Bug #18930 Memory leak in sp-vars.test
Submitted: 10 Apr 2006 5:24 Modified: 8 May 2006 18:32
Reporter: Ingo Strüwing Email Updates:
Status: Closed Impact on me:
None 
Category:MySQL Server: Stored Routines Severity:S2 (Serious)
Version:5.0.21, 5.1.10 OS:Linux (Linux)
Assigned to: Michael Widenius CPU Architecture:Any

[10 Apr 2006 5:24] Ingo Strüwing
Description:
TEST                            RESULT
-------------------------------------------------------
sp-vars                        [ pass ]   
-------------------------------------------------------

Ending Tests
Shutting-down MySQL daemon

Master shutdown finished
Slave shutdown finished
All 1 tests were successful.
WARNING: Got errors/warnings while running tests. Please examine
/home/mydev/mysql-5.0-aid/mysql-test/var/log/warnings for details.

That file contains:
Warning: Not freed memory segments: 4
Warning: Memory that was not free'ed (36000 bytes):

And the error log contains:
Warning: Not freed memory segments: 4
Warning: Memory that was not free'ed (36000 bytes):
          2000 bytes at 0x008f8b008, allocated at line 5420 in 'sql_parse.cc'
         16000 bytes at 0x008f87168, allocated at line 5416 in 'sql_parse.cc'
          2000 bytes at 0x008fa04b0, allocated at line 5420 in 'sql_parse.cc'
         16000 bytes at 0x008f9c610, allocated at line 5416 in 'sql_parse.cc'

How to repeat:
BUILD/compile-pentium-debug-max
cd mysql-test
./mysql-test-run sp-vars
[10 Apr 2006 14:34] Valeriy Kravchuk
Thank you for a problem report. Verified just as described with 5.1-debug-BK (ChangeSet@1.2303.1.1, 2006-04-09 19:43:36-07:00):

TEST                            RESULT
-------------------------------------------------------
sp-vars                        [ pass ]
-------------------------------------------------------

Ending Tests
Shutting-down MySQL daemon

Master shutdown finished
Slave shutdown finished
All 1 tests were successful.
WARNING: Got errors/warnings while running tests. Please examine
/home/openxs/dbs/5.1/mysql-test/var/log/warnings for details.

openxs@suse:~/dbs/5.1/mysql-test> tail var/log/warnings
Warning: Not freed memory segments: 56
Warning: Memory that was not free'ed (38912 bytes):
[18 Apr 2006 7:01] Kristian Nielsen
Saw this problem in Valgrind as well. This is old code, so I would assume it is in 4.1 (and 4.0) as well, though have not checked.

The problem occurs when the parser stack needs to be extended. We #define yyoverflow my_yyoverflow() for that, and use my_realloc() to extend the stack. However, when yyoverflow is defined, Bison does not itself deallocate the stack after parsing, and the MySQL code doesn't either -> 26,000 bytes leaked.

The default parser stack is sufficient for most parsing, so this is not seen too often.

The following patch fixes it and removes the need to use our own yyoverflow definition:

--- sql/sql_yacc.yy.orig        2006-04-09 11:01:05.000000000 +0200
+++ sql/sql_yacc.yy 2006-04-13 09:39:59.000000000 +0200
@@ -45,7 +45,9 @@

 const LEX_STRING null_lex_str={0,0};

-#define yyoverflow(A,B,C,D,E,F) {ulong val= *(F); if (my_yyoverflow((B), (D), &val)) { yyerror((char*) (A)); return 2; } else { *(F)= (YYSIZE_T)val; }}
+#define YYMAXDEPTH 32000
+#define YYMALLOC(bytes) my_malloc(bytes, MYF(0))
+#define YYFREE(ptr) my_free(ptr, MYF(0))

 #define WARN_DEPRECATED(A,B)                                        \
   push_warning_printf(((THD *)yythd), MYSQL_ERROR::WARN_LEVEL_WARN, \

However I would like to check a few things about this patch (required Bison version, clean up my_yyoverflow() etc, ...) to make sure it is a viable approach. Alternatively, we need to keep track ourself of the memory allocated by my_yyoverflow() and free it after parsing is done.

Changing Category, since it is not really a stored procedure problem.
[19 Apr 2006 12:09] Kristian Nielsen
This is actually a fairly serious leak; every client connection that executes this will leak 26,000 bytes:

CREATE PROCEDURE p1()
BEGIN
  DECLARE dummy TINYINT DEFAULT 0;
  CASE 0
    WHEN 0 THEN SET dummy = 0;
    WHEN 1 THEN SET dummy = 0;
    WHEN 2 THEN SET dummy = 0;
    WHEN 3 THEN SET dummy = 0;
    WHEN 4 THEN SET dummy = 0;
    WHEN 5 THEN SET dummy = 0;
    WHEN 6 THEN SET dummy = 0;
    WHEN 7 THEN SET dummy = 0;
    WHEN 8 THEN SET dummy = 0;
    WHEN 9 THEN SET dummy = 0;
    ELSE SET dummy = 0;
  END CASE;
END
[19 Apr 2006 14:20] Kristian Nielsen
Unfortunately, it seems that the simple solution of just #defining YYMALLOC will not work for us (Bison will use alloca() in some cases, which I do not think we want; I've seen segfaults when the stack is grown in this case).

So we need to find somewhere to store the memory returned by my_yyoverflow so that we can free it after parsing.
[19 Apr 2006 20:14] Kristian Nielsen
Digged some more, the problem occurs also with views:

CREATE VIEW v1 AS SELECT 1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(1+(0+0))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))));
SELECT * FROM v1;

This leaks memory in the same was as the stored procedure examples.

There is actually code to free the memory allocated by my_yyoverflow(), in lex_end() in sql_lex.cc. But for stored procedures and views, the call of this function seems to be missing:

db_load_routine(), in sp.cc, calls lex_start() but not lex_end().
mysql_make_view(), in sql_view.cc, calls lex_start() but not lex_end().
sp_head::reset_lex(), in sp_head.cc, also calls lex_start() to push a new lex object on the m_lex list, but it is not clear to me whether that is subsequently called lex_end() upon.

So the proper fix should be to make sure that lex_end() is called in the right places.
[5 May 2006 11:50] Michael Widenius
Thank you for your bug report. This issue has been committed to our
source repository of that product and will be incorporated into the
next release.

If necessary, you can access the source repository and build the latest
available version, including the bugfix, yourself. More information 
about accessing the source trees is available at
    http://www.mysql.com/doc/en/Installing_source_tree.html

Additional info:

Fix will be in 5.0.22 and 5.1.10
[8 May 2006 18:32] Paul Dubois
Noted in 5.0.22, 5.1.10 changelogs.

The parser leaked memory when its stack needed to be extended.
(Bug #18930)