This site is the archived OWASP Foundation Wiki and is no longer accepting Account Requests.
To view the new OWASP Foundation website, please visit https://owasp.org

Reviewing Code for OS Injection

From OWASP
Revision as of 16:13, 9 September 2010 by Thomas Herlea (talk | contribs) (Added navigation to facilitate sequential reading online)

Jump to: navigation, search
«««« Main
(Table of Contents)
»»»»

Introduction

Injection flaws allow attackers to pass malicious code through a web application to another subsystem. Depending on the subsystem, different types of injection attacks can be performed:

RDBMS: SQL Injection
WebBrowser/Appserver: SQL Injection
OS-shell: Operating system commands Calling external applications from your application.

OS Commanding is one of the attack classes that fall into Injection Flaws. In other classifications, it is placed in Input Validation and Representationcategory, OS Commanding threat class or defined as Failure to Sanitize Data into Control Plane weakness and Argument Injection attack pattern enumeration. OS Commanding happens when an application accepts untrusted/insecure input and passes it to external applications (either as the application name itself or arguments) without validation or a proper escaping.

How to Locate the Potentially Vulnerable code

Many developers believe text fields are the only areas for data validation. This is an incorrect assumption. Any external input must be data validated:

Text fields, List boxes, radio buttons, check boxes, cookies, HTTP header data, HTTP post data, hidden fields, parameter names and parameter values. … This is not an exhaustive list.

“Process to process” or “entity-to-entity” communication must be investigated also. Any code that communicates with an upstream or downstream process and accepts input from it must be reviewed.

All injection flaws are input-validation errors. The presence of an injection flaw is an indication of incorrect data validation on the input received from an external source outside the boundary of trust, which gets more blurred every year.

Basically for this type of vulnerability we need to find all input streams into the application. This can be from a user’s browser, CLI or fat client but also from upstream processes that “feed” our application.

An example would be to search the code base for the use of APIs or packages that are normally used for communication purposes.

The java.io, java.sql, java.net, java.rmi, java.xml packages are all used for application communication. Searching for methods from those packages in the code base can yield results. A less “scientific” method is to search for common keywords such as “UserID”, “LoginID” or “Password”.

Vulnerable Patterns for OS Injection

What we should be looking for are relationships between the application and the operating system; the application-utilising functions of the underlying operating system.

In Java using the Runtime object, java.lang.Runtime does this. In .NET calls such as System.Diagnostics.Process.Start are used to call underlying OS functions. In PHP we may look for calls such as exec() or passthru().

Example:

We have a class that eventually gets input from the user via a HTTP request. This class is used to execute some native exe on the application server and return a result.

public class DoStuff {
public string executeCommand(String userName)
{	try {
		String myUid = userName;
		Runtime rt = Runtime.getRuntime();
		rt.exec("cmd.exe /C doStuff.exe " +”-“ +myUid); // Call exe with userID
	}catch(Exception e)
		{
e.printStackTrace();
		}
	}
}


The method executeCommand calls doStuff.exe (utilizing cmd.exe) via the java.lang.runtime static method getRuntime(). The parameter passed is not validated in any way in this class. We are assuming that the data has not been data validated prior to calling this method. Transactional analysis should have encountered any data validation prior to this point. Inputting “Joe69” would result in the following MS DOS command:

doStuff.exe –Joe69

Lets say we input Joe69 & netstat –a we would get the following response:

The exe doStuff would execute passing in the User Id Joe69, but then the DOS command netstat would be called. How this works is the passing of the parameter “&” into the application, which in turn is used as a command appender in MS DOS and hence the command after the & character is executed.

This wouldn't be true, if the code above was written as (here we assume that doStuff.exe doesn't act as an command interpreter, such as cmd.exe or /bin/sh);

public class DoStuff {
public string executeCommand(String userName)
{	try {
		String myUid = userName;
		Runtime rt = Runtime.getRuntime();
		rt.exec("doStuff.exe " +”-“ +myUid); // Call exe with userID
	}catch(Exception e)
		{
e.printStackTrace();
		}
	}
}

Why? From Java 2 documentation;


... More precisely, the given command string is broken into tokens using a StringTokenizer created by the call new StringTokenizer(command) with no further modification of the character categories. The tokens produced by the tokenizer are then placed in the new string array cmdarray, in the same order ...

The produced array contains the executable (the first item) to call and its arguments (the rest of the arguments). So, unless the first item to be called is an application which parses the arguments and interprets them, and further calls other external applications according to them, it wouldn't be possible to execute netstat in the above code snippet. Such a first item to be called would be cmd.exe in Windows boxes or sh in Unix-like boxes.

Most of the out-of-box source code/assembly analyzers would (and some wouldn't!) flag a Command Execution issue when they encounter the dangerous APIs; System.Diagnostics.Process.Start, java.lang.Runtime.exec. However, obviously, the calculated risk should differ. In the first example, the "command injection" is there, whereas, in the second one without any validation nor escaping what can be called as "argument injection" vulnerability exists. The risk is still there, but the severity depends on the command being called. So, the issue needs analysis.

UNIX

An attacker might insert the string “; cat /etc/hosts” and the contents of the UNIX hosts file might be exposed to the attacker if the command is executed through a shell such as /bin/bash or /bin/sh.

.NET Example

namespace ExternalExecution
{
class CallExternal
{
static void Main(string[] args)
{
String arg1=args[0];
System.Diagnostics.Process.Start("doStuff.exe", arg1);
}
}
}

Yet again there is no data validation to speak of here, assuming that there is no upstream validation occurring in another class.

Classic ASP Example

 <% 
   option explicit
   dim wshell
   set wshell = CreateObject("WScript.Shell") 
   wshell.run "c:\file.bat " & Request.Form("Args")
   set wshell = nothing 
 %>
 


These attacks include calls to the operating system via system calls, the use of external programs via shell commands, as well as calls to backend databases via SQL (i.e. SQL injection). Complete scripts written in Perl, Python, shell, bat, and other languages can be injected into poorly designed web applications and executed.

Good Patterns & procedures to prevent OS injection

See the Data Validation section.

Related Articles

Command Injection

Interpreter Injection


«««« Main
(Table of Contents)
»»»»