Thread: Fwd: Bug#396218: bind.c: bad handling of SQL_LEN_DATA_AT_EXEC breaks passing char buffer length at SQLExec time

A Debian bug report ...

----------  Forwarded Message  ----------

Subject: Bug#396218: bind.c: bad handling of SQL_LEN_DATA_AT_EXEC breaks
passing char buffer length at SQLExec time
Date: Monday 30 October 2006 17:00
From: Enrico Zini <enrico@debian.org>
To: Debian Bug Tracking System <submit@bugs.debian.org>

Package: odbc-postgresql
Version: 1:08.01.0200-2
Severity: normal
Tags: patch

Hello,

thanks for maintaining odbc-postgresql.

In my package dballe (apt-get source dballe) I optimize some frequent
query by preparing the ODBC statement for them and binding the input
parameters, so that when I need the results I can just set some of the
bound variables and call SQLExecute.

This is an example of such variable bind code:

    SQLBindParameter(res->istm, 3, SQL_PARAM_INPUT, SQL_C_CHAR,
 SQL_CHAR, 0, 0, &(res->value), 0, &(res->value_ind));

Note that I'm passing a pointer to the variable that will hold the
string length as the last parameter, instead of using SQL_NTS.

Now, this call eventually ends up in this code, found at bind.c:118:

    /* Data at exec macro only valid for C char/binary data */
    if (pcbValue && (*pcbValue == SQL_DATA_AT_EXEC ||
                                     *pcbValue <=
 SQL_LEN_DATA_AT_EXEC_OFFSET)) apdopts->parameters[ipar].data_at_exec =
 TRUE;
    else
        apdopts->parameters[ipar].data_at_exec = FALSE;

This code tries to dereferenciate the (still uninitialised) pointer I
pass it ( that &(res->value_ind) ) and fails.  The test should be like
this instead:

    /* Data at exec macro only valid for C char/binary data */
    if (pcbValue && (pcbValue == SQL_DATA_AT_EXEC ||
                                     pcbValue <=
 SQL_LEN_DATA_AT_EXEC_OFFSET)) apdopts->parameters[ipar].data_at_exec =
 TRUE;
    else
        apdopts->parameters[ipar].data_at_exec = FALSE;

ODBC has this dirty habit of passing negative integers as special
 values for pointers, and SQL_DATA_AT_EXEC is one of those:

from /usr/include/sql.h:29:

#define SQL_NULL_DATA             (-1)
#define SQL_DATA_AT_EXEC          (-2)

same goes for the handling of SQL_LEN_DATA_AT_EXEC_OFFSET.

I regret I can't test the patch with code that uses SQL_DATA_AT_EXEC,
because that is an ODBC feature that I do not use.  However, with the
patch my SQLBindParameter above works fine.


Ciao,

Enrico


-- System Information:
Debian Release: 4.0
  APT prefers unstable
  APT policy: (500, 'unstable')
Architecture: i386 (i686)
Shell:  /bin/sh linked to /bin/bash
Kernel: Linux 2.6.18-1-686
Locale: LANG=it_IT.UTF-8, LC_CTYPE=it_IT.UTF-8 (charmap=UTF-8)

Versions of packages odbc-postgresql depends on:
ii  libc6                        2.3.6.ds1-7 GNU C Library: Shared
 libraries ii  libpq4                       8.1.5-1     PostgreSQL C
 client library ii  odbcinst1debian1             2.2.11-13   Support
 library and helper program

odbc-postgresql recommends no packages.

-- no debconf information

-------------------------------------------------------
--
Peter Eisentraut
http://developer.postgresql.org/~petere/

Attachment

Re: Fwd: Bug#396218: bind.c: bad handling of SQL_LEN_DATA_AT_EXEC

From
Hiroshi Inoue
Date:
Peter Eisentraut wrote:
> A Debian bug report ...

Enrico is right at the point that the evaluation of *pcbValue here is
dangerous and has no meaning. So in my version I already invalidated
the code(#ifdef NOT_USED). Unfortunately his patch seems to have no
meaning.

regards,
Hiroshi Inoue

> ----------  Forwarded Message  ----------
>
> Subject: Bug#396218: bind.c: bad handling of SQL_LEN_DATA_AT_EXEC breaks
> passing char buffer length at SQLExec time
> Date: Monday 30 October 2006 17:00
> From: Enrico Zini <enrico@debian.org>
> To: Debian Bug Tracking System <submit@bugs.debian.org>
>
> Package: odbc-postgresql
> Version: 1:08.01.0200-2
> Severity: normal
> Tags: patch
>
> Hello,
>
> thanks for maintaining odbc-postgresql.
>
> In my package dballe (apt-get source dballe) I optimize some frequent
> query by preparing the ODBC statement for them and binding the input
> parameters, so that when I need the results I can just set some of the
> bound variables and call SQLExecute.
>
> This is an example of such variable bind code:
>
>     SQLBindParameter(res->istm, 3, SQL_PARAM_INPUT, SQL_C_CHAR,
>  SQL_CHAR, 0, 0, &(res->value), 0, &(res->value_ind));
>
> Note that I'm passing a pointer to the variable that will hold the
> string length as the last parameter, instead of using SQL_NTS.
>
> Now, this call eventually ends up in this code, found at bind.c:118:
>
>     /* Data at exec macro only valid for C char/binary data */
>     if (pcbValue && (*pcbValue == SQL_DATA_AT_EXEC ||
>                                      *pcbValue <=
>  SQL_LEN_DATA_AT_EXEC_OFFSET)) apdopts->parameters[ipar].data_at_exec =
>  TRUE;
>     else
>         apdopts->parameters[ipar].data_at_exec = FALSE;
>
> This code tries to dereferenciate the (still uninitialised) pointer I
> pass it ( that &(res->value_ind) ) and fails.  The test should be like
> this instead:
>
>     /* Data at exec macro only valid for C char/binary data */
>     if (pcbValue && (pcbValue == SQL_DATA_AT_EXEC ||
>                                      pcbValue <=
>  SQL_LEN_DATA_AT_EXEC_OFFSET)) apdopts->parameters[ipar].data_at_exec =
>  TRUE;
>     else
>         apdopts->parameters[ipar].data_at_exec = FALSE;
>
> ODBC has this dirty habit of passing negative integers as special
>  values for pointers, and SQL_DATA_AT_EXEC is one of those:
>
> from /usr/include/sql.h:29:
>
> #define SQL_NULL_DATA             (-1)
> #define SQL_DATA_AT_EXEC          (-2)
>
> same goes for the handling of SQL_LEN_DATA_AT_EXEC_OFFSET.
>
> I regret I can't test the patch with code that uses SQL_DATA_AT_EXEC,
> because that is an ODBC feature that I do not use.  However, with the
> patch my SQLBindParameter above works fine.
>
>
> Ciao,
>
> Enrico

Am Dienstag, 7. November 2006 02:10 schrieb Hiroshi Inoue:
> Enrico is right at the point that the evaluation of *pcbValue here is
> dangerous and has no meaning. So in my version I already invalidated
> the code(#ifdef NOT_USED). Unfortunately his patch seems to have no
> meaning.

Which version is your version?

--
Peter Eisentraut
http://developer.postgresql.org/~petere/

Re: Fwd: Bug#396218: bind.c: bad handling of SQL_LEN_DATA_AT_EXEC

From
Hiroshi Inoue
Date:
Peter Eisentraut wrote:
> Am Dienstag, 7. November 2006 02:10 schrieb Hiroshi Inoue:
>
>> Enrico is right at the point that the evaluation of *pcbValue here is
>> dangerous and has no meaning. So in my version I already invalidated
>> the code(#ifdef NOT_USED). Unfortunately his patch seems to have no
>> meaning.
>>
>
> Which version is your version?
>

The current 8.2.xxxx version which is basically different from the
8.1.xxxx version.

regards,
Hiroshi Inoue