Edit report at https://bugs.php.net/bug.php?id=61387&edit=1
ID: 61387
Comment by: brandonkirsch at gmail dot com
Reported by: marec at stringdata dot cz
Summary: NULL valued anonymous column causes segfault in
odbc_fetch_array
Status: Assigned
Type: Bug
Package: ODBC related
Operating System: windows server 2003 R2
PHP Version: 5.3.10
Assigned To: ssb
Block user comment: N
Private report: N
New Comment:
Here's a fix for this issue: https://github.com/php/php-src/pull/193
Please accept it! Would be nice to have a fix before the next 5.5 beta.
Previous Comments:
------------------------------------------------------------------------
[2013-04-09 02:57:22] [email protected]
ssb, are you still active for odbc bugs fix?
------------------------------------------------------------------------
[2012-09-12 23:51:50] brandonkirsch at gmail dot com
This problem is isolated to the SQL Server Native Client ODBC Driver. Most
other
ODBC drivers will automatically provide some type of aliased column name for
arbitrary values that you select.
When using a SELECT 'VALUE' statement with SQL Server ODBC, the
php_odbc_fetch_hash() function returns the column as 'VALUE' => 'VALUE' -- that
is, the value you select is also the key in the array.
The problem occurs at the end of the php_odbc_fetch_hash() function in
php_odbc.c (for me, on 5.2.17 it begins on line 1510) and it looks like this:
if (!*(result->values[i].name)) { // GDB says this evaluates to 1
zend_hash_update(Z_ARRVAL_P(return_value), Z_STRVAL_P(tmp),
Z_STRLEN_P(tmp)+1, &tmp, sizeof(zval *), NULL); // SEGFAULT
This usage of zend_hash_update() causes a segfault in PHP 5.2.17 and PHP
5.3.16.
I guess that result->values[i].name looks legit and/or usable when it is not.
I updated the if statement to read:
if (!*(result->values[i].name) && Z_TYPE_P(tmp) != IS_NULL) { // Evaluates
false
for SQL NULLs that sets ZVAL_NULL(tmp)
This skips the incorrect usage of zend_hash_update with Z_STR* functions and
proceeds to the "else" condition that handles it correctly.
For us, this solution works perfectly and the end-result is array("" => NULL)
instead of a segfault. Other unaliased NULL values from other databases
continue to work the same way as before (iSeries ODBC drivers return
array("0001" => NULL))
Here's the rub: I have no idea how to generate or submit a patch. But this
fix
works. If someone can turn this into a real fix and run with it, great. If
someone would contact me and help me turn it into a real patch, I would be
happy
to help.
------------------------------------------------------------------------
[2012-09-12 21:56:17] brandonkirsch at gmail dot com
I've been fighting with this bug all day long on PHP 5.2.17 *AND* PHP 5.3.16
for
64-bit Linux.
I have isolated the problem to odbc_fetch_array() for result sets that contain
an anonymous (unnamed) column with a NULL value. If NULL value is aliased
("SELECT NULL as SOMETHING") there is no crash. If an anonymous (unnamed)
column contains anything other than NULL, there is no crash.
Test Script:
$sql = 'SELECT NULL';
$c = odbc_connect('Driver=SQL Server Native Client
11.0;server=hpesc1;uid=xxx;pwd=xxx;Database=xxx','','');
$e = odbc_exec($c, $sql);
$row = odbc_fetch_array($e);
Here are different combinations of SQL that will or won't crash PHP:
$sql = 'SELECT NULL'; // PHP SEGFAULT
$sql = 'SELECT NULL as [one]'; // OK
$sql = 'SELECT 1'; // OK
$sql = 'SELECT 1, NULL'; // SEGFAULT
$sql = 'SELECT 1, NULL as [two]'; // OK
And a backtrace:
#0 zend_inline_hash_func (nKeyLength=<optimized out>, arKey=0x0) at
/usr/src/php-5.3.16/Zend/zend_hash.h:283
#1 _zend_hash_add_or_update (ht=0xfce8b8, arKey=0x0, nKeyLength=1,
pData=0x7fffffffab90, nDataSize=8, pDest=0x0, flag=1) at /usr/src/php-
5.3.16/Zend/zend_hash.c:218
#2 0x000000000057528f in php_odbc_fetch_hash (ht=<optimized out>,
return_value=0xfccfe8, result_type=2, return_value_ptr=<optimized out>,
this_ptr=<optimized out>,
return_value_used=<optimized out>) at /usr/src/php-
5.3.16/ext/odbc/php_odbc.c:1775
My comprehension of C & the PHP source code is very limited, but it looks like
php_odbc.c lines 1752 (handling of SQL_NULL_DATA) and 1771 through 1779 are
suspect. I think php_odbc.c may be using zend_hash_update improperly but I
can't tell for sure.
And finally, here is a workaround script that DOES NOT segfault:
<?php
$c = odbc_connect('Driver=SQL Server Native Client
11.0;server=hpesc1;uid=xxx;pwd=xxx;Database=xxx','','');
$e = odbc_exec($c, 'SELECT NULL');
$true = odbc_fetch_row($e);
$row = odbc_result($e, 1);
var_dump($row); // "NULL" -- no segfault
------------------------------------------------------------------------
[2012-03-14 15:19:06] marec at stringdata dot cz
Description:
------------
odbc_fetch_array segfaults when the result consists of one-column anonymous
NULL-valued tuple
our environment:
OS: windows server 2003 R2, apache 2.2 + mod_fcgid + PHP5.3.10 (reproducible
with 5.3.6)
DB: SQL Server 2005
odbc driver: SQL Server Native Client 10.0
tried also with postgres, it seems to me that postgres amends the result with
generated name of that column ('sum' in this case)
Test script:
---------------
$conn = odbc_connect(
"Driver={SQL Server Native Client 10.0};Server=<some microsoft sql
server>;Database=<any db name>;Trusted_Connection=yes"
, ''
, ''
);
$res = odbc_exec(
$conn
, 'SELECT SUM(1)
FROM information_schema.tables
WHERE 3 = 2'
);
$row = odbc_fetch_array($res);
odbc_free_result($res);
odbc_close($conn);
var_dump($row);
Expected result:
----------------
array(1) {[0]=> NULL}
or E_WARNING
Actual result:
--------------
segfault, apache log says:
[warn] [client x.x.x.x] (OS 109)The pipe has been ended. : mod_fcgid: get
overlap result error
[error] [client x.x.x.x] Premature end of script headers: foo.php
[info] mod_fcgid: process C:/Webserver/Apache2.2/php-5.3.10/php-cgi.exe(4420)
exit(communication error), return code -1073741819
eventlog says:
Faulting application php-cgi.exe, version 5.3.10.0, faulting module php5.dll,
version 5.3.10.0, fault address 0x000858df.
PHP itself does not produce any sort of output
------------------------------------------------------------------------
--
Edit this bug report at https://bugs.php.net/bug.php?id=61387&edit=1