- Dec 23, 2004
- Into the House of Logic
- Should Reverse Engineering Be Illegal?
- Reverse Engineering Tools and Concepts
- Approaches to Reverse Engineering
- Methods of the Reverser
- Writing Interactive Disassembler (IDA) Plugins
- Decompiling and Disassembling Software
- Decompilation in Practice: Reversing helpctr.exe
- Automatic, Bulk Auditing for Vulnerabilities
- Writing Your Own Cracking Tools
- Building a Basic Code Coverage Tool
Approaches to Reverse Engineering
As we said earlier, sometimes source code is available for a reverse engineer and sometimes it is not. White box and black box testing and analysis methods both attempt to understand the software, but they use different approaches depending on whether the analyst has access to source code.
Regardless of the method, there are several key areas that an attacker should examine to find vulnerabilities in software:
Functions that do improper (or no) bounds checking
Functions that pass through or consume user-supplied data in a format string
Functions meant to enforce bounds checking in a format string (such as %20s)
Routines that get user input using a loop
Low-level byte copy operations
Routines that use pointer arithmetic on user-supplied buffers
"Trusted" system calls that take dynamic input
This somewhat tactical list is useful when you are "in the weeds" with binary code.
White Box Analysis
White box analysis involves analyzing and understanding source code. Sometimes only binary code is available, but if you decompile a binary to get source code and then study the code, this can be considered a kind of white box analysis as well. White box testing is typically very effective in finding programming errors and implementation errors in software. In some cases this activity amounts to pattern matching and can even be automated with a static analyzer.  One drawback to this kind of whitebox testing is that it may report a potential vulnerability where none actually exists (called a false positive). Nevertheless, using static analysis methods on source code is a good approach to exploiting some kinds of software.
There are two types of white box analysis tools, those that require source code and those that automatically decompile the binary code and continue from there. One powerful and commercially available white box analysis platform, called IDA-Pro, does not require source code access. SourceScope, which includes an extensive database of source code-related problems and issues commonly encountered in Java, C, and C++, does require source code. The knowledge encapsulated in these tools is extremely useful in security analysis (and, of course, in exploiting software).
Black Box Analysis
Black box analysis refers to analyzing a running program by probing it with various inputs. This kind of testing requires only a running program and does not make use of source code analysis of any kind. In the security paradigm, malicious input can be supplied to the program in an effort to cause it to break. If the program does break during a particular test, then a security problem may have been discovered.
Note that black box testing is possible even without access to binary code. That is, a program can be tested remotely over a network. All that is required is a program running somewhere that is accepting input. If the tester can supply input that the program consumes (and can observe the effect of the test), then black box testing is possible. This is one reason that real attackers often resort to black box techniques.
Black box testing is not as effective as white box testing in obtaining knowledge of the code and its behavior, but black box testing is much easier to accomplish and usually requires much less expertise than white box testing. During black box testing, an analyst attempts to evaluate as many meaningful internal code paths as can be directly influenced and observed from outside the system. Black box testing cannot exhaustively search a real program's input space for problems because of theoretical constraints, but a black box test does act more like an actual attack on target software in a real operational environment than a white box test usually can.
Because black box testing happens on a live system, it is often an effective way of understanding and evaluating denial-of-service problems. And because black box testing can validate an application within its runtime environment (if possible), it can be used to determine whether a potential problem area is actually vulnerable in a real production system.  Sometimes problems that are discovered in a white box analysis may not be exploitable in a real, deployed system. A firewall may block the attack, for example. 
Cenzic's Hailstorm is a commercially available black box testing platform for networked software. It can be used to probe live systems for security problems. For testing network routers and switches, special hardware devices are available, such as SmartBits and IXIA. A freeware tool called ISICS can be used to probe TCP/IP stack integrity. Protocol attack systems that use black box techniques include PROTOS and Spike.
Gray Box Analysis
Gray box analysis combines white box techniques with black box input testing. Gray box approaches usually require using several tools together. A good example of a simple gray box analysis is running a target program within a debugger and then supplying particular sets of inputs to the program. In this way, the program is exercised while the debugger is used to detect any failures or faulty behavior. Rational's Purify is a commercial tool that can provide detailed runtime analysis focused on memory use and consumption. This is particularly important for C and C++ programs (in which memory problems are rampant). A freeware debugger that provides runtime analysis for Linux is called Valgrind.
All testing methods can reveal possible software risks and potential exploits. White box analysis directly identifies more bugs, but the actual risk of exploit is hard to measure. Black box analysis identifies real problems that are known to be exploitable. The use of gray box techniques combines both methods in a powerful way. Black box tests can scan programs across networks. White box tests require source code or binaries to analyze statically. In a typical case, white box analysis is used to find potential problem areas, and black box testing is then used to develop working attacks against these areas.
Audit software runtime environment
Audit software code
One problem with almost all kinds of security testing (regardless of whether such testing is black box or white box) is that there really isn't any. That is, most QA organizations concern themselves with functional testing and spend very little time understanding or probing for security risks. The QA process is almost always broken in most commercial software houses anyway because of time and budget constraints and the belief that QA is not an essential part of software development.
As software becomes more important, more emphasis is being placed on software quality managementa unified approach to testing and analysis that encompasses security, reliability, and performance. Software quality management uses both white box and black box techniques to identify and manage software risks as early as possible in the software development life cycle.
Using Gray Box Techniques to Find Vulnerabilities in Microsoft SQL Server 7
Gray box techniques usually leverage several tools. We provide an example using runtime debugging tools combined with a black box input generator. Using runtime error detection and debugging tools is a powerful way of finding problem software. When combined with black box injection tools, debuggers help catch software faults. In many cases, disassembly of the program can determine the exact nature of a software bug like the one we will show you.
One very powerful tool that examines software dynamically as it runs is Rational's Purify. In this example, we perform black box injection against Microsoft's SQL Server 7 using Hailstorm, while monitoring the target instrumented under Purify. By combining Purify and Hailstorm, the test is able to uncover a memory corruption problem occurring in the SQL server as a result of malformed protocol input. The corruption results in a software exception and subsequent failure.
To start, a remote input point is identified in the SQL server. The server listens for connections on TCP port 1433. The protocol used over this port is undocumented for the most part. Instead of reverse engineering the protocol, a simple test is constructed that supplies random inputs interspersed with numerical sequences. These data are played against the TCP port. The result is the generation of many possible "quasilegal" inputs to the port, which thus covers a wide range of input values. The inputs are injected for several minutes at a rate of around 20 per second.
The data injected pass through a number of different code paths inside the SQL server software. These locations, in essence, read the protocol header. After a short time, the test causes a fault, and Purify notes that memory corruption has occurred.
The screen shot in Figure 3-2 illustrates the SQL server failure, the Purify dump, and the Hailstorm testing platform all in one place. The memory corruption noted by Purify occurs before the SQL server crashes. Although the attack does result in a server crash, the point of memory corruption would be hard to determine without the use of Purify. The data supplied by Purify allow us to locate the exact code path that failed.
Figure 3-2 Screen shots of Hailstorm and Purify being used to probe the SQL server software for security problems using a black box paradigm.
The detection of this failure occurs well before an actual exploit has occurred. If we wanted to find this exploit using only black box tools, we might spend days trying input tests before this bug is exercised. The corruption that is occurring might cause a crash in an entirely different code location, making it very hard to identify which input sequence causes the error. Static analysis might have detected a memory corruption problem, but it would never be able to determine whether the bug could be exploited in practice by an attacker. By combining both technologies as we do in this example, we save time and get the best of both worlds.