Commit be811386 authored by Vlad Lesin's avatar Vlad Lesin

MDEV-27025 insert-intention lock conflicts with waiting ORDINARY lock

The code was backported from 10.6 bd03c0e5
commit. See that commit message for details.

Apart from the above commit trx_lock_t::wait_trx was also backported from
MDEV-24738. trx_lock_t::wait_trx is protected with lock_sys.wait_mutex
in 10.6, but that mutex was implemented only in MDEV-24789. As there is no
need to backport MDEV-24789 for MDEV-27025,
trx_lock_t::wait_trx is protected with the same mutexes as
trx_lock_t::wait_lock.

This fix should not break innodb-lock-schedule-algorithm=VATS. This
algorithm uses an Eldest-Transaction-First (ETF) heuristic, which prefers
older transactions over new ones. In this fix we just insert granted lock
just before the last granted lock of the same transaction, what does not
change transactions execution order.

The changes in lock_rec_create_low() should not break Galera Cluster,
there is a big "if" branch for WSREP. This branch is necessary to provide
the correct transactions execution order, and should not be changed for
the current bug fix.
parent e44439ab
#
# 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;
--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
...@@ -283,7 +283,6 @@ connection default; ...@@ -283,7 +283,6 @@ connection default;
update t1 set b = 'foo'; update t1 set b = 'foo';
connection con1; connection con1;
update t1 set a = 'bar'; update t1 set a = 'bar';
ERROR 40001: Deadlock found when trying to get lock; try restarting transaction
disconnect con1; disconnect con1;
connection default; connection default;
drop table t1; drop table t1;
......
...@@ -186,7 +186,9 @@ send update t1 set b = 'foo'; ...@@ -186,7 +186,9 @@ send update t1 set b = 'foo';
connection con1; connection con1;
let $wait_condition= select count(*) from information_schema.innodb_lock_waits; let $wait_condition= select count(*) from information_schema.innodb_lock_waits;
source include/wait_condition.inc; source include/wait_condition.inc;
error ER_LOCK_DEADLOCK; # 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.
update t1 set a = 'bar'; update t1 set a = 'bar';
disconnect con1; disconnect con1;
connection default; connection default;
......
/***************************************************************************** /*****************************************************************************
Copyright (c) 1997, 2016, Oracle and/or its affiliates. All Rights Reserved. Copyright (c) 1997, 2016, Oracle and/or its affiliates. All Rights Reserved.
Copyright (c) 2018, 2020, MariaDB Corporation. Copyright (c) 2018, 2022, MariaDB Corporation.
This program is free software; you can redistribute it and/or modify it under 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 the terms of the GNU General Public License as published by the Free Software
...@@ -28,8 +28,29 @@ Created 5/20/1997 Heikki Tuuri ...@@ -28,8 +28,29 @@ Created 5/20/1997 Heikki Tuuri
#include "ut0rnd.h" #include "ut0rnd.h"
struct hash_table_t; struct hash_table_t;
struct hash_cell_t{ struct hash_cell_t
void* node; /*!< hash chain node, NULL if none */ {
/** 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;
}
}; };
typedef void* hash_node_t; typedef void* hash_node_t;
......
/***************************************************************************** /*****************************************************************************
Copyright (c) 1996, 2016, Oracle and/or its affiliates. All Rights Reserved. Copyright (c) 1996, 2016, Oracle and/or its affiliates. All Rights Reserved.
Copyright (c) 2017, 2021, MariaDB Corporation. Copyright (c) 2017, 2022, MariaDB Corporation.
This program is free software; you can redistribute it and/or modify it under 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 the terms of the GNU General Public License as published by the Free Software
...@@ -832,26 +832,29 @@ class lock_sys_t ...@@ -832,26 +832,29 @@ class lock_sys_t
/*********************************************************************//** /*********************************************************************//**
Creates a new record lock and inserts it to the lock queue. Does NOT check Creates a new record lock and inserts it to the lock queue. Does NOT check
for deadlocks or lock compatibility! 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 */ @return created lock */
UNIV_INLINE UNIV_INLINE
lock_t* lock_t *lock_rec_create(lock_t *c_lock,
lock_rec_create(
/*============*/
#ifdef WITH_WSREP #ifdef WITH_WSREP
lock_t* c_lock, /*!< conflicting lock */ que_thr_t *thr,
que_thr_t* thr, /*!< thread owning trx */
#endif #endif
unsigned type_mode,/*!< in: lock mode and wait unsigned type_mode, const buf_block_t *block,
flag, type is ignored and ulint heap_no, dict_index_t *index, trx_t *trx,
replaced by LOCK_REC */ bool caller_owns_trx_mutex,
const buf_block_t* block, /*!< in: buffer block containing bool insert_before_waiting= false);
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. */ Removes a record lock request, waiting or granted, from the queue. */
...@@ -864,6 +867,7 @@ lock_rec_discard( ...@@ -864,6 +867,7 @@ lock_rec_discard(
/** Create a new record lock and inserts it to the lock queue, /** Create a new record lock and inserts it to the lock queue,
without checking for deadlocks or conflicts. 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 @param[in] type_mode lock mode and wait flag; type will be replaced
with LOCK_REC with LOCK_REC
@param[in] page_id index page number @param[in] page_id index page number
...@@ -872,11 +876,15 @@ without checking for deadlocks or conflicts. ...@@ -872,11 +876,15 @@ without checking for deadlocks or conflicts.
@param[in] index the index tree @param[in] index the index tree
@param[in,out] trx transaction @param[in,out] trx transaction
@param[in] holds_trx_mutex whether the caller holds trx->mutex @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 */ @return created lock */
lock_t* lock_t*
lock_rec_create_low( lock_rec_create_low(
lock_t* c_lock,
#ifdef WITH_WSREP #ifdef WITH_WSREP
lock_t* c_lock, /*!< conflicting lock */
que_thr_t* thr, /*!< thread owning trx */ que_thr_t* thr, /*!< thread owning trx */
#endif #endif
unsigned type_mode, unsigned type_mode,
...@@ -885,7 +893,9 @@ lock_rec_create_low( ...@@ -885,7 +893,9 @@ lock_rec_create_low(
ulint heap_no, ulint heap_no,
dict_index_t* index, dict_index_t* index,
trx_t* trx, trx_t* trx,
bool holds_trx_mutex); bool holds_trx_mutex,
bool insert_before_waiting = false);
/** Enqueue a waiting request for a lock which cannot be granted immediately. /** Enqueue a waiting request for a lock which cannot be granted immediately.
Check for deadlocks. Check for deadlocks.
@param[in] type_mode the requested lock mode (LOCK_S or LOCK_X) @param[in] type_mode the requested lock mode (LOCK_S or LOCK_X)
...@@ -906,9 +916,7 @@ Check for deadlocks. ...@@ -906,9 +916,7 @@ Check for deadlocks.
(or it happened to commit) */ (or it happened to commit) */
dberr_t dberr_t
lock_rec_enqueue_waiting( lock_rec_enqueue_waiting(
#ifdef WITH_WSREP
lock_t* c_lock, /*!< conflicting lock */ lock_t* c_lock, /*!< conflicting lock */
#endif
unsigned type_mode, unsigned type_mode,
const buf_block_t* block, const buf_block_t* block,
ulint heap_no, ulint heap_no,
......
...@@ -72,32 +72,35 @@ lock_hash_get( ...@@ -72,32 +72,35 @@ lock_hash_get(
/*********************************************************************//** /*********************************************************************//**
Creates a new record lock and inserts it to the lock queue. Does NOT check Creates a new record lock and inserts it to the lock queue. Does NOT check
for deadlocks or lock compatibility! 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 */ @return created lock */
UNIV_INLINE UNIV_INLINE
lock_t* lock_t *lock_rec_create(lock_t *c_lock,
lock_rec_create(
/*============*/
#ifdef WITH_WSREP #ifdef WITH_WSREP
lock_t* c_lock, /*!< conflicting lock */ que_thr_t *thr,
que_thr_t* thr, /*!< thread owning trx */
#endif #endif
unsigned type_mode,/*!< in: lock mode and wait unsigned type_mode, const buf_block_t *block,
flag, type is ignored and ulint heap_no, dict_index_t *index, trx_t *trx,
replaced by LOCK_REC */ bool caller_owns_trx_mutex,
const buf_block_t* block, /*!< in: buffer block containing bool insert_before_waiting)
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); btr_assert_not_corrupted(block, index);
return lock_rec_create_low( return lock_rec_create_low(c_lock,
#ifdef WITH_WSREP #ifdef WITH_WSREP
c_lock, thr, thr,
#endif #endif
type_mode, block->page.id(), block->frame, heap_no, type_mode, block->page.id(), block->frame, heap_no,
index, trx, caller_owns_trx_mutex); index, trx, caller_owns_trx_mutex, insert_before_waiting);
} }
...@@ -562,14 +562,13 @@ lock_rec_get_next_const( ...@@ -562,14 +562,13 @@ lock_rec_get_next_const(
/*********************************************************************//** /*********************************************************************//**
Gets the first explicit lock request on a record. 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 */ @return first lock, NULL if none exists */
UNIV_INLINE UNIV_INLINE
lock_t* lock_t *lock_rec_get_first(hash_table_t *hash, page_id_t page_id,
lock_rec_get_first( ulint heap_no);
/*===============*/
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 */
/*********************************************************************//** /*********************************************************************//**
Gets the mode of a lock. Gets the mode of a lock.
...@@ -623,15 +622,26 @@ lock_table_has( ...@@ -623,15 +622,26 @@ lock_table_has(
/** Set the wait status of a lock. /** Set the wait status of a lock.
@param[in,out] lock lock that will be waited for @param[in,out] lock lock that will be waited for
@param[in,out] trx transaction that will wait for the 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) @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)
{ {
ut_ad(lock); ut_ad(lock);
ut_ad(lock->trx == trx); ut_ad(lock->trx == trx);
ut_ad(trx->lock.wait_lock == NULL);
ut_ad(lock_mutex_own()); ut_ad(lock_mutex_own());
ut_ad(trx_mutex_own(trx)); 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; trx->lock.wait_lock = lock;
lock->type_mode |= LOCK_WAIT; lock->type_mode |= LOCK_WAIT;
} }
...@@ -644,6 +654,7 @@ inline void lock_reset_lock_and_trx_wait(lock_t* lock) ...@@ -644,6 +654,7 @@ inline void lock_reset_lock_and_trx_wait(lock_t* lock)
ut_ad(lock_mutex_own()); ut_ad(lock_mutex_own());
ut_ad(lock->trx->lock.wait_lock == NULL ut_ad(lock->trx->lock.wait_lock == NULL
|| lock->trx->lock.wait_lock == lock); || lock->trx->lock.wait_lock == lock);
lock->trx->lock.wait_trx= nullptr;
lock->trx->lock.wait_lock = NULL; lock->trx->lock.wait_lock = NULL;
lock->type_mode &= ~LOCK_WAIT; lock->type_mode &= ~LOCK_WAIT;
} }
......
...@@ -156,16 +156,15 @@ lock_rec_get_next_const( ...@@ -156,16 +156,15 @@ lock_rec_get_next_const(
/*********************************************************************//** /*********************************************************************//**
Gets the first explicit lock request on a record. Gets the first explicit lock request on a record.
@return first lock, NULL if none exists */ @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 UNIV_INLINE
lock_t* lock_t *lock_rec_get_first(hash_table_t *hash, page_id_t page_id,
lock_rec_get_first( ulint heap_no)
/*===============*/
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 */
{ {
for (lock_t *lock= lock_sys.get_first(*hash, block->page.id()); for (lock_t *lock= lock_sys.get_first(*hash, page_id);
lock; lock= lock_rec_get_next_on_page(lock)) lock; lock= lock_rec_get_next_on_page(lock))
if (lock_rec_get_nth_bit(lock, heap_no)) if (lock_rec_get_nth_bit(lock, heap_no))
return lock; return lock;
......
...@@ -426,7 +426,9 @@ struct trx_lock_t { ...@@ -426,7 +426,9 @@ struct trx_lock_t {
trx_que_t que_state; /*!< valid when trx->state trx_que_t que_state; /*!< valid when trx->state
== TRX_STATE_ACTIVE: TRX_QUE_RUNNING, == TRX_STATE_ACTIVE: TRX_QUE_RUNNING,
TRX_QUE_LOCK_WAIT, ... */ 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 lock_t* wait_lock; /*!< if trx execution state is
TRX_QUE_LOCK_WAIT, this points to TRX_QUE_LOCK_WAIT, this points to
the lock request, otherwise this is the lock request, otherwise this is
......
/***************************************************************************** /*****************************************************************************
Copyright (c) 1996, 2017, Oracle and/or its affiliates. All Rights Reserved. Copyright (c) 1996, 2017, Oracle and/or its affiliates. All Rights Reserved.
Copyright (c) 2014, 2021, MariaDB Corporation. Copyright (c) 2014, 2022, MariaDB Corporation.
This program is free software; you can redistribute it and/or modify it under 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 the terms of the GNU General Public License as published by the Free Software
...@@ -970,7 +970,8 @@ lock_rec_has_expl( ...@@ -970,7 +970,8 @@ lock_rec_has_expl(
|| (precise_mode & LOCK_MODE_MASK) == LOCK_X); || (precise_mode & LOCK_MODE_MASK) == LOCK_X);
ut_ad(!(precise_mode & LOCK_INSERT_INTENTION)); ut_ad(!(precise_mode & LOCK_INSERT_INTENTION));
for (lock = lock_rec_get_first(&lock_sys.rec_hash, block, heap_no); for (lock = lock_rec_get_first(&lock_sys.rec_hash, block->page.id(),
heap_no);
lock != NULL; lock != NULL;
lock = lock_rec_get_next(heap_no, lock)) { lock = lock_rec_get_next(heap_no, lock)) {
...@@ -1024,7 +1025,7 @@ lock_rec_other_has_expl_req( ...@@ -1024,7 +1025,7 @@ lock_rec_other_has_expl_req(
} }
for (lock_t* lock = lock_rec_get_first(&lock_sys.rec_hash, for (lock_t* lock = lock_rec_get_first(&lock_sys.rec_hash,
block, heap_no); block->page.id(), heap_no);
lock != NULL; lock != NULL;
lock = lock_rec_get_next(heap_no, lock)) { lock = lock_rec_get_next(heap_no, lock)) {
...@@ -1075,19 +1076,18 @@ static void wsrep_kill_victim(const trx_t * const trx, const lock_t *lock) ...@@ -1075,19 +1076,18 @@ 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 Checks if some other transaction has a conflicting explicit lock request
in the queue, so that we have to wait. 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 */ @return lock or NULL */
static static lock_t *lock_rec_other_has_conflicting(unsigned mode,
lock_t* const buf_block_t *block,
lock_rec_other_has_conflicting( ulint heap_no, const trx_t *trx,
/*===========================*/ bool *was_ignored= nullptr)
unsigned 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; lock_t* lock;
...@@ -1095,10 +1095,21 @@ lock_rec_other_has_conflicting( ...@@ -1095,10 +1095,21 @@ lock_rec_other_has_conflicting(
bool is_supremum = (heap_no == PAGE_HEAP_NO_SUPREMUM); bool is_supremum = (heap_no == PAGE_HEAP_NO_SUPREMUM);
for (lock = lock_rec_get_first(&lock_sys.rec_hash, block, heap_no); for (lock = lock_rec_get_first(&lock_sys.rec_hash, block->page.id(),
heap_no);
lock != NULL; lock != NULL;
lock = lock_rec_get_next(heap_no, lock)) { 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)) { if (lock_rec_has_to_wait(true, trx, mode, lock, is_supremum)) {
#ifdef WITH_WSREP #ifdef WITH_WSREP
if (trx->is_wsrep()) { if (trx->is_wsrep()) {
...@@ -1255,6 +1266,7 @@ static void check_trx_state(const trx_t *trx) ...@@ -1255,6 +1266,7 @@ static void check_trx_state(const trx_t *trx)
/** Create a new record lock and inserts it to the lock queue, /** Create a new record lock and inserts it to the lock queue,
without checking for deadlocks or conflicts. 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 @param[in] type_mode lock mode and wait flag; type will be replaced
with LOCK_REC with LOCK_REC
@param[in] page_id index page number @param[in] page_id index page number
...@@ -1263,11 +1275,15 @@ without checking for deadlocks or conflicts. ...@@ -1263,11 +1275,15 @@ without checking for deadlocks or conflicts.
@param[in] index the index tree @param[in] index the index tree
@param[in,out] trx transaction @param[in,out] trx transaction
@param[in] holds_trx_mutex whether the caller holds trx->mutex @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 */ @return created lock */
lock_t* lock_t*
lock_rec_create_low( lock_rec_create_low(
lock_t* c_lock,
#ifdef WITH_WSREP #ifdef WITH_WSREP
lock_t* c_lock, /*!< conflicting lock */
que_thr_t* thr, /*!< thread owning trx */ que_thr_t* thr, /*!< thread owning trx */
#endif #endif
unsigned type_mode, unsigned type_mode,
...@@ -1276,7 +1292,8 @@ lock_rec_create_low( ...@@ -1276,7 +1292,8 @@ lock_rec_create_low(
ulint heap_no, ulint heap_no,
dict_index_t* index, dict_index_t* index,
trx_t* trx, trx_t* trx,
bool holds_trx_mutex) bool holds_trx_mutex,
bool insert_before_waiting)
{ {
lock_t* lock; lock_t* lock;
ulint n_bits; ulint n_bits;
...@@ -1385,7 +1402,7 @@ lock_rec_create_low( ...@@ -1385,7 +1402,7 @@ lock_rec_create_low(
} }
trx->lock.que_state = TRX_QUE_LOCK_WAIT; trx->lock.que_state = TRX_QUE_LOCK_WAIT;
lock_set_lock_and_trx_wait(lock, trx); lock_set_lock_and_trx_wait(lock, trx, c_lock);
UT_LIST_ADD_LAST(trx->lock.trx_locks, lock); UT_LIST_ADD_LAST(trx->lock.trx_locks, lock);
trx->lock.wait_thr = thr; trx->lock.wait_thr = thr;
...@@ -1413,15 +1430,46 @@ lock_rec_create_low( ...@@ -1413,15 +1430,46 @@ lock_rec_create_low(
trx_mutex_exit(c_lock->trx); trx_mutex_exit(c_lock->trx);
} else } else
#endif /* WITH_WSREP */ #endif /* WITH_WSREP */
if (!(type_mode & (LOCK_WAIT | LOCK_PREDICATE | LOCK_PRDT_PAGE)) if (insert_before_waiting
&& innodb_lock_schedule_algorithm && !(type_mode & (LOCK_PREDICATE | LOCK_PRDT_PAGE))) {
== INNODB_LOCK_SCHEDULE_ALGORITHM_VATS /* Try to insert the lock just after the last non-waiting
&& !thd_is_replication_slave_thread(trx->mysql_thd)) { lock of the current transaction which immediately
HASH_PREPEND(lock_t, hash, &lock_sys.rec_hash, precedes the first waiting lock request. */
page_id.fold(), lock); hash_table_t *lock_hash = lock_hash_get(type_mode);
} else { hash_cell_t& cell = lock_hash->array[lock_hash->calc_hash(
HASH_INSERT(lock_t, hash, lock_hash_get(type_mode), page_id.fold())];
page_id.fold(), lock);
lock_t* last_non_waiting = nullptr;
for (lock_t* l = lock_rec_get_first(lock_hash, page_id,
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
&& !thd_is_replication_slave_thread(trx->mysql_thd)) {
HASH_PREPEND(lock_t, hash, &lock_sys.rec_hash,
page_id.fold(), lock);
} else {
HASH_INSERT(lock_t, hash, lock_hash_get(type_mode),
page_id.fold(), lock);
}
} }
if (!holds_trx_mutex) { if (!holds_trx_mutex) {
...@@ -1429,7 +1477,7 @@ lock_rec_create_low( ...@@ -1429,7 +1477,7 @@ lock_rec_create_low(
} }
ut_ad(trx_mutex_own(trx)); ut_ad(trx_mutex_own(trx));
if (type_mode & LOCK_WAIT) { if (type_mode & LOCK_WAIT) {
lock_set_lock_and_trx_wait(lock, trx); lock_set_lock_and_trx_wait(lock, trx, c_lock);
} }
UT_LIST_ADD_LAST(trx->lock.trx_locks, lock); UT_LIST_ADD_LAST(trx->lock.trx_locks, lock);
if (!holds_trx_mutex) { if (!holds_trx_mutex) {
...@@ -1576,6 +1624,7 @@ lock_rec_insert_to_head( ...@@ -1576,6 +1624,7 @@ lock_rec_insert_to_head(
/** Enqueue a waiting request for a lock which cannot be granted immediately. /** Enqueue a waiting request for a lock which cannot be granted immediately.
Check for deadlocks. Check for deadlocks.
@param[in] c_lock conflicting lock
@param[in] type_mode the requested lock mode (LOCK_S or LOCK_X) @param[in] type_mode the requested lock mode (LOCK_S or LOCK_X)
possibly ORed with LOCK_GAP or possibly ORed with LOCK_GAP or
LOCK_REC_NOT_GAP, ORed with LOCK_REC_NOT_GAP, ORed with
...@@ -1594,9 +1643,7 @@ Check for deadlocks. ...@@ -1594,9 +1643,7 @@ Check for deadlocks.
(or it happened to commit) */ (or it happened to commit) */
dberr_t dberr_t
lock_rec_enqueue_waiting( lock_rec_enqueue_waiting(
#ifdef WITH_WSREP lock_t* c_lock,
lock_t* c_lock, /*!< conflicting lock */
#endif
unsigned type_mode, unsigned type_mode,
const buf_block_t* block, const buf_block_t* block,
ulint heap_no, ulint heap_no,
...@@ -1634,9 +1681,9 @@ lock_rec_enqueue_waiting( ...@@ -1634,9 +1681,9 @@ lock_rec_enqueue_waiting(
/* Enqueue the lock request that will wait to be granted, note that /* Enqueue the lock request that will wait to be granted, note that
we already own the trx mutex. */ we already own the trx mutex. */
lock_t* lock = lock_rec_create( lock_t* lock = lock_rec_create(c_lock,
#ifdef WITH_WSREP #ifdef WITH_WSREP
c_lock, thr, thr,
#endif #endif
type_mode | LOCK_WAIT, block, heap_no, index, trx, TRUE); type_mode | LOCK_WAIT, block, heap_no, index, trx, TRUE);
...@@ -1730,22 +1777,20 @@ on the record, and the request to be added is not a waiting request, we ...@@ -1730,22 +1777,20 @@ 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, 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 just setting the appropriate bit in its bitmap. This is a low-level function
which does NOT check for deadlocks or lock compatibility! 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 */ @return lock where the bit was set */
static static void lock_rec_add_to_queue(unsigned type_mode, const buf_block_t *block,
void ulint heap_no, dict_index_t *index,
lock_rec_add_to_queue( trx_t *trx, bool caller_owns_trx_mutex,
/*==================*/ bool insert_before_waiting= false)
unsigned 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 #ifdef UNIV_DEBUG
ut_ad(lock_mutex_own()); ut_ad(lock_mutex_own());
...@@ -1834,11 +1879,16 @@ lock_rec_add_to_queue( ...@@ -1834,11 +1879,16 @@ lock_rec_add_to_queue(
} }
} }
lock_rec_create( /* 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,
#ifdef WITH_WSREP #ifdef WITH_WSREP
NULL, NULL, NULL,
#endif #endif
type_mode, block, heap_no, index, trx, caller_owns_trx_mutex); type_mode, block, heap_no, index, trx, caller_owns_trx_mutex,
insert_before_waiting);
} }
/*********************************************************************//** /*********************************************************************//**
...@@ -1896,28 +1946,23 @@ lock_rec_lock( ...@@ -1896,28 +1946,23 @@ lock_rec_lock(
/* Do nothing if the trx already has a strong enough lock on rec */ /* Do nothing if the trx already has a strong enough lock on rec */
if (!lock_rec_has_expl(mode, block, heap_no, trx)) if (!lock_rec_has_expl(mode, block, heap_no, trx))
{ {
if ( bool was_ignored = false;
#ifdef WITH_WSREP if (lock_t *c_lock= lock_rec_other_has_conflicting(
lock_t *c_lock= mode, block, heap_no, trx, &was_ignored))
#endif
lock_rec_other_has_conflicting(mode, block, heap_no, trx))
{ {
/* /*
If another transaction has a non-gap conflicting If another transaction has a non-gap conflicting
request in the queue, as this transaction does not request in the queue, as this transaction does not
have a lock strong enough already granted on the have a lock strong enough already granted on the
record, we have to wait. */ record, we have to wait. */
err = lock_rec_enqueue_waiting( err = lock_rec_enqueue_waiting(c_lock, mode, block, heap_no, index,
#ifdef WITH_WSREP thr, NULL);
c_lock,
#endif /* WITH_WSREP */
mode, block, heap_no, index, thr, NULL);
} }
else if (!impl) else if (!impl)
{ {
/* Set the requested lock on the record. */ /* Set the requested lock on the record. */
lock_rec_add_to_queue(LOCK_REC | mode, block, heap_no, index, trx, lock_rec_add_to_queue(LOCK_REC | mode, block, heap_no, index, trx,
true); true, was_ignored);
err= DB_SUCCESS_LOCKED_REC; err= DB_SUCCESS_LOCKED_REC;
} }
} }
...@@ -1943,9 +1988,9 @@ lock_rec_lock( ...@@ -1943,9 +1988,9 @@ lock_rec_lock(
Note that we don't own the trx mutex. Note that we don't own the trx mutex.
*/ */
if (!impl) if (!impl)
lock_rec_create( lock_rec_create(NULL,
#ifdef WITH_WSREP #ifdef WITH_WSREP
NULL, NULL, NULL,
#endif #endif
mode, block, heap_no, index, trx, false); mode, block, heap_no, index, trx, false);
...@@ -2168,8 +2213,17 @@ static void lock_rec_dequeue_from_page(lock_t* in_lock) ...@@ -2168,8 +2213,17 @@ static void lock_rec_dequeue_from_page(lock_t* in_lock)
if (!lock_get_wait(lock)) { if (!lock_get_wait(lock)) {
continue; continue;
} }
const lock_t* c = lock_rec_has_to_wait_in_queue(lock);
if (!c) { 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 {
/* Grant the lock */ /* Grant the lock */
ut_ad(lock->trx != in_lock->trx); ut_ad(lock->trx != in_lock->trx);
lock_grant(lock); lock_grant(lock);
...@@ -2259,7 +2313,7 @@ lock_rec_reset_and_release_wait_low( ...@@ -2259,7 +2313,7 @@ lock_rec_reset_and_release_wait_low(
ut_ad(lock_mutex_own()); ut_ad(lock_mutex_own());
for (lock = lock_rec_get_first(hash, block, heap_no); for (lock = lock_rec_get_first(hash, block->page.id(), heap_no);
lock != NULL; lock != NULL;
lock = lock_rec_get_next(heap_no, lock)) { lock = lock_rec_get_next(heap_no, lock)) {
...@@ -2321,7 +2375,8 @@ lock_rec_inherit_to_gap( ...@@ -2321,7 +2375,8 @@ lock_rec_inherit_to_gap(
DO want S-locks/X-locks(taken for replace) set by a consistency DO want S-locks/X-locks(taken for replace) set by a consistency
constraint to be inherited also then. */ constraint to be inherited also then. */
for (lock = lock_rec_get_first(&lock_sys.rec_hash, block, heap_no); for (lock = lock_rec_get_first(&lock_sys.rec_hash, block->page.id(),
heap_no);
lock != NULL; lock != NULL;
lock = lock_rec_get_next(heap_no, lock)) { lock = lock_rec_get_next(heap_no, lock)) {
...@@ -2357,7 +2412,8 @@ lock_rec_inherit_to_gap_if_gap_lock( ...@@ -2357,7 +2412,8 @@ lock_rec_inherit_to_gap_if_gap_lock(
lock_mutex_enter(); lock_mutex_enter();
for (lock = lock_rec_get_first(&lock_sys.rec_hash, block, heap_no); for (lock = lock_rec_get_first(&lock_sys.rec_hash, block->page.id(),
heap_no);
lock != NULL; lock != NULL;
lock = lock_rec_get_next(heap_no, lock)) { lock = lock_rec_get_next(heap_no, lock)) {
...@@ -2400,12 +2456,12 @@ lock_rec_move_low( ...@@ -2400,12 +2456,12 @@ lock_rec_move_low(
/* If the lock is predicate lock, it resides on INFIMUM record */ /* If the lock is predicate lock, it resides on INFIMUM record */
ut_ad(lock_rec_get_first( ut_ad(lock_rec_get_first(
lock_hash, receiver, receiver_heap_no) == NULL lock_hash, receiver->page.id(), receiver_heap_no) == NULL
|| lock_hash == &lock_sys.prdt_hash || lock_hash == &lock_sys.prdt_hash
|| lock_hash == &lock_sys.prdt_page_hash); || lock_hash == &lock_sys.prdt_page_hash);
for (lock = lock_rec_get_first(lock_hash, for (lock = lock_rec_get_first(lock_hash,
donator, donator_heap_no); donator->page.id(), donator_heap_no);
lock != NULL; lock != NULL;
lock = lock_rec_get_next(donator_heap_no, lock)) { lock = lock_rec_get_next(donator_heap_no, lock)) {
...@@ -2414,7 +2470,8 @@ lock_rec_move_low( ...@@ -2414,7 +2470,8 @@ lock_rec_move_low(
lock_rec_reset_nth_bit(lock, donator_heap_no); lock_rec_reset_nth_bit(lock, donator_heap_no);
if (type_mode & LOCK_WAIT) { if (type_mode & LOCK_WAIT) {
lock_reset_lock_and_trx_wait(lock); ut_ad(lock->trx->lock.wait_lock == lock);
lock->type_mode &= ~LOCK_WAIT;
} }
/* Note that we FIRST reset the bit, and then set the lock: /* Note that we FIRST reset the bit, and then set the lock:
...@@ -2426,7 +2483,7 @@ lock_rec_move_low( ...@@ -2426,7 +2483,7 @@ lock_rec_move_low(
} }
ut_ad(!lock_rec_get_first(&lock_sys.rec_hash, ut_ad(!lock_rec_get_first(&lock_sys.rec_hash,
donator, donator_heap_no)); donator->page.id(), donator_heap_no));
} }
/** Move all the granted locks to the front of the given lock list. /** Move all the granted locks to the front of the given lock list.
...@@ -2531,8 +2588,8 @@ lock_move_reorganize_page( ...@@ -2531,8 +2588,8 @@ lock_move_reorganize_page(
lock_rec_bitmap_reset(lock); lock_rec_bitmap_reset(lock);
if (lock_get_wait(lock)) { if (lock_get_wait(lock)) {
ut_ad(lock->trx->lock.wait_lock == lock);
lock_reset_lock_and_trx_wait(lock); lock->type_mode&= ~LOCK_WAIT;
} }
lock = lock_rec_get_next_on_page(lock); lock = lock_rec_get_next_on_page(lock);
...@@ -2708,7 +2765,9 @@ lock_move_rec_list_end( ...@@ -2708,7 +2765,9 @@ lock_move_rec_list_end(
ut_ad(!page_rec_is_metadata(orec)); ut_ad(!page_rec_is_metadata(orec));
if (type_mode & LOCK_WAIT) { if (type_mode & LOCK_WAIT) {
lock_reset_lock_and_trx_wait(lock); ut_ad(lock->trx->lock.wait_lock ==
lock);
lock->type_mode&= ~LOCK_WAIT;
} }
lock_rec_add_to_queue( lock_rec_add_to_queue(
...@@ -2806,7 +2865,9 @@ lock_move_rec_list_start( ...@@ -2806,7 +2865,9 @@ lock_move_rec_list_start(
ut_ad(!page_rec_is_metadata(prev)); ut_ad(!page_rec_is_metadata(prev));
if (type_mode & LOCK_WAIT) { if (type_mode & LOCK_WAIT) {
lock_reset_lock_and_trx_wait(lock); ut_ad(lock->trx->lock.wait_lock
== lock);
lock->type_mode&= ~LOCK_WAIT;
} }
lock_rec_add_to_queue( lock_rec_add_to_queue(
...@@ -2902,7 +2963,9 @@ lock_rtr_move_rec_list( ...@@ -2902,7 +2963,9 @@ lock_rtr_move_rec_list(
if (rec1_heap_no < lock->un_member.rec_lock.n_bits if (rec1_heap_no < lock->un_member.rec_lock.n_bits
&& lock_rec_reset_nth_bit(lock, rec1_heap_no)) { && lock_rec_reset_nth_bit(lock, rec1_heap_no)) {
if (type_mode & LOCK_WAIT) { if (type_mode & LOCK_WAIT) {
lock_reset_lock_and_trx_wait(lock); ut_ad(lock->trx->lock.wait_lock
== lock);
lock->type_mode&= ~LOCK_WAIT;
} }
lock_rec_add_to_queue( lock_rec_add_to_queue(
...@@ -3348,10 +3411,8 @@ lock_table_create( ...@@ -3348,10 +3411,8 @@ lock_table_create(
in dictionary cache */ in dictionary cache */
unsigned type_mode,/*!< in: lock mode possibly ORed with unsigned type_mode,/*!< in: lock mode possibly ORed with
LOCK_WAIT */ LOCK_WAIT */
trx_t* trx /*!< in: trx */ trx_t* trx, /*!< in: trx */
#ifdef WITH_WSREP lock_t* c_lock = NULL /*!< in: conflicting lock */
, lock_t* c_lock = NULL /*!< in: conflicting lock */
#endif
) )
{ {
lock_t* lock; lock_t* lock;
...@@ -3434,8 +3495,7 @@ lock_table_create( ...@@ -3434,8 +3495,7 @@ lock_table_create(
ut_list_append(table->locks, lock, TableLockGetNode()); ut_list_append(table->locks, lock, TableLockGetNode());
if (type_mode & LOCK_WAIT) { 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); lock->trx->lock.table_locks.push_back(lock);
...@@ -3590,10 +3650,8 @@ lock_table_enqueue_waiting( ...@@ -3590,10 +3650,8 @@ lock_table_enqueue_waiting(
unsigned mode, /*!< in: lock mode this transaction is unsigned mode, /*!< in: lock mode this transaction is
requesting */ requesting */
dict_table_t* table, /*!< in/out: table */ dict_table_t* table, /*!< in/out: table */
que_thr_t* thr /*!< in: query thread */ que_thr_t* thr, /*!< in: query thread */
#ifdef WITH_WSREP lock_t* c_lock /*!< in: conflicting lock or NULL */
, lock_t* c_lock /*!< in: conflicting lock or NULL */
#endif
) )
{ {
trx_t* trx; trx_t* trx;
...@@ -3624,11 +3682,7 @@ lock_table_enqueue_waiting( ...@@ -3624,11 +3682,7 @@ lock_table_enqueue_waiting(
#endif /* WITH_WSREP */ #endif /* WITH_WSREP */
/* Enqueue the lock request that will wait to be granted */ /* Enqueue the lock request that will wait to be granted */
lock = lock_table_create(table, mode | LOCK_WAIT, trx lock = lock_table_create(table, mode | LOCK_WAIT, trx, c_lock);
#ifdef WITH_WSREP
, c_lock
#endif
);
const trx_t* victim_trx = const trx_t* victim_trx =
DeadlockChecker::check_and_resolve(lock, trx); DeadlockChecker::check_and_resolve(lock, trx);
...@@ -3784,11 +3838,7 @@ lock_table( ...@@ -3784,11 +3838,7 @@ lock_table(
if (wait_for != NULL) { if (wait_for != NULL) {
err = lock_table_enqueue_waiting(flags | mode, table, err = lock_table_enqueue_waiting(flags | mode, table,
thr thr, wait_for);
#ifdef WITH_WSREP
, wait_for
#endif
);
} else { } else {
lock_table_create(table, flags | mode, trx); lock_table_create(table, flags | mode, trx);
...@@ -3836,7 +3886,7 @@ lock_table_ix_resurrect( ...@@ -3836,7 +3886,7 @@ lock_table_ix_resurrect(
Checks if a waiting table lock request still has to wait in a queue. Checks if a waiting table lock request still has to wait in a queue.
@return TRUE if still has to wait */ @return TRUE if still has to wait */
static static
bool const lock_t*
lock_table_has_to_wait_in_queue( lock_table_has_to_wait_in_queue(
/*============================*/ /*============================*/
const lock_t* wait_lock) /*!< in: waiting table lock */ const lock_t* wait_lock) /*!< in: waiting table lock */
...@@ -3855,11 +3905,11 @@ lock_table_has_to_wait_in_queue( ...@@ -3855,11 +3905,11 @@ lock_table_has_to_wait_in_queue(
if (lock_has_to_wait(wait_lock, lock)) { if (lock_has_to_wait(wait_lock, lock)) {
return(true); return(lock);
} }
} }
return(false); return(NULL);
} }
/*************************************************************//** /*************************************************************//**
...@@ -3888,9 +3938,17 @@ lock_table_dequeue( ...@@ -3888,9 +3938,17 @@ lock_table_dequeue(
lock != NULL; lock != NULL;
lock = UT_LIST_GET_NEXT(un_member.tab_lock.locks, lock)) { lock = UT_LIST_GET_NEXT(un_member.tab_lock.locks, lock)) {
if (lock_get_wait(lock) if (!lock_get_wait(lock))
&& !lock_table_has_to_wait_in_queue(lock)) { continue;
ut_ad(lock->trx->lock.wait_trx);
ut_ad(lock->trx->lock.wait_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 */ /* Grant the lock */
ut_ad(in_lock->trx != lock->trx); ut_ad(in_lock->trx != lock->trx);
lock_grant(lock); lock_grant(lock);
...@@ -4035,7 +4093,8 @@ lock_rec_unlock( ...@@ -4035,7 +4093,8 @@ lock_rec_unlock(
lock_mutex_enter(); lock_mutex_enter();
trx_mutex_enter(trx); trx_mutex_enter(trx);
first_lock = lock_rec_get_first(&lock_sys.rec_hash, block, heap_no); first_lock = lock_rec_get_first(&lock_sys.rec_hash, block->page.id(),
heap_no);
/* Find the last lock with the same lock_mode and transaction /* Find the last lock with the same lock_mode and transaction
on the record. */ on the record. */
...@@ -4078,8 +4137,16 @@ lock_rec_unlock( ...@@ -4078,8 +4137,16 @@ lock_rec_unlock(
if (!lock_get_wait(lock)) { if (!lock_get_wait(lock)) {
continue; continue;
} }
const lock_t* c = lock_rec_has_to_wait_in_queue(lock); ut_ad(lock->trx->lock.wait_trx);
if (!c) { 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 {
/* Grant the lock */ /* Grant the lock */
ut_ad(trx != lock->trx); ut_ad(trx != lock->trx);
lock_grant(lock); lock_grant(lock);
...@@ -4773,7 +4840,7 @@ lock_rec_queue_validate( ...@@ -4773,7 +4840,7 @@ lock_rec_queue_validate(
if (!page_rec_is_user_rec(rec)) { if (!page_rec_is_user_rec(rec)) {
for (lock = lock_rec_get_first(&lock_sys.rec_hash, for (lock = lock_rec_get_first(&lock_sys.rec_hash,
block, heap_no); block->page.id(), heap_no);
lock != NULL; lock != NULL;
lock = lock_rec_get_next_const(heap_no, lock)) { lock = lock_rec_get_next_const(heap_no, lock)) {
...@@ -4847,7 +4914,7 @@ lock_rec_queue_validate( ...@@ -4847,7 +4914,7 @@ lock_rec_queue_validate(
wsrep_report_bf_lock_wait(impl_trx->mysql_thd, impl_trx->id); 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); wsrep_report_bf_lock_wait(other_lock->trx->mysql_thd, other_lock->trx->id);
if (!lock_rec_has_expl(LOCK_X | LOCK_REC_NOT_GAP, if (!lock_rec_has_expl(LOCK_S | LOCK_REC_NOT_GAP,
block, heap_no, block, heap_no,
impl_trx)) { impl_trx)) {
ib::info() << "WSREP impl BF lock conflict"; ib::info() << "WSREP impl BF lock conflict";
...@@ -4856,7 +4923,20 @@ lock_rec_queue_validate( ...@@ -4856,7 +4923,20 @@ lock_rec_queue_validate(
#endif /* WITH_WSREP */ #endif /* WITH_WSREP */
{ {
ut_ad(lock_get_wait(other_lock)); ut_ad(lock_get_wait(other_lock));
ut_ad(lock_rec_has_expl(LOCK_X | LOCK_REC_NOT_GAP, /* 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,
block, heap_no, impl_trx)); block, heap_no, impl_trx));
} }
} }
...@@ -4864,7 +4944,8 @@ lock_rec_queue_validate( ...@@ -4864,7 +4944,8 @@ lock_rec_queue_validate(
mutex_exit(&impl_trx->mutex); mutex_exit(&impl_trx->mutex);
} }
for (lock = lock_rec_get_first(&lock_sys.rec_hash, block, heap_no); for (lock = lock_rec_get_first(&lock_sys.rec_hash, block->page.id(),
heap_no);
lock != NULL; lock != NULL;
lock = lock_rec_get_next_const(heap_no, lock)) { lock = lock_rec_get_next_const(heap_no, lock)) {
ut_ad(!lock->trx->read_only ut_ad(!lock->trx->read_only
...@@ -5199,7 +5280,8 @@ lock_rec_insert_check_and_lock( ...@@ -5199,7 +5280,8 @@ lock_rec_insert_check_and_lock(
BTR_NO_LOCKING_FLAG and skip the locking altogether. */ BTR_NO_LOCKING_FLAG and skip the locking altogether. */
ut_ad(lock_table_has(trx, index->table, LOCK_IX)); ut_ad(lock_table_has(trx, index->table, LOCK_IX));
lock = lock_rec_get_first(&lock_sys.rec_hash, block, heap_no); lock = lock_rec_get_first(&lock_sys.rec_hash, block->page.id(),
heap_no);
if (lock == NULL) { if (lock == NULL) {
/* We optimize CPU time usage in the simplest case */ /* We optimize CPU time usage in the simplest case */
...@@ -5238,19 +5320,13 @@ lock_rec_insert_check_and_lock( ...@@ -5238,19 +5320,13 @@ lock_rec_insert_check_and_lock(
const unsigned type_mode = LOCK_X | LOCK_GAP | LOCK_INSERT_INTENTION; const unsigned type_mode = LOCK_X | LOCK_GAP | LOCK_INSERT_INTENTION;
if ( if (lock_t* c_lock =
#ifdef WITH_WSREP
lock_t* c_lock =
#endif /* WITH_WSREP */
lock_rec_other_has_conflicting(type_mode, block, heap_no, trx)) { lock_rec_other_has_conflicting(type_mode, block, heap_no, trx)) {
/* Note that we may get DB_SUCCESS also here! */ /* Note that we may get DB_SUCCESS also here! */
trx_mutex_enter(trx); trx_mutex_enter(trx);
err = lock_rec_enqueue_waiting( err = lock_rec_enqueue_waiting(c_lock, type_mode, block,
#ifdef WITH_WSREP heap_no, index, thr, NULL);
c_lock,
#endif /* WITH_WSREP */
type_mode, block, heap_no, index, thr, NULL);
trx_mutex_exit(trx); trx_mutex_exit(trx);
} else { } else {
...@@ -5327,7 +5403,7 @@ lock_rec_convert_impl_to_expl_for_trx( ...@@ -5327,7 +5403,7 @@ lock_rec_convert_impl_to_expl_for_trx(
&& !lock_rec_has_expl(LOCK_X | LOCK_REC_NOT_GAP, && !lock_rec_has_expl(LOCK_X | LOCK_REC_NOT_GAP,
block, heap_no, trx)) { block, heap_no, trx)) {
lock_rec_add_to_queue(LOCK_REC | LOCK_X | LOCK_REC_NOT_GAP, lock_rec_add_to_queue(LOCK_REC | LOCK_X | LOCK_REC_NOT_GAP,
block, heap_no, index, trx, true); block, heap_no, index, trx, true, true);
} }
lock_mutex_exit(); lock_mutex_exit();
......
...@@ -243,7 +243,7 @@ lock_prdt_has_lock( ...@@ -243,7 +243,7 @@ lock_prdt_has_lock(
ut_ad(!(precise_mode & LOCK_INSERT_INTENTION)); ut_ad(!(precise_mode & LOCK_INSERT_INTENTION));
for (lock = lock_rec_get_first( for (lock = lock_rec_get_first(
lock_hash_get(type_mode), block, PRDT_HEAPNO); lock_hash_get(type_mode), block->page.id(), PRDT_HEAPNO);
lock != NULL; lock != NULL;
lock = lock_rec_get_next(PRDT_HEAPNO, lock)) { lock = lock_rec_get_next(PRDT_HEAPNO, lock)) {
ut_ad(lock->type_mode & (LOCK_PREDICATE | LOCK_PRDT_PAGE)); ut_ad(lock->type_mode & (LOCK_PREDICATE | LOCK_PRDT_PAGE));
...@@ -298,7 +298,7 @@ lock_prdt_other_has_conflicting( ...@@ -298,7 +298,7 @@ lock_prdt_other_has_conflicting(
ut_ad(lock_mutex_own()); ut_ad(lock_mutex_own());
for (lock_t* lock = lock_rec_get_first( for (lock_t* lock = lock_rec_get_first(
lock_hash_get(mode), block, PRDT_HEAPNO); lock_hash_get(mode), block->page.id(), PRDT_HEAPNO);
lock != NULL; lock != NULL;
lock = lock_rec_get_next(PRDT_HEAPNO, lock)) { lock = lock_rec_get_next(PRDT_HEAPNO, lock)) {
...@@ -489,9 +489,13 @@ lock_prdt_add_to_queue( ...@@ -489,9 +489,13 @@ lock_prdt_add_to_queue(
} }
} }
lock = lock_rec_create( /* 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,
#ifdef WITH_WSREP #ifdef WITH_WSREP
NULL, NULL, /* FIXME: replicate SPATIAL INDEX locks */ NULL, /* FIXME: replicate SPATIAL INDEX locks */
#endif #endif
type_mode, block, PRDT_HEAPNO, index, trx, type_mode, block, PRDT_HEAPNO, index, trx,
caller_owns_trx_mutex); caller_owns_trx_mutex);
...@@ -543,7 +547,8 @@ lock_prdt_insert_check_and_lock( ...@@ -543,7 +547,8 @@ lock_prdt_insert_check_and_lock(
lock_t* lock; lock_t* lock;
/* Only need to check locks on prdt_hash */ /* Only need to check locks on prdt_hash */
lock = lock_rec_get_first(&lock_sys.prdt_hash, block, PRDT_HEAPNO); lock = lock_rec_get_first(&lock_sys.prdt_hash, block->page.id(),
PRDT_HEAPNO);
if (lock == NULL) { if (lock == NULL) {
lock_mutex_exit(); lock_mutex_exit();
...@@ -581,9 +586,7 @@ lock_prdt_insert_check_and_lock( ...@@ -581,9 +586,7 @@ lock_prdt_insert_check_and_lock(
trx_mutex_enter(trx); trx_mutex_enter(trx);
err = lock_rec_enqueue_waiting( err = lock_rec_enqueue_waiting(
#ifdef WITH_WSREP
NULL, /* FIXME: replicate SPATIAL INDEX locks */ NULL, /* FIXME: replicate SPATIAL INDEX locks */
#endif
LOCK_X | LOCK_PREDICATE | LOCK_INSERT_INTENTION, LOCK_X | LOCK_PREDICATE | LOCK_INSERT_INTENTION,
block, PRDT_HEAPNO, index, thr, prdt); block, PRDT_HEAPNO, index, thr, prdt);
...@@ -822,9 +825,9 @@ lock_prdt_lock( ...@@ -822,9 +825,9 @@ lock_prdt_lock(
lock_t* lock = lock_sys.get_first(hash, block->page.id()); lock_t* lock = lock_sys.get_first(hash, block->page.id());
if (lock == NULL) { if (lock == NULL) {
lock = lock_rec_create( lock = lock_rec_create(NULL,
#ifdef WITH_WSREP #ifdef WITH_WSREP
NULL, NULL, /* FIXME: replicate SPATIAL INDEX locks */ NULL, /* FIXME: replicate SPATIAL INDEX locks */
#endif #endif
prdt_mode, block, PRDT_HEAPNO, prdt_mode, block, PRDT_HEAPNO,
index, trx, FALSE); index, trx, FALSE);
...@@ -854,10 +857,8 @@ lock_prdt_lock( ...@@ -854,10 +857,8 @@ lock_prdt_lock(
if (wait_for != NULL) { if (wait_for != NULL) {
err = lock_rec_enqueue_waiting( err = lock_rec_enqueue_waiting(
#ifdef WITH_WSREP
NULL, /* FIXME: replicate NULL, /* FIXME: replicate
SPATIAL INDEX locks */ SPATIAL INDEX locks */
#endif
prdt_mode, prdt_mode,
block, PRDT_HEAPNO, block, PRDT_HEAPNO,
index, thr, prdt); index, thr, prdt);
...@@ -937,9 +938,9 @@ lock_place_prdt_page_lock( ...@@ -937,9 +938,9 @@ lock_place_prdt_page_lock(
} }
if (lock == NULL) { if (lock == NULL) {
lock = lock_rec_create_low( lock = lock_rec_create_low(NULL,
#ifdef WITH_WSREP #ifdef WITH_WSREP
NULL, NULL, /* FIXME: replicate SPATIAL INDEX locks */ NULL, /* FIXME: replicate SPATIAL INDEX locks */
#endif #endif
mode, page_id, NULL, PRDT_HEAPNO, mode, page_id, NULL, PRDT_HEAPNO,
index, trx, FALSE); index, trx, FALSE);
...@@ -985,7 +986,7 @@ lock_prdt_rec_move( ...@@ -985,7 +986,7 @@ lock_prdt_rec_move(
lock_mutex_enter(); lock_mutex_enter();
for (lock_t *lock = lock_rec_get_first(&lock_sys.prdt_hash, for (lock_t *lock = lock_rec_get_first(&lock_sys.prdt_hash,
donator, PRDT_HEAPNO); donator->page.id(), PRDT_HEAPNO);
lock != NULL; lock != NULL;
lock = lock_rec_get_next(PRDT_HEAPNO, lock)) { lock = lock_rec_get_next(PRDT_HEAPNO, lock)) {
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment