Re: EN_Destructor bug

From: "Inoue, Hiroshi" <inoue(at)tpf(dot)co(dot)jp>
To: mkocherov <mkocherov(at)parallels(dot)com>
Cc: pgsql-odbc(at)postgresql(dot)org
Subject: Re: EN_Destructor bug
Date: 2013-01-23 23:42:58
Message-ID: 51007582.3010601@tpf.co.jp
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-odbc

Thanks for the report.
I would take care fo it.

regards,
Hiroshi Inoue

(2013/01/22 22:32), mkocherov wrote:
> Hello!
>
> 1) We observe crashes in psqlodbc.so library from time to time. Crash dumps
> typically look like this one:
> #0 0x0057f420 in __kernel_vsyscall ()
> #1 0x09ba7df0 in raise () from /lib/libc.so.6
> #2 0x09ba9701 in abort () from /lib/libc.so.6
> #3 0x09bdebcb in __libc_message () from /lib/libc.so.6
> #4 0x09be6fea in _int_free () from /lib/libc.so.6
> #5 0x09be7329 in free () from /lib/libc.so.6
> #6 0x02c601d0 in EN_Destructor () from /usr/pgsql-9.0/lib/psqlodbc.so
> #7 0x02c60224 in PGAPI_FreeEnv () from /usr/pgsql-9.0/lib/psqlodbc.so
> #8 0x02c8f3fe in SQLFreeHandle () from /usr/pgsql-9.0/lib/psqlodbc.so
> #9 0x00e961e5 in ?? () from /usr/lib/libodbc.so.1
> #10 0x00e9666d in ?? () from /usr/lib/libodbc.so.1
> #11 0x00e96760 in ?? () from /usr/lib/libodbc.so.1
> #12 0x00e9d02e in SQLDisconnect () from /usr/lib/libodbc.so.1
> …
> Due to nature of our program it makes frequent connect / disconnects in
> parallel threads. ODBC threading level is set to 2, e.g.:
> [PostgreSQL]
> Description = PostgreSQL driver
> Driver = /usr/pgsql-9.0/lib/psqlodbc.so
> Setup = /usr/pgsql-9.0/lib/psqlodbc.so
> FileUsage = 1
> Threading = 2
> 2) Checking code of driver reveals that it doesn’t protect shared connection
> list when making cleanup tasks in EN_Destructor (PGAPI_FreeEnv ->
> EN_Destructor). In this function there is no ENTER_CONNS_CS /
> LEAVE_CONNS_CS;
> 3) In 2005 Tom Hughes
> (http://postgresql.1045698.n5.nabble.com/EN-Destructor-bug-td2187457.html)
> provided patch that fixed this problem but then his changes were lost
> (starting from psqlodbc-08.02.0100 version).
> 4) The problem was reproduced on latest stock drivers psqlodbc-09.00.0310
> and psqlodbc-09.01.0200.
> 5) When logging to mylog is enrolled then problem is hardly reproduced – it
> seems parallels calls are effectively serialized within mylog() function.
> 6) Problem is somewhat mitigated by the fact that default threading level in
> unixODBC is set to 3 which eventually prevents races in psqlodbc driver.
> 7) Patch:
> --- environ.c.van 2010-08-21 10:29:19.000000000 +0400
> +++ environ.c 2013-01-17 18:47:12.000000000 +0300
> @@ -583,6 +583,7 @@
> */
>
> /* Free any connections belonging to this environment */
> + ENTER_CONNS_CS;
> for (lf = 0, nullcnt = 0; lf < conns_count; lf++)
> {
> if (NULL == conns[lf])
> @@ -603,6 +604,7 @@
> conns = NULL;
> conns_count = 0;
> }
> + LEAVE_CONNS_CS;
> DELETE_ENV_CS(self);
> free(self);
>
> Thanks!
>
> Michael Kocherov
> Software Developer
> Parallels Automation Maintenance

In response to

Browse pgsql-odbc by date

  From Date Subject
Next Message Barry Bell 2013-01-25 14:06:36 ODBC blank date.time setting
Previous Message j.letzel@t-online.de 2013-01-23 15:29:29 SQL_ATTR_QUERY_TIMEOUT not supported