+--echo # following SELECT must show 0 rows.
+SELECT count(*) FROM t;
+DROP TABLE t;
+--source include/wait_until_count_sessions.inc

mysql-test/suite/innodb/t/lock_wait_conflict.test

---source include/have_innodb.inc
---source include/count_sessions.inc
---source include/have_debug.inc
---source include/have_debug_sync.inc
-
---echo #
---echo # MDEV-27025 insert-intention lock conflicts with waiting ORDINARY lock
---echo #
-
-# The test checks the ability to acquire exclusive record lock if the acquiring
-# transaction already holds a shared lock on the record and another transaction
-# is waiting for a lock.
-
-CREATE TABLE t (a INT PRIMARY KEY, b INT NOT NULL UNIQUE) ENGINE=InnoDB;
-
---connect(prevent_purge,localhost,root,,)
-start transaction with consistent snapshot;
-
---connection default
-INSERT INTO t VALUES (20,20);
-DELETE FROM t WHERE b = 20;
-
---connect(con_ins,localhost,root,,)
-SET DEBUG_SYNC = 'row_ins_sec_index_entry_dup_locks_created SIGNAL ins_set_locks WAIT_FOR ins_cont';
-send
-INSERT INTO t VALUES(10, 20);
-
---connect(con_del,localhost,root,,)
-SET DEBUG_SYNC = 'now WAIT_FOR ins_set_locks';
-SET DEBUG_SYNC = 'lock_wait_suspend_thread_enter SIGNAL del_locked';
-###############################################################################
-# This DELETE creates waiting ORDINARY X-lock for heap_no 2 as the record is
-# delete-marked, this lock conflicts with ORDINARY S-lock set by the the last
-# INSERT. After the last INSERT creates insert-intention lock on
-# heap_no 2, this lock will conflict with waiting ORDINARY X-lock of this
-# DELETE, what causes DEADLOCK error for this DELETE.
-###############################################################################
-send
-DELETE FROM t WHERE b = 20;
-
---connection default
-SET DEBUG_SYNC = 'now WAIT_FOR del_locked';
-SET DEBUG_SYNC = 'now SIGNAL ins_cont';
-
---connection con_ins
---reap
---disconnect con_ins
-
---connection con_del
-# Without the fix, ER_LOCK_DEADLOCK would be reported here.
---reap
---disconnect con_del
-
---disconnect prevent_purge
-
---connection default
-
-SET DEBUG_SYNC = 'RESET';
-DROP TABLE t;
---source include/wait_until_count_sessions.inc

mysql-test/suite/versioning/r/update.result

@@ -283,6 +283,7 @@ connection default;
 update t1 set b = 'foo';
 connection con1;
 update t1 set a = 'bar';
+ERROR 40001: Deadlock found when trying to get lock; try restarting transaction
 disconnect con1;
 connection default;
 drop table t1;

mysql-test/suite/versioning/t/update.test

@@ -186,9 +186,7 @@ send update t1 set b = 'foo';
 connection con1;
 let $wait_condition= select count(*) from information_schema.innodb_lock_waits;
 source include/wait_condition.inc;
-# There must no be DEADLOCK here as con1 transaction already holds locks, and
-# default's transaction lock is waiting, so the locks of the following "UPDATE"
-# must not conflict with waiting lock.
+error ER_LOCK_DEADLOCK;
 update t1 set a = 'bar';
 disconnect con1;
 connection default;

storage/innobase/include/hash0hash.h

 /*****************************************************************************
 
 Copyright (c) 1997, 2016, Oracle and/or its affiliates. All Rights Reserved.
-Copyright (c) 2018, 2022, MariaDB Corporation.
+Copyright (c) 2018, MariaDB Corporation.
 
 This program is free software; you can redistribute it and/or modify it under
 the terms of the GNU General Public License as published by the Free Software
@@ -31,31 +31,7 @@ Created 5/20/1997 Heikki Tuuri
 #include "sync0rw.h"
 
 struct hash_table_t;
-
-struct hash_cell_t
-{
-  /** singly-linked, nullptr terminated list of hash buckets */
-  void *node;
-
-  /** Insert an element after another.
-  @tparam T  type of the element
-  @param after   the element after which to insert
-  @param insert  the being-inserted element
-  @param next    the next-element pointer in T */
-  template<typename T>
-  void insert_after(T &after, T &insert, T *T::*next)
-  {
-#ifdef UNIV_DEBUG
-    for (const T *c= static_cast<const T*>(node); c; c= c->*next)
-      if (c == &after)
-        goto found;
-    ut_error;
-  found:
-#endif
-    insert.*next= after.*next;
-    after.*next= &insert;
-  }
-};
+struct hash_cell_t;
 
 typedef void*	hash_node_t;
 
@@ -501,6 +477,10 @@ hash_unlock_x_all_but(
 	hash_table_t*	table,		/*!< in: hash table */
 	rw_lock_t*	keep_lock);	/*!< in: lock to keep */
 
+struct hash_cell_t{
+	void*	node;	/*!< hash chain node, NULL if none */
+};
+
 /* The hash table structure */
 struct hash_table_t {
 	enum hash_table_sync_t	type;	/*<! type of hash_table. */

storage/innobase/include/lock0lock.h

 /*****************************************************************************
 
 Copyright (c) 1996, 2016, Oracle and/or its affiliates. All Rights Reserved.
-Copyright (c) 2017, 2020, 2022, MariaDB Corporation.
+Copyright (c) 2017, 2020, MariaDB Corporation.
 
 This program is free software; you can redistribute it and/or modify it under
 the terms of the GNU General Public License as published by the Free Software
@@ -888,29 +888,26 @@ class lock_sys_t
 /*********************************************************************//**
 Creates a new record lock and inserts it to the lock queue. Does NOT check
 for deadlocks or lock compatibility!
-@param[in] c_lock conflicting lock
-@param[in] thr thread owning trx
-@param[in] type_mode lock mode and wait flag, type is ignored and replaced by
-LOCK_REC
-@param[in] block buffer block containing the record
-@param[in] heap_no heap number of the record
-@param[in] index index of record
-@param[in,out] trx transaction
-@param[in] caller_owns_trx_mutex TRUE if caller owns trx mutex
-@param[in] insert_before_waiting if true, inserts new B-tree record lock
-just after the last non-waiting lock of the current transaction which is
-located before the first waiting for the current transaction lock, otherwise
-the lock is inserted at the end of the queue
 @return created lock */
 UNIV_INLINE
-lock_t *lock_rec_create(lock_t *c_lock,
+lock_t*
+lock_rec_create(
+/*============*/
 #ifdef WITH_WSREP
-                        que_thr_t *thr,
+	lock_t*			c_lock,	/*!< conflicting lock */
+	que_thr_t*		thr,	/*!< thread owning trx */
 #endif
-                        ulint type_mode, const buf_block_t *block,
-                        ulint heap_no, dict_index_t *index, trx_t *trx,
-                        bool caller_owns_trx_mutex,
-                        bool insert_before_waiting= false);
+	ulint			type_mode,/*!< in: lock mode and wait
+					flag, type is ignored and
+					replaced by LOCK_REC */
+	const buf_block_t*	block,	/*!< in: buffer block containing
+					the record */
+	ulint			heap_no,/*!< in: heap number of the record */
+	dict_index_t*		index,	/*!< in: index of record */
+	trx_t*			trx,	/*!< in,out: transaction */
+	bool			caller_owns_trx_mutex);
+					/*!< in: true if caller owns
+					trx mutex */
 
 /*************************************************************//**
 Removes a record lock request, waiting or granted, from the queue. */
@@ -923,7 +920,6 @@ lock_rec_discard(
 
 /** Create a new record lock and inserts it to the lock queue,
 without checking for deadlocks or conflicts.
-@param[in]	c_lock		conflicting lock
 @param[in]	type_mode	lock mode and wait flag; type will be replaced
 				with LOCK_REC
 @param[in]	space		tablespace id
@@ -933,15 +929,11 @@ without checking for deadlocks or conflicts.
 @param[in]	index		the index tree
 @param[in,out]	trx		transaction
 @param[in]	holds_trx_mutex	whether the caller holds trx->mutex
-@param[in]	insert_before_waiting if true, inserts new B-tree record lock
-just after the last non-waiting lock of the current transaction which is
-located before the first waiting for the current transaction lock, otherwise
-the lock is inserted at the end of the queue
 @return created lock */
 lock_t*
 lock_rec_create_low(
-	lock_t*		c_lock,
 #ifdef WITH_WSREP
+	lock_t*		c_lock,	/*!< conflicting lock */
 	que_thr_t*	thr,	/*!< thread owning trx */
 #endif
 	ulint		type_mode,
@@ -951,12 +943,9 @@ lock_rec_create_low(
 	ulint		heap_no,
 	dict_index_t*	index,
 	trx_t*		trx,
-	bool		holds_trx_mutex,
-	bool		insert_before_waiting = false);
-
+	bool		holds_trx_mutex);
 /** Enqueue a waiting request for a lock which cannot be granted immediately.
 Check for deadlocks.
-@param[in]	c_lock		conflicting lock
 @param[in]	type_mode	the requested lock mode (LOCK_S or LOCK_X)
 				possibly ORed with LOCK_GAP or
 				LOCK_REC_NOT_GAP, ORed with
@@ -975,7 +964,9 @@ Check for deadlocks.
 				(or it happened to commit) */
 dberr_t
 lock_rec_enqueue_waiting(
-	lock_t*			c_lock,
+#ifdef WITH_WSREP
+	lock_t*			c_lock,	/*!< conflicting lock */
+#endif
 	ulint			type_mode,
 	const buf_block_t*	block,
 	ulint			heap_no,

storage/innobase/include/lock0lock.inl

@@ -101,37 +101,34 @@ lock_hash_get(
 /*********************************************************************//**
 Creates a new record lock and inserts it to the lock queue. Does NOT check
 for deadlocks or lock compatibility!
-@param[in] c_lock conflicting lock
-@param[in] thr thread owning trx
-@param[in] type_mode lock mode and wait flag, type is ignored and replaced by
-LOCK_REC
-@param[in] block buffer block containing the record
-@param[in] heap_no heap number of the record
-@param[in] index index of record
-@param[in,out] trx transaction
-@param[in] caller_owns_trx_mutex TRUE if caller owns trx mutex
-@param[in] insert_before_waiting if true, inserts new B-tree record lock
-just after the last non-waiting lock of the current transaction which is
-located before the first waiting for the current transaction lock, otherwise
-the lock is inserted at the end of the queue
 @return created lock */
 UNIV_INLINE
-lock_t *lock_rec_create(lock_t *c_lock,
+lock_t*
+lock_rec_create(
+/*============*/
 #ifdef WITH_WSREP
-                        que_thr_t *thr,
+	lock_t*			c_lock,	/*!< conflicting lock */
+	que_thr_t*		thr,	/*!< thread owning trx */
 #endif
-                        ulint type_mode, const buf_block_t *block,
-                        ulint heap_no, dict_index_t *index, trx_t *trx,
-                        bool caller_owns_trx_mutex,
-                        bool insert_before_waiting)
+	ulint			type_mode,/*!< in: lock mode and wait
+					flag, type is ignored and
+					replaced by LOCK_REC */
+	const buf_block_t*	block,	/*!< in: buffer block containing
+					the record */
+	ulint			heap_no,/*!< in: heap number of the record */
+	dict_index_t*		index,	/*!< in: index of record */
+	trx_t*			trx,	/*!< in,out: transaction */
+	bool			caller_owns_trx_mutex)
+					/*!< in: TRUE if caller owns
+					trx mutex */
 {
 	btr_assert_not_corrupted(block, index);
-	return lock_rec_create_low(c_lock,
+	return lock_rec_create_low(
 #ifdef WITH_WSREP
-		thr,
+		c_lock, thr,
 #endif
 		type_mode,
 		block->page.id.space(), block->page.id.page_no(),
-		block->frame, heap_no, index, trx,
-		caller_owns_trx_mutex, insert_before_waiting);
+		block->frame, heap_no,
+		index, trx, caller_owns_trx_mutex);
 }

storage/innobase/include/lock0priv.h

 /*****************************************************************************
 
 Copyright (c) 2007, 2016, Oracle and/or its affiliates. All Rights Reserved.
-Copyright (c) 2015, 2018, 2022 MariaDB Corporation.
+Copyright (c) 2015, 2018, MariaDB Corporation.
 
 This program is free software; you can redistribute it and/or modify it under
 the terms of the GNU General Public License as published by the Free Software
@@ -585,9 +585,6 @@ lock_rec_get_next_const(
 
 /*********************************************************************//**
 Gets the first explicit lock request on a record.
-@param[in] hash hash chain the lock on
-@param[in] page_id page id
-@param[in] heap_no heap number of the record
 @return first lock, NULL if none exists */
 UNIV_INLINE
 lock_t*
@@ -663,26 +660,15 @@ lock_table_has(
 
 /** Set the wait status of a lock.
 @param[in,out]	lock	lock that will be waited for
-@param[in,out]	trx	transaction that will wait for the lock
-@param[in]      c_lock   conflicting lock */
-inline void lock_set_lock_and_trx_wait(lock_t* lock, trx_t* trx,
-	const lock_t *c_lock)
+@param[in,out]	trx	transaction that will wait for the lock */
+inline void lock_set_lock_and_trx_wait(lock_t* lock, trx_t* trx)
 {
 	ut_ad(lock);
 	ut_ad(lock->trx == trx);
+	ut_ad(trx->lock.wait_lock == NULL);
 	ut_ad(lock_mutex_own());
 	ut_ad(trx_mutex_own(trx));
 
-	if (trx->lock.wait_trx) {
-		ut_ad(!c_lock || trx->lock.wait_trx == c_lock->trx);
-		ut_ad(trx->lock.wait_lock);
-		ut_ad((*trx->lock.wait_lock).trx == trx);
-	} else {
-		ut_ad(c_lock);
-		trx->lock.wait_trx = c_lock->trx;
-		ut_ad(!trx->lock.wait_lock);
-	}
-
 	trx->lock.wait_lock = lock;
 	lock->type_mode |= LOCK_WAIT;
 }
@@ -695,7 +681,6 @@ inline void lock_reset_lock_and_trx_wait(lock_t* lock)
 	ut_ad(lock_mutex_own());
 	ut_ad(lock->trx->lock.wait_lock == NULL
 	      || lock->trx->lock.wait_lock == lock);
-	lock->trx->lock.wait_trx= NULL;
 	lock->trx->lock.wait_lock = NULL;
 	lock->type_mode &= ~LOCK_WAIT;
 }

storage/innobase/include/lock0priv.inl

@@ -145,19 +145,22 @@ lock_rec_get_first_on_page_addr(
 	return(NULL);
 }
 
-/** Gets the first record lock on a page, where the page is identified by a
+/*********************************************************************//**
+Gets the first record lock on a page, where the page is identified by a
 pointer to it.
-@param[in] lock_hash lock hash table
-@param[in] space page's space id
-@param[in] page_no page number
-@param[in] hash page's hash value in records hash table
 @return	first lock, NULL if none exists */
 UNIV_INLINE
-lock_t *lock_rec_get_first_on_page(hash_table_t *lock_hash, ulint space,
-                                   ulint page_no, ulint hash)
+lock_t*
+lock_rec_get_first_on_page(
+/*=======================*/
+	hash_table_t*		lock_hash,	/*!< in: lock hash table */
+	const buf_block_t*	block)		/*!< in: buffer block */
 {
 	ut_ad(lock_mutex_own());
 
+	ulint	space	= block->page.id.space();
+	ulint	page_no	= block->page.id.page_no();
+	ulint	hash = buf_block_get_lock_hash_val(block);
 
 	for (lock_t* lock = static_cast<lock_t*>(
 			HASH_GET_FIRST(lock_hash, hash));
@@ -174,20 +177,6 @@ lock_t *lock_rec_get_first_on_page(hash_table_t *lock_hash, ulint space,
 	return(NULL);
 }
 
-/** Gets the first record lock on a page, where the page is identified by a
-pointer to it.
-@param[in] lock_hash lock hash table
-@param[in] block buffer block
-@return	first lock, NULL if none exists */
-UNIV_INLINE
-lock_t *lock_rec_get_first_on_page(hash_table_t *lock_hash,
-                                   const buf_block_t *block)
-{
-  return lock_rec_get_first_on_page(lock_hash, block->page.id.space(),
-                                    block->page.id.page_no(),
-                                    buf_block_get_lock_hash_val(block));
-}
-
 /*********************************************************************//**
 Gets the next explicit lock request on a record.
 @return	next lock, NULL if none exists or if heap_no == ULINT_UNDEFINED */
@@ -221,21 +210,21 @@ lock_rec_get_next_const(
 	return(lock_rec_get_next(heap_no, (lock_t*) lock));
 }
 
-/** Gets the first explicit lock request on a record.
-@param[in] hash hash chain the lock on
-@param[in] space record's space id
-@param[in] page_no record's page number
-@param[in] lock_hash_val page's hash value in records hash table
-@param[in] heap_no heap number of the record
+/*********************************************************************//**
+Gets the first explicit lock request on a record.
 @return	first lock, NULL if none exists */
 UNIV_INLINE
-lock_t *lock_rec_get_first(hash_table_t *hash, ulint space, ulint page_no,
-                           uint32_t lock_hash_val, ulint heap_no)
+lock_t*
+lock_rec_get_first(
+/*===============*/
+	hash_table_t*		hash,	/*!< in: hash chain the lock on */
+	const buf_block_t*	block,	/*!< in: block containing the record */
+	ulint			heap_no)/*!< in: heap number of the record */
 {
 	ut_ad(lock_mutex_own());
 
-	for (lock_t* lock = lock_rec_get_first_on_page(hash, space, page_no,
-	      lock_hash_val); lock; lock = lock_rec_get_next_on_page(lock)) {
+	for (lock_t* lock = lock_rec_get_first_on_page(hash, block); lock;
+	     lock = lock_rec_get_next_on_page(lock)) {
 		if (lock_rec_get_nth_bit(lock, heap_no)) {
 			return(lock);
 		}
@@ -244,20 +233,6 @@ lock_t *lock_rec_get_first(hash_table_t *hash, ulint space, ulint page_no,
 	return(NULL);
 }
 
-/** Gets the first explicit lock request on a record.
-@param[in] hash hash chain the lock on
-@param[in] block block containing the record
-@param[in] heap_no heap number of the record
-@return	first lock, NULL if none exists */
-UNIV_INLINE
-lock_t *lock_rec_get_first(hash_table_t *hash, const buf_block_t *block,
-                           ulint heap_no)
-{
-  return lock_rec_get_first(hash, block->page.id.space(),
-                            block->page.id.page_no(),
-                            buf_block_get_lock_hash_val(block), heap_no);
-}
-
 /*********************************************************************//**
 Gets the nth bit of a record lock.
 @return TRUE if bit set also if i == ULINT_UNDEFINED return FALSE*/

storage/innobase/include/trx0trx.h

@@ -425,9 +425,7 @@ struct trx_lock_t {
 	trx_que_t	que_state;	/*!< valid when trx->state
 					== TRX_STATE_ACTIVE: TRX_QUE_RUNNING,
 					TRX_QUE_LOCK_WAIT, ... */
-	/** Transaction being waited for; protected by the same mutexes as
-	wait_lock */
-	trx_t*		wait_trx;
+
 	lock_t*		wait_lock;	/*!< if trx execution state is
 					TRX_QUE_LOCK_WAIT, this points to
 					the lock request, otherwise this is

storage/innobase/lock/lock0lock.cc

 /*****************************************************************************
 
 Copyright (c) 1996, 2017, Oracle and/or its affiliates. All Rights Reserved.
-Copyright (c) 2014, 2021, 2022, MariaDB Corporation.
+Copyright (c) 2014, 2021, MariaDB Corporation.
 
 This program is free software; you can redistribute it and/or modify it under
 the terms of the GNU General Public License as published by the Free Software
@@ -1141,18 +1141,19 @@ static void wsrep_kill_victim(const trx_t * const trx, const lock_t *lock)
 /*********************************************************************//**
 Checks if some other transaction has a conflicting explicit lock request
 in the queue, so that we have to wait.
-@param[in] mode LOCK_S or LOCK_X, possibly ORed to LOCK_GAP or LOC_REC_NOT_GAP,
-LOCK_INSERT_INTENTION
-@param[in] block buffer block containing the record
-@param[in] heap_no heap number of the record
-@param[in] trx our transaction
-@param[out] was_ignored true if conflicting locks waiting for the current
-transaction were ignored
 @return lock or NULL */
-static lock_t *lock_rec_other_has_conflicting(ulint mode,
-                                              const buf_block_t *block,
-                                              ulint heap_no, const trx_t *trx,
-                                              bool *was_ignored= NULL)
+static
+lock_t*
+lock_rec_other_has_conflicting(
+/*===========================*/
+	ulint			mode,	/*!< in: LOCK_S or LOCK_X,
+					possibly ORed to LOCK_GAP or
+					LOC_REC_NOT_GAP,
+					LOCK_INSERT_INTENTION */
+	const buf_block_t*	block,	/*!< in: buffer block containing
+					the record */
+	ulint			heap_no,/*!< in: heap number of the record */
+	const trx_t*		trx)	/*!< in: our transaction */
 {
 	lock_t*		lock;
 
@@ -1164,16 +1165,6 @@ static lock_t *lock_rec_other_has_conflicting(ulint mode,
 	     lock != NULL;
 	     lock = lock_rec_get_next(heap_no, lock)) {
 
-		/* There can't be lock loops for one record, because
-		all waiting locks of the record  will always wait for the same
-		lock of the record in a cell array, and check for
-		conflicting lock will always start with the first lock for the
-		heap_no, and go ahead with the same order(the order of the
-		locks in the cell array) */
-		if (lock_get_wait(lock) && lock->trx->lock.wait_trx == trx) {
-			if (was_ignored) *was_ignored= true;
-			continue;
-		}
 		if (lock_rec_has_to_wait(true, trx, mode, lock, is_supremum)) {
 #ifdef WITH_WSREP
 			if (trx->is_wsrep()) {
@@ -1330,7 +1321,6 @@ static void check_trx_state(const trx_t *trx)
 
 /** Create a new record lock and inserts it to the lock queue,
 without checking for deadlocks or conflicts.
-@param[in]	c_lock		conflicting lock
 @param[in]	type_mode	lock mode and wait flag; type will be replaced
 				with LOCK_REC
 @param[in]	space		tablespace id
@@ -1340,15 +1330,11 @@ without checking for deadlocks or conflicts.
 @param[in]	index		the index tree
 @param[in,out]	trx		transaction
 @param[in]	holds_trx_mutex	whether the caller holds trx->mutex
-@param[in]	insert_before_waiting if true, inserts new B-tree record lock
-just after the last non-waiting lock of the current transaction which is
-located before the first waiting for the current transaction lock, otherwise
-the lock is inserted at the end of the queue
 @return created lock */
 lock_t*
 lock_rec_create_low(
-	lock_t*		c_lock,
 #ifdef WITH_WSREP
+	lock_t*		c_lock,	/*!< conflicting lock */
 	que_thr_t*	thr,	/*!< thread owning trx */
 #endif
 	ulint		type_mode,
@@ -1358,8 +1344,7 @@ lock_rec_create_low(
 	ulint		heap_no,
 	dict_index_t*	index,
 	trx_t*		trx,
-	bool		holds_trx_mutex,
-	bool		insert_before_waiting)
+	bool		holds_trx_mutex)
 {
 	lock_t*		lock;
 	ulint		n_bits;
@@ -1469,7 +1454,7 @@ lock_rec_create_low(
 			}
 
 			trx->lock.que_state = TRX_QUE_LOCK_WAIT;
-			lock_set_lock_and_trx_wait(lock, trx, c_lock);
+			lock_set_lock_and_trx_wait(lock, trx);
 			UT_LIST_ADD_LAST(trx->lock.trx_locks, lock);
 
 			trx->lock.wait_thr = thr;
@@ -1497,36 +1482,6 @@ lock_rec_create_low(
 		trx_mutex_exit(c_lock->trx);
 	} else
 #endif /* WITH_WSREP */
-	if (insert_before_waiting
-	    && !(type_mode & (LOCK_PREDICATE | LOCK_PRDT_PAGE))) {
-		/* Try to insert the lock just after the last non-waiting
-		lock of the current transaction which immediately
-		precedes the first waiting lock request. */
-		uint32_t lock_hash_val = lock_rec_hash(space, page_no);
-		hash_cell_t& cell = lock_sys.rec_hash->array[lock_hash_val];
-
-		lock_t* last_non_waiting = NULL;
-
-		for (lock_t* l = lock_rec_get_first(lock_sys.rec_hash, space,
-		      page_no, lock_hash_val, heap_no); l;
-		    l = lock_rec_get_next(heap_no, l)) {
-			if (lock_get_wait(lock)
-			    && l->trx->lock.wait_trx == trx) {
-				break;
-			}
-			if (l->trx == trx) {
-				last_non_waiting = l;
-			}
-		}
-
-		if (!last_non_waiting) {
-			goto append_last;
-		}
-
-		cell.insert_after(*last_non_waiting, *lock, &lock_t::hash);
-	}
-	else {
-append_last:
 	if (!(type_mode & (LOCK_WAIT | LOCK_PREDICATE | LOCK_PRDT_PAGE))
 	    && innodb_lock_schedule_algorithm
 	    == INNODB_LOCK_SCHEDULE_ALGORITHM_VATS
@@ -1537,14 +1492,13 @@ lock_rec_create_low(
 		HASH_INSERT(lock_t, hash, lock_hash_get(type_mode),
 			    lock_rec_fold(space, page_no), lock);
 	}
-	}
 
 	if (!holds_trx_mutex) {
 		trx_mutex_enter(trx);
 	}
 	ut_ad(trx_mutex_own(trx));
 	if (type_mode & LOCK_WAIT) {
-		lock_set_lock_and_trx_wait(lock, trx, c_lock);
+		lock_set_lock_and_trx_wait(lock, trx);
 	}
 	UT_LIST_ADD_LAST(trx->lock.trx_locks, lock);
 	if (!holds_trx_mutex) {
@@ -1704,7 +1658,6 @@ lock_rec_insert_to_head(
 
 /** Enqueue a waiting request for a lock which cannot be granted immediately.
 Check for deadlocks.
-@param[in]	c_lock		conflicting lock
 @param[in]	type_mode	the requested lock mode (LOCK_S or LOCK_X)
 				possibly ORed with LOCK_GAP or
 				LOCK_REC_NOT_GAP, ORed with
@@ -1723,7 +1676,9 @@ Check for deadlocks.
 				(or it happened to commit) */
 dberr_t
 lock_rec_enqueue_waiting(
-	lock_t*			c_lock,
+#ifdef WITH_WSREP
+	lock_t*			c_lock,	/*!< conflicting lock */
+#endif
 	ulint			type_mode,
 	const buf_block_t*	block,
 	ulint			heap_no,
@@ -1761,9 +1716,9 @@ lock_rec_enqueue_waiting(
 
 	/* Enqueue the lock request that will wait to be granted, note that
 	we already own the trx mutex. */
-	lock_t* lock = lock_rec_create(c_lock,
+	lock_t* lock = lock_rec_create(
 #ifdef WITH_WSREP
-		thr,
+		c_lock, thr,
 #endif
 		type_mode | LOCK_WAIT, block, heap_no, index, trx, TRUE);
 
@@ -1827,20 +1782,22 @@ on the record, and the request to be added is not a waiting request, we
 can reuse a suitable record lock object already existing on the same page,
 just setting the appropriate bit in its bitmap. This is a low-level function
 which does NOT check for deadlocks or lock compatibility!
-@param[in] type_mode lock mode, wait, gap etc. flags; type is ignored and
-replaced by LOCK_REC
-@param[in] block buffer block containing the record
-@param[in] heap_no heap number of the record
-@param[in] index index of record
-@param[in,out] trx transaction
-@param[in] caller_owns_trx_mutex, TRUE if caller owns the transaction mutex
-@param[in] insert_before_waiting true=insert B-tree record lock right before
-a waiting lock request; false=insert the lock at the end of the queue
 @return lock where the bit was set */
-static void lock_rec_add_to_queue(ulint type_mode, const buf_block_t *block,
-                                  ulint heap_no, dict_index_t *index,
-                                  trx_t *trx, bool caller_owns_trx_mutex,
-                                  bool insert_before_waiting= false)
+static
+void
+lock_rec_add_to_queue(
+/*==================*/
+	ulint			type_mode,/*!< in: lock mode, wait, gap
+					etc. flags; type is ignored
+					and replaced by LOCK_REC */
+	const buf_block_t*	block,	/*!< in: buffer block containing
+					the record */
+	ulint			heap_no,/*!< in: heap number of the record */
+	dict_index_t*		index,	/*!< in: index of record */
+	trx_t*			trx,	/*!< in/out: transaction */
+	bool			caller_owns_trx_mutex)
+					/*!< in: TRUE if caller owns the
+					transaction mutex */
 {
 #ifdef UNIV_DEBUG
 	ut_ad(lock_mutex_own());
@@ -1929,16 +1886,11 @@ static void lock_rec_add_to_queue(ulint type_mode, const buf_block_t *block,
 		}
 	}
 
-	/* Note: We will not pass any conflicting lock to lock_rec_create(),
-	because we should be moving an existing waiting lock request. */
-	ut_ad(!(type_mode & LOCK_WAIT) || trx->lock.wait_trx);
-
-	lock_rec_create(NULL,
+	lock_rec_create(
 #ifdef WITH_WSREP
-		NULL,
+		NULL, NULL,
 #endif
-		type_mode, block, heap_no, index, trx, caller_owns_trx_mutex,
-		insert_before_waiting);
+		type_mode, block, heap_no, index, trx, caller_owns_trx_mutex);
 }
 
 /*********************************************************************//**
@@ -1994,23 +1946,28 @@ lock_rec_lock(
       /* Do nothing if the trx already has a strong enough lock on rec */
       if (!lock_rec_has_expl(mode, block, heap_no, trx))
       {
-        bool was_ignored = false;
-        if (lock_t *c_lock= lock_rec_other_has_conflicting(
-                mode, block, heap_no, trx, &was_ignored))
+        if (
+#ifdef WITH_WSREP
+	    lock_t *c_lock=
+#endif
+	    lock_rec_other_has_conflicting(mode, block, heap_no, trx))
         {
           /*
             If another transaction has a non-gap conflicting
             request in the queue, as this transaction does not
             have a lock strong enough already granted on the
 	    record, we have to wait. */
-	    err = lock_rec_enqueue_waiting(c_lock, mode, block, heap_no, index,
-		thr, NULL);
+	    err = lock_rec_enqueue_waiting(
+#ifdef WITH_WSREP
+			c_lock,
+#endif /* WITH_WSREP */
+			mode, block, heap_no, index, thr, NULL);
         }
         else if (!impl)
         {
           /* Set the requested lock on the record. */
           lock_rec_add_to_queue(LOCK_REC | mode, block, heap_no, index, trx,
-                                true, was_ignored);
+                                true);
           err= DB_SUCCESS_LOCKED_REC;
         }
       }
@@ -2036,9 +1993,9 @@ lock_rec_lock(
       Note that we don't own the trx mutex.
     */
     if (!impl)
-      lock_rec_create(NULL,
+      lock_rec_create(
 #ifdef WITH_WSREP
-         NULL,
+         NULL, NULL,
 #endif
         mode, block, heap_no, index, trx, false);
 
@@ -2277,17 +2234,8 @@ static void lock_rec_dequeue_from_page(lock_t* in_lock)
 			if (!lock_get_wait(lock)) {
 				continue;
 			}
-
-			ut_ad(lock->trx->lock.wait_trx);
-			ut_ad(lock->trx->lock.wait_lock);
-
-			if (const lock_t* c = lock_rec_has_to_wait_in_queue(
-				    lock)) {
-				trx_mutex_enter(lock->trx);
-				lock->trx->lock.wait_trx = c->trx;
-				trx_mutex_exit(lock->trx);
-			}
-			else {
+			const lock_t* c = lock_rec_has_to_wait_in_queue(lock);
+			if (!c) {
 				/* Grant the lock */
 				ut_ad(lock->trx != in_lock->trx);
 				lock_grant(lock);
@@ -2561,8 +2509,7 @@ lock_rec_move_low(
 		lock_rec_reset_nth_bit(lock, donator_heap_no);
 
 		if (type_mode & LOCK_WAIT) {
-			ut_ad(lock->trx->lock.wait_lock == lock);
-			lock->type_mode &= ~LOCK_WAIT;
+			lock_reset_lock_and_trx_wait(lock);
 		}
 
 		/* Note that we FIRST reset the bit, and then set the lock:
@@ -2679,8 +2626,8 @@ lock_move_reorganize_page(
 		lock_rec_bitmap_reset(lock);
 
 		if (lock_get_wait(lock)) {
-			ut_ad(lock->trx->lock.wait_lock == lock);
-			lock->type_mode&= ~LOCK_WAIT;
+
+			lock_reset_lock_and_trx_wait(lock);
 		}
 
 		lock = lock_rec_get_next_on_page(lock);
@@ -2855,9 +2802,7 @@ lock_move_rec_list_end(
 				ut_ad(!page_rec_is_metadata(orec));
 
 				if (type_mode & LOCK_WAIT) {
-					ut_ad(lock->trx->lock.wait_lock ==
-					    lock);
-					lock->type_mode&= ~LOCK_WAIT;
+					lock_reset_lock_and_trx_wait(lock);
 				}
 
 				lock_rec_add_to_queue(
@@ -2954,9 +2899,7 @@ lock_move_rec_list_start(
 				ut_ad(!page_rec_is_metadata(prev));
 
 				if (type_mode & LOCK_WAIT) {
-					ut_ad(lock->trx->lock.wait_lock
-					    == lock);
-					lock->type_mode&= ~LOCK_WAIT;
+					lock_reset_lock_and_trx_wait(lock);
 				}
 
 				lock_rec_add_to_queue(
@@ -3051,9 +2994,7 @@ lock_rtr_move_rec_list(
 			if (rec1_heap_no < lock->un_member.rec_lock.n_bits
 			    && lock_rec_reset_nth_bit(lock, rec1_heap_no)) {
 				if (type_mode & LOCK_WAIT) {
-					ut_ad(lock->trx->lock.wait_lock
-					    == lock);
-					lock->type_mode&= ~LOCK_WAIT;
+					lock_reset_lock_and_trx_wait(lock);
 				}
 
 				lock_rec_add_to_queue(
@@ -3507,8 +3448,10 @@ lock_table_create(
 				in dictionary cache */
 	ulint		type_mode,/*!< in: lock mode possibly ORed with
 				LOCK_WAIT */
-	trx_t*		trx,	/*!< in: trx */
-	lock_t*	c_lock = NULL	/*!< in: conflicting lock */
+	trx_t*		trx	/*!< in: trx */
+#ifdef WITH_WSREP
+	, lock_t*	c_lock = NULL	/*!< in: conflicting lock */
+#endif
 	)
 {
 	lock_t*		lock;
@@ -3591,7 +3534,8 @@ lock_table_create(
 	ut_list_append(table->locks, lock, TableLockGetNode());
 
 	if (type_mode & LOCK_WAIT) {
-		lock_set_lock_and_trx_wait(lock, trx, c_lock);
+
+		lock_set_lock_and_trx_wait(lock, trx);
 	}
 
 	lock->trx->lock.table_locks.push_back(lock);
@@ -3746,8 +3690,10 @@ lock_table_enqueue_waiting(
 	ulint		mode,	/*!< in: lock mode this transaction is
 				requesting */
 	dict_table_t*	table,	/*!< in/out: table */
-	que_thr_t*	thr,	/*!< in: query thread */
-	lock_t*	c_lock	/*!< in: conflicting lock or NULL */
+	que_thr_t*	thr	/*!< in: query thread */
+#ifdef WITH_WSREP
+	, lock_t*	c_lock	/*!< in: conflicting lock or NULL */
+#endif
 )
 {
 	trx_t*		trx;
@@ -3778,7 +3724,11 @@ lock_table_enqueue_waiting(
 #endif /* WITH_WSREP */
 
 	/* Enqueue the lock request that will wait to be granted */
-	lock = lock_table_create(table, ulint(mode) | LOCK_WAIT, trx, c_lock);
+	lock = lock_table_create(table, ulint(mode) | LOCK_WAIT, trx
+#ifdef WITH_WSREP
+				 , c_lock
+#endif
+				 );
 
 	const trx_t*	victim_trx =
 		DeadlockChecker::check_and_resolve(lock, trx);
@@ -3934,7 +3884,11 @@ lock_table(
 
 	if (wait_for != NULL) {
 		err = lock_table_enqueue_waiting(ulint(mode) | flags, table,
-						 thr, wait_for);
+						 thr
+#ifdef WITH_WSREP
+						 , wait_for
+#endif
+						 );
 	} else {
 		lock_table_create(table, ulint(mode) | flags, trx);
 
@@ -3982,7 +3936,7 @@ lock_table_ix_resurrect(
 Checks if a waiting table lock request still has to wait in a queue.
 @return TRUE if still has to wait */
 static
-const lock_t*
+bool
 lock_table_has_to_wait_in_queue(
 /*============================*/
 	const lock_t*	wait_lock)	/*!< in: waiting table lock */
@@ -4001,11 +3955,11 @@ lock_table_has_to_wait_in_queue(
 
 		if (lock_has_to_wait(wait_lock, lock)) {
 
-			return(lock);
+			return(true);
 		}
 	}
 
-	return(NULL);
+	return(false);
 }
 
 /*************************************************************//**
@@ -4034,17 +3988,9 @@ lock_table_dequeue(
 	     lock != NULL;
 	     lock = UT_LIST_GET_NEXT(un_member.tab_lock.locks, lock)) {
 
-		if (!lock_get_wait(lock))
-			continue;
-
-		ut_ad(lock->trx->lock.wait_trx);
-		ut_ad(lock->trx->lock.wait_lock);
+		if (lock_get_wait(lock)
+		    && !lock_table_has_to_wait_in_queue(lock)) {
 
-		if (const lock_t *c = lock_table_has_to_wait_in_queue(lock)) {
-			trx_mutex_enter(lock->trx);
-			lock->trx->lock.wait_trx = c->trx;
-			trx_mutex_exit(lock->trx);
-		} else {
 			/* Grant the lock */
 			ut_ad(in_lock->trx != lock->trx);
 			lock_grant(lock);
@@ -4240,16 +4186,8 @@ lock_rec_unlock(
 			if (!lock_get_wait(lock)) {
 				continue;
 			}
-			ut_ad(lock->trx->lock.wait_trx);
-			ut_ad(lock->trx->lock.wait_lock);
-			if (const lock_t* c = lock_rec_has_to_wait_in_queue(
-			      lock)) {
-				if (lock->trx != trx)
-					trx_mutex_enter(lock->trx);
-				lock->trx->lock.wait_trx = c->trx;
-				if (lock->trx != trx)
-					trx_mutex_exit(lock->trx);
-			} else {
+			const lock_t* c = lock_rec_has_to_wait_in_queue(lock);
+			if (!c) {
 				/* Grant the lock */
 				ut_ad(trx != lock->trx);
 				lock_grant(lock);
@@ -4978,7 +4916,7 @@ lock_rec_queue_validate(
 				wsrep_report_bf_lock_wait(impl_trx->mysql_thd, impl_trx->id);
 				wsrep_report_bf_lock_wait(other_lock->trx->mysql_thd, other_lock->trx->id);
 
-				if (!lock_rec_has_expl(LOCK_S | LOCK_REC_NOT_GAP,
+				if (!lock_rec_has_expl(LOCK_X | LOCK_REC_NOT_GAP,
 						       block, heap_no,
 						       impl_trx)) {
 					ib::info() << "WSREP impl BF lock conflict";
@@ -4987,20 +4925,7 @@ lock_rec_queue_validate(
 #endif /* WITH_WSREP */
 			{
 				ut_ad(lock_get_wait(other_lock));
-				/* After MDEV-27025 fix the following case is
-				possible:
-				1. trx 1 acquires S-lock;
-				2. trx 2 creates X-lock waiting for trx 1;
-				3. trx 1 creates implicit lock, as
-				lock_rec_other_has_conflicting() returns no
-				conflicting trx 2 X-lock, the explicit lock
-				will not be created;
-				4. trx 3 creates waiting X-lock,
-				it will wait for S-lock of trx 1.
-				That is why we relaxing the condition here and
-				check only for S-lock.
-				*/
-				ut_ad(lock_rec_has_expl(LOCK_S | LOCK_REC_NOT_GAP,
+				ut_ad(lock_rec_has_expl(LOCK_X | LOCK_REC_NOT_GAP,
 						        block, heap_no, impl_trx));
 			}
 		}
@@ -5406,13 +5331,19 @@ lock_rec_insert_check_and_lock(
 
 	const ulint	type_mode = LOCK_X | LOCK_GAP | LOCK_INSERT_INTENTION;
 
-	if (lock_t* c_lock =
+	if (
+#ifdef WITH_WSREP
+	    lock_t* c_lock =
+#endif /* WITH_WSREP */
 	    lock_rec_other_has_conflicting(type_mode, block, heap_no, trx)) {
 		/* Note that we may get DB_SUCCESS also here! */
 		trx_mutex_enter(trx);
 
-		err = lock_rec_enqueue_waiting(c_lock, type_mode, block,
-		    heap_no, index, thr, NULL);
+		err = lock_rec_enqueue_waiting(
+#ifdef WITH_WSREP
+			c_lock,
+#endif /* WITH_WSREP */
+			type_mode, block, heap_no, index, thr, NULL);
 
 		trx_mutex_exit(trx);
 	} else {
@@ -5489,7 +5420,7 @@ lock_rec_convert_impl_to_expl_for_trx(
 	    && !lock_rec_has_expl(LOCK_X | LOCK_REC_NOT_GAP,
 				  block, heap_no, trx)) {
 		lock_rec_add_to_queue(LOCK_REC | LOCK_X | LOCK_REC_NOT_GAP,
-				      block, heap_no, index, trx, true, true);
+				      block, heap_no, index, trx, true);
 	}
 
 	lock_mutex_exit();

storage/innobase/lock/lock0prdt.cc

 /*****************************************************************************
 
 Copyright (c) 2014, 2016, Oracle and/or its affiliates. All Rights Reserved.
-Copyright (c) 2018, 2022 MariaDB Corporation.
+Copyright (c) 2018, MariaDB Corporation.
 
 This program is free software; you can redistribute it and/or modify it under
 the terms of the GNU General Public License as published by the Free Software
@@ -487,13 +487,9 @@ lock_prdt_add_to_queue(
 		}
 	}
 
-	/* Note: We will not pass any conflicting lock to lock_rec_create(),
-	because we should be moving an existing waiting lock request. */
-	ut_ad(!(type_mode & LOCK_WAIT) || trx->lock.wait_trx);
-
-	lock = lock_rec_create(NULL,
+	lock = lock_rec_create(
 #ifdef WITH_WSREP
-		NULL, /* FIXME: replicate SPATIAL INDEX locks */
+		NULL, NULL, /* FIXME: replicate SPATIAL INDEX locks */
 #endif
 		type_mode, block, PRDT_HEAPNO, index, trx,
 		caller_owns_trx_mutex);
@@ -583,7 +579,9 @@ lock_prdt_insert_check_and_lock(
 		trx_mutex_enter(trx);
 
 		err = lock_rec_enqueue_waiting(
+#ifdef WITH_WSREP
 			NULL, /* FIXME: replicate SPATIAL INDEX locks */
+#endif
 			LOCK_X | LOCK_PREDICATE | LOCK_INSERT_INTENTION,
 			block, PRDT_HEAPNO, index, thr, prdt);
 
@@ -831,9 +829,9 @@ lock_prdt_lock(
 	lock_t*		lock = lock_rec_get_first_on_page(hash, block);
 
 	if (lock == NULL) {
-		lock = lock_rec_create(NULL,
+		lock = lock_rec_create(
 #ifdef WITH_WSREP
-			NULL, /* FIXME: replicate SPATIAL INDEX locks */
+			NULL, NULL, /* FIXME: replicate SPATIAL INDEX locks */
 #endif
 			ulint(mode) | type_mode, block, PRDT_HEAPNO,
 			index, trx, FALSE);
@@ -863,8 +861,10 @@ lock_prdt_lock(
 				if (wait_for != NULL) {
 
 					err = lock_rec_enqueue_waiting(
+#ifdef WITH_WSREP
 						NULL, /* FIXME: replicate
 						      SPATIAL INDEX locks */
+#endif
 						ulint(mode) | type_mode,
 						block, PRDT_HEAPNO,
 						index, thr, prdt);
@@ -948,9 +948,9 @@ lock_place_prdt_page_lock(
 	}
 
 	if (lock == NULL) {
-		lock = lock_rec_create_low(NULL,
+		lock = lock_rec_create_low(
 #ifdef WITH_WSREP
-			NULL, /* FIXME: replicate SPATIAL INDEX locks */
+			NULL, NULL, /* FIXME: replicate SPATIAL INDEX locks */
 #endif
 			mode, space, page_no, NULL, PRDT_HEAPNO,
 			index, trx, FALSE);