Bug#50561: ALTER PARTITIONS does not have adequate lock, breaks with

concurrent I_S query

There were two problem:
1) MYSQL_LOCK_IGNORE_FLUSH also ignored name locks
2) there was a race between abort_and_upgrade_locks and
   alter_close_tables
   (i.e. remove_table_from_cache and
    close_data_files_and_morph_locks)

Which allowed the table to be opened with MYSQL_LOCK_IGNORE_FLUSH flag
resulting in renaming a partition that was already in use,
which could cause the table to be unusable.

Solution was to not allow IGNORE_FLUSH to skip waiting for
a named locked table.

And to not release the LOCK_open mutex between the
calls to remove_table_from_cache and
close_data_files_and_morph_locks by merging the functions
abort_and_upgrade_locks and alter_close_tables.

mysql-test/suite/parts/r/partition_debug_sync_innodb.result

+#
+# Bug#50561: ALTER PARTITIONS does not have adequate lock, breaks with
+#            concurrent I_S query
+create table t1 (a int)
+engine = innodb
+partition by range (a)
+(partition p0 values less than MAXVALUE);
+insert into t1 values (1), (11), (21), (33);
+SELECT * FROM t1;
+a
+1
+11
+21
+33
+SHOW CREATE TABLE t1;
+Table	Create Table
+t1	CREATE TABLE `t1` (
+  `a` int(11) DEFAULT NULL
+) ENGINE=InnoDB DEFAULT CHARSET=latin1
+/*!50100 PARTITION BY RANGE (a)
+(PARTITION p0 VALUES LESS THAN MAXVALUE ENGINE = InnoDB) */
+t1#P#p0.ibd
+t1.frm
+t1.par
+SET DEBUG_SYNC='before_open_in_get_all_tables SIGNAL parked WAIT_FOR open';
+SET DEBUG_SYNC='partition_open_error SIGNAL alter WAIT_FOR finish';
+SELECT * FROM INFORMATION_SCHEMA.PARTITIONS WHERE TABLE_NAME = 't1' AND TABLE_SCHEMA = 'test';
+SET DEBUG_SYNC = 'now WAIT_FOR parked';
+# When waiting for the name lock in get_all_tables in sql_show.cc
+# this will not be concurrent any more, thus the TIMEOUT
+SET DEBUG_SYNC = 'before_rename_partitions SIGNAL open WAIT_FOR alter TIMEOUT 1';
+# Needs to be executed twice, since first is this 'SET DEBUG_SYNC ...'
+SET DEBUG_SYNC = 'before_close_thread_tables SIGNAL finish EXECUTE 2';
+ALTER TABLE t1 REORGANIZE PARTITION p0 INTO
+(PARTITION p0 VALUES LESS THAN (10),
+PARTITION p10 VALUES LESS THAN MAXVALUE);
+Warnings:
+Warning	1639	debug sync point wait timed out
+TABLE_CATALOG	TABLE_SCHEMA	TABLE_NAME	PARTITION_NAME	SUBPARTITION_NAME	PARTITION_ORDINAL_POSITION	SUBPARTITION_ORDINAL_POSITION	PARTITION_METHOD	SUBPARTITION_METHOD	PARTITION_EXPRESSION	SUBPARTITION_EXPRESSION	PARTITION_DESCRIPTION	TABLE_ROWS	AVG_ROW_LENGTH	DATA_LENGTH	MAX_DATA_LENGTH	INDEX_LENGTH	DATA_FREE	CREATE_TIME	UPDATE_TIME	CHECK_TIME	CHECKSUM	PARTITION_COMMENT	NODEGROUP	TABLESPACE_NAME
+NULL	test	t1	p0	NULL	1	NULL	RANGE	NULL	a	NULL	10	1	16384	16384	NULL	0	0	NULL	NULL	NULL	NULL		default	NULL
+NULL	test	t1	p10	NULL	2	NULL	RANGE	NULL	a	NULL	MAXVALUE	3	5461	16384	NULL	0	0	NULL	NULL	NULL	NULL		default	NULL
+t1#P#p0.ibd
+t1#P#p10.ibd
+t1.frm
+t1.par
+SHOW CREATE TABLE t1;
+Table	Create Table
+t1	CREATE TABLE `t1` (
+  `a` int(11) DEFAULT NULL
+) ENGINE=InnoDB DEFAULT CHARSET=latin1
+/*!50100 PARTITION BY RANGE (a)
+(PARTITION p0 VALUES LESS THAN (10) ENGINE = InnoDB,
+ PARTITION p10 VALUES LESS THAN MAXVALUE ENGINE = InnoDB) */
+SELECT * FROM t1;
+a
+1
+11
+21
+33
+drop table t1;
+SET DEBUG_SYNC = 'RESET';

mysql-test/suite/parts/t/partition_debug_sync_innodb-master.opt

+--innodb_file_per_table=1

mysql-test/suite/parts/t/partition_debug_sync_innodb.test

+--source include/have_innodb.inc
+--source include/have_partition.inc
+--source include/have_debug_sync.inc
+
+let $MYSQLD_DATADIR=`SELECT @@datadir`;
+
+--echo #
+--echo # Bug#50561: ALTER PARTITIONS does not have adequate lock, breaks with
+--echo #            concurrent I_S query
+create table t1 (a int)
+engine = innodb
+partition by range (a)
+(partition p0 values less than MAXVALUE);
+insert into t1 values (1), (11), (21), (33);
+SELECT * FROM t1;
+SHOW CREATE TABLE t1;
+--list_files $MYSQLD_DATADIR/test
+
+SET DEBUG_SYNC='before_open_in_get_all_tables SIGNAL parked WAIT_FOR open';
+SET DEBUG_SYNC='partition_open_error SIGNAL alter WAIT_FOR finish';
+send
+SELECT * FROM INFORMATION_SCHEMA.PARTITIONS WHERE TABLE_NAME = 't1' AND TABLE_SCHEMA = 'test';
+
+connect (con1, localhost, root,,);
+SET DEBUG_SYNC = 'now WAIT_FOR parked';
+--echo # When waiting for the name lock in get_all_tables in sql_show.cc
+--echo # this will not be concurrent any more, thus the TIMEOUT
+SET DEBUG_SYNC = 'before_rename_partitions SIGNAL open WAIT_FOR alter TIMEOUT 1';
+--echo # Needs to be executed twice, since first is this 'SET DEBUG_SYNC ...'
+SET DEBUG_SYNC = 'before_close_thread_tables SIGNAL finish EXECUTE 2';
+--error 0,ER_TABLE_EXISTS_ERROR
+ALTER TABLE t1 REORGANIZE PARTITION p0 INTO
+(PARTITION p0 VALUES LESS THAN (10),
+ PARTITION p10 VALUES LESS THAN MAXVALUE);
+
+disconnect con1;
+connection default;
+--reap
+--list_files $MYSQLD_DATADIR/test
+SHOW CREATE TABLE t1;
+SELECT * FROM t1;
+drop table t1;
+--list_files $MYSQLD_DATADIR/test
+SET DEBUG_SYNC = 'RESET';

sql/authors.h

@@ -76,6 +76,7 @@ struct show_table_authors_st show_table_authors[]= {
   { "Eric Herman", "Amsterdam, Netherlands", "Bug fixing - federated" },
   { "Andrey Hristov", "Walldorf, Germany", "Event scheduler (5.1)" },
   { "Alexander (Alexi) Ivanov", "St. Petersburg, Russia", "Replication" },
+  { "Mattias Jonsson", "Uppsala, Sweden", "Partitioning" },
   { "Alexander (Salle) Keremidarski", "Sofia, Bulgaria",
     "Bug fixing" },
   { "Mats Kindahl", "Storvreta, Sweden", "Replication" },

sql/ha_partition.cc

@@ -58,6 +58,8 @@
 
 #include <mysql/plugin.h>
 
+#include "debug_sync.h"
+
 static const char *ha_par_ext= ".par";
 #ifdef NOT_USED
 static int free_share(PARTITION_SHARE * share);
@@ -690,6 +692,7 @@ int ha_partition::rename_partitions(const char *path)
   DBUG_ASSERT(!strcmp(path, get_canonical_filename(m_file[0], path,
                                                    norm_name_buff)));
 
+  DEBUG_SYNC(ha_thd(), "before_rename_partitions");
   if (temp_partitions)
   {
     /*
@@ -2607,6 +2610,7 @@ int ha_partition::open(const char *name, int mode, uint test_if_locked)
   DBUG_RETURN(0);
 
 err_handler:
+  DEBUG_SYNC(ha_thd(), "partition_open_error");
   while (file-- != m_file)
     (*file)->close();
   bitmap_free(&m_bulk_insert_started);

sql/mysql_priv.h

@@ -1750,7 +1750,7 @@ extern pthread_mutex_t LOCK_gdl;
 #define WFRM_PACK_FRM 4
 #define WFRM_KEEP_SHARE 8
 bool mysql_write_frm(ALTER_PARTITION_PARAM_TYPE *lpt, uint flags);
-int abort_and_upgrade_lock(ALTER_PARTITION_PARAM_TYPE *lpt);
+int abort_and_upgrade_lock_and_close_table(ALTER_PARTITION_PARAM_TYPE *lpt);
 void close_open_tables_and_downgrade(ALTER_PARTITION_PARAM_TYPE *lpt);
 void mysql_wait_completed_table(ALTER_PARTITION_PARAM_TYPE *lpt, TABLE *my_table);
 

sql/sql_base.cc

@@ -1244,6 +1244,7 @@ void close_thread_tables(THD *thd)
                           table->s->table_name.str, (long) table));
 #endif
 
+  DEBUG_SYNC(thd, "before_close_thread_tables");
   /*
     We are assuming here that thd->derived_tables contains ONLY derived
     tables for this substatement. i.e. instead of approach which uses
@@ -2502,7 +2503,7 @@ bool check_if_table_exists(THD *thd, TABLE_LIST *table, bool *exists)
                         put in the thread-open-list.
     flags               Bitmap of flags to modify how open works:
                           MYSQL_LOCK_IGNORE_FLUSH - Open table even if
-                          someone has done a flush or namelock on it.
+                          someone has done a flush on it.
                           No version number checking is done.
                           MYSQL_OPEN_TEMPORARY_ONLY - Open only temporary
                           table not the base table or view.
@@ -2792,8 +2793,10 @@ TABLE *open_table(THD *thd, TABLE_LIST *table_list, MEM_ROOT *mem_root,
                  ("Found table '%s.%s' with different refresh version",
                   table_list->db, table_list->table_name));
 
-      if (flags & MYSQL_LOCK_IGNORE_FLUSH)
+      /* Ignore FLUSH, but not name locks! */
+      if (flags & MYSQL_LOCK_IGNORE_FLUSH && !table->open_placeholder)
       {
+        DBUG_ASSERT(table->db_stat);
         /* Force close at once after usage */
         thd->version= table->s->version;
         continue;
@@ -4448,7 +4451,7 @@ thr_lock_type read_lock_type_for_table(THD *thd, TABLE *table)
     counter - number of opened tables will be return using this parameter
     flags   - bitmap of flags to modify how the tables will be open:
               MYSQL_LOCK_IGNORE_FLUSH - open table even if someone has
-              done a flush or namelock on it.
+              done a flush on it.
 
   NOTE
     Unless we are already in prelocked mode, this function will also precache
@@ -5047,7 +5050,7 @@ int open_and_lock_tables_derived(THD *thd, TABLE_LIST *tables, bool derived)
     tables	- list of tables for open
     flags       - bitmap of flags to modify how the tables will be open:
                   MYSQL_LOCK_IGNORE_FLUSH - open table even if someone has
-                  done a flush or namelock on it.
+                  done a flush on it.
 
   RETURN
     FALSE - ok
@@ -8708,7 +8711,7 @@ bool is_equal(const LEX_STRING *a, const LEX_STRING *b)
 
 /*
   SYNOPSIS
-    abort_and_upgrade_lock()
+    abort_and_upgrade_lock_and_close_table()
     lpt                           Parameter passing struct
     All parameters passed through the ALTER_PARTITION_PARAM_TYPE object
   RETURN VALUE
@@ -8717,7 +8720,7 @@ bool is_equal(const LEX_STRING *a, const LEX_STRING *b)
     Remember old lock level (for possible downgrade later on), abort all
     waiting threads and ensure that all keeping locks currently are
     completed such that we own the lock exclusively and no other interaction
-    is ongoing.
+    is ongoing. Close the table and hold the name lock.
 
     thd                           Thread object
     table                         Table object
@@ -8726,17 +8729,26 @@ bool is_equal(const LEX_STRING *a, const LEX_STRING *b)
     old_lock_level                Old lock level
 */
 
-int abort_and_upgrade_lock(ALTER_PARTITION_PARAM_TYPE *lpt)
+int abort_and_upgrade_lock_and_close_table(ALTER_PARTITION_PARAM_TYPE *lpt)
 {
   uint flags= RTFC_WAIT_OTHER_THREAD_FLAG | RTFC_CHECK_KILLED_FLAG;
-  DBUG_ENTER("abort_and_upgrade_locks");
+  const char *db=         lpt->db;
+  const char *table_name= lpt->table_name;
+  THD *thd=               lpt->thd;
+  DBUG_ENTER("abort_and_upgrade_lock_and_close_table");
 
   lpt->old_lock_type= lpt->table->reginfo.lock_type;
+  safe_mutex_assert_not_owner(&LOCK_open);
   VOID(pthread_mutex_lock(&LOCK_open));
   /* If MERGE child, forward lock handling to parent. */
-  mysql_lock_abort(lpt->thd, lpt->table->parent ? lpt->table->parent :
-                   lpt->table, TRUE);
-  VOID(remove_table_from_cache(lpt->thd, lpt->db, lpt->table_name, flags));
+  mysql_lock_abort(thd, lpt->table->parent ? lpt->table->parent : lpt->table,
+                   TRUE);
+  if (remove_table_from_cache(thd, db, table_name, flags))
+  {
+    VOID(pthread_mutex_unlock(&LOCK_open));
+    DBUG_RETURN(1);
+  }
+  close_data_files_and_morph_locks(thd, db, table_name);
   VOID(pthread_mutex_unlock(&LOCK_open));
   DBUG_RETURN(0);
 }

sql/sql_partition.cc

@@ -5955,32 +5955,6 @@ static void alter_partition_lock_handling(ALTER_PARTITION_PARAM_TYPE *lpt)
   }
 }
 
-/*
-  Unlock and close table before renaming and dropping partitions
-  SYNOPSIS
-    alter_close_tables()
-    lpt                        Struct carrying parameters
-  RETURN VALUES
-    0
-*/
-
-static int alter_close_tables(ALTER_PARTITION_PARAM_TYPE *lpt)
-{
-  THD *thd= lpt->thd;
-  const char *db= lpt->db;
-  const char *table_name= lpt->table_name;
-  DBUG_ENTER("alter_close_tables");
-  /*
-    We need to also unlock tables and close all handlers.
-    We set lock to zero to ensure we don't do this twice
-    and we set db_stat to zero to ensure we don't close twice.
-  */
-  pthread_mutex_lock(&LOCK_open);
-  close_data_files_and_morph_locks(thd, db, table_name);
-  pthread_mutex_unlock(&LOCK_open);
-  DBUG_RETURN(0);
-}
-
 
 /*
   Handle errors for ALTER TABLE for partitioning
@@ -6278,9 +6252,7 @@ uint fast_alter_partition_table(THD *thd, TABLE *table,
         write_log_drop_partition(lpt) ||
         ERROR_INJECT_CRASH("crash_drop_partition_3") ||
         (not_completed= FALSE) ||
-        abort_and_upgrade_lock(lpt) || /* Always returns 0 */
-        ERROR_INJECT_CRASH("crash_drop_partition_4") ||
-        alter_close_tables(lpt) ||
+        abort_and_upgrade_lock_and_close_table(lpt) ||
         ERROR_INJECT_CRASH("crash_drop_partition_5") ||
         ((!thd->lex->no_write_to_binlog) &&
          (write_bin_log(thd, FALSE,
@@ -6345,9 +6317,7 @@ uint fast_alter_partition_table(THD *thd, TABLE *table,
         ERROR_INJECT_CRASH("crash_add_partition_2") ||
         mysql_change_partitions(lpt) ||
         ERROR_INJECT_CRASH("crash_add_partition_3") ||
-        abort_and_upgrade_lock(lpt) || /* Always returns 0 */
-        ERROR_INJECT_CRASH("crash_add_partition_4") ||
-        alter_close_tables(lpt) ||
+        abort_and_upgrade_lock_and_close_table(lpt) ||
         ERROR_INJECT_CRASH("crash_add_partition_5") ||
         ((!thd->lex->no_write_to_binlog) &&
          (write_bin_log(thd, FALSE,
@@ -6435,9 +6405,7 @@ uint fast_alter_partition_table(THD *thd, TABLE *table,
         write_log_final_change_partition(lpt) ||
         ERROR_INJECT_CRASH("crash_change_partition_4") ||
         (not_completed= FALSE) ||
-        abort_and_upgrade_lock(lpt) || /* Always returns 0 */
-        ERROR_INJECT_CRASH("crash_change_partition_5") ||
-        alter_close_tables(lpt) ||
+        abort_and_upgrade_lock_and_close_table(lpt) ||
         ERROR_INJECT_CRASH("crash_change_partition_6") ||
         ((!thd->lex->no_write_to_binlog) &&
          (write_bin_log(thd, FALSE,

sql/sql_show.cc

@@ -30,6 +30,7 @@
 #include "event_data_objects.h"
 #endif
 #include <my_dir.h>
+#include "debug_sync.h"
 
 #define STR_OR_NIL(S) ((S) ? (S) : "<nil>")
 
@@ -3447,6 +3448,7 @@ int get_all_tables(THD *thd, TABLE_LIST *tables, COND *cond)
             lex->sql_command= SQLCOM_SHOW_FIELDS;
             show_table_list->i_s_requested_object=
               schema_table->i_s_requested_object;
+            DEBUG_SYNC(thd, "before_open_in_get_all_tables");
             res= open_normal_and_derived_tables(thd, show_table_list,
                                                 MYSQL_LOCK_IGNORE_FLUSH);
             lex->sql_command= save_sql_command;