Thread: Fwd: Bug#396218: bind.c: bad handling of SQL_LEN_DATA_AT_EXEC breaks passing char buffer length at SQLExec time
Fwd: Bug#396218: bind.c: bad handling of SQL_LEN_DATA_AT_EXEC breaks passing char buffer length at SQLExec time
From
Peter Eisentraut
Date:
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
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
Re: Fwd: Bug#396218: bind.c: bad handling of SQL_LEN_DATA_AT_EXEC breaks passing char buffer length at SQLExec time
From
Peter Eisentraut
Date:
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/
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