The Art and Science of Computer Security

• Authors: Matthew A. Bishop
• Venue: The Art and Science of Computer Security
• Year: 2002

From the Book:Hortensio: Madam, before you touch the instrument/To learn the order of my fingering, I must begin with rudiments of art/To teach you gamouth in a briefer sort, More pleasant, pithy and effectual, Than hath been taught by any of my trade; And there it is in writing, fairly drawn.— The Taming of the Shrew, III, i, 62-68On September 11, 2001, terrorists seized control of four airplanes. Three were flown into buildings, and a fourth crashed, with catastrophic loss of life. In the aftermath, the security and reliability of many aspects of society drew renewed scrutiny. One of these aspects was the widespread use of computers and their interconnecting networks.The issue is not new. In 1988, approximately 5,000 computers throughout the Internet were rendered unusable within four hours by a program called a worm 430. While the spread, and the effects, of this program alarmed computer scientists, most people were not worried because the worm did not affect their lives or their ability to do their jobs. In 1993, more users of computer systems were finally alerted to such dangers when a set of programs called sniffers were placed on many computers run by network service providers and recorded login names and passwords 373.After an attack on Tsutomu Shimomura's computer system, and the fascinating way Shimomura followed the attacker's trail, which led to his arrest 911, the public's interest and apprehension was finally aroused. Computers were now vulnerable. Their once reassuring protections were now viewed as flimsy.A number of films explored these concerns. Movies like War Games and Hackers provided images ofpeople who can, at will, wander throughout computers and networks, maliciously or frivolously corrupting or destroying information it may have taken millions of dollars to amass. (Reality intruded upon Hackers when the World Wide Web page set up by MGM/United Artists was quickly altered to present an irreverent commentary on the movie and suggest viewers see The Net instead. Paramount Pictures denied doing this 446.) Another film, Sneakers, presented a picture of those who test the security of computer (and other) systems for their owners and for the government.GoalsThis book has three goals. The first is to show the importance of theory to practice and practice to theory. All too often practitioners regard theory as irrelevant, and theoreticians think of practice as trivial. In reality, theory and practice are symbiotic. For example, the theory of covert channels, in which the goal is to limit the ability of processes to communicate through shared resources, provides a mechanism to evaluate the effectiveness of mechanisms that confine processes, such as sandboxes and firewalls. Similarly, business practices in the commercial world led to the development of several security policy models such as the Clark-Wilson model and the Chinese Wall model. These models in turn help the designers of security policies better understand and evaluate the mechanisms and procedures needed to secure their sites.The second goal is to emphasize that computer security and cryptography are different. While cryptography is an essential component of computer security, it is by no means the only component. Cryptography provides a mechanism to perform specific functions, such as preventing unauthorized people from reading and altering messages on a network. But unless developers understand the context in which they are using cryptography, and the assumptions underlying the protocol and the cryptographic mechanisms apply to the context, the cryptography may not add to the security of the system. The canonical example is using cryptography to secure communications between two low security systems. If only trusted users can access the two systems, cryptography protects messages in transit. But if untrusted users can access either system (through authorized accounts or, more likely, by breaking in), the cryptography is not sufficient to protect the messages. The attackers can read the messages at either endpoint.The third goal is to demonstrate that computer security is not just a science but also an art. It is an art because no system can be considered secure without first examining how it is to be used. The definition of a "secure computer" necessitates a statement of requirements and an expression of those requirements in the form of authorized actions and authorized users. (A computer engaged in work at a university may be considered "secure" for the purposes of the work done at the university. When moved to a military installation, that same system may not provide sufficient control to be deemed "secure" for the purposes of the work done at that installation.) How will people, as well as other computers, interact with the computer system? How clear and restrictive an interface can a designer create without rendering the system unusable while trying to prevent an unauthorized use or access to the data or resources on the system?Just as an artist paints his view of the world onto canvas, so does a designer of security features articulate his view of the world of man/machine interaction in the security policy and mechanisms of the system. Two designers may use entirely different designs to achieve the same creation, just as two artists may use different subjects to achieve the same concept.Computer security is also a science. Its theory is based on mathematical constructions, analyses and proofs. Its systems are built following the accepted practices of engineering. It uses inductive and deductive reasoning to examine the security of systems from key axioms and to discover underlying principles. These scientific principles can then be applied to untraditional situations and new theories, policies, and mechanisms.PhilosophyKey to understanding the problems that exist in computer security is a recognition that the problems are not new. They are old problems, dating from the beginning of computer security (and, in fact, arising from parallel problems in the non-computer world). But the locus has changed as the field of computing has changed. Before the mid-1980s, mainframe and mid-level computers dominated the market, and computer security problems and solutions were phrased in terms of securing files or processes on a single system. With the rise of networking and the Internet, the arena has changed. Workstations and servers, and the networking infrastructure that connects them, now dominate the market. Computer security problems and solutions now focus on a networked environment. However, if the workstations and servers, and supporting network infrastructure, are viewed as a single system, the models, theories, and problem statements developed for systems before the mid-1980s apply equally well to current systems.As an example, consider the issue of assurance. In the early period, assurance arose in several ways: formal methods and proofs of correctness, validation of policy to requirements, and acquisition of data and programs from trusted sources, to name a few. Those providing assurance analyzed a single system, the code on it, and the sources (vendors and users) from which code could be acquired to ensure that either the sources could be trusted, or the programs could be confined adequately to do minimal damage. In the later period, the same basic principles and techniques apply, except that the scope of some is greatly expanded (from a single system and small set of vendors to the world-wide Internet). The work on proof-carrying code, an exciting development in which the proof that a downloadable program module satisfies a stated policy is incorporated into the program itself, is an example of this. It extends the notion of a proof of consistency with a stated policy. It advances the technology of the earlier period into the later period. But in order to understand it properly, one must understand the ideas underlying the concept of proof-carrying code, and these ideas lie in the earlier period.As another example, consider Saltzer and Schroeder's principles of secure design. Enunciated in 1975, they promote simplicity, confinement, and understanding. When security mechanisms grow too complex, attackers can evade or bypass them. Many programmers and vendors are learning this when attackers break into their systems and servers. The argument that the principles are old, and somehow outdated, rings hollow when the results of their violation are a non-secure system.The work from the earlier period is sometimes cast in terms of systems that no longer exist, and that differ in many ways from modern systems. That does not vitiate the ideas and concepts. The ideas and concepts underlie the work done today. Once these are properly understood, applying them in a multiplicity of environments becomes possible. Further, the current mechanisms and technologies will become obsolete and of historical interest themselves as new forms of computing arise. But the underlying principles will live on, to underlie the next generation, indeed the next era, of computing.The philosophy of this book is that certain key concepts underlie all of computer security, and that the study of all parts of computer security enriches the understanding of all parts. And understanding the theory underlying those applications is critical to understanding the applications of security-related technologies and methodologies.Advances in the theory of computer protection have illuminated the foundations of security systems. Issues of abstract modeling, and modeling to meet specific environments, lead to systems designed to achieve a specific and rewarding goal. Theorems such as the "hook-up" theorem and the undecidability of the general security question have indicated the limits of what can be done. Much work and effort is continuing to extend the borders of those limits.Application of these results has improved the quality in the security of the systems being protected. However, the issue is how compatibly the assumptions of the model (and theory) conform to the environment to which the theory is applied. Although our perceptions of how to apply these abstractions are continually increasing, we still have problems correctly transposing the relevant information from a realistic setting to one in which analyses can than proceed. Such abstraction often eliminates vital information. The omitted data may pertain to security in non-obvious ways. Without this information the analysis is flawed.The practitioner needs to know both the theoretical and practical aspects of the art and science of computer security. The theory demonstrates what is possible. The practical makes known what is feasible. The theoretician needs to understand the constraints under which these theories are used. How their results are translated into practical tools and methods. How realistic are the assumptions underlying the theories. Computer Security: Art and Science tries to meet these needs.Unfortunately, no single work can cover all aspects of computer security, so this book focuses on those parts that are, in the author's opinion, most fundamental and most pervasive. The mechanisms exemplify the applications of these principles.OrganizationThe organization of this book reflects this philosophy. It begins with mathematical fundamentals and principles. These provide boundaries within which security can be modeled and analyzed effectively. The mathematics provides a framework to express and analyze the requirements of the security of a system. These policies constrain what is allowed and not allowed. Mechanisms provide the ability to implement these policies. The degree to which the mechanisms correctly implement the policies, and indeed the degree to which the policies themselves meet the requirements of the organizations using the system, is a question of assurance. Exploiting failures in policy, in implementation, and in assurance come next, as well as mechanisms to provide information on the attack. The book concludes with the applications of both theory and policy focused on realistic situations. This natural progression emphasizes the development and application of the principles existent in computer security.Part 1, "Introduction," describes what computer security is all about, and explores the problems and challenges to be faced. It sets the context for the remainder of the book.Part 2, "Foundations," deals with basic questions such as: how can "security" be clearly and functionally defined? Is it realistic? Is it decidable? If so, under what conditions, and if not, how must the definition be bounded in order to make it decidable?Part 3, "Policy," probes the relationship between policy and security. The definition of "security" depends upon policy. In this section we examine several types of policies, including the ever-present fundamental questions of trust, analysis of policies, and the use of policies to constrain operations and transitions.Part 4, "Implementation I: Cryptography," discusses cryptography and its role in security. This section focuses on applications, and discusses issues such as key management and escrow, key distribution, and how cryptosystems are used in networks. A quick study of authentication completes this section.Part 5, "Implementation II: Systems," considers how to implement the requirements imposed by policies using system-oriented techniques. Certain design principles are fundamental to effective security mechanisms. Policies define who can act, and how, so identity is a critical aspect of implementation. Mechanisms implementing access control and flow control enforce various aspects of policies.Part 6, "Assurance," presents methodologies and technologies for ascertaining how well a system, or product, meets its goals. After setting the background, to explain exactly what "assurance" is, the art of building systems to meet varying levels of assurance is discussed. Formal verification methods play a role. This part shows how the progression of standards enhanced our understanding of assurance techniques.Part 7, "Special Topics," discusses some miscellaneous aspects of computer security. Malicious logic thwarts many mechanisms. Despite our best efforts at high assurance, systems today are replete with vulnerabilities. Why? How can a system be analyzed to detect vulnerabilities? What models might help us improve the state of the art? Given these security holes, how can we detect attackers who exploit them? A discussion of auditing flows naturally into a discussion of intrusion detection, a detection method for such attacks.Part 8, "Practicum," presents examples of how to apply the principles discussed throughout the book. It begins with networks, proceeding to systems, users, and programs. Each chapter states a desired policy, and shows how to translate that policy into a set of mechanisms and procedures that support the policy. This section tries to demonstrate that the material covered elsewhere can be, and should be, used in practice.Each chapter in this book ends with a summary, a description of some research issues, some suggestions for further reading. The summary highlights the important ideas in the chapter. The research issues are current "hot topics" or are topics that may prove fertile ground for advancing the state of the art and science of computer security. Interested readers who wish to pursue the topics in the chapter in more depth can go to some of the suggested readings. They expand on the material in the chapter, or present other interesting avenues.RoadmapThis book is both a reference book and a textbook. Its audience is undergraduate and graduate students as well as practitioners. This section offers some suggestions on approaching the book. Chapter 1 is fundamental to the rest of the book, and should be read first. After that, though, the reader need not follow the chapters in order. The following lists some of the dependencies among chapters. Chapter 3 depends on chapter 2, and requires a fair degree of mathematical maturity. Chapter 2, on the other hand, does not. The material in Chapter 3 is for the most part not used elsewhere (although the existence of the first section's key result, the undecidability theorem, is mentioned repeatedly). It can be safely skipped if the interests of the reader lie elsewhere. The chapters in part 3 build upon one another. The formalisms in chapter 5 are called upon in chapters 19 and 20, but nowhere else. Unless the reader intends to delve into the sections on theorem proving and formal mappings, the formalisms may be skipped. The material in chapter 8 requires a degree of mathematical maturity, and the material in it is used sparingly elsewhere. Like chapter 3, the reader whose interests lie elsewhere can skip it. Chap and 11 also build on one another in order. A reader who has encountered basic cryptography will have an easier time with the material than one who has not, but the chapters do not demand the level of mathematical experience that chapters 3 and 8 do. Chapter 12 does not require material from chapters 10 or 11, but does require material from chapter 9. Chapter 13 is required for all of part 5. A reader who has studied operating systems at the undergraduate level will have no trouble with chapter 15. Chapter 14 uses the material in chapter 11; chapter 16 builds on material in chapters 5, 13, and 15; and chapter 17 uses chapters 4, 13, and 16. Chapter 18 relies upon information in chapter 4. Chapter 19 builds on chapters 5, 13, 15, and 18. Chapter 20 presents highly mathematical concepts, and uses material from chapters 18 and 19. Chapter 21 is based upon material in chapters 5, 18, and 19; it does not require chapter 20. For all of part 5, knowledge of software engineering is very helpful. Chapter 22 draws on ideas and information in chapters 5, 6, 9, 13, 15, and 17 (and for section 22.6, the reader should read section 3.1). Chapter 24 uses information from chapter 4, and chapter 25 uses material from chapter 24. Chapter 23 is self-contained, although it implicitly uses many ideas from assurance. It also assumes a good working knowledge of compilers, operating systems, and in some cases networks. Many of the flaws are drawn from versions of the UNIX operating system, or from Windows systems, so a working knowledge of either or both systems will make some of the material easier to understand.The practicum chapters are self-contained, and do not require any material beyond chapter. However, they point out relevant material in other sections that augments the information, and (we hope) the reader's understanding, of the material in the chapter.BackgroundThe material in this book is at the advanced undergraduate level. Throughout, we assume the reader is familiar with the basics of compilers and computer architecture (such as the use of the program stack) and operating systems. The reader should also be comfortable with modular arithmetic (for the material on cryptography). Some material, such as that on formal methods (chapter 20) and the mathematical theory of computer security (chapter 3 and the formal presentation of policy models) requires considerable mathematical maturity. Other specific recommended background is presented in the previous section. The end matter (chapters 30 through 34) contains material that will be helpful to readers with backgrounds that lack some of the recommended material.Examples are drawn from many systems. Many come from the UNIX operating system or variations of it (such as Linux). Others come from the Windows family of systems. Familiarity with these systems will help the reader understand many examples easily and quickly.Undergraduate LevelAn undergraduate class typically focuses on applications of theory and how students can use the material. The specific arrangement and selection of material depends upon the focus of the class, but all classes should cover some basic material, notably that in chapters 1, 9, and 13, as well as the notion of an access control matrix in sections 2.1 and 2.2.Presenting real problems and solutions often engages undergraduate students more effectively than presenting abstractions. The special topics and practicum provide a wealth of practical problems and ways to deal with them. This leads naturally to the deeper issues of policy, cryptography, non-cryptographic mechanisms, and assurance. The following are sections appropriate for non-mathematical undergraduate courses in these topics: Policy: Sections 4.1 through 4.4 describe the notion of policy. The instructor should select one or two examples from sections 5.1, 5.2.1, 6.2, 6.4, 7.1.1, and 7.2, which describe several policy models informally. Section 7.4 discusses role-based access control. Cryptography: Key distribution is discussed in sections 10,1, 10.2, and a common form of public key infrastructures (called PKI), in 10.4.2. Section 11.1 points out common errors in using cryptography. Section 11.3 shows how cryptography is used in networks, and the instructor should use one of the protocols in 11.4 as an example. Chapter 12 offers a look at various forms of authentication, including noncryptographic methods. Non-cryptographic mechanisms: Identity is the basis for many access control mechanisms. Sections 14.1 through 14.4 discuss identity on a system, and 14.6 discusses identity and anonymity on the web. Sections 15.1 and 15.2 explore two mechanisms to control access to files, and section 15.4 discusses the ring-based mechanism underlying the notion of multiple levels of privilege. If desired, the instructor can cover sandboxes by using sections 17.1 and 17.2, but as 17.2 uses material from sections 4.5 and 4.5.1, the instructor will need to cover that. Assurance: Chapter 18 provides a basic introduct often-overlooked topic of assurance.Graduate LevelA typical introductory graduate class can focus more deeply on the subject than can an undergraduate class. Like an undergraduate class, a graduate class should cover chapters 1, 9, and 13. Also important are the undecidability result in sections 3.1 and 3.2, which requires that chapter 2 be covered. Beyond that, the instructor can choose from a number of topics, and present them to whatever depth is appropriate. The following are sections suitable for graduate study. Policy models: Part 3 covers many common policy models both informally and formally. The formal description is much easier to understand once the informal description is understood, so in all cases both should be covered. The controversy in section 5.4 is particularly illuminating to students who have not considered the role of policy and the nature of a policy. Chapter 8 is a highly formal discussion of the foundations of policy, and is appropriate for students with experience in formal mathematics. Students without that background will find it quite difficult. Cryptography: Part 4 focuses on the applications of cryptography, and not on cryptography's mathematical underpinnings. It discusses areas of interest critical to the use of cryptography, such as key management and some basic cryptographic protocols used in networking. Non-cryptographic mechanisms: Issues of identity and certification are complex and generally poorly understood. Section 14.5 covers these problems. Combining this with the discussion of identity on the web (section 14.6) raises issues of trust and naming. Chapters 165 and 17 explore issu information flow and confining that flow. Assurance: Traditionally, assurance is taught as formal methods, and Chapter 20 serves this purpose. But in practice, assurance is more often accomplished by using structured processes and techniques and informal, but rigorous, arguments of justification, mappings and analysis. Chapter 19 emphasizes these. Chapter 21 discusses evaluation standards, and relies heavily on the material in chapters 18 and 19, and some on the ideas in chapter 20. Miscellaneous Topics: Section 22.6 presents a proof that the generic problem of determining if a generic program is a computer virus is in fact undecidable. The theory of penetration studies in 23.2, and the more formal approach in section 23.5, illuminate the analysis of systems for vulnerabilities. Id the instructor chooses to cover intrusion detection (chapter 25) in depth, that draws heavily on the material on auditing (chapter 24). Practicum: The practicum ties the material in the earlier part of the book to real-world examples, and emphasizes the applications of the theory and methodologies discussed earlier.PractitionersPractitioners in the field of computer security will find much of interest. The table of contents, and index, will help them locate specific topics. A more general approach is to start at Chapter 1, and then proceed to Part 8, the practicum. Each chapter has references to other sections of the text that explain the underpinnings of the material. This will lead the reader to a deeper understanding of the reasons for the policies, settings, configurations, and advice in the practicum. This approach also allows the reader to focus on topics of most interest.Special AcknowledgmentElisabeth Sullivan contributed the assurance part of this book. She wrote several drafts, all of which reflect her extensive knowledge and experience in that aspect of computer security. I am particularly grateful for her contributing her real-world knowledge of how assurance is managed.