Buffer Overflow Defenses

	Author: Jedidiah R. Crandall, crandaj@erau.edu
	This Document was Funded by the National Science Foundation Federal Cyber Service Scholarship For Service Program: Grant No. 0113627
		Distributed July 2002
		Embry-Riddle Aeronautical University • Prescott, Arizona • USA

Buffer Overflow Defenses

	This presents some examples, pros, and cons of various defenses against buffer overflows.
	Caveats:
		1. This is not intended to be a complete list of products that defend against buffer overflows.
		2. There is no silver bullet that will stamp out buffer overflows, but some of these tools may help.

Kinds of Defenses

	Better software engineering practices
	Find-and-patch methods
	Language tools
	Analysis tools
	Compiler tools
	Operating system tools

Better Software Engineering Practices

	Testing – Execution of the software with selected data.
	Code Inspection – Inspection of the code by humans with a checklist to make sure the code meets certain criteria.
	Documentation of vendor code – Documentation of vendor code components that others may reuse in their own projects.

Better Software Engineering Practices - Testing

	Pros:
		Good testing practices should catch most buffer overflows
	Cons:
		Time is money, sometimes it’s a more economically sound solution to allow buffer overflows than to find them
		When using vendor software, you can’t white-box test software that you don’t have the source code or the documentation for
		Data corruption is harder to detect than abnormal program behavior without dynamic analysis tools

Better Software Engineering Practices– Code Inspection

	Pros:
		Code inspection may catch many buffer overflows that testing won’t
	Cons:
		Time is money
		When using vendor software, you can’t do a code inspection if you don’t have the source code

Better Software Engineering Practices - Documentation

	Pros:
		Good documentation of reusable software components will allow people who use your code in their own projects to test and inspect it
	Cons:
		Time is money, and the cost of documenting the code gets passed on to the customers
		Often software companies don’t want to release the source code for libraries that they sell

Find-and-patch Methods

	Software patches – released by vendors a security problem in their software is found to fix the vulnerability.
	Programs that block known attacks – Programs that keep a list of known attacks and watch for those attacks on your system.

Find-and-patch Methods – Software patches

	Example: The vendor, the customer, or a group concerned about software security finds a buffer overflow and a patch is written and released
	Pros:
		Very effective at preventing known buffer overflow attacks for specific vulnerabilities
	Cons:
		No protection against unknown attacks or known attacks for which a patch has not been released
		Not all patches fix the buffer overflow, some are specific to one attack but leave the buffer overflow itself in place
		The customer must regularly check for patches for their system (at the vendor’s website or www.cert.org) and install them.

Find-and-patch Methods– Programs that block known attacks

	Example: An anti-virus program that checks files and other inputs to the system for signatures of known attacks
	Pros:
		Very effective against specific attacks that are known
	Cons:
		Not effective against unknown attacks or attacks for which the anti-virus program does not yet have the signature
		The program must keep a current list of signatures for known attacks and must be updated regularly

Language tools

	Languages less susceptible to buffer overflows – Languages other than C/C++ that are less susceptible to buffer overflows when used properly.
	Languages based on C – Languages like Cyclone that were designed with preventing buffer overflows in mind.
	“Safe” buffers – Buffers that automatically truncate inputs, generate exceptions, are grow bigger.
	Safer library functions – Library functions that are less susceptible to buffer overflows than the standard C library.

Language tools – Languages less susceptible to buffer overflows

	Examples: Ada, Java, Perl, Python, etc.
	Pros:
		Automatic bounds checking makes them less susceptible to the buffer overflow problem
		Exception handling can greatly ameliorate the problem
	Cons:
		Using different languages can increase development cost
		None of these languages give the programmer access to the machine at a low level
		None of these languages give you the performance of  C/C++, most require distributable run-time environments
		C/C++ are popular languages that many programmers are familiar with
		What happens when a string that is too long is entered or an array is referenced out of bounds, is an exception generated, does the buffer grow, does the program just halt, is the user asked to provide different input?
		Programmer still must be aware of buffer overflows to provide exception handlers to do what they want (Exception handling comes with its own set of problems)

Language tools – Languages based on C

	Example: Cyclone is a different dialect of C that handles pointers in a much safer manner
	Pros:
		The transition from C to Cyclone is an easy one because Cyclone is nearly identical to C
	Cons:
		Existing C source code must be recompiled and probably modified
		Code ported to Cyclone must be debugged, and gdb (a commonly used UNIX-based debugger) doesn’t work well with Cyclone
		Using pointers in Cyclone is considerably more complicated than using pointers in C (‘*’ is replaced with ‘*’, ‘@’, and ‘?’)
		Cyclone does not provide object-oriented features

Language tools – “Safe” buffers

	Example:  C++ class objects that do bounds checking like CString,  or “limitless” strings like libmib
	Pros:
		Much safer than standard string handling in C
		Exceptions can be handled instead of a program halt
	Cons:
		Require the use of different library functions, meaning that existing code has to be modified or interfaced with in a low-level way
		A “limitless” string has to continually be reallocated meaning a bigger heap and a performance cost
		What if you don’t want the buffer to grow and accept a bigger input?

Language tools – Safer library functions

	Example: Use of a different library than the standard C libraries
	Pros:
		Eliminates problems with unsafe library function calls in C/C++
	Cons:
		Existing code has to be modified
		Programmers have to become familiar with a different set of libraries
		Often string and memory handling libraries are replaced, but not standard library functions specific to an operating system like file handling and environment variable functions which can also lead to buffer overflows
		Not all buffer overflows are caused by library functions
		What happens when a buffer’s limit is reached?  Does the program halt?  Is the string truncated?  Is an exception generated?

Analysis tools

	Static analysis – Tools that find possible defects in the source code.
	Dynamic analysis – Tools that find possible defects by analyzing things like memory usage during execution of the program.

Analysis tools - Static

	Examples: Software that searches source code for unsafe library function calls like ITS4
	Pros:
		Can be a very effective tool during code inspection by finding unsafe library function calls and making recommendations
	Cons:
		Only effective against buffer overflows caused by unsafe standard C library function calls
		Produces many false positives, only a fraction of the library function calls that are reported are actually unsafe

Analysis tools - Dynamic

	Examples: Tools that analyze memory use of a program during testing, like Purify
	Pros:
		Can detect buffer overflows that occur during testing
		Sometimes testing won’t catch buffer overflows where data is corrupted but program behavior is not affected, dynamic analysis will
	Cons:
		Buffer overflows that lead to erratic program behavior can usually be found during testing without dynamic analysis tools

Compiler tools

	Add bounds checking to all buffers
	Protect the return pointer on the stack

Compiler tools – Bounds checking

	Example: Attempts to add bounds checking to gcc
	Pros:
		Does not require modification of the source code, although you do still have to recompile
	Cons:
		Very significant decrease in performance, code size and execution time can double
		All of the programs that a systems administrator wants to protect must be recompiled
		Cannot prevent every possible buffer overflow

Compiler tools – Protect the return pointer

	Examples: Placing a canary on the stack to detect buffer overflows such as StackGuard, or adding automatic bounds checking for all strings on the stack like libsafe
	Pros:
		Does not require that existing code be modified (although it sometimes must be recompiled)
		Will effectively prevent stack smashing attacks
	Cons:
		Not all buffer overflow attacks are stack smashing attacks, program execution can be hijacked using heap-based attacks and data can always be corrupted
		Significant performance overhead
		StackGuard causes the program to halt upon detection of a buffer overflow leaving it open to denial-of-service attacks
		StackGuard requires that the target program to be protected is recompiled, libsafe doesn’t

Operating system tools

	Disable execution of code outside the code space – It is possible on some architectures to distinguish between code and data and not allow data to be executed as code.
	Intrusion detection systems – These are programs that watch for abnormal behavior or behavior that is similar to attack behavior.
	Generation of an interrupt – With hardware support it is possible to set bounds on a buffer and generate an interrupt when an attempt is made to access or change memory outside those bounds.

Operating system tools – Disable code execution outside the code space

	Example: A patch for Linux that disables execution of code on the stack as well as maps library function calls to addresses with a zero byte in them
	Pros:
		Currently, the most common and most devastating buffer overflow exploit is stack smashing and this patch makes stack smashing much more difficult
		Does not require that existing software be modified or recompiled
		A zero byte in the address of a system call forces the attacker to have a null character in the attack string
	Cons:
		Does not prevent all stack smashing attacks, often attack code can be placed in global variables or on the heap, or library code to spin a shell already exists in the code space (i.e., system() or execv())
		Crashing still leaves programs open to denial-of-service and core dump attacks
		A null character in just the right place in an attack string is not always impossible for an attacker to accomplish, and they can always jump to a small piece of code in variable space that contains a second jump to the desired location
		Some legitimate programs execute code on the stack, but very few, and there is a work-around for this

Operating system tools – Intrusion detection

	Example: An intrusion detection system could keep track of what patterns of system calls programs usually exhibit, and then report or react to anomalies such as an “execv()” call when the next system call is usually to close a file
	Pros:
		Could be able to detect a variety of hijacking attacks, not just stack smashing
		Could be able to detect many attacks on unknown vulnerabilities
	Cons:
		Intrusion detection is a developing technology
		The offending process will probably be killed leaving it open to a denial-of-service attack

Operating system tools – Generation of an Interrupt

	Example: With hardware support the program could set the bounds of every buffer and an interrupt would be generated if an attempt was made to access or change memory outside of those bounds
	Pros:
		Would prevent many buffer overflows if done properly
	Cons:
		Pointer arithmetic would still be unbounded as a pointer might be pointing to an array of 100 bytes, and array of 50 bytes, or to the 40th byte of an array of 50 bytes
		Programmers would still have to be educated about buffer overflows because they need to write an interrupt handler to do what they want it to (halt, truncate the buffer, ask the user for different input?)

About this Project

	This presentation is part of a larger package of materials on buffer overflow vulnerabilities, defenses, and software practices. For more  information, go to: http://nsfsecurity.pr.erau.edu
	Also available are:
		Demonstrations of how buffer overflows occur (Java applets)
		PowerPoint lecture-style presentations on an introduction to buffer overflows, preventing buffer overflows (for C programmers), and a case study of Code Red
		Checklists and Points to Remember for C Programmers
		An interactive module and quiz set with alternative paths for journalists/analysts and IT managers as well as programmers and testers
		A scavenger hunt on implications of the buffer overflow vulnerability
	Please complete a feedback form at http://nsfsecurity.pr.erau.edu/feedback.html to tell us how you used this material and to offer suggestions for improvements.