Simplistic, experimental framework for Stored Procedures (SPs).
Implements creation and dropping of PROCEDUREs, IN, OUT, and INOUT parameters,
single-statement procedures, rudimentary multi-statement (begin-end) prodedures
(when the client can handle it), and local variables.
Missing most of the embedded SQL language, all attributes, FUNCTIONs, error handling,
reparses procedures at each call (no caching), etc, etc.
Certainly buggy too, but procedures can actually be created and called....
2002-12-08 19:59:22 +01:00
|
|
|
/* Copyright (C) 2002 MySQL AB
|
|
|
|
|
|
|
|
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
|
2006-12-23 20:17:15 +01:00
|
|
|
the Free Software Foundation; version 2 of the License.
|
Simplistic, experimental framework for Stored Procedures (SPs).
Implements creation and dropping of PROCEDUREs, IN, OUT, and INOUT parameters,
single-statement procedures, rudimentary multi-statement (begin-end) prodedures
(when the client can handle it), and local variables.
Missing most of the embedded SQL language, all attributes, FUNCTIONs, error handling,
reparses procedures at each call (no caching), etc, etc.
Certainly buggy too, but procedures can actually be created and called....
2002-12-08 19:59:22 +01:00
|
|
|
|
|
|
|
This program is distributed in the hope that it will be useful,
|
|
|
|
but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
|
|
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
|
|
|
GNU General Public License for more details.
|
|
|
|
|
|
|
|
You should have received a copy of the GNU General Public License
|
|
|
|
along with this program; if not, write to the Free Software
|
|
|
|
Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA */
|
|
|
|
|
2005-06-05 16:20:22 +02:00
|
|
|
#include "mysql_priv.h"
|
2005-05-27 12:03:37 +02:00
|
|
|
#ifdef USE_PRAGMA_IMPLEMENTATION
|
Simplistic, experimental framework for Stored Procedures (SPs).
Implements creation and dropping of PROCEDUREs, IN, OUT, and INOUT parameters,
single-statement procedures, rudimentary multi-statement (begin-end) prodedures
(when the client can handle it), and local variables.
Missing most of the embedded SQL language, all attributes, FUNCTIONs, error handling,
reparses procedures at each call (no caching), etc, etc.
Certainly buggy too, but procedures can actually be created and called....
2002-12-08 19:59:22 +01:00
|
|
|
#pragma implementation
|
|
|
|
#endif
|
|
|
|
#include "sp_head.h"
|
2002-12-12 13:14:23 +01:00
|
|
|
#include "sp.h"
|
Simplistic, experimental framework for Stored Procedures (SPs).
Implements creation and dropping of PROCEDUREs, IN, OUT, and INOUT parameters,
single-statement procedures, rudimentary multi-statement (begin-end) prodedures
(when the client can handle it), and local variables.
Missing most of the embedded SQL language, all attributes, FUNCTIONs, error handling,
reparses procedures at each call (no caching), etc, etc.
Certainly buggy too, but procedures can actually be created and called....
2002-12-08 19:59:22 +01:00
|
|
|
#include "sp_pcontext.h"
|
|
|
|
#include "sp_rcontext.h"
|
2005-02-08 20:52:50 +01:00
|
|
|
#include "sp_cache.h"
|
Simplistic, experimental framework for Stored Procedures (SPs).
Implements creation and dropping of PROCEDUREs, IN, OUT, and INOUT parameters,
single-statement procedures, rudimentary multi-statement (begin-end) prodedures
(when the client can handle it), and local variables.
Missing most of the embedded SQL language, all attributes, FUNCTIONs, error handling,
reparses procedures at each call (no caching), etc, etc.
Certainly buggy too, but procedures can actually be created and called....
2002-12-08 19:59:22 +01:00
|
|
|
|
2006-01-05 23:47:49 +01:00
|
|
|
/*
|
|
|
|
Sufficient max length of printed destinations and frame offsets (all uints).
|
|
|
|
*/
|
|
|
|
#define SP_INSTR_UINT_MAXLEN 8
|
|
|
|
#define SP_STMT_PRINT_MAXLEN 40
|
|
|
|
|
|
|
|
|
2006-01-11 00:07:40 +01:00
|
|
|
#include <my_user.h>
|
|
|
|
|
2003-02-26 19:22:29 +01:00
|
|
|
Item_result
|
|
|
|
sp_map_result_type(enum enum_field_types type)
|
|
|
|
{
|
2005-12-07 15:01:17 +01:00
|
|
|
switch (type) {
|
Bug#12976 (stored procedures local variables of type bit)
Before this change, a local variables in stored procedures / stored functions
or triggers, when declared with a type of bit(N), would not evaluate their
value properly.
The problem was that the data was incorrectly typed as a string,
causing for example bit b'1', implemented as a byte 0x01, to be interpreted
as a string starting with the character 0x01. This later would cause
implicit conversions to integers or booleans to fail.
The root cause of this problem was an incorrect translation between field
types, like bit(N), and internal types used when representing values in Item
objects.
Also, before this change, the function HEX() would sometime print extra "0"
characters when invoked with bit(N) values.
With this fix, the type translation (sp_map_result_type, sp_map_item_type)
has been changed so that bit(N) fields are represented with integer values.
A consequence is that, for the function HEX(), when called with a stored
procedure local variable of type bit(N) as argument, HEX() is provided with an
integer instead of a string, and therefore does not print "0" padding.
A test case for Bug 12976 was present in the test suite, and has been updated.
2007-02-07 00:01:22 +01:00
|
|
|
case MYSQL_TYPE_BIT:
|
2003-02-26 19:22:29 +01:00
|
|
|
case MYSQL_TYPE_TINY:
|
|
|
|
case MYSQL_TYPE_SHORT:
|
|
|
|
case MYSQL_TYPE_LONG:
|
|
|
|
case MYSQL_TYPE_LONGLONG:
|
|
|
|
case MYSQL_TYPE_INT24:
|
|
|
|
return INT_RESULT;
|
|
|
|
case MYSQL_TYPE_DECIMAL:
|
2005-02-08 23:50:45 +01:00
|
|
|
case MYSQL_TYPE_NEWDECIMAL:
|
|
|
|
return DECIMAL_RESULT;
|
2003-02-26 19:22:29 +01:00
|
|
|
case MYSQL_TYPE_FLOAT:
|
|
|
|
case MYSQL_TYPE_DOUBLE:
|
|
|
|
return REAL_RESULT;
|
|
|
|
default:
|
|
|
|
return STRING_RESULT;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2005-12-07 15:01:17 +01:00
|
|
|
|
|
|
|
Item::Type
|
|
|
|
sp_map_item_type(enum enum_field_types type)
|
|
|
|
{
|
|
|
|
switch (type) {
|
Bug#12976 (stored procedures local variables of type bit)
Before this change, a local variables in stored procedures / stored functions
or triggers, when declared with a type of bit(N), would not evaluate their
value properly.
The problem was that the data was incorrectly typed as a string,
causing for example bit b'1', implemented as a byte 0x01, to be interpreted
as a string starting with the character 0x01. This later would cause
implicit conversions to integers or booleans to fail.
The root cause of this problem was an incorrect translation between field
types, like bit(N), and internal types used when representing values in Item
objects.
Also, before this change, the function HEX() would sometime print extra "0"
characters when invoked with bit(N) values.
With this fix, the type translation (sp_map_result_type, sp_map_item_type)
has been changed so that bit(N) fields are represented with integer values.
A consequence is that, for the function HEX(), when called with a stored
procedure local variable of type bit(N) as argument, HEX() is provided with an
integer instead of a string, and therefore does not print "0" padding.
A test case for Bug 12976 was present in the test suite, and has been updated.
2007-02-07 00:01:22 +01:00
|
|
|
case MYSQL_TYPE_BIT:
|
2005-12-07 15:01:17 +01:00
|
|
|
case MYSQL_TYPE_TINY:
|
|
|
|
case MYSQL_TYPE_SHORT:
|
|
|
|
case MYSQL_TYPE_LONG:
|
|
|
|
case MYSQL_TYPE_LONGLONG:
|
|
|
|
case MYSQL_TYPE_INT24:
|
|
|
|
return Item::INT_ITEM;
|
|
|
|
case MYSQL_TYPE_DECIMAL:
|
|
|
|
case MYSQL_TYPE_NEWDECIMAL:
|
|
|
|
return Item::DECIMAL_ITEM;
|
|
|
|
case MYSQL_TYPE_FLOAT:
|
|
|
|
case MYSQL_TYPE_DOUBLE:
|
|
|
|
return Item::REAL_ITEM;
|
|
|
|
default:
|
|
|
|
return Item::STRING_ITEM;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
Return a string representation of the Item value.
|
|
|
|
|
2006-03-21 14:35:49 +01:00
|
|
|
NOTE: If the item has a string result type, the string is escaped
|
|
|
|
according to its character set.
|
2005-12-07 15:01:17 +01:00
|
|
|
|
|
|
|
SYNOPSIS
|
|
|
|
item a pointer to the Item
|
|
|
|
str string buffer for representation of the value
|
|
|
|
|
|
|
|
RETURN
|
|
|
|
NULL on error
|
|
|
|
a pointer to valid a valid string on success
|
|
|
|
*/
|
|
|
|
|
|
|
|
static String *
|
2006-11-09 11:27:34 +01:00
|
|
|
sp_get_item_value(THD *thd, Item *item, String *str)
|
2005-12-07 15:01:17 +01:00
|
|
|
{
|
|
|
|
switch (item->result_type()) {
|
|
|
|
case REAL_RESULT:
|
|
|
|
case INT_RESULT:
|
|
|
|
case DECIMAL_RESULT:
|
|
|
|
return item->val_str(str);
|
|
|
|
|
|
|
|
case STRING_RESULT:
|
|
|
|
{
|
|
|
|
String *result= item->val_str(str);
|
|
|
|
|
|
|
|
if (!result)
|
|
|
|
return NULL;
|
|
|
|
|
2006-01-20 13:59:22 +01:00
|
|
|
{
|
|
|
|
char buf_holder[STRING_BUFFER_USUAL_SIZE];
|
|
|
|
String buf(buf_holder, sizeof(buf_holder), result->charset());
|
2006-11-09 11:27:34 +01:00
|
|
|
CHARSET_INFO *cs= thd->variables.character_set_client;
|
2006-01-20 13:59:22 +01:00
|
|
|
|
|
|
|
/* We must reset length of the buffer, because of String specificity. */
|
|
|
|
buf.length(0);
|
|
|
|
|
|
|
|
buf.append('_');
|
|
|
|
buf.append(result->charset()->csname);
|
2006-11-09 11:27:34 +01:00
|
|
|
if (cs->escape_with_backslash_is_dangerous)
|
2006-03-21 14:35:49 +01:00
|
|
|
buf.append(' ');
|
2006-11-09 11:27:34 +01:00
|
|
|
append_query_string(cs, result, &buf);
|
2006-01-20 13:59:22 +01:00
|
|
|
str->copy(buf);
|
|
|
|
|
|
|
|
return str;
|
|
|
|
}
|
2005-12-07 15:01:17 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
case ROW_RESULT:
|
|
|
|
default:
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
|
2004-08-06 13:47:01 +02:00
|
|
|
/*
|
Implement WL#2661 "Prepared Statements: Dynamic SQL in Stored Procedures".
The idea of the patch is to separate statement processing logic,
such as parsing, validation of the parsed tree, execution and cleanup,
from global query processing logic, such as logging, resetting
priorities of a thread, resetting stored procedure cache, resetting
thread count of errors and warnings.
This makes PREPARE and EXECUTE behave similarly to the rest of SQL
statements and allows their use in stored procedures.
This patch contains a change in behaviour:
until recently for each SQL prepared statement command, 2 queries
were written to the general log, e.g.
[Query] prepare stmt from @stmt_text;
[Prepare] select * from t1 <-- contents of @stmt_text
The chagne was necessary to prevent [Prepare] commands from being written
to the general log when executing a stored procedure with Dynamic SQL.
We should consider whether the old behavior is preferrable and probably
restore it.
This patch refixes Bug#7115, Bug#10975 (partially), Bug#10605 (various bugs
in Dynamic SQL reported before it was disabled).
2005-09-03 01:13:18 +02:00
|
|
|
SYNOPSIS
|
|
|
|
sp_get_flags_for_command()
|
|
|
|
|
|
|
|
DESCRIPTION
|
|
|
|
Returns a combination of:
|
|
|
|
* sp_head::MULTI_RESULTS: added if the 'cmd' is a command that might
|
|
|
|
result in multiple result sets being sent back.
|
|
|
|
* sp_head::CONTAINS_DYNAMIC_SQL: added if 'cmd' is one of PREPARE,
|
|
|
|
EXECUTE, DEALLOCATE.
|
|
|
|
*/
|
|
|
|
|
|
|
|
uint
|
|
|
|
sp_get_flags_for_command(LEX *lex)
|
2004-08-06 13:47:01 +02:00
|
|
|
{
|
Implement WL#2661 "Prepared Statements: Dynamic SQL in Stored Procedures".
The idea of the patch is to separate statement processing logic,
such as parsing, validation of the parsed tree, execution and cleanup,
from global query processing logic, such as logging, resetting
priorities of a thread, resetting stored procedure cache, resetting
thread count of errors and warnings.
This makes PREPARE and EXECUTE behave similarly to the rest of SQL
statements and allows their use in stored procedures.
This patch contains a change in behaviour:
until recently for each SQL prepared statement command, 2 queries
were written to the general log, e.g.
[Query] prepare stmt from @stmt_text;
[Prepare] select * from t1 <-- contents of @stmt_text
The chagne was necessary to prevent [Prepare] commands from being written
to the general log when executing a stored procedure with Dynamic SQL.
We should consider whether the old behavior is preferrable and probably
restore it.
This patch refixes Bug#7115, Bug#10975 (partially), Bug#10605 (various bugs
in Dynamic SQL reported before it was disabled).
2005-09-03 01:13:18 +02:00
|
|
|
uint flags;
|
|
|
|
|
|
|
|
switch (lex->sql_command) {
|
|
|
|
case SQLCOM_SELECT:
|
|
|
|
if (lex->result)
|
|
|
|
{
|
|
|
|
flags= 0; /* This is a SELECT with INTO clause */
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
/* fallthrough */
|
2004-08-06 13:47:01 +02:00
|
|
|
case SQLCOM_ANALYZE:
|
2005-12-03 15:02:09 +01:00
|
|
|
case SQLCOM_OPTIMIZE:
|
|
|
|
case SQLCOM_PRELOAD_KEYS:
|
|
|
|
case SQLCOM_ASSIGN_TO_KEYCACHE:
|
2004-08-30 14:52:21 +02:00
|
|
|
case SQLCOM_CHECKSUM:
|
2005-12-03 15:02:09 +01:00
|
|
|
case SQLCOM_CHECK:
|
2004-08-06 13:47:01 +02:00
|
|
|
case SQLCOM_HA_READ:
|
|
|
|
case SQLCOM_SHOW_BINLOGS:
|
|
|
|
case SQLCOM_SHOW_BINLOG_EVENTS:
|
|
|
|
case SQLCOM_SHOW_CHARSETS:
|
|
|
|
case SQLCOM_SHOW_COLLATIONS:
|
|
|
|
case SQLCOM_SHOW_COLUMN_TYPES:
|
|
|
|
case SQLCOM_SHOW_CREATE:
|
|
|
|
case SQLCOM_SHOW_CREATE_DB:
|
|
|
|
case SQLCOM_SHOW_CREATE_FUNC:
|
|
|
|
case SQLCOM_SHOW_CREATE_PROC:
|
|
|
|
case SQLCOM_SHOW_DATABASES:
|
|
|
|
case SQLCOM_SHOW_ERRORS:
|
|
|
|
case SQLCOM_SHOW_FIELDS:
|
|
|
|
case SQLCOM_SHOW_GRANTS:
|
|
|
|
case SQLCOM_SHOW_INNODB_STATUS:
|
|
|
|
case SQLCOM_SHOW_KEYS:
|
|
|
|
case SQLCOM_SHOW_LOGS:
|
|
|
|
case SQLCOM_SHOW_MASTER_STAT:
|
2005-02-25 15:21:05 +01:00
|
|
|
case SQLCOM_SHOW_MUTEX_STATUS:
|
2004-08-06 13:47:01 +02:00
|
|
|
case SQLCOM_SHOW_NEW_MASTER:
|
|
|
|
case SQLCOM_SHOW_OPEN_TABLES:
|
|
|
|
case SQLCOM_SHOW_PRIVILEGES:
|
|
|
|
case SQLCOM_SHOW_PROCESSLIST:
|
|
|
|
case SQLCOM_SHOW_SLAVE_HOSTS:
|
|
|
|
case SQLCOM_SHOW_SLAVE_STAT:
|
|
|
|
case SQLCOM_SHOW_STATUS:
|
|
|
|
case SQLCOM_SHOW_STATUS_FUNC:
|
|
|
|
case SQLCOM_SHOW_STATUS_PROC:
|
|
|
|
case SQLCOM_SHOW_STORAGE_ENGINES:
|
|
|
|
case SQLCOM_SHOW_TABLES:
|
|
|
|
case SQLCOM_SHOW_VARIABLES:
|
|
|
|
case SQLCOM_SHOW_WARNS:
|
2005-11-17 11:11:48 +01:00
|
|
|
case SQLCOM_SHOW_PROC_CODE:
|
|
|
|
case SQLCOM_SHOW_FUNC_CODE:
|
2005-12-02 22:59:45 +01:00
|
|
|
case SQLCOM_REPAIR:
|
|
|
|
case SQLCOM_BACKUP_TABLE:
|
|
|
|
case SQLCOM_RESTORE_TABLE:
|
Implement WL#2661 "Prepared Statements: Dynamic SQL in Stored Procedures".
The idea of the patch is to separate statement processing logic,
such as parsing, validation of the parsed tree, execution and cleanup,
from global query processing logic, such as logging, resetting
priorities of a thread, resetting stored procedure cache, resetting
thread count of errors and warnings.
This makes PREPARE and EXECUTE behave similarly to the rest of SQL
statements and allows their use in stored procedures.
This patch contains a change in behaviour:
until recently for each SQL prepared statement command, 2 queries
were written to the general log, e.g.
[Query] prepare stmt from @stmt_text;
[Prepare] select * from t1 <-- contents of @stmt_text
The chagne was necessary to prevent [Prepare] commands from being written
to the general log when executing a stored procedure with Dynamic SQL.
We should consider whether the old behavior is preferrable and probably
restore it.
This patch refixes Bug#7115, Bug#10975 (partially), Bug#10605 (various bugs
in Dynamic SQL reported before it was disabled).
2005-09-03 01:13:18 +02:00
|
|
|
flags= sp_head::MULTI_RESULTS;
|
|
|
|
break;
|
|
|
|
/*
|
|
|
|
EXECUTE statement may return a result set, but doesn't have to.
|
|
|
|
We can't, however, know it in advance, and therefore must add
|
|
|
|
this statement here. This is ok, as is equivalent to a result-set
|
|
|
|
statement within an IF condition.
|
|
|
|
*/
|
|
|
|
case SQLCOM_EXECUTE:
|
|
|
|
flags= sp_head::MULTI_RESULTS | sp_head::CONTAINS_DYNAMIC_SQL;
|
|
|
|
break;
|
|
|
|
case SQLCOM_PREPARE:
|
|
|
|
case SQLCOM_DEALLOCATE_PREPARE:
|
|
|
|
flags= sp_head::CONTAINS_DYNAMIC_SQL;
|
|
|
|
break;
|
2005-11-16 13:09:06 +01:00
|
|
|
case SQLCOM_CREATE_TABLE:
|
|
|
|
if (lex->create_info.options & HA_LEX_CREATE_TMP_TABLE)
|
|
|
|
flags= 0;
|
|
|
|
else
|
|
|
|
flags= sp_head::HAS_COMMIT_OR_ROLLBACK;
|
|
|
|
break;
|
|
|
|
case SQLCOM_DROP_TABLE:
|
|
|
|
if (lex->drop_temporary)
|
|
|
|
flags= 0;
|
|
|
|
else
|
|
|
|
flags= sp_head::HAS_COMMIT_OR_ROLLBACK;
|
|
|
|
break;
|
2006-08-25 15:51:29 +02:00
|
|
|
case SQLCOM_FLUSH:
|
|
|
|
flags= sp_head::HAS_SQLCOM_FLUSH;
|
|
|
|
break;
|
|
|
|
case SQLCOM_RESET:
|
|
|
|
flags= sp_head::HAS_SQLCOM_RESET;
|
|
|
|
break;
|
2005-11-16 13:09:06 +01:00
|
|
|
case SQLCOM_CREATE_INDEX:
|
|
|
|
case SQLCOM_CREATE_DB:
|
|
|
|
case SQLCOM_CREATE_VIEW:
|
|
|
|
case SQLCOM_CREATE_TRIGGER:
|
|
|
|
case SQLCOM_CREATE_USER:
|
|
|
|
case SQLCOM_ALTER_TABLE:
|
|
|
|
case SQLCOM_BEGIN:
|
|
|
|
case SQLCOM_RENAME_TABLE:
|
|
|
|
case SQLCOM_RENAME_USER:
|
|
|
|
case SQLCOM_DROP_INDEX:
|
|
|
|
case SQLCOM_DROP_DB:
|
|
|
|
case SQLCOM_DROP_USER:
|
|
|
|
case SQLCOM_DROP_VIEW:
|
|
|
|
case SQLCOM_DROP_TRIGGER:
|
|
|
|
case SQLCOM_TRUNCATE:
|
|
|
|
case SQLCOM_COMMIT:
|
|
|
|
case SQLCOM_ROLLBACK:
|
2006-05-23 16:29:58 +02:00
|
|
|
case SQLCOM_LOAD:
|
2005-11-16 13:09:06 +01:00
|
|
|
case SQLCOM_LOAD_MASTER_DATA:
|
|
|
|
case SQLCOM_LOCK_TABLES:
|
|
|
|
case SQLCOM_CREATE_PROCEDURE:
|
|
|
|
case SQLCOM_CREATE_SPFUNCTION:
|
|
|
|
case SQLCOM_ALTER_PROCEDURE:
|
|
|
|
case SQLCOM_ALTER_FUNCTION:
|
|
|
|
case SQLCOM_DROP_PROCEDURE:
|
|
|
|
case SQLCOM_DROP_FUNCTION:
|
|
|
|
flags= sp_head::HAS_COMMIT_OR_ROLLBACK;
|
|
|
|
break;
|
2004-08-06 13:47:01 +02:00
|
|
|
default:
|
Implement WL#2661 "Prepared Statements: Dynamic SQL in Stored Procedures".
The idea of the patch is to separate statement processing logic,
such as parsing, validation of the parsed tree, execution and cleanup,
from global query processing logic, such as logging, resetting
priorities of a thread, resetting stored procedure cache, resetting
thread count of errors and warnings.
This makes PREPARE and EXECUTE behave similarly to the rest of SQL
statements and allows their use in stored procedures.
This patch contains a change in behaviour:
until recently for each SQL prepared statement command, 2 queries
were written to the general log, e.g.
[Query] prepare stmt from @stmt_text;
[Prepare] select * from t1 <-- contents of @stmt_text
The chagne was necessary to prevent [Prepare] commands from being written
to the general log when executing a stored procedure with Dynamic SQL.
We should consider whether the old behavior is preferrable and probably
restore it.
This patch refixes Bug#7115, Bug#10975 (partially), Bug#10605 (various bugs
in Dynamic SQL reported before it was disabled).
2005-09-03 01:13:18 +02:00
|
|
|
flags= 0;
|
|
|
|
break;
|
2004-08-06 13:47:01 +02:00
|
|
|
}
|
Implement WL#2661 "Prepared Statements: Dynamic SQL in Stored Procedures".
The idea of the patch is to separate statement processing logic,
such as parsing, validation of the parsed tree, execution and cleanup,
from global query processing logic, such as logging, resetting
priorities of a thread, resetting stored procedure cache, resetting
thread count of errors and warnings.
This makes PREPARE and EXECUTE behave similarly to the rest of SQL
statements and allows their use in stored procedures.
This patch contains a change in behaviour:
until recently for each SQL prepared statement command, 2 queries
were written to the general log, e.g.
[Query] prepare stmt from @stmt_text;
[Prepare] select * from t1 <-- contents of @stmt_text
The chagne was necessary to prevent [Prepare] commands from being written
to the general log when executing a stored procedure with Dynamic SQL.
We should consider whether the old behavior is preferrable and probably
restore it.
This patch refixes Bug#7115, Bug#10975 (partially), Bug#10605 (various bugs
in Dynamic SQL reported before it was disabled).
2005-09-03 01:13:18 +02:00
|
|
|
return flags;
|
2004-08-06 13:47:01 +02:00
|
|
|
}
|
|
|
|
|
2005-05-09 00:59:10 +02:00
|
|
|
|
|
|
|
/*
|
2005-12-07 15:01:17 +01:00
|
|
|
Prepare an Item for evaluation (call of fix_fields).
|
2005-05-09 00:59:10 +02:00
|
|
|
|
|
|
|
SYNOPSIS
|
|
|
|
sp_prepare_func_item()
|
|
|
|
thd thread handler
|
|
|
|
it_addr pointer on item refernce
|
|
|
|
|
|
|
|
RETURN
|
|
|
|
NULL error
|
|
|
|
prepared item
|
|
|
|
*/
|
|
|
|
|
2005-12-07 15:01:17 +01:00
|
|
|
Item *
|
2005-05-09 00:59:10 +02:00
|
|
|
sp_prepare_func_item(THD* thd, Item **it_addr)
|
|
|
|
{
|
|
|
|
DBUG_ENTER("sp_prepare_func_item");
|
2005-12-07 15:01:17 +01:00
|
|
|
it_addr= (*it_addr)->this_item_addr(thd, it_addr);
|
2005-05-09 00:59:10 +02:00
|
|
|
|
2005-12-07 15:01:17 +01:00
|
|
|
if (!(*it_addr)->fixed &&
|
|
|
|
((*it_addr)->fix_fields(thd, it_addr) ||
|
|
|
|
(*it_addr)->check_cols(1)))
|
2005-05-09 00:59:10 +02:00
|
|
|
{
|
|
|
|
DBUG_PRINT("info", ("fix_fields() failed"));
|
|
|
|
DBUG_RETURN(NULL);
|
|
|
|
}
|
|
|
|
DBUG_RETURN(*it_addr);
|
|
|
|
}
|
|
|
|
|
|
|
|
|
2005-08-18 11:23:54 +02:00
|
|
|
/*
|
2005-12-07 15:01:17 +01:00
|
|
|
Evaluate an expression and store the result in the field.
|
2005-08-18 11:23:54 +02:00
|
|
|
|
|
|
|
SYNOPSIS
|
2005-12-07 15:01:17 +01:00
|
|
|
sp_eval_expr()
|
|
|
|
thd - current thread object
|
|
|
|
expr_item - the root item of the expression
|
|
|
|
result_field - the field to store the result
|
2005-08-18 11:23:54 +02:00
|
|
|
|
|
|
|
RETURN VALUES
|
2005-12-07 15:01:17 +01:00
|
|
|
FALSE on success
|
|
|
|
TRUE on error
|
Simplistic, experimental framework for Stored Procedures (SPs).
Implements creation and dropping of PROCEDUREs, IN, OUT, and INOUT parameters,
single-statement procedures, rudimentary multi-statement (begin-end) prodedures
(when the client can handle it), and local variables.
Missing most of the embedded SQL language, all attributes, FUNCTIONs, error handling,
reparses procedures at each call (no caching), etc, etc.
Certainly buggy too, but procedures can actually be created and called....
2002-12-08 19:59:22 +01:00
|
|
|
*/
|
2005-08-18 11:23:54 +02:00
|
|
|
|
2005-12-07 15:01:17 +01:00
|
|
|
bool
|
2006-05-15 12:01:55 +02:00
|
|
|
sp_eval_expr(THD *thd, Field *result_field, Item **expr_item_ptr)
|
Simplistic, experimental framework for Stored Procedures (SPs).
Implements creation and dropping of PROCEDUREs, IN, OUT, and INOUT parameters,
single-statement procedures, rudimentary multi-statement (begin-end) prodedures
(when the client can handle it), and local variables.
Missing most of the embedded SQL language, all attributes, FUNCTIONs, error handling,
reparses procedures at each call (no caching), etc, etc.
Certainly buggy too, but procedures can actually be created and called....
2002-12-08 19:59:22 +01:00
|
|
|
{
|
2006-05-15 12:01:55 +02:00
|
|
|
Item *expr_item;
|
|
|
|
|
2005-12-07 15:01:17 +01:00
|
|
|
DBUG_ENTER("sp_eval_expr");
|
Simplistic, experimental framework for Stored Procedures (SPs).
Implements creation and dropping of PROCEDUREs, IN, OUT, and INOUT parameters,
single-statement procedures, rudimentary multi-statement (begin-end) prodedures
(when the client can handle it), and local variables.
Missing most of the embedded SQL language, all attributes, FUNCTIONs, error handling,
reparses procedures at each call (no caching), etc, etc.
Certainly buggy too, but procedures can actually be created and called....
2002-12-08 19:59:22 +01:00
|
|
|
|
2006-05-15 12:01:55 +02:00
|
|
|
if (!(expr_item= sp_prepare_func_item(thd, expr_item_ptr)))
|
2005-12-07 15:01:17 +01:00
|
|
|
DBUG_RETURN(TRUE);
|
2005-09-22 00:11:21 +02:00
|
|
|
|
2005-12-07 15:01:17 +01:00
|
|
|
bool err_status= FALSE;
|
2002-12-13 18:25:36 +01:00
|
|
|
|
2005-12-07 15:01:17 +01:00
|
|
|
/*
|
|
|
|
Set THD flags to emit warnings/errors in case of overflow/type errors
|
|
|
|
during saving the item into the field.
|
2005-08-22 00:13:37 +02:00
|
|
|
|
2005-12-07 15:01:17 +01:00
|
|
|
Save original values and restore them after save.
|
|
|
|
*/
|
|
|
|
|
|
|
|
enum_check_fields save_count_cuted_fields= thd->count_cuted_fields;
|
|
|
|
bool save_abort_on_warning= thd->abort_on_warning;
|
2007-03-23 16:12:58 +01:00
|
|
|
bool save_no_trans_update_stmt= thd->no_trans_update.stmt;
|
2005-08-18 11:23:54 +02:00
|
|
|
|
2005-12-07 15:01:17 +01:00
|
|
|
thd->count_cuted_fields= CHECK_FIELD_ERROR_FOR_NULL;
|
|
|
|
thd->abort_on_warning=
|
|
|
|
thd->variables.sql_mode &
|
|
|
|
(MODE_STRICT_TRANS_TABLES | MODE_STRICT_ALL_TABLES);
|
2007-03-23 16:12:58 +01:00
|
|
|
thd->no_trans_update.stmt= FALSE;
|
2005-09-08 18:25:42 +02:00
|
|
|
|
2005-12-07 15:01:17 +01:00
|
|
|
/* Save the value in the field. Convert the value if needed. */
|
2005-10-19 14:54:54 +02:00
|
|
|
|
2005-12-07 15:01:17 +01:00
|
|
|
expr_item->save_in_field(result_field, 0);
|
2005-08-18 11:23:54 +02:00
|
|
|
|
2005-12-07 15:01:17 +01:00
|
|
|
thd->count_cuted_fields= save_count_cuted_fields;
|
|
|
|
thd->abort_on_warning= save_abort_on_warning;
|
2007-03-23 16:12:58 +01:00
|
|
|
thd->no_trans_update.stmt= save_no_trans_update_stmt;
|
2005-08-18 11:23:54 +02:00
|
|
|
|
2005-12-07 15:01:17 +01:00
|
|
|
if (thd->net.report_error)
|
2005-09-22 00:11:21 +02:00
|
|
|
{
|
2005-12-07 15:01:17 +01:00
|
|
|
/* Return error status if something went wrong. */
|
|
|
|
err_status= TRUE;
|
2005-09-22 00:11:21 +02:00
|
|
|
}
|
2005-12-07 15:01:17 +01:00
|
|
|
|
|
|
|
DBUG_RETURN(err_status);
|
Simplistic, experimental framework for Stored Procedures (SPs).
Implements creation and dropping of PROCEDUREs, IN, OUT, and INOUT parameters,
single-statement procedures, rudimentary multi-statement (begin-end) prodedures
(when the client can handle it), and local variables.
Missing most of the embedded SQL language, all attributes, FUNCTIONs, error handling,
reparses procedures at each call (no caching), etc, etc.
Certainly buggy too, but procedures can actually be created and called....
2002-12-08 19:59:22 +01:00
|
|
|
}
|
|
|
|
|
2004-02-17 17:36:53 +01:00
|
|
|
|
|
|
|
/*
|
|
|
|
*
|
|
|
|
* sp_name
|
|
|
|
*
|
|
|
|
*/
|
|
|
|
|
|
|
|
void
|
|
|
|
sp_name::init_qname(THD *thd)
|
|
|
|
{
|
2005-07-09 19:51:59 +02:00
|
|
|
m_sroutines_key.length= m_db.length + m_name.length + 2;
|
|
|
|
if (!(m_sroutines_key.str= thd->alloc(m_sroutines_key.length + 1)))
|
|
|
|
return;
|
|
|
|
m_qname.length= m_sroutines_key.length - 1;
|
|
|
|
m_qname.str= m_sroutines_key.str + 1;
|
2005-10-07 02:37:24 +02:00
|
|
|
sprintf(m_qname.str, "%.*s.%.*s",
|
2004-02-17 17:36:53 +01:00
|
|
|
m_db.length, (m_db.length ? m_db.str : ""),
|
|
|
|
m_name.length, m_name.str);
|
|
|
|
}
|
|
|
|
|
2004-03-11 17:18:59 +01:00
|
|
|
|
2006-01-11 15:11:05 +01:00
|
|
|
/*
|
2006-01-19 16:13:04 +01:00
|
|
|
Check that the name 'ident' is ok. It's assumed to be an 'ident'
|
|
|
|
from the parser, so we only have to check length and trailing spaces.
|
|
|
|
The former is a standard requirement (and 'show status' assumes a
|
|
|
|
non-empty name), the latter is a mysql:ism as trailing spaces are
|
|
|
|
removed by get_field().
|
|
|
|
|
|
|
|
RETURN
|
|
|
|
TRUE - bad name
|
|
|
|
FALSE - name is ok
|
|
|
|
*/
|
2006-01-11 15:11:05 +01:00
|
|
|
|
|
|
|
bool
|
2006-01-19 16:13:04 +01:00
|
|
|
check_routine_name(LEX_STRING ident)
|
2006-01-11 15:11:05 +01:00
|
|
|
{
|
|
|
|
return (!ident.str || !ident.str[0] || ident.str[ident.length-1] == ' ');
|
|
|
|
}
|
2004-03-11 17:18:59 +01:00
|
|
|
|
2004-02-17 17:36:53 +01:00
|
|
|
/* ------------------------------------------------------------------ */
|
|
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
*
|
|
|
|
* sp_head
|
|
|
|
*
|
|
|
|
*/
|
|
|
|
|
2003-06-29 18:15:17 +02:00
|
|
|
void *
|
|
|
|
sp_head::operator new(size_t size)
|
|
|
|
{
|
|
|
|
DBUG_ENTER("sp_head::operator new");
|
|
|
|
MEM_ROOT own_root;
|
|
|
|
sp_head *sp;
|
|
|
|
|
|
|
|
init_alloc_root(&own_root, MEM_ROOT_BLOCK_SIZE, MEM_ROOT_PREALLOC);
|
2004-11-09 02:58:44 +01:00
|
|
|
sp= (sp_head *) alloc_root(&own_root, size);
|
|
|
|
sp->main_mem_root= own_root;
|
2004-05-20 01:02:49 +02:00
|
|
|
DBUG_PRINT("info", ("mem_root 0x%lx", (ulong) &sp->mem_root));
|
2003-06-29 18:15:17 +02:00
|
|
|
DBUG_RETURN(sp);
|
|
|
|
}
|
|
|
|
|
|
|
|
void
|
|
|
|
sp_head::operator delete(void *ptr, size_t size)
|
|
|
|
{
|
|
|
|
DBUG_ENTER("sp_head::operator delete");
|
|
|
|
MEM_ROOT own_root;
|
2004-11-09 02:58:44 +01:00
|
|
|
sp_head *sp= (sp_head *) ptr;
|
2003-06-29 18:15:17 +02:00
|
|
|
|
2004-11-09 02:58:44 +01:00
|
|
|
/* Make a copy of main_mem_root as free_root will free the sp */
|
|
|
|
own_root= sp->main_mem_root;
|
2004-05-20 01:02:49 +02:00
|
|
|
DBUG_PRINT("info", ("mem_root 0x%lx moved to 0x%lx",
|
|
|
|
(ulong) &sp->mem_root, (ulong) &own_root));
|
2003-06-29 18:15:17 +02:00
|
|
|
free_root(&own_root, MYF(0));
|
|
|
|
|
|
|
|
DBUG_VOID_RETURN;
|
|
|
|
}
|
|
|
|
|
2004-05-20 01:02:49 +02:00
|
|
|
|
2003-07-01 17:19:48 +02:00
|
|
|
sp_head::sp_head()
|
2005-06-22 09:59:13 +02:00
|
|
|
:Query_arena(&main_mem_root, INITIALIZED_FOR_SP),
|
2005-12-07 15:01:17 +01:00
|
|
|
m_flags(0), m_recursion_level(0), m_next_cached_sp(0),
|
2006-01-16 15:37:25 +01:00
|
|
|
m_cont_level(0)
|
Simplistic, experimental framework for Stored Procedures (SPs).
Implements creation and dropping of PROCEDUREs, IN, OUT, and INOUT parameters,
single-statement procedures, rudimentary multi-statement (begin-end) prodedures
(when the client can handle it), and local variables.
Missing most of the embedded SQL language, all attributes, FUNCTIONs, error handling,
reparses procedures at each call (no caching), etc, etc.
Certainly buggy too, but procedures can actually be created and called....
2002-12-08 19:59:22 +01:00
|
|
|
{
|
2006-11-30 17:25:05 +01:00
|
|
|
m_first_instance= this;
|
|
|
|
m_first_free_instance= this;
|
|
|
|
m_last_cached_sp= this;
|
|
|
|
|
2005-12-07 15:01:17 +01:00
|
|
|
m_return_field_def.charset = NULL;
|
|
|
|
|
2005-02-08 20:52:50 +01:00
|
|
|
extern byte *
|
|
|
|
sp_table_key(const byte *ptr, uint *plen, my_bool first);
|
2003-04-03 20:00:52 +02:00
|
|
|
DBUG_ENTER("sp_head::sp_head");
|
2003-07-01 17:19:48 +02:00
|
|
|
|
|
|
|
m_backpatch.empty();
|
2005-11-04 15:37:39 +01:00
|
|
|
m_cont_backpatch.empty();
|
2003-07-01 17:19:48 +02:00
|
|
|
m_lex.empty();
|
2005-02-08 20:52:50 +01:00
|
|
|
hash_init(&m_sptabs, system_charset_info, 0, 0, 0, sp_table_key, 0, 0);
|
2005-07-09 19:51:59 +02:00
|
|
|
hash_init(&m_sroutines, system_charset_info, 0, 0, 0, sp_sroutine_key, 0, 0);
|
2003-07-01 17:19:48 +02:00
|
|
|
DBUG_VOID_RETURN;
|
|
|
|
}
|
|
|
|
|
2004-05-20 01:02:49 +02:00
|
|
|
|
2003-07-01 17:19:48 +02:00
|
|
|
void
|
2003-12-12 14:05:29 +01:00
|
|
|
sp_head::init(LEX *lex)
|
2003-07-01 17:19:48 +02:00
|
|
|
{
|
|
|
|
DBUG_ENTER("sp_head::init");
|
2003-12-12 14:05:29 +01:00
|
|
|
|
Bug#26503 (Illegal SQL exception handler code causes the server to crash)
Before this fix, the parser would accept illegal code in SQL exceptions
handlers, that later causes the runtime to crash when executing the code,
due to memory violations in the exception handler stack.
The root cause of the problem is instructions within an exception handler
that jumps to code located outside of the handler. This is illegal according
to the SQL 2003 standard, since labels located outside the handler are not
supposed to be visible (they are "out of scope"), so any instruction that
jumps to these labels, like ITERATE or LEAVE, should not parse.
The section of the standard that is relevant for this is :
SQL:2003 SQL/PSM (ISO/IEC 9075-4:2003)
section 13.1 <compound statement>,
syntax rule 4
<quote>
The scope of the <beginning label> is CS excluding every <SQL schema
statement> contained in CS and excluding every
<local handler declaration list> contained in CS. <beginning label> shall
not be equivalent to any other <beginning label>s within that scope.
</quote>
With this fix, the C++ class sp_pcontext, which represent the "parsing
context" tree (a.k.a symbol table) of a stored procedure, has been changed
as follows:
- constructors have been cleaned up, so that only building a root node for
the tree is public; building nodes inside a tree is not public.
- a new member, m_label_scope, indicates if a given syntactic context
belongs to a DECLARE HANDLER block,
- label resolution, in the method find_label(), has been changed to
implement the restriction of scope regarding labels used in a compound
statement.
The actions in the parser, when parsing the body of a SQL exception handler,
have been changed as follows:
- the implementation of an exception handler (DECLARE HANDLER) now creates
explicitly a new sp_pcontext, to isolate the code inside the handler from
the containing compound statement context.
- registering exception handlers as a result occurs in the parent context,
see the rule sp_hcond_element
- the code in sp_hcond_list has been cleaned up, to avoid code duplication
In addition, the flags IN_SIMPLE_CASE and IN_HANDLER, declared in sp_head.h
have been removed, since they are unused and broken by design (as seen with
Bug 19194 (Right recursion in parser for CASE causes excessive stack usage,
limitation), representing a stack in a single flag is not possible.
Tests in sp-error have been added to show that illegal constructs are now
rejected.
Tests in sp have been added for code coverage, to show that ITERATE or LEAVE
statements are legal when jumping to a label in scope, inside the body of
an exception handler.
2007-03-14 19:02:32 +01:00
|
|
|
lex->spcont= m_pcont= new sp_pcontext();
|
2005-12-07 15:01:17 +01:00
|
|
|
|
2004-11-24 10:24:02 +01:00
|
|
|
/*
|
|
|
|
Altough trg_table_fields list is used only in triggers we init for all
|
|
|
|
types of stored procedures to simplify reset_lex()/restore_lex() code.
|
|
|
|
*/
|
|
|
|
lex->trg_table_fields.empty();
|
2003-12-12 14:05:29 +01:00
|
|
|
my_init_dynamic_array(&m_instr, sizeof(sp_instr *), 16, 8);
|
A fix and a test case for Bug#26141 mixing table types in trigger
causes full table lock on innodb table.
Also fixes Bug#28502 Triggers that update another innodb table
will block on X lock unnecessarily (duplciate).
Code review fixes.
Both bugs' synopses are misleading: InnoDB table is
not X locked. The statements, however, cannot proceed concurrently,
but this happens due to lock conflicts for tables used in triggers,
not for the InnoDB table.
If a user had an InnoDB table, and two triggers, AFTER UPDATE and
AFTER INSERT, competing for different resources (e.g. two distinct
MyISAM tables), then these two triggers would not be able to execute
concurrently. Moreover, INSERTS/UPDATES of the InnoDB table would
not be able to run concurrently.
The problem had other side-effects (see respective bug reports).
This behavior was a consequence of a shortcoming of the pre-locking
algorithm, which would not distinguish between different DML operations
(e.g. INSERT and DELETE) and pre-lock all the tables
that are used by any trigger defined on the subject table.
The idea of the fix is to extend the pre-locking algorithm to keep track,
for each table, what DML operation it is used for and not
load triggers that are known to never be fired.
2007-07-12 20:26:41 +02:00
|
|
|
|
|
|
|
m_param_begin= NULL;
|
|
|
|
m_param_end= NULL;
|
|
|
|
|
|
|
|
m_body_begin= NULL ;
|
|
|
|
|
|
|
|
m_qname.str= NULL;
|
|
|
|
m_qname.length= 0;
|
|
|
|
|
|
|
|
m_db.str= NULL;
|
|
|
|
m_db.length= 0;
|
|
|
|
|
|
|
|
m_name.str= NULL;
|
|
|
|
m_name.length= 0;
|
|
|
|
|
|
|
|
m_params.str= NULL;
|
|
|
|
m_params.length= 0;
|
|
|
|
|
|
|
|
m_body.str= NULL;
|
|
|
|
m_body.length= 0;
|
|
|
|
|
|
|
|
m_defstr.str= NULL;
|
|
|
|
m_defstr.length= 0;
|
|
|
|
|
|
|
|
m_sroutines_key.str= NULL;
|
|
|
|
m_sroutines_key.length= 0;
|
|
|
|
|
2005-12-07 15:01:17 +01:00
|
|
|
m_return_field_def.charset= NULL;
|
A fix and a test case for Bug#26141 mixing table types in trigger
causes full table lock on innodb table.
Also fixes Bug#28502 Triggers that update another innodb table
will block on X lock unnecessarily (duplciate).
Code review fixes.
Both bugs' synopses are misleading: InnoDB table is
not X locked. The statements, however, cannot proceed concurrently,
but this happens due to lock conflicts for tables used in triggers,
not for the InnoDB table.
If a user had an InnoDB table, and two triggers, AFTER UPDATE and
AFTER INSERT, competing for different resources (e.g. two distinct
MyISAM tables), then these two triggers would not be able to execute
concurrently. Moreover, INSERTS/UPDATES of the InnoDB table would
not be able to run concurrently.
The problem had other side-effects (see respective bug reports).
This behavior was a consequence of a shortcoming of the pre-locking
algorithm, which would not distinguish between different DML operations
(e.g. INSERT and DELETE) and pre-lock all the tables
that are used by any trigger defined on the subject table.
The idea of the fix is to extend the pre-locking algorithm to keep track,
for each table, what DML operation it is used for and not
load triggers that are known to never be fired.
2007-07-12 20:26:41 +02:00
|
|
|
|
2003-12-12 14:05:29 +01:00
|
|
|
DBUG_VOID_RETURN;
|
|
|
|
}
|
|
|
|
|
2006-07-27 15:57:43 +02:00
|
|
|
|
2003-12-12 14:05:29 +01:00
|
|
|
void
|
2006-07-27 15:57:43 +02:00
|
|
|
sp_head::init_sp_name(THD *thd, sp_name *spname)
|
|
|
|
{
|
|
|
|
DBUG_ENTER("sp_head::init_sp_name");
|
|
|
|
|
|
|
|
/* Must be initialized in the parser. */
|
|
|
|
|
|
|
|
DBUG_ASSERT(spname && spname->m_db.str && spname->m_db.length);
|
|
|
|
|
|
|
|
/* We have to copy strings to get them into the right memroot. */
|
|
|
|
|
|
|
|
m_db.length= spname->m_db.length;
|
|
|
|
m_db.str= strmake_root(thd->mem_root, spname->m_db.str, spname->m_db.length);
|
|
|
|
|
|
|
|
m_name.length= spname->m_name.length;
|
|
|
|
m_name.str= strmake_root(thd->mem_root, spname->m_name.str,
|
|
|
|
spname->m_name.length);
|
|
|
|
|
|
|
|
if (spname->m_qname.length == 0)
|
|
|
|
spname->init_qname(thd);
|
|
|
|
|
A fix and a test case for Bug#26141 mixing table types in trigger
causes full table lock on innodb table.
Also fixes Bug#28502 Triggers that update another innodb table
will block on X lock unnecessarily (duplciate).
Code review fixes.
Both bugs' synopses are misleading: InnoDB table is
not X locked. The statements, however, cannot proceed concurrently,
but this happens due to lock conflicts for tables used in triggers,
not for the InnoDB table.
If a user had an InnoDB table, and two triggers, AFTER UPDATE and
AFTER INSERT, competing for different resources (e.g. two distinct
MyISAM tables), then these two triggers would not be able to execute
concurrently. Moreover, INSERTS/UPDATES of the InnoDB table would
not be able to run concurrently.
The problem had other side-effects (see respective bug reports).
This behavior was a consequence of a shortcoming of the pre-locking
algorithm, which would not distinguish between different DML operations
(e.g. INSERT and DELETE) and pre-lock all the tables
that are used by any trigger defined on the subject table.
The idea of the fix is to extend the pre-locking algorithm to keep track,
for each table, what DML operation it is used for and not
load triggers that are known to never be fired.
2007-07-12 20:26:41 +02:00
|
|
|
m_sroutines_key.length= spname->m_sroutines_key.length;
|
|
|
|
m_sroutines_key.str= memdup_root(thd->mem_root,
|
|
|
|
spname->m_sroutines_key.str,
|
|
|
|
spname->m_sroutines_key.length + 1);
|
|
|
|
m_sroutines_key.str[0]= static_cast<char>(m_type);
|
|
|
|
|
|
|
|
m_qname.length= m_sroutines_key.length - 1;
|
|
|
|
m_qname.str= m_sroutines_key.str + 1;
|
2007-05-11 22:03:50 +02:00
|
|
|
|
|
|
|
DBUG_VOID_RETURN;
|
2006-07-27 15:57:43 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
void
|
|
|
|
sp_head::init_strings(THD *thd, LEX *lex)
|
2003-12-12 14:05:29 +01:00
|
|
|
{
|
|
|
|
DBUG_ENTER("sp_head::init_strings");
|
Bug#25411 (trigger code truncated), PART I
The issue found with bug 25411 is due to the function skip_rear_comments()
which damages the source code while implementing a work around.
The root cause of the problem is in the lexical analyser, which does not
process special comments properly.
For special comments like :
[1] aaa /*!50000 bbb */ ccc
since 5.0 is a version older that the current code, the parser is in lining
the content of the special comment, so that the query to process is
[2] aaa bbb ccc
However, the text of the query captured when processing a stored procedure,
stored function or trigger (or event in 5.1), can be after rebuilding it:
[3] aaa bbb */ ccc
which is wrong.
To fix bug 25411 properly, the lexical analyser needs to return [2] when
in lining special comments.
In order to implement this, some preliminary cleanup is required in the code,
which is implemented by this patch.
Before this change, the structure named LEX (or st_lex) contains attributes
that belong to lexical analysis, as well as attributes that represents the
abstract syntax tree (AST) of a statement.
Creating a new LEX structure for each statements (which makes sense for the
AST part) also re-initialized the lexical analysis phase each time, which
is conceptually wrong.
With this patch, the previous st_lex structure has been split in two:
- st_lex represents the Abstract Syntax Tree for a statement. The name "lex"
has not been changed to avoid a bigger impact in the code base.
- class lex_input_stream represents the internal state of the lexical
analyser, which by definition should *not* be reinitialized when parsing
multiple statements from the same input stream.
This change is a pre-requisite for bug 25411, since the implementation of
lex_input_stream will later improve to deal properly with special comments,
and this processing can not be done with the current implementation of
sp_head::reset_lex and sp_head::restore_lex, which interfere with the lexer.
This change set alone does not fix bug 25411.
2007-04-24 17:24:21 +02:00
|
|
|
const char *endp; /* Used to trim the end */
|
2003-12-15 13:24:16 +01:00
|
|
|
/* During parsing, we must use thd->mem_root */
|
2004-11-09 02:58:44 +01:00
|
|
|
MEM_ROOT *root= thd->mem_root;
|
Bug#25411 (trigger code truncated), PART I
The issue found with bug 25411 is due to the function skip_rear_comments()
which damages the source code while implementing a work around.
The root cause of the problem is in the lexical analyser, which does not
process special comments properly.
For special comments like :
[1] aaa /*!50000 bbb */ ccc
since 5.0 is a version older that the current code, the parser is in lining
the content of the special comment, so that the query to process is
[2] aaa bbb ccc
However, the text of the query captured when processing a stored procedure,
stored function or trigger (or event in 5.1), can be after rebuilding it:
[3] aaa bbb */ ccc
which is wrong.
To fix bug 25411 properly, the lexical analyser needs to return [2] when
in lining special comments.
In order to implement this, some preliminary cleanup is required in the code,
which is implemented by this patch.
Before this change, the structure named LEX (or st_lex) contains attributes
that belong to lexical analysis, as well as attributes that represents the
abstract syntax tree (AST) of a statement.
Creating a new LEX structure for each statements (which makes sense for the
AST part) also re-initialized the lexical analysis phase each time, which
is conceptually wrong.
With this patch, the previous st_lex structure has been split in two:
- st_lex represents the Abstract Syntax Tree for a statement. The name "lex"
has not been changed to avoid a bigger impact in the code base.
- class lex_input_stream represents the internal state of the lexical
analyser, which by definition should *not* be reinitialized when parsing
multiple statements from the same input stream.
This change is a pre-requisite for bug 25411, since the implementation of
lex_input_stream will later improve to deal properly with special comments,
and this processing can not be done with the current implementation of
sp_head::reset_lex and sp_head::restore_lex, which interfere with the lexer.
This change set alone does not fix bug 25411.
2007-04-24 17:24:21 +02:00
|
|
|
Lex_input_stream *lip=thd->m_lip;
|
Simplistic, experimental framework for Stored Procedures (SPs).
Implements creation and dropping of PROCEDUREs, IN, OUT, and INOUT parameters,
single-statement procedures, rudimentary multi-statement (begin-end) prodedures
(when the client can handle it), and local variables.
Missing most of the embedded SQL language, all attributes, FUNCTIONs, error handling,
reparses procedures at each call (no caching), etc, etc.
Certainly buggy too, but procedures can actually be created and called....
2002-12-08 19:59:22 +01:00
|
|
|
|
2004-09-07 14:29:46 +02:00
|
|
|
if (m_param_begin && m_param_end)
|
2004-02-17 17:36:53 +01:00
|
|
|
{
|
2004-09-07 14:29:46 +02:00
|
|
|
m_params.length= m_param_end - m_param_begin;
|
|
|
|
m_params.str= strmake_root(root,
|
|
|
|
(char *)m_param_begin, m_params.length);
|
2004-02-17 17:36:53 +01:00
|
|
|
}
|
2004-09-09 17:52:10 +02:00
|
|
|
|
2005-11-11 11:10:52 +01:00
|
|
|
/* If ptr has overrun end_of_query then end_of_query is the end */
|
Bug#25411 (trigger code truncated), PART I
The issue found with bug 25411 is due to the function skip_rear_comments()
which damages the source code while implementing a work around.
The root cause of the problem is in the lexical analyser, which does not
process special comments properly.
For special comments like :
[1] aaa /*!50000 bbb */ ccc
since 5.0 is a version older that the current code, the parser is in lining
the content of the special comment, so that the query to process is
[2] aaa bbb ccc
However, the text of the query captured when processing a stored procedure,
stored function or trigger (or event in 5.1), can be after rebuilding it:
[3] aaa bbb */ ccc
which is wrong.
To fix bug 25411 properly, the lexical analyser needs to return [2] when
in lining special comments.
In order to implement this, some preliminary cleanup is required in the code,
which is implemented by this patch.
Before this change, the structure named LEX (or st_lex) contains attributes
that belong to lexical analysis, as well as attributes that represents the
abstract syntax tree (AST) of a statement.
Creating a new LEX structure for each statements (which makes sense for the
AST part) also re-initialized the lexical analysis phase each time, which
is conceptually wrong.
With this patch, the previous st_lex structure has been split in two:
- st_lex represents the Abstract Syntax Tree for a statement. The name "lex"
has not been changed to avoid a bigger impact in the code base.
- class lex_input_stream represents the internal state of the lexical
analyser, which by definition should *not* be reinitialized when parsing
multiple statements from the same input stream.
This change is a pre-requisite for bug 25411, since the implementation of
lex_input_stream will later improve to deal properly with special comments,
and this processing can not be done with the current implementation of
sp_head::reset_lex and sp_head::restore_lex, which interfere with the lexer.
This change set alone does not fix bug 25411.
2007-04-24 17:24:21 +02:00
|
|
|
endp= (lip->ptr > lip->end_of_query ? lip->end_of_query : lip->ptr);
|
2005-11-11 11:10:52 +01:00
|
|
|
/*
|
|
|
|
Trim "garbage" at the end. This is sometimes needed with the
|
|
|
|
"/ * ! VERSION... * /" wrapper in dump files.
|
|
|
|
*/
|
Bug#27876 (SF with cyrillic variable name fails during execution (regression))
The root cause of this bug is related to the function skip_rear_comments,
in sql_lex.cc
Recent code changes in skip_rear_comments changed the prototype from
"const uchar*" to "const char*", which had an unforseen impact on this test:
(endp[-1] < ' ')
With unsigned characters, this code filters bytes of value [0x00 - 0x20]
With *signed* characters, this also filters bytes of value [0x80 - 0xFF].
This caused the regression reported, considering cyrillic characters in the
parameter name to be whitespace, and truncated.
Note that the regression is present both in 5.0 and 5.1.
With this fix:
- [0x80 - 0xFF] bytes are no longer considered whitespace.
This alone fixes the regression.
In addition, filtering [0x00 - 0x20] was found bogus and abusive,
so that the code now filters uses my_isspace when looking for whitespace.
Note that this fix is only addressing the regression affecting UTF-8
in general, but does not address a more fundamental problem with
skip_rear_comments: parsing a string *backwards*, starting at end[-1],
is not safe with multi-bytes characters, so that end[-1] can confuse the
last byte of a multi-byte characters with a characters to filter out.
The only known impact of this remaining issue affects objects that have to
meet all the conditions below:
- the object is a FUNCTION / PROCEDURE / TRIGGER / EVENT / VIEW
- the body consist of only *1* instruction, and does *not* contain a
BEGIN-END block
- the instruction ends, lexically, with <ident> <whitespace>* ';'?
For example, "select <ident>;" or "return <ident>;"
- The last character of <ident> is a multi-byte character
- the last byte of this character is ';' '*', '/' or whitespace
In this case, the body of the object will be truncated after parsing,
and stored in an invalid format.
This last issue has not been fixed in this patch, since the real fix
will be implemented by Bug 25411 (trigger code truncated), which is caused
by the very same code.
The real problem is that the function skip_rear_comments is only a
work-around, and should be removed entirely: see the proposed patch for
bug 25411 for details.
2007-05-25 22:36:01 +02:00
|
|
|
endp= skip_rear_comments(thd->charset(), (char*) m_body_begin, (char*) endp);
|
2005-11-11 11:10:52 +01:00
|
|
|
|
|
|
|
m_body.length= endp - m_body_begin;
|
Bug#25411 (trigger code truncated), PART I
The issue found with bug 25411 is due to the function skip_rear_comments()
which damages the source code while implementing a work around.
The root cause of the problem is in the lexical analyser, which does not
process special comments properly.
For special comments like :
[1] aaa /*!50000 bbb */ ccc
since 5.0 is a version older that the current code, the parser is in lining
the content of the special comment, so that the query to process is
[2] aaa bbb ccc
However, the text of the query captured when processing a stored procedure,
stored function or trigger (or event in 5.1), can be after rebuilding it:
[3] aaa bbb */ ccc
which is wrong.
To fix bug 25411 properly, the lexical analyser needs to return [2] when
in lining special comments.
In order to implement this, some preliminary cleanup is required in the code,
which is implemented by this patch.
Before this change, the structure named LEX (or st_lex) contains attributes
that belong to lexical analysis, as well as attributes that represents the
abstract syntax tree (AST) of a statement.
Creating a new LEX structure for each statements (which makes sense for the
AST part) also re-initialized the lexical analysis phase each time, which
is conceptually wrong.
With this patch, the previous st_lex structure has been split in two:
- st_lex represents the Abstract Syntax Tree for a statement. The name "lex"
has not been changed to avoid a bigger impact in the code base.
- class lex_input_stream represents the internal state of the lexical
analyser, which by definition should *not* be reinitialized when parsing
multiple statements from the same input stream.
This change is a pre-requisite for bug 25411, since the implementation of
lex_input_stream will later improve to deal properly with special comments,
and this processing can not be done with the current implementation of
sp_head::reset_lex and sp_head::restore_lex, which interfere with the lexer.
This change set alone does not fix bug 25411.
2007-04-24 17:24:21 +02:00
|
|
|
m_body.str= strmake_root(root, m_body_begin, m_body.length);
|
|
|
|
m_defstr.length= endp - lip->buf;
|
|
|
|
m_defstr.str= strmake_root(root, lip->buf, m_defstr.length);
|
2003-04-03 20:00:52 +02:00
|
|
|
DBUG_VOID_RETURN;
|
Simplistic, experimental framework for Stored Procedures (SPs).
Implements creation and dropping of PROCEDUREs, IN, OUT, and INOUT parameters,
single-statement procedures, rudimentary multi-statement (begin-end) prodedures
(when the client can handle it), and local variables.
Missing most of the embedded SQL language, all attributes, FUNCTIONs, error handling,
reparses procedures at each call (no caching), etc, etc.
Certainly buggy too, but procedures can actually be created and called....
2002-12-08 19:59:22 +01:00
|
|
|
}
|
|
|
|
|
2005-12-07 15:01:17 +01:00
|
|
|
|
|
|
|
static TYPELIB *
|
|
|
|
create_typelib(MEM_ROOT *mem_root, create_field *field_def, List<String> *src)
|
2005-03-04 22:14:35 +01:00
|
|
|
{
|
|
|
|
TYPELIB *result= NULL;
|
2005-12-07 15:01:17 +01:00
|
|
|
CHARSET_INFO *cs= field_def->charset;
|
|
|
|
DBUG_ENTER("create_typelib");
|
2005-03-04 22:14:35 +01:00
|
|
|
if (src->elements)
|
|
|
|
{
|
|
|
|
result= (TYPELIB*) alloc_root(mem_root, sizeof(TYPELIB));
|
|
|
|
result->count= src->elements;
|
|
|
|
result->name= "";
|
|
|
|
if (!(result->type_names=(const char **)
|
2005-04-20 19:08:42 +02:00
|
|
|
alloc_root(mem_root,(sizeof(char *)+sizeof(int))*(result->count+1))))
|
2006-06-14 20:18:42 +02:00
|
|
|
DBUG_RETURN(0);
|
2005-04-20 19:08:42 +02:00
|
|
|
result->type_lengths= (unsigned int *)(result->type_names + result->count+1);
|
2005-03-04 22:14:35 +01:00
|
|
|
List_iterator<String> it(*src);
|
2005-05-06 10:39:30 +02:00
|
|
|
String conv;
|
|
|
|
for (uint i=0; i < result->count; i++)
|
2005-04-19 10:09:25 +02:00
|
|
|
{
|
2005-05-06 10:39:30 +02:00
|
|
|
uint32 dummy;
|
|
|
|
uint length;
|
|
|
|
String *tmp= it++;
|
|
|
|
|
2005-04-19 10:09:25 +02:00
|
|
|
if (String::needs_conversion(tmp->length(), tmp->charset(),
|
|
|
|
cs, &dummy))
|
|
|
|
{
|
|
|
|
uint cnv_errs;
|
|
|
|
conv.copy(tmp->ptr(), tmp->length(), tmp->charset(), cs, &cnv_errs);
|
2005-05-06 10:39:30 +02:00
|
|
|
|
|
|
|
length= conv.length();
|
|
|
|
result->type_names[i]= (char*) strmake_root(mem_root, conv.ptr(),
|
|
|
|
length);
|
2005-04-19 10:09:25 +02:00
|
|
|
}
|
2005-05-06 10:39:30 +02:00
|
|
|
else
|
|
|
|
{
|
|
|
|
length= tmp->length();
|
|
|
|
result->type_names[i]= strmake_root(mem_root, tmp->ptr(), length);
|
2005-04-20 19:08:42 +02:00
|
|
|
}
|
2005-04-19 10:09:25 +02:00
|
|
|
|
|
|
|
// Strip trailing spaces.
|
2005-05-06 10:39:30 +02:00
|
|
|
length= cs->cset->lengthsp(cs, result->type_names[i], length);
|
|
|
|
result->type_lengths[i]= length;
|
|
|
|
((uchar *)result->type_names[i])[length]= '\0';
|
2005-04-19 10:09:25 +02:00
|
|
|
}
|
2005-03-04 22:14:35 +01:00
|
|
|
result->type_names[result->count]= 0;
|
2005-04-20 19:08:42 +02:00
|
|
|
result->type_lengths[result->count]= 0;
|
2005-03-04 22:14:35 +01:00
|
|
|
}
|
2006-06-14 20:18:42 +02:00
|
|
|
DBUG_RETURN(result);
|
2005-03-04 22:14:35 +01:00
|
|
|
}
|
|
|
|
|
2005-12-07 15:01:17 +01:00
|
|
|
|
Simplistic, experimental framework for Stored Procedures (SPs).
Implements creation and dropping of PROCEDUREs, IN, OUT, and INOUT parameters,
single-statement procedures, rudimentary multi-statement (begin-end) prodedures
(when the client can handle it), and local variables.
Missing most of the embedded SQL language, all attributes, FUNCTIONs, error handling,
reparses procedures at each call (no caching), etc, etc.
Certainly buggy too, but procedures can actually be created and called....
2002-12-08 19:59:22 +01:00
|
|
|
int
|
|
|
|
sp_head::create(THD *thd)
|
|
|
|
{
|
2003-02-12 16:17:03 +01:00
|
|
|
DBUG_ENTER("sp_head::create");
|
2003-02-21 17:37:05 +01:00
|
|
|
int ret;
|
|
|
|
|
2003-12-12 14:05:29 +01:00
|
|
|
DBUG_PRINT("info", ("type: %d name: %s params: %s body: %s",
|
|
|
|
m_type, m_name.str, m_params.str, m_body.str));
|
2004-08-02 18:05:31 +02:00
|
|
|
|
2003-02-21 17:37:05 +01:00
|
|
|
if (m_type == TYPE_ENUM_FUNCTION)
|
2003-12-12 14:05:29 +01:00
|
|
|
ret= sp_create_function(thd, this);
|
2003-02-21 17:37:05 +01:00
|
|
|
else
|
2003-12-12 14:05:29 +01:00
|
|
|
ret= sp_create_procedure(thd, this);
|
Simplistic, experimental framework for Stored Procedures (SPs).
Implements creation and dropping of PROCEDUREs, IN, OUT, and INOUT parameters,
single-statement procedures, rudimentary multi-statement (begin-end) prodedures
(when the client can handle it), and local variables.
Missing most of the embedded SQL language, all attributes, FUNCTIONs, error handling,
reparses procedures at each call (no caching), etc, etc.
Certainly buggy too, but procedures can actually be created and called....
2002-12-08 19:59:22 +01:00
|
|
|
|
2003-02-21 17:37:05 +01:00
|
|
|
DBUG_RETURN(ret);
|
Simplistic, experimental framework for Stored Procedures (SPs).
Implements creation and dropping of PROCEDUREs, IN, OUT, and INOUT parameters,
single-statement procedures, rudimentary multi-statement (begin-end) prodedures
(when the client can handle it), and local variables.
Missing most of the embedded SQL language, all attributes, FUNCTIONs, error handling,
reparses procedures at each call (no caching), etc, etc.
Certainly buggy too, but procedures can actually be created and called....
2002-12-08 19:59:22 +01:00
|
|
|
}
|
|
|
|
|
2003-06-29 18:15:17 +02:00
|
|
|
sp_head::~sp_head()
|
|
|
|
{
|
|
|
|
destroy();
|
2005-11-23 00:11:19 +01:00
|
|
|
delete m_next_cached_sp;
|
2003-06-29 18:15:17 +02:00
|
|
|
if (m_thd)
|
|
|
|
restore_thd_mem_root(m_thd);
|
|
|
|
}
|
|
|
|
|
2003-04-02 20:42:28 +02:00
|
|
|
void
|
|
|
|
sp_head::destroy()
|
|
|
|
{
|
2003-06-29 18:15:17 +02:00
|
|
|
sp_instr *i;
|
|
|
|
LEX *lex;
|
2005-02-05 15:05:46 +01:00
|
|
|
DBUG_ENTER("sp_head::destroy");
|
|
|
|
DBUG_PRINT("info", ("name: %s", m_name.str));
|
2003-06-29 18:15:17 +02:00
|
|
|
|
|
|
|
for (uint ip = 0 ; (i = get_instr(ip)) ; ip++)
|
|
|
|
delete i;
|
2003-04-02 20:42:28 +02:00
|
|
|
delete_dynamic(&m_instr);
|
|
|
|
m_pcont->destroy();
|
2005-06-23 18:22:08 +02:00
|
|
|
free_items();
|
2005-03-05 14:31:58 +01:00
|
|
|
|
|
|
|
/*
|
|
|
|
If we have non-empty LEX stack then we just came out of parser with
|
|
|
|
error. Now we should delete all auxilary LEXes and restore original
|
|
|
|
THD::lex (In this case sp_head::restore_thd_mem_root() was not called
|
|
|
|
too, so m_thd points to the current thread context).
|
|
|
|
It is safe to not update LEX::ptr because further query string parsing
|
|
|
|
and execution will be stopped anyway.
|
|
|
|
*/
|
|
|
|
DBUG_ASSERT(m_lex.is_empty() || m_thd);
|
2003-06-29 18:15:17 +02:00
|
|
|
while ((lex= (LEX *)m_lex.pop()))
|
|
|
|
{
|
2006-05-04 14:30:38 +02:00
|
|
|
lex_end(m_thd->lex);
|
2005-03-05 14:31:58 +01:00
|
|
|
delete m_thd->lex;
|
|
|
|
m_thd->lex= lex;
|
2003-06-29 18:15:17 +02:00
|
|
|
}
|
2005-03-05 14:31:58 +01:00
|
|
|
|
2005-03-04 14:35:28 +01:00
|
|
|
hash_free(&m_sptabs);
|
2005-07-09 19:51:59 +02:00
|
|
|
hash_free(&m_sroutines);
|
2003-04-03 20:00:52 +02:00
|
|
|
DBUG_VOID_RETURN;
|
2003-04-02 20:42:28 +02:00
|
|
|
}
|
2003-02-26 19:22:29 +01:00
|
|
|
|
2005-03-04 22:14:35 +01:00
|
|
|
|
2005-06-10 16:14:01 +02:00
|
|
|
/*
|
2005-08-22 00:13:37 +02:00
|
|
|
This is only used for result fields from functions (both during
|
|
|
|
fix_length_and_dec() and evaluation).
|
|
|
|
*/
|
|
|
|
|
2005-03-04 22:14:35 +01:00
|
|
|
Field *
|
2005-12-07 15:01:17 +01:00
|
|
|
sp_head::create_result_field(uint field_max_length, const char *field_name,
|
|
|
|
TABLE *table)
|
2005-03-04 22:14:35 +01:00
|
|
|
{
|
2005-12-07 15:01:17 +01:00
|
|
|
uint field_length;
|
2005-03-04 22:14:35 +01:00
|
|
|
Field *field;
|
2005-12-07 15:01:17 +01:00
|
|
|
|
|
|
|
DBUG_ENTER("sp_head::create_result_field");
|
|
|
|
|
|
|
|
field_length= !m_return_field_def.length ?
|
|
|
|
field_max_length : m_return_field_def.length;
|
|
|
|
|
|
|
|
field= ::make_field((char*) 0, /* field ptr */
|
|
|
|
field_length, /* field [max] length */
|
|
|
|
(uchar*) "", /* null ptr */
|
|
|
|
0, /* null bit */
|
|
|
|
m_return_field_def.pack_flag,
|
|
|
|
m_return_field_def.sql_type,
|
|
|
|
m_return_field_def.charset,
|
|
|
|
m_return_field_def.geom_type,
|
|
|
|
Field::NONE, /* unreg check */
|
|
|
|
m_return_field_def.interval,
|
|
|
|
field_name ? field_name : (const char *) m_name.str,
|
|
|
|
table);
|
|
|
|
|
2005-03-04 22:14:35 +01:00
|
|
|
DBUG_RETURN(field);
|
|
|
|
}
|
|
|
|
|
2005-08-25 15:34:34 +02:00
|
|
|
|
|
|
|
int cmp_splocal_locations(Item_splocal * const *a, Item_splocal * const *b)
|
|
|
|
{
|
|
|
|
return (int)((*a)->pos_in_query - (*b)->pos_in_query);
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
StoredRoutinesBinlogging
|
|
|
|
Top-down overview:
|
|
|
|
|
|
|
|
1. Statements
|
|
|
|
|
|
|
|
Statements that have is_update_query(stmt) == TRUE are written into the
|
|
|
|
binary log verbatim.
|
|
|
|
Examples:
|
|
|
|
UPDATE tbl SET tbl.x = spfunc_w_side_effects()
|
|
|
|
UPDATE tbl SET tbl.x=1 WHERE spfunc_w_side_effect_that_returns_false(tbl.y)
|
|
|
|
|
|
|
|
Statements that have is_update_query(stmt) == FALSE (e.g. SELECTs) are not
|
|
|
|
written into binary log. Instead we catch function calls the statement
|
|
|
|
makes and write it into binary log separately (see #3).
|
2006-02-18 17:26:30 +01:00
|
|
|
|
2005-08-25 15:34:34 +02:00
|
|
|
2. PROCEDURE calls
|
|
|
|
|
|
|
|
CALL statements are not written into binary log. Instead
|
|
|
|
* Any FUNCTION invocation (in SET, IF, WHILE, OPEN CURSOR and other SP
|
|
|
|
instructions) is written into binlog separately.
|
|
|
|
|
|
|
|
* Each statement executed in SP is binlogged separately, according to rules
|
|
|
|
in #1, with the exception that we modify query string: we replace uses
|
|
|
|
of SP local variables with NAME_CONST('spvar_name', <spvar-value>) calls.
|
|
|
|
This substitution is done in subst_spvars().
|
|
|
|
|
|
|
|
3. FUNCTION calls
|
|
|
|
|
|
|
|
In sp_head::execute_function(), we check
|
|
|
|
* If this function invocation is done from a statement that is written
|
|
|
|
into the binary log.
|
|
|
|
* If there were any attempts to write events to the binary log during
|
2005-09-07 17:39:47 +02:00
|
|
|
function execution (grep for start_union_events and stop_union_events)
|
|
|
|
|
2005-08-25 15:34:34 +02:00
|
|
|
If the answers are No and Yes, we write the function call into the binary
|
2006-02-18 17:26:30 +01:00
|
|
|
log as "SELECT spfunc(<param1value>, <param2value>, ...)".
|
2005-09-07 17:39:47 +02:00
|
|
|
|
|
|
|
|
|
|
|
4. Miscellaneous issues.
|
|
|
|
|
|
|
|
4.1 User variables.
|
|
|
|
|
|
|
|
When we call mysql_bin_log.write() for an SP statement, thd->user_var_events
|
|
|
|
must hold set<{var_name, value}> pairs for all user variables used during
|
|
|
|
the statement execution.
|
|
|
|
This set is produced by tracking user variable reads during statement
|
|
|
|
execution.
|
|
|
|
|
|
|
|
Fo SPs, this has the following implications:
|
|
|
|
1) thd->user_var_events may contain events from several SP statements and
|
|
|
|
needs to be valid after exection of these statements was finished. In
|
|
|
|
order to achieve that, we
|
|
|
|
* Allocate user_var_events array elements on appropriate mem_root (grep
|
|
|
|
for user_var_events_alloc).
|
|
|
|
* Use is_query_in_union() to determine if user_var_event is created.
|
|
|
|
|
|
|
|
2) We need to empty thd->user_var_events after we have wrote a function
|
|
|
|
call. This is currently done by making
|
|
|
|
reset_dynamic(&thd->user_var_events);
|
|
|
|
calls in several different places. (TODO cosider moving this into
|
|
|
|
mysql_bin_log.write() function)
|
2005-08-25 15:34:34 +02:00
|
|
|
*/
|
|
|
|
|
|
|
|
|
|
|
|
/*
|
2007-07-28 13:01:29 +02:00
|
|
|
Replace thd->query{_length} with a string that one can write to the binlog
|
|
|
|
or the query cache.
|
2005-08-25 15:34:34 +02:00
|
|
|
|
|
|
|
SYNOPSIS
|
|
|
|
subst_spvars()
|
|
|
|
thd Current thread.
|
|
|
|
instr Instruction (we look for Item_splocal instances in
|
|
|
|
instr->free_list)
|
|
|
|
query_str Original query string
|
|
|
|
|
|
|
|
DESCRIPTION
|
|
|
|
|
|
|
|
The binlog-suitable string is produced by replacing references to SP local
|
2007-07-28 13:01:29 +02:00
|
|
|
variables with NAME_CONST('sp_var_name', value) calls. To make this string
|
|
|
|
suitable for the query cache this function allocates some additional space
|
|
|
|
for the query cache flags.
|
2005-08-25 15:34:34 +02:00
|
|
|
|
|
|
|
RETURN
|
2005-12-07 15:01:17 +01:00
|
|
|
FALSE on success
|
|
|
|
thd->query{_length} either has been appropriately replaced or there
|
|
|
|
is no need for replacements.
|
|
|
|
TRUE out of memory error.
|
2005-08-25 15:34:34 +02:00
|
|
|
*/
|
|
|
|
|
2005-12-07 15:01:17 +01:00
|
|
|
static bool
|
|
|
|
subst_spvars(THD *thd, sp_instr *instr, LEX_STRING *query_str)
|
2005-08-25 15:34:34 +02:00
|
|
|
{
|
|
|
|
DBUG_ENTER("subst_spvars");
|
|
|
|
|
2007-07-28 13:01:29 +02:00
|
|
|
Dynamic_array<Item_splocal*> sp_vars_uses;
|
|
|
|
char *pbuf, *cur, buffer[512];
|
|
|
|
String qbuf(buffer, sizeof(buffer), &my_charset_bin);
|
|
|
|
int prev_pos, res, buf_len;
|
|
|
|
|
|
|
|
/* Find all instances of Item_splocal used in this statement */
|
|
|
|
for (Item *item= instr->free_list; item; item= item->next)
|
|
|
|
{
|
|
|
|
if (item->is_splocal())
|
2005-08-25 15:34:34 +02:00
|
|
|
{
|
2007-07-28 13:01:29 +02:00
|
|
|
Item_splocal *item_spl= (Item_splocal*)item;
|
|
|
|
if (item_spl->pos_in_query)
|
|
|
|
sp_vars_uses.append(item_spl);
|
2005-08-25 15:34:34 +02:00
|
|
|
}
|
2007-07-28 13:01:29 +02:00
|
|
|
}
|
|
|
|
if (!sp_vars_uses.elements())
|
|
|
|
DBUG_RETURN(FALSE);
|
|
|
|
|
|
|
|
/* Sort SP var refs by their occurences in the query */
|
|
|
|
sp_vars_uses.sort(cmp_splocal_locations);
|
|
|
|
|
|
|
|
/*
|
|
|
|
Construct a statement string where SP local var refs are replaced
|
|
|
|
with "NAME_CONST(name, value)"
|
|
|
|
*/
|
|
|
|
qbuf.length(0);
|
|
|
|
cur= query_str->str;
|
|
|
|
prev_pos= res= 0;
|
|
|
|
for (Item_splocal **splocal= sp_vars_uses.front();
|
|
|
|
splocal < sp_vars_uses.back(); splocal++)
|
|
|
|
{
|
|
|
|
Item *val;
|
|
|
|
|
|
|
|
char str_buffer[STRING_BUFFER_USUAL_SIZE];
|
|
|
|
String str_value_holder(str_buffer, sizeof(str_buffer),
|
|
|
|
&my_charset_latin1);
|
|
|
|
String *str_value;
|
|
|
|
|
|
|
|
/* append the text between sp ref occurences */
|
|
|
|
res|= qbuf.append(cur + prev_pos, (*splocal)->pos_in_query - prev_pos);
|
|
|
|
prev_pos= (*splocal)->pos_in_query + (*splocal)->m_name.length;
|
|
|
|
|
|
|
|
/* append the spvar substitute */
|
|
|
|
res|= qbuf.append(STRING_WITH_LEN(" NAME_CONST('"));
|
|
|
|
res|= qbuf.append((*splocal)->m_name.str, (*splocal)->m_name.length);
|
|
|
|
res|= qbuf.append(STRING_WITH_LEN("',"));
|
|
|
|
res|= (*splocal)->fix_fields(thd, (Item **) splocal);
|
2005-12-07 15:01:17 +01:00
|
|
|
|
2005-08-25 15:34:34 +02:00
|
|
|
if (res)
|
2007-07-28 13:01:29 +02:00
|
|
|
break;
|
2005-08-25 15:34:34 +02:00
|
|
|
|
2007-07-28 13:01:29 +02:00
|
|
|
val= (*splocal)->this_item();
|
|
|
|
DBUG_PRINT("info", ("print %p", val));
|
|
|
|
str_value= sp_get_item_value(thd, val, &str_value_holder);
|
|
|
|
if (str_value)
|
|
|
|
res|= qbuf.append(*str_value);
|
|
|
|
else
|
|
|
|
res|= qbuf.append(STRING_WITH_LEN("NULL"));
|
|
|
|
res|= qbuf.append(')');
|
|
|
|
if (res)
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
res|= qbuf.append(cur + prev_pos, query_str->length - prev_pos);
|
|
|
|
if (res)
|
|
|
|
DBUG_RETURN(TRUE);
|
2005-08-25 15:34:34 +02:00
|
|
|
|
2007-07-28 13:01:29 +02:00
|
|
|
/*
|
|
|
|
Allocate additional space at the end of the new query string for the
|
|
|
|
query_cache_send_result_to_client function.
|
|
|
|
*/
|
|
|
|
buf_len= qbuf.length() + thd->db_length + 1 + QUERY_CACHE_FLAGS_SIZE + 1;
|
|
|
|
if ((pbuf= alloc_root(thd->mem_root, buf_len)))
|
|
|
|
{
|
|
|
|
memcpy(pbuf, qbuf.ptr(), qbuf.length());
|
|
|
|
pbuf[qbuf.length()]= 0;
|
2005-08-25 15:34:34 +02:00
|
|
|
}
|
2007-07-28 13:01:29 +02:00
|
|
|
else
|
|
|
|
DBUG_RETURN(TRUE);
|
|
|
|
|
|
|
|
thd->query= pbuf;
|
|
|
|
thd->query_length= qbuf.length();
|
|
|
|
|
2005-12-07 15:01:17 +01:00
|
|
|
DBUG_RETURN(FALSE);
|
2005-08-25 15:34:34 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
|
2005-11-23 00:11:19 +01:00
|
|
|
/*
|
|
|
|
Return appropriate error about recursion limit reaching
|
|
|
|
|
|
|
|
SYNOPSIS
|
|
|
|
sp_head::recursion_level_error()
|
2006-01-05 23:47:49 +01:00
|
|
|
thd Thread handle
|
2005-11-23 00:11:19 +01:00
|
|
|
|
|
|
|
NOTE
|
|
|
|
For functions and triggers we return error about prohibited recursion.
|
|
|
|
For stored procedures we return about reaching recursion limit.
|
|
|
|
*/
|
|
|
|
|
2006-01-05 23:47:49 +01:00
|
|
|
void sp_head::recursion_level_error(THD *thd)
|
2005-11-23 00:11:19 +01:00
|
|
|
{
|
|
|
|
if (m_type == TYPE_ENUM_PROCEDURE)
|
|
|
|
{
|
|
|
|
my_error(ER_SP_RECURSION_LIMIT, MYF(0),
|
|
|
|
thd->variables.max_sp_recursion_depth,
|
2005-12-12 12:57:35 +01:00
|
|
|
m_name.str);
|
2005-11-23 00:11:19 +01:00
|
|
|
}
|
|
|
|
else
|
|
|
|
my_error(ER_SP_NO_RECURSION, MYF(0));
|
|
|
|
}
|
|
|
|
|
|
|
|
|
2005-08-25 15:34:34 +02:00
|
|
|
/*
|
|
|
|
Execute the routine. The main instruction jump loop is there
|
|
|
|
Assume the parameters already set.
|
|
|
|
|
|
|
|
RETURN
|
2005-12-07 15:01:17 +01:00
|
|
|
FALSE on success
|
|
|
|
TRUE on error
|
2005-08-25 15:34:34 +02:00
|
|
|
|
|
|
|
*/
|
|
|
|
|
2005-12-07 15:01:17 +01:00
|
|
|
bool
|
|
|
|
sp_head::execute(THD *thd)
|
Simplistic, experimental framework for Stored Procedures (SPs).
Implements creation and dropping of PROCEDUREs, IN, OUT, and INOUT parameters,
single-statement procedures, rudimentary multi-statement (begin-end) prodedures
(when the client can handle it), and local variables.
Missing most of the embedded SQL language, all attributes, FUNCTIONs, error handling,
reparses procedures at each call (no caching), etc, etc.
Certainly buggy too, but procedures can actually be created and called....
2002-12-08 19:59:22 +01:00
|
|
|
{
|
2003-02-12 16:17:03 +01:00
|
|
|
DBUG_ENTER("sp_head::execute");
|
2006-09-27 16:21:29 +02:00
|
|
|
char old_db_buf[NAME_LEN+1];
|
A fix and a test case for
Bug#19022 "Memory bug when switching db during trigger execution"
Bug#17199 "Problem when view calls function from another database."
Bug#18444 "Fully qualified stored function names don't work correctly in
SELECT statements"
Documentation note: this patch introduces a change in behaviour of prepared
statements.
This patch adds a few new invariants with regard to how THD::db should
be used. These invariants should be preserved in future:
- one should never refer to THD::db by pointer and always make a deep copy
(strmake, strdup)
- one should never compare two databases by pointer, but use strncmp or
my_strncasecmp
- TABLE_LIST object table->db should be always initialized in the parser or
by creator of the object.
For prepared statements it means that if the current database is changed
after a statement is prepared, the database that was current at prepare
remains active. This also means that you can not prepare a statement that
implicitly refers to the current database if the latter is not set.
This is not documented, and therefore needs documentation. This is NOT a
change in behavior for almost all SQL statements except:
- ALTER TABLE t1 RENAME t2
- OPTIMIZE TABLE t1
- ANALYZE TABLE t1
- TRUNCATE TABLE t1 --
until this patch t1 or t2 could be evaluated at the first execution of
prepared statement.
CURRENT_DATABASE() still works OK and is evaluated at every execution
of prepared statement.
Note, that in stored routines this is not an issue as the default
database is the database of the stored procedure and "use" statement
is prohibited in stored routines.
This patch makes obsolete the use of check_db_used (it was never used in the
old code too) and all other places that check for table->db and assign it
from THD::db if it's NULL, except the parser.
How this patch was created: THD::{db,db_length} were replaced with a
LEX_STRING, THD::db. All the places that refer to THD::{db,db_length} were
manually checked and:
- if the place uses thd->db by pointer, it was fixed to make a deep copy
- if a place compared two db pointers, it was fixed to compare them by value
(via strcmp/my_strcasecmp, whatever was approproate)
Then this intermediate patch was used to write a smaller patch that does the
same thing but without a rename.
TODO in 5.1:
- remove check_db_used
- deploy THD::set_db in mysql_change_db
See also comments to individual files.
2006-06-26 22:47:52 +02:00
|
|
|
LEX_STRING old_db= { old_db_buf, sizeof(old_db_buf) };
|
2004-06-08 18:41:18 +02:00
|
|
|
bool dbchanged;
|
2004-07-22 16:46:59 +02:00
|
|
|
sp_rcontext *ctx;
|
2005-12-07 15:01:17 +01:00
|
|
|
bool err_status= FALSE;
|
2003-02-26 19:22:29 +01:00
|
|
|
uint ip= 0;
|
2005-07-28 21:39:11 +02:00
|
|
|
ulong save_sql_mode;
|
2006-04-19 12:27:59 +02:00
|
|
|
bool save_abort_on_warning;
|
2005-06-15 19:58:35 +02:00
|
|
|
Query_arena *old_arena;
|
2005-08-18 11:23:54 +02:00
|
|
|
/* per-instruction arena */
|
|
|
|
MEM_ROOT execute_mem_root;
|
|
|
|
Query_arena execute_arena(&execute_mem_root, INITIALIZED_FOR_SP),
|
2005-09-02 15:21:19 +02:00
|
|
|
backup_arena;
|
2005-03-04 15:46:45 +01:00
|
|
|
query_id_t old_query_id;
|
2005-03-04 14:35:28 +01:00
|
|
|
TABLE *old_derived_tables;
|
|
|
|
LEX *old_lex;
|
|
|
|
Item_change_list old_change_list;
|
|
|
|
String old_packet;
|
2004-07-22 16:46:59 +02:00
|
|
|
|
2005-05-26 20:36:14 +02:00
|
|
|
/* Use some extra margin for possible SP recursion and functions */
|
2005-11-23 00:11:19 +01:00
|
|
|
if (check_stack_overrun(thd, 8 * STACK_MIN_SIZE, (char*)&old_packet))
|
2005-12-07 15:01:17 +01:00
|
|
|
DBUG_RETURN(TRUE);
|
2004-03-11 17:18:59 +01:00
|
|
|
|
2005-11-23 00:11:19 +01:00
|
|
|
/* init per-instruction memroot */
|
|
|
|
init_alloc_root(&execute_mem_root, MEM_ROOT_BLOCK_SIZE, 0);
|
|
|
|
|
|
|
|
DBUG_ASSERT(!(m_flags & IS_INVOKED));
|
Implement WL#2661 "Prepared Statements: Dynamic SQL in Stored Procedures".
The idea of the patch is to separate statement processing logic,
such as parsing, validation of the parsed tree, execution and cleanup,
from global query processing logic, such as logging, resetting
priorities of a thread, resetting stored procedure cache, resetting
thread count of errors and warnings.
This makes PREPARE and EXECUTE behave similarly to the rest of SQL
statements and allows their use in stored procedures.
This patch contains a change in behaviour:
until recently for each SQL prepared statement command, 2 queries
were written to the general log, e.g.
[Query] prepare stmt from @stmt_text;
[Prepare] select * from t1 <-- contents of @stmt_text
The chagne was necessary to prevent [Prepare] commands from being written
to the general log when executing a stored procedure with Dynamic SQL.
We should consider whether the old behavior is preferrable and probably
restore it.
This patch refixes Bug#7115, Bug#10975 (partially), Bug#10605 (various bugs
in Dynamic SQL reported before it was disabled).
2005-09-03 01:13:18 +02:00
|
|
|
m_flags|= IS_INVOKED;
|
2005-11-23 00:11:19 +01:00
|
|
|
m_first_instance->m_first_free_instance= m_next_cached_sp;
|
2006-01-05 23:47:49 +01:00
|
|
|
if (m_next_cached_sp)
|
|
|
|
{
|
|
|
|
DBUG_PRINT("info",
|
|
|
|
("first free for 0x%lx ++: 0x%lx->0x%lx level: %lu flags %x",
|
|
|
|
(ulong)m_first_instance, (ulong) this,
|
|
|
|
(ulong) m_next_cached_sp,
|
|
|
|
m_next_cached_sp->m_recursion_level,
|
|
|
|
m_next_cached_sp->m_flags));
|
|
|
|
}
|
2005-11-23 00:11:19 +01:00
|
|
|
/*
|
|
|
|
Check that if there are not any instances after this one then
|
|
|
|
pointer to the last instance points on this instance or if there are
|
|
|
|
some instances after this one then recursion level of next instance
|
|
|
|
greater then recursion level of current instance on 1
|
|
|
|
*/
|
|
|
|
DBUG_ASSERT((m_next_cached_sp == 0 &&
|
|
|
|
m_first_instance->m_last_cached_sp == this) ||
|
|
|
|
(m_recursion_level + 1 == m_next_cached_sp->m_recursion_level));
|
2005-07-01 11:01:46 +02:00
|
|
|
|
2007-04-06 18:21:30 +02:00
|
|
|
/*
|
|
|
|
NOTE: The SQL Standard does not specify the context that should be
|
|
|
|
preserved for stored routines. However, at SAP/Walldorf meeting it was
|
|
|
|
decided that current database should be preserved.
|
|
|
|
*/
|
|
|
|
|
2004-09-08 14:23:14 +02:00
|
|
|
if (m_db.length &&
|
A fix and a test case for
Bug#19022 "Memory bug when switching db during trigger execution"
Bug#17199 "Problem when view calls function from another database."
Bug#18444 "Fully qualified stored function names don't work correctly in
SELECT statements"
Documentation note: this patch introduces a change in behaviour of prepared
statements.
This patch adds a few new invariants with regard to how THD::db should
be used. These invariants should be preserved in future:
- one should never refer to THD::db by pointer and always make a deep copy
(strmake, strdup)
- one should never compare two databases by pointer, but use strncmp or
my_strncasecmp
- TABLE_LIST object table->db should be always initialized in the parser or
by creator of the object.
For prepared statements it means that if the current database is changed
after a statement is prepared, the database that was current at prepare
remains active. This also means that you can not prepare a statement that
implicitly refers to the current database if the latter is not set.
This is not documented, and therefore needs documentation. This is NOT a
change in behavior for almost all SQL statements except:
- ALTER TABLE t1 RENAME t2
- OPTIMIZE TABLE t1
- ANALYZE TABLE t1
- TRUNCATE TABLE t1 --
until this patch t1 or t2 could be evaluated at the first execution of
prepared statement.
CURRENT_DATABASE() still works OK and is evaluated at every execution
of prepared statement.
Note, that in stored routines this is not an issue as the default
database is the database of the stored procedure and "use" statement
is prohibited in stored routines.
This patch makes obsolete the use of check_db_used (it was never used in the
old code too) and all other places that check for table->db and assign it
from THD::db if it's NULL, except the parser.
How this patch was created: THD::{db,db_length} were replaced with a
LEX_STRING, THD::db. All the places that refer to THD::{db,db_length} were
manually checked and:
- if the place uses thd->db by pointer, it was fixed to make a deep copy
- if a place compared two db pointers, it was fixed to compare them by value
(via strcmp/my_strcasecmp, whatever was approproate)
Then this intermediate patch was used to write a smaller patch that does the
same thing but without a rename.
TODO in 5.1:
- remove check_db_used
- deploy THD::set_db in mysql_change_db
See also comments to individual files.
2006-06-26 22:47:52 +02:00
|
|
|
(err_status= sp_use_new_db(thd, m_db, &old_db, 0, &dbchanged)))
|
2004-03-11 17:18:59 +01:00
|
|
|
goto done;
|
2003-03-26 15:02:48 +01:00
|
|
|
|
2004-07-22 16:46:59 +02:00
|
|
|
if ((ctx= thd->spcont))
|
2003-09-16 14:26:08 +02:00
|
|
|
ctx->clear_handler();
|
2004-01-09 09:36:37 +01:00
|
|
|
thd->query_error= 0;
|
2005-09-02 15:21:19 +02:00
|
|
|
old_arena= thd->stmt_arena;
|
2004-07-22 16:46:59 +02:00
|
|
|
|
2005-03-04 14:35:28 +01:00
|
|
|
/*
|
|
|
|
We have to save/restore this info when we are changing call level to
|
|
|
|
be able properly do close_thread_tables() in instructions.
|
|
|
|
*/
|
|
|
|
old_query_id= thd->query_id;
|
|
|
|
old_derived_tables= thd->derived_tables;
|
|
|
|
thd->derived_tables= 0;
|
2005-07-28 21:39:11 +02:00
|
|
|
save_sql_mode= thd->variables.sql_mode;
|
|
|
|
thd->variables.sql_mode= m_sql_mode;
|
2006-04-19 12:27:59 +02:00
|
|
|
save_abort_on_warning= thd->abort_on_warning;
|
2006-10-19 20:39:51 +02:00
|
|
|
thd->abort_on_warning= 0;
|
2006-04-19 12:27:59 +02:00
|
|
|
|
2005-03-04 14:35:28 +01:00
|
|
|
/*
|
|
|
|
It is also more efficient to save/restore current thd->lex once when
|
|
|
|
do it in each instruction
|
|
|
|
*/
|
|
|
|
old_lex= thd->lex;
|
|
|
|
/*
|
|
|
|
We should also save Item tree change list to avoid rollback something
|
|
|
|
too early in the calling query.
|
|
|
|
*/
|
|
|
|
old_change_list= thd->change_list;
|
|
|
|
thd->change_list.empty();
|
|
|
|
/*
|
|
|
|
Cursors will use thd->packet, so they may corrupt data which was prepared
|
|
|
|
for sending by upper level. OTOH cursors in the same routine can share this
|
|
|
|
buffer safely so let use use routine-local packet instead of having own
|
|
|
|
packet buffer for each cursor.
|
|
|
|
|
|
|
|
It is probably safe to use same thd->convert_buff everywhere.
|
|
|
|
*/
|
|
|
|
old_packet.swap(thd->packet);
|
|
|
|
|
2005-08-18 11:23:54 +02:00
|
|
|
/*
|
|
|
|
Switch to per-instruction arena here. We can do it since we cleanup
|
|
|
|
arena after every instruction.
|
|
|
|
*/
|
2005-09-02 15:21:19 +02:00
|
|
|
thd->set_n_backup_active_arena(&execute_arena, &backup_arena);
|
2005-08-18 11:23:54 +02:00
|
|
|
|
|
|
|
/*
|
|
|
|
Save callers arena in order to store instruction results and out
|
|
|
|
parameters in it later during sp_eval_func_item()
|
|
|
|
*/
|
2005-09-02 15:21:19 +02:00
|
|
|
thd->spcont->callers_arena= &backup_arena;
|
2005-08-18 11:23:54 +02:00
|
|
|
|
2003-02-26 19:22:29 +01:00
|
|
|
do
|
|
|
|
{
|
|
|
|
sp_instr *i;
|
2003-09-16 14:26:08 +02:00
|
|
|
uint hip; // Handler ip
|
2003-02-26 19:22:29 +01:00
|
|
|
|
|
|
|
i = get_instr(ip); // Returns NULL when we're done.
|
|
|
|
if (i == NULL)
|
|
|
|
break;
|
|
|
|
DBUG_PRINT("execute", ("Instruction %u", ip));
|
2005-08-15 17:15:12 +02:00
|
|
|
/* Don't change NOW() in FUNCTION or TRIGGER */
|
|
|
|
if (!thd->in_sub_stmt)
|
|
|
|
thd->set_time(); // Make current_time() et al work
|
2005-09-07 17:39:47 +02:00
|
|
|
|
2005-06-23 18:22:08 +02:00
|
|
|
/*
|
2005-09-02 15:21:19 +02:00
|
|
|
We have to set thd->stmt_arena before executing the instruction
|
2005-06-23 18:22:08 +02:00
|
|
|
to store in the instruction free_list all new items, created
|
|
|
|
during the first execution (for example expanding of '*' or the
|
|
|
|
items made during other permanent subquery transformations).
|
|
|
|
*/
|
2005-09-02 15:21:19 +02:00
|
|
|
thd->stmt_arena= i;
|
2005-09-07 17:39:47 +02:00
|
|
|
|
2005-09-09 18:09:37 +02:00
|
|
|
/*
|
|
|
|
Will write this SP statement into binlog separately
|
|
|
|
(TODO: consider changing the condition to "not inside event union")
|
|
|
|
*/
|
|
|
|
if (thd->prelocked_mode == NON_PRELOCKED)
|
2005-09-07 17:39:47 +02:00
|
|
|
thd->user_var_events_alloc= thd->mem_root;
|
|
|
|
|
2005-12-07 15:01:17 +01:00
|
|
|
err_status= i->execute(thd, &ip);
|
2005-08-18 11:23:54 +02:00
|
|
|
|
2005-07-05 01:00:23 +02:00
|
|
|
/*
|
2005-07-05 01:40:01 +02:00
|
|
|
If this SP instruction have sent eof, it has caused no_send_error to be
|
|
|
|
set. Clear it back to allow the next instruction to send error. (multi-
|
|
|
|
statement execution code clears no_send_error between statements too)
|
2005-07-05 01:00:23 +02:00
|
|
|
*/
|
|
|
|
thd->net.no_send_error= 0;
|
2004-05-26 13:28:35 +02:00
|
|
|
if (i->free_list)
|
|
|
|
cleanup_items(i->free_list);
|
2005-09-09 18:09:37 +02:00
|
|
|
|
|
|
|
/*
|
|
|
|
If we've set thd->user_var_events_alloc to mem_root of this SP
|
|
|
|
statement, clean all the events allocated in it.
|
|
|
|
*/
|
|
|
|
if (thd->prelocked_mode == NON_PRELOCKED)
|
|
|
|
{
|
|
|
|
reset_dynamic(&thd->user_var_events);
|
|
|
|
thd->user_var_events_alloc= NULL;//DEBUG
|
|
|
|
}
|
2005-06-23 18:22:08 +02:00
|
|
|
|
2005-08-18 11:23:54 +02:00
|
|
|
/* we should cleanup free_list and memroot, used by instruction */
|
2005-12-01 11:26:46 +01:00
|
|
|
thd->cleanup_after_query();
|
2005-09-09 18:09:37 +02:00
|
|
|
free_root(&execute_mem_root, MYF(0));
|
2005-08-18 11:23:54 +02:00
|
|
|
|
2005-08-11 14:58:15 +02:00
|
|
|
/*
|
|
|
|
Check if an exception has occurred and a handler has been found
|
2005-12-07 15:01:17 +01:00
|
|
|
Note: We have to check even if err_status == FALSE, since warnings (and
|
|
|
|
some errors) don't return a non-zero value. We also have to check even
|
|
|
|
if thd->killed != 0, since some errors return with this even when a
|
|
|
|
handler has been found (e.g. "bad data").
|
2005-08-11 14:58:15 +02:00
|
|
|
*/
|
2005-04-05 15:49:10 +02:00
|
|
|
if (ctx)
|
2003-09-16 14:26:08 +02:00
|
|
|
{
|
|
|
|
uint hf;
|
|
|
|
|
2005-08-15 17:15:12 +02:00
|
|
|
switch (ctx->found_handler(&hip, &hf)) {
|
2003-09-16 14:26:08 +02:00
|
|
|
case SP_HANDLER_NONE:
|
|
|
|
break;
|
|
|
|
case SP_HANDLER_CONTINUE:
|
2005-09-02 15:21:19 +02:00
|
|
|
thd->restore_active_arena(&execute_arena, &backup_arena);
|
|
|
|
thd->set_n_backup_active_arena(&execute_arena, &backup_arena);
|
2007-03-07 17:53:46 +01:00
|
|
|
ctx->push_hstack(i->get_cont_dest());
|
2004-10-20 03:04:37 +02:00
|
|
|
// Fall through
|
2003-09-16 14:26:08 +02:00
|
|
|
default:
|
|
|
|
ip= hip;
|
2005-12-07 15:01:17 +01:00
|
|
|
err_status= FALSE;
|
2003-09-16 14:26:08 +02:00
|
|
|
ctx->clear_handler();
|
2005-09-26 18:22:00 +02:00
|
|
|
ctx->enter_handler(hip);
|
2004-10-20 03:04:37 +02:00
|
|
|
thd->clear_error();
|
2005-04-05 15:49:10 +02:00
|
|
|
thd->killed= THD::NOT_KILLED;
|
2007-03-15 09:51:35 +01:00
|
|
|
thd->mysys_var->abort= 0;
|
2003-09-16 14:26:08 +02:00
|
|
|
continue;
|
|
|
|
}
|
|
|
|
}
|
2005-12-07 15:01:17 +01:00
|
|
|
} while (!err_status && !thd->killed);
|
2003-03-26 15:02:48 +01:00
|
|
|
|
2005-09-02 15:21:19 +02:00
|
|
|
thd->restore_active_arena(&execute_arena, &backup_arena);
|
2005-08-18 11:23:54 +02:00
|
|
|
|
2006-07-13 15:12:31 +02:00
|
|
|
thd->spcont->pop_all_cursors(); // To avoid memory leaks after an error
|
2005-08-18 11:23:54 +02:00
|
|
|
|
2005-03-04 14:35:28 +01:00
|
|
|
/* Restore all saved */
|
|
|
|
old_packet.swap(thd->packet);
|
|
|
|
DBUG_ASSERT(thd->change_list.is_empty());
|
|
|
|
thd->change_list= old_change_list;
|
|
|
|
/* To avoid wiping out thd->change_list on old_change_list destruction */
|
|
|
|
old_change_list.empty();
|
|
|
|
thd->lex= old_lex;
|
|
|
|
thd->query_id= old_query_id;
|
|
|
|
DBUG_ASSERT(!thd->derived_tables);
|
|
|
|
thd->derived_tables= old_derived_tables;
|
2005-07-28 21:39:11 +02:00
|
|
|
thd->variables.sql_mode= save_sql_mode;
|
2006-04-19 12:27:59 +02:00
|
|
|
thd->abort_on_warning= save_abort_on_warning;
|
2005-03-04 14:35:28 +01:00
|
|
|
|
2005-09-02 15:21:19 +02:00
|
|
|
thd->stmt_arena= old_arena;
|
2005-06-21 19:15:21 +02:00
|
|
|
state= EXECUTED;
|
2004-07-22 16:46:59 +02:00
|
|
|
|
2004-03-11 17:18:59 +01:00
|
|
|
done:
|
2006-01-06 00:08:48 +01:00
|
|
|
DBUG_PRINT("info", ("err_status: %d killed: %d query_error: %d",
|
2005-12-07 15:01:17 +01:00
|
|
|
err_status, thd->killed, thd->query_error));
|
2004-05-26 13:28:35 +02:00
|
|
|
|
2004-10-20 03:04:37 +02:00
|
|
|
if (thd->killed)
|
2005-12-07 15:01:17 +01:00
|
|
|
err_status= TRUE;
|
2006-01-06 00:08:48 +01:00
|
|
|
/*
|
|
|
|
If the DB has changed, the pointer has changed too, but the
|
|
|
|
original thd->db will then have been freed
|
|
|
|
*/
|
2004-06-08 18:41:18 +02:00
|
|
|
if (dbchanged)
|
2003-03-26 15:02:48 +01:00
|
|
|
{
|
2005-12-06 22:57:15 +01:00
|
|
|
/*
|
|
|
|
No access check when changing back to where we came from.
|
A fix and a test case for
Bug#19022 "Memory bug when switching db during trigger execution"
Bug#17199 "Problem when view calls function from another database."
Bug#18444 "Fully qualified stored function names don't work correctly in
SELECT statements"
Documentation note: this patch introduces a change in behaviour of prepared
statements.
This patch adds a few new invariants with regard to how THD::db should
be used. These invariants should be preserved in future:
- one should never refer to THD::db by pointer and always make a deep copy
(strmake, strdup)
- one should never compare two databases by pointer, but use strncmp or
my_strncasecmp
- TABLE_LIST object table->db should be always initialized in the parser or
by creator of the object.
For prepared statements it means that if the current database is changed
after a statement is prepared, the database that was current at prepare
remains active. This also means that you can not prepare a statement that
implicitly refers to the current database if the latter is not set.
This is not documented, and therefore needs documentation. This is NOT a
change in behavior for almost all SQL statements except:
- ALTER TABLE t1 RENAME t2
- OPTIMIZE TABLE t1
- ANALYZE TABLE t1
- TRUNCATE TABLE t1 --
until this patch t1 or t2 could be evaluated at the first execution of
prepared statement.
CURRENT_DATABASE() still works OK and is evaluated at every execution
of prepared statement.
Note, that in stored routines this is not an issue as the default
database is the database of the stored procedure and "use" statement
is prohibited in stored routines.
This patch makes obsolete the use of check_db_used (it was never used in the
old code too) and all other places that check for table->db and assign it
from THD::db if it's NULL, except the parser.
How this patch was created: THD::{db,db_length} were replaced with a
LEX_STRING, THD::db. All the places that refer to THD::{db,db_length} were
manually checked and:
- if the place uses thd->db by pointer, it was fixed to make a deep copy
- if a place compared two db pointers, it was fixed to compare them by value
(via strcmp/my_strcasecmp, whatever was approproate)
Then this intermediate patch was used to write a smaller patch that does the
same thing but without a rename.
TODO in 5.1:
- remove check_db_used
- deploy THD::set_db in mysql_change_db
See also comments to individual files.
2006-06-26 22:47:52 +02:00
|
|
|
(It would generate an error from mysql_change_db() when old_db=="")
|
2005-12-06 22:57:15 +01:00
|
|
|
*/
|
2003-03-28 17:02:31 +01:00
|
|
|
if (! thd->killed)
|
2007-03-27 19:55:01 +02:00
|
|
|
err_status|= mysql_change_db(thd, &old_db, TRUE);
|
2003-03-26 15:02:48 +01:00
|
|
|
}
|
Implement WL#2661 "Prepared Statements: Dynamic SQL in Stored Procedures".
The idea of the patch is to separate statement processing logic,
such as parsing, validation of the parsed tree, execution and cleanup,
from global query processing logic, such as logging, resetting
priorities of a thread, resetting stored procedure cache, resetting
thread count of errors and warnings.
This makes PREPARE and EXECUTE behave similarly to the rest of SQL
statements and allows their use in stored procedures.
This patch contains a change in behaviour:
until recently for each SQL prepared statement command, 2 queries
were written to the general log, e.g.
[Query] prepare stmt from @stmt_text;
[Prepare] select * from t1 <-- contents of @stmt_text
The chagne was necessary to prevent [Prepare] commands from being written
to the general log when executing a stored procedure with Dynamic SQL.
We should consider whether the old behavior is preferrable and probably
restore it.
This patch refixes Bug#7115, Bug#10975 (partially), Bug#10605 (various bugs
in Dynamic SQL reported before it was disabled).
2005-09-03 01:13:18 +02:00
|
|
|
m_flags&= ~IS_INVOKED;
|
2006-01-06 00:08:48 +01:00
|
|
|
DBUG_PRINT("info",
|
|
|
|
("first free for 0x%lx --: 0x%lx->0x%lx, level: %lu, flags %x",
|
|
|
|
(ulong) m_first_instance,
|
|
|
|
(ulong) m_first_instance->m_first_free_instance,
|
|
|
|
(ulong) this, m_recursion_level, m_flags));
|
2005-11-23 00:11:19 +01:00
|
|
|
/*
|
|
|
|
Check that we have one of following:
|
|
|
|
|
|
|
|
1) there are not free instances which means that this instance is last
|
|
|
|
in the list of instances (pointer to the last instance point on it and
|
|
|
|
ther are not other instances after this one in the list)
|
|
|
|
|
|
|
|
2) There are some free instances which mean that first free instance
|
|
|
|
should go just after this one and recursion level of that free instance
|
2006-01-05 23:47:49 +01:00
|
|
|
should be on 1 more then recursion level of this instance.
|
2005-11-23 00:11:19 +01:00
|
|
|
*/
|
|
|
|
DBUG_ASSERT((m_first_instance->m_first_free_instance == 0 &&
|
|
|
|
this == m_first_instance->m_last_cached_sp &&
|
|
|
|
m_next_cached_sp == 0) ||
|
|
|
|
(m_first_instance->m_first_free_instance != 0 &&
|
|
|
|
m_first_instance->m_first_free_instance == m_next_cached_sp &&
|
|
|
|
m_first_instance->m_first_free_instance->m_recursion_level ==
|
|
|
|
m_recursion_level + 1));
|
|
|
|
m_first_instance->m_first_free_instance= this;
|
2006-07-13 15:12:31 +02:00
|
|
|
|
|
|
|
DBUG_RETURN(err_status);
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
#ifndef NO_EMBEDDED_ACCESS_CHECKS
|
|
|
|
/*
|
|
|
|
set_routine_security_ctx() changes routine security context, and
|
|
|
|
checks if there is an EXECUTE privilege in new context. If there is
|
|
|
|
no EXECUTE privilege, it changes the context back and returns a
|
|
|
|
error.
|
|
|
|
|
|
|
|
SYNOPSIS
|
|
|
|
set_routine_security_ctx()
|
|
|
|
thd thread handle
|
|
|
|
sp stored routine to change the context for
|
|
|
|
is_proc TRUE is procedure, FALSE if function
|
|
|
|
save_ctx pointer to an old security context
|
|
|
|
|
|
|
|
RETURN
|
|
|
|
TRUE if there was a error, and the context wasn't changed.
|
|
|
|
FALSE if the context was changed.
|
|
|
|
*/
|
|
|
|
|
|
|
|
bool
|
|
|
|
set_routine_security_ctx(THD *thd, sp_head *sp, bool is_proc,
|
|
|
|
Security_context **save_ctx)
|
|
|
|
{
|
|
|
|
*save_ctx= 0;
|
|
|
|
if (sp_change_security_context(thd, sp, save_ctx))
|
|
|
|
return TRUE;
|
|
|
|
|
|
|
|
/*
|
|
|
|
If we changed context to run as another user, we need to check the
|
|
|
|
access right for the new context again as someone may have revoked
|
|
|
|
the right to use the procedure from this user.
|
|
|
|
|
|
|
|
TODO:
|
|
|
|
Cache if the definer has the right to use the object on the
|
|
|
|
first usage and only reset the cache if someone does a GRANT
|
|
|
|
statement that 'may' affect this.
|
|
|
|
*/
|
|
|
|
if (*save_ctx &&
|
|
|
|
check_routine_access(thd, EXECUTE_ACL,
|
|
|
|
sp->m_db.str, sp->m_name.str, is_proc, FALSE))
|
|
|
|
{
|
|
|
|
sp_restore_security_context(thd, *save_ctx);
|
|
|
|
*save_ctx= 0;
|
|
|
|
return TRUE;
|
|
|
|
}
|
|
|
|
|
|
|
|
return FALSE;
|
|
|
|
}
|
|
|
|
#endif // ! NO_EMBEDDED_ACCESS_CHECKS
|
|
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
Execute a trigger:
|
|
|
|
- changes security context for triggers
|
|
|
|
- switch to new memroot
|
|
|
|
- call sp_head::execute
|
|
|
|
- restore old memroot
|
|
|
|
- restores security context
|
|
|
|
|
|
|
|
SYNOPSIS
|
|
|
|
sp_head::execute_trigger()
|
|
|
|
thd Thread handle
|
|
|
|
db database name
|
|
|
|
table table name
|
|
|
|
grant_info GRANT_INFO structure to be filled with
|
|
|
|
information about definer's privileges
|
|
|
|
on subject table
|
|
|
|
|
|
|
|
RETURN
|
|
|
|
FALSE on success
|
|
|
|
TRUE on error
|
|
|
|
*/
|
|
|
|
|
|
|
|
bool
|
|
|
|
sp_head::execute_trigger(THD *thd, const char *db, const char *table,
|
|
|
|
GRANT_INFO *grant_info)
|
|
|
|
{
|
|
|
|
sp_rcontext *octx = thd->spcont;
|
|
|
|
sp_rcontext *nctx = NULL;
|
|
|
|
bool err_status= FALSE;
|
|
|
|
MEM_ROOT call_mem_root;
|
|
|
|
Query_arena call_arena(&call_mem_root, Query_arena::INITIALIZED_FOR_SP);
|
|
|
|
Query_arena backup_arena;
|
|
|
|
|
|
|
|
DBUG_ENTER("sp_head::execute_trigger");
|
|
|
|
DBUG_PRINT("info", ("trigger %s", m_name.str));
|
|
|
|
|
|
|
|
#ifndef NO_EMBEDDED_ACCESS_CHECKS
|
|
|
|
Security_context *save_ctx;
|
|
|
|
if (sp_change_security_context(thd, this, &save_ctx))
|
|
|
|
DBUG_RETURN(TRUE);
|
|
|
|
|
|
|
|
/*
|
|
|
|
NOTE: TRIGGER_ACL should be used here.
|
|
|
|
*/
|
|
|
|
if (check_global_access(thd, SUPER_ACL))
|
|
|
|
{
|
|
|
|
sp_restore_security_context(thd, save_ctx);
|
|
|
|
DBUG_RETURN(TRUE);
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
Fetch information about table-level privileges to GRANT_INFO
|
|
|
|
structure for subject table. Check of privileges that will use it
|
|
|
|
and information about column-level privileges will happen in
|
|
|
|
Item_trigger_field::fix_fields().
|
|
|
|
*/
|
|
|
|
fill_effective_table_privileges(thd, grant_info, db, table);
|
|
|
|
#endif // NO_EMBEDDED_ACCESS_CHECKS
|
|
|
|
|
|
|
|
/*
|
|
|
|
Prepare arena and memroot for objects which lifetime is whole
|
|
|
|
duration of trigger call (sp_rcontext, it's tables and items,
|
|
|
|
sp_cursor and Item_cache holders for case expressions). We can't
|
|
|
|
use caller's arena/memroot for those objects because in this case
|
|
|
|
some fixed amount of memory will be consumed for each trigger
|
|
|
|
invocation and so statements which involve lot of them will hog
|
|
|
|
memory.
|
|
|
|
|
|
|
|
TODO: we should create sp_rcontext once per command and reuse it
|
|
|
|
on subsequent executions of a trigger.
|
|
|
|
*/
|
|
|
|
init_sql_alloc(&call_mem_root, MEM_ROOT_BLOCK_SIZE, 0);
|
|
|
|
thd->set_n_backup_active_arena(&call_arena, &backup_arena);
|
|
|
|
|
|
|
|
if (!(nctx= new sp_rcontext(m_pcont, 0, octx)) ||
|
|
|
|
nctx->init(thd))
|
|
|
|
{
|
|
|
|
err_status= TRUE;
|
|
|
|
goto err_with_cleanup;
|
|
|
|
}
|
|
|
|
|
|
|
|
#ifndef DBUG_OFF
|
|
|
|
nctx->sp= this;
|
|
|
|
#endif
|
|
|
|
|
|
|
|
thd->spcont= nctx;
|
|
|
|
|
|
|
|
err_status= execute(thd);
|
|
|
|
|
|
|
|
err_with_cleanup:
|
|
|
|
thd->restore_active_arena(&call_arena, &backup_arena);
|
|
|
|
#ifndef NO_EMBEDDED_ACCESS_CHECKS
|
|
|
|
sp_restore_security_context(thd, save_ctx);
|
|
|
|
#endif // NO_EMBEDDED_ACCESS_CHECKS
|
|
|
|
delete nctx;
|
|
|
|
call_arena.free_items();
|
|
|
|
free_root(&call_mem_root, MYF(0));
|
|
|
|
thd->spcont= octx;
|
|
|
|
|
2007-05-23 21:24:16 +02:00
|
|
|
if (thd->killed)
|
|
|
|
thd->send_kill_message();
|
|
|
|
|
2005-12-07 15:01:17 +01:00
|
|
|
DBUG_RETURN(err_status);
|
2003-02-26 19:22:29 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
|
2005-08-25 15:34:34 +02:00
|
|
|
/*
|
|
|
|
Execute a function:
|
|
|
|
- evaluate parameters
|
2006-07-13 15:12:31 +02:00
|
|
|
- changes security context for SUID routines
|
|
|
|
- switch to new memroot
|
2005-08-25 15:34:34 +02:00
|
|
|
- call sp_head::execute
|
2006-07-13 15:12:31 +02:00
|
|
|
- restore old memroot
|
2005-08-25 15:34:34 +02:00
|
|
|
- evaluate the return value
|
2006-07-13 15:12:31 +02:00
|
|
|
- restores security context
|
2005-08-25 15:34:34 +02:00
|
|
|
|
|
|
|
SYNOPSIS
|
|
|
|
sp_head::execute_function()
|
2005-12-07 15:01:17 +01:00
|
|
|
thd Thread handle
|
|
|
|
argp Passed arguments (these are items from containing
|
|
|
|
statement?)
|
|
|
|
argcount Number of passed arguments. We need to check if this is
|
|
|
|
correct.
|
|
|
|
return_value_fld Save result here.
|
2005-08-25 15:34:34 +02:00
|
|
|
|
|
|
|
RETURN
|
2005-12-07 15:01:17 +01:00
|
|
|
FALSE on success
|
|
|
|
TRUE on error
|
2005-08-25 15:34:34 +02:00
|
|
|
*/
|
|
|
|
|
2005-12-07 15:01:17 +01:00
|
|
|
bool
|
|
|
|
sp_head::execute_function(THD *thd, Item **argp, uint argcount,
|
|
|
|
Field *return_value_fld)
|
2003-02-26 19:22:29 +01:00
|
|
|
{
|
2005-08-25 15:34:34 +02:00
|
|
|
ulonglong binlog_save_options;
|
|
|
|
bool need_binlog_call;
|
2006-05-06 11:51:35 +02:00
|
|
|
uint arg_no;
|
2003-02-26 19:22:29 +01:00
|
|
|
sp_rcontext *octx = thd->spcont;
|
|
|
|
sp_rcontext *nctx = NULL;
|
2006-05-06 11:51:35 +02:00
|
|
|
char buf[STRING_BUFFER_USUAL_SIZE];
|
|
|
|
String binlog_buf(buf, sizeof(buf), &my_charset_bin);
|
2005-12-07 15:01:17 +01:00
|
|
|
bool err_status= FALSE;
|
2006-05-06 11:51:35 +02:00
|
|
|
MEM_ROOT call_mem_root;
|
|
|
|
Query_arena call_arena(&call_mem_root, Query_arena::INITIALIZED_FOR_SP);
|
|
|
|
Query_arena backup_arena;
|
2005-12-07 15:01:17 +01:00
|
|
|
|
|
|
|
DBUG_ENTER("sp_head::execute_function");
|
|
|
|
DBUG_PRINT("info", ("function %s", m_name.str));
|
|
|
|
|
|
|
|
/*
|
|
|
|
Check that the function is called with all specified arguments.
|
|
|
|
|
|
|
|
If it is not, use my_error() to report an error, or it will not terminate
|
|
|
|
the invoking query properly.
|
|
|
|
*/
|
2006-05-06 11:51:35 +02:00
|
|
|
if (argcount != m_pcont->context_var_count())
|
2003-04-17 13:20:02 +02:00
|
|
|
{
|
2005-08-11 14:58:15 +02:00
|
|
|
/*
|
2005-08-22 00:13:37 +02:00
|
|
|
Need to use my_error here, or it will not terminate the
|
2005-08-11 14:58:15 +02:00
|
|
|
invoking query properly.
|
|
|
|
*/
|
2004-11-13 18:35:51 +01:00
|
|
|
my_error(ER_SP_WRONG_NO_OF_ARGS, MYF(0),
|
2006-05-06 11:51:35 +02:00
|
|
|
"FUNCTION", m_qname.str, m_pcont->context_var_count(), argcount);
|
2005-12-07 15:01:17 +01:00
|
|
|
DBUG_RETURN(TRUE);
|
2003-04-17 13:20:02 +02:00
|
|
|
}
|
2006-05-06 11:51:35 +02:00
|
|
|
/*
|
|
|
|
Prepare arena and memroot for objects which lifetime is whole
|
|
|
|
duration of function call (sp_rcontext, it's tables and items,
|
|
|
|
sp_cursor and Item_cache holders for case expressions).
|
|
|
|
We can't use caller's arena/memroot for those objects because
|
|
|
|
in this case some fixed amount of memory will be consumed for
|
|
|
|
each function/trigger invocation and so statements which involve
|
|
|
|
lot of them will hog memory.
|
|
|
|
TODO: we should create sp_rcontext once per command and reuse
|
|
|
|
it on subsequent executions of a function/trigger.
|
|
|
|
*/
|
|
|
|
init_sql_alloc(&call_mem_root, MEM_ROOT_BLOCK_SIZE, 0);
|
|
|
|
thd->set_n_backup_active_arena(&call_arena, &backup_arena);
|
2005-12-07 15:01:17 +01:00
|
|
|
|
|
|
|
if (!(nctx= new sp_rcontext(m_pcont, return_value_fld, octx)) ||
|
|
|
|
nctx->init(thd))
|
|
|
|
{
|
2006-05-06 11:51:35 +02:00
|
|
|
thd->restore_active_arena(&call_arena, &backup_arena);
|
|
|
|
err_status= TRUE;
|
|
|
|
goto err_with_cleanup;
|
2005-12-07 15:01:17 +01:00
|
|
|
}
|
|
|
|
|
2006-05-06 11:51:35 +02:00
|
|
|
/*
|
|
|
|
We have to switch temporarily back to callers arena/memroot.
|
|
|
|
Function arguments belong to the caller and so the may reference
|
|
|
|
memory which they will allocate during calculation long after
|
|
|
|
this function call will be finished (e.g. in Item::cleanup()).
|
|
|
|
*/
|
|
|
|
thd->restore_active_arena(&call_arena, &backup_arena);
|
|
|
|
|
2005-11-22 23:50:37 +01:00
|
|
|
#ifndef DBUG_OFF
|
2005-12-07 15:01:17 +01:00
|
|
|
nctx->sp= this;
|
2005-11-22 23:50:37 +01:00
|
|
|
#endif
|
2005-12-07 15:01:17 +01:00
|
|
|
|
|
|
|
/* Pass arguments. */
|
2006-05-06 11:51:35 +02:00
|
|
|
for (arg_no= 0; arg_no < argcount; arg_no++)
|
|
|
|
{
|
|
|
|
/* Arguments must be fixed in Item_func_sp::fix_fields */
|
|
|
|
DBUG_ASSERT(argp[arg_no]->fixed);
|
2005-12-07 15:01:17 +01:00
|
|
|
|
2006-05-15 19:57:09 +02:00
|
|
|
if ((err_status= nctx->set_variable(thd, arg_no, &(argp[arg_no]))))
|
2006-05-06 11:51:35 +02:00
|
|
|
goto err_with_cleanup;
|
|
|
|
}
|
|
|
|
|
|
|
|
need_binlog_call= mysql_bin_log.is_open() && (thd->options & OPTION_BIN_LOG);
|
2005-12-07 15:01:17 +01:00
|
|
|
|
2006-05-06 11:51:35 +02:00
|
|
|
/*
|
|
|
|
Remember the original arguments for unrolled replication of functions
|
|
|
|
before they are changed by execution.
|
|
|
|
*/
|
|
|
|
if (need_binlog_call)
|
2003-02-26 19:22:29 +01:00
|
|
|
{
|
2006-05-06 11:51:35 +02:00
|
|
|
binlog_buf.length(0);
|
|
|
|
binlog_buf.append(STRING_WITH_LEN("SELECT "));
|
Fix for BUG#19725 "Calls to SF in other database are not replicated
correctly in some cases".
In short, calls to a stored function located in another database
than the default database, may fail to replicate if the call was made
by SET, SELECT, or DO.
Longer: when a stored function is called from a statement which does not go
to binlog ("SET @a=somedb.myfunc()", "SELECT somedb.myfunc()",
"DO somedb.myfunc()"), this crafted statement is binlogged:
"SELECT myfunc();" (accompanied with a mention of the default database
if there is one). So, if "somedb" is not the default database,
the slave would fail to find myfunc(). The fix is to specify the
function's database name in the crafted binlogged statement, like this:
"SELECT somedb.myfunc();". Test added in rpl_sp.test.
2007-01-08 22:01:06 +01:00
|
|
|
append_identifier(thd, &binlog_buf, m_db.str, m_db.length);
|
|
|
|
binlog_buf.append('.');
|
2006-05-06 11:51:35 +02:00
|
|
|
append_identifier(thd, &binlog_buf, m_name.str, m_name.length);
|
|
|
|
binlog_buf.append('(');
|
|
|
|
for (arg_no= 0; arg_no < argcount; arg_no++)
|
2005-12-07 15:01:17 +01:00
|
|
|
{
|
2006-05-06 11:51:35 +02:00
|
|
|
String str_value_holder;
|
|
|
|
String *str_value;
|
2003-02-26 19:22:29 +01:00
|
|
|
|
2006-05-06 11:51:35 +02:00
|
|
|
if (arg_no)
|
|
|
|
binlog_buf.append(',');
|
2005-08-22 00:13:37 +02:00
|
|
|
|
2006-11-09 11:27:34 +01:00
|
|
|
str_value= sp_get_item_value(thd, nctx->get_item(arg_no),
|
2006-05-06 11:51:35 +02:00
|
|
|
&str_value_holder);
|
2005-09-13 12:50:21 +02:00
|
|
|
|
2006-05-06 11:51:35 +02:00
|
|
|
if (str_value)
|
|
|
|
binlog_buf.append(*str_value);
|
|
|
|
else
|
|
|
|
binlog_buf.append(STRING_WITH_LEN("NULL"));
|
|
|
|
}
|
|
|
|
binlog_buf.append(')');
|
2005-12-07 15:01:17 +01:00
|
|
|
}
|
2003-02-26 19:22:29 +01:00
|
|
|
thd->spcont= nctx;
|
|
|
|
|
2006-07-13 15:12:31 +02:00
|
|
|
#ifndef NO_EMBEDDED_ACCESS_CHECKS
|
|
|
|
Security_context *save_security_ctx;
|
|
|
|
if (set_routine_security_ctx(thd, this, FALSE, &save_security_ctx))
|
|
|
|
{
|
|
|
|
err_status= TRUE;
|
|
|
|
goto err_with_cleanup;
|
|
|
|
}
|
|
|
|
#endif
|
|
|
|
|
2005-08-25 15:34:34 +02:00
|
|
|
binlog_save_options= thd->options;
|
|
|
|
if (need_binlog_call)
|
2005-09-07 17:39:47 +02:00
|
|
|
{
|
2007-02-23 18:58:56 +01:00
|
|
|
query_id_t q;
|
2005-09-07 17:39:47 +02:00
|
|
|
reset_dynamic(&thd->user_var_events);
|
2007-02-23 18:58:56 +01:00
|
|
|
/*
|
|
|
|
In case of artificially constructed events for function calls
|
|
|
|
we have separate union for each such event and hence can't use
|
|
|
|
query_id of real calling statement as the start of all these
|
|
|
|
unions (this will break logic of replication of user-defined
|
|
|
|
variables). So we use artifical value which is guaranteed to
|
|
|
|
be greater than all query_id's of all statements belonging
|
|
|
|
to previous events/unions.
|
|
|
|
Possible alternative to this is logging of all function invocations
|
|
|
|
as one select and not resetting THD::user_var_events before
|
|
|
|
each invocation.
|
|
|
|
*/
|
|
|
|
VOID(pthread_mutex_lock(&LOCK_thread_count));
|
2007-02-27 16:32:45 +01:00
|
|
|
q= global_query_id;
|
2007-02-23 18:58:56 +01:00
|
|
|
VOID(pthread_mutex_unlock(&LOCK_thread_count));
|
|
|
|
mysql_bin_log.start_union_events(thd, q + 1);
|
2005-09-07 17:39:47 +02:00
|
|
|
}
|
2006-05-06 11:51:35 +02:00
|
|
|
|
|
|
|
/*
|
|
|
|
Switch to call arena/mem_root so objects like sp_cursor or
|
|
|
|
Item_cache holders for case expressions can be allocated on it.
|
|
|
|
|
|
|
|
TODO: In future we should associate call arena/mem_root with
|
|
|
|
sp_rcontext and allocate all these objects (and sp_rcontext
|
|
|
|
itself) on it directly rather than juggle with arenas.
|
|
|
|
*/
|
|
|
|
thd->set_n_backup_active_arena(&call_arena, &backup_arena);
|
|
|
|
|
2005-08-25 15:34:34 +02:00
|
|
|
thd->options&= ~OPTION_BIN_LOG;
|
2005-12-07 15:01:17 +01:00
|
|
|
err_status= execute(thd);
|
2005-08-25 15:34:34 +02:00
|
|
|
thd->options= binlog_save_options;
|
2006-05-06 11:51:35 +02:00
|
|
|
|
|
|
|
thd->restore_active_arena(&call_arena, &backup_arena);
|
|
|
|
|
2005-08-25 15:34:34 +02:00
|
|
|
if (need_binlog_call)
|
|
|
|
mysql_bin_log.stop_union_events(thd);
|
|
|
|
|
2005-08-28 18:25:53 +02:00
|
|
|
if (need_binlog_call && thd->binlog_evt_union.unioned_events)
|
2005-08-25 15:34:34 +02:00
|
|
|
{
|
2006-05-06 11:51:35 +02:00
|
|
|
Query_log_event qinfo(thd, binlog_buf.ptr(), binlog_buf.length(),
|
2005-08-27 00:33:06 +02:00
|
|
|
thd->binlog_evt_union.unioned_events_trans, FALSE);
|
2006-05-06 11:51:35 +02:00
|
|
|
if (mysql_bin_log.write(&qinfo) &&
|
2005-08-27 00:33:06 +02:00
|
|
|
thd->binlog_evt_union.unioned_events_trans)
|
2005-08-25 15:34:34 +02:00
|
|
|
{
|
2005-08-27 00:33:06 +02:00
|
|
|
push_warning(thd, MYSQL_ERROR::WARN_LEVEL_WARN, ER_UNKNOWN_ERROR,
|
|
|
|
"Invoked ROUTINE modified a transactional table but MySQL "
|
|
|
|
"failed to reflect this change in the binary log");
|
2005-08-25 15:34:34 +02:00
|
|
|
}
|
2005-09-07 17:39:47 +02:00
|
|
|
reset_dynamic(&thd->user_var_events);
|
2005-08-25 15:34:34 +02:00
|
|
|
}
|
2004-09-07 14:29:46 +02:00
|
|
|
|
2006-07-13 15:12:31 +02:00
|
|
|
if (!err_status)
|
2003-10-03 17:38:12 +02:00
|
|
|
{
|
2004-09-07 14:29:46 +02:00
|
|
|
/* We need result only in function but not in trigger */
|
2003-10-03 17:38:12 +02:00
|
|
|
|
2005-12-07 15:01:17 +01:00
|
|
|
if (!nctx->is_return_value_set())
|
2003-10-03 17:38:12 +02:00
|
|
|
{
|
2004-11-13 18:35:51 +01:00
|
|
|
my_error(ER_SP_NORETURNEND, MYF(0), m_name.str);
|
2005-12-07 15:01:17 +01:00
|
|
|
err_status= TRUE;
|
2003-10-03 17:38:12 +02:00
|
|
|
}
|
|
|
|
}
|
2003-02-26 19:22:29 +01:00
|
|
|
|
2006-07-13 15:12:31 +02:00
|
|
|
#ifndef NO_EMBEDDED_ACCESS_CHECKS
|
|
|
|
sp_restore_security_context(thd, save_security_ctx);
|
|
|
|
#endif
|
2006-05-06 11:51:35 +02:00
|
|
|
|
|
|
|
err_with_cleanup:
|
2005-12-07 15:01:17 +01:00
|
|
|
delete nctx;
|
2006-05-06 11:51:35 +02:00
|
|
|
call_arena.free_items();
|
|
|
|
free_root(&call_mem_root, MYF(0));
|
2003-02-26 19:22:29 +01:00
|
|
|
thd->spcont= octx;
|
2005-06-03 19:21:12 +02:00
|
|
|
|
2005-12-07 15:01:17 +01:00
|
|
|
DBUG_RETURN(err_status);
|
2003-02-26 19:22:29 +01:00
|
|
|
}
|
|
|
|
|
2005-08-22 00:13:37 +02:00
|
|
|
|
2005-08-25 15:34:34 +02:00
|
|
|
/*
|
|
|
|
Execute a procedure.
|
|
|
|
SYNOPSIS
|
|
|
|
sp_head::execute_procedure()
|
|
|
|
thd Thread handle
|
|
|
|
args List of values passed as arguments.
|
|
|
|
|
|
|
|
DESCRIPTION
|
|
|
|
|
|
|
|
The function does the following steps:
|
|
|
|
- Set all parameters
|
2006-07-13 15:12:31 +02:00
|
|
|
- changes security context for SUID routines
|
2005-08-25 15:34:34 +02:00
|
|
|
- call sp_head::execute
|
|
|
|
- copy back values of INOUT and OUT parameters
|
2006-07-13 15:12:31 +02:00
|
|
|
- restores security context
|
2005-08-25 15:34:34 +02:00
|
|
|
|
|
|
|
RETURN
|
2005-12-07 15:01:17 +01:00
|
|
|
FALSE on success
|
|
|
|
TRUE on error
|
2005-08-25 15:34:34 +02:00
|
|
|
*/
|
|
|
|
|
2005-12-07 15:01:17 +01:00
|
|
|
bool
|
|
|
|
sp_head::execute_procedure(THD *thd, List<Item> *args)
|
2003-02-26 19:22:29 +01:00
|
|
|
{
|
2005-12-07 15:01:17 +01:00
|
|
|
bool err_status= FALSE;
|
2006-04-07 16:53:15 +02:00
|
|
|
uint params = m_pcont->context_var_count();
|
2005-08-22 00:13:37 +02:00
|
|
|
sp_rcontext *save_spcont, *octx;
|
Simplistic, experimental framework for Stored Procedures (SPs).
Implements creation and dropping of PROCEDUREs, IN, OUT, and INOUT parameters,
single-statement procedures, rudimentary multi-statement (begin-end) prodedures
(when the client can handle it), and local variables.
Missing most of the embedded SQL language, all attributes, FUNCTIONs, error handling,
reparses procedures at each call (no caching), etc, etc.
Certainly buggy too, but procedures can actually be created and called....
2002-12-08 19:59:22 +01:00
|
|
|
sp_rcontext *nctx = NULL;
|
2005-08-22 00:13:37 +02:00
|
|
|
DBUG_ENTER("sp_head::execute_procedure");
|
|
|
|
DBUG_PRINT("info", ("procedure %s", m_name.str));
|
Simplistic, experimental framework for Stored Procedures (SPs).
Implements creation and dropping of PROCEDUREs, IN, OUT, and INOUT parameters,
single-statement procedures, rudimentary multi-statement (begin-end) prodedures
(when the client can handle it), and local variables.
Missing most of the embedded SQL language, all attributes, FUNCTIONs, error handling,
reparses procedures at each call (no caching), etc, etc.
Certainly buggy too, but procedures can actually be created and called....
2002-12-08 19:59:22 +01:00
|
|
|
|
2003-04-17 13:20:02 +02:00
|
|
|
if (args->elements != params)
|
|
|
|
{
|
2004-11-13 18:35:51 +01:00
|
|
|
my_error(ER_SP_WRONG_NO_OF_ARGS, MYF(0), "PROCEDURE",
|
2005-04-14 14:52:35 +02:00
|
|
|
m_qname.str, params, args->elements);
|
2005-12-07 15:01:17 +01:00
|
|
|
DBUG_RETURN(TRUE);
|
2003-04-17 13:20:02 +02:00
|
|
|
}
|
|
|
|
|
2005-08-22 00:13:37 +02:00
|
|
|
save_spcont= octx= thd->spcont;
|
2005-08-18 11:23:54 +02:00
|
|
|
if (! octx)
|
|
|
|
{ // Create a temporary old context
|
2005-12-07 15:01:17 +01:00
|
|
|
if (!(octx= new sp_rcontext(m_pcont, NULL, octx)) ||
|
|
|
|
octx->init(thd))
|
|
|
|
{
|
|
|
|
delete octx; /* Delete octx if it was init() that failed. */
|
|
|
|
DBUG_RETURN(TRUE);
|
|
|
|
}
|
|
|
|
|
2005-11-22 23:50:37 +01:00
|
|
|
#ifndef DBUG_OFF
|
2005-12-07 15:01:17 +01:00
|
|
|
octx->sp= 0;
|
2005-11-22 23:50:37 +01:00
|
|
|
#endif
|
2005-08-18 11:23:54 +02:00
|
|
|
thd->spcont= octx;
|
|
|
|
|
|
|
|
/* set callers_arena to thd, for upper-level function to work */
|
|
|
|
thd->spcont->callers_arena= thd;
|
|
|
|
}
|
|
|
|
|
2005-12-07 15:01:17 +01:00
|
|
|
if (!(nctx= new sp_rcontext(m_pcont, NULL, octx)) ||
|
|
|
|
nctx->init(thd))
|
2005-08-22 00:13:37 +02:00
|
|
|
{
|
2005-12-07 15:01:17 +01:00
|
|
|
delete nctx; /* Delete nctx if it was init() that failed. */
|
2005-08-22 00:13:37 +02:00
|
|
|
thd->spcont= save_spcont;
|
2005-12-07 15:01:17 +01:00
|
|
|
DBUG_RETURN(TRUE);
|
2005-08-22 00:13:37 +02:00
|
|
|
}
|
2005-11-22 23:50:37 +01:00
|
|
|
#ifndef DBUG_OFF
|
2005-12-07 15:01:17 +01:00
|
|
|
nctx->sp= this;
|
2005-11-22 23:50:37 +01:00
|
|
|
#endif
|
Implement WL#2661 "Prepared Statements: Dynamic SQL in Stored Procedures".
The idea of the patch is to separate statement processing logic,
such as parsing, validation of the parsed tree, execution and cleanup,
from global query processing logic, such as logging, resetting
priorities of a thread, resetting stored procedure cache, resetting
thread count of errors and warnings.
This makes PREPARE and EXECUTE behave similarly to the rest of SQL
statements and allows their use in stored procedures.
This patch contains a change in behaviour:
until recently for each SQL prepared statement command, 2 queries
were written to the general log, e.g.
[Query] prepare stmt from @stmt_text;
[Prepare] select * from t1 <-- contents of @stmt_text
The chagne was necessary to prevent [Prepare] commands from being written
to the general log when executing a stored procedure with Dynamic SQL.
We should consider whether the old behavior is preferrable and probably
restore it.
This patch refixes Bug#7115, Bug#10975 (partially), Bug#10605 (various bugs
in Dynamic SQL reported before it was disabled).
2005-09-03 01:13:18 +02:00
|
|
|
|
2005-12-07 15:01:17 +01:00
|
|
|
if (params > 0)
|
Simplistic, experimental framework for Stored Procedures (SPs).
Implements creation and dropping of PROCEDUREs, IN, OUT, and INOUT parameters,
single-statement procedures, rudimentary multi-statement (begin-end) prodedures
(when the client can handle it), and local variables.
Missing most of the embedded SQL language, all attributes, FUNCTIONs, error handling,
reparses procedures at each call (no caching), etc, etc.
Certainly buggy too, but procedures can actually be created and called....
2002-12-08 19:59:22 +01:00
|
|
|
{
|
2005-12-07 15:01:17 +01:00
|
|
|
List_iterator<Item> it_args(*args);
|
Simplistic, experimental framework for Stored Procedures (SPs).
Implements creation and dropping of PROCEDUREs, IN, OUT, and INOUT parameters,
single-statement procedures, rudimentary multi-statement (begin-end) prodedures
(when the client can handle it), and local variables.
Missing most of the embedded SQL language, all attributes, FUNCTIONs, error handling,
reparses procedures at each call (no caching), etc, etc.
Certainly buggy too, but procedures can actually be created and called....
2002-12-08 19:59:22 +01:00
|
|
|
|
2005-07-30 10:19:57 +02:00
|
|
|
DBUG_PRINT("info",(" %.*s: eval args", m_name.length, m_name.str));
|
2005-12-07 15:01:17 +01:00
|
|
|
|
|
|
|
for (uint i= 0 ; i < params ; i++)
|
Simplistic, experimental framework for Stored Procedures (SPs).
Implements creation and dropping of PROCEDUREs, IN, OUT, and INOUT parameters,
single-statement procedures, rudimentary multi-statement (begin-end) prodedures
(when the client can handle it), and local variables.
Missing most of the embedded SQL language, all attributes, FUNCTIONs, error handling,
reparses procedures at each call (no caching), etc, etc.
Certainly buggy too, but procedures can actually be created and called....
2002-12-08 19:59:22 +01:00
|
|
|
{
|
2005-12-07 15:01:17 +01:00
|
|
|
Item *arg_item= it_args++;
|
Simplistic, experimental framework for Stored Procedures (SPs).
Implements creation and dropping of PROCEDUREs, IN, OUT, and INOUT parameters,
single-statement procedures, rudimentary multi-statement (begin-end) prodedures
(when the client can handle it), and local variables.
Missing most of the embedded SQL language, all attributes, FUNCTIONs, error handling,
reparses procedures at each call (no caching), etc, etc.
Certainly buggy too, but procedures can actually be created and called....
2002-12-08 19:59:22 +01:00
|
|
|
|
2005-12-07 15:01:17 +01:00
|
|
|
if (!arg_item)
|
|
|
|
break;
|
|
|
|
|
2006-05-12 11:55:21 +02:00
|
|
|
sp_variable_t *spvar= m_pcont->find_variable(i);
|
|
|
|
|
2006-04-07 16:53:15 +02:00
|
|
|
if (!spvar)
|
2005-12-07 15:01:17 +01:00
|
|
|
continue;
|
|
|
|
|
2006-04-07 16:53:15 +02:00
|
|
|
if (spvar->mode != sp_param_in)
|
2002-12-11 14:24:29 +01:00
|
|
|
{
|
2006-05-12 11:55:21 +02:00
|
|
|
Settable_routine_parameter *srp=
|
|
|
|
arg_item->get_settable_routine_parameter();
|
|
|
|
|
|
|
|
if (!srp)
|
2005-12-07 15:01:17 +01:00
|
|
|
{
|
|
|
|
my_error(ER_SP_NOT_VAR_ARG, MYF(0), i+1, m_qname.str);
|
|
|
|
err_status= TRUE;
|
|
|
|
break;
|
|
|
|
}
|
2006-05-12 11:55:21 +02:00
|
|
|
|
|
|
|
srp->set_required_privilege(spvar->mode == sp_param_inout);
|
2005-12-07 15:01:17 +01:00
|
|
|
}
|
|
|
|
|
2006-04-07 16:53:15 +02:00
|
|
|
if (spvar->mode == sp_param_out)
|
2005-12-07 15:01:17 +01:00
|
|
|
{
|
|
|
|
Item_null *null_item= new Item_null();
|
|
|
|
|
|
|
|
if (!null_item ||
|
2006-11-30 17:25:05 +01:00
|
|
|
nctx->set_variable(thd, i, (Item **)&null_item))
|
2005-12-07 15:01:17 +01:00
|
|
|
{
|
|
|
|
err_status= TRUE;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
else
|
|
|
|
{
|
2006-05-15 12:01:55 +02:00
|
|
|
if (nctx->set_variable(thd, i, it_args.ref()))
|
2005-12-07 15:01:17 +01:00
|
|
|
{
|
|
|
|
err_status= TRUE;
|
|
|
|
break;
|
|
|
|
}
|
2002-12-11 14:24:29 +01:00
|
|
|
}
|
Simplistic, experimental framework for Stored Procedures (SPs).
Implements creation and dropping of PROCEDUREs, IN, OUT, and INOUT parameters,
single-statement procedures, rudimentary multi-statement (begin-end) prodedures
(when the client can handle it), and local variables.
Missing most of the embedded SQL language, all attributes, FUNCTIONs, error handling,
reparses procedures at each call (no caching), etc, etc.
Certainly buggy too, but procedures can actually be created and called....
2002-12-08 19:59:22 +01:00
|
|
|
}
|
2003-04-23 21:31:47 +02:00
|
|
|
|
2005-07-30 10:19:57 +02:00
|
|
|
/*
|
|
|
|
Okay, got values for all arguments. Close tables that might be used by
|
2005-08-03 05:37:32 +02:00
|
|
|
arguments evaluation. If arguments evaluation required prelocking mode,
|
|
|
|
we'll leave it here.
|
2005-07-30 10:19:57 +02:00
|
|
|
*/
|
|
|
|
if (!thd->in_sub_stmt)
|
2005-08-09 10:56:55 +02:00
|
|
|
close_thread_tables(thd, 0, 0);
|
2005-07-30 10:19:57 +02:00
|
|
|
|
|
|
|
DBUG_PRINT("info",(" %.*s: eval args done", m_name.length, m_name.str));
|
Simplistic, experimental framework for Stored Procedures (SPs).
Implements creation and dropping of PROCEDUREs, IN, OUT, and INOUT parameters,
single-statement procedures, rudimentary multi-statement (begin-end) prodedures
(when the client can handle it), and local variables.
Missing most of the embedded SQL language, all attributes, FUNCTIONs, error handling,
reparses procedures at each call (no caching), etc, etc.
Certainly buggy too, but procedures can actually be created and called....
2002-12-08 19:59:22 +01:00
|
|
|
}
|
|
|
|
|
2005-08-18 11:23:54 +02:00
|
|
|
thd->spcont= nctx;
|
|
|
|
|
2006-07-13 15:12:31 +02:00
|
|
|
#ifndef NO_EMBEDDED_ACCESS_CHECKS
|
|
|
|
Security_context *save_security_ctx= 0;
|
|
|
|
if (!err_status)
|
|
|
|
err_status= set_routine_security_ctx(thd, this, TRUE, &save_security_ctx);
|
|
|
|
#endif
|
|
|
|
|
2005-12-07 15:01:17 +01:00
|
|
|
if (!err_status)
|
|
|
|
err_status= execute(thd);
|
2005-08-18 11:23:54 +02:00
|
|
|
|
|
|
|
/*
|
|
|
|
In the case when we weren't able to employ reuse mechanism for
|
|
|
|
OUT/INOUT paranmeters, we should reallocate memory. This
|
|
|
|
allocation should be done on the arena which will live through
|
|
|
|
all execution of calling routine.
|
|
|
|
*/
|
|
|
|
thd->spcont->callers_arena= octx->callers_arena;
|
2005-06-03 19:21:12 +02:00
|
|
|
|
2005-12-07 15:01:17 +01:00
|
|
|
if (!err_status && params > 0)
|
Simplistic, experimental framework for Stored Procedures (SPs).
Implements creation and dropping of PROCEDUREs, IN, OUT, and INOUT parameters,
single-statement procedures, rudimentary multi-statement (begin-end) prodedures
(when the client can handle it), and local variables.
Missing most of the embedded SQL language, all attributes, FUNCTIONs, error handling,
reparses procedures at each call (no caching), etc, etc.
Certainly buggy too, but procedures can actually be created and called....
2002-12-08 19:59:22 +01:00
|
|
|
{
|
2005-12-07 15:01:17 +01:00
|
|
|
List_iterator<Item> it_args(*args);
|
2003-02-02 17:44:39 +01:00
|
|
|
|
2005-08-11 14:58:15 +02:00
|
|
|
/*
|
|
|
|
Copy back all OUT or INOUT values to the previous frame, or
|
|
|
|
set global user variables
|
|
|
|
*/
|
2005-12-07 15:01:17 +01:00
|
|
|
for (uint i= 0 ; i < params ; i++)
|
Simplistic, experimental framework for Stored Procedures (SPs).
Implements creation and dropping of PROCEDUREs, IN, OUT, and INOUT parameters,
single-statement procedures, rudimentary multi-statement (begin-end) prodedures
(when the client can handle it), and local variables.
Missing most of the embedded SQL language, all attributes, FUNCTIONs, error handling,
reparses procedures at each call (no caching), etc, etc.
Certainly buggy too, but procedures can actually be created and called....
2002-12-08 19:59:22 +01:00
|
|
|
{
|
2005-12-07 15:01:17 +01:00
|
|
|
Item *arg_item= it_args++;
|
|
|
|
|
|
|
|
if (!arg_item)
|
|
|
|
break;
|
|
|
|
|
2006-04-07 16:53:15 +02:00
|
|
|
sp_variable_t *spvar= m_pcont->find_variable(i);
|
Simplistic, experimental framework for Stored Procedures (SPs).
Implements creation and dropping of PROCEDUREs, IN, OUT, and INOUT parameters,
single-statement procedures, rudimentary multi-statement (begin-end) prodedures
(when the client can handle it), and local variables.
Missing most of the embedded SQL language, all attributes, FUNCTIONs, error handling,
reparses procedures at each call (no caching), etc, etc.
Certainly buggy too, but procedures can actually be created and called....
2002-12-08 19:59:22 +01:00
|
|
|
|
2006-04-07 16:53:15 +02:00
|
|
|
if (spvar->mode == sp_param_in)
|
2005-12-07 15:01:17 +01:00
|
|
|
continue;
|
|
|
|
|
2006-05-12 11:55:21 +02:00
|
|
|
Settable_routine_parameter *srp=
|
|
|
|
arg_item->get_settable_routine_parameter();
|
|
|
|
|
|
|
|
DBUG_ASSERT(srp);
|
|
|
|
|
2006-05-15 19:57:10 +02:00
|
|
|
if (srp->set_value(thd, octx, nctx->get_item_addr(i)))
|
2005-12-07 15:01:17 +01:00
|
|
|
{
|
2006-05-12 11:55:21 +02:00
|
|
|
err_status= TRUE;
|
|
|
|
break;
|
2003-02-02 17:44:39 +01:00
|
|
|
}
|
Simplistic, experimental framework for Stored Procedures (SPs).
Implements creation and dropping of PROCEDUREs, IN, OUT, and INOUT parameters,
single-statement procedures, rudimentary multi-statement (begin-end) prodedures
(when the client can handle it), and local variables.
Missing most of the embedded SQL language, all attributes, FUNCTIONs, error handling,
reparses procedures at each call (no caching), etc, etc.
Certainly buggy too, but procedures can actually be created and called....
2002-12-08 19:59:22 +01:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2006-07-13 15:12:31 +02:00
|
|
|
#ifndef NO_EMBEDDED_ACCESS_CHECKS
|
|
|
|
if (save_security_ctx)
|
|
|
|
sp_restore_security_context(thd, save_security_ctx);
|
|
|
|
#endif
|
|
|
|
|
2005-08-22 00:13:37 +02:00
|
|
|
if (!save_spcont)
|
2005-12-07 15:01:17 +01:00
|
|
|
delete octx;
|
2003-10-10 16:57:21 +02:00
|
|
|
|
2005-12-07 15:01:17 +01:00
|
|
|
delete nctx;
|
2005-08-22 00:13:37 +02:00
|
|
|
thd->spcont= save_spcont;
|
Implement WL#2661 "Prepared Statements: Dynamic SQL in Stored Procedures".
The idea of the patch is to separate statement processing logic,
such as parsing, validation of the parsed tree, execution and cleanup,
from global query processing logic, such as logging, resetting
priorities of a thread, resetting stored procedure cache, resetting
thread count of errors and warnings.
This makes PREPARE and EXECUTE behave similarly to the rest of SQL
statements and allows their use in stored procedures.
This patch contains a change in behaviour:
until recently for each SQL prepared statement command, 2 queries
were written to the general log, e.g.
[Query] prepare stmt from @stmt_text;
[Prepare] select * from t1 <-- contents of @stmt_text
The chagne was necessary to prevent [Prepare] commands from being written
to the general log when executing a stored procedure with Dynamic SQL.
We should consider whether the old behavior is preferrable and probably
restore it.
This patch refixes Bug#7115, Bug#10975 (partially), Bug#10605 (various bugs
in Dynamic SQL reported before it was disabled).
2005-09-03 01:13:18 +02:00
|
|
|
|
2005-12-07 15:01:17 +01:00
|
|
|
DBUG_RETURN(err_status);
|
Simplistic, experimental framework for Stored Procedures (SPs).
Implements creation and dropping of PROCEDUREs, IN, OUT, and INOUT parameters,
single-statement procedures, rudimentary multi-statement (begin-end) prodedures
(when the client can handle it), and local variables.
Missing most of the embedded SQL language, all attributes, FUNCTIONs, error handling,
reparses procedures at each call (no caching), etc, etc.
Certainly buggy too, but procedures can actually be created and called....
2002-12-08 19:59:22 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
|
2002-12-12 13:14:23 +01:00
|
|
|
// Reset lex during parsing, before we parse a sub statement.
|
Simplistic, experimental framework for Stored Procedures (SPs).
Implements creation and dropping of PROCEDUREs, IN, OUT, and INOUT parameters,
single-statement procedures, rudimentary multi-statement (begin-end) prodedures
(when the client can handle it), and local variables.
Missing most of the embedded SQL language, all attributes, FUNCTIONs, error handling,
reparses procedures at each call (no caching), etc, etc.
Certainly buggy too, but procedures can actually be created and called....
2002-12-08 19:59:22 +01:00
|
|
|
void
|
|
|
|
sp_head::reset_lex(THD *thd)
|
|
|
|
{
|
2003-05-23 15:32:31 +02:00
|
|
|
DBUG_ENTER("sp_head::reset_lex");
|
|
|
|
LEX *sublex;
|
2003-06-29 18:15:17 +02:00
|
|
|
LEX *oldlex= thd->lex;
|
2003-05-23 15:32:31 +02:00
|
|
|
|
2003-06-29 18:15:17 +02:00
|
|
|
(void)m_lex.push_front(oldlex);
|
2003-05-23 15:32:31 +02:00
|
|
|
thd->lex= sublex= new st_lex;
|
2004-07-22 00:26:33 +02:00
|
|
|
|
Bug#25411 (trigger code truncated), PART I
The issue found with bug 25411 is due to the function skip_rear_comments()
which damages the source code while implementing a work around.
The root cause of the problem is in the lexical analyser, which does not
process special comments properly.
For special comments like :
[1] aaa /*!50000 bbb */ ccc
since 5.0 is a version older that the current code, the parser is in lining
the content of the special comment, so that the query to process is
[2] aaa bbb ccc
However, the text of the query captured when processing a stored procedure,
stored function or trigger (or event in 5.1), can be after rebuilding it:
[3] aaa bbb */ ccc
which is wrong.
To fix bug 25411 properly, the lexical analyser needs to return [2] when
in lining special comments.
In order to implement this, some preliminary cleanup is required in the code,
which is implemented by this patch.
Before this change, the structure named LEX (or st_lex) contains attributes
that belong to lexical analysis, as well as attributes that represents the
abstract syntax tree (AST) of a statement.
Creating a new LEX structure for each statements (which makes sense for the
AST part) also re-initialized the lexical analysis phase each time, which
is conceptually wrong.
With this patch, the previous st_lex structure has been split in two:
- st_lex represents the Abstract Syntax Tree for a statement. The name "lex"
has not been changed to avoid a bigger impact in the code base.
- class lex_input_stream represents the internal state of the lexical
analyser, which by definition should *not* be reinitialized when parsing
multiple statements from the same input stream.
This change is a pre-requisite for bug 25411, since the implementation of
lex_input_stream will later improve to deal properly with special comments,
and this processing can not be done with the current implementation of
sp_head::reset_lex and sp_head::restore_lex, which interfere with the lexer.
This change set alone does not fix bug 25411.
2007-04-24 17:24:21 +02:00
|
|
|
/* Reset most stuff. */
|
|
|
|
lex_start(thd);
|
2004-07-22 00:26:33 +02:00
|
|
|
|
Simplistic, experimental framework for Stored Procedures (SPs).
Implements creation and dropping of PROCEDUREs, IN, OUT, and INOUT parameters,
single-statement procedures, rudimentary multi-statement (begin-end) prodedures
(when the client can handle it), and local variables.
Missing most of the embedded SQL language, all attributes, FUNCTIONs, error handling,
reparses procedures at each call (no caching), etc, etc.
Certainly buggy too, but procedures can actually be created and called....
2002-12-08 19:59:22 +01:00
|
|
|
/* And keep the SP stuff too */
|
2003-06-29 18:15:17 +02:00
|
|
|
sublex->sphead= oldlex->sphead;
|
|
|
|
sublex->spcont= oldlex->spcont;
|
2004-09-07 14:29:46 +02:00
|
|
|
/* And trigger related stuff too */
|
|
|
|
sublex->trg_chistics= oldlex->trg_chistics;
|
2004-11-24 10:24:02 +01:00
|
|
|
sublex->trg_table_fields.empty();
|
2003-06-29 18:15:17 +02:00
|
|
|
sublex->sp_lex_in_use= FALSE;
|
2005-12-07 15:01:17 +01:00
|
|
|
|
2006-02-09 11:05:28 +01:00
|
|
|
sublex->in_comment= oldlex->in_comment;
|
|
|
|
|
2005-12-07 15:01:17 +01:00
|
|
|
/* Reset type info. */
|
|
|
|
|
|
|
|
sublex->charset= NULL;
|
|
|
|
sublex->length= NULL;
|
|
|
|
sublex->dec= NULL;
|
|
|
|
sublex->interval_list.empty();
|
|
|
|
sublex->type= 0;
|
|
|
|
|
2003-05-23 15:32:31 +02:00
|
|
|
DBUG_VOID_RETURN;
|
Simplistic, experimental framework for Stored Procedures (SPs).
Implements creation and dropping of PROCEDUREs, IN, OUT, and INOUT parameters,
single-statement procedures, rudimentary multi-statement (begin-end) prodedures
(when the client can handle it), and local variables.
Missing most of the embedded SQL language, all attributes, FUNCTIONs, error handling,
reparses procedures at each call (no caching), etc, etc.
Certainly buggy too, but procedures can actually be created and called....
2002-12-08 19:59:22 +01:00
|
|
|
}
|
|
|
|
|
2002-12-12 13:14:23 +01:00
|
|
|
// Restore lex during parsing, after we have parsed a sub statement.
|
Simplistic, experimental framework for Stored Procedures (SPs).
Implements creation and dropping of PROCEDUREs, IN, OUT, and INOUT parameters,
single-statement procedures, rudimentary multi-statement (begin-end) prodedures
(when the client can handle it), and local variables.
Missing most of the embedded SQL language, all attributes, FUNCTIONs, error handling,
reparses procedures at each call (no caching), etc, etc.
Certainly buggy too, but procedures can actually be created and called....
2002-12-08 19:59:22 +01:00
|
|
|
void
|
|
|
|
sp_head::restore_lex(THD *thd)
|
|
|
|
{
|
2003-05-23 15:32:31 +02:00
|
|
|
DBUG_ENTER("sp_head::restore_lex");
|
|
|
|
LEX *sublex= thd->lex;
|
A fix and a test case for Bug#26141 mixing table types in trigger
causes full table lock on innodb table.
Also fixes Bug#28502 Triggers that update another innodb table
will block on X lock unnecessarily (duplciate).
Code review fixes.
Both bugs' synopses are misleading: InnoDB table is
not X locked. The statements, however, cannot proceed concurrently,
but this happens due to lock conflicts for tables used in triggers,
not for the InnoDB table.
If a user had an InnoDB table, and two triggers, AFTER UPDATE and
AFTER INSERT, competing for different resources (e.g. two distinct
MyISAM tables), then these two triggers would not be able to execute
concurrently. Moreover, INSERTS/UPDATES of the InnoDB table would
not be able to run concurrently.
The problem had other side-effects (see respective bug reports).
This behavior was a consequence of a shortcoming of the pre-locking
algorithm, which would not distinguish between different DML operations
(e.g. INSERT and DELETE) and pre-lock all the tables
that are used by any trigger defined on the subject table.
The idea of the fix is to extend the pre-locking algorithm to keep track,
for each table, what DML operation it is used for and not
load triggers that are known to never be fired.
2007-07-12 20:26:41 +02:00
|
|
|
LEX *oldlex;
|
|
|
|
|
|
|
|
sublex->set_trg_event_type_for_tables();
|
2003-06-29 18:15:17 +02:00
|
|
|
|
A fix and a test case for Bug#26141 mixing table types in trigger
causes full table lock on innodb table.
Also fixes Bug#28502 Triggers that update another innodb table
will block on X lock unnecessarily (duplciate).
Code review fixes.
Both bugs' synopses are misleading: InnoDB table is
not X locked. The statements, however, cannot proceed concurrently,
but this happens due to lock conflicts for tables used in triggers,
not for the InnoDB table.
If a user had an InnoDB table, and two triggers, AFTER UPDATE and
AFTER INSERT, competing for different resources (e.g. two distinct
MyISAM tables), then these two triggers would not be able to execute
concurrently. Moreover, INSERTS/UPDATES of the InnoDB table would
not be able to run concurrently.
The problem had other side-effects (see respective bug reports).
This behavior was a consequence of a shortcoming of the pre-locking
algorithm, which would not distinguish between different DML operations
(e.g. INSERT and DELETE) and pre-lock all the tables
that are used by any trigger defined on the subject table.
The idea of the fix is to extend the pre-locking algorithm to keep track,
for each table, what DML operation it is used for and not
load triggers that are known to never be fired.
2007-07-12 20:26:41 +02:00
|
|
|
oldlex= (LEX *)m_lex.pop();
|
2003-06-29 18:15:17 +02:00
|
|
|
if (! oldlex)
|
|
|
|
return; // Nothing to restore
|
2003-05-23 15:32:31 +02:00
|
|
|
|
2004-11-24 10:24:02 +01:00
|
|
|
oldlex->trg_table_fields.push_back(&sublex->trg_table_fields);
|
2002-12-13 18:25:36 +01:00
|
|
|
|
2005-03-04 14:35:28 +01:00
|
|
|
/*
|
2005-07-09 19:51:59 +02:00
|
|
|
Add routines which are used by statement to respective set for
|
|
|
|
this routine.
|
2005-03-04 14:35:28 +01:00
|
|
|
*/
|
2005-07-09 19:51:59 +02:00
|
|
|
sp_update_sp_used_routines(&m_sroutines, &sublex->sroutines);
|
2005-03-04 14:35:28 +01:00
|
|
|
/*
|
|
|
|
Merge tables used by this statement (but not by its functions or
|
|
|
|
procedures) to multiset of tables used by this routine.
|
|
|
|
*/
|
|
|
|
merge_table_list(thd, sublex->query_tables, sublex);
|
2003-06-29 18:15:17 +02:00
|
|
|
if (! sublex->sp_lex_in_use)
|
2006-05-04 14:30:38 +02:00
|
|
|
{
|
|
|
|
lex_end(sublex);
|
2003-06-29 18:15:17 +02:00
|
|
|
delete sublex;
|
2006-05-04 14:30:38 +02:00
|
|
|
}
|
2003-06-29 18:15:17 +02:00
|
|
|
thd->lex= oldlex;
|
2003-05-23 15:32:31 +02:00
|
|
|
DBUG_VOID_RETURN;
|
Simplistic, experimental framework for Stored Procedures (SPs).
Implements creation and dropping of PROCEDUREs, IN, OUT, and INOUT parameters,
single-statement procedures, rudimentary multi-statement (begin-end) prodedures
(when the client can handle it), and local variables.
Missing most of the embedded SQL language, all attributes, FUNCTIONs, error handling,
reparses procedures at each call (no caching), etc, etc.
Certainly buggy too, but procedures can actually be created and called....
2002-12-08 19:59:22 +01:00
|
|
|
}
|
|
|
|
|
2002-12-11 14:24:29 +01:00
|
|
|
void
|
2002-12-16 15:40:44 +01:00
|
|
|
sp_head::push_backpatch(sp_instr *i, sp_label_t *lab)
|
2002-12-11 14:24:29 +01:00
|
|
|
{
|
2003-04-02 20:42:28 +02:00
|
|
|
bp_t *bp= (bp_t *)sql_alloc(sizeof(bp_t));
|
2002-12-16 15:40:44 +01:00
|
|
|
|
|
|
|
if (bp)
|
|
|
|
{
|
|
|
|
bp->lab= lab;
|
|
|
|
bp->instr= i;
|
|
|
|
(void)m_backpatch.push_front(bp);
|
|
|
|
}
|
2002-12-11 14:24:29 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
void
|
2002-12-16 15:40:44 +01:00
|
|
|
sp_head::backpatch(sp_label_t *lab)
|
Simplistic, experimental framework for Stored Procedures (SPs).
Implements creation and dropping of PROCEDUREs, IN, OUT, and INOUT parameters,
single-statement procedures, rudimentary multi-statement (begin-end) prodedures
(when the client can handle it), and local variables.
Missing most of the embedded SQL language, all attributes, FUNCTIONs, error handling,
reparses procedures at each call (no caching), etc, etc.
Certainly buggy too, but procedures can actually be created and called....
2002-12-08 19:59:22 +01:00
|
|
|
{
|
2002-12-16 15:40:44 +01:00
|
|
|
bp_t *bp;
|
2002-12-12 13:14:23 +01:00
|
|
|
uint dest= instructions();
|
2002-12-16 15:40:44 +01:00
|
|
|
List_iterator_fast<bp_t> li(m_backpatch);
|
Simplistic, experimental framework for Stored Procedures (SPs).
Implements creation and dropping of PROCEDUREs, IN, OUT, and INOUT parameters,
single-statement procedures, rudimentary multi-statement (begin-end) prodedures
(when the client can handle it), and local variables.
Missing most of the embedded SQL language, all attributes, FUNCTIONs, error handling,
reparses procedures at each call (no caching), etc, etc.
Certainly buggy too, but procedures can actually be created and called....
2002-12-08 19:59:22 +01:00
|
|
|
|
2002-12-16 15:40:44 +01:00
|
|
|
while ((bp= li++))
|
2004-08-17 20:20:58 +02:00
|
|
|
{
|
2006-04-18 11:07:34 +02:00
|
|
|
if (bp->lab == lab)
|
|
|
|
bp->instr->backpatch(dest, lab->ctx);
|
2004-08-17 20:20:58 +02:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2005-12-07 15:01:17 +01:00
|
|
|
/*
|
|
|
|
Prepare an instance of create_field for field creation (fill all necessary
|
|
|
|
attributes).
|
|
|
|
|
|
|
|
SYNOPSIS
|
|
|
|
sp_head::fill_field_definition()
|
|
|
|
thd [IN] Thread handle
|
|
|
|
lex [IN] Yacc parsing context
|
|
|
|
field_type [IN] Field type
|
|
|
|
field_def [OUT] An instance of create_field to be filled
|
|
|
|
|
|
|
|
RETURN
|
|
|
|
FALSE on success
|
|
|
|
TRUE on error
|
|
|
|
*/
|
|
|
|
|
|
|
|
bool
|
|
|
|
sp_head::fill_field_definition(THD *thd, LEX *lex,
|
|
|
|
enum enum_field_types field_type,
|
|
|
|
create_field *field_def)
|
|
|
|
{
|
2006-07-27 15:57:43 +02:00
|
|
|
HA_CREATE_INFO sp_db_info;
|
2005-12-07 15:01:17 +01:00
|
|
|
LEX_STRING cmt = { 0, 0 };
|
|
|
|
uint unused1= 0;
|
|
|
|
int unused2= 0;
|
|
|
|
|
2006-07-27 15:57:43 +02:00
|
|
|
load_db_opt_by_name(thd, m_db.str, &sp_db_info);
|
|
|
|
|
2005-12-07 15:01:17 +01:00
|
|
|
if (field_def->init(thd, (char*) "", field_type, lex->length, lex->dec,
|
|
|
|
lex->type, (Item*) 0, (Item*) 0, &cmt, 0,
|
|
|
|
&lex->interval_list,
|
2006-07-27 15:57:43 +02:00
|
|
|
(lex->charset ? lex->charset :
|
|
|
|
sp_db_info.default_table_charset),
|
2005-12-07 15:01:17 +01:00
|
|
|
lex->uint_geom_type))
|
|
|
|
return TRUE;
|
|
|
|
|
|
|
|
if (field_def->interval_list.elements)
|
|
|
|
field_def->interval= create_typelib(mem_root, field_def,
|
|
|
|
&field_def->interval_list);
|
|
|
|
|
|
|
|
sp_prepare_create_field(thd, field_def);
|
|
|
|
|
|
|
|
if (prepare_create_field(field_def, &unused1, &unused2, &unused2,
|
|
|
|
HA_CAN_GEOMETRY))
|
|
|
|
{
|
|
|
|
return TRUE;
|
|
|
|
}
|
|
|
|
|
|
|
|
return FALSE;
|
|
|
|
}
|
|
|
|
|
|
|
|
|
2005-11-04 15:37:39 +01:00
|
|
|
void
|
2006-01-26 17:26:25 +01:00
|
|
|
sp_head::new_cont_backpatch(sp_instr_opt_meta *i)
|
2005-11-04 15:37:39 +01:00
|
|
|
{
|
|
|
|
m_cont_level+= 1;
|
|
|
|
if (i)
|
|
|
|
{
|
|
|
|
/* Use the cont. destination slot to store the level */
|
|
|
|
i->m_cont_dest= m_cont_level;
|
|
|
|
(void)m_cont_backpatch.push_front(i);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
void
|
2006-01-26 17:26:25 +01:00
|
|
|
sp_head::add_cont_backpatch(sp_instr_opt_meta *i)
|
2005-11-04 15:37:39 +01:00
|
|
|
{
|
|
|
|
i->m_cont_dest= m_cont_level;
|
|
|
|
(void)m_cont_backpatch.push_front(i);
|
|
|
|
}
|
|
|
|
|
|
|
|
void
|
|
|
|
sp_head::do_cont_backpatch()
|
|
|
|
{
|
|
|
|
uint dest= instructions();
|
|
|
|
uint lev= m_cont_level--;
|
2006-01-26 17:26:25 +01:00
|
|
|
sp_instr_opt_meta *i;
|
2005-11-04 15:37:39 +01:00
|
|
|
|
|
|
|
while ((i= m_cont_backpatch.head()) && i->m_cont_dest == lev)
|
|
|
|
{
|
|
|
|
i->m_cont_dest= dest;
|
|
|
|
(void)m_cont_backpatch.pop();
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2003-12-13 16:40:52 +01:00
|
|
|
void
|
2005-11-10 20:25:03 +01:00
|
|
|
sp_head::set_info(longlong created, longlong modified,
|
2004-06-09 14:19:43 +02:00
|
|
|
st_sp_chistics *chistics, ulong sql_mode)
|
2003-12-13 16:40:52 +01:00
|
|
|
{
|
|
|
|
m_created= created;
|
|
|
|
m_modified= modified;
|
2004-11-09 02:58:44 +01:00
|
|
|
m_chistics= (st_sp_chistics *) memdup_root(mem_root, (char*) chistics,
|
|
|
|
sizeof(*chistics));
|
2003-12-13 16:40:52 +01:00
|
|
|
if (m_chistics->comment.length == 0)
|
|
|
|
m_chistics->comment.str= 0;
|
|
|
|
else
|
2004-11-09 02:58:44 +01:00
|
|
|
m_chistics->comment.str= strmake_root(mem_root,
|
2003-12-13 16:40:52 +01:00
|
|
|
m_chistics->comment.str,
|
|
|
|
m_chistics->comment.length);
|
2004-06-09 14:19:43 +02:00
|
|
|
m_sql_mode= sql_mode;
|
2003-12-13 16:40:52 +01:00
|
|
|
}
|
|
|
|
|
2005-11-10 20:25:03 +01:00
|
|
|
|
|
|
|
void
|
2006-01-05 23:47:49 +01:00
|
|
|
sp_head::set_definer(const char *definer, uint definerlen)
|
2005-11-10 20:25:03 +01:00
|
|
|
{
|
2006-09-27 16:21:29 +02:00
|
|
|
char user_name_holder[USERNAME_LENGTH + 1];
|
|
|
|
LEX_STRING_WITH_INIT user_name(user_name_holder, USERNAME_LENGTH);
|
2005-11-10 20:25:03 +01:00
|
|
|
|
2006-03-02 13:18:49 +01:00
|
|
|
char host_name_holder[HOSTNAME_LENGTH + 1];
|
|
|
|
LEX_STRING_WITH_INIT host_name(host_name_holder, HOSTNAME_LENGTH);
|
2005-11-10 20:25:03 +01:00
|
|
|
|
2006-03-02 13:18:49 +01:00
|
|
|
parse_user(definer, definerlen, user_name.str, &user_name.length,
|
|
|
|
host_name.str, &host_name.length);
|
2005-11-10 20:25:03 +01:00
|
|
|
|
2006-03-02 13:18:49 +01:00
|
|
|
set_definer(&user_name, &host_name);
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
void
|
|
|
|
sp_head::set_definer(const LEX_STRING *user_name, const LEX_STRING *host_name)
|
|
|
|
{
|
|
|
|
m_definer_user.str= strmake_root(mem_root, user_name->str, user_name->length);
|
|
|
|
m_definer_user.length= user_name->length;
|
|
|
|
|
|
|
|
m_definer_host.str= strmake_root(mem_root, host_name->str, host_name->length);
|
|
|
|
m_definer_host.length= host_name->length;
|
2005-11-10 20:25:03 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
|
2004-03-11 17:18:59 +01:00
|
|
|
void
|
|
|
|
sp_head::reset_thd_mem_root(THD *thd)
|
|
|
|
{
|
2004-05-20 01:02:49 +02:00
|
|
|
DBUG_ENTER("sp_head::reset_thd_mem_root");
|
2004-03-11 17:18:59 +01:00
|
|
|
m_thd_root= thd->mem_root;
|
2004-11-09 02:58:44 +01:00
|
|
|
thd->mem_root= &main_mem_root;
|
2004-05-20 01:02:49 +02:00
|
|
|
DBUG_PRINT("info", ("mem_root 0x%lx moved to thd mem root 0x%lx",
|
|
|
|
(ulong) &mem_root, (ulong) &thd->mem_root));
|
|
|
|
free_list= thd->free_list; // Keep the old list
|
2004-03-11 17:18:59 +01:00
|
|
|
thd->free_list= NULL; // Start a new one
|
|
|
|
m_thd= thd;
|
2004-05-20 01:02:49 +02:00
|
|
|
DBUG_VOID_RETURN;
|
2004-03-11 17:18:59 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
void
|
|
|
|
sp_head::restore_thd_mem_root(THD *thd)
|
|
|
|
{
|
2004-05-20 01:02:49 +02:00
|
|
|
DBUG_ENTER("sp_head::restore_thd_mem_root");
|
|
|
|
Item *flist= free_list; // The old list
|
2005-09-02 15:21:19 +02:00
|
|
|
set_query_arena(thd); // Get new free_list and mem_root
|
2005-06-07 12:11:36 +02:00
|
|
|
state= INITIALIZED_FOR_SP;
|
2004-09-09 05:59:26 +02:00
|
|
|
|
2004-05-20 01:02:49 +02:00
|
|
|
DBUG_PRINT("info", ("mem_root 0x%lx returned from thd mem root 0x%lx",
|
|
|
|
(ulong) &mem_root, (ulong) &thd->mem_root));
|
2004-03-11 17:18:59 +01:00
|
|
|
thd->free_list= flist; // Restore the old one
|
|
|
|
thd->mem_root= m_thd_root;
|
|
|
|
m_thd= NULL;
|
2004-05-20 01:02:49 +02:00
|
|
|
DBUG_VOID_RETURN;
|
2004-03-11 17:18:59 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
|
2005-03-15 15:07:28 +01:00
|
|
|
/*
|
|
|
|
Check if a user has access right to a routine
|
|
|
|
|
|
|
|
SYNOPSIS
|
|
|
|
check_show_routine_access()
|
|
|
|
thd Thread handler
|
|
|
|
sp SP
|
|
|
|
full_access Set to 1 if the user has SELECT right to the
|
|
|
|
'mysql.proc' able or is the owner of the routine
|
|
|
|
RETURN
|
|
|
|
0 ok
|
|
|
|
1 error
|
|
|
|
*/
|
|
|
|
|
|
|
|
bool check_show_routine_access(THD *thd, sp_head *sp, bool *full_access)
|
2005-03-05 12:35:32 +01:00
|
|
|
{
|
|
|
|
TABLE_LIST tables;
|
|
|
|
bzero((char*) &tables,sizeof(tables));
|
|
|
|
tables.db= (char*) "mysql";
|
|
|
|
tables.table_name= tables.alias= (char*) "proc";
|
2005-03-15 15:07:28 +01:00
|
|
|
*full_access= (!check_table_access(thd, SELECT_ACL, &tables, 1) ||
|
2005-09-15 21:29:07 +02:00
|
|
|
(!strcmp(sp->m_definer_user.str,
|
|
|
|
thd->security_ctx->priv_user) &&
|
|
|
|
!strcmp(sp->m_definer_host.str,
|
|
|
|
thd->security_ctx->priv_host)));
|
2005-03-15 15:07:28 +01:00
|
|
|
if (!*full_access)
|
2005-05-17 20:54:20 +02:00
|
|
|
return check_some_routine_access(thd, sp->m_db.str, sp->m_name.str,
|
|
|
|
sp->m_type == TYPE_ENUM_PROCEDURE);
|
2005-03-05 12:35:32 +01:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
|
2003-11-17 18:21:36 +01:00
|
|
|
int
|
|
|
|
sp_head::show_create_procedure(THD *thd)
|
|
|
|
{
|
|
|
|
Protocol *protocol= thd->protocol;
|
|
|
|
char buff[2048];
|
|
|
|
String buffer(buff, sizeof(buff), system_charset_info);
|
|
|
|
int res;
|
|
|
|
List<Item> field_list;
|
2004-06-09 14:19:43 +02:00
|
|
|
byte *sql_mode_str;
|
|
|
|
ulong sql_mode_len;
|
2005-03-05 12:35:32 +01:00
|
|
|
bool full_access;
|
2003-11-17 18:21:36 +01:00
|
|
|
DBUG_ENTER("sp_head::show_create_procedure");
|
|
|
|
DBUG_PRINT("info", ("procedure %s", m_name.str));
|
2006-01-05 23:47:49 +01:00
|
|
|
|
2004-07-21 14:53:09 +02:00
|
|
|
LINT_INIT(sql_mode_str);
|
|
|
|
LINT_INIT(sql_mode_len);
|
2005-03-05 12:35:32 +01:00
|
|
|
|
2005-03-15 15:07:28 +01:00
|
|
|
if (check_show_routine_access(thd, this, &full_access))
|
2005-11-17 11:11:48 +01:00
|
|
|
DBUG_RETURN(1);
|
2004-06-09 14:19:43 +02:00
|
|
|
|
2005-07-28 21:39:11 +02:00
|
|
|
sql_mode_str=
|
|
|
|
sys_var_thd_sql_mode::symbolic_mode_representation(thd,
|
|
|
|
m_sql_mode,
|
|
|
|
&sql_mode_len);
|
2004-06-09 14:19:43 +02:00
|
|
|
field_list.push_back(new Item_empty_string("Procedure", NAME_LEN));
|
2005-07-28 21:39:11 +02:00
|
|
|
field_list.push_back(new Item_empty_string("sql_mode", sql_mode_len));
|
2003-11-17 18:21:36 +01:00
|
|
|
// 1024 is for not to confuse old clients
|
2006-06-29 22:21:55 +02:00
|
|
|
Item_empty_string *definition=
|
|
|
|
new Item_empty_string("Create Procedure", max(buffer.length(),1024));
|
|
|
|
definition->maybe_null= TRUE;
|
|
|
|
field_list.push_back(definition);
|
|
|
|
|
2004-08-03 12:32:21 +02:00
|
|
|
if (protocol->send_fields(&field_list, Protocol::SEND_NUM_ROWS |
|
|
|
|
Protocol::SEND_EOF))
|
2005-11-17 11:11:48 +01:00
|
|
|
DBUG_RETURN(1);
|
2003-11-17 18:21:36 +01:00
|
|
|
protocol->prepare_for_resend();
|
|
|
|
protocol->store(m_name.str, m_name.length, system_charset_info);
|
2005-07-28 21:39:11 +02:00
|
|
|
protocol->store((char*) sql_mode_str, sql_mode_len, system_charset_info);
|
2005-03-05 12:35:32 +01:00
|
|
|
if (full_access)
|
|
|
|
protocol->store(m_defstr.str, m_defstr.length, system_charset_info);
|
2006-06-29 22:21:55 +02:00
|
|
|
else
|
|
|
|
protocol->store_null();
|
2003-11-17 18:21:36 +01:00
|
|
|
res= protocol->write();
|
|
|
|
send_eof(thd);
|
2004-06-09 14:19:43 +02:00
|
|
|
|
2003-11-17 18:21:36 +01:00
|
|
|
DBUG_RETURN(res);
|
|
|
|
}
|
|
|
|
|
2004-05-26 13:28:35 +02:00
|
|
|
|
|
|
|
/*
|
|
|
|
Add instruction to SP
|
|
|
|
|
|
|
|
SYNOPSIS
|
|
|
|
sp_head::add_instr()
|
|
|
|
instr Instruction
|
|
|
|
*/
|
|
|
|
|
|
|
|
void sp_head::add_instr(sp_instr *instr)
|
|
|
|
{
|
|
|
|
instr->free_list= m_thd->free_list;
|
|
|
|
m_thd->free_list= 0;
|
2005-06-23 18:22:08 +02:00
|
|
|
/*
|
|
|
|
Memory root of every instruction is designated for permanent
|
|
|
|
transformations (optimizations) made on the parsed tree during
|
|
|
|
the first execution. It points to the memory root of the
|
|
|
|
entire stored procedure, as their life span is equal.
|
|
|
|
*/
|
|
|
|
instr->mem_root= &main_mem_root;
|
2004-05-26 13:28:35 +02:00
|
|
|
insert_dynamic(&m_instr, (gptr)&instr);
|
|
|
|
}
|
|
|
|
|
|
|
|
|
2003-11-17 18:21:36 +01:00
|
|
|
int
|
|
|
|
sp_head::show_create_function(THD *thd)
|
|
|
|
{
|
|
|
|
Protocol *protocol= thd->protocol;
|
|
|
|
char buff[2048];
|
|
|
|
String buffer(buff, sizeof(buff), system_charset_info);
|
|
|
|
int res;
|
|
|
|
List<Item> field_list;
|
2004-06-09 14:19:43 +02:00
|
|
|
byte *sql_mode_str;
|
|
|
|
ulong sql_mode_len;
|
2005-03-05 12:35:32 +01:00
|
|
|
bool full_access;
|
2003-11-17 18:21:36 +01:00
|
|
|
DBUG_ENTER("sp_head::show_create_function");
|
|
|
|
DBUG_PRINT("info", ("procedure %s", m_name.str));
|
2004-07-21 14:53:09 +02:00
|
|
|
LINT_INIT(sql_mode_str);
|
|
|
|
LINT_INIT(sql_mode_len);
|
Simplistic, experimental framework for Stored Procedures (SPs).
Implements creation and dropping of PROCEDUREs, IN, OUT, and INOUT parameters,
single-statement procedures, rudimentary multi-statement (begin-end) prodedures
(when the client can handle it), and local variables.
Missing most of the embedded SQL language, all attributes, FUNCTIONs, error handling,
reparses procedures at each call (no caching), etc, etc.
Certainly buggy too, but procedures can actually be created and called....
2002-12-08 19:59:22 +01:00
|
|
|
|
2005-03-15 15:07:28 +01:00
|
|
|
if (check_show_routine_access(thd, this, &full_access))
|
2005-11-17 11:11:48 +01:00
|
|
|
DBUG_RETURN(1);
|
2005-03-05 12:35:32 +01:00
|
|
|
|
2005-07-28 21:39:11 +02:00
|
|
|
sql_mode_str=
|
|
|
|
sys_var_thd_sql_mode::symbolic_mode_representation(thd,
|
|
|
|
m_sql_mode,
|
|
|
|
&sql_mode_len);
|
2003-11-17 18:21:36 +01:00
|
|
|
field_list.push_back(new Item_empty_string("Function",NAME_LEN));
|
2005-08-02 00:43:40 +02:00
|
|
|
field_list.push_back(new Item_empty_string("sql_mode", sql_mode_len));
|
2006-06-29 22:21:55 +02:00
|
|
|
Item_empty_string *definition=
|
|
|
|
new Item_empty_string("Create Function", max(buffer.length(),1024));
|
|
|
|
definition->maybe_null= TRUE;
|
|
|
|
field_list.push_back(definition);
|
|
|
|
|
2004-08-03 12:32:21 +02:00
|
|
|
if (protocol->send_fields(&field_list,
|
|
|
|
Protocol::SEND_NUM_ROWS | Protocol::SEND_EOF))
|
2005-11-17 11:11:48 +01:00
|
|
|
DBUG_RETURN(1);
|
2003-11-17 18:21:36 +01:00
|
|
|
protocol->prepare_for_resend();
|
|
|
|
protocol->store(m_name.str, m_name.length, system_charset_info);
|
2005-07-28 21:39:11 +02:00
|
|
|
protocol->store((char*) sql_mode_str, sql_mode_len, system_charset_info);
|
2005-03-05 12:35:32 +01:00
|
|
|
if (full_access)
|
|
|
|
protocol->store(m_defstr.str, m_defstr.length, system_charset_info);
|
2006-06-29 22:21:55 +02:00
|
|
|
else
|
|
|
|
protocol->store_null();
|
2003-11-17 18:21:36 +01:00
|
|
|
res= protocol->write();
|
|
|
|
send_eof(thd);
|
2004-06-09 14:19:43 +02:00
|
|
|
|
2003-11-17 18:21:36 +01:00
|
|
|
DBUG_RETURN(res);
|
|
|
|
}
|
2004-08-02 18:05:31 +02:00
|
|
|
|
2005-08-25 15:34:34 +02:00
|
|
|
|
|
|
|
/*
|
2005-11-04 15:37:39 +01:00
|
|
|
Do some minimal optimization of the code:
|
|
|
|
1) Mark used instructions
|
|
|
|
1.1) While doing this, shortcut jumps to jump instructions
|
|
|
|
2) Compact the code, removing unused instructions
|
2006-01-25 15:11:49 +01:00
|
|
|
|
|
|
|
This is the main mark and move loop; it relies on the following methods
|
|
|
|
in sp_instr and its subclasses:
|
|
|
|
|
Bug#19194 (Right recursion in parser for CASE causes excessive stack usage,
limitation)
Note to the reviewer
====================
Warning: reviewing this patch is somewhat involved.
Due to the nature of several issues all affecting the same area,
fixing separately each issue is not practical, since each fix can not be
implemented and tested independently.
In particular, the issues with
- rule recursion
- nested case statements
- forward jump resolution (backpatch list)
are tightly coupled (see below).
Definitions
===========
The expression
CASE expr
WHEN expr THEN expr
WHEN expr THEN expr
...
END
is a "Simple Case Expression".
The expression
CASE
WHEN expr THEN expr
WHEN expr THEN expr
...
END
is a "Searched Case Expression".
The statement
CASE expr
WHEN expr THEN stmts
WHEN expr THEN stmts
...
END CASE
is a "Simple Case Statement".
The statement
CASE
WHEN expr THEN stmts
WHEN expr THEN stmts
...
END CASE
is a "Searched Case Statement".
A "Left Recursive" rule is like
list:
element
| list element
;
A "Right Recursive" rule is like
list:
element
| element list
;
Left and right recursion produces the same language, the difference only
affects the *order* in which the text is parsed.
In a descendant parser (usually written manually), right recursion works
very well, and is typically implemented with a while loop.
In an ascendant parser (yacc/bison) left recursion works very well,
and is implemented naturally by the parser stack.
In both cases, using the wrong type or recursion is very bad and should be
avoided, as it causes technical issues with the parser implementation.
Before this change
==================
The "Simple Case Expression" and "Searched Case Expression" were both
implemented by the "when_list" and "when_list2" rules, which are left
recursive (ok).
These rules, however, used lex->when_list instead of using the parser stack,
which is more complex that necessary, and potentially dangerous because
of other rules using THD::reset_lex.
The "Simple Case Statement" and "Searched Case Statements" were implemented
by the "sp_case", "sp_whens" and in part by "sp_proc_stmt" rules.
Both cases were right recursive (bad).
The grammar involved was convoluted, and is assumed to be the results of
tweaks to get the code generation to work, but is not what someone would
naturally write.
In addition, using a common rule for both "Simple" and "Searched" case
statements was implemented with sp_head::m_flags |= IN_SIMPLE_CASE,
which is a flag and not a stack, and therefore does not take into account
*nested* case statements. This leads to incorrect generated code, and either
a server crash or an incorrect result.
With regards to the backpatch mechanism, a *different* backpatch list was
created for each jump from "WHEN expr THEN stmt" to "END CASE", which
relied on the grammar to be right recursive.
This is a mis-use of the backpatch list, since this list can resolve
multiple references to the same target at once.
The optimizer algorithm used to detect dead code in the "assembly" SQL
instructions, implemented by sp_head::opt_mark(uint ip), was recursive
in some cases (a conditional jump pointing forward to another conditional
jump).
In case of specially crafted code, like
- a long list of "IF expr THEN stmt END IF"
- a long CASE statement
this would actually cause a server crash with a stack overflow.
In general, having a stack that grows proportionally with user data (the
SQL code given by the client in a CREATE PROCEDURE) is to be avoided.
In debug builds only, creating a SP / SF / Trigger which had a significant
amount of code would spend --literally-- several minutes in sp_head::create,
because of the debug code involved with DBUG_PRINT("info", ("Code %s ...
There are several issues with this code:
- in a CASE with 5 000 WHEN, there are 15 000 instructions generated,
which create a sting representation of the code which is 500 000 bytes
long,
- using a String instead of an io stream causes performances to degrade
to a total server freeze, as time is spent doing realloc of a buffer
always too short,
- Printing a 500 000 long string in the debug log is too verbose,
- Generating this string even when DBUG_PRINT is off is useless,
- Having code that potentially can affect the server behavior, used with
#ifdef / #endif is useful in some cases, but is also a bad practice.
After this change
=================
"Case Expressions" (both simple and searched) have been simplified to
not use LEX::when_list, which has been removed.
Considering all the issues affecting case statements, the grammar for these
has been totally re written.
The existing actions, used to generate "assembly" sp_inst* code, have been
preserved but moved in the new grammar, with the following changes:
a) Bison rules are no longer shared between "Simple" and "Searched" case
statements, because a stack instead of a flag is required to handle them.
Nested statements are handled naturally by the parser stack, which by
definition uses the correct rule in the correct context.
Nested statements of the opposite type (simple vs searched) works correctly.
The flag sp_head::IN_SIMPLE_CASE is no longer used.
This is a step towards resolution of WL#2999, which correctly identified
that temporary parsing flags do not belong to sp_head.
The code in the action is shared by mean of the case_stmt_action_xxx()
helpers.
b) The backpatch mechanism, used to resolve forward jumps in the generated
code, has been changed to:
- create a label for the instruction following 'END CASE',
- register each jump at the end of a "WHEN expr THEN stmt" in a *unique*
backpatch list associated with the 'END CASE' label
- resolve all the forward jumps for this label at once.
In addition, the code involving backpatch has been commented, so that a
reader can now understand by reading matching "Registering" and "Resolving"
comments how the forward jumps are resolved and what target they resolve to,
as this is far from evident when reading the code alone.
The implementation of sp_head::opt_mark() has been revised to avoid
recursive calls from jump instructions, and instead add the jump location
to the list of paths to explore during the flow analysis of the instruction
graph, with a call to sp_head::add_mark_lead().
In addition, the flow analysis will stop if an instruction has already
been marked as reachable, which the previous code failed to do in the
recursive case.
sp_head::opt_mark() is now private, to prevent new calls to this method from
being introduced.
The debug code present in sp_head::create() has been removed.
Considering that SHOW PROCEDURE CODE is also available in debug builds,
and can be used anytime regardless of the trace level, as opposed to
"CREATE PROCEDURE" time and only if the trace was on,
removing the code actually makes debugging easier (usable trace).
Tests have been written to cover the parser overflow (big CASE),
and to cover nested CASE statements.
2006-11-17 20:14:29 +01:00
|
|
|
opt_mark() Mark instruction as reachable
|
2006-01-25 15:11:49 +01:00
|
|
|
opt_shortcut_jump() Shortcut jumps to the final destination;
|
|
|
|
used by opt_mark().
|
|
|
|
opt_move() Update moved instruction
|
|
|
|
set_destination() Set the new destination (jump instructions only)
|
2005-08-25 15:34:34 +02:00
|
|
|
*/
|
|
|
|
|
|
|
|
void sp_head::optimize()
|
2004-08-02 18:05:31 +02:00
|
|
|
{
|
|
|
|
List<sp_instr> bp;
|
|
|
|
sp_instr *i;
|
|
|
|
uint src, dst;
|
|
|
|
|
Bug#19194 (Right recursion in parser for CASE causes excessive stack usage,
limitation)
Note to the reviewer
====================
Warning: reviewing this patch is somewhat involved.
Due to the nature of several issues all affecting the same area,
fixing separately each issue is not practical, since each fix can not be
implemented and tested independently.
In particular, the issues with
- rule recursion
- nested case statements
- forward jump resolution (backpatch list)
are tightly coupled (see below).
Definitions
===========
The expression
CASE expr
WHEN expr THEN expr
WHEN expr THEN expr
...
END
is a "Simple Case Expression".
The expression
CASE
WHEN expr THEN expr
WHEN expr THEN expr
...
END
is a "Searched Case Expression".
The statement
CASE expr
WHEN expr THEN stmts
WHEN expr THEN stmts
...
END CASE
is a "Simple Case Statement".
The statement
CASE
WHEN expr THEN stmts
WHEN expr THEN stmts
...
END CASE
is a "Searched Case Statement".
A "Left Recursive" rule is like
list:
element
| list element
;
A "Right Recursive" rule is like
list:
element
| element list
;
Left and right recursion produces the same language, the difference only
affects the *order* in which the text is parsed.
In a descendant parser (usually written manually), right recursion works
very well, and is typically implemented with a while loop.
In an ascendant parser (yacc/bison) left recursion works very well,
and is implemented naturally by the parser stack.
In both cases, using the wrong type or recursion is very bad and should be
avoided, as it causes technical issues with the parser implementation.
Before this change
==================
The "Simple Case Expression" and "Searched Case Expression" were both
implemented by the "when_list" and "when_list2" rules, which are left
recursive (ok).
These rules, however, used lex->when_list instead of using the parser stack,
which is more complex that necessary, and potentially dangerous because
of other rules using THD::reset_lex.
The "Simple Case Statement" and "Searched Case Statements" were implemented
by the "sp_case", "sp_whens" and in part by "sp_proc_stmt" rules.
Both cases were right recursive (bad).
The grammar involved was convoluted, and is assumed to be the results of
tweaks to get the code generation to work, but is not what someone would
naturally write.
In addition, using a common rule for both "Simple" and "Searched" case
statements was implemented with sp_head::m_flags |= IN_SIMPLE_CASE,
which is a flag and not a stack, and therefore does not take into account
*nested* case statements. This leads to incorrect generated code, and either
a server crash or an incorrect result.
With regards to the backpatch mechanism, a *different* backpatch list was
created for each jump from "WHEN expr THEN stmt" to "END CASE", which
relied on the grammar to be right recursive.
This is a mis-use of the backpatch list, since this list can resolve
multiple references to the same target at once.
The optimizer algorithm used to detect dead code in the "assembly" SQL
instructions, implemented by sp_head::opt_mark(uint ip), was recursive
in some cases (a conditional jump pointing forward to another conditional
jump).
In case of specially crafted code, like
- a long list of "IF expr THEN stmt END IF"
- a long CASE statement
this would actually cause a server crash with a stack overflow.
In general, having a stack that grows proportionally with user data (the
SQL code given by the client in a CREATE PROCEDURE) is to be avoided.
In debug builds only, creating a SP / SF / Trigger which had a significant
amount of code would spend --literally-- several minutes in sp_head::create,
because of the debug code involved with DBUG_PRINT("info", ("Code %s ...
There are several issues with this code:
- in a CASE with 5 000 WHEN, there are 15 000 instructions generated,
which create a sting representation of the code which is 500 000 bytes
long,
- using a String instead of an io stream causes performances to degrade
to a total server freeze, as time is spent doing realloc of a buffer
always too short,
- Printing a 500 000 long string in the debug log is too verbose,
- Generating this string even when DBUG_PRINT is off is useless,
- Having code that potentially can affect the server behavior, used with
#ifdef / #endif is useful in some cases, but is also a bad practice.
After this change
=================
"Case Expressions" (both simple and searched) have been simplified to
not use LEX::when_list, which has been removed.
Considering all the issues affecting case statements, the grammar for these
has been totally re written.
The existing actions, used to generate "assembly" sp_inst* code, have been
preserved but moved in the new grammar, with the following changes:
a) Bison rules are no longer shared between "Simple" and "Searched" case
statements, because a stack instead of a flag is required to handle them.
Nested statements are handled naturally by the parser stack, which by
definition uses the correct rule in the correct context.
Nested statements of the opposite type (simple vs searched) works correctly.
The flag sp_head::IN_SIMPLE_CASE is no longer used.
This is a step towards resolution of WL#2999, which correctly identified
that temporary parsing flags do not belong to sp_head.
The code in the action is shared by mean of the case_stmt_action_xxx()
helpers.
b) The backpatch mechanism, used to resolve forward jumps in the generated
code, has been changed to:
- create a label for the instruction following 'END CASE',
- register each jump at the end of a "WHEN expr THEN stmt" in a *unique*
backpatch list associated with the 'END CASE' label
- resolve all the forward jumps for this label at once.
In addition, the code involving backpatch has been commented, so that a
reader can now understand by reading matching "Registering" and "Resolving"
comments how the forward jumps are resolved and what target they resolve to,
as this is far from evident when reading the code alone.
The implementation of sp_head::opt_mark() has been revised to avoid
recursive calls from jump instructions, and instead add the jump location
to the list of paths to explore during the flow analysis of the instruction
graph, with a call to sp_head::add_mark_lead().
In addition, the flow analysis will stop if an instruction has already
been marked as reachable, which the previous code failed to do in the
recursive case.
sp_head::opt_mark() is now private, to prevent new calls to this method from
being introduced.
The debug code present in sp_head::create() has been removed.
Considering that SHOW PROCEDURE CODE is also available in debug builds,
and can be used anytime regardless of the trace level, as opposed to
"CREATE PROCEDURE" time and only if the trace was on,
removing the code actually makes debugging easier (usable trace).
Tests have been written to cover the parser overflow (big CASE),
and to cover nested CASE statements.
2006-11-17 20:14:29 +01:00
|
|
|
opt_mark();
|
2004-08-02 18:05:31 +02:00
|
|
|
|
|
|
|
bp.empty();
|
|
|
|
src= dst= 0;
|
|
|
|
while ((i= get_instr(src)))
|
|
|
|
{
|
|
|
|
if (! i->marked)
|
|
|
|
{
|
|
|
|
delete i;
|
|
|
|
src+= 1;
|
|
|
|
}
|
|
|
|
else
|
|
|
|
{
|
|
|
|
if (src != dst)
|
2005-11-04 15:37:39 +01:00
|
|
|
{ // Move the instruction and update prev. jumps
|
2004-08-02 18:05:31 +02:00
|
|
|
sp_instr *ibp;
|
|
|
|
List_iterator_fast<sp_instr> li(bp);
|
|
|
|
|
|
|
|
set_dynamic(&m_instr, (gptr)&i, dst);
|
|
|
|
while ((ibp= li++))
|
2006-01-26 17:26:25 +01:00
|
|
|
{
|
|
|
|
sp_instr_opt_meta *im= static_cast<sp_instr_opt_meta *>(ibp);
|
|
|
|
im->set_destination(src, dst);
|
|
|
|
}
|
2004-08-02 18:05:31 +02:00
|
|
|
}
|
|
|
|
i->opt_move(dst, &bp);
|
|
|
|
src+= 1;
|
|
|
|
dst+= 1;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
m_instr.elements= dst;
|
|
|
|
bp.empty();
|
|
|
|
}
|
|
|
|
|
Bug#19194 (Right recursion in parser for CASE causes excessive stack usage,
limitation)
Note to the reviewer
====================
Warning: reviewing this patch is somewhat involved.
Due to the nature of several issues all affecting the same area,
fixing separately each issue is not practical, since each fix can not be
implemented and tested independently.
In particular, the issues with
- rule recursion
- nested case statements
- forward jump resolution (backpatch list)
are tightly coupled (see below).
Definitions
===========
The expression
CASE expr
WHEN expr THEN expr
WHEN expr THEN expr
...
END
is a "Simple Case Expression".
The expression
CASE
WHEN expr THEN expr
WHEN expr THEN expr
...
END
is a "Searched Case Expression".
The statement
CASE expr
WHEN expr THEN stmts
WHEN expr THEN stmts
...
END CASE
is a "Simple Case Statement".
The statement
CASE
WHEN expr THEN stmts
WHEN expr THEN stmts
...
END CASE
is a "Searched Case Statement".
A "Left Recursive" rule is like
list:
element
| list element
;
A "Right Recursive" rule is like
list:
element
| element list
;
Left and right recursion produces the same language, the difference only
affects the *order* in which the text is parsed.
In a descendant parser (usually written manually), right recursion works
very well, and is typically implemented with a while loop.
In an ascendant parser (yacc/bison) left recursion works very well,
and is implemented naturally by the parser stack.
In both cases, using the wrong type or recursion is very bad and should be
avoided, as it causes technical issues with the parser implementation.
Before this change
==================
The "Simple Case Expression" and "Searched Case Expression" were both
implemented by the "when_list" and "when_list2" rules, which are left
recursive (ok).
These rules, however, used lex->when_list instead of using the parser stack,
which is more complex that necessary, and potentially dangerous because
of other rules using THD::reset_lex.
The "Simple Case Statement" and "Searched Case Statements" were implemented
by the "sp_case", "sp_whens" and in part by "sp_proc_stmt" rules.
Both cases were right recursive (bad).
The grammar involved was convoluted, and is assumed to be the results of
tweaks to get the code generation to work, but is not what someone would
naturally write.
In addition, using a common rule for both "Simple" and "Searched" case
statements was implemented with sp_head::m_flags |= IN_SIMPLE_CASE,
which is a flag and not a stack, and therefore does not take into account
*nested* case statements. This leads to incorrect generated code, and either
a server crash or an incorrect result.
With regards to the backpatch mechanism, a *different* backpatch list was
created for each jump from "WHEN expr THEN stmt" to "END CASE", which
relied on the grammar to be right recursive.
This is a mis-use of the backpatch list, since this list can resolve
multiple references to the same target at once.
The optimizer algorithm used to detect dead code in the "assembly" SQL
instructions, implemented by sp_head::opt_mark(uint ip), was recursive
in some cases (a conditional jump pointing forward to another conditional
jump).
In case of specially crafted code, like
- a long list of "IF expr THEN stmt END IF"
- a long CASE statement
this would actually cause a server crash with a stack overflow.
In general, having a stack that grows proportionally with user data (the
SQL code given by the client in a CREATE PROCEDURE) is to be avoided.
In debug builds only, creating a SP / SF / Trigger which had a significant
amount of code would spend --literally-- several minutes in sp_head::create,
because of the debug code involved with DBUG_PRINT("info", ("Code %s ...
There are several issues with this code:
- in a CASE with 5 000 WHEN, there are 15 000 instructions generated,
which create a sting representation of the code which is 500 000 bytes
long,
- using a String instead of an io stream causes performances to degrade
to a total server freeze, as time is spent doing realloc of a buffer
always too short,
- Printing a 500 000 long string in the debug log is too verbose,
- Generating this string even when DBUG_PRINT is off is useless,
- Having code that potentially can affect the server behavior, used with
#ifdef / #endif is useful in some cases, but is also a bad practice.
After this change
=================
"Case Expressions" (both simple and searched) have been simplified to
not use LEX::when_list, which has been removed.
Considering all the issues affecting case statements, the grammar for these
has been totally re written.
The existing actions, used to generate "assembly" sp_inst* code, have been
preserved but moved in the new grammar, with the following changes:
a) Bison rules are no longer shared between "Simple" and "Searched" case
statements, because a stack instead of a flag is required to handle them.
Nested statements are handled naturally by the parser stack, which by
definition uses the correct rule in the correct context.
Nested statements of the opposite type (simple vs searched) works correctly.
The flag sp_head::IN_SIMPLE_CASE is no longer used.
This is a step towards resolution of WL#2999, which correctly identified
that temporary parsing flags do not belong to sp_head.
The code in the action is shared by mean of the case_stmt_action_xxx()
helpers.
b) The backpatch mechanism, used to resolve forward jumps in the generated
code, has been changed to:
- create a label for the instruction following 'END CASE',
- register each jump at the end of a "WHEN expr THEN stmt" in a *unique*
backpatch list associated with the 'END CASE' label
- resolve all the forward jumps for this label at once.
In addition, the code involving backpatch has been commented, so that a
reader can now understand by reading matching "Registering" and "Resolving"
comments how the forward jumps are resolved and what target they resolve to,
as this is far from evident when reading the code alone.
The implementation of sp_head::opt_mark() has been revised to avoid
recursive calls from jump instructions, and instead add the jump location
to the list of paths to explore during the flow analysis of the instruction
graph, with a call to sp_head::add_mark_lead().
In addition, the flow analysis will stop if an instruction has already
been marked as reachable, which the previous code failed to do in the
recursive case.
sp_head::opt_mark() is now private, to prevent new calls to this method from
being introduced.
The debug code present in sp_head::create() has been removed.
Considering that SHOW PROCEDURE CODE is also available in debug builds,
and can be used anytime regardless of the trace level, as opposed to
"CREATE PROCEDURE" time and only if the trace was on,
removing the code actually makes debugging easier (usable trace).
Tests have been written to cover the parser overflow (big CASE),
and to cover nested CASE statements.
2006-11-17 20:14:29 +01:00
|
|
|
void sp_head::add_mark_lead(uint ip, List<sp_instr> *leads)
|
|
|
|
{
|
|
|
|
sp_instr *i= get_instr(ip);
|
|
|
|
|
|
|
|
if (i && ! i->marked)
|
|
|
|
leads->push_front(i);
|
|
|
|
}
|
|
|
|
|
2004-08-02 18:05:31 +02:00
|
|
|
void
|
Bug#19194 (Right recursion in parser for CASE causes excessive stack usage,
limitation)
Note to the reviewer
====================
Warning: reviewing this patch is somewhat involved.
Due to the nature of several issues all affecting the same area,
fixing separately each issue is not practical, since each fix can not be
implemented and tested independently.
In particular, the issues with
- rule recursion
- nested case statements
- forward jump resolution (backpatch list)
are tightly coupled (see below).
Definitions
===========
The expression
CASE expr
WHEN expr THEN expr
WHEN expr THEN expr
...
END
is a "Simple Case Expression".
The expression
CASE
WHEN expr THEN expr
WHEN expr THEN expr
...
END
is a "Searched Case Expression".
The statement
CASE expr
WHEN expr THEN stmts
WHEN expr THEN stmts
...
END CASE
is a "Simple Case Statement".
The statement
CASE
WHEN expr THEN stmts
WHEN expr THEN stmts
...
END CASE
is a "Searched Case Statement".
A "Left Recursive" rule is like
list:
element
| list element
;
A "Right Recursive" rule is like
list:
element
| element list
;
Left and right recursion produces the same language, the difference only
affects the *order* in which the text is parsed.
In a descendant parser (usually written manually), right recursion works
very well, and is typically implemented with a while loop.
In an ascendant parser (yacc/bison) left recursion works very well,
and is implemented naturally by the parser stack.
In both cases, using the wrong type or recursion is very bad and should be
avoided, as it causes technical issues with the parser implementation.
Before this change
==================
The "Simple Case Expression" and "Searched Case Expression" were both
implemented by the "when_list" and "when_list2" rules, which are left
recursive (ok).
These rules, however, used lex->when_list instead of using the parser stack,
which is more complex that necessary, and potentially dangerous because
of other rules using THD::reset_lex.
The "Simple Case Statement" and "Searched Case Statements" were implemented
by the "sp_case", "sp_whens" and in part by "sp_proc_stmt" rules.
Both cases were right recursive (bad).
The grammar involved was convoluted, and is assumed to be the results of
tweaks to get the code generation to work, but is not what someone would
naturally write.
In addition, using a common rule for both "Simple" and "Searched" case
statements was implemented with sp_head::m_flags |= IN_SIMPLE_CASE,
which is a flag and not a stack, and therefore does not take into account
*nested* case statements. This leads to incorrect generated code, and either
a server crash or an incorrect result.
With regards to the backpatch mechanism, a *different* backpatch list was
created for each jump from "WHEN expr THEN stmt" to "END CASE", which
relied on the grammar to be right recursive.
This is a mis-use of the backpatch list, since this list can resolve
multiple references to the same target at once.
The optimizer algorithm used to detect dead code in the "assembly" SQL
instructions, implemented by sp_head::opt_mark(uint ip), was recursive
in some cases (a conditional jump pointing forward to another conditional
jump).
In case of specially crafted code, like
- a long list of "IF expr THEN stmt END IF"
- a long CASE statement
this would actually cause a server crash with a stack overflow.
In general, having a stack that grows proportionally with user data (the
SQL code given by the client in a CREATE PROCEDURE) is to be avoided.
In debug builds only, creating a SP / SF / Trigger which had a significant
amount of code would spend --literally-- several minutes in sp_head::create,
because of the debug code involved with DBUG_PRINT("info", ("Code %s ...
There are several issues with this code:
- in a CASE with 5 000 WHEN, there are 15 000 instructions generated,
which create a sting representation of the code which is 500 000 bytes
long,
- using a String instead of an io stream causes performances to degrade
to a total server freeze, as time is spent doing realloc of a buffer
always too short,
- Printing a 500 000 long string in the debug log is too verbose,
- Generating this string even when DBUG_PRINT is off is useless,
- Having code that potentially can affect the server behavior, used with
#ifdef / #endif is useful in some cases, but is also a bad practice.
After this change
=================
"Case Expressions" (both simple and searched) have been simplified to
not use LEX::when_list, which has been removed.
Considering all the issues affecting case statements, the grammar for these
has been totally re written.
The existing actions, used to generate "assembly" sp_inst* code, have been
preserved but moved in the new grammar, with the following changes:
a) Bison rules are no longer shared between "Simple" and "Searched" case
statements, because a stack instead of a flag is required to handle them.
Nested statements are handled naturally by the parser stack, which by
definition uses the correct rule in the correct context.
Nested statements of the opposite type (simple vs searched) works correctly.
The flag sp_head::IN_SIMPLE_CASE is no longer used.
This is a step towards resolution of WL#2999, which correctly identified
that temporary parsing flags do not belong to sp_head.
The code in the action is shared by mean of the case_stmt_action_xxx()
helpers.
b) The backpatch mechanism, used to resolve forward jumps in the generated
code, has been changed to:
- create a label for the instruction following 'END CASE',
- register each jump at the end of a "WHEN expr THEN stmt" in a *unique*
backpatch list associated with the 'END CASE' label
- resolve all the forward jumps for this label at once.
In addition, the code involving backpatch has been commented, so that a
reader can now understand by reading matching "Registering" and "Resolving"
comments how the forward jumps are resolved and what target they resolve to,
as this is far from evident when reading the code alone.
The implementation of sp_head::opt_mark() has been revised to avoid
recursive calls from jump instructions, and instead add the jump location
to the list of paths to explore during the flow analysis of the instruction
graph, with a call to sp_head::add_mark_lead().
In addition, the flow analysis will stop if an instruction has already
been marked as reachable, which the previous code failed to do in the
recursive case.
sp_head::opt_mark() is now private, to prevent new calls to this method from
being introduced.
The debug code present in sp_head::create() has been removed.
Considering that SHOW PROCEDURE CODE is also available in debug builds,
and can be used anytime regardless of the trace level, as opposed to
"CREATE PROCEDURE" time and only if the trace was on,
removing the code actually makes debugging easier (usable trace).
Tests have been written to cover the parser overflow (big CASE),
and to cover nested CASE statements.
2006-11-17 20:14:29 +01:00
|
|
|
sp_head::opt_mark()
|
2004-08-02 18:05:31 +02:00
|
|
|
{
|
Bug#19194 (Right recursion in parser for CASE causes excessive stack usage,
limitation)
Note to the reviewer
====================
Warning: reviewing this patch is somewhat involved.
Due to the nature of several issues all affecting the same area,
fixing separately each issue is not practical, since each fix can not be
implemented and tested independently.
In particular, the issues with
- rule recursion
- nested case statements
- forward jump resolution (backpatch list)
are tightly coupled (see below).
Definitions
===========
The expression
CASE expr
WHEN expr THEN expr
WHEN expr THEN expr
...
END
is a "Simple Case Expression".
The expression
CASE
WHEN expr THEN expr
WHEN expr THEN expr
...
END
is a "Searched Case Expression".
The statement
CASE expr
WHEN expr THEN stmts
WHEN expr THEN stmts
...
END CASE
is a "Simple Case Statement".
The statement
CASE
WHEN expr THEN stmts
WHEN expr THEN stmts
...
END CASE
is a "Searched Case Statement".
A "Left Recursive" rule is like
list:
element
| list element
;
A "Right Recursive" rule is like
list:
element
| element list
;
Left and right recursion produces the same language, the difference only
affects the *order* in which the text is parsed.
In a descendant parser (usually written manually), right recursion works
very well, and is typically implemented with a while loop.
In an ascendant parser (yacc/bison) left recursion works very well,
and is implemented naturally by the parser stack.
In both cases, using the wrong type or recursion is very bad and should be
avoided, as it causes technical issues with the parser implementation.
Before this change
==================
The "Simple Case Expression" and "Searched Case Expression" were both
implemented by the "when_list" and "when_list2" rules, which are left
recursive (ok).
These rules, however, used lex->when_list instead of using the parser stack,
which is more complex that necessary, and potentially dangerous because
of other rules using THD::reset_lex.
The "Simple Case Statement" and "Searched Case Statements" were implemented
by the "sp_case", "sp_whens" and in part by "sp_proc_stmt" rules.
Both cases were right recursive (bad).
The grammar involved was convoluted, and is assumed to be the results of
tweaks to get the code generation to work, but is not what someone would
naturally write.
In addition, using a common rule for both "Simple" and "Searched" case
statements was implemented with sp_head::m_flags |= IN_SIMPLE_CASE,
which is a flag and not a stack, and therefore does not take into account
*nested* case statements. This leads to incorrect generated code, and either
a server crash or an incorrect result.
With regards to the backpatch mechanism, a *different* backpatch list was
created for each jump from "WHEN expr THEN stmt" to "END CASE", which
relied on the grammar to be right recursive.
This is a mis-use of the backpatch list, since this list can resolve
multiple references to the same target at once.
The optimizer algorithm used to detect dead code in the "assembly" SQL
instructions, implemented by sp_head::opt_mark(uint ip), was recursive
in some cases (a conditional jump pointing forward to another conditional
jump).
In case of specially crafted code, like
- a long list of "IF expr THEN stmt END IF"
- a long CASE statement
this would actually cause a server crash with a stack overflow.
In general, having a stack that grows proportionally with user data (the
SQL code given by the client in a CREATE PROCEDURE) is to be avoided.
In debug builds only, creating a SP / SF / Trigger which had a significant
amount of code would spend --literally-- several minutes in sp_head::create,
because of the debug code involved with DBUG_PRINT("info", ("Code %s ...
There are several issues with this code:
- in a CASE with 5 000 WHEN, there are 15 000 instructions generated,
which create a sting representation of the code which is 500 000 bytes
long,
- using a String instead of an io stream causes performances to degrade
to a total server freeze, as time is spent doing realloc of a buffer
always too short,
- Printing a 500 000 long string in the debug log is too verbose,
- Generating this string even when DBUG_PRINT is off is useless,
- Having code that potentially can affect the server behavior, used with
#ifdef / #endif is useful in some cases, but is also a bad practice.
After this change
=================
"Case Expressions" (both simple and searched) have been simplified to
not use LEX::when_list, which has been removed.
Considering all the issues affecting case statements, the grammar for these
has been totally re written.
The existing actions, used to generate "assembly" sp_inst* code, have been
preserved but moved in the new grammar, with the following changes:
a) Bison rules are no longer shared between "Simple" and "Searched" case
statements, because a stack instead of a flag is required to handle them.
Nested statements are handled naturally by the parser stack, which by
definition uses the correct rule in the correct context.
Nested statements of the opposite type (simple vs searched) works correctly.
The flag sp_head::IN_SIMPLE_CASE is no longer used.
This is a step towards resolution of WL#2999, which correctly identified
that temporary parsing flags do not belong to sp_head.
The code in the action is shared by mean of the case_stmt_action_xxx()
helpers.
b) The backpatch mechanism, used to resolve forward jumps in the generated
code, has been changed to:
- create a label for the instruction following 'END CASE',
- register each jump at the end of a "WHEN expr THEN stmt" in a *unique*
backpatch list associated with the 'END CASE' label
- resolve all the forward jumps for this label at once.
In addition, the code involving backpatch has been commented, so that a
reader can now understand by reading matching "Registering" and "Resolving"
comments how the forward jumps are resolved and what target they resolve to,
as this is far from evident when reading the code alone.
The implementation of sp_head::opt_mark() has been revised to avoid
recursive calls from jump instructions, and instead add the jump location
to the list of paths to explore during the flow analysis of the instruction
graph, with a call to sp_head::add_mark_lead().
In addition, the flow analysis will stop if an instruction has already
been marked as reachable, which the previous code failed to do in the
recursive case.
sp_head::opt_mark() is now private, to prevent new calls to this method from
being introduced.
The debug code present in sp_head::create() has been removed.
Considering that SHOW PROCEDURE CODE is also available in debug builds,
and can be used anytime regardless of the trace level, as opposed to
"CREATE PROCEDURE" time and only if the trace was on,
removing the code actually makes debugging easier (usable trace).
Tests have been written to cover the parser overflow (big CASE),
and to cover nested CASE statements.
2006-11-17 20:14:29 +01:00
|
|
|
uint ip;
|
2004-08-02 18:05:31 +02:00
|
|
|
sp_instr *i;
|
Bug#19194 (Right recursion in parser for CASE causes excessive stack usage,
limitation)
Note to the reviewer
====================
Warning: reviewing this patch is somewhat involved.
Due to the nature of several issues all affecting the same area,
fixing separately each issue is not practical, since each fix can not be
implemented and tested independently.
In particular, the issues with
- rule recursion
- nested case statements
- forward jump resolution (backpatch list)
are tightly coupled (see below).
Definitions
===========
The expression
CASE expr
WHEN expr THEN expr
WHEN expr THEN expr
...
END
is a "Simple Case Expression".
The expression
CASE
WHEN expr THEN expr
WHEN expr THEN expr
...
END
is a "Searched Case Expression".
The statement
CASE expr
WHEN expr THEN stmts
WHEN expr THEN stmts
...
END CASE
is a "Simple Case Statement".
The statement
CASE
WHEN expr THEN stmts
WHEN expr THEN stmts
...
END CASE
is a "Searched Case Statement".
A "Left Recursive" rule is like
list:
element
| list element
;
A "Right Recursive" rule is like
list:
element
| element list
;
Left and right recursion produces the same language, the difference only
affects the *order* in which the text is parsed.
In a descendant parser (usually written manually), right recursion works
very well, and is typically implemented with a while loop.
In an ascendant parser (yacc/bison) left recursion works very well,
and is implemented naturally by the parser stack.
In both cases, using the wrong type or recursion is very bad and should be
avoided, as it causes technical issues with the parser implementation.
Before this change
==================
The "Simple Case Expression" and "Searched Case Expression" were both
implemented by the "when_list" and "when_list2" rules, which are left
recursive (ok).
These rules, however, used lex->when_list instead of using the parser stack,
which is more complex that necessary, and potentially dangerous because
of other rules using THD::reset_lex.
The "Simple Case Statement" and "Searched Case Statements" were implemented
by the "sp_case", "sp_whens" and in part by "sp_proc_stmt" rules.
Both cases were right recursive (bad).
The grammar involved was convoluted, and is assumed to be the results of
tweaks to get the code generation to work, but is not what someone would
naturally write.
In addition, using a common rule for both "Simple" and "Searched" case
statements was implemented with sp_head::m_flags |= IN_SIMPLE_CASE,
which is a flag and not a stack, and therefore does not take into account
*nested* case statements. This leads to incorrect generated code, and either
a server crash or an incorrect result.
With regards to the backpatch mechanism, a *different* backpatch list was
created for each jump from "WHEN expr THEN stmt" to "END CASE", which
relied on the grammar to be right recursive.
This is a mis-use of the backpatch list, since this list can resolve
multiple references to the same target at once.
The optimizer algorithm used to detect dead code in the "assembly" SQL
instructions, implemented by sp_head::opt_mark(uint ip), was recursive
in some cases (a conditional jump pointing forward to another conditional
jump).
In case of specially crafted code, like
- a long list of "IF expr THEN stmt END IF"
- a long CASE statement
this would actually cause a server crash with a stack overflow.
In general, having a stack that grows proportionally with user data (the
SQL code given by the client in a CREATE PROCEDURE) is to be avoided.
In debug builds only, creating a SP / SF / Trigger which had a significant
amount of code would spend --literally-- several minutes in sp_head::create,
because of the debug code involved with DBUG_PRINT("info", ("Code %s ...
There are several issues with this code:
- in a CASE with 5 000 WHEN, there are 15 000 instructions generated,
which create a sting representation of the code which is 500 000 bytes
long,
- using a String instead of an io stream causes performances to degrade
to a total server freeze, as time is spent doing realloc of a buffer
always too short,
- Printing a 500 000 long string in the debug log is too verbose,
- Generating this string even when DBUG_PRINT is off is useless,
- Having code that potentially can affect the server behavior, used with
#ifdef / #endif is useful in some cases, but is also a bad practice.
After this change
=================
"Case Expressions" (both simple and searched) have been simplified to
not use LEX::when_list, which has been removed.
Considering all the issues affecting case statements, the grammar for these
has been totally re written.
The existing actions, used to generate "assembly" sp_inst* code, have been
preserved but moved in the new grammar, with the following changes:
a) Bison rules are no longer shared between "Simple" and "Searched" case
statements, because a stack instead of a flag is required to handle them.
Nested statements are handled naturally by the parser stack, which by
definition uses the correct rule in the correct context.
Nested statements of the opposite type (simple vs searched) works correctly.
The flag sp_head::IN_SIMPLE_CASE is no longer used.
This is a step towards resolution of WL#2999, which correctly identified
that temporary parsing flags do not belong to sp_head.
The code in the action is shared by mean of the case_stmt_action_xxx()
helpers.
b) The backpatch mechanism, used to resolve forward jumps in the generated
code, has been changed to:
- create a label for the instruction following 'END CASE',
- register each jump at the end of a "WHEN expr THEN stmt" in a *unique*
backpatch list associated with the 'END CASE' label
- resolve all the forward jumps for this label at once.
In addition, the code involving backpatch has been commented, so that a
reader can now understand by reading matching "Registering" and "Resolving"
comments how the forward jumps are resolved and what target they resolve to,
as this is far from evident when reading the code alone.
The implementation of sp_head::opt_mark() has been revised to avoid
recursive calls from jump instructions, and instead add the jump location
to the list of paths to explore during the flow analysis of the instruction
graph, with a call to sp_head::add_mark_lead().
In addition, the flow analysis will stop if an instruction has already
been marked as reachable, which the previous code failed to do in the
recursive case.
sp_head::opt_mark() is now private, to prevent new calls to this method from
being introduced.
The debug code present in sp_head::create() has been removed.
Considering that SHOW PROCEDURE CODE is also available in debug builds,
and can be used anytime regardless of the trace level, as opposed to
"CREATE PROCEDURE" time and only if the trace was on,
removing the code actually makes debugging easier (usable trace).
Tests have been written to cover the parser overflow (big CASE),
and to cover nested CASE statements.
2006-11-17 20:14:29 +01:00
|
|
|
List<sp_instr> leads;
|
2004-08-02 18:05:31 +02:00
|
|
|
|
Bug#19194 (Right recursion in parser for CASE causes excessive stack usage,
limitation)
Note to the reviewer
====================
Warning: reviewing this patch is somewhat involved.
Due to the nature of several issues all affecting the same area,
fixing separately each issue is not practical, since each fix can not be
implemented and tested independently.
In particular, the issues with
- rule recursion
- nested case statements
- forward jump resolution (backpatch list)
are tightly coupled (see below).
Definitions
===========
The expression
CASE expr
WHEN expr THEN expr
WHEN expr THEN expr
...
END
is a "Simple Case Expression".
The expression
CASE
WHEN expr THEN expr
WHEN expr THEN expr
...
END
is a "Searched Case Expression".
The statement
CASE expr
WHEN expr THEN stmts
WHEN expr THEN stmts
...
END CASE
is a "Simple Case Statement".
The statement
CASE
WHEN expr THEN stmts
WHEN expr THEN stmts
...
END CASE
is a "Searched Case Statement".
A "Left Recursive" rule is like
list:
element
| list element
;
A "Right Recursive" rule is like
list:
element
| element list
;
Left and right recursion produces the same language, the difference only
affects the *order* in which the text is parsed.
In a descendant parser (usually written manually), right recursion works
very well, and is typically implemented with a while loop.
In an ascendant parser (yacc/bison) left recursion works very well,
and is implemented naturally by the parser stack.
In both cases, using the wrong type or recursion is very bad and should be
avoided, as it causes technical issues with the parser implementation.
Before this change
==================
The "Simple Case Expression" and "Searched Case Expression" were both
implemented by the "when_list" and "when_list2" rules, which are left
recursive (ok).
These rules, however, used lex->when_list instead of using the parser stack,
which is more complex that necessary, and potentially dangerous because
of other rules using THD::reset_lex.
The "Simple Case Statement" and "Searched Case Statements" were implemented
by the "sp_case", "sp_whens" and in part by "sp_proc_stmt" rules.
Both cases were right recursive (bad).
The grammar involved was convoluted, and is assumed to be the results of
tweaks to get the code generation to work, but is not what someone would
naturally write.
In addition, using a common rule for both "Simple" and "Searched" case
statements was implemented with sp_head::m_flags |= IN_SIMPLE_CASE,
which is a flag and not a stack, and therefore does not take into account
*nested* case statements. This leads to incorrect generated code, and either
a server crash or an incorrect result.
With regards to the backpatch mechanism, a *different* backpatch list was
created for each jump from "WHEN expr THEN stmt" to "END CASE", which
relied on the grammar to be right recursive.
This is a mis-use of the backpatch list, since this list can resolve
multiple references to the same target at once.
The optimizer algorithm used to detect dead code in the "assembly" SQL
instructions, implemented by sp_head::opt_mark(uint ip), was recursive
in some cases (a conditional jump pointing forward to another conditional
jump).
In case of specially crafted code, like
- a long list of "IF expr THEN stmt END IF"
- a long CASE statement
this would actually cause a server crash with a stack overflow.
In general, having a stack that grows proportionally with user data (the
SQL code given by the client in a CREATE PROCEDURE) is to be avoided.
In debug builds only, creating a SP / SF / Trigger which had a significant
amount of code would spend --literally-- several minutes in sp_head::create,
because of the debug code involved with DBUG_PRINT("info", ("Code %s ...
There are several issues with this code:
- in a CASE with 5 000 WHEN, there are 15 000 instructions generated,
which create a sting representation of the code which is 500 000 bytes
long,
- using a String instead of an io stream causes performances to degrade
to a total server freeze, as time is spent doing realloc of a buffer
always too short,
- Printing a 500 000 long string in the debug log is too verbose,
- Generating this string even when DBUG_PRINT is off is useless,
- Having code that potentially can affect the server behavior, used with
#ifdef / #endif is useful in some cases, but is also a bad practice.
After this change
=================
"Case Expressions" (both simple and searched) have been simplified to
not use LEX::when_list, which has been removed.
Considering all the issues affecting case statements, the grammar for these
has been totally re written.
The existing actions, used to generate "assembly" sp_inst* code, have been
preserved but moved in the new grammar, with the following changes:
a) Bison rules are no longer shared between "Simple" and "Searched" case
statements, because a stack instead of a flag is required to handle them.
Nested statements are handled naturally by the parser stack, which by
definition uses the correct rule in the correct context.
Nested statements of the opposite type (simple vs searched) works correctly.
The flag sp_head::IN_SIMPLE_CASE is no longer used.
This is a step towards resolution of WL#2999, which correctly identified
that temporary parsing flags do not belong to sp_head.
The code in the action is shared by mean of the case_stmt_action_xxx()
helpers.
b) The backpatch mechanism, used to resolve forward jumps in the generated
code, has been changed to:
- create a label for the instruction following 'END CASE',
- register each jump at the end of a "WHEN expr THEN stmt" in a *unique*
backpatch list associated with the 'END CASE' label
- resolve all the forward jumps for this label at once.
In addition, the code involving backpatch has been commented, so that a
reader can now understand by reading matching "Registering" and "Resolving"
comments how the forward jumps are resolved and what target they resolve to,
as this is far from evident when reading the code alone.
The implementation of sp_head::opt_mark() has been revised to avoid
recursive calls from jump instructions, and instead add the jump location
to the list of paths to explore during the flow analysis of the instruction
graph, with a call to sp_head::add_mark_lead().
In addition, the flow analysis will stop if an instruction has already
been marked as reachable, which the previous code failed to do in the
recursive case.
sp_head::opt_mark() is now private, to prevent new calls to this method from
being introduced.
The debug code present in sp_head::create() has been removed.
Considering that SHOW PROCEDURE CODE is also available in debug builds,
and can be used anytime regardless of the trace level, as opposed to
"CREATE PROCEDURE" time and only if the trace was on,
removing the code actually makes debugging easier (usable trace).
Tests have been written to cover the parser overflow (big CASE),
and to cover nested CASE statements.
2006-11-17 20:14:29 +01:00
|
|
|
/*
|
|
|
|
Forward flow analysis algorithm in the instruction graph:
|
|
|
|
- first, add the entry point in the graph (the first instruction) to the
|
|
|
|
'leads' list of paths to explore.
|
|
|
|
- while there are still leads to explore:
|
|
|
|
- pick one lead, and follow the path forward. Mark instruction reached.
|
|
|
|
Stop only if the end of the routine is reached, or the path converge
|
|
|
|
to code already explored (marked).
|
|
|
|
- while following a path, collect in the 'leads' list any fork to
|
|
|
|
another path (caused by conditional jumps instructions), so that these
|
|
|
|
paths can be explored as well.
|
|
|
|
*/
|
|
|
|
|
|
|
|
/* Add the entry point */
|
|
|
|
i= get_instr(0);
|
|
|
|
leads.push_front(i);
|
|
|
|
|
|
|
|
/* For each path of code ... */
|
|
|
|
while (leads.elements != 0)
|
|
|
|
{
|
|
|
|
i= leads.pop();
|
|
|
|
|
|
|
|
/* Mark the entire path, collecting new leads. */
|
|
|
|
while (i && ! i->marked)
|
|
|
|
{
|
|
|
|
ip= i->opt_mark(this, & leads);
|
|
|
|
i= get_instr(ip);
|
|
|
|
}
|
|
|
|
}
|
2004-08-02 18:05:31 +02:00
|
|
|
}
|
|
|
|
|
2005-03-04 14:35:28 +01:00
|
|
|
|
2005-11-17 11:11:48 +01:00
|
|
|
#ifndef DBUG_OFF
|
2006-01-26 17:26:25 +01:00
|
|
|
/*
|
|
|
|
Return the routine instructions as a result set.
|
|
|
|
Returns 0 if ok, !=0 on error.
|
|
|
|
*/
|
2005-11-17 11:11:48 +01:00
|
|
|
int
|
|
|
|
sp_head::show_routine_code(THD *thd)
|
|
|
|
{
|
|
|
|
Protocol *protocol= thd->protocol;
|
|
|
|
char buff[2048];
|
|
|
|
String buffer(buff, sizeof(buff), system_charset_info);
|
|
|
|
List<Item> field_list;
|
2005-11-18 16:30:27 +01:00
|
|
|
sp_instr *i;
|
2005-11-17 11:11:48 +01:00
|
|
|
bool full_access;
|
2005-11-22 14:25:44 +01:00
|
|
|
int res= 0;
|
2005-11-17 11:11:48 +01:00
|
|
|
uint ip;
|
|
|
|
DBUG_ENTER("sp_head::show_routine_code");
|
2005-11-18 16:30:27 +01:00
|
|
|
DBUG_PRINT("info", ("procedure: %s", m_name.str));
|
2005-11-17 11:11:48 +01:00
|
|
|
|
|
|
|
if (check_show_routine_access(thd, this, &full_access) || !full_access)
|
|
|
|
DBUG_RETURN(1);
|
|
|
|
|
|
|
|
field_list.push_back(new Item_uint("Pos", 9));
|
|
|
|
// 1024 is for not to confuse old clients
|
|
|
|
field_list.push_back(new Item_empty_string("Instruction",
|
|
|
|
max(buffer.length(), 1024)));
|
|
|
|
if (protocol->send_fields(&field_list, Protocol::SEND_NUM_ROWS |
|
|
|
|
Protocol::SEND_EOF))
|
|
|
|
DBUG_RETURN(1);
|
|
|
|
|
|
|
|
for (ip= 0; (i = get_instr(ip)) ; ip++)
|
|
|
|
{
|
2006-01-26 17:26:25 +01:00
|
|
|
/*
|
|
|
|
Consistency check. If these are different something went wrong
|
|
|
|
during optimization.
|
|
|
|
*/
|
|
|
|
if (ip != i->m_ip)
|
|
|
|
{
|
|
|
|
const char *format= "Instruction at position %u has m_ip=%u";
|
|
|
|
char tmp[sizeof(format) + 2*SP_INSTR_UINT_MAXLEN + 1];
|
|
|
|
|
|
|
|
sprintf(tmp, format, ip, i->m_ip);
|
|
|
|
/*
|
|
|
|
Since this is for debugging purposes only, we don't bother to
|
|
|
|
introduce a special error code for it.
|
|
|
|
*/
|
|
|
|
push_warning(thd, MYSQL_ERROR::WARN_LEVEL_WARN, ER_UNKNOWN_ERROR, tmp);
|
|
|
|
}
|
2005-11-17 11:11:48 +01:00
|
|
|
protocol->prepare_for_resend();
|
|
|
|
protocol->store((longlong)ip);
|
|
|
|
|
|
|
|
buffer.set("", 0, system_charset_info);
|
|
|
|
i->print(&buffer);
|
2005-11-18 16:30:27 +01:00
|
|
|
protocol->store(buffer.ptr(), buffer.length(), system_charset_info);
|
2005-11-17 11:11:48 +01:00
|
|
|
if ((res= protocol->write()))
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
send_eof(thd);
|
|
|
|
|
|
|
|
DBUG_RETURN(res);
|
|
|
|
}
|
|
|
|
#endif // ifndef DBUG_OFF
|
|
|
|
|
|
|
|
|
2005-03-04 14:35:28 +01:00
|
|
|
/*
|
|
|
|
Prepare LEX and thread for execution of instruction, if requested open
|
|
|
|
and lock LEX's tables, execute instruction's core function, perform
|
|
|
|
cleanup afterwards.
|
|
|
|
|
|
|
|
SYNOPSIS
|
|
|
|
reset_lex_and_exec_core()
|
|
|
|
thd - thread context
|
|
|
|
nextp - out - next instruction
|
|
|
|
open_tables - if TRUE then check read access to tables in LEX's table
|
|
|
|
list and open and lock them (used in instructions which
|
|
|
|
need to calculate some expression and don't execute
|
|
|
|
complete statement).
|
|
|
|
sp_instr - instruction for which we prepare context, and which core
|
|
|
|
function execute by calling its exec_core() method.
|
|
|
|
|
|
|
|
NOTE
|
|
|
|
We are not saving/restoring some parts of THD which may need this because
|
|
|
|
we do this once for whole routine execution in sp_head::execute().
|
|
|
|
|
|
|
|
RETURN VALUE
|
|
|
|
0/non-0 - Success/Failure
|
|
|
|
*/
|
|
|
|
|
|
|
|
int
|
|
|
|
sp_lex_keeper::reset_lex_and_exec_core(THD *thd, uint *nextp,
|
|
|
|
bool open_tables, sp_instr* instr)
|
|
|
|
{
|
|
|
|
int res= 0;
|
|
|
|
|
|
|
|
DBUG_ASSERT(!thd->derived_tables);
|
|
|
|
DBUG_ASSERT(thd->change_list.is_empty());
|
|
|
|
/*
|
|
|
|
Use our own lex.
|
|
|
|
We should not save old value since it is saved/restored in
|
|
|
|
sp_head::execute() when we are entering/leaving routine.
|
|
|
|
*/
|
|
|
|
thd->lex= m_lex;
|
|
|
|
|
|
|
|
VOID(pthread_mutex_lock(&LOCK_thread_count));
|
2005-03-04 15:46:45 +01:00
|
|
|
thd->query_id= next_query_id();
|
2005-03-04 14:35:28 +01:00
|
|
|
VOID(pthread_mutex_unlock(&LOCK_thread_count));
|
|
|
|
|
2005-07-30 10:19:57 +02:00
|
|
|
if (thd->prelocked_mode == NON_PRELOCKED)
|
|
|
|
{
|
|
|
|
/*
|
|
|
|
This statement will enter/leave prelocked mode on its own.
|
|
|
|
Entering prelocked mode changes table list and related members
|
|
|
|
of LEX, so we'll need to restore them.
|
|
|
|
*/
|
|
|
|
if (lex_query_tables_own_last)
|
|
|
|
{
|
|
|
|
/*
|
|
|
|
We've already entered/left prelocked mode with this statement.
|
|
|
|
Attach the list of tables that need to be prelocked and mark m_lex
|
|
|
|
as having such list attached.
|
|
|
|
*/
|
|
|
|
*lex_query_tables_own_last= prelocking_tables;
|
|
|
|
m_lex->mark_as_requiring_prelocking(lex_query_tables_own_last);
|
|
|
|
}
|
|
|
|
}
|
Bug#8407 (Stored functions/triggers ignore exception handler)
Bug 18914 (Calling certain SPs from triggers fail)
Bug 20713 (Functions will not not continue for SQLSTATE VALUE '42S02')
Bug 21825 (Incorrect message error deleting records in a table with a
trigger for inserting)
Bug 22580 (DROP TABLE in nested stored procedure causes strange dependency
error)
Bug 25345 (Cursors from Functions)
This fix resolves a long standing issue originally reported with bug 8407,
which affect the behavior of Stored Procedures, Stored Functions and Trigger
in many different ways, causing symptoms reported by all the bugs listed.
In all cases, the root cause of the problem traces back to 8407 and how the
server locks tables involved with sub statements.
Prior to this fix, the implementation of stored routines would:
- compute the transitive closure of all the tables referenced by a top level
statement
- open and lock all the tables involved
- execute the top level statement
"transitive closure of tables" means collecting:
- all the tables,
- all the stored functions,
- all the views,
- all the table triggers
- all the stored procedures
involved, and recursively inspect these objects definition to find more
references to more objects, until the list of every object referenced does
not grow any more.
This mechanism is known as "pre-locking" tables before execution.
The motivation for locking all the tables (possibly) used at once is to
prevent dead locks.
One problem with this approach is that, if the execution path the code
really takes during runtime does not use a given table, and if the table is
missing, the server would not execute the statement.
This in particular has a major impact on triggers, since a missing table
referenced by an update/delete trigger would prevent an insert trigger to run.
Another problem is that stored routines might define SQL exception handlers
to deal with missing tables, but the server implementation would never give
user code a chance to execute this logic, since the routine is never
executed when a missing table cause the pre-locking code to fail.
With this fix, the internal implementation of the pre-locking code has been
relaxed of some constraints, so that failure to open a table does not
necessarily prevent execution of a stored routine.
In particular, the pre-locking mechanism is now behaving as follows:
1) the first step, to compute the transitive closure of all the tables
possibly referenced by a statement, is unchanged.
2) the next step, which is to open all the tables involved, only attempts
to open the tables added by the pre-locking code, but silently fails without
reporting any error or invoking any exception handler is the table is not
present. This is achieved by trapping internal errors with
Prelock_error_handler
3) the locking step only locks tables that were successfully opened.
4) when executing sub statements, the list of tables used by each statements
is evaluated as before. The tables needed by the sub statement are expected
to be already opened and locked. Statement referencing tables that were not
opened in step 2) will fail to find the table in the open list, and only at
this point will execution of the user code fail.
5) when a runtime exception is raised at 4), the instruction continuation
destination (the next instruction to execute in case of SQL continue
handlers) is evaluated.
This is achieved with sp_instr::exec_open_and_lock_tables()
6) if a user exception handler is present in the stored routine, that
handler is invoked as usual, so that ER_NO_SUCH_TABLE exceptions can be
trapped by stored routines. If no handler exists, then the runtime execution
will fail as expected.
With all these changes, a side effect is that view security is impacted, in
two different ways.
First, a view defined as "select stored_function()", where the stored
function references a table that may not exist, is considered valid.
The rationale is that, because the stored function might trap exceptions
during execution and still return a valid result, there is no way to decide
when the view is created if a missing table really cause the view to be invalid.
Secondly, testing for existence of tables is now done later during
execution. View security, which consist of trapping errors and return a
generic ER_VIEW_INVALID (to prevent disclosing information) was only
implemented at very specific phases covering *opening* tables, but not
covering the runtime execution. Because of this existing limitation,
errors that were previously trapped and converted into ER_VIEW_INVALID are
not trapped, causing table names to be reported to the user.
This change is exposing an existing problem, which is independent and will
be resolved separately.
2007-03-06 03:42:07 +01:00
|
|
|
|
2005-07-30 10:19:57 +02:00
|
|
|
reinit_stmt_before_use(thd, m_lex);
|
Bug#8407 (Stored functions/triggers ignore exception handler)
Bug 18914 (Calling certain SPs from triggers fail)
Bug 20713 (Functions will not not continue for SQLSTATE VALUE '42S02')
Bug 21825 (Incorrect message error deleting records in a table with a
trigger for inserting)
Bug 22580 (DROP TABLE in nested stored procedure causes strange dependency
error)
Bug 25345 (Cursors from Functions)
This fix resolves a long standing issue originally reported with bug 8407,
which affect the behavior of Stored Procedures, Stored Functions and Trigger
in many different ways, causing symptoms reported by all the bugs listed.
In all cases, the root cause of the problem traces back to 8407 and how the
server locks tables involved with sub statements.
Prior to this fix, the implementation of stored routines would:
- compute the transitive closure of all the tables referenced by a top level
statement
- open and lock all the tables involved
- execute the top level statement
"transitive closure of tables" means collecting:
- all the tables,
- all the stored functions,
- all the views,
- all the table triggers
- all the stored procedures
involved, and recursively inspect these objects definition to find more
references to more objects, until the list of every object referenced does
not grow any more.
This mechanism is known as "pre-locking" tables before execution.
The motivation for locking all the tables (possibly) used at once is to
prevent dead locks.
One problem with this approach is that, if the execution path the code
really takes during runtime does not use a given table, and if the table is
missing, the server would not execute the statement.
This in particular has a major impact on triggers, since a missing table
referenced by an update/delete trigger would prevent an insert trigger to run.
Another problem is that stored routines might define SQL exception handlers
to deal with missing tables, but the server implementation would never give
user code a chance to execute this logic, since the routine is never
executed when a missing table cause the pre-locking code to fail.
With this fix, the internal implementation of the pre-locking code has been
relaxed of some constraints, so that failure to open a table does not
necessarily prevent execution of a stored routine.
In particular, the pre-locking mechanism is now behaving as follows:
1) the first step, to compute the transitive closure of all the tables
possibly referenced by a statement, is unchanged.
2) the next step, which is to open all the tables involved, only attempts
to open the tables added by the pre-locking code, but silently fails without
reporting any error or invoking any exception handler is the table is not
present. This is achieved by trapping internal errors with
Prelock_error_handler
3) the locking step only locks tables that were successfully opened.
4) when executing sub statements, the list of tables used by each statements
is evaluated as before. The tables needed by the sub statement are expected
to be already opened and locked. Statement referencing tables that were not
opened in step 2) will fail to find the table in the open list, and only at
this point will execution of the user code fail.
5) when a runtime exception is raised at 4), the instruction continuation
destination (the next instruction to execute in case of SQL continue
handlers) is evaluated.
This is achieved with sp_instr::exec_open_and_lock_tables()
6) if a user exception handler is present in the stored routine, that
handler is invoked as usual, so that ER_NO_SUCH_TABLE exceptions can be
trapped by stored routines. If no handler exists, then the runtime execution
will fail as expected.
With all these changes, a side effect is that view security is impacted, in
two different ways.
First, a view defined as "select stored_function()", where the stored
function references a table that may not exist, is considered valid.
The rationale is that, because the stored function might trap exceptions
during execution and still return a valid result, there is no way to decide
when the view is created if a missing table really cause the view to be invalid.
Secondly, testing for existence of tables is now done later during
execution. View security, which consist of trapping errors and return a
generic ER_VIEW_INVALID (to prevent disclosing information) was only
implemented at very specific phases covering *opening* tables, but not
covering the runtime execution. Because of this existing limitation,
errors that were previously trapped and converted into ER_VIEW_INVALID are
not trapped, causing table names to be reported to the user.
This change is exposing an existing problem, which is independent and will
be resolved separately.
2007-03-06 03:42:07 +01:00
|
|
|
|
|
|
|
if (open_tables)
|
2007-03-07 17:53:46 +01:00
|
|
|
res= instr->exec_open_and_lock_tables(thd, m_lex->query_tables);
|
2005-03-04 14:35:28 +01:00
|
|
|
|
|
|
|
if (!res)
|
|
|
|
res= instr->exec_core(thd, nextp);
|
|
|
|
|
|
|
|
m_lex->unit.cleanup();
|
|
|
|
|
|
|
|
thd->proc_info="closing tables";
|
|
|
|
close_thread_tables(thd);
|
2005-08-11 14:58:15 +02:00
|
|
|
thd->proc_info= 0;
|
2005-03-04 14:35:28 +01:00
|
|
|
|
2005-08-03 05:37:32 +02:00
|
|
|
if (m_lex->query_tables_own_last)
|
2005-07-30 10:19:57 +02:00
|
|
|
{
|
2005-08-03 05:37:32 +02:00
|
|
|
/*
|
|
|
|
We've entered and left prelocking mode when executing statement
|
|
|
|
stored in m_lex.
|
|
|
|
m_lex->query_tables(->next_global)* list now has a 'tail' - a list
|
|
|
|
of tables that are added for prelocking. (If this is the first
|
|
|
|
execution, the 'tail' was added by open_tables(), otherwise we've
|
|
|
|
attached it above in this function).
|
|
|
|
Now we'll save the 'tail', and detach it.
|
|
|
|
*/
|
|
|
|
lex_query_tables_own_last= m_lex->query_tables_own_last;
|
|
|
|
prelocking_tables= *lex_query_tables_own_last;
|
|
|
|
*lex_query_tables_own_last= NULL;
|
|
|
|
m_lex->mark_as_requiring_prelocking(NULL);
|
2005-07-30 10:19:57 +02:00
|
|
|
}
|
2005-03-04 14:35:28 +01:00
|
|
|
thd->rollback_item_tree_changes();
|
2006-06-22 17:29:48 +02:00
|
|
|
/* Update the state of the active arena. */
|
|
|
|
thd->stmt_arena->state= Query_arena::EXECUTED;
|
|
|
|
|
2005-03-04 14:35:28 +01:00
|
|
|
|
|
|
|
/*
|
|
|
|
Unlike for PS we should not call Item's destructors for newly created
|
|
|
|
items after execution of each instruction in stored routine. This is
|
|
|
|
because SP often create Item (like Item_int, Item_string etc...) when
|
|
|
|
they want to store some value in local variable, pass return value and
|
|
|
|
etc... So their life time should be longer than one instruction.
|
|
|
|
|
|
|
|
cleanup_items() is called in sp_head::execute()
|
|
|
|
*/
|
2005-12-06 22:57:15 +01:00
|
|
|
return res || thd->net.report_error;
|
2005-03-04 14:35:28 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
|
2005-08-11 14:58:15 +02:00
|
|
|
/*
|
|
|
|
sp_instr class functions
|
|
|
|
*/
|
|
|
|
|
2007-03-07 17:53:46 +01:00
|
|
|
int sp_instr::exec_open_and_lock_tables(THD *thd, TABLE_LIST *tables)
|
Bug#8407 (Stored functions/triggers ignore exception handler)
Bug 18914 (Calling certain SPs from triggers fail)
Bug 20713 (Functions will not not continue for SQLSTATE VALUE '42S02')
Bug 21825 (Incorrect message error deleting records in a table with a
trigger for inserting)
Bug 22580 (DROP TABLE in nested stored procedure causes strange dependency
error)
Bug 25345 (Cursors from Functions)
This fix resolves a long standing issue originally reported with bug 8407,
which affect the behavior of Stored Procedures, Stored Functions and Trigger
in many different ways, causing symptoms reported by all the bugs listed.
In all cases, the root cause of the problem traces back to 8407 and how the
server locks tables involved with sub statements.
Prior to this fix, the implementation of stored routines would:
- compute the transitive closure of all the tables referenced by a top level
statement
- open and lock all the tables involved
- execute the top level statement
"transitive closure of tables" means collecting:
- all the tables,
- all the stored functions,
- all the views,
- all the table triggers
- all the stored procedures
involved, and recursively inspect these objects definition to find more
references to more objects, until the list of every object referenced does
not grow any more.
This mechanism is known as "pre-locking" tables before execution.
The motivation for locking all the tables (possibly) used at once is to
prevent dead locks.
One problem with this approach is that, if the execution path the code
really takes during runtime does not use a given table, and if the table is
missing, the server would not execute the statement.
This in particular has a major impact on triggers, since a missing table
referenced by an update/delete trigger would prevent an insert trigger to run.
Another problem is that stored routines might define SQL exception handlers
to deal with missing tables, but the server implementation would never give
user code a chance to execute this logic, since the routine is never
executed when a missing table cause the pre-locking code to fail.
With this fix, the internal implementation of the pre-locking code has been
relaxed of some constraints, so that failure to open a table does not
necessarily prevent execution of a stored routine.
In particular, the pre-locking mechanism is now behaving as follows:
1) the first step, to compute the transitive closure of all the tables
possibly referenced by a statement, is unchanged.
2) the next step, which is to open all the tables involved, only attempts
to open the tables added by the pre-locking code, but silently fails without
reporting any error or invoking any exception handler is the table is not
present. This is achieved by trapping internal errors with
Prelock_error_handler
3) the locking step only locks tables that were successfully opened.
4) when executing sub statements, the list of tables used by each statements
is evaluated as before. The tables needed by the sub statement are expected
to be already opened and locked. Statement referencing tables that were not
opened in step 2) will fail to find the table in the open list, and only at
this point will execution of the user code fail.
5) when a runtime exception is raised at 4), the instruction continuation
destination (the next instruction to execute in case of SQL continue
handlers) is evaluated.
This is achieved with sp_instr::exec_open_and_lock_tables()
6) if a user exception handler is present in the stored routine, that
handler is invoked as usual, so that ER_NO_SUCH_TABLE exceptions can be
trapped by stored routines. If no handler exists, then the runtime execution
will fail as expected.
With all these changes, a side effect is that view security is impacted, in
two different ways.
First, a view defined as "select stored_function()", where the stored
function references a table that may not exist, is considered valid.
The rationale is that, because the stored function might trap exceptions
during execution and still return a valid result, there is no way to decide
when the view is created if a missing table really cause the view to be invalid.
Secondly, testing for existence of tables is now done later during
execution. View security, which consist of trapping errors and return a
generic ER_VIEW_INVALID (to prevent disclosing information) was only
implemented at very specific phases covering *opening* tables, but not
covering the runtime execution. Because of this existing limitation,
errors that were previously trapped and converted into ER_VIEW_INVALID are
not trapped, causing table names to be reported to the user.
This change is exposing an existing problem, which is independent and will
be resolved separately.
2007-03-06 03:42:07 +01:00
|
|
|
{
|
|
|
|
int result;
|
|
|
|
|
|
|
|
/*
|
|
|
|
Check whenever we have access to tables for this statement
|
|
|
|
and open and lock them before executing instructions core function.
|
|
|
|
*/
|
|
|
|
if (check_table_access(thd, SELECT_ACL, tables, 0)
|
|
|
|
|| open_and_lock_tables(thd, tables))
|
|
|
|
result= -1;
|
|
|
|
else
|
|
|
|
result= 0;
|
|
|
|
|
|
|
|
return result;
|
|
|
|
}
|
|
|
|
|
2007-03-07 17:53:46 +01:00
|
|
|
uint sp_instr::get_cont_dest()
|
Bug#8407 (Stored functions/triggers ignore exception handler)
Bug 18914 (Calling certain SPs from triggers fail)
Bug 20713 (Functions will not not continue for SQLSTATE VALUE '42S02')
Bug 21825 (Incorrect message error deleting records in a table with a
trigger for inserting)
Bug 22580 (DROP TABLE in nested stored procedure causes strange dependency
error)
Bug 25345 (Cursors from Functions)
This fix resolves a long standing issue originally reported with bug 8407,
which affect the behavior of Stored Procedures, Stored Functions and Trigger
in many different ways, causing symptoms reported by all the bugs listed.
In all cases, the root cause of the problem traces back to 8407 and how the
server locks tables involved with sub statements.
Prior to this fix, the implementation of stored routines would:
- compute the transitive closure of all the tables referenced by a top level
statement
- open and lock all the tables involved
- execute the top level statement
"transitive closure of tables" means collecting:
- all the tables,
- all the stored functions,
- all the views,
- all the table triggers
- all the stored procedures
involved, and recursively inspect these objects definition to find more
references to more objects, until the list of every object referenced does
not grow any more.
This mechanism is known as "pre-locking" tables before execution.
The motivation for locking all the tables (possibly) used at once is to
prevent dead locks.
One problem with this approach is that, if the execution path the code
really takes during runtime does not use a given table, and if the table is
missing, the server would not execute the statement.
This in particular has a major impact on triggers, since a missing table
referenced by an update/delete trigger would prevent an insert trigger to run.
Another problem is that stored routines might define SQL exception handlers
to deal with missing tables, but the server implementation would never give
user code a chance to execute this logic, since the routine is never
executed when a missing table cause the pre-locking code to fail.
With this fix, the internal implementation of the pre-locking code has been
relaxed of some constraints, so that failure to open a table does not
necessarily prevent execution of a stored routine.
In particular, the pre-locking mechanism is now behaving as follows:
1) the first step, to compute the transitive closure of all the tables
possibly referenced by a statement, is unchanged.
2) the next step, which is to open all the tables involved, only attempts
to open the tables added by the pre-locking code, but silently fails without
reporting any error or invoking any exception handler is the table is not
present. This is achieved by trapping internal errors with
Prelock_error_handler
3) the locking step only locks tables that were successfully opened.
4) when executing sub statements, the list of tables used by each statements
is evaluated as before. The tables needed by the sub statement are expected
to be already opened and locked. Statement referencing tables that were not
opened in step 2) will fail to find the table in the open list, and only at
this point will execution of the user code fail.
5) when a runtime exception is raised at 4), the instruction continuation
destination (the next instruction to execute in case of SQL continue
handlers) is evaluated.
This is achieved with sp_instr::exec_open_and_lock_tables()
6) if a user exception handler is present in the stored routine, that
handler is invoked as usual, so that ER_NO_SUCH_TABLE exceptions can be
trapped by stored routines. If no handler exists, then the runtime execution
will fail as expected.
With all these changes, a side effect is that view security is impacted, in
two different ways.
First, a view defined as "select stored_function()", where the stored
function references a table that may not exist, is considered valid.
The rationale is that, because the stored function might trap exceptions
during execution and still return a valid result, there is no way to decide
when the view is created if a missing table really cause the view to be invalid.
Secondly, testing for existence of tables is now done later during
execution. View security, which consist of trapping errors and return a
generic ER_VIEW_INVALID (to prevent disclosing information) was only
implemented at very specific phases covering *opening* tables, but not
covering the runtime execution. Because of this existing limitation,
errors that were previously trapped and converted into ER_VIEW_INVALID are
not trapped, causing table names to be reported to the user.
This change is exposing an existing problem, which is independent and will
be resolved separately.
2007-03-06 03:42:07 +01:00
|
|
|
{
|
2007-03-07 17:53:46 +01:00
|
|
|
return (m_ip+1);
|
Bug#8407 (Stored functions/triggers ignore exception handler)
Bug 18914 (Calling certain SPs from triggers fail)
Bug 20713 (Functions will not not continue for SQLSTATE VALUE '42S02')
Bug 21825 (Incorrect message error deleting records in a table with a
trigger for inserting)
Bug 22580 (DROP TABLE in nested stored procedure causes strange dependency
error)
Bug 25345 (Cursors from Functions)
This fix resolves a long standing issue originally reported with bug 8407,
which affect the behavior of Stored Procedures, Stored Functions and Trigger
in many different ways, causing symptoms reported by all the bugs listed.
In all cases, the root cause of the problem traces back to 8407 and how the
server locks tables involved with sub statements.
Prior to this fix, the implementation of stored routines would:
- compute the transitive closure of all the tables referenced by a top level
statement
- open and lock all the tables involved
- execute the top level statement
"transitive closure of tables" means collecting:
- all the tables,
- all the stored functions,
- all the views,
- all the table triggers
- all the stored procedures
involved, and recursively inspect these objects definition to find more
references to more objects, until the list of every object referenced does
not grow any more.
This mechanism is known as "pre-locking" tables before execution.
The motivation for locking all the tables (possibly) used at once is to
prevent dead locks.
One problem with this approach is that, if the execution path the code
really takes during runtime does not use a given table, and if the table is
missing, the server would not execute the statement.
This in particular has a major impact on triggers, since a missing table
referenced by an update/delete trigger would prevent an insert trigger to run.
Another problem is that stored routines might define SQL exception handlers
to deal with missing tables, but the server implementation would never give
user code a chance to execute this logic, since the routine is never
executed when a missing table cause the pre-locking code to fail.
With this fix, the internal implementation of the pre-locking code has been
relaxed of some constraints, so that failure to open a table does not
necessarily prevent execution of a stored routine.
In particular, the pre-locking mechanism is now behaving as follows:
1) the first step, to compute the transitive closure of all the tables
possibly referenced by a statement, is unchanged.
2) the next step, which is to open all the tables involved, only attempts
to open the tables added by the pre-locking code, but silently fails without
reporting any error or invoking any exception handler is the table is not
present. This is achieved by trapping internal errors with
Prelock_error_handler
3) the locking step only locks tables that were successfully opened.
4) when executing sub statements, the list of tables used by each statements
is evaluated as before. The tables needed by the sub statement are expected
to be already opened and locked. Statement referencing tables that were not
opened in step 2) will fail to find the table in the open list, and only at
this point will execution of the user code fail.
5) when a runtime exception is raised at 4), the instruction continuation
destination (the next instruction to execute in case of SQL continue
handlers) is evaluated.
This is achieved with sp_instr::exec_open_and_lock_tables()
6) if a user exception handler is present in the stored routine, that
handler is invoked as usual, so that ER_NO_SUCH_TABLE exceptions can be
trapped by stored routines. If no handler exists, then the runtime execution
will fail as expected.
With all these changes, a side effect is that view security is impacted, in
two different ways.
First, a view defined as "select stored_function()", where the stored
function references a table that may not exist, is considered valid.
The rationale is that, because the stored function might trap exceptions
during execution and still return a valid result, there is no way to decide
when the view is created if a missing table really cause the view to be invalid.
Secondly, testing for existence of tables is now done later during
execution. View security, which consist of trapping errors and return a
generic ER_VIEW_INVALID (to prevent disclosing information) was only
implemented at very specific phases covering *opening* tables, but not
covering the runtime execution. Because of this existing limitation,
errors that were previously trapped and converted into ER_VIEW_INVALID are
not trapped, causing table names to be reported to the user.
This change is exposing an existing problem, which is independent and will
be resolved separately.
2007-03-06 03:42:07 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
|
2005-03-04 14:35:28 +01:00
|
|
|
int sp_instr::exec_core(THD *thd, uint *nextp)
|
2003-06-29 18:15:17 +02:00
|
|
|
{
|
2005-03-04 14:35:28 +01:00
|
|
|
DBUG_ASSERT(0);
|
|
|
|
return 0;
|
2003-06-29 18:15:17 +02:00
|
|
|
}
|
|
|
|
|
2005-08-11 14:58:15 +02:00
|
|
|
/*
|
|
|
|
sp_instr_stmt class functions
|
|
|
|
*/
|
|
|
|
|
Simplistic, experimental framework for Stored Procedures (SPs).
Implements creation and dropping of PROCEDUREs, IN, OUT, and INOUT parameters,
single-statement procedures, rudimentary multi-statement (begin-end) prodedures
(when the client can handle it), and local variables.
Missing most of the embedded SQL language, all attributes, FUNCTIONs, error handling,
reparses procedures at each call (no caching), etc, etc.
Certainly buggy too, but procedures can actually be created and called....
2002-12-08 19:59:22 +01:00
|
|
|
int
|
2002-12-11 14:24:29 +01:00
|
|
|
sp_instr_stmt::execute(THD *thd, uint *nextp)
|
Simplistic, experimental framework for Stored Procedures (SPs).
Implements creation and dropping of PROCEDUREs, IN, OUT, and INOUT parameters,
single-statement procedures, rudimentary multi-statement (begin-end) prodedures
(when the client can handle it), and local variables.
Missing most of the embedded SQL language, all attributes, FUNCTIONs, error handling,
reparses procedures at each call (no caching), etc, etc.
Certainly buggy too, but procedures can actually be created and called....
2002-12-08 19:59:22 +01:00
|
|
|
{
|
2004-09-17 15:40:38 +02:00
|
|
|
char *query;
|
|
|
|
uint32 query_length;
|
2005-08-25 15:34:34 +02:00
|
|
|
int res;
|
2003-02-12 16:17:03 +01:00
|
|
|
DBUG_ENTER("sp_instr_stmt::execute");
|
2005-03-04 14:35:28 +01:00
|
|
|
DBUG_PRINT("info", ("command: %d", m_lex_keeper.sql_command()));
|
2004-09-17 15:40:38 +02:00
|
|
|
|
|
|
|
query= thd->query;
|
|
|
|
query_length= thd->query_length;
|
2005-08-27 00:33:06 +02:00
|
|
|
if (!(res= alloc_query(thd, m_query.str, m_query.length+1)) &&
|
|
|
|
!(res=subst_spvars(thd, this, &m_query)))
|
2004-09-17 15:40:38 +02:00
|
|
|
{
|
2005-08-27 00:33:06 +02:00
|
|
|
/*
|
|
|
|
(the order of query cache and subst_spvars calls is irrelevant because
|
|
|
|
queries with SP vars can't be cached)
|
|
|
|
*/
|
2004-09-17 15:40:38 +02:00
|
|
|
if (query_cache_send_result_to_client(thd,
|
|
|
|
thd->query, thd->query_length) <= 0)
|
|
|
|
{
|
2005-03-04 14:35:28 +01:00
|
|
|
res= m_lex_keeper.reset_lex_and_exec_core(thd, nextp, FALSE, this);
|
2004-09-17 15:40:38 +02:00
|
|
|
query_cache_end_of_result(thd);
|
|
|
|
}
|
2005-03-04 14:35:28 +01:00
|
|
|
else
|
|
|
|
*nextp= m_ip+1;
|
2004-09-17 15:40:38 +02:00
|
|
|
thd->query= query;
|
|
|
|
thd->query_length= query_length;
|
|
|
|
}
|
2003-10-10 16:57:21 +02:00
|
|
|
DBUG_RETURN(res);
|
|
|
|
}
|
|
|
|
|
2005-11-22 13:06:52 +01:00
|
|
|
|
2004-03-29 11:16:45 +02:00
|
|
|
void
|
|
|
|
sp_instr_stmt::print(String *str)
|
|
|
|
{
|
2005-11-17 11:11:48 +01:00
|
|
|
uint i, len;
|
|
|
|
|
2005-11-22 13:06:52 +01:00
|
|
|
/* stmt CMD "..." */
|
|
|
|
if (str->reserve(SP_STMT_PRINT_MAXLEN+SP_INSTR_UINT_MAXLEN+8))
|
2005-11-18 16:30:27 +01:00
|
|
|
return;
|
2005-11-22 14:25:44 +01:00
|
|
|
str->qs_append(STRING_WITH_LEN("stmt "));
|
2005-03-04 14:35:28 +01:00
|
|
|
str->qs_append((uint)m_lex_keeper.sql_command());
|
2005-11-22 14:25:44 +01:00
|
|
|
str->qs_append(STRING_WITH_LEN(" \""));
|
2005-11-17 11:11:48 +01:00
|
|
|
len= m_query.length;
|
|
|
|
/*
|
|
|
|
Print the query string (but not too much of it), just to indicate which
|
|
|
|
statement it is.
|
|
|
|
*/
|
2005-11-22 13:06:52 +01:00
|
|
|
if (len > SP_STMT_PRINT_MAXLEN)
|
|
|
|
len= SP_STMT_PRINT_MAXLEN-3;
|
2005-11-17 11:11:48 +01:00
|
|
|
/* Copy the query string and replace '\n' with ' ' in the process */
|
|
|
|
for (i= 0 ; i < len ; i++)
|
2005-11-18 16:30:27 +01:00
|
|
|
{
|
2006-01-05 23:47:49 +01:00
|
|
|
char c= m_query.str[i];
|
|
|
|
if (c == '\n')
|
|
|
|
c= ' ';
|
|
|
|
str->qs_append(c);
|
2005-11-18 16:30:27 +01:00
|
|
|
}
|
2005-11-22 13:06:52 +01:00
|
|
|
if (m_query.length > SP_STMT_PRINT_MAXLEN)
|
2005-11-22 14:25:44 +01:00
|
|
|
str->qs_append(STRING_WITH_LEN("...")); /* Indicate truncated string */
|
2005-11-18 16:30:27 +01:00
|
|
|
str->qs_append('"');
|
2004-03-29 11:16:45 +02:00
|
|
|
}
|
2006-01-05 23:47:49 +01:00
|
|
|
|
2004-03-29 11:16:45 +02:00
|
|
|
|
2003-10-10 16:57:21 +02:00
|
|
|
int
|
2005-03-04 14:35:28 +01:00
|
|
|
sp_instr_stmt::exec_core(THD *thd, uint *nextp)
|
2003-10-10 16:57:21 +02:00
|
|
|
{
|
2005-03-04 14:35:28 +01:00
|
|
|
int res= mysql_execute_command(thd);
|
|
|
|
*nextp= m_ip+1;
|
2003-10-10 16:57:21 +02:00
|
|
|
return res;
|
Simplistic, experimental framework for Stored Procedures (SPs).
Implements creation and dropping of PROCEDUREs, IN, OUT, and INOUT parameters,
single-statement procedures, rudimentary multi-statement (begin-end) prodedures
(when the client can handle it), and local variables.
Missing most of the embedded SQL language, all attributes, FUNCTIONs, error handling,
reparses procedures at each call (no caching), etc, etc.
Certainly buggy too, but procedures can actually be created and called....
2002-12-08 19:59:22 +01:00
|
|
|
}
|
|
|
|
|
2005-08-11 14:58:15 +02:00
|
|
|
|
|
|
|
/*
|
|
|
|
sp_instr_set class functions
|
|
|
|
*/
|
|
|
|
|
Simplistic, experimental framework for Stored Procedures (SPs).
Implements creation and dropping of PROCEDUREs, IN, OUT, and INOUT parameters,
single-statement procedures, rudimentary multi-statement (begin-end) prodedures
(when the client can handle it), and local variables.
Missing most of the embedded SQL language, all attributes, FUNCTIONs, error handling,
reparses procedures at each call (no caching), etc, etc.
Certainly buggy too, but procedures can actually be created and called....
2002-12-08 19:59:22 +01:00
|
|
|
int
|
2002-12-11 14:24:29 +01:00
|
|
|
sp_instr_set::execute(THD *thd, uint *nextp)
|
Simplistic, experimental framework for Stored Procedures (SPs).
Implements creation and dropping of PROCEDUREs, IN, OUT, and INOUT parameters,
single-statement procedures, rudimentary multi-statement (begin-end) prodedures
(when the client can handle it), and local variables.
Missing most of the embedded SQL language, all attributes, FUNCTIONs, error handling,
reparses procedures at each call (no caching), etc, etc.
Certainly buggy too, but procedures can actually be created and called....
2002-12-08 19:59:22 +01:00
|
|
|
{
|
2003-02-12 16:17:03 +01:00
|
|
|
DBUG_ENTER("sp_instr_set::execute");
|
|
|
|
DBUG_PRINT("info", ("offset: %u", m_offset));
|
2005-03-04 14:35:28 +01:00
|
|
|
|
|
|
|
DBUG_RETURN(m_lex_keeper.reset_lex_and_exec_core(thd, nextp, TRUE, this));
|
|
|
|
}
|
|
|
|
|
2005-08-11 14:58:15 +02:00
|
|
|
|
2005-03-04 14:35:28 +01:00
|
|
|
int
|
|
|
|
sp_instr_set::exec_core(THD *thd, uint *nextp)
|
|
|
|
{
|
2006-05-15 12:01:55 +02:00
|
|
|
int res= thd->spcont->set_variable(thd, m_offset, &m_value);
|
2004-07-21 14:53:09 +02:00
|
|
|
|
2005-12-07 15:01:17 +01:00
|
|
|
if (res && thd->spcont->found_handler_here())
|
2005-11-08 14:47:33 +01:00
|
|
|
{
|
|
|
|
/*
|
2005-12-07 15:01:17 +01:00
|
|
|
Failed to evaluate the value, and a handler has been found. Reset the
|
|
|
|
variable to NULL.
|
2005-11-08 14:47:33 +01:00
|
|
|
*/
|
|
|
|
|
2005-12-07 15:01:17 +01:00
|
|
|
if (thd->spcont->set_variable(thd, m_offset, 0))
|
|
|
|
{
|
|
|
|
/* If this also failed, let's abort. */
|
2005-11-08 14:47:33 +01:00
|
|
|
|
2005-12-07 15:01:17 +01:00
|
|
|
sp_rcontext *spcont= thd->spcont;
|
|
|
|
|
2007-06-18 23:54:35 +02:00
|
|
|
thd->spcont= NULL; /* Avoid handlers */
|
2005-11-08 14:47:33 +01:00
|
|
|
my_error(ER_OUT_OF_RESOURCES, MYF(0));
|
|
|
|
spcont->clear_handler();
|
|
|
|
thd->spcont= spcont;
|
|
|
|
}
|
|
|
|
}
|
2005-12-07 15:01:17 +01:00
|
|
|
|
2002-12-11 14:24:29 +01:00
|
|
|
*nextp = m_ip+1;
|
2005-03-04 14:35:28 +01:00
|
|
|
return res;
|
2002-12-11 14:24:29 +01:00
|
|
|
}
|
|
|
|
|
2004-03-29 11:16:45 +02:00
|
|
|
void
|
|
|
|
sp_instr_set::print(String *str)
|
|
|
|
{
|
2005-11-22 13:06:52 +01:00
|
|
|
/* set name@offset ... */
|
|
|
|
int rsrv = SP_INSTR_UINT_MAXLEN+6;
|
2006-04-07 16:53:15 +02:00
|
|
|
sp_variable_t *var = m_ctx->find_variable(m_offset);
|
2005-11-17 11:11:48 +01:00
|
|
|
|
|
|
|
/* 'var' should always be non-null, but just in case... */
|
|
|
|
if (var)
|
|
|
|
rsrv+= var->name.length;
|
2005-11-18 16:30:27 +01:00
|
|
|
if (str->reserve(rsrv))
|
|
|
|
return;
|
2005-11-22 13:24:53 +01:00
|
|
|
str->qs_append(STRING_WITH_LEN("set "));
|
2005-11-17 11:11:48 +01:00
|
|
|
if (var)
|
|
|
|
{
|
2005-11-18 16:30:27 +01:00
|
|
|
str->qs_append(var->name.str, var->name.length);
|
|
|
|
str->qs_append('@');
|
2005-11-17 11:11:48 +01:00
|
|
|
}
|
2004-03-29 11:16:45 +02:00
|
|
|
str->qs_append(m_offset);
|
2005-11-18 16:30:27 +01:00
|
|
|
str->qs_append(' ');
|
2004-03-29 11:16:45 +02:00
|
|
|
m_value->print(str);
|
|
|
|
}
|
|
|
|
|
2004-09-07 14:29:46 +02:00
|
|
|
|
2005-08-11 14:58:15 +02:00
|
|
|
/*
|
|
|
|
sp_instr_set_trigger_field class functions
|
|
|
|
*/
|
|
|
|
|
2004-09-07 14:29:46 +02:00
|
|
|
int
|
|
|
|
sp_instr_set_trigger_field::execute(THD *thd, uint *nextp)
|
|
|
|
{
|
|
|
|
DBUG_ENTER("sp_instr_set_trigger_field::execute");
|
2005-07-09 19:51:59 +02:00
|
|
|
DBUG_RETURN(m_lex_keeper.reset_lex_and_exec_core(thd, nextp, TRUE, this));
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
int
|
|
|
|
sp_instr_set_trigger_field::exec_core(THD *thd, uint *nextp)
|
|
|
|
{
|
2006-05-15 19:57:10 +02:00
|
|
|
const int res= (trigger_field->set_value(thd, &value) ? -1 : 0);
|
2005-07-09 19:51:59 +02:00
|
|
|
*nextp = m_ip+1;
|
|
|
|
return res;
|
2004-09-07 14:29:46 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
void
|
|
|
|
sp_instr_set_trigger_field::print(String *str)
|
|
|
|
{
|
2005-11-22 13:24:53 +01:00
|
|
|
str->append(STRING_WITH_LEN("set_trigger_field "));
|
2005-05-27 12:15:17 +02:00
|
|
|
trigger_field->print(str);
|
2005-11-20 19:47:07 +01:00
|
|
|
str->append(STRING_WITH_LEN(":="));
|
2004-09-07 14:29:46 +02:00
|
|
|
value->print(str);
|
|
|
|
}
|
|
|
|
|
Bug#8407 (Stored functions/triggers ignore exception handler)
Bug 18914 (Calling certain SPs from triggers fail)
Bug 20713 (Functions will not not continue for SQLSTATE VALUE '42S02')
Bug 21825 (Incorrect message error deleting records in a table with a
trigger for inserting)
Bug 22580 (DROP TABLE in nested stored procedure causes strange dependency
error)
Bug 25345 (Cursors from Functions)
This fix resolves a long standing issue originally reported with bug 8407,
which affect the behavior of Stored Procedures, Stored Functions and Trigger
in many different ways, causing symptoms reported by all the bugs listed.
In all cases, the root cause of the problem traces back to 8407 and how the
server locks tables involved with sub statements.
Prior to this fix, the implementation of stored routines would:
- compute the transitive closure of all the tables referenced by a top level
statement
- open and lock all the tables involved
- execute the top level statement
"transitive closure of tables" means collecting:
- all the tables,
- all the stored functions,
- all the views,
- all the table triggers
- all the stored procedures
involved, and recursively inspect these objects definition to find more
references to more objects, until the list of every object referenced does
not grow any more.
This mechanism is known as "pre-locking" tables before execution.
The motivation for locking all the tables (possibly) used at once is to
prevent dead locks.
One problem with this approach is that, if the execution path the code
really takes during runtime does not use a given table, and if the table is
missing, the server would not execute the statement.
This in particular has a major impact on triggers, since a missing table
referenced by an update/delete trigger would prevent an insert trigger to run.
Another problem is that stored routines might define SQL exception handlers
to deal with missing tables, but the server implementation would never give
user code a chance to execute this logic, since the routine is never
executed when a missing table cause the pre-locking code to fail.
With this fix, the internal implementation of the pre-locking code has been
relaxed of some constraints, so that failure to open a table does not
necessarily prevent execution of a stored routine.
In particular, the pre-locking mechanism is now behaving as follows:
1) the first step, to compute the transitive closure of all the tables
possibly referenced by a statement, is unchanged.
2) the next step, which is to open all the tables involved, only attempts
to open the tables added by the pre-locking code, but silently fails without
reporting any error or invoking any exception handler is the table is not
present. This is achieved by trapping internal errors with
Prelock_error_handler
3) the locking step only locks tables that were successfully opened.
4) when executing sub statements, the list of tables used by each statements
is evaluated as before. The tables needed by the sub statement are expected
to be already opened and locked. Statement referencing tables that were not
opened in step 2) will fail to find the table in the open list, and only at
this point will execution of the user code fail.
5) when a runtime exception is raised at 4), the instruction continuation
destination (the next instruction to execute in case of SQL continue
handlers) is evaluated.
This is achieved with sp_instr::exec_open_and_lock_tables()
6) if a user exception handler is present in the stored routine, that
handler is invoked as usual, so that ER_NO_SUCH_TABLE exceptions can be
trapped by stored routines. If no handler exists, then the runtime execution
will fail as expected.
With all these changes, a side effect is that view security is impacted, in
two different ways.
First, a view defined as "select stored_function()", where the stored
function references a table that may not exist, is considered valid.
The rationale is that, because the stored function might trap exceptions
during execution and still return a valid result, there is no way to decide
when the view is created if a missing table really cause the view to be invalid.
Secondly, testing for existence of tables is now done later during
execution. View security, which consist of trapping errors and return a
generic ER_VIEW_INVALID (to prevent disclosing information) was only
implemented at very specific phases covering *opening* tables, but not
covering the runtime execution. Because of this existing limitation,
errors that were previously trapped and converted into ER_VIEW_INVALID are
not trapped, causing table names to be reported to the user.
This change is exposing an existing problem, which is independent and will
be resolved separately.
2007-03-06 03:42:07 +01:00
|
|
|
/*
|
|
|
|
sp_instr_opt_meta
|
|
|
|
*/
|
|
|
|
|
2007-03-07 17:53:46 +01:00
|
|
|
uint sp_instr_opt_meta::get_cont_dest()
|
Bug#8407 (Stored functions/triggers ignore exception handler)
Bug 18914 (Calling certain SPs from triggers fail)
Bug 20713 (Functions will not not continue for SQLSTATE VALUE '42S02')
Bug 21825 (Incorrect message error deleting records in a table with a
trigger for inserting)
Bug 22580 (DROP TABLE in nested stored procedure causes strange dependency
error)
Bug 25345 (Cursors from Functions)
This fix resolves a long standing issue originally reported with bug 8407,
which affect the behavior of Stored Procedures, Stored Functions and Trigger
in many different ways, causing symptoms reported by all the bugs listed.
In all cases, the root cause of the problem traces back to 8407 and how the
server locks tables involved with sub statements.
Prior to this fix, the implementation of stored routines would:
- compute the transitive closure of all the tables referenced by a top level
statement
- open and lock all the tables involved
- execute the top level statement
"transitive closure of tables" means collecting:
- all the tables,
- all the stored functions,
- all the views,
- all the table triggers
- all the stored procedures
involved, and recursively inspect these objects definition to find more
references to more objects, until the list of every object referenced does
not grow any more.
This mechanism is known as "pre-locking" tables before execution.
The motivation for locking all the tables (possibly) used at once is to
prevent dead locks.
One problem with this approach is that, if the execution path the code
really takes during runtime does not use a given table, and if the table is
missing, the server would not execute the statement.
This in particular has a major impact on triggers, since a missing table
referenced by an update/delete trigger would prevent an insert trigger to run.
Another problem is that stored routines might define SQL exception handlers
to deal with missing tables, but the server implementation would never give
user code a chance to execute this logic, since the routine is never
executed when a missing table cause the pre-locking code to fail.
With this fix, the internal implementation of the pre-locking code has been
relaxed of some constraints, so that failure to open a table does not
necessarily prevent execution of a stored routine.
In particular, the pre-locking mechanism is now behaving as follows:
1) the first step, to compute the transitive closure of all the tables
possibly referenced by a statement, is unchanged.
2) the next step, which is to open all the tables involved, only attempts
to open the tables added by the pre-locking code, but silently fails without
reporting any error or invoking any exception handler is the table is not
present. This is achieved by trapping internal errors with
Prelock_error_handler
3) the locking step only locks tables that were successfully opened.
4) when executing sub statements, the list of tables used by each statements
is evaluated as before. The tables needed by the sub statement are expected
to be already opened and locked. Statement referencing tables that were not
opened in step 2) will fail to find the table in the open list, and only at
this point will execution of the user code fail.
5) when a runtime exception is raised at 4), the instruction continuation
destination (the next instruction to execute in case of SQL continue
handlers) is evaluated.
This is achieved with sp_instr::exec_open_and_lock_tables()
6) if a user exception handler is present in the stored routine, that
handler is invoked as usual, so that ER_NO_SUCH_TABLE exceptions can be
trapped by stored routines. If no handler exists, then the runtime execution
will fail as expected.
With all these changes, a side effect is that view security is impacted, in
two different ways.
First, a view defined as "select stored_function()", where the stored
function references a table that may not exist, is considered valid.
The rationale is that, because the stored function might trap exceptions
during execution and still return a valid result, there is no way to decide
when the view is created if a missing table really cause the view to be invalid.
Secondly, testing for existence of tables is now done later during
execution. View security, which consist of trapping errors and return a
generic ER_VIEW_INVALID (to prevent disclosing information) was only
implemented at very specific phases covering *opening* tables, but not
covering the runtime execution. Because of this existing limitation,
errors that were previously trapped and converted into ER_VIEW_INVALID are
not trapped, causing table names to be reported to the user.
This change is exposing an existing problem, which is independent and will
be resolved separately.
2007-03-06 03:42:07 +01:00
|
|
|
{
|
2007-03-07 17:53:46 +01:00
|
|
|
return m_cont_dest;
|
Bug#8407 (Stored functions/triggers ignore exception handler)
Bug 18914 (Calling certain SPs from triggers fail)
Bug 20713 (Functions will not not continue for SQLSTATE VALUE '42S02')
Bug 21825 (Incorrect message error deleting records in a table with a
trigger for inserting)
Bug 22580 (DROP TABLE in nested stored procedure causes strange dependency
error)
Bug 25345 (Cursors from Functions)
This fix resolves a long standing issue originally reported with bug 8407,
which affect the behavior of Stored Procedures, Stored Functions and Trigger
in many different ways, causing symptoms reported by all the bugs listed.
In all cases, the root cause of the problem traces back to 8407 and how the
server locks tables involved with sub statements.
Prior to this fix, the implementation of stored routines would:
- compute the transitive closure of all the tables referenced by a top level
statement
- open and lock all the tables involved
- execute the top level statement
"transitive closure of tables" means collecting:
- all the tables,
- all the stored functions,
- all the views,
- all the table triggers
- all the stored procedures
involved, and recursively inspect these objects definition to find more
references to more objects, until the list of every object referenced does
not grow any more.
This mechanism is known as "pre-locking" tables before execution.
The motivation for locking all the tables (possibly) used at once is to
prevent dead locks.
One problem with this approach is that, if the execution path the code
really takes during runtime does not use a given table, and if the table is
missing, the server would not execute the statement.
This in particular has a major impact on triggers, since a missing table
referenced by an update/delete trigger would prevent an insert trigger to run.
Another problem is that stored routines might define SQL exception handlers
to deal with missing tables, but the server implementation would never give
user code a chance to execute this logic, since the routine is never
executed when a missing table cause the pre-locking code to fail.
With this fix, the internal implementation of the pre-locking code has been
relaxed of some constraints, so that failure to open a table does not
necessarily prevent execution of a stored routine.
In particular, the pre-locking mechanism is now behaving as follows:
1) the first step, to compute the transitive closure of all the tables
possibly referenced by a statement, is unchanged.
2) the next step, which is to open all the tables involved, only attempts
to open the tables added by the pre-locking code, but silently fails without
reporting any error or invoking any exception handler is the table is not
present. This is achieved by trapping internal errors with
Prelock_error_handler
3) the locking step only locks tables that were successfully opened.
4) when executing sub statements, the list of tables used by each statements
is evaluated as before. The tables needed by the sub statement are expected
to be already opened and locked. Statement referencing tables that were not
opened in step 2) will fail to find the table in the open list, and only at
this point will execution of the user code fail.
5) when a runtime exception is raised at 4), the instruction continuation
destination (the next instruction to execute in case of SQL continue
handlers) is evaluated.
This is achieved with sp_instr::exec_open_and_lock_tables()
6) if a user exception handler is present in the stored routine, that
handler is invoked as usual, so that ER_NO_SUCH_TABLE exceptions can be
trapped by stored routines. If no handler exists, then the runtime execution
will fail as expected.
With all these changes, a side effect is that view security is impacted, in
two different ways.
First, a view defined as "select stored_function()", where the stored
function references a table that may not exist, is considered valid.
The rationale is that, because the stored function might trap exceptions
during execution and still return a valid result, there is no way to decide
when the view is created if a missing table really cause the view to be invalid.
Secondly, testing for existence of tables is now done later during
execution. View security, which consist of trapping errors and return a
generic ER_VIEW_INVALID (to prevent disclosing information) was only
implemented at very specific phases covering *opening* tables, but not
covering the runtime execution. Because of this existing limitation,
errors that were previously trapped and converted into ER_VIEW_INVALID are
not trapped, causing table names to be reported to the user.
This change is exposing an existing problem, which is independent and will
be resolved separately.
2007-03-06 03:42:07 +01:00
|
|
|
}
|
|
|
|
|
2005-08-11 14:58:15 +02:00
|
|
|
|
|
|
|
/*
|
|
|
|
sp_instr_jump class functions
|
|
|
|
*/
|
|
|
|
|
2003-03-05 19:45:17 +01:00
|
|
|
int
|
|
|
|
sp_instr_jump::execute(THD *thd, uint *nextp)
|
|
|
|
{
|
|
|
|
DBUG_ENTER("sp_instr_jump::execute");
|
|
|
|
DBUG_PRINT("info", ("destination: %u", m_dest));
|
|
|
|
|
|
|
|
*nextp= m_dest;
|
|
|
|
DBUG_RETURN(0);
|
|
|
|
}
|
|
|
|
|
2004-03-29 11:16:45 +02:00
|
|
|
void
|
|
|
|
sp_instr_jump::print(String *str)
|
|
|
|
{
|
2005-11-22 13:06:52 +01:00
|
|
|
/* jump dest */
|
|
|
|
if (str->reserve(SP_INSTR_UINT_MAXLEN+5))
|
2005-11-18 16:30:27 +01:00
|
|
|
return;
|
2005-11-22 13:24:53 +01:00
|
|
|
str->qs_append(STRING_WITH_LEN("jump "));
|
2004-03-29 11:16:45 +02:00
|
|
|
str->qs_append(m_dest);
|
|
|
|
}
|
|
|
|
|
2004-08-02 18:05:31 +02:00
|
|
|
uint
|
Bug#19194 (Right recursion in parser for CASE causes excessive stack usage,
limitation)
Note to the reviewer
====================
Warning: reviewing this patch is somewhat involved.
Due to the nature of several issues all affecting the same area,
fixing separately each issue is not practical, since each fix can not be
implemented and tested independently.
In particular, the issues with
- rule recursion
- nested case statements
- forward jump resolution (backpatch list)
are tightly coupled (see below).
Definitions
===========
The expression
CASE expr
WHEN expr THEN expr
WHEN expr THEN expr
...
END
is a "Simple Case Expression".
The expression
CASE
WHEN expr THEN expr
WHEN expr THEN expr
...
END
is a "Searched Case Expression".
The statement
CASE expr
WHEN expr THEN stmts
WHEN expr THEN stmts
...
END CASE
is a "Simple Case Statement".
The statement
CASE
WHEN expr THEN stmts
WHEN expr THEN stmts
...
END CASE
is a "Searched Case Statement".
A "Left Recursive" rule is like
list:
element
| list element
;
A "Right Recursive" rule is like
list:
element
| element list
;
Left and right recursion produces the same language, the difference only
affects the *order* in which the text is parsed.
In a descendant parser (usually written manually), right recursion works
very well, and is typically implemented with a while loop.
In an ascendant parser (yacc/bison) left recursion works very well,
and is implemented naturally by the parser stack.
In both cases, using the wrong type or recursion is very bad and should be
avoided, as it causes technical issues with the parser implementation.
Before this change
==================
The "Simple Case Expression" and "Searched Case Expression" were both
implemented by the "when_list" and "when_list2" rules, which are left
recursive (ok).
These rules, however, used lex->when_list instead of using the parser stack,
which is more complex that necessary, and potentially dangerous because
of other rules using THD::reset_lex.
The "Simple Case Statement" and "Searched Case Statements" were implemented
by the "sp_case", "sp_whens" and in part by "sp_proc_stmt" rules.
Both cases were right recursive (bad).
The grammar involved was convoluted, and is assumed to be the results of
tweaks to get the code generation to work, but is not what someone would
naturally write.
In addition, using a common rule for both "Simple" and "Searched" case
statements was implemented with sp_head::m_flags |= IN_SIMPLE_CASE,
which is a flag and not a stack, and therefore does not take into account
*nested* case statements. This leads to incorrect generated code, and either
a server crash or an incorrect result.
With regards to the backpatch mechanism, a *different* backpatch list was
created for each jump from "WHEN expr THEN stmt" to "END CASE", which
relied on the grammar to be right recursive.
This is a mis-use of the backpatch list, since this list can resolve
multiple references to the same target at once.
The optimizer algorithm used to detect dead code in the "assembly" SQL
instructions, implemented by sp_head::opt_mark(uint ip), was recursive
in some cases (a conditional jump pointing forward to another conditional
jump).
In case of specially crafted code, like
- a long list of "IF expr THEN stmt END IF"
- a long CASE statement
this would actually cause a server crash with a stack overflow.
In general, having a stack that grows proportionally with user data (the
SQL code given by the client in a CREATE PROCEDURE) is to be avoided.
In debug builds only, creating a SP / SF / Trigger which had a significant
amount of code would spend --literally-- several minutes in sp_head::create,
because of the debug code involved with DBUG_PRINT("info", ("Code %s ...
There are several issues with this code:
- in a CASE with 5 000 WHEN, there are 15 000 instructions generated,
which create a sting representation of the code which is 500 000 bytes
long,
- using a String instead of an io stream causes performances to degrade
to a total server freeze, as time is spent doing realloc of a buffer
always too short,
- Printing a 500 000 long string in the debug log is too verbose,
- Generating this string even when DBUG_PRINT is off is useless,
- Having code that potentially can affect the server behavior, used with
#ifdef / #endif is useful in some cases, but is also a bad practice.
After this change
=================
"Case Expressions" (both simple and searched) have been simplified to
not use LEX::when_list, which has been removed.
Considering all the issues affecting case statements, the grammar for these
has been totally re written.
The existing actions, used to generate "assembly" sp_inst* code, have been
preserved but moved in the new grammar, with the following changes:
a) Bison rules are no longer shared between "Simple" and "Searched" case
statements, because a stack instead of a flag is required to handle them.
Nested statements are handled naturally by the parser stack, which by
definition uses the correct rule in the correct context.
Nested statements of the opposite type (simple vs searched) works correctly.
The flag sp_head::IN_SIMPLE_CASE is no longer used.
This is a step towards resolution of WL#2999, which correctly identified
that temporary parsing flags do not belong to sp_head.
The code in the action is shared by mean of the case_stmt_action_xxx()
helpers.
b) The backpatch mechanism, used to resolve forward jumps in the generated
code, has been changed to:
- create a label for the instruction following 'END CASE',
- register each jump at the end of a "WHEN expr THEN stmt" in a *unique*
backpatch list associated with the 'END CASE' label
- resolve all the forward jumps for this label at once.
In addition, the code involving backpatch has been commented, so that a
reader can now understand by reading matching "Registering" and "Resolving"
comments how the forward jumps are resolved and what target they resolve to,
as this is far from evident when reading the code alone.
The implementation of sp_head::opt_mark() has been revised to avoid
recursive calls from jump instructions, and instead add the jump location
to the list of paths to explore during the flow analysis of the instruction
graph, with a call to sp_head::add_mark_lead().
In addition, the flow analysis will stop if an instruction has already
been marked as reachable, which the previous code failed to do in the
recursive case.
sp_head::opt_mark() is now private, to prevent new calls to this method from
being introduced.
The debug code present in sp_head::create() has been removed.
Considering that SHOW PROCEDURE CODE is also available in debug builds,
and can be used anytime regardless of the trace level, as opposed to
"CREATE PROCEDURE" time and only if the trace was on,
removing the code actually makes debugging easier (usable trace).
Tests have been written to cover the parser overflow (big CASE),
and to cover nested CASE statements.
2006-11-17 20:14:29 +01:00
|
|
|
sp_instr_jump::opt_mark(sp_head *sp, List<sp_instr> *leads)
|
2004-08-02 18:05:31 +02:00
|
|
|
{
|
2004-08-26 12:54:30 +02:00
|
|
|
m_dest= opt_shortcut_jump(sp, this);
|
2004-08-17 20:20:58 +02:00
|
|
|
if (m_dest != m_ip+1) /* Jumping to following instruction? */
|
|
|
|
marked= 1;
|
2004-08-02 18:05:31 +02:00
|
|
|
m_optdest= sp->get_instr(m_dest);
|
|
|
|
return m_dest;
|
|
|
|
}
|
|
|
|
|
|
|
|
uint
|
2004-08-26 12:54:30 +02:00
|
|
|
sp_instr_jump::opt_shortcut_jump(sp_head *sp, sp_instr *start)
|
2004-08-02 18:05:31 +02:00
|
|
|
{
|
|
|
|
uint dest= m_dest;
|
|
|
|
sp_instr *i;
|
|
|
|
|
|
|
|
while ((i= sp->get_instr(dest)))
|
|
|
|
{
|
2004-08-26 12:54:30 +02:00
|
|
|
uint ndest;
|
2004-08-02 18:05:31 +02:00
|
|
|
|
2005-04-20 15:37:07 +02:00
|
|
|
if (start == i || this == i)
|
2004-08-26 12:54:30 +02:00
|
|
|
break;
|
|
|
|
ndest= i->opt_shortcut_jump(sp, start);
|
2004-08-02 18:05:31 +02:00
|
|
|
if (ndest == dest)
|
|
|
|
break;
|
|
|
|
dest= ndest;
|
|
|
|
}
|
|
|
|
return dest;
|
|
|
|
}
|
|
|
|
|
|
|
|
void
|
|
|
|
sp_instr_jump::opt_move(uint dst, List<sp_instr> *bp)
|
|
|
|
{
|
|
|
|
if (m_dest > m_ip)
|
|
|
|
bp->push_back(this); // Forward
|
|
|
|
else if (m_optdest)
|
|
|
|
m_dest= m_optdest->m_ip; // Backward
|
|
|
|
m_ip= dst;
|
|
|
|
}
|
|
|
|
|
2005-08-11 14:58:15 +02:00
|
|
|
|
|
|
|
/*
|
|
|
|
sp_instr_jump_if_not class functions
|
|
|
|
*/
|
|
|
|
|
2002-12-11 14:24:29 +01:00
|
|
|
int
|
|
|
|
sp_instr_jump_if_not::execute(THD *thd, uint *nextp)
|
|
|
|
{
|
2003-02-12 16:17:03 +01:00
|
|
|
DBUG_ENTER("sp_instr_jump_if_not::execute");
|
|
|
|
DBUG_PRINT("info", ("destination: %u", m_dest));
|
2005-03-04 14:35:28 +01:00
|
|
|
DBUG_RETURN(m_lex_keeper.reset_lex_and_exec_core(thd, nextp, TRUE, this));
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
int
|
|
|
|
sp_instr_jump_if_not::exec_core(THD *thd, uint *nextp)
|
|
|
|
{
|
2004-08-24 16:07:39 +02:00
|
|
|
Item *it;
|
|
|
|
int res;
|
2002-12-11 14:24:29 +01:00
|
|
|
|
2005-05-09 00:59:10 +02:00
|
|
|
it= sp_prepare_func_item(thd, &m_expr);
|
2004-08-24 16:07:39 +02:00
|
|
|
if (! it)
|
2005-11-04 15:37:39 +01:00
|
|
|
{
|
2004-08-24 16:07:39 +02:00
|
|
|
res= -1;
|
2005-11-04 15:37:39 +01:00
|
|
|
}
|
2002-12-11 14:24:29 +01:00
|
|
|
else
|
2004-08-24 16:07:39 +02:00
|
|
|
{
|
|
|
|
res= 0;
|
2005-05-09 00:59:10 +02:00
|
|
|
if (! it->val_bool())
|
2004-08-24 16:07:39 +02:00
|
|
|
*nextp = m_dest;
|
|
|
|
else
|
|
|
|
*nextp = m_ip+1;
|
|
|
|
}
|
2005-03-04 14:35:28 +01:00
|
|
|
|
|
|
|
return res;
|
Simplistic, experimental framework for Stored Procedures (SPs).
Implements creation and dropping of PROCEDUREs, IN, OUT, and INOUT parameters,
single-statement procedures, rudimentary multi-statement (begin-end) prodedures
(when the client can handle it), and local variables.
Missing most of the embedded SQL language, all attributes, FUNCTIONs, error handling,
reparses procedures at each call (no caching), etc, etc.
Certainly buggy too, but procedures can actually be created and called....
2002-12-08 19:59:22 +01:00
|
|
|
}
|
2003-02-26 19:22:29 +01:00
|
|
|
|
2005-08-11 14:58:15 +02:00
|
|
|
|
2004-03-29 11:16:45 +02:00
|
|
|
void
|
|
|
|
sp_instr_jump_if_not::print(String *str)
|
|
|
|
{
|
2006-01-26 17:26:25 +01:00
|
|
|
/* jump_if_not dest(cont) ... */
|
2006-01-16 15:37:25 +01:00
|
|
|
if (str->reserve(2*SP_INSTR_UINT_MAXLEN+14+32)) // Add some for the expr. too
|
2005-11-18 16:30:27 +01:00
|
|
|
return;
|
2005-11-22 13:24:53 +01:00
|
|
|
str->qs_append(STRING_WITH_LEN("jump_if_not "));
|
2004-03-29 11:16:45 +02:00
|
|
|
str->qs_append(m_dest);
|
2006-01-26 17:26:25 +01:00
|
|
|
str->qs_append('(');
|
2005-11-04 15:37:39 +01:00
|
|
|
str->qs_append(m_cont_dest);
|
2006-01-26 17:26:25 +01:00
|
|
|
str->qs_append(STRING_WITH_LEN(") "));
|
2004-03-29 11:16:45 +02:00
|
|
|
m_expr->print(str);
|
|
|
|
}
|
|
|
|
|
2005-08-11 14:58:15 +02:00
|
|
|
|
2004-08-02 18:05:31 +02:00
|
|
|
uint
|
Bug#19194 (Right recursion in parser for CASE causes excessive stack usage,
limitation)
Note to the reviewer
====================
Warning: reviewing this patch is somewhat involved.
Due to the nature of several issues all affecting the same area,
fixing separately each issue is not practical, since each fix can not be
implemented and tested independently.
In particular, the issues with
- rule recursion
- nested case statements
- forward jump resolution (backpatch list)
are tightly coupled (see below).
Definitions
===========
The expression
CASE expr
WHEN expr THEN expr
WHEN expr THEN expr
...
END
is a "Simple Case Expression".
The expression
CASE
WHEN expr THEN expr
WHEN expr THEN expr
...
END
is a "Searched Case Expression".
The statement
CASE expr
WHEN expr THEN stmts
WHEN expr THEN stmts
...
END CASE
is a "Simple Case Statement".
The statement
CASE
WHEN expr THEN stmts
WHEN expr THEN stmts
...
END CASE
is a "Searched Case Statement".
A "Left Recursive" rule is like
list:
element
| list element
;
A "Right Recursive" rule is like
list:
element
| element list
;
Left and right recursion produces the same language, the difference only
affects the *order* in which the text is parsed.
In a descendant parser (usually written manually), right recursion works
very well, and is typically implemented with a while loop.
In an ascendant parser (yacc/bison) left recursion works very well,
and is implemented naturally by the parser stack.
In both cases, using the wrong type or recursion is very bad and should be
avoided, as it causes technical issues with the parser implementation.
Before this change
==================
The "Simple Case Expression" and "Searched Case Expression" were both
implemented by the "when_list" and "when_list2" rules, which are left
recursive (ok).
These rules, however, used lex->when_list instead of using the parser stack,
which is more complex that necessary, and potentially dangerous because
of other rules using THD::reset_lex.
The "Simple Case Statement" and "Searched Case Statements" were implemented
by the "sp_case", "sp_whens" and in part by "sp_proc_stmt" rules.
Both cases were right recursive (bad).
The grammar involved was convoluted, and is assumed to be the results of
tweaks to get the code generation to work, but is not what someone would
naturally write.
In addition, using a common rule for both "Simple" and "Searched" case
statements was implemented with sp_head::m_flags |= IN_SIMPLE_CASE,
which is a flag and not a stack, and therefore does not take into account
*nested* case statements. This leads to incorrect generated code, and either
a server crash or an incorrect result.
With regards to the backpatch mechanism, a *different* backpatch list was
created for each jump from "WHEN expr THEN stmt" to "END CASE", which
relied on the grammar to be right recursive.
This is a mis-use of the backpatch list, since this list can resolve
multiple references to the same target at once.
The optimizer algorithm used to detect dead code in the "assembly" SQL
instructions, implemented by sp_head::opt_mark(uint ip), was recursive
in some cases (a conditional jump pointing forward to another conditional
jump).
In case of specially crafted code, like
- a long list of "IF expr THEN stmt END IF"
- a long CASE statement
this would actually cause a server crash with a stack overflow.
In general, having a stack that grows proportionally with user data (the
SQL code given by the client in a CREATE PROCEDURE) is to be avoided.
In debug builds only, creating a SP / SF / Trigger which had a significant
amount of code would spend --literally-- several minutes in sp_head::create,
because of the debug code involved with DBUG_PRINT("info", ("Code %s ...
There are several issues with this code:
- in a CASE with 5 000 WHEN, there are 15 000 instructions generated,
which create a sting representation of the code which is 500 000 bytes
long,
- using a String instead of an io stream causes performances to degrade
to a total server freeze, as time is spent doing realloc of a buffer
always too short,
- Printing a 500 000 long string in the debug log is too verbose,
- Generating this string even when DBUG_PRINT is off is useless,
- Having code that potentially can affect the server behavior, used with
#ifdef / #endif is useful in some cases, but is also a bad practice.
After this change
=================
"Case Expressions" (both simple and searched) have been simplified to
not use LEX::when_list, which has been removed.
Considering all the issues affecting case statements, the grammar for these
has been totally re written.
The existing actions, used to generate "assembly" sp_inst* code, have been
preserved but moved in the new grammar, with the following changes:
a) Bison rules are no longer shared between "Simple" and "Searched" case
statements, because a stack instead of a flag is required to handle them.
Nested statements are handled naturally by the parser stack, which by
definition uses the correct rule in the correct context.
Nested statements of the opposite type (simple vs searched) works correctly.
The flag sp_head::IN_SIMPLE_CASE is no longer used.
This is a step towards resolution of WL#2999, which correctly identified
that temporary parsing flags do not belong to sp_head.
The code in the action is shared by mean of the case_stmt_action_xxx()
helpers.
b) The backpatch mechanism, used to resolve forward jumps in the generated
code, has been changed to:
- create a label for the instruction following 'END CASE',
- register each jump at the end of a "WHEN expr THEN stmt" in a *unique*
backpatch list associated with the 'END CASE' label
- resolve all the forward jumps for this label at once.
In addition, the code involving backpatch has been commented, so that a
reader can now understand by reading matching "Registering" and "Resolving"
comments how the forward jumps are resolved and what target they resolve to,
as this is far from evident when reading the code alone.
The implementation of sp_head::opt_mark() has been revised to avoid
recursive calls from jump instructions, and instead add the jump location
to the list of paths to explore during the flow analysis of the instruction
graph, with a call to sp_head::add_mark_lead().
In addition, the flow analysis will stop if an instruction has already
been marked as reachable, which the previous code failed to do in the
recursive case.
sp_head::opt_mark() is now private, to prevent new calls to this method from
being introduced.
The debug code present in sp_head::create() has been removed.
Considering that SHOW PROCEDURE CODE is also available in debug builds,
and can be used anytime regardless of the trace level, as opposed to
"CREATE PROCEDURE" time and only if the trace was on,
removing the code actually makes debugging easier (usable trace).
Tests have been written to cover the parser overflow (big CASE),
and to cover nested CASE statements.
2006-11-17 20:14:29 +01:00
|
|
|
sp_instr_jump_if_not::opt_mark(sp_head *sp, List<sp_instr> *leads)
|
2004-08-02 18:05:31 +02:00
|
|
|
{
|
|
|
|
sp_instr *i;
|
|
|
|
|
|
|
|
marked= 1;
|
|
|
|
if ((i= sp->get_instr(m_dest)))
|
|
|
|
{
|
2004-08-26 12:54:30 +02:00
|
|
|
m_dest= i->opt_shortcut_jump(sp, this);
|
2004-08-02 18:05:31 +02:00
|
|
|
m_optdest= sp->get_instr(m_dest);
|
|
|
|
}
|
Bug#19194 (Right recursion in parser for CASE causes excessive stack usage,
limitation)
Note to the reviewer
====================
Warning: reviewing this patch is somewhat involved.
Due to the nature of several issues all affecting the same area,
fixing separately each issue is not practical, since each fix can not be
implemented and tested independently.
In particular, the issues with
- rule recursion
- nested case statements
- forward jump resolution (backpatch list)
are tightly coupled (see below).
Definitions
===========
The expression
CASE expr
WHEN expr THEN expr
WHEN expr THEN expr
...
END
is a "Simple Case Expression".
The expression
CASE
WHEN expr THEN expr
WHEN expr THEN expr
...
END
is a "Searched Case Expression".
The statement
CASE expr
WHEN expr THEN stmts
WHEN expr THEN stmts
...
END CASE
is a "Simple Case Statement".
The statement
CASE
WHEN expr THEN stmts
WHEN expr THEN stmts
...
END CASE
is a "Searched Case Statement".
A "Left Recursive" rule is like
list:
element
| list element
;
A "Right Recursive" rule is like
list:
element
| element list
;
Left and right recursion produces the same language, the difference only
affects the *order* in which the text is parsed.
In a descendant parser (usually written manually), right recursion works
very well, and is typically implemented with a while loop.
In an ascendant parser (yacc/bison) left recursion works very well,
and is implemented naturally by the parser stack.
In both cases, using the wrong type or recursion is very bad and should be
avoided, as it causes technical issues with the parser implementation.
Before this change
==================
The "Simple Case Expression" and "Searched Case Expression" were both
implemented by the "when_list" and "when_list2" rules, which are left
recursive (ok).
These rules, however, used lex->when_list instead of using the parser stack,
which is more complex that necessary, and potentially dangerous because
of other rules using THD::reset_lex.
The "Simple Case Statement" and "Searched Case Statements" were implemented
by the "sp_case", "sp_whens" and in part by "sp_proc_stmt" rules.
Both cases were right recursive (bad).
The grammar involved was convoluted, and is assumed to be the results of
tweaks to get the code generation to work, but is not what someone would
naturally write.
In addition, using a common rule for both "Simple" and "Searched" case
statements was implemented with sp_head::m_flags |= IN_SIMPLE_CASE,
which is a flag and not a stack, and therefore does not take into account
*nested* case statements. This leads to incorrect generated code, and either
a server crash or an incorrect result.
With regards to the backpatch mechanism, a *different* backpatch list was
created for each jump from "WHEN expr THEN stmt" to "END CASE", which
relied on the grammar to be right recursive.
This is a mis-use of the backpatch list, since this list can resolve
multiple references to the same target at once.
The optimizer algorithm used to detect dead code in the "assembly" SQL
instructions, implemented by sp_head::opt_mark(uint ip), was recursive
in some cases (a conditional jump pointing forward to another conditional
jump).
In case of specially crafted code, like
- a long list of "IF expr THEN stmt END IF"
- a long CASE statement
this would actually cause a server crash with a stack overflow.
In general, having a stack that grows proportionally with user data (the
SQL code given by the client in a CREATE PROCEDURE) is to be avoided.
In debug builds only, creating a SP / SF / Trigger which had a significant
amount of code would spend --literally-- several minutes in sp_head::create,
because of the debug code involved with DBUG_PRINT("info", ("Code %s ...
There are several issues with this code:
- in a CASE with 5 000 WHEN, there are 15 000 instructions generated,
which create a sting representation of the code which is 500 000 bytes
long,
- using a String instead of an io stream causes performances to degrade
to a total server freeze, as time is spent doing realloc of a buffer
always too short,
- Printing a 500 000 long string in the debug log is too verbose,
- Generating this string even when DBUG_PRINT is off is useless,
- Having code that potentially can affect the server behavior, used with
#ifdef / #endif is useful in some cases, but is also a bad practice.
After this change
=================
"Case Expressions" (both simple and searched) have been simplified to
not use LEX::when_list, which has been removed.
Considering all the issues affecting case statements, the grammar for these
has been totally re written.
The existing actions, used to generate "assembly" sp_inst* code, have been
preserved but moved in the new grammar, with the following changes:
a) Bison rules are no longer shared between "Simple" and "Searched" case
statements, because a stack instead of a flag is required to handle them.
Nested statements are handled naturally by the parser stack, which by
definition uses the correct rule in the correct context.
Nested statements of the opposite type (simple vs searched) works correctly.
The flag sp_head::IN_SIMPLE_CASE is no longer used.
This is a step towards resolution of WL#2999, which correctly identified
that temporary parsing flags do not belong to sp_head.
The code in the action is shared by mean of the case_stmt_action_xxx()
helpers.
b) The backpatch mechanism, used to resolve forward jumps in the generated
code, has been changed to:
- create a label for the instruction following 'END CASE',
- register each jump at the end of a "WHEN expr THEN stmt" in a *unique*
backpatch list associated with the 'END CASE' label
- resolve all the forward jumps for this label at once.
In addition, the code involving backpatch has been commented, so that a
reader can now understand by reading matching "Registering" and "Resolving"
comments how the forward jumps are resolved and what target they resolve to,
as this is far from evident when reading the code alone.
The implementation of sp_head::opt_mark() has been revised to avoid
recursive calls from jump instructions, and instead add the jump location
to the list of paths to explore during the flow analysis of the instruction
graph, with a call to sp_head::add_mark_lead().
In addition, the flow analysis will stop if an instruction has already
been marked as reachable, which the previous code failed to do in the
recursive case.
sp_head::opt_mark() is now private, to prevent new calls to this method from
being introduced.
The debug code present in sp_head::create() has been removed.
Considering that SHOW PROCEDURE CODE is also available in debug builds,
and can be used anytime regardless of the trace level, as opposed to
"CREATE PROCEDURE" time and only if the trace was on,
removing the code actually makes debugging easier (usable trace).
Tests have been written to cover the parser overflow (big CASE),
and to cover nested CASE statements.
2006-11-17 20:14:29 +01:00
|
|
|
sp->add_mark_lead(m_dest, leads);
|
2005-11-04 15:37:39 +01:00
|
|
|
if ((i= sp->get_instr(m_cont_dest)))
|
|
|
|
{
|
|
|
|
m_cont_dest= i->opt_shortcut_jump(sp, this);
|
|
|
|
m_cont_optdest= sp->get_instr(m_cont_dest);
|
|
|
|
}
|
Bug#19194 (Right recursion in parser for CASE causes excessive stack usage,
limitation)
Note to the reviewer
====================
Warning: reviewing this patch is somewhat involved.
Due to the nature of several issues all affecting the same area,
fixing separately each issue is not practical, since each fix can not be
implemented and tested independently.
In particular, the issues with
- rule recursion
- nested case statements
- forward jump resolution (backpatch list)
are tightly coupled (see below).
Definitions
===========
The expression
CASE expr
WHEN expr THEN expr
WHEN expr THEN expr
...
END
is a "Simple Case Expression".
The expression
CASE
WHEN expr THEN expr
WHEN expr THEN expr
...
END
is a "Searched Case Expression".
The statement
CASE expr
WHEN expr THEN stmts
WHEN expr THEN stmts
...
END CASE
is a "Simple Case Statement".
The statement
CASE
WHEN expr THEN stmts
WHEN expr THEN stmts
...
END CASE
is a "Searched Case Statement".
A "Left Recursive" rule is like
list:
element
| list element
;
A "Right Recursive" rule is like
list:
element
| element list
;
Left and right recursion produces the same language, the difference only
affects the *order* in which the text is parsed.
In a descendant parser (usually written manually), right recursion works
very well, and is typically implemented with a while loop.
In an ascendant parser (yacc/bison) left recursion works very well,
and is implemented naturally by the parser stack.
In both cases, using the wrong type or recursion is very bad and should be
avoided, as it causes technical issues with the parser implementation.
Before this change
==================
The "Simple Case Expression" and "Searched Case Expression" were both
implemented by the "when_list" and "when_list2" rules, which are left
recursive (ok).
These rules, however, used lex->when_list instead of using the parser stack,
which is more complex that necessary, and potentially dangerous because
of other rules using THD::reset_lex.
The "Simple Case Statement" and "Searched Case Statements" were implemented
by the "sp_case", "sp_whens" and in part by "sp_proc_stmt" rules.
Both cases were right recursive (bad).
The grammar involved was convoluted, and is assumed to be the results of
tweaks to get the code generation to work, but is not what someone would
naturally write.
In addition, using a common rule for both "Simple" and "Searched" case
statements was implemented with sp_head::m_flags |= IN_SIMPLE_CASE,
which is a flag and not a stack, and therefore does not take into account
*nested* case statements. This leads to incorrect generated code, and either
a server crash or an incorrect result.
With regards to the backpatch mechanism, a *different* backpatch list was
created for each jump from "WHEN expr THEN stmt" to "END CASE", which
relied on the grammar to be right recursive.
This is a mis-use of the backpatch list, since this list can resolve
multiple references to the same target at once.
The optimizer algorithm used to detect dead code in the "assembly" SQL
instructions, implemented by sp_head::opt_mark(uint ip), was recursive
in some cases (a conditional jump pointing forward to another conditional
jump).
In case of specially crafted code, like
- a long list of "IF expr THEN stmt END IF"
- a long CASE statement
this would actually cause a server crash with a stack overflow.
In general, having a stack that grows proportionally with user data (the
SQL code given by the client in a CREATE PROCEDURE) is to be avoided.
In debug builds only, creating a SP / SF / Trigger which had a significant
amount of code would spend --literally-- several minutes in sp_head::create,
because of the debug code involved with DBUG_PRINT("info", ("Code %s ...
There are several issues with this code:
- in a CASE with 5 000 WHEN, there are 15 000 instructions generated,
which create a sting representation of the code which is 500 000 bytes
long,
- using a String instead of an io stream causes performances to degrade
to a total server freeze, as time is spent doing realloc of a buffer
always too short,
- Printing a 500 000 long string in the debug log is too verbose,
- Generating this string even when DBUG_PRINT is off is useless,
- Having code that potentially can affect the server behavior, used with
#ifdef / #endif is useful in some cases, but is also a bad practice.
After this change
=================
"Case Expressions" (both simple and searched) have been simplified to
not use LEX::when_list, which has been removed.
Considering all the issues affecting case statements, the grammar for these
has been totally re written.
The existing actions, used to generate "assembly" sp_inst* code, have been
preserved but moved in the new grammar, with the following changes:
a) Bison rules are no longer shared between "Simple" and "Searched" case
statements, because a stack instead of a flag is required to handle them.
Nested statements are handled naturally by the parser stack, which by
definition uses the correct rule in the correct context.
Nested statements of the opposite type (simple vs searched) works correctly.
The flag sp_head::IN_SIMPLE_CASE is no longer used.
This is a step towards resolution of WL#2999, which correctly identified
that temporary parsing flags do not belong to sp_head.
The code in the action is shared by mean of the case_stmt_action_xxx()
helpers.
b) The backpatch mechanism, used to resolve forward jumps in the generated
code, has been changed to:
- create a label for the instruction following 'END CASE',
- register each jump at the end of a "WHEN expr THEN stmt" in a *unique*
backpatch list associated with the 'END CASE' label
- resolve all the forward jumps for this label at once.
In addition, the code involving backpatch has been commented, so that a
reader can now understand by reading matching "Registering" and "Resolving"
comments how the forward jumps are resolved and what target they resolve to,
as this is far from evident when reading the code alone.
The implementation of sp_head::opt_mark() has been revised to avoid
recursive calls from jump instructions, and instead add the jump location
to the list of paths to explore during the flow analysis of the instruction
graph, with a call to sp_head::add_mark_lead().
In addition, the flow analysis will stop if an instruction has already
been marked as reachable, which the previous code failed to do in the
recursive case.
sp_head::opt_mark() is now private, to prevent new calls to this method from
being introduced.
The debug code present in sp_head::create() has been removed.
Considering that SHOW PROCEDURE CODE is also available in debug builds,
and can be used anytime regardless of the trace level, as opposed to
"CREATE PROCEDURE" time and only if the trace was on,
removing the code actually makes debugging easier (usable trace).
Tests have been written to cover the parser overflow (big CASE),
and to cover nested CASE statements.
2006-11-17 20:14:29 +01:00
|
|
|
sp->add_mark_lead(m_cont_dest, leads);
|
2004-08-02 18:05:31 +02:00
|
|
|
return m_ip+1;
|
|
|
|
}
|
|
|
|
|
2005-11-04 15:37:39 +01:00
|
|
|
void
|
|
|
|
sp_instr_jump_if_not::opt_move(uint dst, List<sp_instr> *bp)
|
|
|
|
{
|
|
|
|
/*
|
|
|
|
cont. destinations may point backwards after shortcutting jumps
|
|
|
|
during the mark phase. If it's still pointing forwards, only
|
|
|
|
push this for backpatching if sp_instr_jump::opt_move() will not
|
|
|
|
do it (i.e. if the m_dest points backwards).
|
|
|
|
*/
|
|
|
|
if (m_cont_dest > m_ip)
|
|
|
|
{ // Forward
|
|
|
|
if (m_dest < m_ip)
|
|
|
|
bp->push_back(this);
|
|
|
|
}
|
|
|
|
else if (m_cont_optdest)
|
|
|
|
m_cont_dest= m_cont_optdest->m_ip; // Backward
|
|
|
|
/* This will take care of m_dest and m_ip */
|
|
|
|
sp_instr_jump::opt_move(dst, bp);
|
|
|
|
}
|
|
|
|
|
2005-08-11 14:58:15 +02:00
|
|
|
|
|
|
|
/*
|
|
|
|
sp_instr_freturn class functions
|
|
|
|
*/
|
2005-03-04 14:35:28 +01:00
|
|
|
|
2003-02-26 19:22:29 +01:00
|
|
|
int
|
2003-09-16 14:26:08 +02:00
|
|
|
sp_instr_freturn::execute(THD *thd, uint *nextp)
|
2003-02-26 19:22:29 +01:00
|
|
|
{
|
2003-09-16 14:26:08 +02:00
|
|
|
DBUG_ENTER("sp_instr_freturn::execute");
|
2005-03-04 14:35:28 +01:00
|
|
|
DBUG_RETURN(m_lex_keeper.reset_lex_and_exec_core(thd, nextp, TRUE, this));
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
int
|
|
|
|
sp_instr_freturn::exec_core(THD *thd, uint *nextp)
|
|
|
|
{
|
2005-12-07 15:01:17 +01:00
|
|
|
/*
|
|
|
|
Change <next instruction pointer>, so that this will be the last
|
|
|
|
instruction in the stored function.
|
|
|
|
*/
|
2004-07-21 14:53:09 +02:00
|
|
|
|
2003-02-26 19:22:29 +01:00
|
|
|
*nextp= UINT_MAX;
|
2005-03-04 14:35:28 +01:00
|
|
|
|
2005-12-07 15:01:17 +01:00
|
|
|
/*
|
|
|
|
Evaluate the value of return expression and store it in current runtime
|
|
|
|
context.
|
|
|
|
|
|
|
|
NOTE: It's necessary to evaluate result item right here, because we must
|
|
|
|
do it in scope of execution the current context/block.
|
|
|
|
*/
|
|
|
|
|
2006-05-15 12:01:55 +02:00
|
|
|
return thd->spcont->set_return_value(thd, &m_value);
|
2003-02-26 19:22:29 +01:00
|
|
|
}
|
2003-09-16 14:26:08 +02:00
|
|
|
|
2004-03-29 11:16:45 +02:00
|
|
|
void
|
|
|
|
sp_instr_freturn::print(String *str)
|
|
|
|
{
|
2005-11-22 13:06:52 +01:00
|
|
|
/* freturn type expr... */
|
2006-11-30 17:25:05 +01:00
|
|
|
if (str->reserve(1024+8+32)) // Add some for the expr. too
|
2005-11-18 16:30:27 +01:00
|
|
|
return;
|
2005-11-22 13:24:53 +01:00
|
|
|
str->qs_append(STRING_WITH_LEN("freturn "));
|
2004-04-06 13:26:53 +02:00
|
|
|
str->qs_append((uint)m_type);
|
2005-11-18 16:30:27 +01:00
|
|
|
str->qs_append(' ');
|
2004-03-29 11:16:45 +02:00
|
|
|
m_value->print(str);
|
|
|
|
}
|
|
|
|
|
2005-08-11 14:58:15 +02:00
|
|
|
/*
|
|
|
|
sp_instr_hpush_jump class functions
|
|
|
|
*/
|
|
|
|
|
2003-09-16 14:26:08 +02:00
|
|
|
int
|
|
|
|
sp_instr_hpush_jump::execute(THD *thd, uint *nextp)
|
|
|
|
{
|
|
|
|
DBUG_ENTER("sp_instr_hpush_jump::execute");
|
|
|
|
List_iterator_fast<sp_cond_type_t> li(m_cond);
|
|
|
|
sp_cond_type_t *p;
|
|
|
|
|
|
|
|
while ((p= li++))
|
2005-08-19 15:03:21 +02:00
|
|
|
thd->spcont->push_handler(p, m_ip+1, m_type, m_frame);
|
2003-09-16 14:26:08 +02:00
|
|
|
|
|
|
|
*nextp= m_dest;
|
|
|
|
DBUG_RETURN(0);
|
|
|
|
}
|
|
|
|
|
2006-01-05 23:47:49 +01:00
|
|
|
|
2004-03-29 11:16:45 +02:00
|
|
|
void
|
|
|
|
sp_instr_hpush_jump::print(String *str)
|
|
|
|
{
|
2005-11-22 13:06:52 +01:00
|
|
|
/* hpush_jump dest fsize type */
|
|
|
|
if (str->reserve(SP_INSTR_UINT_MAXLEN*2 + 21))
|
2005-11-18 16:30:27 +01:00
|
|
|
return;
|
2005-11-22 13:24:53 +01:00
|
|
|
str->qs_append(STRING_WITH_LEN("hpush_jump "));
|
2004-04-05 17:01:19 +02:00
|
|
|
str->qs_append(m_dest);
|
2005-11-18 16:30:27 +01:00
|
|
|
str->qs_append(' ');
|
2004-03-29 11:16:45 +02:00
|
|
|
str->qs_append(m_frame);
|
2006-01-05 23:47:49 +01:00
|
|
|
switch (m_type) {
|
2005-11-17 11:11:48 +01:00
|
|
|
case SP_HANDLER_NONE:
|
2005-11-22 14:25:44 +01:00
|
|
|
str->qs_append(STRING_WITH_LEN(" NONE")); // This would be a bug
|
2005-11-17 11:11:48 +01:00
|
|
|
break;
|
|
|
|
case SP_HANDLER_EXIT:
|
2005-11-22 14:25:44 +01:00
|
|
|
str->qs_append(STRING_WITH_LEN(" EXIT"));
|
2005-11-17 11:11:48 +01:00
|
|
|
break;
|
|
|
|
case SP_HANDLER_CONTINUE:
|
2005-11-22 14:25:44 +01:00
|
|
|
str->qs_append(STRING_WITH_LEN(" CONTINUE"));
|
2005-11-17 11:11:48 +01:00
|
|
|
break;
|
|
|
|
case SP_HANDLER_UNDO:
|
2005-11-22 14:25:44 +01:00
|
|
|
str->qs_append(STRING_WITH_LEN(" UNDO"));
|
2005-11-17 11:11:48 +01:00
|
|
|
break;
|
|
|
|
default:
|
2006-01-05 23:47:49 +01:00
|
|
|
// This would be a bug as well
|
|
|
|
str->qs_append(STRING_WITH_LEN(" UNKNOWN:"));
|
2005-11-17 11:11:48 +01:00
|
|
|
str->qs_append(m_type);
|
|
|
|
}
|
2004-03-29 11:16:45 +02:00
|
|
|
}
|
|
|
|
|
2006-01-05 23:47:49 +01:00
|
|
|
|
2004-08-02 18:05:31 +02:00
|
|
|
uint
|
Bug#19194 (Right recursion in parser for CASE causes excessive stack usage,
limitation)
Note to the reviewer
====================
Warning: reviewing this patch is somewhat involved.
Due to the nature of several issues all affecting the same area,
fixing separately each issue is not practical, since each fix can not be
implemented and tested independently.
In particular, the issues with
- rule recursion
- nested case statements
- forward jump resolution (backpatch list)
are tightly coupled (see below).
Definitions
===========
The expression
CASE expr
WHEN expr THEN expr
WHEN expr THEN expr
...
END
is a "Simple Case Expression".
The expression
CASE
WHEN expr THEN expr
WHEN expr THEN expr
...
END
is a "Searched Case Expression".
The statement
CASE expr
WHEN expr THEN stmts
WHEN expr THEN stmts
...
END CASE
is a "Simple Case Statement".
The statement
CASE
WHEN expr THEN stmts
WHEN expr THEN stmts
...
END CASE
is a "Searched Case Statement".
A "Left Recursive" rule is like
list:
element
| list element
;
A "Right Recursive" rule is like
list:
element
| element list
;
Left and right recursion produces the same language, the difference only
affects the *order* in which the text is parsed.
In a descendant parser (usually written manually), right recursion works
very well, and is typically implemented with a while loop.
In an ascendant parser (yacc/bison) left recursion works very well,
and is implemented naturally by the parser stack.
In both cases, using the wrong type or recursion is very bad and should be
avoided, as it causes technical issues with the parser implementation.
Before this change
==================
The "Simple Case Expression" and "Searched Case Expression" were both
implemented by the "when_list" and "when_list2" rules, which are left
recursive (ok).
These rules, however, used lex->when_list instead of using the parser stack,
which is more complex that necessary, and potentially dangerous because
of other rules using THD::reset_lex.
The "Simple Case Statement" and "Searched Case Statements" were implemented
by the "sp_case", "sp_whens" and in part by "sp_proc_stmt" rules.
Both cases were right recursive (bad).
The grammar involved was convoluted, and is assumed to be the results of
tweaks to get the code generation to work, but is not what someone would
naturally write.
In addition, using a common rule for both "Simple" and "Searched" case
statements was implemented with sp_head::m_flags |= IN_SIMPLE_CASE,
which is a flag and not a stack, and therefore does not take into account
*nested* case statements. This leads to incorrect generated code, and either
a server crash or an incorrect result.
With regards to the backpatch mechanism, a *different* backpatch list was
created for each jump from "WHEN expr THEN stmt" to "END CASE", which
relied on the grammar to be right recursive.
This is a mis-use of the backpatch list, since this list can resolve
multiple references to the same target at once.
The optimizer algorithm used to detect dead code in the "assembly" SQL
instructions, implemented by sp_head::opt_mark(uint ip), was recursive
in some cases (a conditional jump pointing forward to another conditional
jump).
In case of specially crafted code, like
- a long list of "IF expr THEN stmt END IF"
- a long CASE statement
this would actually cause a server crash with a stack overflow.
In general, having a stack that grows proportionally with user data (the
SQL code given by the client in a CREATE PROCEDURE) is to be avoided.
In debug builds only, creating a SP / SF / Trigger which had a significant
amount of code would spend --literally-- several minutes in sp_head::create,
because of the debug code involved with DBUG_PRINT("info", ("Code %s ...
There are several issues with this code:
- in a CASE with 5 000 WHEN, there are 15 000 instructions generated,
which create a sting representation of the code which is 500 000 bytes
long,
- using a String instead of an io stream causes performances to degrade
to a total server freeze, as time is spent doing realloc of a buffer
always too short,
- Printing a 500 000 long string in the debug log is too verbose,
- Generating this string even when DBUG_PRINT is off is useless,
- Having code that potentially can affect the server behavior, used with
#ifdef / #endif is useful in some cases, but is also a bad practice.
After this change
=================
"Case Expressions" (both simple and searched) have been simplified to
not use LEX::when_list, which has been removed.
Considering all the issues affecting case statements, the grammar for these
has been totally re written.
The existing actions, used to generate "assembly" sp_inst* code, have been
preserved but moved in the new grammar, with the following changes:
a) Bison rules are no longer shared between "Simple" and "Searched" case
statements, because a stack instead of a flag is required to handle them.
Nested statements are handled naturally by the parser stack, which by
definition uses the correct rule in the correct context.
Nested statements of the opposite type (simple vs searched) works correctly.
The flag sp_head::IN_SIMPLE_CASE is no longer used.
This is a step towards resolution of WL#2999, which correctly identified
that temporary parsing flags do not belong to sp_head.
The code in the action is shared by mean of the case_stmt_action_xxx()
helpers.
b) The backpatch mechanism, used to resolve forward jumps in the generated
code, has been changed to:
- create a label for the instruction following 'END CASE',
- register each jump at the end of a "WHEN expr THEN stmt" in a *unique*
backpatch list associated with the 'END CASE' label
- resolve all the forward jumps for this label at once.
In addition, the code involving backpatch has been commented, so that a
reader can now understand by reading matching "Registering" and "Resolving"
comments how the forward jumps are resolved and what target they resolve to,
as this is far from evident when reading the code alone.
The implementation of sp_head::opt_mark() has been revised to avoid
recursive calls from jump instructions, and instead add the jump location
to the list of paths to explore during the flow analysis of the instruction
graph, with a call to sp_head::add_mark_lead().
In addition, the flow analysis will stop if an instruction has already
been marked as reachable, which the previous code failed to do in the
recursive case.
sp_head::opt_mark() is now private, to prevent new calls to this method from
being introduced.
The debug code present in sp_head::create() has been removed.
Considering that SHOW PROCEDURE CODE is also available in debug builds,
and can be used anytime regardless of the trace level, as opposed to
"CREATE PROCEDURE" time and only if the trace was on,
removing the code actually makes debugging easier (usable trace).
Tests have been written to cover the parser overflow (big CASE),
and to cover nested CASE statements.
2006-11-17 20:14:29 +01:00
|
|
|
sp_instr_hpush_jump::opt_mark(sp_head *sp, List<sp_instr> *leads)
|
2004-08-02 18:05:31 +02:00
|
|
|
{
|
|
|
|
sp_instr *i;
|
|
|
|
|
|
|
|
marked= 1;
|
|
|
|
if ((i= sp->get_instr(m_dest)))
|
|
|
|
{
|
2004-08-26 12:54:30 +02:00
|
|
|
m_dest= i->opt_shortcut_jump(sp, this);
|
2004-08-02 18:05:31 +02:00
|
|
|
m_optdest= sp->get_instr(m_dest);
|
|
|
|
}
|
Bug#19194 (Right recursion in parser for CASE causes excessive stack usage,
limitation)
Note to the reviewer
====================
Warning: reviewing this patch is somewhat involved.
Due to the nature of several issues all affecting the same area,
fixing separately each issue is not practical, since each fix can not be
implemented and tested independently.
In particular, the issues with
- rule recursion
- nested case statements
- forward jump resolution (backpatch list)
are tightly coupled (see below).
Definitions
===========
The expression
CASE expr
WHEN expr THEN expr
WHEN expr THEN expr
...
END
is a "Simple Case Expression".
The expression
CASE
WHEN expr THEN expr
WHEN expr THEN expr
...
END
is a "Searched Case Expression".
The statement
CASE expr
WHEN expr THEN stmts
WHEN expr THEN stmts
...
END CASE
is a "Simple Case Statement".
The statement
CASE
WHEN expr THEN stmts
WHEN expr THEN stmts
...
END CASE
is a "Searched Case Statement".
A "Left Recursive" rule is like
list:
element
| list element
;
A "Right Recursive" rule is like
list:
element
| element list
;
Left and right recursion produces the same language, the difference only
affects the *order* in which the text is parsed.
In a descendant parser (usually written manually), right recursion works
very well, and is typically implemented with a while loop.
In an ascendant parser (yacc/bison) left recursion works very well,
and is implemented naturally by the parser stack.
In both cases, using the wrong type or recursion is very bad and should be
avoided, as it causes technical issues with the parser implementation.
Before this change
==================
The "Simple Case Expression" and "Searched Case Expression" were both
implemented by the "when_list" and "when_list2" rules, which are left
recursive (ok).
These rules, however, used lex->when_list instead of using the parser stack,
which is more complex that necessary, and potentially dangerous because
of other rules using THD::reset_lex.
The "Simple Case Statement" and "Searched Case Statements" were implemented
by the "sp_case", "sp_whens" and in part by "sp_proc_stmt" rules.
Both cases were right recursive (bad).
The grammar involved was convoluted, and is assumed to be the results of
tweaks to get the code generation to work, but is not what someone would
naturally write.
In addition, using a common rule for both "Simple" and "Searched" case
statements was implemented with sp_head::m_flags |= IN_SIMPLE_CASE,
which is a flag and not a stack, and therefore does not take into account
*nested* case statements. This leads to incorrect generated code, and either
a server crash or an incorrect result.
With regards to the backpatch mechanism, a *different* backpatch list was
created for each jump from "WHEN expr THEN stmt" to "END CASE", which
relied on the grammar to be right recursive.
This is a mis-use of the backpatch list, since this list can resolve
multiple references to the same target at once.
The optimizer algorithm used to detect dead code in the "assembly" SQL
instructions, implemented by sp_head::opt_mark(uint ip), was recursive
in some cases (a conditional jump pointing forward to another conditional
jump).
In case of specially crafted code, like
- a long list of "IF expr THEN stmt END IF"
- a long CASE statement
this would actually cause a server crash with a stack overflow.
In general, having a stack that grows proportionally with user data (the
SQL code given by the client in a CREATE PROCEDURE) is to be avoided.
In debug builds only, creating a SP / SF / Trigger which had a significant
amount of code would spend --literally-- several minutes in sp_head::create,
because of the debug code involved with DBUG_PRINT("info", ("Code %s ...
There are several issues with this code:
- in a CASE with 5 000 WHEN, there are 15 000 instructions generated,
which create a sting representation of the code which is 500 000 bytes
long,
- using a String instead of an io stream causes performances to degrade
to a total server freeze, as time is spent doing realloc of a buffer
always too short,
- Printing a 500 000 long string in the debug log is too verbose,
- Generating this string even when DBUG_PRINT is off is useless,
- Having code that potentially can affect the server behavior, used with
#ifdef / #endif is useful in some cases, but is also a bad practice.
After this change
=================
"Case Expressions" (both simple and searched) have been simplified to
not use LEX::when_list, which has been removed.
Considering all the issues affecting case statements, the grammar for these
has been totally re written.
The existing actions, used to generate "assembly" sp_inst* code, have been
preserved but moved in the new grammar, with the following changes:
a) Bison rules are no longer shared between "Simple" and "Searched" case
statements, because a stack instead of a flag is required to handle them.
Nested statements are handled naturally by the parser stack, which by
definition uses the correct rule in the correct context.
Nested statements of the opposite type (simple vs searched) works correctly.
The flag sp_head::IN_SIMPLE_CASE is no longer used.
This is a step towards resolution of WL#2999, which correctly identified
that temporary parsing flags do not belong to sp_head.
The code in the action is shared by mean of the case_stmt_action_xxx()
helpers.
b) The backpatch mechanism, used to resolve forward jumps in the generated
code, has been changed to:
- create a label for the instruction following 'END CASE',
- register each jump at the end of a "WHEN expr THEN stmt" in a *unique*
backpatch list associated with the 'END CASE' label
- resolve all the forward jumps for this label at once.
In addition, the code involving backpatch has been commented, so that a
reader can now understand by reading matching "Registering" and "Resolving"
comments how the forward jumps are resolved and what target they resolve to,
as this is far from evident when reading the code alone.
The implementation of sp_head::opt_mark() has been revised to avoid
recursive calls from jump instructions, and instead add the jump location
to the list of paths to explore during the flow analysis of the instruction
graph, with a call to sp_head::add_mark_lead().
In addition, the flow analysis will stop if an instruction has already
been marked as reachable, which the previous code failed to do in the
recursive case.
sp_head::opt_mark() is now private, to prevent new calls to this method from
being introduced.
The debug code present in sp_head::create() has been removed.
Considering that SHOW PROCEDURE CODE is also available in debug builds,
and can be used anytime regardless of the trace level, as opposed to
"CREATE PROCEDURE" time and only if the trace was on,
removing the code actually makes debugging easier (usable trace).
Tests have been written to cover the parser overflow (big CASE),
and to cover nested CASE statements.
2006-11-17 20:14:29 +01:00
|
|
|
sp->add_mark_lead(m_dest, leads);
|
2004-08-02 18:05:31 +02:00
|
|
|
return m_ip+1;
|
|
|
|
}
|
|
|
|
|
2005-08-11 14:58:15 +02:00
|
|
|
|
|
|
|
/*
|
|
|
|
sp_instr_hpop class functions
|
|
|
|
*/
|
|
|
|
|
2003-09-16 14:26:08 +02:00
|
|
|
int
|
|
|
|
sp_instr_hpop::execute(THD *thd, uint *nextp)
|
|
|
|
{
|
|
|
|
DBUG_ENTER("sp_instr_hpop::execute");
|
|
|
|
thd->spcont->pop_handlers(m_count);
|
|
|
|
*nextp= m_ip+1;
|
|
|
|
DBUG_RETURN(0);
|
|
|
|
}
|
|
|
|
|
2004-03-29 11:16:45 +02:00
|
|
|
void
|
|
|
|
sp_instr_hpop::print(String *str)
|
|
|
|
{
|
2005-11-22 13:06:52 +01:00
|
|
|
/* hpop count */
|
|
|
|
if (str->reserve(SP_INSTR_UINT_MAXLEN+5))
|
2005-11-18 16:30:27 +01:00
|
|
|
return;
|
2005-11-22 13:24:53 +01:00
|
|
|
str->qs_append(STRING_WITH_LEN("hpop "));
|
2004-03-29 11:16:45 +02:00
|
|
|
str->qs_append(m_count);
|
|
|
|
}
|
|
|
|
|
2004-08-26 12:54:30 +02:00
|
|
|
|
2005-08-11 14:58:15 +02:00
|
|
|
/*
|
|
|
|
sp_instr_hreturn class functions
|
|
|
|
*/
|
|
|
|
|
2003-09-16 14:26:08 +02:00
|
|
|
int
|
|
|
|
sp_instr_hreturn::execute(THD *thd, uint *nextp)
|
|
|
|
{
|
|
|
|
DBUG_ENTER("sp_instr_hreturn::execute");
|
2004-09-10 11:11:52 +02:00
|
|
|
if (m_dest)
|
|
|
|
*nextp= m_dest;
|
|
|
|
else
|
|
|
|
{
|
|
|
|
*nextp= thd->spcont->pop_hstack();
|
|
|
|
}
|
2005-09-26 18:22:00 +02:00
|
|
|
thd->spcont->exit_handler();
|
2003-09-16 14:26:08 +02:00
|
|
|
DBUG_RETURN(0);
|
|
|
|
}
|
2003-10-10 16:57:21 +02:00
|
|
|
|
2005-08-11 14:58:15 +02:00
|
|
|
|
2004-03-29 11:16:45 +02:00
|
|
|
void
|
|
|
|
sp_instr_hreturn::print(String *str)
|
|
|
|
{
|
2005-11-22 13:06:52 +01:00
|
|
|
/* hreturn framesize dest */
|
|
|
|
if (str->reserve(SP_INSTR_UINT_MAXLEN*2 + 9))
|
2005-11-18 16:30:27 +01:00
|
|
|
return;
|
2005-11-22 13:24:53 +01:00
|
|
|
str->qs_append(STRING_WITH_LEN("hreturn "));
|
2004-03-29 11:16:45 +02:00
|
|
|
str->qs_append(m_frame);
|
2004-09-10 11:11:52 +02:00
|
|
|
if (m_dest)
|
2005-09-26 18:46:31 +02:00
|
|
|
{
|
2005-11-18 16:30:27 +01:00
|
|
|
str->qs_append(' ');
|
2004-09-10 11:11:52 +02:00
|
|
|
str->qs_append(m_dest);
|
2005-09-26 18:46:31 +02:00
|
|
|
}
|
2004-09-10 11:11:52 +02:00
|
|
|
}
|
|
|
|
|
2005-08-11 14:58:15 +02:00
|
|
|
|
2004-09-10 11:11:52 +02:00
|
|
|
uint
|
Bug#19194 (Right recursion in parser for CASE causes excessive stack usage,
limitation)
Note to the reviewer
====================
Warning: reviewing this patch is somewhat involved.
Due to the nature of several issues all affecting the same area,
fixing separately each issue is not practical, since each fix can not be
implemented and tested independently.
In particular, the issues with
- rule recursion
- nested case statements
- forward jump resolution (backpatch list)
are tightly coupled (see below).
Definitions
===========
The expression
CASE expr
WHEN expr THEN expr
WHEN expr THEN expr
...
END
is a "Simple Case Expression".
The expression
CASE
WHEN expr THEN expr
WHEN expr THEN expr
...
END
is a "Searched Case Expression".
The statement
CASE expr
WHEN expr THEN stmts
WHEN expr THEN stmts
...
END CASE
is a "Simple Case Statement".
The statement
CASE
WHEN expr THEN stmts
WHEN expr THEN stmts
...
END CASE
is a "Searched Case Statement".
A "Left Recursive" rule is like
list:
element
| list element
;
A "Right Recursive" rule is like
list:
element
| element list
;
Left and right recursion produces the same language, the difference only
affects the *order* in which the text is parsed.
In a descendant parser (usually written manually), right recursion works
very well, and is typically implemented with a while loop.
In an ascendant parser (yacc/bison) left recursion works very well,
and is implemented naturally by the parser stack.
In both cases, using the wrong type or recursion is very bad and should be
avoided, as it causes technical issues with the parser implementation.
Before this change
==================
The "Simple Case Expression" and "Searched Case Expression" were both
implemented by the "when_list" and "when_list2" rules, which are left
recursive (ok).
These rules, however, used lex->when_list instead of using the parser stack,
which is more complex that necessary, and potentially dangerous because
of other rules using THD::reset_lex.
The "Simple Case Statement" and "Searched Case Statements" were implemented
by the "sp_case", "sp_whens" and in part by "sp_proc_stmt" rules.
Both cases were right recursive (bad).
The grammar involved was convoluted, and is assumed to be the results of
tweaks to get the code generation to work, but is not what someone would
naturally write.
In addition, using a common rule for both "Simple" and "Searched" case
statements was implemented with sp_head::m_flags |= IN_SIMPLE_CASE,
which is a flag and not a stack, and therefore does not take into account
*nested* case statements. This leads to incorrect generated code, and either
a server crash or an incorrect result.
With regards to the backpatch mechanism, a *different* backpatch list was
created for each jump from "WHEN expr THEN stmt" to "END CASE", which
relied on the grammar to be right recursive.
This is a mis-use of the backpatch list, since this list can resolve
multiple references to the same target at once.
The optimizer algorithm used to detect dead code in the "assembly" SQL
instructions, implemented by sp_head::opt_mark(uint ip), was recursive
in some cases (a conditional jump pointing forward to another conditional
jump).
In case of specially crafted code, like
- a long list of "IF expr THEN stmt END IF"
- a long CASE statement
this would actually cause a server crash with a stack overflow.
In general, having a stack that grows proportionally with user data (the
SQL code given by the client in a CREATE PROCEDURE) is to be avoided.
In debug builds only, creating a SP / SF / Trigger which had a significant
amount of code would spend --literally-- several minutes in sp_head::create,
because of the debug code involved with DBUG_PRINT("info", ("Code %s ...
There are several issues with this code:
- in a CASE with 5 000 WHEN, there are 15 000 instructions generated,
which create a sting representation of the code which is 500 000 bytes
long,
- using a String instead of an io stream causes performances to degrade
to a total server freeze, as time is spent doing realloc of a buffer
always too short,
- Printing a 500 000 long string in the debug log is too verbose,
- Generating this string even when DBUG_PRINT is off is useless,
- Having code that potentially can affect the server behavior, used with
#ifdef / #endif is useful in some cases, but is also a bad practice.
After this change
=================
"Case Expressions" (both simple and searched) have been simplified to
not use LEX::when_list, which has been removed.
Considering all the issues affecting case statements, the grammar for these
has been totally re written.
The existing actions, used to generate "assembly" sp_inst* code, have been
preserved but moved in the new grammar, with the following changes:
a) Bison rules are no longer shared between "Simple" and "Searched" case
statements, because a stack instead of a flag is required to handle them.
Nested statements are handled naturally by the parser stack, which by
definition uses the correct rule in the correct context.
Nested statements of the opposite type (simple vs searched) works correctly.
The flag sp_head::IN_SIMPLE_CASE is no longer used.
This is a step towards resolution of WL#2999, which correctly identified
that temporary parsing flags do not belong to sp_head.
The code in the action is shared by mean of the case_stmt_action_xxx()
helpers.
b) The backpatch mechanism, used to resolve forward jumps in the generated
code, has been changed to:
- create a label for the instruction following 'END CASE',
- register each jump at the end of a "WHEN expr THEN stmt" in a *unique*
backpatch list associated with the 'END CASE' label
- resolve all the forward jumps for this label at once.
In addition, the code involving backpatch has been commented, so that a
reader can now understand by reading matching "Registering" and "Resolving"
comments how the forward jumps are resolved and what target they resolve to,
as this is far from evident when reading the code alone.
The implementation of sp_head::opt_mark() has been revised to avoid
recursive calls from jump instructions, and instead add the jump location
to the list of paths to explore during the flow analysis of the instruction
graph, with a call to sp_head::add_mark_lead().
In addition, the flow analysis will stop if an instruction has already
been marked as reachable, which the previous code failed to do in the
recursive case.
sp_head::opt_mark() is now private, to prevent new calls to this method from
being introduced.
The debug code present in sp_head::create() has been removed.
Considering that SHOW PROCEDURE CODE is also available in debug builds,
and can be used anytime regardless of the trace level, as opposed to
"CREATE PROCEDURE" time and only if the trace was on,
removing the code actually makes debugging easier (usable trace).
Tests have been written to cover the parser overflow (big CASE),
and to cover nested CASE statements.
2006-11-17 20:14:29 +01:00
|
|
|
sp_instr_hreturn::opt_mark(sp_head *sp, List<sp_instr> *leads)
|
2004-09-10 11:11:52 +02:00
|
|
|
{
|
Bug#19194 (Right recursion in parser for CASE causes excessive stack usage,
limitation)
Note to the reviewer
====================
Warning: reviewing this patch is somewhat involved.
Due to the nature of several issues all affecting the same area,
fixing separately each issue is not practical, since each fix can not be
implemented and tested independently.
In particular, the issues with
- rule recursion
- nested case statements
- forward jump resolution (backpatch list)
are tightly coupled (see below).
Definitions
===========
The expression
CASE expr
WHEN expr THEN expr
WHEN expr THEN expr
...
END
is a "Simple Case Expression".
The expression
CASE
WHEN expr THEN expr
WHEN expr THEN expr
...
END
is a "Searched Case Expression".
The statement
CASE expr
WHEN expr THEN stmts
WHEN expr THEN stmts
...
END CASE
is a "Simple Case Statement".
The statement
CASE
WHEN expr THEN stmts
WHEN expr THEN stmts
...
END CASE
is a "Searched Case Statement".
A "Left Recursive" rule is like
list:
element
| list element
;
A "Right Recursive" rule is like
list:
element
| element list
;
Left and right recursion produces the same language, the difference only
affects the *order* in which the text is parsed.
In a descendant parser (usually written manually), right recursion works
very well, and is typically implemented with a while loop.
In an ascendant parser (yacc/bison) left recursion works very well,
and is implemented naturally by the parser stack.
In both cases, using the wrong type or recursion is very bad and should be
avoided, as it causes technical issues with the parser implementation.
Before this change
==================
The "Simple Case Expression" and "Searched Case Expression" were both
implemented by the "when_list" and "when_list2" rules, which are left
recursive (ok).
These rules, however, used lex->when_list instead of using the parser stack,
which is more complex that necessary, and potentially dangerous because
of other rules using THD::reset_lex.
The "Simple Case Statement" and "Searched Case Statements" were implemented
by the "sp_case", "sp_whens" and in part by "sp_proc_stmt" rules.
Both cases were right recursive (bad).
The grammar involved was convoluted, and is assumed to be the results of
tweaks to get the code generation to work, but is not what someone would
naturally write.
In addition, using a common rule for both "Simple" and "Searched" case
statements was implemented with sp_head::m_flags |= IN_SIMPLE_CASE,
which is a flag and not a stack, and therefore does not take into account
*nested* case statements. This leads to incorrect generated code, and either
a server crash or an incorrect result.
With regards to the backpatch mechanism, a *different* backpatch list was
created for each jump from "WHEN expr THEN stmt" to "END CASE", which
relied on the grammar to be right recursive.
This is a mis-use of the backpatch list, since this list can resolve
multiple references to the same target at once.
The optimizer algorithm used to detect dead code in the "assembly" SQL
instructions, implemented by sp_head::opt_mark(uint ip), was recursive
in some cases (a conditional jump pointing forward to another conditional
jump).
In case of specially crafted code, like
- a long list of "IF expr THEN stmt END IF"
- a long CASE statement
this would actually cause a server crash with a stack overflow.
In general, having a stack that grows proportionally with user data (the
SQL code given by the client in a CREATE PROCEDURE) is to be avoided.
In debug builds only, creating a SP / SF / Trigger which had a significant
amount of code would spend --literally-- several minutes in sp_head::create,
because of the debug code involved with DBUG_PRINT("info", ("Code %s ...
There are several issues with this code:
- in a CASE with 5 000 WHEN, there are 15 000 instructions generated,
which create a sting representation of the code which is 500 000 bytes
long,
- using a String instead of an io stream causes performances to degrade
to a total server freeze, as time is spent doing realloc of a buffer
always too short,
- Printing a 500 000 long string in the debug log is too verbose,
- Generating this string even when DBUG_PRINT is off is useless,
- Having code that potentially can affect the server behavior, used with
#ifdef / #endif is useful in some cases, but is also a bad practice.
After this change
=================
"Case Expressions" (both simple and searched) have been simplified to
not use LEX::when_list, which has been removed.
Considering all the issues affecting case statements, the grammar for these
has been totally re written.
The existing actions, used to generate "assembly" sp_inst* code, have been
preserved but moved in the new grammar, with the following changes:
a) Bison rules are no longer shared between "Simple" and "Searched" case
statements, because a stack instead of a flag is required to handle them.
Nested statements are handled naturally by the parser stack, which by
definition uses the correct rule in the correct context.
Nested statements of the opposite type (simple vs searched) works correctly.
The flag sp_head::IN_SIMPLE_CASE is no longer used.
This is a step towards resolution of WL#2999, which correctly identified
that temporary parsing flags do not belong to sp_head.
The code in the action is shared by mean of the case_stmt_action_xxx()
helpers.
b) The backpatch mechanism, used to resolve forward jumps in the generated
code, has been changed to:
- create a label for the instruction following 'END CASE',
- register each jump at the end of a "WHEN expr THEN stmt" in a *unique*
backpatch list associated with the 'END CASE' label
- resolve all the forward jumps for this label at once.
In addition, the code involving backpatch has been commented, so that a
reader can now understand by reading matching "Registering" and "Resolving"
comments how the forward jumps are resolved and what target they resolve to,
as this is far from evident when reading the code alone.
The implementation of sp_head::opt_mark() has been revised to avoid
recursive calls from jump instructions, and instead add the jump location
to the list of paths to explore during the flow analysis of the instruction
graph, with a call to sp_head::add_mark_lead().
In addition, the flow analysis will stop if an instruction has already
been marked as reachable, which the previous code failed to do in the
recursive case.
sp_head::opt_mark() is now private, to prevent new calls to this method from
being introduced.
The debug code present in sp_head::create() has been removed.
Considering that SHOW PROCEDURE CODE is also available in debug builds,
and can be used anytime regardless of the trace level, as opposed to
"CREATE PROCEDURE" time and only if the trace was on,
removing the code actually makes debugging easier (usable trace).
Tests have been written to cover the parser overflow (big CASE),
and to cover nested CASE statements.
2006-11-17 20:14:29 +01:00
|
|
|
marked= 1;
|
2007-05-07 10:23:10 +02:00
|
|
|
|
|
|
|
if (m_dest)
|
|
|
|
{
|
|
|
|
/*
|
|
|
|
This is an EXIT handler; next instruction step is in m_dest.
|
|
|
|
*/
|
|
|
|
return m_dest;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
This is a CONTINUE handler; next instruction step will come from
|
|
|
|
the handler stack and not from opt_mark.
|
|
|
|
*/
|
Bug#19194 (Right recursion in parser for CASE causes excessive stack usage,
limitation)
Note to the reviewer
====================
Warning: reviewing this patch is somewhat involved.
Due to the nature of several issues all affecting the same area,
fixing separately each issue is not practical, since each fix can not be
implemented and tested independently.
In particular, the issues with
- rule recursion
- nested case statements
- forward jump resolution (backpatch list)
are tightly coupled (see below).
Definitions
===========
The expression
CASE expr
WHEN expr THEN expr
WHEN expr THEN expr
...
END
is a "Simple Case Expression".
The expression
CASE
WHEN expr THEN expr
WHEN expr THEN expr
...
END
is a "Searched Case Expression".
The statement
CASE expr
WHEN expr THEN stmts
WHEN expr THEN stmts
...
END CASE
is a "Simple Case Statement".
The statement
CASE
WHEN expr THEN stmts
WHEN expr THEN stmts
...
END CASE
is a "Searched Case Statement".
A "Left Recursive" rule is like
list:
element
| list element
;
A "Right Recursive" rule is like
list:
element
| element list
;
Left and right recursion produces the same language, the difference only
affects the *order* in which the text is parsed.
In a descendant parser (usually written manually), right recursion works
very well, and is typically implemented with a while loop.
In an ascendant parser (yacc/bison) left recursion works very well,
and is implemented naturally by the parser stack.
In both cases, using the wrong type or recursion is very bad and should be
avoided, as it causes technical issues with the parser implementation.
Before this change
==================
The "Simple Case Expression" and "Searched Case Expression" were both
implemented by the "when_list" and "when_list2" rules, which are left
recursive (ok).
These rules, however, used lex->when_list instead of using the parser stack,
which is more complex that necessary, and potentially dangerous because
of other rules using THD::reset_lex.
The "Simple Case Statement" and "Searched Case Statements" were implemented
by the "sp_case", "sp_whens" and in part by "sp_proc_stmt" rules.
Both cases were right recursive (bad).
The grammar involved was convoluted, and is assumed to be the results of
tweaks to get the code generation to work, but is not what someone would
naturally write.
In addition, using a common rule for both "Simple" and "Searched" case
statements was implemented with sp_head::m_flags |= IN_SIMPLE_CASE,
which is a flag and not a stack, and therefore does not take into account
*nested* case statements. This leads to incorrect generated code, and either
a server crash or an incorrect result.
With regards to the backpatch mechanism, a *different* backpatch list was
created for each jump from "WHEN expr THEN stmt" to "END CASE", which
relied on the grammar to be right recursive.
This is a mis-use of the backpatch list, since this list can resolve
multiple references to the same target at once.
The optimizer algorithm used to detect dead code in the "assembly" SQL
instructions, implemented by sp_head::opt_mark(uint ip), was recursive
in some cases (a conditional jump pointing forward to another conditional
jump).
In case of specially crafted code, like
- a long list of "IF expr THEN stmt END IF"
- a long CASE statement
this would actually cause a server crash with a stack overflow.
In general, having a stack that grows proportionally with user data (the
SQL code given by the client in a CREATE PROCEDURE) is to be avoided.
In debug builds only, creating a SP / SF / Trigger which had a significant
amount of code would spend --literally-- several minutes in sp_head::create,
because of the debug code involved with DBUG_PRINT("info", ("Code %s ...
There are several issues with this code:
- in a CASE with 5 000 WHEN, there are 15 000 instructions generated,
which create a sting representation of the code which is 500 000 bytes
long,
- using a String instead of an io stream causes performances to degrade
to a total server freeze, as time is spent doing realloc of a buffer
always too short,
- Printing a 500 000 long string in the debug log is too verbose,
- Generating this string even when DBUG_PRINT is off is useless,
- Having code that potentially can affect the server behavior, used with
#ifdef / #endif is useful in some cases, but is also a bad practice.
After this change
=================
"Case Expressions" (both simple and searched) have been simplified to
not use LEX::when_list, which has been removed.
Considering all the issues affecting case statements, the grammar for these
has been totally re written.
The existing actions, used to generate "assembly" sp_inst* code, have been
preserved but moved in the new grammar, with the following changes:
a) Bison rules are no longer shared between "Simple" and "Searched" case
statements, because a stack instead of a flag is required to handle them.
Nested statements are handled naturally by the parser stack, which by
definition uses the correct rule in the correct context.
Nested statements of the opposite type (simple vs searched) works correctly.
The flag sp_head::IN_SIMPLE_CASE is no longer used.
This is a step towards resolution of WL#2999, which correctly identified
that temporary parsing flags do not belong to sp_head.
The code in the action is shared by mean of the case_stmt_action_xxx()
helpers.
b) The backpatch mechanism, used to resolve forward jumps in the generated
code, has been changed to:
- create a label for the instruction following 'END CASE',
- register each jump at the end of a "WHEN expr THEN stmt" in a *unique*
backpatch list associated with the 'END CASE' label
- resolve all the forward jumps for this label at once.
In addition, the code involving backpatch has been commented, so that a
reader can now understand by reading matching "Registering" and "Resolving"
comments how the forward jumps are resolved and what target they resolve to,
as this is far from evident when reading the code alone.
The implementation of sp_head::opt_mark() has been revised to avoid
recursive calls from jump instructions, and instead add the jump location
to the list of paths to explore during the flow analysis of the instruction
graph, with a call to sp_head::add_mark_lead().
In addition, the flow analysis will stop if an instruction has already
been marked as reachable, which the previous code failed to do in the
recursive case.
sp_head::opt_mark() is now private, to prevent new calls to this method from
being introduced.
The debug code present in sp_head::create() has been removed.
Considering that SHOW PROCEDURE CODE is also available in debug builds,
and can be used anytime regardless of the trace level, as opposed to
"CREATE PROCEDURE" time and only if the trace was on,
removing the code actually makes debugging easier (usable trace).
Tests have been written to cover the parser overflow (big CASE),
and to cover nested CASE statements.
2006-11-17 20:14:29 +01:00
|
|
|
return UINT_MAX;
|
2004-03-29 11:16:45 +02:00
|
|
|
}
|
|
|
|
|
2004-09-10 11:11:52 +02:00
|
|
|
|
2005-08-11 14:58:15 +02:00
|
|
|
/*
|
|
|
|
sp_instr_cpush class functions
|
|
|
|
*/
|
|
|
|
|
2003-10-10 16:57:21 +02:00
|
|
|
int
|
|
|
|
sp_instr_cpush::execute(THD *thd, uint *nextp)
|
|
|
|
{
|
2005-09-02 15:21:19 +02:00
|
|
|
Query_arena backup_arena;
|
2003-10-10 16:57:21 +02:00
|
|
|
DBUG_ENTER("sp_instr_cpush::execute");
|
2005-08-18 11:23:54 +02:00
|
|
|
|
|
|
|
/*
|
|
|
|
We should create cursors in the callers arena, as
|
|
|
|
it could be (and usually is) used in several instructions.
|
|
|
|
*/
|
2005-09-02 15:21:19 +02:00
|
|
|
thd->set_n_backup_active_arena(thd->spcont->callers_arena, &backup_arena);
|
2005-08-18 11:23:54 +02:00
|
|
|
|
2005-06-30 18:07:06 +02:00
|
|
|
thd->spcont->push_cursor(&m_lex_keeper, this);
|
2005-08-18 11:23:54 +02:00
|
|
|
|
2005-09-02 15:21:19 +02:00
|
|
|
thd->restore_active_arena(thd->spcont->callers_arena, &backup_arena);
|
2005-08-18 11:23:54 +02:00
|
|
|
|
2003-10-10 16:57:21 +02:00
|
|
|
*nextp= m_ip+1;
|
2005-08-18 11:23:54 +02:00
|
|
|
|
2003-10-10 16:57:21 +02:00
|
|
|
DBUG_RETURN(0);
|
|
|
|
}
|
|
|
|
|
2005-08-11 14:58:15 +02:00
|
|
|
|
2004-03-29 11:16:45 +02:00
|
|
|
void
|
|
|
|
sp_instr_cpush::print(String *str)
|
|
|
|
{
|
2005-11-17 11:11:48 +01:00
|
|
|
LEX_STRING n;
|
|
|
|
my_bool found= m_ctx->find_cursor(m_cursor, &n);
|
2005-11-22 13:06:52 +01:00
|
|
|
/* cpush name@offset */
|
|
|
|
uint rsrv= SP_INSTR_UINT_MAXLEN+7;
|
2005-11-17 11:11:48 +01:00
|
|
|
|
2005-11-18 16:30:27 +01:00
|
|
|
if (found)
|
|
|
|
rsrv+= n.length;
|
|
|
|
if (str->reserve(rsrv))
|
|
|
|
return;
|
2005-11-22 14:25:44 +01:00
|
|
|
str->qs_append(STRING_WITH_LEN("cpush "));
|
2005-11-17 11:11:48 +01:00
|
|
|
if (found)
|
|
|
|
{
|
2005-11-18 16:30:27 +01:00
|
|
|
str->qs_append(n.str, n.length);
|
|
|
|
str->qs_append('@');
|
2005-11-17 11:11:48 +01:00
|
|
|
}
|
|
|
|
str->qs_append(m_cursor);
|
2004-03-29 11:16:45 +02:00
|
|
|
}
|
|
|
|
|
2005-08-11 14:58:15 +02:00
|
|
|
|
|
|
|
/*
|
|
|
|
sp_instr_cpop class functions
|
|
|
|
*/
|
|
|
|
|
2003-10-10 16:57:21 +02:00
|
|
|
int
|
|
|
|
sp_instr_cpop::execute(THD *thd, uint *nextp)
|
|
|
|
{
|
|
|
|
DBUG_ENTER("sp_instr_cpop::execute");
|
|
|
|
thd->spcont->pop_cursors(m_count);
|
|
|
|
*nextp= m_ip+1;
|
|
|
|
DBUG_RETURN(0);
|
|
|
|
}
|
|
|
|
|
2005-08-11 14:58:15 +02:00
|
|
|
|
2004-03-29 11:16:45 +02:00
|
|
|
void
|
|
|
|
sp_instr_cpop::print(String *str)
|
|
|
|
{
|
2005-11-22 13:06:52 +01:00
|
|
|
/* cpop count */
|
|
|
|
if (str->reserve(SP_INSTR_UINT_MAXLEN+5))
|
2005-11-18 16:30:27 +01:00
|
|
|
return;
|
2005-11-22 13:24:53 +01:00
|
|
|
str->qs_append(STRING_WITH_LEN("cpop "));
|
2004-03-29 11:16:45 +02:00
|
|
|
str->qs_append(m_count);
|
|
|
|
}
|
|
|
|
|
2005-08-11 14:58:15 +02:00
|
|
|
|
|
|
|
/*
|
|
|
|
sp_instr_copen class functions
|
|
|
|
*/
|
|
|
|
|
2003-10-10 16:57:21 +02:00
|
|
|
int
|
|
|
|
sp_instr_copen::execute(THD *thd, uint *nextp)
|
|
|
|
{
|
2005-09-22 00:11:21 +02:00
|
|
|
/*
|
|
|
|
We don't store a pointer to the cursor in the instruction to be
|
|
|
|
able to reuse the same instruction among different threads in future.
|
|
|
|
*/
|
2003-10-10 16:57:21 +02:00
|
|
|
sp_cursor *c= thd->spcont->get_cursor(m_cursor);
|
|
|
|
int res;
|
|
|
|
DBUG_ENTER("sp_instr_copen::execute");
|
|
|
|
|
|
|
|
if (! c)
|
|
|
|
res= -1;
|
|
|
|
else
|
|
|
|
{
|
2005-09-22 00:11:21 +02:00
|
|
|
sp_lex_keeper *lex_keeper= c->get_lex_keeper();
|
|
|
|
Query_arena *old_arena= thd->stmt_arena;
|
2005-06-30 18:07:06 +02:00
|
|
|
|
2005-09-22 00:11:21 +02:00
|
|
|
/*
|
|
|
|
Get the Query_arena from the cpush instruction, which contains
|
|
|
|
the free_list of the query, so new items (if any) are stored in
|
|
|
|
the right free_list, and we can cleanup after each open.
|
|
|
|
*/
|
|
|
|
thd->stmt_arena= c->get_instr();
|
|
|
|
res= lex_keeper->reset_lex_and_exec_core(thd, nextp, FALSE, this);
|
|
|
|
/* Cleanup the query's items */
|
|
|
|
if (thd->stmt_arena->free_list)
|
|
|
|
cleanup_items(thd->stmt_arena->free_list);
|
|
|
|
thd->stmt_arena= old_arena;
|
|
|
|
/*
|
|
|
|
Work around the fact that errors in selects are not returned properly
|
|
|
|
(but instead converted into a warning), so if a condition handler
|
|
|
|
caught, we have lost the result code.
|
|
|
|
*/
|
|
|
|
if (!res)
|
|
|
|
{
|
|
|
|
uint dummy1, dummy2;
|
2005-06-01 12:18:41 +02:00
|
|
|
|
2005-09-22 00:11:21 +02:00
|
|
|
if (thd->spcont->found_handler(&dummy1, &dummy2))
|
|
|
|
res= -1;
|
2005-06-01 12:18:41 +02:00
|
|
|
}
|
2005-09-22 00:11:21 +02:00
|
|
|
/* TODO: Assert here that we either have an error or a cursor */
|
2003-10-10 16:57:21 +02:00
|
|
|
}
|
|
|
|
DBUG_RETURN(res);
|
|
|
|
}
|
|
|
|
|
2005-06-05 16:01:20 +02:00
|
|
|
|
2005-03-04 14:35:28 +01:00
|
|
|
int
|
|
|
|
sp_instr_copen::exec_core(THD *thd, uint *nextp)
|
|
|
|
{
|
2005-09-22 00:11:21 +02:00
|
|
|
sp_cursor *c= thd->spcont->get_cursor(m_cursor);
|
|
|
|
int res= c->open(thd);
|
2005-03-04 14:35:28 +01:00
|
|
|
*nextp= m_ip+1;
|
|
|
|
return res;
|
|
|
|
}
|
|
|
|
|
2004-03-29 11:16:45 +02:00
|
|
|
void
|
|
|
|
sp_instr_copen::print(String *str)
|
|
|
|
{
|
2005-11-17 11:11:48 +01:00
|
|
|
LEX_STRING n;
|
|
|
|
my_bool found= m_ctx->find_cursor(m_cursor, &n);
|
2005-11-22 13:06:52 +01:00
|
|
|
/* copen name@offset */
|
|
|
|
uint rsrv= SP_INSTR_UINT_MAXLEN+7;
|
2005-11-17 11:11:48 +01:00
|
|
|
|
2005-11-18 16:30:27 +01:00
|
|
|
if (found)
|
|
|
|
rsrv+= n.length;
|
|
|
|
if (str->reserve(rsrv))
|
|
|
|
return;
|
2005-11-22 13:24:53 +01:00
|
|
|
str->qs_append(STRING_WITH_LEN("copen "));
|
2005-11-17 11:11:48 +01:00
|
|
|
if (found)
|
|
|
|
{
|
2005-11-18 16:30:27 +01:00
|
|
|
str->qs_append(n.str, n.length);
|
|
|
|
str->qs_append('@');
|
2005-11-17 11:11:48 +01:00
|
|
|
}
|
2004-03-29 11:16:45 +02:00
|
|
|
str->qs_append(m_cursor);
|
|
|
|
}
|
|
|
|
|
2005-08-11 14:58:15 +02:00
|
|
|
|
|
|
|
/*
|
|
|
|
sp_instr_cclose class functions
|
|
|
|
*/
|
|
|
|
|
2003-10-10 16:57:21 +02:00
|
|
|
int
|
|
|
|
sp_instr_cclose::execute(THD *thd, uint *nextp)
|
|
|
|
{
|
|
|
|
sp_cursor *c= thd->spcont->get_cursor(m_cursor);
|
|
|
|
int res;
|
|
|
|
DBUG_ENTER("sp_instr_cclose::execute");
|
|
|
|
|
|
|
|
if (! c)
|
|
|
|
res= -1;
|
|
|
|
else
|
|
|
|
res= c->close(thd);
|
|
|
|
*nextp= m_ip+1;
|
|
|
|
DBUG_RETURN(res);
|
|
|
|
}
|
|
|
|
|
2005-08-11 14:58:15 +02:00
|
|
|
|
2004-03-29 11:16:45 +02:00
|
|
|
void
|
|
|
|
sp_instr_cclose::print(String *str)
|
|
|
|
{
|
2005-11-17 11:11:48 +01:00
|
|
|
LEX_STRING n;
|
|
|
|
my_bool found= m_ctx->find_cursor(m_cursor, &n);
|
2005-11-22 13:06:52 +01:00
|
|
|
/* cclose name@offset */
|
|
|
|
uint rsrv= SP_INSTR_UINT_MAXLEN+8;
|
2005-11-17 11:11:48 +01:00
|
|
|
|
2005-11-18 16:30:27 +01:00
|
|
|
if (found)
|
|
|
|
rsrv+= n.length;
|
|
|
|
if (str->reserve(rsrv))
|
|
|
|
return;
|
2005-11-22 13:24:53 +01:00
|
|
|
str->qs_append(STRING_WITH_LEN("cclose "));
|
2005-11-17 11:11:48 +01:00
|
|
|
if (found)
|
|
|
|
{
|
2005-11-18 16:30:27 +01:00
|
|
|
str->qs_append(n.str, n.length);
|
|
|
|
str->qs_append('@');
|
2005-11-17 11:11:48 +01:00
|
|
|
}
|
2004-03-29 11:16:45 +02:00
|
|
|
str->qs_append(m_cursor);
|
|
|
|
}
|
|
|
|
|
2005-08-11 14:58:15 +02:00
|
|
|
|
|
|
|
/*
|
|
|
|
sp_instr_cfetch class functions
|
|
|
|
*/
|
|
|
|
|
2003-10-10 16:57:21 +02:00
|
|
|
int
|
|
|
|
sp_instr_cfetch::execute(THD *thd, uint *nextp)
|
|
|
|
{
|
|
|
|
sp_cursor *c= thd->spcont->get_cursor(m_cursor);
|
|
|
|
int res;
|
2005-09-02 15:21:19 +02:00
|
|
|
Query_arena backup_arena;
|
2003-10-10 16:57:21 +02:00
|
|
|
DBUG_ENTER("sp_instr_cfetch::execute");
|
|
|
|
|
2005-09-22 00:11:21 +02:00
|
|
|
res= c ? c->fetch(thd, &m_varlist) : -1;
|
2005-08-18 11:23:54 +02:00
|
|
|
|
2003-10-10 16:57:21 +02:00
|
|
|
*nextp= m_ip+1;
|
|
|
|
DBUG_RETURN(res);
|
|
|
|
}
|
2003-12-13 16:40:52 +01:00
|
|
|
|
2005-08-11 14:58:15 +02:00
|
|
|
|
2004-03-29 11:16:45 +02:00
|
|
|
void
|
|
|
|
sp_instr_cfetch::print(String *str)
|
|
|
|
{
|
2006-04-07 16:53:15 +02:00
|
|
|
List_iterator_fast<struct sp_variable> li(m_varlist);
|
|
|
|
sp_variable_t *pv;
|
2005-11-17 11:11:48 +01:00
|
|
|
LEX_STRING n;
|
|
|
|
my_bool found= m_ctx->find_cursor(m_cursor, &n);
|
2005-11-22 13:06:52 +01:00
|
|
|
/* cfetch name@offset vars... */
|
|
|
|
uint rsrv= SP_INSTR_UINT_MAXLEN+8;
|
2004-03-29 11:16:45 +02:00
|
|
|
|
2005-11-18 16:30:27 +01:00
|
|
|
if (found)
|
|
|
|
rsrv+= n.length;
|
|
|
|
if (str->reserve(rsrv))
|
|
|
|
return;
|
2005-11-22 13:24:53 +01:00
|
|
|
str->qs_append(STRING_WITH_LEN("cfetch "));
|
2005-11-17 11:11:48 +01:00
|
|
|
if (found)
|
|
|
|
{
|
2005-11-18 16:30:27 +01:00
|
|
|
str->qs_append(n.str, n.length);
|
|
|
|
str->qs_append('@');
|
2005-11-17 11:11:48 +01:00
|
|
|
}
|
2004-03-29 11:16:45 +02:00
|
|
|
str->qs_append(m_cursor);
|
|
|
|
while ((pv= li++))
|
|
|
|
{
|
2005-11-22 13:06:52 +01:00
|
|
|
if (str->reserve(pv->name.length+SP_INSTR_UINT_MAXLEN+2))
|
2005-11-18 16:30:27 +01:00
|
|
|
return;
|
|
|
|
str->qs_append(' ');
|
|
|
|
str->qs_append(pv->name.str, pv->name.length);
|
|
|
|
str->qs_append('@');
|
2004-03-29 11:16:45 +02:00
|
|
|
str->qs_append(pv->offset);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2005-08-11 14:58:15 +02:00
|
|
|
|
|
|
|
/*
|
|
|
|
sp_instr_error class functions
|
|
|
|
*/
|
|
|
|
|
2004-03-29 11:16:45 +02:00
|
|
|
int
|
|
|
|
sp_instr_error::execute(THD *thd, uint *nextp)
|
|
|
|
{
|
|
|
|
DBUG_ENTER("sp_instr_error::execute");
|
|
|
|
|
2004-11-12 13:34:00 +01:00
|
|
|
my_message(m_errcode, ER(m_errcode), MYF(0));
|
2004-03-29 11:16:45 +02:00
|
|
|
*nextp= m_ip+1;
|
|
|
|
DBUG_RETURN(-1);
|
|
|
|
}
|
|
|
|
|
2005-08-11 14:58:15 +02:00
|
|
|
|
2004-03-29 11:16:45 +02:00
|
|
|
void
|
|
|
|
sp_instr_error::print(String *str)
|
|
|
|
{
|
2005-11-22 13:06:52 +01:00
|
|
|
/* error code */
|
|
|
|
if (str->reserve(SP_INSTR_UINT_MAXLEN+6))
|
2005-11-18 16:30:27 +01:00
|
|
|
return;
|
2005-11-22 13:24:53 +01:00
|
|
|
str->qs_append(STRING_WITH_LEN("error "));
|
2004-03-29 11:16:45 +02:00
|
|
|
str->qs_append(m_errcode);
|
|
|
|
}
|
|
|
|
|
2004-02-17 17:36:53 +01:00
|
|
|
|
2005-12-07 15:01:17 +01:00
|
|
|
/**************************************************************************
|
|
|
|
sp_instr_set_case_expr class implementation
|
|
|
|
**************************************************************************/
|
|
|
|
|
|
|
|
int
|
|
|
|
sp_instr_set_case_expr::execute(THD *thd, uint *nextp)
|
|
|
|
{
|
|
|
|
DBUG_ENTER("sp_instr_set_case_expr::execute");
|
|
|
|
|
|
|
|
DBUG_RETURN(m_lex_keeper.reset_lex_and_exec_core(thd, nextp, TRUE, this));
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
int
|
|
|
|
sp_instr_set_case_expr::exec_core(THD *thd, uint *nextp)
|
|
|
|
{
|
2006-05-15 12:01:55 +02:00
|
|
|
int res= thd->spcont->set_case_expr(thd, m_case_expr_id, &m_case_expr);
|
2005-12-07 15:01:17 +01:00
|
|
|
|
|
|
|
if (res &&
|
|
|
|
!thd->spcont->get_case_expr(m_case_expr_id) &&
|
|
|
|
thd->spcont->found_handler_here())
|
|
|
|
{
|
|
|
|
/*
|
|
|
|
Failed to evaluate the value, the case expression is still not
|
|
|
|
initialized, and a handler has been found. Set to NULL so we can continue.
|
|
|
|
*/
|
|
|
|
|
|
|
|
Item *null_item= new Item_null();
|
|
|
|
|
|
|
|
if (!null_item ||
|
2006-05-15 12:01:55 +02:00
|
|
|
thd->spcont->set_case_expr(thd, m_case_expr_id, &null_item))
|
2005-12-07 15:01:17 +01:00
|
|
|
{
|
|
|
|
/* If this also failed, we have to abort. */
|
|
|
|
|
|
|
|
sp_rcontext *spcont= thd->spcont;
|
|
|
|
|
2007-06-18 23:54:35 +02:00
|
|
|
thd->spcont= NULL; /* Avoid handlers */
|
2005-12-07 15:01:17 +01:00
|
|
|
my_error(ER_OUT_OF_RESOURCES, MYF(0));
|
|
|
|
spcont->clear_handler();
|
|
|
|
thd->spcont= spcont;
|
|
|
|
}
|
|
|
|
}
|
2006-01-26 17:26:25 +01:00
|
|
|
else
|
|
|
|
*nextp= m_ip+1;
|
2005-12-07 15:01:17 +01:00
|
|
|
|
2006-01-26 17:26:25 +01:00
|
|
|
return res;
|
2005-12-07 15:01:17 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
void
|
|
|
|
sp_instr_set_case_expr::print(String *str)
|
|
|
|
{
|
2006-01-26 17:26:25 +01:00
|
|
|
/* set_case_expr (cont) id ... */
|
|
|
|
str->reserve(2*SP_INSTR_UINT_MAXLEN+18+32); // Add some extra for expr too
|
|
|
|
str->qs_append(STRING_WITH_LEN("set_case_expr ("));
|
|
|
|
str->qs_append(m_cont_dest);
|
|
|
|
str->qs_append(STRING_WITH_LEN(") "));
|
2005-12-07 15:01:17 +01:00
|
|
|
str->qs_append(m_case_expr_id);
|
2006-01-20 13:59:22 +01:00
|
|
|
str->qs_append(' ');
|
2005-12-07 15:01:17 +01:00
|
|
|
m_case_expr->print(str);
|
|
|
|
}
|
|
|
|
|
2006-01-26 17:26:25 +01:00
|
|
|
uint
|
Bug#19194 (Right recursion in parser for CASE causes excessive stack usage,
limitation)
Note to the reviewer
====================
Warning: reviewing this patch is somewhat involved.
Due to the nature of several issues all affecting the same area,
fixing separately each issue is not practical, since each fix can not be
implemented and tested independently.
In particular, the issues with
- rule recursion
- nested case statements
- forward jump resolution (backpatch list)
are tightly coupled (see below).
Definitions
===========
The expression
CASE expr
WHEN expr THEN expr
WHEN expr THEN expr
...
END
is a "Simple Case Expression".
The expression
CASE
WHEN expr THEN expr
WHEN expr THEN expr
...
END
is a "Searched Case Expression".
The statement
CASE expr
WHEN expr THEN stmts
WHEN expr THEN stmts
...
END CASE
is a "Simple Case Statement".
The statement
CASE
WHEN expr THEN stmts
WHEN expr THEN stmts
...
END CASE
is a "Searched Case Statement".
A "Left Recursive" rule is like
list:
element
| list element
;
A "Right Recursive" rule is like
list:
element
| element list
;
Left and right recursion produces the same language, the difference only
affects the *order* in which the text is parsed.
In a descendant parser (usually written manually), right recursion works
very well, and is typically implemented with a while loop.
In an ascendant parser (yacc/bison) left recursion works very well,
and is implemented naturally by the parser stack.
In both cases, using the wrong type or recursion is very bad and should be
avoided, as it causes technical issues with the parser implementation.
Before this change
==================
The "Simple Case Expression" and "Searched Case Expression" were both
implemented by the "when_list" and "when_list2" rules, which are left
recursive (ok).
These rules, however, used lex->when_list instead of using the parser stack,
which is more complex that necessary, and potentially dangerous because
of other rules using THD::reset_lex.
The "Simple Case Statement" and "Searched Case Statements" were implemented
by the "sp_case", "sp_whens" and in part by "sp_proc_stmt" rules.
Both cases were right recursive (bad).
The grammar involved was convoluted, and is assumed to be the results of
tweaks to get the code generation to work, but is not what someone would
naturally write.
In addition, using a common rule for both "Simple" and "Searched" case
statements was implemented with sp_head::m_flags |= IN_SIMPLE_CASE,
which is a flag and not a stack, and therefore does not take into account
*nested* case statements. This leads to incorrect generated code, and either
a server crash or an incorrect result.
With regards to the backpatch mechanism, a *different* backpatch list was
created for each jump from "WHEN expr THEN stmt" to "END CASE", which
relied on the grammar to be right recursive.
This is a mis-use of the backpatch list, since this list can resolve
multiple references to the same target at once.
The optimizer algorithm used to detect dead code in the "assembly" SQL
instructions, implemented by sp_head::opt_mark(uint ip), was recursive
in some cases (a conditional jump pointing forward to another conditional
jump).
In case of specially crafted code, like
- a long list of "IF expr THEN stmt END IF"
- a long CASE statement
this would actually cause a server crash with a stack overflow.
In general, having a stack that grows proportionally with user data (the
SQL code given by the client in a CREATE PROCEDURE) is to be avoided.
In debug builds only, creating a SP / SF / Trigger which had a significant
amount of code would spend --literally-- several minutes in sp_head::create,
because of the debug code involved with DBUG_PRINT("info", ("Code %s ...
There are several issues with this code:
- in a CASE with 5 000 WHEN, there are 15 000 instructions generated,
which create a sting representation of the code which is 500 000 bytes
long,
- using a String instead of an io stream causes performances to degrade
to a total server freeze, as time is spent doing realloc of a buffer
always too short,
- Printing a 500 000 long string in the debug log is too verbose,
- Generating this string even when DBUG_PRINT is off is useless,
- Having code that potentially can affect the server behavior, used with
#ifdef / #endif is useful in some cases, but is also a bad practice.
After this change
=================
"Case Expressions" (both simple and searched) have been simplified to
not use LEX::when_list, which has been removed.
Considering all the issues affecting case statements, the grammar for these
has been totally re written.
The existing actions, used to generate "assembly" sp_inst* code, have been
preserved but moved in the new grammar, with the following changes:
a) Bison rules are no longer shared between "Simple" and "Searched" case
statements, because a stack instead of a flag is required to handle them.
Nested statements are handled naturally by the parser stack, which by
definition uses the correct rule in the correct context.
Nested statements of the opposite type (simple vs searched) works correctly.
The flag sp_head::IN_SIMPLE_CASE is no longer used.
This is a step towards resolution of WL#2999, which correctly identified
that temporary parsing flags do not belong to sp_head.
The code in the action is shared by mean of the case_stmt_action_xxx()
helpers.
b) The backpatch mechanism, used to resolve forward jumps in the generated
code, has been changed to:
- create a label for the instruction following 'END CASE',
- register each jump at the end of a "WHEN expr THEN stmt" in a *unique*
backpatch list associated with the 'END CASE' label
- resolve all the forward jumps for this label at once.
In addition, the code involving backpatch has been commented, so that a
reader can now understand by reading matching "Registering" and "Resolving"
comments how the forward jumps are resolved and what target they resolve to,
as this is far from evident when reading the code alone.
The implementation of sp_head::opt_mark() has been revised to avoid
recursive calls from jump instructions, and instead add the jump location
to the list of paths to explore during the flow analysis of the instruction
graph, with a call to sp_head::add_mark_lead().
In addition, the flow analysis will stop if an instruction has already
been marked as reachable, which the previous code failed to do in the
recursive case.
sp_head::opt_mark() is now private, to prevent new calls to this method from
being introduced.
The debug code present in sp_head::create() has been removed.
Considering that SHOW PROCEDURE CODE is also available in debug builds,
and can be used anytime regardless of the trace level, as opposed to
"CREATE PROCEDURE" time and only if the trace was on,
removing the code actually makes debugging easier (usable trace).
Tests have been written to cover the parser overflow (big CASE),
and to cover nested CASE statements.
2006-11-17 20:14:29 +01:00
|
|
|
sp_instr_set_case_expr::opt_mark(sp_head *sp, List<sp_instr> *leads)
|
2006-01-26 17:26:25 +01:00
|
|
|
{
|
|
|
|
sp_instr *i;
|
|
|
|
|
|
|
|
marked= 1;
|
|
|
|
if ((i= sp->get_instr(m_cont_dest)))
|
|
|
|
{
|
|
|
|
m_cont_dest= i->opt_shortcut_jump(sp, this);
|
|
|
|
m_cont_optdest= sp->get_instr(m_cont_dest);
|
|
|
|
}
|
Bug#19194 (Right recursion in parser for CASE causes excessive stack usage,
limitation)
Note to the reviewer
====================
Warning: reviewing this patch is somewhat involved.
Due to the nature of several issues all affecting the same area,
fixing separately each issue is not practical, since each fix can not be
implemented and tested independently.
In particular, the issues with
- rule recursion
- nested case statements
- forward jump resolution (backpatch list)
are tightly coupled (see below).
Definitions
===========
The expression
CASE expr
WHEN expr THEN expr
WHEN expr THEN expr
...
END
is a "Simple Case Expression".
The expression
CASE
WHEN expr THEN expr
WHEN expr THEN expr
...
END
is a "Searched Case Expression".
The statement
CASE expr
WHEN expr THEN stmts
WHEN expr THEN stmts
...
END CASE
is a "Simple Case Statement".
The statement
CASE
WHEN expr THEN stmts
WHEN expr THEN stmts
...
END CASE
is a "Searched Case Statement".
A "Left Recursive" rule is like
list:
element
| list element
;
A "Right Recursive" rule is like
list:
element
| element list
;
Left and right recursion produces the same language, the difference only
affects the *order* in which the text is parsed.
In a descendant parser (usually written manually), right recursion works
very well, and is typically implemented with a while loop.
In an ascendant parser (yacc/bison) left recursion works very well,
and is implemented naturally by the parser stack.
In both cases, using the wrong type or recursion is very bad and should be
avoided, as it causes technical issues with the parser implementation.
Before this change
==================
The "Simple Case Expression" and "Searched Case Expression" were both
implemented by the "when_list" and "when_list2" rules, which are left
recursive (ok).
These rules, however, used lex->when_list instead of using the parser stack,
which is more complex that necessary, and potentially dangerous because
of other rules using THD::reset_lex.
The "Simple Case Statement" and "Searched Case Statements" were implemented
by the "sp_case", "sp_whens" and in part by "sp_proc_stmt" rules.
Both cases were right recursive (bad).
The grammar involved was convoluted, and is assumed to be the results of
tweaks to get the code generation to work, but is not what someone would
naturally write.
In addition, using a common rule for both "Simple" and "Searched" case
statements was implemented with sp_head::m_flags |= IN_SIMPLE_CASE,
which is a flag and not a stack, and therefore does not take into account
*nested* case statements. This leads to incorrect generated code, and either
a server crash or an incorrect result.
With regards to the backpatch mechanism, a *different* backpatch list was
created for each jump from "WHEN expr THEN stmt" to "END CASE", which
relied on the grammar to be right recursive.
This is a mis-use of the backpatch list, since this list can resolve
multiple references to the same target at once.
The optimizer algorithm used to detect dead code in the "assembly" SQL
instructions, implemented by sp_head::opt_mark(uint ip), was recursive
in some cases (a conditional jump pointing forward to another conditional
jump).
In case of specially crafted code, like
- a long list of "IF expr THEN stmt END IF"
- a long CASE statement
this would actually cause a server crash with a stack overflow.
In general, having a stack that grows proportionally with user data (the
SQL code given by the client in a CREATE PROCEDURE) is to be avoided.
In debug builds only, creating a SP / SF / Trigger which had a significant
amount of code would spend --literally-- several minutes in sp_head::create,
because of the debug code involved with DBUG_PRINT("info", ("Code %s ...
There are several issues with this code:
- in a CASE with 5 000 WHEN, there are 15 000 instructions generated,
which create a sting representation of the code which is 500 000 bytes
long,
- using a String instead of an io stream causes performances to degrade
to a total server freeze, as time is spent doing realloc of a buffer
always too short,
- Printing a 500 000 long string in the debug log is too verbose,
- Generating this string even when DBUG_PRINT is off is useless,
- Having code that potentially can affect the server behavior, used with
#ifdef / #endif is useful in some cases, but is also a bad practice.
After this change
=================
"Case Expressions" (both simple and searched) have been simplified to
not use LEX::when_list, which has been removed.
Considering all the issues affecting case statements, the grammar for these
has been totally re written.
The existing actions, used to generate "assembly" sp_inst* code, have been
preserved but moved in the new grammar, with the following changes:
a) Bison rules are no longer shared between "Simple" and "Searched" case
statements, because a stack instead of a flag is required to handle them.
Nested statements are handled naturally by the parser stack, which by
definition uses the correct rule in the correct context.
Nested statements of the opposite type (simple vs searched) works correctly.
The flag sp_head::IN_SIMPLE_CASE is no longer used.
This is a step towards resolution of WL#2999, which correctly identified
that temporary parsing flags do not belong to sp_head.
The code in the action is shared by mean of the case_stmt_action_xxx()
helpers.
b) The backpatch mechanism, used to resolve forward jumps in the generated
code, has been changed to:
- create a label for the instruction following 'END CASE',
- register each jump at the end of a "WHEN expr THEN stmt" in a *unique*
backpatch list associated with the 'END CASE' label
- resolve all the forward jumps for this label at once.
In addition, the code involving backpatch has been commented, so that a
reader can now understand by reading matching "Registering" and "Resolving"
comments how the forward jumps are resolved and what target they resolve to,
as this is far from evident when reading the code alone.
The implementation of sp_head::opt_mark() has been revised to avoid
recursive calls from jump instructions, and instead add the jump location
to the list of paths to explore during the flow analysis of the instruction
graph, with a call to sp_head::add_mark_lead().
In addition, the flow analysis will stop if an instruction has already
been marked as reachable, which the previous code failed to do in the
recursive case.
sp_head::opt_mark() is now private, to prevent new calls to this method from
being introduced.
The debug code present in sp_head::create() has been removed.
Considering that SHOW PROCEDURE CODE is also available in debug builds,
and can be used anytime regardless of the trace level, as opposed to
"CREATE PROCEDURE" time and only if the trace was on,
removing the code actually makes debugging easier (usable trace).
Tests have been written to cover the parser overflow (big CASE),
and to cover nested CASE statements.
2006-11-17 20:14:29 +01:00
|
|
|
sp->add_mark_lead(m_cont_dest, leads);
|
2006-01-26 17:26:25 +01:00
|
|
|
return m_ip+1;
|
|
|
|
}
|
|
|
|
|
|
|
|
void
|
|
|
|
sp_instr_set_case_expr::opt_move(uint dst, List<sp_instr> *bp)
|
|
|
|
{
|
|
|
|
if (m_cont_dest > m_ip)
|
|
|
|
bp->push_back(this); // Forward
|
|
|
|
else if (m_cont_optdest)
|
|
|
|
m_cont_dest= m_cont_optdest->m_ip; // Backward
|
|
|
|
m_ip= dst;
|
|
|
|
}
|
|
|
|
|
2005-12-07 15:01:17 +01:00
|
|
|
|
2005-08-11 14:58:15 +02:00
|
|
|
/* ------------------------------------------------------------------ */
|
2003-12-13 16:40:52 +01:00
|
|
|
|
2005-08-11 14:58:15 +02:00
|
|
|
/*
|
|
|
|
Security context swapping
|
|
|
|
*/
|
2004-02-17 17:36:53 +01:00
|
|
|
|
2003-12-16 14:15:27 +01:00
|
|
|
#ifndef NO_EMBEDDED_ACCESS_CHECKS
|
2005-09-15 21:29:07 +02:00
|
|
|
bool
|
2005-09-20 20:20:38 +02:00
|
|
|
sp_change_security_context(THD *thd, sp_head *sp, Security_context **backup)
|
2003-12-13 16:40:52 +01:00
|
|
|
{
|
2005-09-15 21:29:07 +02:00
|
|
|
*backup= 0;
|
2005-09-20 20:20:38 +02:00
|
|
|
if (sp->m_chistics->suid != SP_IS_NOT_SUID &&
|
|
|
|
(strcmp(sp->m_definer_user.str,
|
|
|
|
thd->security_ctx->priv_user) ||
|
|
|
|
my_strcasecmp(system_charset_info, sp->m_definer_host.str,
|
|
|
|
thd->security_ctx->priv_host)))
|
2003-12-13 16:40:52 +01:00
|
|
|
{
|
2005-09-15 21:29:07 +02:00
|
|
|
if (acl_getroot_no_password(&sp->m_security_ctx, sp->m_definer_user.str,
|
|
|
|
sp->m_definer_host.str,
|
|
|
|
sp->m_definer_host.str,
|
|
|
|
sp->m_db.str))
|
|
|
|
{
|
|
|
|
my_error(ER_NO_SUCH_USER, MYF(0), sp->m_definer_user.str,
|
|
|
|
sp->m_definer_host.str);
|
|
|
|
return TRUE;
|
2003-12-13 16:40:52 +01:00
|
|
|
}
|
2005-09-15 21:29:07 +02:00
|
|
|
*backup= thd->security_ctx;
|
|
|
|
thd->security_ctx= &sp->m_security_ctx;
|
2003-12-13 16:40:52 +01:00
|
|
|
}
|
2005-09-15 21:29:07 +02:00
|
|
|
return FALSE;
|
2003-12-13 16:40:52 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
void
|
2005-09-20 20:20:38 +02:00
|
|
|
sp_restore_security_context(THD *thd, Security_context *backup)
|
2003-12-13 16:40:52 +01:00
|
|
|
{
|
2005-09-15 21:29:07 +02:00
|
|
|
if (backup)
|
|
|
|
thd->security_ctx= backup;
|
2003-12-13 16:40:52 +01:00
|
|
|
}
|
2003-12-16 14:15:27 +01:00
|
|
|
|
|
|
|
#endif /* NO_EMBEDDED_ACCESS_CHECKS */
|
2005-02-08 20:52:50 +01:00
|
|
|
|
|
|
|
/*
|
2005-03-04 14:35:28 +01:00
|
|
|
Structure that represent all instances of one table
|
|
|
|
in optimized multi-set of tables used by routine.
|
|
|
|
*/
|
|
|
|
|
2005-02-08 20:52:50 +01:00
|
|
|
typedef struct st_sp_table
|
|
|
|
{
|
2006-01-12 23:51:56 +01:00
|
|
|
/*
|
|
|
|
Multi-set key:
|
|
|
|
db_name\0table_name\0alias\0 - for normal tables
|
|
|
|
db_name\0table_name\0 - for temporary tables
|
|
|
|
Note that in both cases we don't take last '\0' into account when
|
|
|
|
we count length of key.
|
|
|
|
*/
|
|
|
|
LEX_STRING qname;
|
2005-05-17 17:08:43 +02:00
|
|
|
uint db_length, table_name_length;
|
|
|
|
bool temp; /* true if corresponds to a temporary table */
|
|
|
|
thr_lock_type lock_type; /* lock type used for prelocking */
|
2005-03-04 14:35:28 +01:00
|
|
|
uint lock_count;
|
|
|
|
uint query_lock_count;
|
A fix and a test case for Bug#26141 mixing table types in trigger
causes full table lock on innodb table.
Also fixes Bug#28502 Triggers that update another innodb table
will block on X lock unnecessarily (duplciate).
Code review fixes.
Both bugs' synopses are misleading: InnoDB table is
not X locked. The statements, however, cannot proceed concurrently,
but this happens due to lock conflicts for tables used in triggers,
not for the InnoDB table.
If a user had an InnoDB table, and two triggers, AFTER UPDATE and
AFTER INSERT, competing for different resources (e.g. two distinct
MyISAM tables), then these two triggers would not be able to execute
concurrently. Moreover, INSERTS/UPDATES of the InnoDB table would
not be able to run concurrently.
The problem had other side-effects (see respective bug reports).
This behavior was a consequence of a shortcoming of the pre-locking
algorithm, which would not distinguish between different DML operations
(e.g. INSERT and DELETE) and pre-lock all the tables
that are used by any trigger defined on the subject table.
The idea of the fix is to extend the pre-locking algorithm to keep track,
for each table, what DML operation it is used for and not
load triggers that are known to never be fired.
2007-07-12 20:26:41 +02:00
|
|
|
uint8 trg_event_map;
|
2005-02-08 20:52:50 +01:00
|
|
|
} SP_TABLE;
|
|
|
|
|
|
|
|
byte *
|
|
|
|
sp_table_key(const byte *ptr, uint *plen, my_bool first)
|
|
|
|
{
|
|
|
|
SP_TABLE *tab= (SP_TABLE *)ptr;
|
|
|
|
*plen= tab->qname.length;
|
|
|
|
return (byte *)tab->qname.str;
|
|
|
|
}
|
|
|
|
|
2005-03-04 14:35:28 +01:00
|
|
|
|
2005-02-08 20:52:50 +01:00
|
|
|
/*
|
2005-03-04 14:35:28 +01:00
|
|
|
Merge the list of tables used by some query into the multi-set of
|
|
|
|
tables used by routine.
|
|
|
|
|
|
|
|
SYNOPSIS
|
|
|
|
merge_table_list()
|
|
|
|
thd - thread context
|
|
|
|
table - table list
|
|
|
|
lex_for_tmp_check - LEX of the query for which we are merging
|
|
|
|
table list.
|
|
|
|
|
|
|
|
NOTE
|
|
|
|
This method will use LEX provided to check whenever we are creating
|
|
|
|
temporary table and mark it as such in target multi-set.
|
|
|
|
|
|
|
|
RETURN VALUE
|
|
|
|
TRUE - Success
|
|
|
|
FALSE - Error
|
|
|
|
*/
|
|
|
|
|
2005-02-08 20:52:50 +01:00
|
|
|
bool
|
2005-03-04 14:35:28 +01:00
|
|
|
sp_head::merge_table_list(THD *thd, TABLE_LIST *table, LEX *lex_for_tmp_check)
|
2005-02-08 20:52:50 +01:00
|
|
|
{
|
2005-03-04 14:35:28 +01:00
|
|
|
SP_TABLE *tab;
|
|
|
|
|
|
|
|
if (lex_for_tmp_check->sql_command == SQLCOM_DROP_TABLE &&
|
|
|
|
lex_for_tmp_check->drop_temporary)
|
|
|
|
return TRUE;
|
|
|
|
|
|
|
|
for (uint i= 0 ; i < m_sptabs.records ; i++)
|
|
|
|
{
|
|
|
|
tab= (SP_TABLE *)hash_element(&m_sptabs, i);
|
|
|
|
tab->query_lock_count= 0;
|
|
|
|
}
|
|
|
|
|
2005-02-08 20:52:50 +01:00
|
|
|
for (; table ; table= table->next_global)
|
2005-03-04 14:35:28 +01:00
|
|
|
if (!table->derived && !table->schema_table)
|
2005-02-08 20:52:50 +01:00
|
|
|
{
|
2005-05-17 17:08:43 +02:00
|
|
|
char tname[(NAME_LEN + 1) * 3]; // db\0table\0alias\0
|
2005-02-08 20:52:50 +01:00
|
|
|
uint tlen, alen;
|
|
|
|
|
|
|
|
tlen= table->db_length;
|
|
|
|
memcpy(tname, table->db, tlen);
|
2005-05-17 17:08:43 +02:00
|
|
|
tname[tlen++]= '\0';
|
2005-02-08 20:52:50 +01:00
|
|
|
memcpy(tname+tlen, table->table_name, table->table_name_length);
|
|
|
|
tlen+= table->table_name_length;
|
2005-05-17 17:08:43 +02:00
|
|
|
tname[tlen++]= '\0';
|
2005-02-08 20:52:50 +01:00
|
|
|
alen= strlen(table->alias);
|
|
|
|
memcpy(tname+tlen, table->alias, alen);
|
|
|
|
tlen+= alen;
|
|
|
|
tname[tlen]= '\0';
|
|
|
|
|
A fix and a test case for
Bug#21483 "Server abort or deadlock on INSERT DELAYED with another
implicit insert"
Also fixes and adds test cases for bugs:
20497 "Trigger with INSERT DELAYED causes Error 1165"
21714 "Wrong NEW.value and server abort on INSERT DELAYED to a
table with a trigger".
Post-review fixes.
Problem:
In MySQL INSERT DELAYED is a way to pipe all inserts into a
given table through a dedicated thread. This is necessary for
simplistic storage engines like MyISAM, which do not have internal
concurrency control or threading and thus can not
achieve efficient INSERT throughput without support from SQL layer.
DELAYED INSERT works as follows:
For every distinct table, which can accept DELAYED inserts and has
pending data to insert, a dedicated thread is created to write data
to disk. All user connection threads that attempt to
delayed-insert into this table interact with the dedicated thread in
producer/consumer fashion: all records to-be inserted are pushed
into a queue of the dedicated thread, which fetches the records and
writes them.
In this design, client connection threads never open or lock
the delayed insert table.
This functionality was introduced in version 3.23 and does not take
into account existence of triggers, views, or pre-locking.
E.g. if INSERT DELAYED is called from a stored function, which,
in turn, is called from another stored function that uses the delayed
table, a deadlock can occur, because delayed locking by-passes
pre-locking. Besides:
* the delayed thread works directly with the subject table through
the storage engine API and does not invoke triggers
* even if it was patched to invoke triggers, if triggers,
in turn, used other tables, the delayed thread would
have to open and lock involved tables (use pre-locking).
* even if it was patched to use pre-locking, without deadlock
detection the delayed thread could easily lock out user
connection threads in case when the same table is used both
in a trigger and on the right side of the insert query:
the delayed thread would not release locks until all inserts
are complete, and user connection can not complete inserts
without having locks on the tables used on the right side of the
query.
Solution:
These considerations suggest two general alternatives for the
future of INSERT DELAYED:
* it is considered a full-fledged alternative to normal INSERT
* it is regarded as an optimisation that is only relevant
for simplistic engines.
Since we missed our chance to provide complete support of new
features when 5.0 was in development, the first alternative
currently renders infeasible.
However, even the second alternative, which is to detect
new features and convert DELAYED insert into a normal insert,
is not easy to implement.
The catch-22 is that we don't know if the subject table has triggers
or is a view before we open it, and we only open it in the
delayed thread. We don't know if the query involves pre-locking
until we have opened all tables, and we always first create
the delayed thread, and only then open the remaining tables.
This patch detects the problematic scenarios and converts
DELAYED INSERT to a normal INSERT using the following approach:
* if the statement is executed under pre-locking (e.g. from
within a stored function or trigger) or the right
side may require pre-locking, we detect the situation
before creating a delayed insert thread and convert the statement
to a conventional INSERT.
* if the subject table is a view or has triggers, we shutdown
the delayed thread and convert the statement to a conventional
INSERT.
2007-05-16 07:51:05 +02:00
|
|
|
/*
|
|
|
|
Upgrade the lock type because this table list will be used
|
|
|
|
only in pre-locked mode, in which DELAYED inserts are always
|
|
|
|
converted to normal inserts.
|
|
|
|
*/
|
|
|
|
if (table->lock_type == TL_WRITE_DELAYED)
|
|
|
|
table->lock_type= TL_WRITE;
|
|
|
|
|
2005-03-04 14:35:28 +01:00
|
|
|
/*
|
2006-01-12 23:51:56 +01:00
|
|
|
We ignore alias when we check if table was already marked as temporary
|
|
|
|
(and therefore should not be prelocked). Otherwise we will erroneously
|
|
|
|
treat table with same name but with different alias as non-temporary.
|
2005-03-04 14:35:28 +01:00
|
|
|
*/
|
2006-01-12 23:51:56 +01:00
|
|
|
if ((tab= (SP_TABLE *)hash_search(&m_sptabs, (byte *)tname, tlen)) ||
|
|
|
|
((tab= (SP_TABLE *)hash_search(&m_sptabs, (byte *)tname,
|
|
|
|
tlen - alen - 1)) &&
|
|
|
|
tab->temp))
|
2005-02-08 20:52:50 +01:00
|
|
|
{
|
2005-04-15 18:31:47 +02:00
|
|
|
if (tab->lock_type < table->lock_type)
|
|
|
|
tab->lock_type= table->lock_type; // Use the table with the highest lock type
|
2005-03-04 14:35:28 +01:00
|
|
|
tab->query_lock_count++;
|
|
|
|
if (tab->query_lock_count > tab->lock_count)
|
|
|
|
tab->lock_count++;
|
A fix and a test case for Bug#26141 mixing table types in trigger
causes full table lock on innodb table.
Also fixes Bug#28502 Triggers that update another innodb table
will block on X lock unnecessarily (duplciate).
Code review fixes.
Both bugs' synopses are misleading: InnoDB table is
not X locked. The statements, however, cannot proceed concurrently,
but this happens due to lock conflicts for tables used in triggers,
not for the InnoDB table.
If a user had an InnoDB table, and two triggers, AFTER UPDATE and
AFTER INSERT, competing for different resources (e.g. two distinct
MyISAM tables), then these two triggers would not be able to execute
concurrently. Moreover, INSERTS/UPDATES of the InnoDB table would
not be able to run concurrently.
The problem had other side-effects (see respective bug reports).
This behavior was a consequence of a shortcoming of the pre-locking
algorithm, which would not distinguish between different DML operations
(e.g. INSERT and DELETE) and pre-lock all the tables
that are used by any trigger defined on the subject table.
The idea of the fix is to extend the pre-locking algorithm to keep track,
for each table, what DML operation it is used for and not
load triggers that are known to never be fired.
2007-07-12 20:26:41 +02:00
|
|
|
tab->trg_event_map|= table->trg_event_map;
|
2005-02-08 20:52:50 +01:00
|
|
|
}
|
|
|
|
else
|
|
|
|
{
|
|
|
|
if (!(tab= (SP_TABLE *)thd->calloc(sizeof(SP_TABLE))))
|
|
|
|
return FALSE;
|
2005-03-04 14:35:28 +01:00
|
|
|
if (lex_for_tmp_check->sql_command == SQLCOM_CREATE_TABLE &&
|
2005-02-08 20:52:50 +01:00
|
|
|
lex_for_tmp_check->query_tables == table &&
|
|
|
|
lex_for_tmp_check->create_info.options & HA_LEX_CREATE_TMP_TABLE)
|
2006-01-12 23:51:56 +01:00
|
|
|
{
|
2005-02-08 20:52:50 +01:00
|
|
|
tab->temp= TRUE;
|
2006-01-12 23:51:56 +01:00
|
|
|
tab->qname.length= tlen - alen - 1;
|
|
|
|
}
|
|
|
|
else
|
|
|
|
tab->qname.length= tlen;
|
|
|
|
tab->qname.str= (char*) thd->memdup(tname, tab->qname.length + 1);
|
|
|
|
if (!tab->qname.str)
|
|
|
|
return FALSE;
|
2005-05-17 17:08:43 +02:00
|
|
|
tab->table_name_length= table->table_name_length;
|
|
|
|
tab->db_length= table->db_length;
|
2005-04-15 18:31:47 +02:00
|
|
|
tab->lock_type= table->lock_type;
|
2005-03-04 14:35:28 +01:00
|
|
|
tab->lock_count= tab->query_lock_count= 1;
|
A fix and a test case for Bug#26141 mixing table types in trigger
causes full table lock on innodb table.
Also fixes Bug#28502 Triggers that update another innodb table
will block on X lock unnecessarily (duplciate).
Code review fixes.
Both bugs' synopses are misleading: InnoDB table is
not X locked. The statements, however, cannot proceed concurrently,
but this happens due to lock conflicts for tables used in triggers,
not for the InnoDB table.
If a user had an InnoDB table, and two triggers, AFTER UPDATE and
AFTER INSERT, competing for different resources (e.g. two distinct
MyISAM tables), then these two triggers would not be able to execute
concurrently. Moreover, INSERTS/UPDATES of the InnoDB table would
not be able to run concurrently.
The problem had other side-effects (see respective bug reports).
This behavior was a consequence of a shortcoming of the pre-locking
algorithm, which would not distinguish between different DML operations
(e.g. INSERT and DELETE) and pre-lock all the tables
that are used by any trigger defined on the subject table.
The idea of the fix is to extend the pre-locking algorithm to keep track,
for each table, what DML operation it is used for and not
load triggers that are known to never be fired.
2007-07-12 20:26:41 +02:00
|
|
|
tab->trg_event_map= table->trg_event_map;
|
2005-03-04 14:35:28 +01:00
|
|
|
my_hash_insert(&m_sptabs, (byte *)tab);
|
2005-02-08 20:52:50 +01:00
|
|
|
}
|
|
|
|
}
|
|
|
|
return TRUE;
|
|
|
|
}
|
|
|
|
|
|
|
|
|
2005-03-04 14:35:28 +01:00
|
|
|
/*
|
|
|
|
Add tables used by routine to the table list.
|
2005-02-08 20:52:50 +01:00
|
|
|
|
2005-03-04 14:35:28 +01:00
|
|
|
SYNOPSIS
|
|
|
|
add_used_tables_to_table_list()
|
2005-12-07 10:27:17 +01:00
|
|
|
thd [in] Thread context
|
|
|
|
query_tables_last_ptr [in/out] Pointer to the next_global member of
|
|
|
|
last element of the list where tables
|
|
|
|
will be added (or to its root).
|
|
|
|
belong_to_view [in] Uppermost view which uses this routine,
|
|
|
|
0 if none.
|
2005-02-08 20:52:50 +01:00
|
|
|
|
2005-03-04 14:35:28 +01:00
|
|
|
DESCRIPTION
|
|
|
|
Converts multi-set of tables used by this routine to table list and adds
|
|
|
|
this list to the end of table list specified by 'query_tables_last_ptr'.
|
2005-02-08 20:52:50 +01:00
|
|
|
|
2005-03-04 14:35:28 +01:00
|
|
|
Elements of list will be allocated in PS memroot, so this list will be
|
|
|
|
persistent between PS executions.
|
2005-02-08 20:52:50 +01:00
|
|
|
|
2005-03-04 14:35:28 +01:00
|
|
|
RETURN VALUE
|
|
|
|
TRUE - if some elements were added, FALSE - otherwise.
|
|
|
|
*/
|
2005-02-08 20:52:50 +01:00
|
|
|
|
2005-03-04 14:35:28 +01:00
|
|
|
bool
|
|
|
|
sp_head::add_used_tables_to_table_list(THD *thd,
|
2005-12-07 10:27:17 +01:00
|
|
|
TABLE_LIST ***query_tables_last_ptr,
|
|
|
|
TABLE_LIST *belong_to_view)
|
2005-02-08 20:52:50 +01:00
|
|
|
{
|
|
|
|
uint i;
|
2005-06-15 19:58:35 +02:00
|
|
|
Query_arena *arena, backup;
|
2005-03-04 14:35:28 +01:00
|
|
|
bool result= FALSE;
|
|
|
|
DBUG_ENTER("sp_head::add_used_tables_to_table_list");
|
|
|
|
|
|
|
|
/*
|
2005-11-17 01:51:14 +01:00
|
|
|
Use persistent arena for table list allocation to be PS/SP friendly.
|
|
|
|
Note that we also have to copy database/table names and alias to PS/SP
|
|
|
|
memory since current instance of sp_head object can pass away before
|
|
|
|
next execution of PS/SP for which tables are added to prelocking list.
|
|
|
|
This will be fixed by introducing of proper invalidation mechanism
|
|
|
|
once new TDC is ready.
|
2005-03-04 14:35:28 +01:00
|
|
|
*/
|
2005-09-02 15:21:19 +02:00
|
|
|
arena= thd->activate_stmt_arena_if_needed(&backup);
|
2005-02-08 20:52:50 +01:00
|
|
|
|
2005-03-04 14:35:28 +01:00
|
|
|
for (i=0 ; i < m_sptabs.records ; i++)
|
2005-02-08 20:52:50 +01:00
|
|
|
{
|
2005-11-17 01:51:14 +01:00
|
|
|
char *tab_buff, *key_buff;
|
2005-05-17 17:08:43 +02:00
|
|
|
TABLE_LIST *table;
|
2005-03-04 14:35:28 +01:00
|
|
|
SP_TABLE *stab= (SP_TABLE *)hash_element(&m_sptabs, i);
|
2005-02-08 20:52:50 +01:00
|
|
|
if (stab->temp)
|
|
|
|
continue;
|
|
|
|
|
2005-03-04 14:35:28 +01:00
|
|
|
if (!(tab_buff= (char *)thd->calloc(ALIGN_SIZE(sizeof(TABLE_LIST)) *
|
2005-11-17 01:51:14 +01:00
|
|
|
stab->lock_count)) ||
|
|
|
|
!(key_buff= (char*)thd->memdup(stab->qname.str,
|
|
|
|
stab->qname.length + 1)))
|
2005-03-04 14:35:28 +01:00
|
|
|
DBUG_RETURN(FALSE);
|
2005-02-08 20:52:50 +01:00
|
|
|
|
2005-03-04 14:35:28 +01:00
|
|
|
for (uint j= 0; j < stab->lock_count; j++)
|
|
|
|
{
|
|
|
|
table= (TABLE_LIST *)tab_buff;
|
|
|
|
|
2005-11-17 01:51:14 +01:00
|
|
|
table->db= key_buff;
|
2005-05-17 17:08:43 +02:00
|
|
|
table->db_length= stab->db_length;
|
|
|
|
table->table_name= table->db + table->db_length + 1;
|
|
|
|
table->table_name_length= stab->table_name_length;
|
|
|
|
table->alias= table->table_name + table->table_name_length + 1;
|
2005-04-15 18:31:47 +02:00
|
|
|
table->lock_type= stab->lock_type;
|
2005-03-04 14:35:28 +01:00
|
|
|
table->cacheable_table= 1;
|
|
|
|
table->prelocking_placeholder= 1;
|
2005-12-07 10:27:17 +01:00
|
|
|
table->belong_to_view= belong_to_view;
|
A fix and a test case for Bug#26141 mixing table types in trigger
causes full table lock on innodb table.
Also fixes Bug#28502 Triggers that update another innodb table
will block on X lock unnecessarily (duplciate).
Code review fixes.
Both bugs' synopses are misleading: InnoDB table is
not X locked. The statements, however, cannot proceed concurrently,
but this happens due to lock conflicts for tables used in triggers,
not for the InnoDB table.
If a user had an InnoDB table, and two triggers, AFTER UPDATE and
AFTER INSERT, competing for different resources (e.g. two distinct
MyISAM tables), then these two triggers would not be able to execute
concurrently. Moreover, INSERTS/UPDATES of the InnoDB table would
not be able to run concurrently.
The problem had other side-effects (see respective bug reports).
This behavior was a consequence of a shortcoming of the pre-locking
algorithm, which would not distinguish between different DML operations
(e.g. INSERT and DELETE) and pre-lock all the tables
that are used by any trigger defined on the subject table.
The idea of the fix is to extend the pre-locking algorithm to keep track,
for each table, what DML operation it is used for and not
load triggers that are known to never be fired.
2007-07-12 20:26:41 +02:00
|
|
|
table->trg_event_map= stab->trg_event_map;
|
2005-03-04 14:35:28 +01:00
|
|
|
|
|
|
|
/* Everyting else should be zeroed */
|
|
|
|
|
|
|
|
**query_tables_last_ptr= table;
|
|
|
|
table->prev_global= *query_tables_last_ptr;
|
|
|
|
*query_tables_last_ptr= &table->next_global;
|
|
|
|
|
|
|
|
tab_buff+= ALIGN_SIZE(sizeof(TABLE_LIST));
|
|
|
|
result= TRUE;
|
|
|
|
}
|
2005-02-08 20:52:50 +01:00
|
|
|
}
|
2005-03-04 14:35:28 +01:00
|
|
|
|
|
|
|
if (arena)
|
2005-09-02 15:21:19 +02:00
|
|
|
thd->restore_active_arena(arena, &backup);
|
2005-03-04 14:35:28 +01:00
|
|
|
|
|
|
|
DBUG_RETURN(result);
|
2005-02-08 20:52:50 +01:00
|
|
|
}
|
|
|
|
|
2005-08-11 14:58:15 +02:00
|
|
|
|
2005-02-08 20:52:50 +01:00
|
|
|
/*
|
2005-08-11 14:58:15 +02:00
|
|
|
Simple function for adding an explicetly named (systems) table to
|
|
|
|
the global table list, e.g. "mysql", "proc".
|
|
|
|
*/
|
|
|
|
|
2005-02-08 20:52:50 +01:00
|
|
|
TABLE_LIST *
|
|
|
|
sp_add_to_query_tables(THD *thd, LEX *lex,
|
|
|
|
const char *db, const char *name,
|
|
|
|
thr_lock_type locktype)
|
|
|
|
{
|
|
|
|
TABLE_LIST *table;
|
|
|
|
|
|
|
|
if (!(table= (TABLE_LIST *)thd->calloc(sizeof(TABLE_LIST))))
|
|
|
|
{
|
|
|
|
my_error(ER_OUTOFMEMORY, MYF(0), sizeof(TABLE_LIST));
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
table->db_length= strlen(db);
|
|
|
|
table->db= thd->strmake(db, table->db_length);
|
|
|
|
table->table_name_length= strlen(name);
|
|
|
|
table->table_name= thd->strmake(name, table->table_name_length);
|
|
|
|
table->alias= thd->strdup(name);
|
|
|
|
table->lock_type= locktype;
|
|
|
|
table->select_lex= lex->current_select; // QQ?
|
|
|
|
table->cacheable_table= 1;
|
|
|
|
|
|
|
|
lex->add_to_query_tables(table);
|
|
|
|
return table;
|
|
|
|
}
|
2005-03-04 14:35:28 +01:00
|
|
|
|