{
    "pages": [
        {
            "page_number": 1,
            "text": " \n \n"
        },
        {
            "page_number": 2,
            "text": "Testing Web Security-Assessing the Security of Web \nSites and Applications  \nSteven Splaine \n \nWiley Publishing, Inc. \nPublisher: Robert Ipsen \nEditor: Carol Long \nDevelopmental Editor: Scott Amerman \nManaging Editor: John Atkins \nNew Media Editor: Brian Snapp \nText Design & Composition: Wiley Composition Services \nDesignations used by companies to distinguish their products are often claimed as \ntrademarks. In all instances where Wiley Publishing, Inc., is aware of a claim, the product \nnames appear in initial capital or ALL CAPITAL LETTERS. Readers, however, should \ncontact the appropriate companies for more complete information regarding trademarks \nand registration. \nThis book is printed on acid-free paper.  \nCopyright © 2002 by Steven Splaine. \nISBN:0471232815 \nAll rights reserved. \nPublished by Wiley Publishing, Inc., Indianapolis, Indiana \nPublished simultaneously in Canada \nNo part of this publication may be reproduced, stored in a retrieval system, or transmitted in \nany form or by any means, electronic, mechanical, photocopying, recording, scanning, or \notherwise, except as permitted under Section 107 or 108 of the 1976 United States \nCopyright Act, without either the prior written permission of the Publisher, or authorization \nthrough payment of the appropriate per-copy fee to the Copyright Clearance Center, Inc., \n222 Rosewood Drive, Danvers, MA 01923, (978) 750-8400, fax (978) 750-4470. Requests \nto the Publisher for permission should be addressed to the Legal Department, Wiley \nPublishing, Inc., 10475 Crosspointe Blvd., Indianapolis, IN 46256, (317) 572-3447, fax \n(317) 572-4447, Email: <permcoordinator@wiley.com>. \nLimit of Liability/Disclaimer of Warranty: While the publisher and author have used their \nbest efforts in preparing this book, they make no representations or warranties with respect \nto the accuracy or completeness of the contents of this book and specifically disclaim any \nimplied warranties of merchantability or fitness for a particular purpose. No warranty may be \ncreated or extended by sales representatives or written sales materials. The advice and \nstrategies contained herein may not be suitable for your situation. You should consult with a \nprofessional where appropriate. Neither the publisher nor author shall be liable for any loss \nof profit or any other commercial damages, including but not limited to special, incidental, \nconsequential, or other damages. \n"
        },
        {
            "page_number": 3,
            "text": "For general information on our other products and services, please contact our Customer \nCare Department within the United States at (800) 762-2974, outside the United States at \n(317) 572-3993 or fax (317) 572-4002. \nWiley also publishes its books in a variety of electronic formats. Some content that appears \nin print may not be available in electronic books. \nLibrary of Congress Cataloging-in-Publication Data:  \n0-471-23281-5 \nPrinted in the United States of America \n10 9 8 7 6 5 4 3 2 1  \nTo my wife Darlene and our sons, Jack and Sam, who every day remind me of just how \nfortunate I am.  \nTo the victims and heroes of September 11, 2001, lest we forget that freedom must always \nbe vigilant.  \nAcknowledgments  \nThe topic of Web security is so large and the content is so frequently changing that it is \nimpossible for a single person to understand every aspect of security testing. For this \nreason alone, this book would not have been possible without the help of the many security \nconsultants, \ntesters, \nwebmasters, \nproject \nmanagers, \ndevelopers, \nDBAs, \nLAN \nadministrators, firewall engineers, technical writers, academics, and tool vendors who were \nkind enough to offer their suggestions and constructive criticisms, which made this book \nmore comprehensive, accurate, and easier to digest. \nMany thanks to the following team of friends and colleagues who willingly spent many hours \nof what should have been their free time reviewing this book and/or advising me on how \nbest to proceed with this project. \nJames Bach \nJoey Maier \nRex Black \nBrian McCaughey \nRoss Collard \nWayne Middleton \nRick Craig \nClaudette Moore \nDan Crawford \nDavid Parks \nYves de Montcheuil \nEric Patel \nMickey Epperson \nRoger Rivest \nDanny Faught \nMartin Ryan \nPaul Gerrard \nJohn Smentowski \nStefan Jaskiel \nJohn Splaine \nJeff Jones \nHerbert Thompson \nPhilip Joung \nMichael Waldmann \nA special thank-you goes to my wife Darlene and our sons Jack and Sam, for their love and \ncontinued support while I was writing this book. I would especially like to thank Jack for \nunderstanding why Daddy couldn't go play ball on so many evenings. \n \n"
        },
        {
            "page_number": 4,
            "text": "Professional Acknowledgment  \nI would like to thank everyone who helped me create and then extend Software Quality \nEngineering's Web Security Testing course (www.sqe.com), the source that provided much \nof the structure and content for this book. Specifically, many of SQE's staff, students, and \nclients provided me with numerous suggestions for improving the training course, many of \nwhich were subsequently incorporated into this book. \nSTEVEN SPLAINE is a chartered software engineer with more than twenty years of \nexperience in project management, software testing and product development. He is a \nregular speaker at software testing conferences and lead author of The Web Testing \nHandbook. \n"
        },
        {
            "page_number": 5,
            "text": "Foreword \nAs more and more organizations move to Internet-based and intranet-based applications, \nthey find themselves exposed to new or increased risks to system quality, especially in the \nareas of performance and security. Steven Splaine's last book, The Web Testing \nHandbook, provided the reader with tips and techniques for testing performance along with \nmany other important considerations for Web testing, such as functionality. Now Steve \ntakes on the critical issue of testing Web security. \nToo many users and even testers of Web applications believe that solving their security \nproblems merely entails buying a firewall and connecting the various cables. In this book, \nSteve identifies this belief as the firewall myth, and I have seen victims of this myth in my \nown testing, consulting, and training work. This book not only helps dispel this myth, but it \nalso provides practical steps you can take that really will allow you to find and resolve \nsecurity problems throughout the network. Client-side, server-side, Internet, intranet, \noutside hackers and inside jobs, software, hardware, networks, and social engineering, it's \nall covered here. How should you run a penetration test? How can you assess the level of \nrisk inherent in each potential security vulnerability, and test appropriately? When \nconfronted with an existing system or building a new one, how do you keep track of \neverything that's out there that could conceivably become an entryway for trouble? In a \nreadable way, Steve will show you the ins and outs of Web security testing. This book will \nbe an important resource for me on my next Web testing project. If you are responsible for \nthe testing or security of a Web system, I bet it will be helpful to you, too. \nRex Black \nRex Black Consulting \nBulverde, Texas \n"
        },
        {
            "page_number": 6,
            "text": "Preface \nAs the Internet continues to evolve, more and more organizations are replacing their \nplaceholder or brochureware Web sites with mission-critical Web applications designed to \ngenerate revenue and integrate with their existing systems. One of the toughest challenges \nfacing those charged with implementing these corporate goals is ensuring that these new \nstorefronts are safe from attack and misuse. \nCurrently, the number of Web sites and Web applications that need to be tested for security \nvulnerabilities far exceeds the number of security professionals who are sufficiently \nexperienced to carry out such an assessment. Unfortunately, this means that many Web \nsites and applications are either inadequately tested or simply not tested at all. These \norganizations are, in effect, playing a game of hacker roulette, just hoping to stay lucky. \nA significant reason that not enough professionals are able to test the security of a Web site \nor application is the lack of introductory-level educational material. Much of the educational \nmaterial available today is either high-level/strategic in nature and aimed at senior \nmanagement and chief architects who are designing the high-level functionality of the \nsystem, or low-level/extremely technical in nature and aimed at experienced developers \nand network engineers charged with implementing these designs. \nTesting Web Security is an attempt to fill the need for a straightforward, easy-to-follow book \nthat can be used by anyone who is new to the security-testing field. Readers of my first \nbook that I coauthored with Stefan Jaskiel will find I have retained in this book the checklist \nformat that we found to be so popular with The Web Testing Handbook (Splaine and \nJaskiel, 2001) and will thereby hopefully make it easier for security testers to ensure that \nthe developers and network engineers have implemented a system that meets the explicit \n(and implied) security objectives envisioned by the system's architects and owners. \nSteven Splaine \nTampa, Florida \n"
        },
        {
            "page_number": 7,
            "text": " \nTable of Contents  \n \nTesting Web Security—Assessing the Security of Web\nSites and Applications  \n Foreword  \n Preface  \n Part I - An Introduction to the Book \n Chapter 1 - Introduction \n Part II - Planning the Testing Effort \n Chapter 2 - Test Planning \n Part III - Test Design \n Chapter 3 - Network Security \n Chapter 4 - System Software Security \n Chapter 5 - Client-Side Application Security \n Chapter 6 - Server-Side Application Security \n Chapter 7 - Sneak Attacks: Guarding Against the Less-\nThought-of Security Threats \n Chapter 8 - Intruder \nConfusion, \nDetection, \nand\nResponse \n Part IV - Test Implementation \n Chapter 9 - Assessment and Penetration Options \n Chapter 10 - Risk Analysis \n Epilogue  \n Part V - Appendixes \n Appendix A - An \nOverview \nof \nNetwork \nProtocols,\nAddresses, and Devices \n Appendix B - SANS Institute Top 20 Critical Internet\nSecurity Vulnerabilities \n Appendix C - Test-Deliverable Templates \n Additional Resources  \n Index  \n List of Figures  \n List of Tables  \n List of Sidebars  \n \n"
        },
        {
            "page_number": 8,
            "text": " \n1 \nPart I: An Introduction to the Book \nChapter List \nChapter 1: Introduction  \n"
        },
        {
            "page_number": 9,
            "text": " \n2 \nChapter 1: Introduction \nOverview \nThe following are some sobering statistics and stories that seek to illustrate the growing \nneed to assess the security of Web sites and applications. The 2002 Computer Crime and \nSecurity Survey conducted by the Computer Security Institute (in conjunction with the San \nFrancisco Federal Bureau of Investigation) reported the following statistics (available free of \ncharge via www.gocsi.com): \nƒ \nNinety percent of respondents (primarily large corporations and government \nagencies) detected computer security breaches within the last 12 months. \nƒ \nSeventy-four percent of respondents cited their Internet connection as a frequent \npoint of attack, and 40 percent detected system penetration from the outside. \nƒ \nSeventy-five percent of respondents estimated that disgruntled employees were the \nlikely source of some of the attacks that they experienced. \nThe following lists the number of security-related incidents reported to the CERT \nCoordination Center (www.cert.org) for the previous 4 1/2 years: \nƒ \n2002 (Q1 and Q2)-43,136 \nƒ \n2001-52,658 \nƒ \n2000-21,756 \nƒ \n1999-9,859 \nƒ \n1998-3,734 \nIn February 2002, Reuters (www.reuters.co.uk) reported that \"hackers\" forced CloudNine \nCommunications-one of Britain's oldest Internet service providers (ISPs) -out of business. \nCloudNine came to the conclusion that the cost of recovering from the attack was too great \nfor the company to bear, and instead elected to hand over their customers to a rival ISP. \nIn May 2002, CNN/Money (www.money.cnn.com) reported that the financing division of a \nlarge U.S. automobile manufacturer was warning 13,000 people to be aware of identity theft \nafter the automaker discovered \"hackers\" had posed as their employees in order to gain \naccess to consumer credit reports. \n \nThe Goals of This Book \nThe world of security, especially Web security, is a very complex and extensive knowledge \ndomain to attempt to master-one where the consequences of failure can be extremely high. \nPractitioners can spend years studying this discipline only to realize that the more they \nknow, the more they realize they need to know. In fact, the challenge may seem to be so \ndaunting that many choose to shy away from the subject altogether and deny any \nresponsibility for the security of the system they are working on. \"We're not responsible for \nsecurity-somebody else looks after that\" is a common reason many members of the project \nteam give for not testing a system's security. Of course, when asked who the somebody \nelse is, all too often the reply is \"I don't know,\" which probably means that the security \ntesting is fragmented or, worse still, nonexistent. \nA second hindrance to effective security testing is the naive belief held by many owners and \nsenior managers that all they have to do to secure their internal network and its applications \nis purchase a firewall appliance and plug it into the socket that the organization uses to \nconnect to the Internet. Although a firewall is, without doubt, an indispensable defense for a \nWeb site, it should not be the only defense that an organization deploys to protect its Web \nassets. The protection afforded by the most sophisticated firewalls can be negated by a \n"
        },
        {
            "page_number": 10,
            "text": " \n3 \npoorly designed Web application running on the Web site, an oversight in the firewall's \nconfiguration, or a disgruntled employee working from the inside. \n \nTHE FIREWALL MYTH \nThe firewall myth is alive and well, as the following two true conversations illustrate. \nAnthony is a director at a European software-testing consultancy, and Kevin is the owner of \na mass-marketing firm based in Florida. \nƒ Anthony: We just paid for someone to come in and install three top-of-the-line \nfirewalls, so we're all safe now. \nƒ Security tester: Has anybody tested them to make sure they are configured \ncorrectly? \nƒ Anthony: No, why should we? \nƒ Kevin: We're installing a new wireless network for the entire company. \nƒ Security tester: Are you encrypting the data transmissions? \nƒ Kevin: I don't know; what difference does it make? No one would want to hack us, \nand even if they did, our firewall will protect us. \nThis book has two goals. The first goal is to raise the awareness of those managers \nresponsible for the security of a Web site, conveying that a firewall should be part of the \nsecurity solution, but not the solution. This information can assist them in identifying and \nplanning the activities needed to test all of the possible avenues that an intruder could use \nto compromise a Web site. The second goal is aimed at the growing number of individuals \nwho are new to the area of security testing, but are still expected to evaluate the security of \na Web site. Although no book can be a substitute for years of experience, this book \nprovides descriptions and checklists for hundreds of tests that can be adapted and used as \na set of candidate test cases. These tests can be included in a Web site's security test \nplan(s), making the testing effort more comprehensive than it would have been otherwise. \nWhere applicable, each section also references tools that can be used to automate many of \nthese tasks in order to speed up the testing process. \n \nThe Approach of This Book \nTesting techniques can be categorized in many different ways; white box versus black box \nis one of the most common categorizations. Black-box testing (also known as behavioral \ntesting) treats the system being tested as a black box into which testers can't see. As a \nresult, all the testing must be conducted via the system's external interfaces (for example, \nvia an application's Web pages), and tests need to be designed based on what the system \nis expected to do and in accordance with its explicit or implied requirements. White-box \ntesting assumes that the tester has direct access to the source code and can look into the \nbox and see the inner workings of the system. This is why white-box testing is sometimes \nreferred to as clear-box, glass-box, translucent, or structural testing. Having access to the \nsource code helps testers to understand how the system works, enabling them to design \ntests that will exercise specific program execution paths. Input data can be submitted via \nexternal or internal interfaces. Test results do not need to be based solely on external \noutputs; they can also be deduced from examining internal data stores (such as records in \nan application's database or entries in an operating system's registry). \nIn general, neither testing approach should be considered inherently more effective at \nfinding defects than the other, but depending upon the specific context of an individual \ntesting project (for example, the background of the people who will be doing the testing-\ndeveloper oriented versus end-user oriented), one approach could be easier or more cost-\neffective to implement than the other. Beizer (1995), Craig et al. (2002), Jorgensen (2002), \n"
        },
        {
            "page_number": 11,
            "text": " \n4 \nand Kaner et al. (1999) provide additional information on black-box and white-box testing \ntechniques. \nGray-box testing techniques can be regarded as a hybrid approach. In other words, a tester \nstill tests the system as a black box, but the tests are designed based on the knowledge \ngained by using white-box-like investigative techniques. Gray-box testers using the \nknowledge gained from examining the system's internal structure are able to design more \naccurate/focused tests, which yield higher defect detection rates than those achieved using \na purely traditional black-box testing approach. At the same time, however, gray-box testers \nare also able to execute these tests without having to use resource-consuming white-box \ntesting infrastructures. \nGRAY-BOX TESTING \nGray-box testing incorporates elements of both black-box and white-box testing. It consists \nof methods and tools derived from having some knowledge of the internal workings of the \napplication and the environment with which it interacts. This extra knowledge can be \napplied in black-box testing to enhance testing productivity, bug finding, and bug-analyzing \nefficiency. \nSource: Nguyen (2000). \nWherever possible, this book attempts to adopt a gray-box approach to security testing. By \ncovering the technologies used to build and deploy the systems that will be tested and then \nexplaining the potential pitfalls (or vulnerabilities) of each technology design or \nimplementation strategy, the reader will be able to create more effective tests that can still \nbe executed in a resource-friendly black-box manner. \nThis book stops short of describing platform- and threat-specific test execution details, such \nas how to check that a Web site's Windows 2000/IIS v5.0 servers have been protected from \nan attack by the Nimda worm (for detailed information on this specific threat, refer to CERT \nadvisory CA-2001-26-www.cert.org). Rather than trying to describe in detail the specifics of \nthe thousands of different security threats that exist today (in the first half of 2002 alone, the \nCERT Coordination Center recorded 2,148 reported vulnerabilities), this book describes \ngeneric tests that can be extrapolated and customized by the reader to accommodate \nindividual and unique needs. In addition, this book does not expand on how a security \nvulnerability could be exploited (information that is likely to be more useful to a security \nabuser than a security tester) and endeavors to avoid making specific recommendations on \nhow to fix a security vulnerability, since the most appropriate remedy will vary from \norganization to organization and such a decision (and subsequent implementation) would \ngenerally be considered to be the role of a security designer. \n \nHow This Book Is Organized \nAlthough most readers will probably find it easier to read the chapters in sequential order, \nthis book has been organized in a manner that permits readers to read any of the chapters \nin any order. Depending on the background and objectives of different readers, some may \neven choose to skip some of the chapters. For example, a test manager who is well versed \nin writing test plans used to test the functionality of a Web application may decide to skip \nthe chapter on test planning and focus on the chapters that describe some of the new types \nof tests that could be included in his or her test plans. In the case of an application \ndeveloper, he or she may not be concerned with the chapter on testing a Web site's \nphysical security because someone else looks after that (just so long as someone actually \ndoes) and may be most interested in the chapters on application security. \n"
        },
        {
            "page_number": 12,
            "text": " \n5 \nTo make it easier for readers to hone in on the chapters that are of most interest to them, \nthis book has been divided into four parts. Part 1 is comprised of this chapter and provides \nan introduction and explanation of the framework used to construct this book. \nChapter 2, \"Test Planning,\" provides the material for Part 2, \"Planning the Testing Effort,\" \nand looks at the issues surrounding the planning of the testing effort. \nPart 3, \"Test Design,\" is the focus of this book and therefore forms the bulk of its content by \nitemizing the various candidate tests that the testing team should consider when evaluating \nwhat they are actually going to test as part of the security-testing effort of a Web site and its \nassociated Web application(s). Because the testing is likely to require a variety of different \nskill sets, it's quite probable that different people will execute different groups of tests. With \nthis consideration in mind, the tests have been grouped together based on the typical skill \nsets and backgrounds of the people who might be expected to execute them. This part \nincludes the following chapters: \nƒ \nChapter 3: Network Security \nƒ \nChapter 4: System Software Security \nƒ \nChapter 5: Client-Side Application Security \nƒ \nChapter 6: Server-Side Application Security \nƒ \nChapter 7: Sneak Attacks: Guarding against the Less-Thought-of Security Threats \nƒ \nChapter 8: Intruder Confusion, Detection, and Response \nHaving discussed what needs to be tested, Part 4, \"Test Implementation,\" addresses the \nissue of how to best execute these tests in terms of who should actually do the work, what \ntools should be used, and what order the tests should be performed in (ranking test \npriority). This part includes the following chapters: \nƒ \nChapter 9: Assessment and Penetration Options \nƒ \nChapter 10: Risk Analysis \nAs a means of support for these 10 chapters, the appendix provides some additional \nbackground information, specifically: a brief introduction to the basics of computer networks \nas utilized by many Web sites (in case some of the readers of this book are unfamiliar with \nthe components used to build Web sites), a summarized list of the top-20 critical Internet \nsecurity vulnerabilities (as determined by the SANS Institute), and some sample test \ndeliverable templates (which a security-testing team could use as a starting point for \ndeveloping their own customized documentation). \nFinally, the resources section not only serves as a bibliography of all the books and Web \nsites referenced in this book, but it also lists other reference books that readers interested \nin testing Web security may find useful in their quest for knowledge. \n \nTerminology Used in This Book \nThe following two sections describe some of the terms used in this book to describe the \nindividuals who might seek to exploit a security vulnerability on a Web site-and hence the \npeople that a security tester is trying to inhibit-and the names given to some of the more \ncommon deliverables that a security tester is likely to produce. \nHackers, Crackers, Script Kiddies, and Disgruntled Insiders \nThe term computer hacker was originally used to describe someone who really knew how \nthe internals of a computer (hardware and/or software) worked and could be relied on to \ncome up with ingenious workarounds (hacks) to either fix a problem with the system or \nextend its original capabilities. Somewhere along the line, the popular press relabeled this \n"
        },
        {
            "page_number": 13,
            "text": " \n6 \nterm to describe someone who tries to acquire unauthorized access to a computer or \nnetwork of computers. \nThe terminology has become further blurred by the effort of some practitioners to \ndifferentiate the skill levels of those seeking unauthorized access. The term cracker is \ntypically used to label an attacker who is knowledgeable enough to create his or her own \nhacks, whereas the term script kiddie is used to describe a person who primarily relies on \nthe hacks of others (often passed around as a script or executable). The situation becomes \neven less clear if you try to pigeonhole disgruntled employees who don't need to gain \nunauthorized access in order to accomplish their malicious goals because they are already \nauthorized to access the system. \nNot all attackers are viewed equally. Aside from their varying technical expertise, they also \nmay be differentiated by their ethics. Crudely speaking, based on their actions and \nintentions, attackers are often be categorized into one of the following color-coded groups: \nƒ \nWhite-hat hackers. These are individuals who are authorized by the owner of a \nWeb site or Web-accessible product to ascertain whether or not the site or product is \nadequately protected from known security loopholes and common generic exploits. \nThey are also known as ethical hackers, or are part of a group known as a tiger team \nor red team. \nƒ \nGray-hat hackers. Also sometimes known as wackers, gray-hat hackers attack a \nnew product or technology on their own initiative to determine if the product has any \nnew security loopholes, to further their own education, or to satisfy their own curiosity. \nAlthough their often-stated aim is to improve the quality of the new technology or their \nown knowledge without directly causing harm to anyone, their methods can at times \nbe disruptive. For example, some of these attackers will not inform the product's \nowner of a newly discovered security hole until they have had time to build and \npublicize a tool that enables the hole to be easily exploited by others. \n \nƒ HACKER \nƒ Webster's II New Riverside Dictionary offers three alternative definitions for the word \nhacker, the first two of which are relevant for our purposes: \no 1 a. Computer buff \no 1b. One who illegally gains access to another's electronic system \n \n \nƒ COLOR-CODING ATTACKERS \nƒ The reference to colored hats comes from Hollywood's use of hats in old black-and-\nwhite cowboy movies to help an audience differentiate between the good guys \n(white hats) and the bad guys (black hats). \n \n \n \n• Black-hat hackers. Also known as crackers, these are attackers who typically \nseek to exploit known (and occasionally unknown) security holes for their own \npersonal gain. Script kiddies are often considered to be the subset of black-\nhatters, whose limited knowledge forces them to be dependent almost exclusively \nupon the tools developed by more experienced attackers. Honeynet Project \n(2001) provides additional insight into the motives of black-hat hackers. \nOf course, assigning a particular person a single designation can be somewhat arbitrary \nand these terms are by no means used consistently across the industry; many people have \nslightly different definitions for each category. The confusion is compounded further when \n"
        },
        {
            "page_number": 14,
            "text": " \n7 \nconsidering individuals who do not always follow the actions of just one definition. For \ninstance, if an attacker secretly practices the black art at night, but also publicly fights the \ngood fight during the day, what kind of hatter does that make him? \nRather than use terms that potentially carry different meanings to different readers (such as \nhacker), this book will use the terms attacker, intruder, or assailant to describe someone \nwho is up to no good on a Web site. \nTesting Vocabulary \nMany people who are new to the discipline of software testing are sometimes confused \nover exactly what is meant by some of the common terminology used to describe various \nsoftware-testing artifacts. For example, they might ask the question, \"What's the difference \nbetween a test case and a test run?\" This confusion is in part due to various practitioners, \norganizations, book authors, and professional societies using slightly different vocabularies \nand often subtly different definitions for the terms defined within their own respective \nvocabularies. These terms and definitions vary for many reasons. Some definitions are \nembryonic (defined early in this discipline's history), whereas others reflect the desire by \nsome practitioners to push the envelope of software testing to new areas. \nThe following simple definitions are for the testing artifacts more frequently referenced in \nthis book. They are not intended to compete with or replace the more verbose and exacting \ndefinitions already defined in industry standards and other published materials, such as \nthose defined by the Institute of Electrical and Electronics Engineers (www.ieee.org), the \nProject \nManagement \nInstitute \n(www.pmi.org), \nor \nRational's \nUnified \nProcess \n(www.rational.com). Rather, they are intended to provide the reader with a convenient \nreference of how these terms are used in this book. Figure 1.1 graphically summarizes the \nrelationship between each of the documents. \n \n \n \nFigure 1.1: Testing documents.  \n \n \nƒ \nTest plan. A test plan is a document that describes the what, why, who, when, and \nhow of a testing project. Some testing teams may choose to describe their entire \ntesting effort within a single test plan, whereas others find it easier to organize groups \nof tests into two or more test plans, with each test plan focusing on a different aspect \nof the testing effort. \nTo foster better communication across projects, many organizations have defined test \nplan templates. These templates are then used as a starting point for each new test \nplan, and the testing team refines and customizes each plan(s) to fit the unique needs \nof their project. \n \n"
        },
        {
            "page_number": 15,
            "text": " \n8 \nƒ \nTest item. A test item is a hardware device or software program that is the subject of \nthe testing effort. The term system under test is often used to refer to the collection of \nall test items. \nƒ \nTest. A test is an evaluation with a clearly defined objective of one or more test \nitems. A sample objective could look like the following: \"Check that no unneeded \nservices are running on any of the system's servers.\" \nƒ \nTest case. A test case is a detailed description of a test. Some tests may \nnecessitate utilizing several test cases in order to satisfy the stated objective of a \nsingle test. The description of the test case could be as simple as the following: \n\"Check that NetBIOS has been disabled on the Web server.\" It could also provide \nadditional details on how the test should be executed, such as the following: \"Using \nthe tool nmap, an external port scan will be performed against the Web server to \ndetermine if ports 137-139 have been closed.\" \nDepending on the number and complexity of the test cases, a testing team may choose \nto specify their test cases in multiple test case documents, consolidate them into a \nsingle document, or possibly even embed them into the test plan itself. \nƒ \nTest script. A test script is a series of steps that need to be performed in order to \nexecute a test case. Depending on whether the test has been automated, this series \nof steps may be expressed as a sequence of tasks that need to be performed \nmanually or as the source code used by an automated testing tool to run the test. \nNote that some practitioners reserve the term test script for automated scripts and \nuse the term test procedure for the manual components. \nƒ \nTest run. A test run is the actual execution of a test script. Each time a test case is \nexecuted, it creates a new instance of a test run. \n \nWho Should Read This Book? \nThis book is aimed at three groups of people. The first group consists of the owners, CIOs, \nmanagers, and security officers of a Web site who are ultimately responsible for the security \nof their site. Because these people might not have a strong technical background and, \nconsequently, not be aware of all the types of threats that their site faces, this book seeks \nto make these critical decision makers aware of what security testing entails and thereby \nenable them to delegate (and fund) a security-testing effort in a knowledgeable fashion. \nThe second group of individuals who should find this book useful are the architects and \nimplementers of a Web site and application (local area network [LAN] administrators, \ndevelopers, database administrators [DBAs], and so on) who may be aware of some (or all) \nof the security factors that should be considered when designing and building a Web site, \nbut would appreciate having a checklist of security issues that they could use as they \nconstruct the site. These checklists can be used in much the same way that an experienced \nairplane pilot goes through a mandated preflight checklist before taking off. These are \nhelpful because the consequences of overlooking a single item can be catastrophic. \nThe final group consists of the people who may be asked to complete an independent \nsecurity assessment of the Web site (in-house testers, Q/A analysts, end users, or outside \nconsultants), but may not be as familiar with the technology (and its associated \nvulnerabilities) as the implementation group. For the benefit of these people, this book \nattempts to describe the technologies commonly used by implementers to build Web sites \nto a level of detail that will enable them to test the technology effectively but without getting \nas detailed as a book on how to build a Web site. \n \n"
        },
        {
            "page_number": 16,
            "text": " \n9 \nSummary \nWith the heightened awareness for the need to securely protect an organization's electronic \nassets, the supply of available career security veterans is quickly becoming tapped out, \nwhich has resulted in an influx of new people into the field of security testing. This book \nseeks to provide an introduction to Web security testing for those people with relatively little \nexperience in the world of information security (infosec), allowing them to hit the ground \nrunning. It also serves as an easy-to-use reference book that is full of checklists to assist \ncareer veterans such as the growing number of certified information systems security \nprofessionals (CISSPs) in making sure their security assessments are as comprehensive \nas they can be. Bragg (2002), Endorf (2001), Harris (2001), Krutz et al. (2001 and 2002), \nPeltier (2002), the CISSP Web portal (www.cissp.com), and the International Information \nSystems Security Certifications Consortium (www.isc2.org) provide additional information \non CISSP certification. \n"
        },
        {
            "page_number": 17,
            "text": " \n10 \nPart II: Planning the Testing Effort \nChapter List \nChapter 2: Test Planning  \n"
        },
        {
            "page_number": 18,
            "text": " \n11 \nChapter 2: Test Planning \nFailing to adequately plan a testing effort will frequently result in the project's sponsors \nbeing unpleasantly surprised. The surprise could be in the form of an unexpected cost \noverrun by the testing team, or finding out that a critical component of a Web site wasn't \ntested and consequently permitted an intruder to gain unauthorized access to company \nconfidential information. \nThis chapter looks at the key decisions that a security-testing team needs to make while \nplanning their project, such as agreeing on the scope of the testing effort, assessing the \nrisks (and mitigating contingencies) that the project may face, spelling out any rules of \nengagement (terms of reference) for interacting with a production environment, and \nspecifying which configuration management practices to use. Failing to acknowledge any \none of these considerations could have potentially dire consequences to the success of the \ntesting effort and should therefore be addressed as early as possible in the project. Black \n(2002), Craig et al. (2002), Gerrard et al. (2002), Kaner et al. (1999, 2001), the Ideahamster \nOrganization (www.ideahamster.org), and the Rational Unified Process (www.rational.com) \nprovide additional information on planning a testing project. \nRequirements \nA common practice among testing teams charged with evaluating how closely a system will \nmeet its user's (or owner's) expectations is to design a set of tests that confirm whether or \nnot all of the features explicitly documented in a system's requirements specification have \nbeen implemented correctly. In other words, the objectives of the testing effort are \ndependent upon on the system's stated requirements. For example, if the system is \nrequired to do 10 things and the testing team runs a series of tests that confirm that the \nsystem can indeed accurately perform all 10 desired tasks, then the system will typically be \nconsidered to have passed. Unfortunately, as the following sections seek to illustrate, this \nprocess is nowhere near as simple a task to accomplish as the previous statement would \nlead you to believe. \nClarifying Requirements \nIdeally, a system's requirements should be clearly and explicitly documented in order for the \nsystem to be evaluated to determine how closely it matches the expectations of the \nsystem's users and owners (as enshrined by the requirements documentation). \nUnfortunately, a testing team rarely inherits a comprehensive, unambiguous set of \nrequirements; often the requirements team-or their surrogates, who in some instances may \nend up being the testing team-ends up having to clarify these requirements before the \ntesting effort can be completed (or in some cases started). The following are just a few \nsituations that may necessitate revisiting the system's requirements: \nƒ \nImplied requirements. Sometimes requirements are so obvious (to the \nrequirements author) that the documentation of these requirements is deemed to be a \nwaste of time. For example, it's rare to see a requirement such as \"no spelling \nmistakes are to be permitted in the intruder response manual\" explicitly documented, \nbut at the same time, few organizations would regard spelling mistakes as desirable. \nƒ \nIncomplete or ambiguous requirements. A requirement that states, \"all the Web \nservers should have service pack 3 installed,\" is ambiguous. It does not make it clear \nwhether the service pack relates to the operating system or to the Web service \n(potentially different products) or which specific brand of system software is required. \n"
        },
        {
            "page_number": 19,
            "text": " \n12 \nƒ \nNonspecific requirements. Specifying \"strong passwords must be used\" may \nsound like a good requirement, but from a testing perspective, what exactly is a \nstrong password: a password longer than 7 characters or one longer than 10? To be \nconsidered strong, can the password use all uppercase or all lowercase characters, \nor must a mixture of both types of letters be used? \nƒ \nGlobal requirements. Faced with the daunting task of specifying everything that a \nsystem should not do, some requirements authors resort to all-encompassing \nstatements like the following: \"The Web site must be secure.\" Although everyone \nwould agree that this is a good thing, the reality is that the only way the Web site \ncould be made utterly secure is to disconnect it from any other network (including the \nInternet) and lock it behind a sealed door in a room to which no one has access. \nUndoubtedly, this is not what the author of the requirement had in mind. \nFailing to ensure that a system's requirements are verifiable before the construction of the \nsystem is started (and consequently open to interpretation) is one of the leading reasons \nwhy systems need to be reworked or, worse still, a system enters service only for its users \n(or owners) to realize in production that the system is not actually doing what they need it to \ndo. An organization would therefore be well advised to involve in the requirements \ngathering process the individuals who will be charged with verifying the system's capability. \nThese individuals (ideally professional testers) may then review any documented \nrequirement to ensure that it has been specified in such a way that it can be easily and \nimpartially tested. \nMore clearly defined requirements should not only result in less rework on the part of \ndevelopment, but also speed the testing effort, as specific tests not only can be designed \nearlier, but their results are likely to require much less interpretation (debate). Barman \n(2001), Peltier (2001), and Wood (2001) provide additional information on writing security \nrequirements. \nSecurity Policies \nDocumenting requirements that are not ambiguous, incomplete, nonquantifiable, or even \ncontradictory is not a trivial task, but even with clearly defined requirements, a security-\ntesting team faces an additional challenge. Security testing is primarily concerned with \ntesting that a system does not do something (negative testing)-as opposed to confirming \nthat the system can do something (positive testing). Unfortunately, the list of things that a \nsystem (or someone) should not do is potentially infinite in comparison to a finite set of \nthings that a system should do (as depicted in Figure 2.1). Therefore, security requirements \n(often referred to as security policies) are by their very nature extremely hard to test, \nbecause the number of things a system should not do far exceeds the things it should do. \n"
        },
        {
            "page_number": 20,
            "text": " \n13 \n \nFigure 2.1: System capabilities.  \nWhen testing security requirements, a tester is likely to have to focus on deciding what \nnegative tests should be performed to ascertain if the system is capable of doing something \nit should not do (capabilities that are rarely well documented-if at all). Since the number of \ntests needed to prove that a system does not do what it isn't supposed to is potentially \nenormous, and the testing effort is not, it is critically important that the security-testing team \nnot only clarify any vague requirements, but also conduct a risk analysis (the subject of \nChapter 10) to determine what subset of the limitless number of negative tests will be \nperformed by the testing effort. They should then document exactly what (positive and \nnegative tests) will and will not be covered and subsequently ensure that the sponsor of the \neffort approves of this proposed scope. \n \nThe Anatomy of a Test Plan \nOnce a set of requirements has been agreed upon (and where needed, clarified), thereby \nproviding the testing team with a solid foundation for them to build upon, the testing team \ncan then focus its attention on the test-planning decisions that the team will have to make \nbefore selecting and designing the tests that they intend to execute. These decisions and \nthe rationale for making them are typically recorded in a document referred to as a test \nplan.  \nA test plan could be structured according to an industry standard such as the Institute of \nElectrical and Electronics Engineers (IEEE) Standard for Software Documentation-Std. 829, \nbased on an internal template, or even be pioneering in its layout. What's more important \nthan its specific layout is the process that building a test plan forces the testing team to go \nthrough. Put simply, filling in the blank spaces under the various section headings of the \ntest plan should generate constructive debates within the testing team and with other \ninterested parties. As a result, issues can be brought to the surface early before they \nbecome more costly to fix (measured in terms of additional resources, delayed release, or \nsystem quality). For some testing projects, the layout of the test plan is extremely important. \nFor example, a regulatory agency, insurance underwriter, or mandated corporate policy \nmay require that the test plan be structured in a specific way. For those testing teams that \nare not required to use a particular layout, using an existing organizational template or \nindustry standard (such as the Rational's Unified Process [RUP]) may foster better \n"
        },
        {
            "page_number": 21,
            "text": " \n14 \ninterproject communication. At the same time, the testing team should be permitted to \ncustomize the template or standard to reflect the needs of their specific project and not feel \nobliged to generate superfluous documentation purely because it's suggested in a specific \ntemplate or standard. Craig et al. (2002) and Kaner et al. (2002) both provide additional \nguidance on customizing a test plan to better fit the unique needs of each testing project. \nA test plan can be as large as several hundred pages in length or as simple as a single \npiece of paper (such as the one-page test plan described in Nguyen [2000]). A voluminous \ntest plan can be a double-edged sword. A copiously documented test plan may contain a \ncomprehensive analysis of the system to be tested and be extremely helpful to the testing \nteam in the later stages of the project, but it could also represent the proverbial millstone \nthat is hung around the resource neck of the testing team, consuming ever-increasing \namounts of effort to keep up-to-date with the latest project developments or risk becoming \nobsolete. Contractual and regulatory obligations aside, the testing team should decide at \nwhat level of detail a test plan ceases to be an aid and starts to become a net drag on the \nproject's productivity. \nThe testing team should be willing and able (contractual obligations aside) to modify the \ntest plan in light of newly discovered information (such as the test results of some of the \nearlier scheduled tests), allowing the testing effort to hone in on the areas of the system \nthat this newly discovered information indicates needs more testing. This is especially true if \nthe testing effort is to adopt an iterative approach to testing, where the later iterations won't \nbe planned in any great detail until the results of the earlier iterations are known. \nAs previously mentioned in this section, the content (meat) of a test plan is far more \nimportant that the structure (skeleton) that this information is hung on. The testing team \nshould therefore always consider adapting their test plan(s) to meet the specific needs of \neach project. For example, before developing their initial test plan outline, the testing team \nmay wish to review the test plan templates or checklists described by Kaner et al. (2002), \nNguyen (2000), Perry (2000), Stottlemyer (2001), the Open Source Security Testing \nMethodology (www.osstmm.org), IEEE Std. 829 (www.standards.ieee.org), and the \nRational unified process (www.rational.com). The testing team may then select and \ncustomize an existing template, or embark on constructing a brand-new structure and \nthereby produce the test plan that will best fit the unique needs of the project in hand. \nOne of the most widely referenced software-testing documentation standards to date is that \nof the IEEE Std. 829 (this standard can be downloaded for a fee from \nwww.standards.ieee.org). For this reason, this chapter will discuss the content of a security \ntest plan, in the context of an adapted version of the IEEE Std. 829-1998 (the 1998 version \nis a revision of the original 1983 standard). \n \nIEEE STD. 829-1998 SECTION HEADINGS \nFor reference purposes, the sections that the IEEE 829-1998 standard recommends have \nbeen listed below: \na. Test plan identifier \nb. Introduction \nc. Test items \nd. Features to be tested \ne. Features not to be tested \nf. Approach \ng. Item pass/fail criteria \nh. Suspension criteria and resumption requirements \ni. Test deliverables \n"
        },
        {
            "page_number": 22,
            "text": " \n15 \nj. Testing tasks \nk. Environmental needs \nl. Responsibilities \nm. Staffing and training needs \nn. Schedule \no. Risks and contingencies \np. Approvals \nTest Plan Identifier \nEach test plan and, more importantly, each version of a test plan should be assigned an \nidentifier that is unique within the organization. Assuming the organization already has a \ndocumentation configuration management process (manual or automated) in place, the \nmethod for determining the ID should already have been determined. If such a process has \nyet to be implemented, then it may pay to spend a little time trying to improve this situation \nbefore generating additional documentation (configuration management is discussed in \nmore detail later in this chapter in the section Configuration Management). \nIntroduction \nGiven that test-planning documentation is not normally considered exciting reading, this \nsection may be the only part of the plan that many of the intended readers of the plan \nactually read. If this is likely to be the case, then this section may need to be written in an \nexecutive summary style, providing the casual reader with a clear and concise \nunderstanding of the exact goal of this project and how the testing team intends to meet \nthat goal. Depending upon the anticipated audience, it may be necessary to explain basic \nconcepts such as why security testing is needed or highlight significant items of information \nburied in later sections of the document, such as under whose authority this testing effort is \nbeing initiated. The key consideration when writing this section is to anticipate what the \ntargeted reader wants (and needs) to know. \nProject Scope \nAssuming that a high-level description of the project's testing objectives (or goals) was \nexplicitly defined in the test plan's introduction, this section can be used to restate those \nobjectives in much more detail. For example, the introduction may have stated that security \ntesting will be performed on the wiley.com Web site, whereas in this section, the specific \nhardware and software items that make up the wiley.com Web site may be listed. For \nsmaller Web sites, the difference may be trivial, but for larger sites that have been \nintegrated into an organization's existing enterprise network or that share assets with other \nWeb sites or organizations, the exact edge of the testing project's scope may not be \nobvious and should therefore be documented. Chapter 3 describes some of the techniques \nthat can be used to build an inventory of the devices that need to be tested. These \ntechniques can also precisely define the scope of the testing covered by this test plan. \nIt is often a good idea to list the items that will not be tested by the activities covered by this \ntest plan. This could be because the items will be tested under the auspices of another test \nplan (either planned or previously executed), sufficient resources were unavailable to test \nevery item, or other reasons. Whatever the rationale used to justify a particular item's \nexclusion from a test plan, the justification should be clearly documented as this section is \nlikely to be heavily scrutinized in the event that a future security failure occurs with an item \nthat was for some reason excluded from the testing effort. Perhaps because of this \nconcern, the \"out of scope\" section of a test plan may generate more debate with senior \nmanagement than the \"in scope\" section of the plan. \n"
        },
        {
            "page_number": 23,
            "text": " \n16 \nChange Control Process \nThe scope of a testing effort is often defined very early in the testing project, often when \ncomparatively little is known about the robustness and complexity of the system to be \ntested. Because changing the scope of a project often results in project delays and budget \noverruns, many teams attempt to freeze the scope of the project. However, if during the \ncourse of the testing effort, a situation arises that potentially warrants a change in the \nproject's scope, then many organizations will decide whether or not to accommodate this \nchange based on the recommendation of a change control board (CCB). For example, \ndiscovering halfway through the testing effort that a mirror Web site was planned to go into \nservice next month (but had not yet been built) would raise the question \"who is going to \ntest the mirror site?\" and consequently result in a change request being submitted to the \nCCB. \nWhen applying a CCB-like process to changes in the scope of the security-testing effort in \norder to provide better project control, the members of a security-testing CCB should bear \nin mind that unlike the typical end user, an attacker is not bound by a project's scope or the \ndecisions of a CCB. This requires them to perhaps be a little more flexible than they would \nnormally be when faced with a nonsecurity orientation change request. After all, the testing \nproject will most likely be considered a failure if an intruder is able to compromise a system \nusing a route that had not been tested, just because it had been deemed to have been \nconsidered out of scope by the CCB. \nA variation of the CCB change control process implementation is to break the projects up \ninto small increments so that modifying the scope for the increment currently being tested \nbecomes unnecessary because the change request can be included in the next scheduled \nincrement. The role of the CCB is effectively performed by the group responsible for \ndetermining the content of future increments. \n \nTHE ROLE OF THE CCB \nThe CCB (also sometimes known as a configuration control board) is the group of \nindividuals responsible for evaluating and deciding whether or not a requested change \nshould be permitted and subsequently ensuring that any approved changes are \nimplemented appropriately. \nIn some organizations, the CCB may be made up of a group of people drawn from different \nproject roles, such as the product manager, project sponsor, system owner, internal \nsecurity testers, local area network (LAN) administrators, and external consultants, and \nhave elaborate approval processes. In other organizations, the role of the CCB may be \nperformed by a single individual such as the project leader who simply gives a nod to the \nrequest. Regardless of who performs this role, the authority to change the scope of the \ntesting effort should be documented in the test plan. \nFeatures to Be Tested \nA system's security is only as strong as its weakest link. Although this may be an obvious \nstatement, it's surprising how frequently a security-testing effort is directed to only test some \nand not all of the following features of a Web site: \nƒ \nNetwork security (covered in Chapter 3) \nƒ \nSystem software security (covered in Chapter 4) \nƒ \nClient-side application security (covered in Chapter 5) \nƒ \nClient-side to server-side application communication security (covered in Chapter 5) \nƒ \nServer-side application security (covered in Chapter 6) \n"
        },
        {
            "page_number": 24,
            "text": " \n17 \nƒ \nSocial engineering (covered in Chapter 7) \nƒ \nDumpster diving (covered in Chapter 7) \nƒ \nInside accomplices (covered in Chapter 7) \nƒ \nPhysical security (covered in Chapter 7) \nƒ \nMother nature (covered in Chapter 7) \nƒ \nSabotage (covered in Chapter 7) \nƒ \nIntruder confusion (covered in Chapter 8) \nƒ \nIntrusion detection (covered in Chapter 8) \nƒ \nIntrusion response (covered in Chapter 8) \nBefore embarking on an extended research and planning phase that encompasses every \nfeature of security testing, the security-testing team should take a reality check. Just how \nlikely is it that they have the sufficient time and funding to test everything? Most likely the \nsecurity-testing team will not have all the resources they would like, in which case choices \nmust be made to decide which areas of the system will be drilled and which areas will \nreceive comparatively light testing. Ideally, this selection process should be systematic and \nimpartial in nature. A common way of achieving this is through the use of a risk analysis \n(the subject of Chapter 10), the outcome of which should be a set of candidate tests that \nhave been prioritized so that the tests that are anticipated to provide the greatest benefit \nare scheduled first and the ones that provide more marginal assistance are executed last (if \nat all). \nFeatures Not to Be Tested \nIf the testing effort is to be spread across multiple test plans, there is a significant risk that \nsome tests may drop through the proverbial cracks in the floor, because the respective \nscopes of the test plans do not dovetail together perfectly. A potentially much more \ndangerous situation is the scenario of an entire feature of the system going completely \nuntested because everyone in the organization thought someone else was responsible for \ntesting this facet of the system. \nTherefore, it is a good practice to not only document what items will be tested by a specific \ntest plan, but also what features of these items will be tested and what features will fall \noutside the scope of this test plan, thereby making it explicitly clear what is and is not \ncovered by the scope of an individual test plan. \nApproach \nThis section of the test plan is normally used to describe the strategy that will be used by \nthe testing team to meet the test objectives that have been previously defined. It's not \nnecessary to get into the nitty-gritty of every test strategy decision, but the major decisions \nsuch as what levels of testing (described later in this section) will be performed and when \n(or how frequently) in the system's life cycle the testing will be performed should be \ndetermined. \nLevels of Testing \nMany security tests can be conducted without having to recreate an entire replica of the \nsystem under test. The consequence of this mutual dependency (or lack of) on other \ncomponents being completed impacts when and how some tests can be run. \nOne strategy for grouping tests into multiple testing phases (or levels) is to divide up the \ntests based on how complete the system must be before the test can be run. Tests that can \nbe executed on a single component of the system are typically referred to as unit- or \nmodule-level tests, tests that are designed to test the communication between two or more \n"
        },
        {
            "page_number": 25,
            "text": " \n18 \ncomponents of the system are often referred to as integration-, string- or link-level tests, \nand finally those that would benefit from being executed in a full replica of the system are \noften called system-level tests. For example, checking that a server has had the latest \nsecurity patch applied to its operating system can be performed in isolation and can be \nconsidered a unit-level test. Testing for the potential existence of a buffer overflow occurring \nin any of the server-side components of a Web application (possibly as a result of a \nmalicious user entering an abnormally large string via the Web application's front-end) \nwould be considered an integration- or system-level test depending upon how much of the \nsystem needed to be in place for the test to be executed and for the testing team to have a \nhigh degree of confidence in the ensuing test results. \nOne of the advantages of unit-level testing is that it can be conducted much earlier in a \nsystem's development life cycle since the testing is not dependent upon the completion or \ninstallation of any other component. Because of the fact that the earlier that a defect is \ndetected, the easier (and therefore more cheaply) it can be fixed, an obvious advantage \nexists to executing as many tests as possible at the unit level instead of postponing these \ntests until system-level testing is conducted, which because of its inherent dependencies \ntypically must occur later in the development life cycle. \nUnfortunately, many organizations do not conduct as many security tests at the unit level as \nthey could. The reasons for this are many and vary from organization to organization. \nHowever, one recurring theme that is cited in nearly every organization where unit testing is \nunderutilized is that the people who are best situated to conduct this level of testing are \noften unaware of what should be tested and how to best accomplish this task. Although the \nhow is often resolved through education (instructor-led training, books, mentoring, and so \non), the what can to a large part be addressed by documenting the security tests that need \nto be performed in a unit-level checklist or more formally in a unit-level test plan-a step that \nis particularly important if the people who will be conducting these unit-level tests are not \nmembers of the team responsible for identifying all of the security tests that need to be \nperformed. \nDividing tests up into phases based upon component dependencies is just one way a \ntesting team may strategize their testing effort. Alternative or complementary strategies \ninclude breaking the testing objectives up into small increments, basing the priority and type \nof tests in later increments on information gleaned from running earlier tests (an heuristic or \nexploratory approach), and grouping the tests based on who would actually do the testing, \nwhether it be developers, outsourced testing firms, or end users. The large variety of \npossible testing strategies in part explains the proliferation of testing level names that are in \npractice today, such as unit, integration, build, alpha, beta, system, acceptance, staging, \nand post-implementation to name but a few. Black (2003), Craig et al. (2002), Kaner et al. \n(2001), Gerrard et al. (2002), and Perry (2000) provide additional information on the various \nalternate testing strategies that could be employed by a testing team. \nFor some projects, it may make more sense to combine two (or more) levels of testing into \na single test plan. The situation that usually prompts this test plan cohabitation is when the \ntesting levels have a great deal in common. For example, on one project, the set of unit-\nlevel tests might be grouped with the set of integration-level tests because the people who \nwill be conducting the tests are the same, both sets of tests are scheduled to occur at \napproximately the same time, or the testing environments are almost identical. \nRelying on only a single level of testing to capture all of a system's security defects is likely \nto be less efficient than segregating the tests into two (or more) levels; it may quite possibly \nincrease the probability that security holes will be missed. This is one of the reasons why \nmany organizations choose to utilize two or more levels of testing. \n"
        },
        {
            "page_number": 26,
            "text": " \n19 \nWhen to Test \nFor many in the software industry, testing is the activity that happens somewhere in the \nsoftware development life cycle between coding and going live. Security tests are often \nsome of the very last tests to be executed. This view might be an accurate observation of \nyesterday's system development, when development cycles were measured in years and \nthe tasks that the system was being developed to perform were well understood and rarely \nchanged, but it most certainly should not be the case today. \nIn today's world of ever-changing business requirements, rapid application development, \nand extreme programming, testing should occur throughout the software development life \ncycle (SDLC) rather than as a single-step activity that occurs toward the end of the process, \nwhen all too often too little time (or budget) is left to adequately test the product or fix a \nmajor flaw in the system. \nWhen to Retest \nAlthough many foresighted project managers have scheduled testing activities to occur \nearly in the development cycle, it is less likely that as much thought will be given to planning \nthe continuous testing that will be needed once the system goes live. Even if the functional \nrequirements of a system remain unchanged, a system that was deemed secure last week \nmay become insecure next week. The following are just a few examples of why this could \nhappen: \nƒ \nA previously unknown exploit in an operating system used by the system becomes \nknown to the attacker community. \nƒ \nAdditional devices (firewalls, servers, routers, and so on) are added to the system to \nenable it to meet higher usage demands. Unfortunately, these newly added devices \nmay not have been configured in exactly the same way as the existing devices. \nƒ \nA service pack installed to patch a recently discovered security hole also resets other \nconfiguration settings back to their default values. \nƒ \nDue to the large number of false alarms, the on-duty security personnel have \nbecome desensitized to intruder alerts and subsequently do not respond to any \nautomated security warnings. \nƒ \nUser-defined passwords that expire after a period of time and were originally long \nand cryptic have become short, easy to remember, and recycled. \nƒ \nLog files have grown to the point that no free disk space is left, thereby inhibiting the \ncapability of an intruder detection system to detect an attack. \nSecurity testing should not be regarded as a one-time event, but rather as a recurring \nactivity that will be ongoing as long as the system remains active. The frequency with which \nthe retests occur will to a large part be driven by the availability of resources to conduct the \ntests (cost) and the degree to which the system changes over time. Some events may, \nhowever, warrant an immediate (if limited in scope) retest. For example, the organization \nmay decide to upgrade the operating system used by a number of the servers on the Web \nsite, or a firewall vendor releases a \"hot fix\" for its product. \nWhat to Retest \nAs a starting point, the testing team should consider each test that was utilized during the \nsystem's initial testing effort as a potential candidate for inclusion into a future set of tests \nthat will be reexecuted on a regular basis after the system goes into production (sometimes \nreferred to as a postdeployment regression test set) to ensure that vulnerabilities that were \nsupposedly fixed (or never existed) do not subsequently appear. \n"
        },
        {
            "page_number": 27,
            "text": " \n20 \nFor tests that have been automated, there may be very little overhead in keeping these \ntests as part of a regression test set, especially if the automated test script is being \nmaintained by another organization at no additional cost, which may well be the case for a \nsecurity assessment tool (such as those listed in Table 9.4) that an organization has a \nmaintenance agreement for, or is available free of charge. \n \nTHE REGRESSION TEST SET \nRegression tests are usually intended to be executed many times and are designed to \nconfirm that previously identified defects have been fixed and stay fixed, that functionality \nthat should not have changed has indeed remained unaffected by any other changes to the \nsystem, or both. \n \nWith regard to manual tests, the determination as to whether or not to repeat a test will to a \nlarge part depend upon how problems previously detected by the test were fixed (and \nconsequently what the likelihood is that the problem will reappear). For example, if the \ntesting team had originally found that weak passwords were being used and the solution \nwas to send an email telling everyone to clean up their act, then chances are within a \ncouple of userID/password cycles, weak (easy to remember) passwords will again start to \nshow up, necessitating the testing team to be ever vigilant for this potential vulnerability. If, \non the other hand, a single user-sign-on system was implemented with tough password \nrequirements, then the same issue is not likely to occur again and therefore may not \nwarrant the original tests being included in future regression tests. \nPass/Fail Criteria \nA standard testing practice is to document the expected or desired results of an individual \ntest case prior to actually executing the test. As a result, a conscious (or subconscious) \ntemptation to modify the pass criteria for a test based on its now known result is avoided. \nUnfortunately, determining whether security is good enough is a very subjective measure-\none that is best left to the project's sponsor (or the surrogate) rather than the testing team. \nMaking a system more secure all too often means making the system perform more slowly, \nbe less user-friendly, harder to maintain, or more costly to implement. Therefore, unlike \ntraditional functional requirements, where the theoretical goal is absolute functional \ncorrectness, an organization may not want its system to be as secure as it could be \nbecause of the detrimental impact that such a secure implementation would have on \nanother aspect of the system. For example, suppose a Web site requires perspective new \nclients to go through an elaborate client authentication process the first time they register \nwith the Web site. (It might even involve mailing user IDs and first-time passwords \nseparately through the postal service.) Such a requirement might reduce the number of \nfraudulent instances, but it also might have a far more drastic business impact on the \nnumber of new clients willing to go through this process, especially if a competitor Web site \noffers a far more user-friendly (but potentially less secure) process. The net result is that \nthe right amount of security for each system is subjective and will vary from system to \nsystem and from organization to organization. \nInstead of trying to make this subjective call, the testing team might be better advised to \nconcentrate on how to present the findings of their testing effort to the individual(s) \nresponsible for making this decision. For example, presenting the commissioner of a \nsecurity assessment with the raw output of an automated security assessment tool that had \nperformed several hundred checks and found a dozen irregularities is probably not as \n"
        },
        {
            "page_number": 28,
            "text": " \n21 \nhelpful as a handcrafted report that lists the security vulnerabilities detected (or suspected) \nand their potential consequences if the system goes into service (or remains in service) as \nis. \nIf an organization's testing methodology mandates that a pass/fail criteria be specified for a \nsecurity-testing test effort, it may be more appropriate for the test plan to use a criteria such \nas the following: \"The IS Director will retain the decision as to whether the total and/or \ncriticality of any or all detected vulnerabilities warrant the rework and/or retesting of the \nWeb site.\" This is more useful than using a dubious pass criteria such as the following: \"95 \npercent of the test cases must pass before the system can be deemed to have passed \ntesting.\" \nSuspension Criteria and Resumption Requirements \nThis section of the test plan may be used to identify the circumstances under which it would \nbe prudent to suspend the entire testing effort (or just portions of it) and what requirements \nmust subsequently be met in order to reinitiate the suspended activities. For example, \nrunning a penetration test would not be advisable just before the operating systems on the \nmajority of the Web site's servers are scheduled to be upgraded with the latest service \npack. Instead, testing these items would be more effective if it was suspended until after the \nservers have been upgraded and reconfigured. \nTest Deliverables \nEach of the deliverables that the testing team generates as a result of the security-testing \neffort should be documented in the test plan. The variety and content of these deliverables \nwill vary from project to project and to a large extent depend on whether the documents \nthemselves are a by-product or an end product of the testing effort. \nAs part of its contractual obligations, a company specializing in security testing may need to \nprovide a client with detailed accounts of all the penetration tests that were attempted \n(regardless of their success) against the client's Web site. For example, the specific layout \nof the test log may have been specified as part of the statement of work that the testing \ncompany proposed to the client while bidding for the job. In this case, the test log is an end \nproduct and will need to be diligently (and time-consumingly) populated by the penetration-\ntesting team or they risk not being paid in full for their work. \nIn comparison, a team of in-house testers trying to find a vulnerability in a Web application's \nuser login procedure may use a screen-capture utility to record their test execution. In the \nevent that a suspected defect is found, the tool could be used to play back the sequence of \nevents that led up to the point of failure, thereby assisting the tester with filling out an \nincident or defect report. Once the report has been completed, the test execution recording \ncould be attached to the defect (providing further assistance to the employee assigned to fix \nthis defect) or be simply discarded along with all the recordings of test executions that didn't \nfind anything unusual. In this case, the test log was produced as a by-product of the testing \neffort and improved the project's productivity. \nBefore a testing team commits to producing any deliverable, it should consider which \ndeliverables will assist them in managing and executing the testing effort and which ones \nare likely to increase their documentation burden. It's not unheard of for testing teams who \nneed to comply with some contractual documentary obligation to write up test designs and \ncreatively populate test logs well after test execution has been completed. \nThe following sections provide brief overviews of some of the more common deliverables \ncreated by testing teams. Their relationships are depicted in Figure 2.2. \n"
        },
        {
            "page_number": 29,
            "text": " \n22 \n \n \n \nFigure 2.2: Testing deliverables.  \nTest Log \nThe test log is intended to record the events that occur during test execution in a \nchronological order. The log can take the form of shorthand notes on the back of an \nenvelope, a central database repository manually populated via a graphical user interface \n(GUI) front end, or a bitmap screen-capture utility unobtrusively running in the background \ntaking screen snapshots every few seconds. Appendix C contains a sample layout for a test \nlog. \nTest Incident Report \nAn incident is something that happens during the course of test execution that merits further \ninvestigation. The incident may be an observable symptom of a defect in the item being \ntested, unexpected but acceptable behavior, a defect in the test itself, or an incident that is \nso trivial in nature or impossible to recreate that its exact cause is never diagnosed. \nThe test incident report is a critical project communication document because the majority \nof incidents are likely to be investigated by someone other than the person who initially \nobserved (and presumable wrote up) the incident report. For this reason, it is important to \nuse a clear and consistent report format and an agreement be reached between \nrepresentatives of those who are likely to encounter and report the incidents and those who \nare likely to be charged with investigating them. Appendix C contains an example layout of \na test incident report. \nAlthough the exact fields used on the report may vary from project to project, depending on \nlocal needs, conceptually the report needs to do one thing: accurately document the \nincident that has just occurred in such a way that someone else is able to understand what \nhappened, thereby enabling the reader to thoroughly investigate the incident (typically by \ntrying to reproduce it) and determine the exact cause of the event. Craig et al. (2002) and \nKaner et al. (1999) provide additional information on the content and use of test incident \nreports. \nDefect-Tracking Reports \nA defect differs from an incident in that a defect is an actual flaw (or bug) in the system, \nwhereas an incident is just an indicator that a defect may exist. At the moment an incident \nis initially recorded, it is typically not clear whether this incident is the result of a defect or \nsome other cause (such as a human error in the testing). Therefore, it's a common practice \nto include incident reports that have not yet been investigated along with identified defects \n"
        },
        {
            "page_number": 30,
            "text": " \n23 \nin a single defect-tracking report. In effect, a guilty-until-proven-innocent mentality is applied \nto the incidents. \nThe question of who should be assigned ownership of a defect (who is responsible for \nmaking sure it is resolved) may be a politically charged issue. Often the testing team has \nthe task of acting as the custodian of all known incident and defect reports. To make \nmanaging these reports easier, most testing teams utilize an automated tool such as one of \nthose listed in Table 2.1. \n \nTable 2.1: Sample Defect-Tracking Tools  \nNAME  \nASSOCIATED WEB SITE  \nBug Cracker \nwww.fesoft.com  \nBugbase \nwww.threerock.com  \nBugCollector \nwww.nesbitt.com  \nBuggy \nwww.novosys.de  \nBUGtrack \nwww.skyeytech.com  \nBugLink \nwww.pandawave.com  \nBugzilla \nwww.mozilla.org  \nClearQuest \nwww.rational.com  \nD-Tracker \nwww.empirix.com  \nElementool \nwww.elementool.com  \nPT BugTracker \nwww.avensoft.com  \nRazor \nwww.visible.com  \nSWBTracker \nwww.softwarewithbrains.com  \nTeam Remote Debugger \nwww.remotedebugger.com  \nTeam Tracker \nwww.hstech.com.au  \nTestDirector \nwww.mercuryinteractive.com  \nTestTrack Pro \nwww.seapine.com  \nVisual Intercept \nwww.elsitech.com  \nZeroDefect \nwww.prostyle.com  \nUsing a commercial defect-tracking tool (such as one of the tools listed in Table 2.1) or an \nin-house-developed tool typically enables the testing team to automatically produce all sorts \nof defect-tracking reports. Examples include project status reports showing the status of \nevery reported incident/defect; progress reports showing the number of defects that have \nbeen found, newly assigned, or fixed since the last progress report was produced; agendas \nfor the next project meeting where defect-fixing priorities will be assessed; and many \ninteresting defect statistics. \nJust because a tool can produce a particular report does not necessarily mean that the \nreport will be useful to distribute (via paper or email). Too many reports produced too \n"
        },
        {
            "page_number": 31,
            "text": " \n24 \nfrequently can often generate so much paper that the reports that are truly useful get lost in \nthe paper shredder. Therefore, a testing team should consider what reports are actually \ngoing to be useful to the team itself and/or useful methods of communication to individuals \noutside of the testing team, and then document in this section of the test plan which reports \nthey initially intend to produce. If the needs of the project change midway through the \ntesting effort, requiring new reports, modifications to existing ones, or the retirement of \nunneeded ones, then the test plan should be updated to reflect the use of this new set of \nreports. \nMetrics \nSome defect metrics are so easy to collect that it's almost impossible to avoid publishing \nthem. For example, the metrics of the number of new incidents found this week, the mean \nnumber of defects found per tested Web page, or the defects found per testing hour are \neasily found and gathered. The problem with statistics is that they can sometimes cause \nmore problems than they solve. For instance, if 15 new bugs were found this week, 25 last \nweek, and 40 the previous week, would senior management then determine that based on \nthese statistics that the system being tested was nearly ready for release? The reality could \nbe quite different. If test execution was prioritized so that the critical tests were run first, \nmoderately important tests were run next, and the low-priority tests were run last, then with \nthis additional information, it would be revealed that the unimportant stuff works relatively \nwell compared to the system's critical components. This situation is hardly as desirable as \nmight have been interpreted from the numbers at first glance. \nBefore cracking out metric after metric just because a tool produces it or because it just \nseems like the right thing to do, the testing team should first consider the value of these \nmetrics to this project or to future projects. Moller et al. (1993) cites six general uses for \nsoftware metrics: (1) goal setting, (2) improving quality, (3) improving productivity, (4) \nproject planning, (5) managing, or (6) improving customer confidence. More specific goals \nmay include identifying training needs, measuring test effectiveness, or pinpointing \nparticularly error-prone areas of the system. If a proposed metric is not a measure that can \nbe directly used to support one or more of these uses, then it runs the risk of being \nirrelevant or, worse still, misinterpreted. Black (2002, 2003), Craig et al. (2002), and Kan \n(1995) provide additional information on collecting and reporting software-testing metrics. \nThis section of the test plan can be used to document what metrics will be collected and \n(optionally) published during the testing effort. Note that some metrics associated with \nprocess improvement may not be analyzed until after this specific project has been \ncompleted. The results are then used to improve future projects. \nTest Summary Report \nThe test summary report-also sometimes known as a test analysis report-enables the \ntesting team to summarize all of its findings. The report typically contains information such \nas the following: \nƒ \nA summary of the testing activities that actually took place. This may vary from the \noriginally planned activities due to changes such as a reduction (or expansion) of \ntesting resources, an altered testing scope, or discoveries that were made during \ntesting (this will be especially true if extensive heuristic or exploratory testing was \nutilized). \nƒ \nA comprehensive list of all of the defects and limitations that were found (sometimes \neuphemistically referred to as a features list). The list may also include all the \nsignificant incidents that could still not be explained after investigation. \n"
        },
        {
            "page_number": 32,
            "text": " \n25 \nƒ \nHigh-level project control information such as the number of hours and/or elapsed \ntime expended on the testing effort, capital expenditure on the test environment, and \nany variance from the budget that was originally approved. \nƒ \nOptionally, an assessment of the accumulative severity of all the known defects and \npossibly an estimation of the number and severity of the defects that may still be \nlurking in the system undetected. \nƒ \nFinally, some test summary reports also include a recommendation as to whether or \nnot the system is in a good enough state to be placed into or remain in production. \nAlthough the ultimate decision for approval should reside with the system's owner or \nsurrogate, often the testing team has the most intimate knowledge of the strengths \nand weaknesses of the system. Therefore, the team is perhaps the best group to \nmake an objective assessment of just how good, or bad, the system actually is. \nThe completion of the test summary report is often on a project plan's critical path. \nTherefore, the testing team may want to build this document in parallel with test execution \nrather than writing it at the end of the execution phase. Depending upon how long test \nexecution is expected to take, it may be helpful to those charged with fixing the system's \nvulnerabilities to see beta versions of the report prior to its final publication. These beta \nversions often take the form of a weekly (or daily) status report, with the test summary \nreport ultimately being the very last status report. Appendix C contains an example layout of \na test summary report. \nEnvironmental Needs \nA test environment is a prerequisite if the security-testing team wants to be proactive and \nattempt to catch security defects before they are deployed in a production environment. In \naddition, tests can be devised and executed without worrying about whether or not \nexecuting the tests might inadvertently have an adverse effect on the system being tested, \nsuch as crashing a critical program. Indeed, some tests may be specifically designed to try \nand bring down the target system (a technique sometimes referred to as destructive \ntesting). For example, a test that tried to emulate a denial-of-service (DoS) attack would be \nmuch safer to evaluate in a controlled test environment, than against a production system \n(even if in theory the production system had safeguards in place that should protect it \nagainst such an attack). \nIt would certainly be convenient if the testing team had a dedicated test lab that was an \nexact full-scale replica of the production environment, which they could use for testing. \nUnfortunately, usually as a result of budgetary constraints, the test environment is often not \nquite the same as the production environment it is meant to duplicate (in an extreme \nsituation it could solely consist of an individual desktop PC). For example, instead of using \nfour servers (as in the production environment) dedicated to running each of the following \ncomponents-Web server, proxy server, application server, and database server-the test \nenvironment may consist of only one machine, which regrettably cannot be simultaneously \nconfigured four different ways. Even if a test environment can be created with an equivalent \nnumber of network devices, some of the devices used in the test lab may be cheap \nimitations of the products actually used in the production environment and therefore behave \nslightly differently. For example, a $100 firewall might be used for a test instead of the \n$50,000 one used in production. \nIf the test environment is not expected to be an exact replica of the production environment, \nconsideration should be given to which tests will need to be rerun on the production system, \nas running them on the imperfect test environment without incident will not guarantee the \nsame results for the production environment. A second consideration is that the test \nenvironment could be too perfect. For example, if the implementation process involves any \n"
        },
        {
            "page_number": 33,
            "text": " \n26 \nsteps that are prone to human error, then just because the proxy server in the test lab has \nbeen configured to implement every security policy correctly does not mean that the \nproduction version has also been implemented correctly. \nIn all probability, some critical site infrastructure security tests will need to be rerun on the \nproduction environment (such as checking the strength of system administrators' \npasswords, or the correct implementation of a set of firewall rules). If the Web site being \ntested is brand-new, this extra step should not pose a problem because these tests can be \nrun on the production environment prior to the site going live. For a Web site that has \nalready gone live, the security-testing team must develop some rules of engagement (terms \nof reference) that specify when and how the site may be prodded and probed, especially if \nthe site was previously undertested or not tested at all. These rules serve as a means of \neliminating false intruder alarms, avoiding accidental service outages during peak site \nusage, and inadvertently ignoring legitimate intruder alarms (because it was thought that \nthe alarm was triggered by the security-testing team and not a real intruder). \nConfiguration Management \nConfiguration management is the process of identifying (what), controlling (library \nmanagement), tracking (who and when), and reporting (who needs to know) the \ncomponents of a system at discrete points in time for the primary purpose of maintaining \nthe integrity of the system. A good configuration management process is not so obtrusive, \nbureaucratic, or all encompassing that it has a net negative effect on development or \ntesting productivity, but rather speeds such activities. \nDeveloping, testing, or maintaining anything but the most simplistic of Web sites is virtually \nimpossible without implementing some kind of configuration management process and \nshould therefore be a prerequisite for any significant testing effort, and consequently be \naddressed by the associated test plan. For example, a Webmaster may choose to install an \noperating system service pack midway through a penetration test, or a developer may \ndirectly perform a quick fix on the Web application in production. Both decisions can cause \na great deal of confusion and consequently prove to be quite costly to straighten out. \nMany organizations have become accustomed to using a configuration management tool to \nhelp them manage the ever-increasing number of software components that need to be \ncombined in order to build a fully functional Web application. A typical source code \npromotion process would require all developers to check in and check out application \nsource code from a central development library. At regular intervals, the configuration \nmanager (the build manager or librarian) baselines the application and then builds and \npromotes the application into a system-testing environment for the testing team to evaluate. \nIf all goes well, the application is finally promoted into the production environment. Note that \nsome organizations use an additional staging area between the system-testing and \nproduction environments, which is a particularly useful extra step if for any reason the \nsystem test environment is not an exact match to the production environment. Table 2.2 \nlists some sample configuration management tools that could be used to assist with this \nprocess. \n \nTable 2.2: Sample Configuration Management Tools  \nTOOL  \nASSOCIATED WEB SITE  \nChangeMan \nwww.serena.com  \nClearCase \nwww.rational.com  \n"
        },
        {
            "page_number": 34,
            "text": " \n27 \nTable 2.2: Sample Configuration Management Tools  \nTOOL  \nASSOCIATED WEB SITE  \nCM Synergy \nwww.telelogic.com  \nEndevor/Harvest \nwww.ca.com  \nPVCS \nwww.merant.com  \nRazor \nwww.visible.com  \nSource Integrity \nwww.mks.com  \nStarTeam \nwww.starbase.com  \nTRUEchange \nwww.mccabe.com  \nVisual SourceSafe \nwww.microsoft.com  \nA second area of software that is a candidate for configuration management (potentially \nusing the same tools to manage the application's source code) would be the test scripts and \ntest data used to run automated tests (sometimes collectively referred to as testware). This \nis a situation that becomes increasingly important as the size and scope of any test sets \ngrow. \nIt is less common to see a similar configuration management process being applied to \nsystem software installation and configuration options. For example, the current set of \nservers may have been thoroughly tested to ensure that they are all configured correctly \nand have had all the relevant security patches applied, but when new servers are added to \nthe system to increase the system's capacity or existing servers are reformatted to fix \ncorrupted files, the new system software installations are not exactly the same as the \nconfiguration that was previously tested; it can be something as simple as two security \npatches being applied in a different order or the installer forgetting to uncheck the install \nsample files option during the install. \nRather than relying on a manual process to install system software, many organizations \nnow choose to implement a configuration management process for system software by \nusing some sort of disk replication tool. (Table 2.3 lists some sample software- and \nhardware-based disk replication tools.) The process works something like the following: a \nmaster image is first made of a thoroughly tested system software install. Then each time a \nnew installation is required (for a new machine or to replace a corrupted version), the \nreplication tool copies the master image onto the target machine, reducing the potential for \nhuman error. \n \nTable 2.3: Sample Disk Replication Tools  \nTOOL  \nASSOCIATED WEB SITE  \nDrive2Drive \nwww.highergroundsoftware.com  \nGhost \nwww.symantec.com  \nImageCast \nwww.storagesoftsolutions.com  \nImage MASSter/ImageCast \nwww.ics-iq.com  \nLabExpert/RapiDeploy \nwww.altiris.com  \n"
        },
        {
            "page_number": 35,
            "text": " \n28 \nTable 2.3: Sample Disk Replication Tools  \nTOOL  \nASSOCIATED WEB SITE  \nOmniClone \nwww.logicube.com  \nPC Relocator \nwww.alohabob.com  \nOne undesirable drawback of disk replication tools is their dependence on the target \nmachine having the same hardware configuration as the master machine. This would be no \nproblem if all the servers were bought at the same time from the same vendor. However, it \nwould be problematic if they were acquired over a period of time and therefore use different \ndevice drivers to communicate to the machine's various hardware components. \nUnfortunately, due to the lack of automated tools for some platforms, some devices still \nneed to be configured manually-for example, when modifying the default network traffic \nfiltering rules for a firewall appliance. Given the potential for human error, it's imperative that \nall manual installation procedures be well documented and the installation process be \nregularly checked to ensure that these human-error-prone manual steps are followed \nclosely. \nBefore embarking on any significant testing effort, the security-testing team should confirm \ntwo things: that all of the Web site and application components that are to be tested are \nunder some form of configuration management process (manual or automated) and that \nunder normal circumstances these configurations will not be changed while test execution \nis taking place. Common exceptions to this ideal scenario include fixing defects that actually \ninhibit further testing and plugging newly discovered holes that present a serious risk to the \nsecurity of the production system. \nIf the items to be tested are not under any form of configuration management process, then \nthe security-testing team should not only try to hasten the demise of this undesirable \nsituation, but they should also budget additional time to handle any delays or setbacks \ncaused by an unstable target. Also, where possible, the team should try to schedule the \ntesting effort in a way that minimizes the probability that a serious defect will make its way \ninto production, especially one that results from a change being made to part of the system \nthat had already been tested and wasn't retested. \nBrown et al. (1999), Dart (2000), Haug et al. (1999), Leon (2000), Lyon (2000), and White \n(2000) all provide a good starting point for those attempting to define and implement a new \nconfiguration management process. \nResponsibilities \nWho will be responsible for making sure all the key testing activities take place on \nschedule? This list of activities may also include tasks that are not directly part of the testing \neffort, but that the testing team depends upon being completed in a timely manner. For \ninstance, who is responsible for acquiring the office space that will be used to house the \nadditional testers called for by the test plan? Or, if hardware procurement is handled \ncentrally, who is responsible for purchasing and delivering the machines that will be needed \nto build the test lab? \nIdeally, an escalation process should also be mapped out, so that in the event that \nsomeone doesn't fulfill their obligations to support the testing team for whatever reason, the \nsituation gets escalated up the management chain until it is resolved (hopefully as quick \nand painless as possible). \n"
        },
        {
            "page_number": 36,
            "text": " \n29 \nAnother decision that needs to be made is how to reference those responsible for these key \nactivities. Any of the reference methods in Table 2.4 would work. \n \nTable 2.4: Options for Referencing Key Personnel  \nKEY PERSON \nREFERENCED BY \n...  \nEXAMPLE  \nCompany level \nThe xyz company is responsible for conducting a physical \nsecurity assessment of the mirror Web site hosted at the \nLondon facility. \nDepartment level \nNetwork support is responsible for installing and configuring \nthe test environment. \nJob/role title \nThe application database administrator (DBA) is responsible \nfor ensuring that the database schema and security settings \ncreated in the test environment are identical to those that will \nbe used in the production environment. \nIndividual \nJohnny Goodspeed is responsible for testing the security of \nall application communications. \nWhen listing these activities, the testing team will need to decide on how granular this list of \nthings to do should be. The more granular the tasks, the greater the accountability, but also \nthe greater the effort needed to draft the test plan and subsequently keep it up to date. The \ntest plan also runs the risk of needlessly duplicating the who information contained in the \nassociated test schedule (described later in this chapter in the section Schedule). \nStaffing and Training Needs \nIf outside experts are used to conduct penetration testing (covered in more detail in Chapter \n9), is it cost effective for the internal staff to first conduct their own security tests? If the \noutside experts are a scarce commodity-and thus correspondingly expensive or hard to \nschedule-then it may make sense for the less experienced internal staff to first run the \neasy-to-execute security assessment tests; costly experts should only be brought in after all \nthe obvious flaws have been fixed. In effect, the in-house staff would be used to run a set of \ncomparatively cheap entry-criteria tests (also sometimes referred to as smoke tests) that \nmust pass before more expensive, thorough testing is performed. \nOne consideration that will have a huge impact on the effectiveness of any internally staffed \nsecurity-testing effort is the choice of who will actually do the testing. The dilemma that \nmany organizations face is that their security-testing needs are sporadic. Often extensive \nsecurity-oriented testing is not needed for several months and then suddenly a team of \nseveral testers is needed for a few weeks. In such an environment, an organization is going \nto be hard pressed to justify hiring security gurus and equally challenged to retain them. An \nalternative to maintaining a permanent team of security testers is to appropriate employees \nfrom other areas such as network engineers, business users, Web masters, and \ndevelopers. Unfortunately, many of these candidates may not be familiar with generic \ntesting practices, let alone security-specific considerations. This could result in a longer-\nlasting testing effort and less dependable test results. \nFor organizations that maintain a permanent staff of functional testers, Q/A analysts, test \nengineers, or other similarly skilled employees, one possible solution is to train these \n"
        },
        {
            "page_number": 37,
            "text": " \n30 \nindividuals in the basics of security testing and use them to form the core of a temporary \nsecurity-testing team. Such a team would, in all probability, still need to draw upon the skills \nof other employees such as the firewall administrators and DBAs in order to conduct the \nsecurity testing. But having such a team conduct many of the security tests that need to be \nperformed may be more cost effective than outsourcing the entire testing task to an outside \nconsulting firm. This would be especially beneficial for the set of tests that are expected to \nbe regularly rerun after the system goes into production. \nThe degree to which some (or all) of the security-testing effort can be handled in house to a \nlarge extent depends on the steepness of the learning curve that the organization's \nemployees will face. One way to reduce this learning curve is to make use of the ever-\ngrowing supply of security-testing tools. The decision on how much of the testing should be \ndone in house and what tools should be acquired will therefore have a major impact on how \nthe security-testing effort is structured. These topics are expanded on further in Chapter 9. \nSchedule \nUnless the testing effort is trivial in size, the actual details of the test schedule are probably \nbest documented in a separate deliverable and generated with the assistance of a project-\nscheduling tool such as one of those listed in Table 2.5. \n \nTable 2.5: Sample Project-Scheduling Tools  \nNAME  \nASSOCIATED WEB SITE  \nEasy Schedule Maker \nwww.patrena.com  \nFastTrack Schedule \nwww.aecsoft.com  \nGigaPlan.net \nwww.gigaplan.com  \nManagePro \nwww.performancesolutionstech.com  \nMicrosoft Project \nwww.microsoft.com  \nNiku \nwww.niku.com  \nOpenAir \nwww.openair.com  \nPlanView \nwww.planview.com  \nProj-Net \nwww.rationalconcepts.com  \nProjectKickStart \nwww.projectkickstart.com  \nProject Dashboard \nwww.itgroupusa.com  \nProject Office \nwww.pacificedge.com  \nTime Disciple \nwww.timedisciple.com  \nVarious \nwww.primavera.com  \nXcolla \nwww.axista.com  \nWith the scheduling details documented elsewhere, this section of the test plan can be \nused to highlight significant scheduling dates such as the planned start and end of the \ntesting effort and the expected dates when any intermediate milestones are expected to be \nreached. \n"
        },
        {
            "page_number": 38,
            "text": " \n31 \nSince many testing projects are themselves subprojects of larger projects, a factor to \nconsider when evaluating or implementing a scheduling tool is how easy it is to roll up the \ndetails of several subprojects into a larger master project schedule, thereby allowing for \neasier coordination of tasks or resources that span multiple subprojects. \nProject Closure \nAlthough itmight be desirable from a security perspective to keep a security-testing project \nrunning indefinitely, financial reality may mean that such a project ultimately must be \nbrought to closure (if only to be superseded by a replacement project). \nWhen winding down a security testing project, great care must be exercised to ensure that \nconfidential information (such as security assessment reports or a defect-tracking database \nthat contains a list of all the defects that were not fixed because of monitory pressures) \ngenerated by the testing effort does not fall into the wrong hands. This is especially relevant \nif going forward nobody is going to be directly accountable for protecting this information, or \nif some of this information was generated by (or shared with) third parties. \nThe test plan should therefore outline how the project should be decommissioned, itemizing \nimportant tasks such as who will reset (or void) any user accounts that were set up \nspecifically for the testing effort, making sure no assessment tools were left installed on a \nproduction machine, and that any paper deliverables are safely destroyed. \nPlanning Risks and Contingencies \nA planning risk can be any event that adversely affects the planned testing effort (the \nschedule, completeness, quality, and so on). Examples would include the late delivery of \napplication software, the lead security tester quitting to take a better-paid job (leaving a \nhuge gap in the testing team's knowledge base), or the planned test environment not being \nbuilt due to unexpected infrastructure shortages (budget cuts). \nThe primary purpose of identifying in the test plan the most significant planning risks is to \nenable contingency plans to be proactively developed ahead of time and ready for \nimplementation in the event that the potential risk becomes a reality. Table 2.6 lists some \nexample contingency plans. \n \nTable 2.6: Example Contingency Plans  \nPLANNING RISK  \nCONTINGENCY PLAN  \nDon't install the service pack. (Keep the \nscope the same.) \nInstall the service pack and reexecute \nany of the test cases whose results \nhave now been invalidated. (More time \nor resources are needed.) \nInstall the service pack, but don't \nchange the test plan. (The quality of the \ntesting is reduced.) \nMidway \nthrough \nthe \ntesting \neffort, \nMicrosoft releases a new service pack \nfor the operating system installed on a \nlarge number of the servers used by the \nWeb site. \nRedo some of the highly critical tests \nthat have been invalidated and drop \nsome of the lower, as-yet-unexecuted \n"
        },
        {
            "page_number": 39,
            "text": " \n32 \nTable 2.6: Example Contingency Plans  \nPLANNING RISK  \nCONTINGENCY PLAN  \ntests. (The quality of the testing is \nreduced.) \nDo nothing, as the test environment \nbecomes \nless \nlike \nthe \nproduction \nenvironment. (The quality of the testing \nis reduced.) \nBuy a new firewall for the test \nenvironment. (Increase resources.) \nThe production environment upgrades its \nfirewall to a more expensive/higher-\ncapacity version. \nReduce firewall testing in the test \nenvironment and increase testing in the \nproduction environment. (Change the \nscope of the testing.) \nThe entire testing team wins the state \nlottery. \nMake sure you are in the syndicate. \nFor any given risk, typically numerous contingencies could be considered. However, in \nmost cases, the contingencies can be categorized as either extending the time required for \ntesting, reducing the scope of the testing (for example, reducing the number of test items \nthat will be tested), adding additional resources to the testing effort, or reducing the quality \nof the testing (for example, running fewer or less well designed tests), thereby increasing \nthe risk of the system failing. These contingency categories can be illustrated by the quality \ntrade-off triangle depicted in Figure 2.3. Reducing one side of the triangle without \nincreasing at least one of the other sides reduces the quality of the testing (as represented \nby the area inside the triangle). \n \nFigure 2.3: Quality trade-off triangle.  \nNone of these options may sound like a good idea to senior management and they may \ndecide that all these contingencies are unacceptable. Unfortunately, if management does \nnot make a proactive contingency decision, the decisions (and their consequences) do not \ngo away. Instead, they are implicitly passed down to the individual members of the testing \nteam. This results in unplanned consequences such as a tester unilaterally deciding to skip \nan entire series of tests, skimping on the details of an incident report (requiring the incident \ninvestigator to spend more time trying to recreate the problem), or working extra unpaid \nhours (while at the same time looking for another job). None of these consequences are \nlikely to be more desirable than the options that senior management previously decided \nwere unacceptable. \n"
        },
        {
            "page_number": 40,
            "text": " \n33 \nIssues \nMany people may hold the view that each and every issue is merely a risk that needs to be \nmitigated by developing one or more contingencies, resulting in any issues that the testing \nteam faces being documented in the \"planning risks\" section of the test plan. \nAlternatively, issues that have highly undesirable or impractical contingencies (such as a \nhuge increase in the cost of the testing effort), maybe siphoned off from the planning risks \nsection and thereby highlighted in their own section, allowing management to focus on \nthese unresolved issues. \nAssumptions \nIn a perfect world, a test plan would not contain any assumptions, because any assumption \nthat the testing team had to make would be investigated to determine the validity of the \nassumption. Once thoroughly researched, the assumption would be deleted or transferred \nto another section (such as the Planning Risks section). \nUnfortunately, many assumptions may not be possible to prove or disprove because of the \ntime needed to investigate them, or because the people who could confirm the assumption \nare unwilling to do so. For example, the testing team may need to assume that the \ninformation provided by bug- and incident-tracking center Web sites (such as those listed in \nTable 4.2) is accurate, because the effort needed to reconfirm this information would take \ntoo long and consume too many resources. \nConstraints and Dependencies \nThe testing team may find it useful to list all the major constraints that they are bound by. \nObvious constraints include the project's budget or the deadline for its completion. Less \nobvious constraints include a corporate \"no new hires\" mandate (which means that if the \ntesting is to be done in house, it must be performed using the existing staff), or a corporate \nprocurement process that requires the testing team to purchase any hardware or software \nthat costs more than $1,000 through a central purchasing unit (a unit that typically runs six \nto eight weeks behind). \nAcronyms and Definitions \nThis section of the test plan can be used to provide a glossary of terms and acronyms \nreferenced by the test plan and are not normally found in the everyday language of the \nplan's anticipated readership. \nReferences \nIt's generally considered a good practice to include a summary list of all the other \ndocuments that are referenced, explicitly or implicitly, by the test plan (such as the project \nschedule or requirements documentation). Placing the list in its own section towards the \nend of the test plan will improve the readability of this section-and hence improve the \nchances that it is actually used. \nApprovals \nA test plan should identify two groups of approvers. The first group will be made up of those \nindividuals who will decide whether or not the proposed test plan is acceptable and meets \nthe security-testing needs of the organization, whereas the second group (which may be \ncomposed of the same individuals as the first group) will decide whether or not the \n"
        },
        {
            "page_number": 41,
            "text": " \n34 \ndeliverables specified in the test plan and subsequently produced and delivered by the \ntesting team (for example, the test summary report) are acceptable. \nBeing asked to approve something is not the same as being kept informed about it. There \nmay be additional interested parties (stakeholders) who need to be kept informed about the \nongoing and ultimate status of the testing project, but do not really have the organizational \npower to approve or disapprove any of the deliverables listed in the test plan. For example, \nthe configuration management librarian may need to know what the testing team needs in \nterms of configuration management support, but it is unlikely to be able to veto a particular \nset of tests. Rather than listing these individuals as approvers, it may be more accurate to \nidentify them as stakeholders and indicate that their acceptance of the test plan merely \nindicates that they believe they have been adequately informed of the testing project's \nplans. \n \nMaster Test Plan (MTP) \nFor small simple testing projects, a single, short test plan may be all that is needed to \nsufficiently convey the intended activities of the testing team to other interested parties. \nHowever, for larger projects, where the work may be divided across several teams working \nat separate locations for different managers and at different points in the system's \ndevelopment, it may be easier to create several focused test plans rather than one large all-\nencompassing plan. For example, one plan may focus on testing the physical security of \nthe computer facilities that the Web site will be housed in, another may describe the \npenetration testing that will be performed by an outsourced security-testing firm, and a third \nmay concentrate on the unit-level tests that the Web application development team is \nexpected to do. \nIf multiple test plans will be used, the activities within each plan need to be coordinated. For \ninstance, it does not make sense for all of the test plans to schedule the creation of a \ncommon test environment; instead, the first plan that will need this capability should include \nthis information. The higher the number of test plans, the easier it is to manage each \nindividual plan, but the harder it becomes to coordinate all of these distributed activities, \nespecially if the organization's culture does not lend itself to nonhierarchical lines of \norganizational communication. \nOne solution to the problem of multiple test plan coordination that many practitioners \nchoose to utilize is the master test plan (MTP). The MTP is a test plan that provides a high-\nlevel summary of all the other test plans, thereby coordinating and documenting how the \nentire security-testing effort has been divided up into smaller, more manageable units of \nwork (as depicted in Figure 2.4). \n \n \n"
        },
        {
            "page_number": 42,
            "text": " \n35 \n \nFigure 2.4: MTP.  \nA by-product of defining multiple test plans is that such an approach may facilitate several \nsecurity-testing teams working in parallel, which provides a significant advantage when \nworking in Web time. Additionally, having several documented and well-scoped groups of \ntests makes outsourcing some or all of the testing effort much more controllable. \nAs is the case with each individual test plan, it is often the process of developing an MTP \nrather than the actual end product that is of greater help to the testing team. Creating the \nMTP should facilitate discussions on what testing objectives should be assigned to each \nindividual test plan as part of an overall scheme, rather than relying on the recognizance of \nhead-down individuals working in isolation on separate test plans. Craig et al. (2002) and \nGerrard (2002) provide additional information on the concept of a master test plan. \n \nSummary \nWhether a person chooses to use a test plan format based on an industry standard, an \ninternal template, or a unique layout customized for this specific project, the test plan and \nits associated documents should be reviewed to make sure that it adequately addresses \nthe test-planning considerations summarized in Table 2.7. \n \n \nTable 2.7: Test-Planning Consideration Checklist  \nYES  \nNO  \nDESCRIPTION  \n□  \n□  \nHave the system's security requirements been clarified and \nunambiguously documented? \n"
        },
        {
            "page_number": 43,
            "text": " \n36 \nTable 2.7: Test-Planning Consideration Checklist  \nYES  \nNO  \nDESCRIPTION  \n□  \n□  \nHas the goal (and therefore scope) of the testing effort been clearly \ndefined? \n□  \n□  \nHave all the items (and their versions) that need to be tested been \nidentified? \n□  \n□  \nHave any significant items that will not be tested been listed? \n□  \n□  \nHas a change control process for the project been defined and have \nall the individuals who will approve changes to the scope of the \ntesting been identified? \n□  \n□  \nHave all the features that need to be tested been identified? \n□  \n□  \nHave any significant features that will not be tested been listed? \n□  \n□  \nHas the testing approach (strategy) been documented? \n□  \n□  \nHave the criteria (if any) by which the system will be deemed to have \npassed security testing been documented? \n□  \n□  \nHave the criteria (if any) for halting (and resuming) the testing effort \nbeen documented? \n□  \n□  \nHave the deliverables that the testing effort is expected to produce \nbeen documented? \n□  \n□  \nHave all the environmental needs of the testing effort been \nresearched and documented? \n□  \n□  \nHas a configuration management strategy for the items that are to be \ntested been documented? \n□  \n□  \nHas a configuration management strategy for the test scripts and test \ndata (testware) been documented? \n□  \n  \nHave responsibilities for all the testing activities been assigned? \n□  \n□  \nHave responsibilities for all the activities that the testing effort is \ndependent upon been assigned? \n□  \n□  \nHave staffing needs been identified and resourced? \n□  \n□  \nHave any training needs been identified and resourced? \n□  \n□  \nHas a test schedule been created? \n□  \n□  \nHave the steps necessary to bring the project to a graceful closure \nbeen considered? \n□  \n□  \nHave the most significant planning risks been identified? \n□  \n□  \nHave contingency plans for the most significant planning risks been \ndevised and approved? \n□  \n□  \nHave all issues, assumptions, constraints, and dependencies been \n"
        },
        {
            "page_number": 44,
            "text": " \n37 \nTable 2.7: Test-Planning Consideration Checklist  \nYES  \nNO  \nDESCRIPTION  \ndocumented? \n□  \n□  \nHave any unusual acronyms and terms been defined? \n□  \n□  \nHave any supporting documents been identified and cross-\nreferenced? \n□  \n□  \nHave those individuals responsible for approving the test plans been \nidentified? \n□  \n□  \nHave those individuals responsible for accepting the results of the \ntesting effort been identified? \n□  \n□  \nHave those individuals who need to be kept informed of the testing \neffort's plans been identified? \n \n"
        },
        {
            "page_number": 45,
            "text": " \n38 \nPart III: Test Design \nChapter List \nChapter 3: Network Security  \nChapter 4: System Software Security  \nChapter 5: Client-Side Application Security  \nChapter 6: Server-Side Application Security  \nChapter 7: Sneak Attacks: Guarding Against the Less-Thought-of Security Threats  \nChapter 8: Intruder Confusion, Detection, and Response  \n"
        },
        {
            "page_number": 46,
            "text": " \n39 \nChapter 3: Network Security \nOverview \nWhen asked to assess the security of a Web site, the first question that needs to be \nanswered is, \"What is the scope of the assessment?\" Often the answer is not as obvious as \nit would seem. Should the assessment include just the servers that are dedicated to hosting \nthe Web site? Or should the assessment be expanded to include other machines that \nreside on the organization's network? What about the routers that reside upstream at the \nWeb site's Internet service provider (ISP), or even the machines running legacy applications \nthat interface to one of the Web applications running on the Web site? Therefore, one of the \nfirst tasks the testing team should accomplish when starting a security assessment is to \ndefine the scope of the testing effort and get approval for the scope that they have \nproposed. \nThis chapter discusses how a security assessment effort can be scoped by referencing a \nset of network segments. The network devices attached to these segments collectively form \nthe network under test. Adding the physical locations used to house these devices, the \nbusiness process aimed at ensuring their security, and the system software and \napplications that run on any of these devices may then form the collection of test items that \nwill ultimately comprise the system that will be tested by the security-testing effort. Figure \n3.1 graphically depicts this relationship. \n \nFigure 3.1: Scope of security testing.  \nThe subsequent sections of this chapter explains an approach that may be used by the \ntesting team to ensure that the network defined by the scoping effort has been designed \n"
        },
        {
            "page_number": 47,
            "text": " \n40 \nand implemented in a manner that minimizes the probability of a security vulnerability being \nexploited by an attacker (summarized in Figure 3.5). \nMany of the networking terms used in this chapter may not be readily familiar to some \nreaders of this book. Appendix A provides a basic explanation of what each of the network \ndevices referenced in this chapter does and gives an overview of the networking protocols \nused by these components to communicate to each other across the network. Readers who \nare unfamiliar with what a firewall does or what the Transmission Control Protocol/Internet \nProtocol (TCP/IP) is should review this appendix first before continuing with the rest of this \nchapter. For those readers looking for a more detailed explanation of networking concepts, \nthe following books provide a good introduction to this topic: Brooks 2001, Nguyen 2000, \nand Skoudis 2001. \n \nScoping Approach \nA security assessment can be scoped in several ways. The assessment can focus entirely \non the software components that compromise a single Web application or restrict itself to \nonly testing the devices that are dedicated to supporting the Web site. The problem with \nthese and other similar approaches is that they ignore the fact that no matter how secure a \nsingle component is (software or physical device), if the component's neighbor is vulnerable \nto attack and the neighbor is able to communicate to the allegedly secure component \nunfettered, then each component is only as secure as the most insecure member of the \ngroup. To use an analogy, suppose two parents are hoping to spare the youngest of their \nthree children from a nasty stomach bug that's currently going around school. The parents \nwould be deluding themselves if they thought that their youngest child would be protected \nfrom this threat if he were kept home from school and his older sisters still went to school. If \nthe elder siblings were eventually infected, there would be little to stop them from passing it \non to their younger brother. \nThe approach this book uses to define the scope of a security assessment is based on \nidentifying an appropriate set of network segments. Because the term network segment \nmay be interpreted differently by readers with different backgrounds, for the purposes of \ndefining the testing scope, this book defines a network segment as a collection of \nnetworked components that have the capability to freely communicate with each other-such \nas several servers that are connected to a single network hub. \nMany organizations choose to use network components such as firewalls, gateways, proxy \nservers, and routers to restrict network communications. For the purposes of defining a \ntesting scope, these components can be considered to represent candidate boundaries for \neach of the network segments. They can be considered this way because these devices \ngive an organization the opportunity to partition a large network into smaller segments that \ncan be insulated from one another (as depicted in Figure 3.2), potentially isolating (or \ndelaying) a successful intrusion. \n \n"
        },
        {
            "page_number": 48,
            "text": " \n41 \n \nFigure 3.2: Example network segments.  \nThe scope of a security assessment can therefore be documented by referencing a \ncollection of one or more network segments. For some small Web sites, the scope could be \nas simple as a single multipurpose server and a companion firewall appliance. For larger \norganizations, the scope could encompass dozens (or even hundreds) of servers and \nappliances scattered across multiple network segments in several different physical \nlocations. \nDepending upon how the network engineers and local area network (LAN) administrators \nhave (or propose to) physically constructed the network, these network segments may be \neasily referenced (for example, stating that the scope of the testing effort will be restricted \nto the network segments ebiz.tampa, crm.tampa, and dmz.tampa) or each of the physical \ncomponents that compromise the network segments may have to be individually itemized. \nAs previously discussed, including only a portion of a network segment within the scope \nshould be avoided because establishing the security of only a portion of a segment is of \nlittle value unless the remaining portion has already been (or will be) tested by another \ntesting effort. \nOnce the network segments that comprise the scope of the testing effort have been \ndefined, the scope can be further annotated to specify the hardware devices that make up \nthese network segments, the physical facilities that are used to house this hardware, and \nany system software or application software that resides on this hardware, and finally any \nbusiness processes (such as an intruder response process) that are intended to protect the \nsecurity of the devices. \n \nScoping Examples \nThe specific approach used to identify the scope of the testing effort is very dependent on \nthe size of the task and the culture of the organization, as the following scenarios illustrate. \nHotel Chain \nA small hotel chain has decided to place its Web site (the originally stated target of the \ntesting effort) on a handful of servers, which they own and administer, at an off-site facility \nowned and managed by their ISP. On occasion, the Web application running at this site \nneeds to upload new reservations and download revised pricing structures from the \norganization's legacy reservation processing system that resides on the corporate network. \nThe Web application and legacy reservation system communicates via the Internet. Access \nto the corporate network is via the same firewall-protected Internet connection used by the \n"
        },
        {
            "page_number": 49,
            "text": " \n42 \nhotel chain for several other Internet services (such as employee emails and Internet \nbrowsing). Figure 3.3 illustrates this configuration. \n \nFigure 3.3: Hotel chain configuration.  \nCommunication between the Web site and corporate network is via a firewall. Therefore, it \nwould not be unreasonable to restrict the scope of the Web site security-testing effort to that \nof the two network segments that the hotel chain administers at the ISP facility (a \ndemilitarized zone [DMZ] and back-end Web application). On the other hand, had the \ncommunication to the legacy system been via an unfiltered direct network connection to the \ncorporate network, it would have been hard to justify not including the corporate network in \nthe scope (unless it was covered by another testing project). A security breach at the \ncorporate network could easily provide a back-door method of entry to the Web site, \ncircumventing any front-door precautions that may have been implemented between the \nWeb application and the Internet. \nFurniture Manufacturer \nA medium-sized furniture manufacturer has decided to develop its own Web application in \nhouse using contracted resources. Its entire Web site, however, will be hosted at an ISP's \nfacility that offers low-cost Web hosting-so low cost that parts of the Web application \n(specifically the database) will be installed on a server shared with several other clients of \nthe ISP. Assume that the ISP is unwilling (or perhaps unable) to provide the furniture \nmanufacturer with the schematic of its network infrastructure and that it would not \nappreciate any of its clients conducting their own, unsolicited security assessments. The \nfurniture manufacturer should restrict its security-testing activities to testing the in-house-\ndeveloped Web application using its own test lab. The risk of the production version being \nattacked physically or via a system software vulnerability would be mitigated by requiring \nthe ISP to produce evidence that it has already tested its infrastructure to ensure that it is \nwell defended. Ideally, some form of guarantee or insurance policy should back up this \nassurance. \nAccounting Firm \nA small accounting firm, whose senior partner doubles as the firm's system administrator, \nhas hosted its Web site on the partnership's one-and-only file and print server. (This is a \nquestionable decision that the security assessment process should highlight.) This server is \naccessible indirectly from the Internet via a cheap firewall appliance and directly from any \none of the dozen PCs used by the firm's employees. Figure 3.4 illustrates this configuration. \n \n"
        },
        {
            "page_number": 50,
            "text": " \n43 \n \nFigure 3.4: Accounting firm configuration.  \nBecause of the lack of any interior firewall or other devices to prohibit accessing the Web \nsite from a desktop machine, the Web site is only as safe as the least secure PC. (Such a \nPC would be one that, unbeknownst to the rest of the firm, has a remote-access software \npackage installed upon it, so the PC's owner can transfer files back and forth from his or her \nhome over the weekend.) Such a situation would merit including all the PCs in the scope of \nthe security assessment or suspending the security testing until an alternate network \nconfiguration is devised. \nSearch Engine \nA large Internet search engine Web site uses several identical clusters of servers scattered \nacross multiple geographic locations in order to provide its visitors with comprehensive and \nfast search results. The Web site's LAN administrator is able to provide the testing team \nwith a list of all the network segments used by the distributed Web site, and the devices \nconnect to these different segments. \nBecause of the size of this security assessment, the testing team may decide to break the \nassessment up into two projects. The first project would concentrate on testing a single \ncluster of servers for vulnerabilities. Once any vulnerabilities identified by the first project \nhave been fixed, the second phase focuses on ensuring that this previously assessed \nconfiguration has now been implemented identically on all the other clusters. \nThe Test Lab \nAn area that is often overlooked when deciding upon what should be covered by a security-\ntesting effort is the devices and network segment(s) used by the testing team itself to test a \nnonproduction version of the system. Typically, these environments are referred to as test \nlabs. If they are connected to the production environment or to the outside world (for \ninstance, via the Internet), they might pose a potential security threat to the organization \nunless they are included in the security assessment. \nTest labs are notorious for having weak security. They are therefore often the target of \nattackers trying to gain access to another network segment using the testing lab as a \nstepping stone to their ultimate goal. The following are just two of the scenarios that have \ncontributed to this reputation. Test lab machines are often reconfigured or reinstalled so \nfrequently that normal security policies and access controls are often disregarded for the \nsake of convenience. For example, very simple or even blank administrator passwords \nmight be used because the machines are constantly being reformatted, or protective \nsoftware such as antivirus programs are not installed because they generate too many false \nalarms during functional testing and potentially skew the test results obtained during \nperformance testing. Secondly, minimum access controls are used in order to make \nautomated test scripts more robust and less likely to fail midway through a test because the \ntesting tool did not have sufficient privileges. \n"
        },
        {
            "page_number": 51,
            "text": " \n44 \nThe scope of the security assessment should therefore always explicitly state whether or \nnot a system's associated test lab is included in the testing effort and, if not, why it has been \nexcluded. All too often, these test labs' only line of defense is the assumption that no \nattacker knows of their existence; solely relying on a security-by-obscurity strategy is a \ndangerous thing to do. \nSuspension Criteria \nIf, during the scoping of a security assessment, clearly defining a testing scope proves to be \ncompletely impossible, then it might be wise to temporarily suspend the testing effort until \nthis issue can be resolved. Such a situation could have come about because the \ninformation needed to make an informed decision about the network topology could not be \nobtained. This could also occur because the topology that has actually been implemented \nappears to allow such liberal access between multiple network segments that the size of \nsecurity assessment needed to ensure the security of the network would become too vast, \nor if it is restricted to a single segment, it could not ensure the security of the segment \nbecause of the uncertainly associated with other adjacent segments. \nAlternatively, a huge disclaimer could be added to the security assessment report stating \nthat someone else has presumably already thoroughly tested (or will soon test) these \nadjacent segments using the same stringent security policies that this testing project uses. \nThis so-called solution ultimately presents a lot of opportunity for mis-communication and \npotential finger-pointing at a later date, but may also provide the impetus for convincing \nmanagement to devote more time and energy toward remedying the situation. \n \nDevice Inventory \nOnce the network segments that will form the subject of the security assessment have been \nidentified, the next step is to identify the devices that are connected to these segments. A \ndevice inventory is a collection of network devices, together with some pertinent information \nabout each device, that are recorded in a document. \nA device can be referenced in a number of ways, such as its physical location or by any \none of several different network protocol addresses (such as its hostname, IP address, or \nEthernet media access control [MAC] address). Therefore, the inventory should be \ncomprehensive enough to record these different means of identification. The security-\ntesting team will need this information later in order to verify that the network has been \nconstructed as designed. In addition, if a test lab is to be built, much of this information will \nbe needed in order to make the test lab as realistic as possible. Table 3.1 depicts the \npertinent information that the testing team should consider acquiring for each of the devices \nin the device inventory. Appendix A provides background information on the differences \nbetween these different network protocol addresses. \n"
        },
        {
            "page_number": 52,
            "text": " \n45 \n \nTable 3.1: Example Layout Structure for a Device Inventory  \nDEVICE ID  \nDEVICE \nDESCRIPTIO\nN  \nPHYSICAL LOCATION  \nNETWORK \nACCESSIBILITY  \nHOSTNAME(S)  \nIP ADDRESS(ES)  \nMAC ADDRESS(ES)  \n1 \nISP router \nISP facility \n  \n  \n123.456.789.123 \n  \n2 \nInternal router \nTelecom room \n  \n  \n123.456.789.124 \n  \n3 \nPerimeter \nfirewall \nTelecom room \n  \n  \n123.456.789.130 \n  \n4 \nWeb server \n#1 \nMain server room \n  \nweb1.dmz.miami \n123.456.789.131 \naa.bb.cc.dd.bb.aa \n5 \nWeb server \n#2 \nMain server room \n  \nweb2.dmz.miami \n123.456.789.132 \naa.bb.cc.dd.bb.bb \n6 \nFTP server \nMain server room \n  \nftp1.dmz.miami \n123.456.789.133 \naa.bb.cc.dd.bb.cc \n7 \nLoad \nbalancer \nMain server room \n  \n  \n123.456.789.134 \n  \n8 \nDMZ firewall \nMain server room \n  \n  \n123.456.789.135 \n  \n9 \nInternal \nswitch \nMain server room \n  \n  \n123.456.789.136 \n  \n10 \nApplication \nserver \nMain server room \n  \nweblogic1.main.miami \n123.456.789.140 \naa.bb.cc.dd.cc.aa \n11 \nDatabase \nserver \nMain server room \n  \nsybase1.main.miami \n123.456.789.141 \naa.bb.cc.dd.cc.bb \n12 \nComputer \nroom PC # 1 \nMain server room \n  \njoshua.main.miami \n123.456.789.150 \naa.bb.cc.dd.aa.aa \n13 \nComputer \nroom PC #2 \nMain server room \n  \ndavid.main.miami \n123.456.789.151 \naa.bb.cc.dd.aa.bb \n14 \nComputer \nroom PC #3 \nMain server room \n  \nmatthew.main.miami \n123.456.789.152 \naa.bb.cc.dd.aa.cc \n"
        },
        {
            "page_number": 53,
            "text": " \n46 \nTable 3.1: Example Layout Structure for a Device Inventory  \nDEVICE ID  \nDEVICE \nDESCRIPTIO\nN  \nPHYSICAL LOCATION  \nNETWORK \nACCESSIBILITY  \nHOSTNAME(S)  \nIP ADDRESS(ES)  \nMAC ADDRESS(ES)  \nNote that instead of assigning a new inventory ID to each device, the testing team may find it more convenient to use the original equipment manufacturer (OEM) \nserial number or the inventory tracking number (sometimes represented as a barcode) internally assigned by the owning organization. \n \n"
        },
        {
            "page_number": 54,
            "text": " \n47 \n \nAn organization's LAN administrator should be able to provide the testing team with all the \ninformation needed to construct a device inventory by either automatically running an \nalready-installed network auditing tool (such as those listed in Table 3.6) or by referencing \nthe network specification used to construct the network (or if it is not already built, the \nnetwork design that will be used to build the network). If such documentation is not \nimmediately available or if it can't be recreated in a timely manner, this may be a symptom \nof an unmanaged network and/or LAN administrators who are too busy administrating the \nnetwork to keep track of its structure. Either way, such an undesirable scenario is likely to \nbe indicative of a system prone to security vulnerabilities due to an undocumented structure \nand overtaxed administrators. \nIf the information needed to build a device inventory cannot be provided to the testing team, \nthe testing team must decide whether the testing effort should be suspended until the \nsituation can be remedied or attempt to acquire the raw information needed to construct the \ndevice inventory themselves (possible using one or more of the techniques described later \nin the verifying device inventory section of this chapter). Table 3.2 provides a checklist for \nbuilding a device inventory. \n \nTable 3.2: Network Device Inventory Checklist  \nYES  \nNO  \nDESCRIPTION  \n□  \n□  \nHas a list been acquired of all the network devices (servers, routers, \nswitches, hubs, and so on) attached to any of the network segments \nthat form the scope of the security assessment? \n□  \n□  \nHas the physical location of each network device been identified? \n□  \n□  \nHave the network addresses used to reference each network device \nbeen identified? \n \nWIRELESS SEGMENTS \nWhen documenting a network that utilizes wireless communications (such as Bluetooth or \nIEEE 802.11), the desired effective range of the communication should also be recorded \nand the method of encryption (if any) that will be used. \nAlthough a wireless standard may stipulate that the broadcasting device only have a short \nrange, in practice this range may be significantly larger, giving potential intruders the \nopportunity to attach their wireless-enabled laptops to a network by parking their cars \nacross the road from the targeted organizations. \n \nNetwork Topology \nFor simplicity, the LAN administrator may have connected (or if the network is yet to be \nbuilt, may be intending to connect) all the devices that comprise the network under test to a \nsingle network hub, switch, or router, enabling any device to directly communicate to any \nother device on the network. A more compartmentalized approach would be to break the \nnetwork up into several separate network segments. Such an approach should, if \nconfigured properly, improve security (and possibly network performance) by keeping \nnetwork traffic localized at the cost of increased network complexity. If a compartmentalized \napproach is adopted, the network's intended topological configuration should be \n"
        },
        {
            "page_number": 55,
            "text": " \n48 \ndocumented and subsequently verified to ensure that the network has actually been \nconfigured as desired. That would be advisable because an improper implementation may \nprovide not only suboptimal performance, but it also may give attackers more options for \ncircumventing network security precautions. \nFor a small network, this information can be displayed in a neat, concise diagram, which is \nsimilar to the ones depicted in Figures 3.3 and 3.4. This illustration might look nice on the \nLAN administrator's office wall, but unfortunately it would also pose a security leak. \nFor larger networks, a two-dimensional matrix may prove to be a more appropriate method \nof documenting links. Either way, this information should be kept under lock and key and \nonly distributed to authorized personnel as an up-to-date network map would save an \nintruder an awful lot of time and effort. \nDevice Accessibility \nBy restricting the number of devices that can be seen (directly communicated to) by a \nmachine located outside of the organization network, the network offers an external intruder \nfewer potential targets. Anyone looking for security vulnerabilities would be thwarted; \nhence, the network as a whole would be made more secure. The same is true when \nconsidering the visibility of a device to other internal networks: The fewer devices that can \nbe accessed from other internal networks, the fewer opportunities an internal attacker has \nto compromise the network. \nDevice accessibility (or visibility) is an additional attribute that can be added to the device \ninventory. This attribute can be documented in detail, explaining under what specific \ncircumstances a device may be accessed (for example, the database server should only be \naccessible to the application server and the network's management and backup devices) or \nmay be defined in more general terms. For example, each device can be characterized in \nthree ways: as (1) a public device that is visible to the outside world (for instance, the \nInternet), (2) a protected device that can be seen from other internal networks but not \nexternally, or (3) a private device that can only be accessed by devices on the same \nnetwork segment. \nOnce the network designers have specified under what circumstances a device may be \naccessed by another device (and subsequently documented as a network security policy), \nnetwork traffic-filtering devices such as firewalls and routers can be added to the network \ndesign to restrict any undesired communication. Barman (2001), Peltier (2001), and Wood \n(2001) provide additional information on the process of defining security policies. The lack \nof a documented network security policy (and a process for updating this document) is often \nan indicator that different policies are likely to have been implemented on different filtering \ndevices and can cause a potential process issue when the existing staff leaves. \n \nEXAMPLE NETWORK SECURITY POLICIES \nNetwork security polices can be as straightforward as \"only permit access to IP address \n123.456.789.123 via port 80\" or as smart as \"don't permit any incoming network traffic that \nhas a source IP address that matches the IP address of an internal machine.\" Such a \nscenario should not occur in the legitimate world, but an external attacker might alter \n(spoof) his or her originating IP address in an effort to fool an internal machine into thinking \nthat it was communicating to another friendly internal machine. \n \nBLOCK AND THEN OPEN VERSUS OPEN AND THEN BLOCK \n"
        },
        {
            "page_number": 56,
            "text": " \n49 \nSome LAN administrators first deploy a filtering device with all accesses permitted and then \nselectively block/filter potentially harmful ones. A more secure strategy is to block all \naccesses and then only open up the required ones, as an implementation error using the \nsecond approach is likely to eventually be spotted by a legitimate user being denied access \nto a resource. However, an implementation error in the former approach may go undetected \nuntil an intruder has successfully breached the filter device, been detected, and \nsubsequently been traced back to his or her entry point, which is a much more dire scenario \nand one that should therefore be checked when inspecting the network filtering rules used \nby a filtering device. \nDocumenting the conditions (or rules) under which a device may be accessed in an \norganization's network security policy is one thing; however, implementing these rules is \nmore difficult because each network security policy must be converted into an access-\ncontrol rule that can be programmed into the appropriate network-filtering device. Zwicky, et \nal. (2000) provide information on how to implement network-filtering rules using a firewall. \nTable 3.3 summarizes the network topology decisions that should ideally be documented \nbefore a network design (proposed or implemented) can be validated effectively. \n \nTable 3.3: Network Topology Checklist  \nYES  \nNO  \nDESCRIPTION  \n□  \n□  \nHas a diagram (or alternate documentation) depicting the network's \ntopology been acquired? \n□  \n□  \nHave the effective ranges of any wireless network segments been \ndefined? \n□  \n□  \nHas the network accessibility of each device been defined? \n□  \n□  \nHave the network security polices needed to restrict undesired \nnetwork traffic been defined? \n□  \n□  \nIs there a documented policy in place that describes how these \nnetwork security policies will be updated? \n \n \nValidating Network Design \nOnce the network's proposed design (if the network is yet to be built) or implemented \ndesign (if the assessment is occurring after the network has been built) has been accurately \ndocumented, it is possible to review the design to ascertain how prone the design is to a \nsecurity breach. Often the goals of security run counter to the goals of easy maintenance. \n(A network that is easy for a LAN administrator to administer may also be easy for an \nintruder to navigate.) In some cases, stricter security may improve network performance; in \nothers, it may reduce the capacity of the network. For these reasons, it's unlikely that a \nnetwork design can be categorized simply as right or wrong (blatant errors aside); instead, \nit can be said to have been optimized for one of the following attributes: performance, \ncapacity, security, availability, robustness/fault tolerance, scalability, maintainability, cost, \nor, more likely, a balancing act among all of these attributes. Therefore, before starting any \ndesign review, the respective priorities of each of these design attributes should be \ndetermined and approved by the network's owner. \n"
        },
        {
            "page_number": 57,
            "text": " \n50 \nNetwork Design Reviews \nApplication developers commonly get together to review application code, but much less \ncommonly does this technique actually get applied to network designs. Perhaps this \nhappens because the number of people within an organization qualified to conduct a \nnetwork review is smaller than the number able to review an application. Maybe it's \nbecause the prerequisite of documenting and circulating the item to be discussed is much \neasier when the object of the review is a standalone application program rather than a (as \nof yet undocumented) network design. Or maybe it's simply a question of the organization's \nculture: \"We've never done one before, so why start now?\" Whatever the reason for not \ndoing them, reviews have been found by those organizations that do perform them to be \none of the most cost-effective methods of identifying defects. A network design review \nshould therefore always be considered for incorporation into a network's construction or, if \nalready built, its assessment. \nThe first step in conducting a network design review is to identify the potential participants. \nObviously, the LAN administrator, the network designer, and a representative of the \nsecurity-testing team should always be included, if possible. Other candidates include the \nnetwork's owner, internal or external peers of the LAN administrator, network security \nconsultants, and end users. (At times, a knowledgeable end user can assist with giving the \nuser's perspective on network design priorities as they are the ones that will primarily be \nusing the network.) \nOnce the participants have been identified, they should be sent a copy of the network \ntopology and device inventory in advance of the review plus any networking requirements \nthat have been explicitly specified. (It goes without saying that these review packages \nshould be treated with the same confidentiality as the master network documentation \nbecause an intruder would find these copies as useful as the originals.) Once the \nparticipants have had a chance to review the network design, a meeting can be scheduled \nfor everybody to discuss their findings. Oppenheimer (1999) and Dimarzio (2001) provide \ninformation on network design concepts and implementations. \nNetwork Design Inspections \nAn inspection differs from a review in that an inspection compares the network design \nagainst a predefined checklist. The checklist could be based on anything from an industry \nstandard defined by an external organization to a homegrown set of best-practice \nguidelines. Table 3.4 lists some questions that may be used as part of a network design \ninspection when evaluating the network from a security perspective. \n \nTable 3.4: Network Design Security Inspection Checklist  \nYES  \nNO  \nDESCRIPTION  \n□  \n□  \nIs the number of network segments appropriate? For example, \nshould a network be segmented or an existing network segment \nfurther divided? or should one or more segments be merged? \n□  \n□  \nIs each network device connected to the most appropriate network \nsegment(s)? \n□  \n□  \nAre the most appropriate types of equipment being used to connect \nthe devices on a network segment together, such as switches, hubs, \ndirect connections, and so on? \n"
        },
        {
            "page_number": 58,
            "text": " \n51 \nTable 3.4: Network Design Security Inspection Checklist  \nYES  \nNO  \nDESCRIPTION  \n□  \n□  \nAre the most appropriate types of equipment being used to connect \ndifferent segments together such as bridges, routers, gateways, and \nso on? \n□  \n□  \nIs each connection between the network segments absolutely \nnecessary? \n□  \n□  \nDoes each network device have an appropriate number of network \nconnections (network interface cards [NICs] and IP addresses)? \n□  \n□  \nHas each device been made accessible/visible to only the \nappropriate network segments? \n□  \n□  \nWill the network security policies that have been defined ensure that \nonly the appropriate devices are accessible? \n \n \nVerifying Device Inventory \nOnce the network design has been reviewed and any changes have been agreed upon and \nimplemented, the testing team can use the revised device inventory to verify that any \nchanges that were identified as part of the network review process have been implemented \nby the networking team correctly. \nPhysical Location \nFor small networks, the task of confirming the precise physical location of each network \ndevice inventory may be as simple as going into the server room and counting two boxes. \nFor larger, more complex network infrastructures, this process is not as straightforward and \nthe answers are not as obvious. For instance, the device inventory may have specified that \nthe router that connects an organization's Web site to its ISP be housed in the secure \nserver room, but in reality this device is located in the telephone switching room and \nprotected by a door that is rarely locked. \nEven if a room's head count matches the number of devices that were expected to be found \nat this location (and the OEM serial numbers or internal inventory tracking numbers are not \navailable), it is not guaranteed that two or more entries in the device inventory were not \ntransposed. For instance, according to the device inventory, Web server 1 (with a hostname \nof web1.corp) is supposed to be located at the downtown facility, whereas Web server 5 \n(with a hostname of web5.corp) is at the midtown facility. However, in reality, web1 is at the \nmidtown facility and web2 is at the downtown facility. \nVerifying that devices that have a built-in user interface (such as a general-purpose server) \nare physically located where they are expected to be can be as simple as logging into each \ndevice in a room and confirming that its hostname matches the one specified in the device \ninventory. For example, on a Windows-based machine, this could be done via the control \npanel, or on a Unix system, via the hostname command. For devices such as printers and \nnetwork routers that don't necessarily have built-in user interfaces, their true identity will \nneed to be ascertained by probing it from a more user-friendly device directly attached to it. \nThe physical inventory is intended to do two things: confirm that all of the devices listed in \nthe device inventory area are where they are supposed to be and identify any extra devices \n"
        },
        {
            "page_number": 59,
            "text": " \n52 \nthat are currently residing in a restricted area such as a server room. For instance, an old \npowered-down server residing in the back corner of a server room still poses a security risk. \nIt should be either added to the device inventory (and hence added to the scope of the \ntesting effort) or removed from the secure area because an intruder who was able to gain \nphysical access to such a device could easily load a set of hacking tools on to the machine \nand temporarily add it to the network via a spare port on a nearby network device. \n \nSTICKY LABELS \nAdding an easily viewable sticky label to the outside of each device, indicating its inventory \nidentifier, should help speed up the process of confirming the physical location of each \nnetwork device should this task need to be repeated in the near future. \nIf sticky labels are used, care should be taken to ensure that the selected identifier does not \nprovide a potential attacker with any useful information. For example, using an \norganization's internal inventory tracking number (possibly in the form of a barcode) would \nbe more secure than displaying a device's IP address(es) in plain view. \nUnauthorized Devices \nAside from physically walking around looking for devices that should not be connected to \nthe network, another approach to discovering unwanted network devices is to perform an IP \naddress sweep on each network segment included in the security-testing efforts scope. \nThe network protocol of choice for conducting an IP address sweep is the Internet Control \nMessage Protocol (ICMP). (Appendix A provides more details on network protocols \ncommonly used by Web applications.) ICMP is also known as ping; hence, the term ping \nsweep is often used to describe the activity of identifying all the IP addresses active on a \ngiven network segment. In the following Windows command-line example, the IP address \n123.456.789.123 successfully acknowledges the ping request four times, taking on average \n130 milliseconds: \n  C:\\>ping 123.456.789.123 \n \n  Pinging 123.456.789.123 with 32 bytes of data: \n \n  Reply from 123.456.789.123: bytes=32 time=97ms TTL=110 \n  Reply from 123.456.789.123: bytes=32 time=82ms TTL=110 \n  Reply from 123.456.789.123: bytes=32 time=151ms TTL=110 \n  Reply from 123.456.789.123: bytes=32 time=193ms TTL=110 \n \n  Ping statistics for 123.456.789.123: \n       Packets: Sent = 4, Received = 4, Lost = 0 (0% loss), \n  Approximate round trip times in milliseconds: \n       Minimum = 82ms, Maximum = 193ms, Average = 130ms \nUnfortunately (from an IP address sweeping perspective), some LAN administrators may \nhave configured one or more of the network devices not to respond to an ICMP (ping) \nprotocol request. If this is the case, it may still be possible to conduct an IP address sweep \nusing another network protocol such as TCP or the User Datagram Protocol (UDP). \n"
        },
        {
            "page_number": 60,
            "text": " \n53 \n \nIP ADDRESS SWEEP \nOne way to determine what devices are active on the network is to direct a \"Hello is anyone \nthere?\" message at every IP address that might be used by a device connected to the \nnetwork. A \"Yes, I'm here\" reply would indicate that the IP address is being used by a \ndevice. Since any active device needs an IP address to communicate to other network \ndevices (a few passive devices may not use an IP address), the sum total of positive replies \nshould comprise all of the powered-up devices connected to the network segment being \nswept. \nThese sweeps may prove to be quite time consuming if the entire range of potential IP \naddresses needs to be scanned manually. Fortunately, numerous IP-address-sweeping \ntools exist that can be used to automate this often tedious task (see Table 3.5). Klevinsky \n(2002) and Scambray (2001) both provide more information on how to conduct a ping \nsweep. \n \nTable 3.5: Sample List of IP-Address-Sweeping Tools and Services  \nNAME  \nASSOCIATED WEB SITE  \nFping \nwww.deter.com  \nHping \nwww.kyuzz.org/antirez  \nIcmpenum & Pinger \nwww.nmrc.org  \nNetScanTools \nwww.nwpsw.com  \nNmap \nwww.insecure.org  \nNmapNT \nwww.eeye.com  \nPing \nwww.procheckup.com and www.trulan.com  \nPing Sweep/SolarWinds \nwww.solarwinds.net  \nWS_Ping Pro Pack \nwww.ipswitch.com  \nAn IP address sweep is often the first indication of an intruder sniffing around, as an \nexternal intruder typically does not know what IP addresses are valid and active, and is \ntherefore often forced to grope in the dark hoping to illuminate (or enumerate) IP addresses \nthat the LAN administrator has not restricted access to. The testing team should therefore \nmake sure that it informs any interested parties before it conducts their (often blatant) IP \naddress sweeps, so as not to be confused with a real attack, especially if an organization \nhas an intruder detection system that is sensitive to this kind of reconnaissance work. \nNetwork Addresses \nSeveral different techniques can be used to verify that the networking team has assigned \neach network device the network addresses specified by the device inventory. In addition, \nthese techniques can also be used to confirm that no unauthorized network addresses have \nbeen assigned to legitimate (or unauthorize) devices. Appendix A provides additional \ninformation on network addresses and on some of the network-addressing scenarios that a \ntesting team may encounter (and that could possibly cause them confusion) while trying to \nverify or build an inventory of network addresses used by a Web site. \n"
        },
        {
            "page_number": 61,
            "text": " \n54 \nCommercial Tools \nThe organization may have already invested in a commercial network-auditing or \nmanagement tool (such as those listed in Table 3.6) that has the capability to produce a \nreport documenting all the network addresses that each network device currently has \nassigned. Because some of these tools may require a software component to be installed \non each network device, some of these tools may not be particularly capable of detecting \ndevices that are not supposed to be attached to the network. Care should also be taken to \nmake sure than when these tools are used, all the devices that are to be audited are \npowered up, as powered-down machines could easily be omitted. \n \n \nTable 3.6: Sample List of Network-Auditing Tools  \nNAME  \nASSOCIATED WEB SITE  \nDiscovery \nwww.centennial.co.uk  \nLANauditor \nwww.lanauditor.com  \nLan-Inspector \nwww.vislogic.org  \nNetwork Software Audit \nwww.mfxr.com  \nOpenManage \nwww.dell.com  \nPC Audit Plus \nwww.eurotek.co.uk  \nSystems Management Server (SMS) \nwww.microsoft.com  \nTivoli \nwww.ibm.com  \nToptools and OpenView \nwww.hp.com  \nTrackBird \nwww.trackbird.com  \nUnicenter \nwww.ca.com  \nZAC Suite \nwww.nai.com/magicsolutions.com  \nDomain Name System (DNS) Zone Transfers \nAn extremely convenient way of obtaining information on all the network addresses used by \na network is to request the common device used by the network to resolve network address \ntranslations to transfer en masse these details to the device making the request. A request \nthat can be accomplished using a technique called a domain name system (DNS) zone \ntransfer. DNS transfers have legitimate uses, such as when a LAN administrator is setting \nup a new LAN at a branch location and does not want to manually reenter all the network \naddresses used by the corporate network. Unfortunately, this capability is open to abuse \nand if it is made available locally, it may still be blocked by any network-filtering device such \nas a perimeter firewall. \nA DNS zone transfer can either be initiated from a built-in operating system command such \nas nslookup or via a tool such as the ones listed in Table 3.7. Scambray (2001) and \nSkoudis (2001) both provide more information on how to attempt a DNS zone transfer. \n \n"
        },
        {
            "page_number": 62,
            "text": " \n55 \nTable 3.7: Sample List of DNS Zone Transfer Tools and Services  \nNAME  \nASSOCIATED WEB SITE  \nDig \nftp.cerias.purdue.edu \nDNS Audit/Solarwinds \nsolarwinds.net \nDnscan \nftp.technotronic.com \nDnswalk \nvisi.com/~barr \nDomtools \ndomtools.com \nHost \nftp.uu.net/networking/ip/dns \nSam Spade \nsamspade.org \nManual \nIf an automated approach cannot be used because of powered-down machines or \nsuspicions that stealthy hidden devices have been connected to the network, then while \nconfirming the physical location of each network device, the testing team may also want to \nmanually confirm the network addresses assigned to each device. For Windows- or Unix-\nbased devices, the network addresses can be determined using one or more of the \ncommands (such as those listed in Table 3.8) that are built into the operating system. \nNetwork devices that do not offer built-in commands to support these kind of inquires (such \nas a network printer) may require probing from a more user-friendly device directly \nconnected to it. \n \nTable 3.8: Sample List of Built-In Operating System Commands  \nOPERATING SYSTEM  \nCOMMANDS  \nUnix \nhostname, ifconfig, or nslookup \nWindows \narp, ipconfig, net, nslookup, route, winipcfg, or wntipcfg \nNote that not all commands have been implemented on every version of the \noperating system. \nTable 3.9 summarizes the checks that can be used to verify that each device \nimplementation matches its corresponding device inventory entry. \n \nTable 3.9: Verifying Network Device Inventory Implementation Checklist  \nYES  \nNO  \nDESCRIPTION  \n□  \n□  \nAre all of the devices listed in the device inventory physically located \nwhere they should be? \n□  \n□  \nHave unauthorized devices been checked for? \n□  \n□  \nHave all of the devices listed in the device inventory been assigned \ntheir appropriate network addresses? \n \n"
        },
        {
            "page_number": 63,
            "text": " \n56 \nVerifying Network Topology \nOnce the testing team has verified that each individual device has been configured \ncorrectly, the next step is to confirm that the physical connections between these devices \nhave been implemented exactly as specified and that any network-filtering rules have been \napplied correctly. \nNetwork Connections \nFor most small networks, a visual inspection may be all that is needed to confirm that each \ndevice has been physically connected to its network peers. For larger, more complicated \nand/or dispersed networks, verifying that all of the network connections have been \nimplemented correctly and no unneeded connections established is probably most \nproductively done electronically. \nThe network's topological map (or matrix) can be manually verified by logging into each \ndevice on the network and using built-in operating system commands such as tracert \n(Windows) or traceroute (Unix). These commands show the path taken by an ICMP request \nas it traverses the network (hopping from device to device) to its ultimate destination. In the \nfollowing Windows command-line example, it appears that the initiating device (IP address \n123.456.789.123) is directly connected to the device with IP address 123.456.789.124 and \nindirectly connected to the target of the request (IP address 123.456.789.125). \n  C::>tracert web1.tampa \n  Tracing route to web1.tampa [123.456.789.125] \n  over a maximum of 30 hops: \n \n  1    69 ms      27 ms      14 ms      bigboy.tampa \n[123.456.789.123] \n  2    28 ms      <10 ms     14 ms      123.456.789.124 \n  3    41 ms      27 ms      14 ms      web1.tampa \n[123.456.789.125] \n \n  Trace complete. \nTable 3.10 lists some tools and services that provide more advanced and user-friendly \nversions of these trace route commands. In addition, some of the network auditng tools \nlisted in Table 3.6 provide features for constructing a topological map of the network they \nare managing. \n \nTable 3.10: Sample List of Trace Route Tools and Services  \nNAME  \nASSOCIATED WEB SITE  \nCheops \nwww.marko.net  \nNeoTrace \nwww.mcafee.com/neoworx.com  \nQcheck \nwww.netiq.com  \nSolarWinds \nwww.solarwinds.net  \nTrace \nwww.network-tools.com  \n"
        },
        {
            "page_number": 64,
            "text": " \n57 \nTable 3.10: Sample List of Trace Route Tools and Services  \nNAME  \nASSOCIATED WEB SITE  \nTraceRoute \nwww.procheckup.com  \nTracert \nwww.trulan.com  \nTracerX \nwww.packetfactory.net  \nVisualRoute \nwww.visualroute.com  \nDevice Accessibility \nContrary to popular belief, once a network-traffic-filtering device such as a firewall is \nconnected to a network, it will not immediately start protecting a network; rather, the filtering \ndevice must first be configured to permit only the network traffic to pass through that is \nappropriate for the specific network it has just been attached to. Although the \nmanufacturer's default configuration may be a good starting point, relying on default \nsettings is a risky business. For instance, default outgoing network traffic policies are often \ntoo liberal, perhaps due to the mindset of an external intruder who does not consider the \npossibility of an attack being initiated from within an organization or that of an external \nintruder wanting to send confidential information out of an organization. \nTherefore, each traffic-filtering device should be checked to make sure that it has been \nconfigured according to the filtering rules defined by the network's security policies and that \nonly the network devices that should be visible to devices on other network segments are \nactually accessible. These network security policies are often implemented as a series of \nrules in a firewall's access control list (ACL). These rules can either be manually inspected \nvia a peer-level review and/or checked by executing a series of tests designed to confirm \nthat each rule has indeed been implemented correctly. \nA filtering device can be tested by using a device directly connected to the outward-facing \nside of a network-filtering device. The testing should try to communicate to each of the \ndevices located on the network segment(s) that the filtering device is intended to protect. \nAlthough many organizations may test their network-filtering defenses by trying to break \ninto a network, fewer organizations run tests designed to make sure that unauthorized \nnetwork traffic cannot break out of their network. The testing team should therefore \nconsider reversing the testing situation and attempt to communicate to devices located in \nthe outside world from each device located on the network segment(s) being protected by \nthe filtering device. Allen (2001) and Nguyen (2000) both provide additional information on \nhow to test a firewall. \nTesting Network-Filtering Devices \nAlthough in many cases filtering implementations are too lax, in others a filter may be too \nstrict, which restricts legitimate traffic. Therefore, tests should also be considered to make \nsure that all approved traffic can pass unfettered by the filter. \nIf multiple filters are to be used—such as a DMZ configuration that uses two firewalls (a \nscenario expended upon in Appendix A)—each filter should be tested to ensure that it has \nbeen correctly configured. This procedure is recommended as it is unwise to only check a \nnetwork's perimeter firewall and assume that just because it is configured correctly (and \nhence blocks all inappropriate traffic that it sees) every other filter is also configured \ncorrectly. This is particularly true when internal firewalls that block communication between \ntwo internal network segments are considered. The most restrictive perimeter firewall does \n"
        },
        {
            "page_number": 65,
            "text": " \n58 \nnothing to prohibit a network intrusion initiated from within an organization (which is a more \ncommon scenario than an externally launched attack). \nIf some of the filter's rules are dependent not only on the destination address of the network \ntraffic, but also on the source address, then in addition to requesting access to permitted \nand restricted devices, it may also be necessary to vary (spoof) the source address in order \nto create authorized and unauthorized network traffic data for inbound and outbound tests. \nFirewalls often have the capability to log inbound and/or outbound requests. This feature \ncan be useful if evidence is needed to prosecute an offender or the network security team is \ninterested in receiving an early warning that someone is attempting to gain unauthorized \nentry. If logging is enabled and unmonitored, aside from slowing down a firewall (another \ncase of a performance optimization conflicting with a security consideration) and \nunmonitored, the logs may grow to a point where the firewall's functionality integrity is \ncompromised. Endurance tests should therefore be considered to make sure that any \nlogging activity does not interfere with the firewall's filtering capabilities over time. \n \nSPOOFING \nSpoofing refers to the technique of changing the original network address of a network \nmessage, typically from an untrusted address to one that is trusted by a firewall (such as \nthe address of the firewall itself). Of course, one of the side effects of changing the \norigination address is that the target machine will now reply to the spoofed address and not \nthe original address. Although it might be possible for an intruder to alter a network's \nconfiguration to set up bidirectional communication (or source routing), spoofing will \ntypically result in the intruder only having unidirectional communication (or blind spoofing). \nUnfortunately, unidirectional communication is all an intruder needs if he or she is only \ninterested in executing system commands (such as creating a new user account) and is not \ninterested in (or does not need to be) receiving any confirmation responses. \nSome filtering devices (such as proxy servers) have a harder time deciding what should \nand should not be filtered as the load placed on them increases. At high load levels, the \ndevice may be so stressed that it starts to miss data packets that it should be blocking. \nRunning stress tests against the filtering device should be considered to ascertain whether \nor not they exhibit this behavior. If they do, consider inserting a network load governor to \nensure that such a heavy load will not be placed on the susceptible device in a production \nenvironment. \nA firewall only works if it's enabled. Forgetting to change or disable any default user IDs and \npasswords or removing any remote login capability that might have been enabled by the \nmanufacturer may allow an intruder to disable the firewall or selectively remove a pesky \nfiltering rule that is preventing him or her from accessing the network behind the firewall. \nTherefore, checks should be considered to make sure that neither of these potential \noversights makes it into production (see Table 3.11). \n \nTable 3.11: Network Topology Verification Checklist  \nYES  \nNO  \nDESCRIPTION  \n□  \n□  \nDoes the implemented network topology match the topology specified \nby the approved network topology design? \n□  \n□  \nHave default configuration settings for each network-traffic-filtering \ndevice been reviewed and, if necessary, changed (for example, \n"
        },
        {
            "page_number": 66,
            "text": " \n59 \nTable 3.11: Network Topology Verification Checklist  \nYES  \nNO  \nDESCRIPTION  \nassigning new and different user IDs and passwords)? \n□  \n□  \nHave all the inbound network-traffic-filtering rules been implemented \ncorrectly on every filtering device? \n□  \n□  \nHave all the outbound network-traffic-filtering rules been implemented \ncorrectly on every filtering device? \n□  \n□  \nDo all of the filtering devices still work correctly when exposed to \nheavy network loads? \n□  \n□  \nIf network-traffic-filtering logs are being used, are the logs being \nmonitored for signs of an intruder at work, lack of free disk space, or \nother noteworthy events? \n \nSupplemental Network Security \nIn addition to the basic security measures described in the preceding sections, some \norganizations implement additional network security measures to make it harder for any \nattacker to compromise the security of the network. Unfortunately, these measures come at \na price, typically one of additional network complexity, which in turn means additional \nnetwork administration—an overhead that may not be justified for every network. However, \nif such additional measures are deemed desirable, then the security-testing team should \nconsider running tests to check that these extra precautions have indeed been \nimplemented correctly. \nNetwork Address Corruption \nTo facilitate more compatible networks, most network protocols utilize several different \nnetwork addresses (for instance, the Web sites typically use three addresses: a \nhost/domain name, an IP address, and a MAC address). Each time a data packet is passed \nfrom one network device to another, the sending device typically must convert an address \nfrom one address format to another (for example, the domain name wiley.com must first be \nconverted to the IP address 123.456.789.123 before the data can be passed across the \nInternet). Ordinarily, this translation process occurs without incident, each device \nremembering (caching) the translations that it repeatedly has to perform and occasionally \nrefreshing this information or requesting a new network address mapping for a network \naddress it has not had to translate before (or recently) from a network controller. \nUnfortunately, if intruders are able to gain access to one or more devices on a network, \nthey may be able to corrupt these mappings (a technique often referred to as spoofing or \npoisoning). They may misdirect network traffic to alternate devices (often a device that is \nbeing used by an intruder to eavesdrop on the misdirected network traffic). To reduce the \npossibility of this form of subversion, some LAN administrators permanently set critical \nnetwork address mappings on some devices (such as the network address of the Web \nserver on a firewall), making these network address translations static. Permanent (static) \nentries are much less prone to manipulation than entries that are dynamically resolved (and \nmay even improve network performance ever so slightly, as fewer translation lookups need \nto be performed). However, manually setting network address mappings can be time \nconsuming and is therefore not typically implemented for every probable translation. \n"
        },
        {
            "page_number": 67,
            "text": " \n60 \nHostname-to-IP-Address Corruption \nA device needing to resolve a hostname-to-IP-address translation typically calls a local \nDNS server on an as-needed basis (dynamically). To make sure that erroneous or \nunauthorized entries are not present, DNS mappings can be checked using a built-in \noperating system command such as nslookup or using a tool such as the ones listed in \nTable 3.7 . Alternatively, a LAN administrator may have hardcoded some critical DNS \nmappings using a device's hosts file (note that this file has no file-type extension), thereby \nremoving the need for the device to use a DNS server and mitigating the possibility of this \nlookup being corrupted (improving network performance ever so slightly). The following is a \nsample layout of a hosts file that might be found on a Windows-based device: \n  # This is a HOSTS file used by Microsoft TCP/IP for Windows. \n  # This file contains the mappings of IP addresses to host names. \nEach \n  # entry should be kept on an individual line. The IP address \nshould \n  # be placed in the first column followed by the corresponding \nhost name. \n  # The IP address and the host name should be separated by at \nleast one \n  # space. \n  # Additionally, comments (such as these) may be inserted on \nindividual \n  # lines or following the machine name denoted by a '#' symbol. \n \n  127.0.0.1       localhost \n \n  123.456.789.123 wiley.com \nThe static hostname-to-IP-address mappings on each device can be tested by either a \nvisual inspection of the hosts file in which the static mappings contained in this file can be \nviewed (or edited) using a simple text-based editor such as Notepad (Windows) or vi (Unix), \nor by using a simple networking utility that must resolve the mapping before it is able to \nperform its designated task. (For example, entering ping wiley.com from a command-line \nprompt requires the host device to convert wiley.com to an IP address before being able to \nping its intended target.) \nIP Address Forwarding Corruption \nInstead of corrupting a network address mapping, an intruder may attempt to misdirect \nnetwork traffic by modifying the routing tables used by network devices to forward network \ntraffic to their ultimate destination. To confirm that a network device such as a router has \nnot had its IP routing tables misconfigured or altered by an intruder, these tables can either \nbe manually inspected (typically using the utility originally used to configure these routing \ntables) or verified by sending network traffic destined for all probable network destinations \nvia the device being tested and then monitoring the IP address that the device actually \nforwards the test network traffic to. \n"
        },
        {
            "page_number": 68,
            "text": " \n61 \nIP-Address-to-MAC-Address Corruption \nThe Address Resolution Protocol (ARP) is the protocol used to convert IP addresses to \nphysical network addresses. For Ethernet-based LANs, the physical network address is \nknown as a MAC address. \nAs with hostname-to-IP-address mappings, a LAN administrator may choose to selectively \nuse static mappings for IP-address-to-MAC-address mappings. If static ARP entries are \nsupposed to have been implemented, each device should be checked to see which ARP \nentries are static and which are dynamic. This can be done by using a tool such as \narpwatch (www.ee.lbl.gov) or manually visiting every device and using a built-in operating \nsystem command such as arp. In the following Windows command-line example, only one \nof the ARP entries has been set permanently (statically): \n  C:\\>arp -a \n  Interface: 123.456.789.123 on Interface 0x2 \n       Internet Address      Physical Address     Type \n       123.456.789.124       aa.bb.cc.dd.ee.ff    static \n       123.456.789.125       aa.bb.cc.dd.ee.aa    dynamic \nTable 3.12 provides a checklist for verifying that static network addresses have been \nimplemented correctly. \n \nTable 3.12: Network Address Corruption Checklist  \nYES  \nNO  \nDESCRIPTION  \n□  \n□  \nIs there a documented policy in place that describes which devices \nare to use static network addresses and which specific addresses are \nto be statically defined? \n□  \n□  \nAre all of the devices that should be using static network addresses \nactually using static addressing? \nSecure LAN Communications \nUnless encrypted, any data transmitted across a LAN can potentially be eavesdropped \n(sniffed) by either installing a sniffer application onto a compromised device or attaching a \nsniffing appliance to the cabling that makes up the LAN (an exercise that is made a lot \neasier if the network uses wireless connections). \nTo protect against internal sniffing, sensitive data (such as application user IDs and \npasswords) transmitted between these internal devices should be encrypted and/or \ntransmitted only over physically secured cabling. For example, a direct connection between \ntwo servers locked behind a secure door would require an intruder to first compromise one \nof the servers before he or she could listen to any of the communications. \nTo check for sensitive data being transmitted across a LAN in cleartext (unencrypted), a \nnetwork- or host-based network-sniffing device (such as one of the tools listed in Table \n3.13) can be placed on different network segments and devices to sniff for insecure data \ntransmissions. Due to the large amount of background traffic (for example, ARP requests) \nthat typically occurs on larger LANs, the sniffing tool should be configured to filter out this \nnoise, making the analysis of the data communication much easier. \n"
        },
        {
            "page_number": 69,
            "text": " \n62 \n \nTable 3.13: Sample List of Network-Sniffing Tools  \nNAME  \nASSOCIATED WEB SITE  \nAgilent Advisor \nwww.onenetworks.comms.agilent.com  \nDragonware (Carnivore) \nwww.fbi.gov  \nCommView \nwww.tamos.com  \nDistinct Network Monitor \nwww.distinct.com  \nEsniff/Linsniff/Solsniff \nwww.rootshell.com  \nEthereal \nwww.zing.org  \nEthertest \nwww.fte.com  \nIris \nwww.eeye.com  \nNetBoy \nwww.ndgssoftware.com  \nNetMon & Windows Network \nMonitor \nwww.microsoft.com  \nSniff'em \nwww.sniff-em.com  \nSniffer \nwww.sniffer.com  \nTCPDump \nwww.ee.lbl.gov and www.tcpdump.org  \nWinDump \nwww.netgroup-serv.polito.it  \nNote that some of the functional testing tools listed in Table 6.12 can also be used to \nsniff network traffic entering or leaving the device they are installed on. \nA more rigorous test would be to input selectively sniffed data into a decryption tool (such \nas the ones listed in Table 4.16) to ascertain whether or not the data was sufficiently \nencrypted and not easily decipherable. Table 3.14 is a sample checklist for testing the \nsafety of LAN network communications. \n \nTable 3.14: Secure LAN Communication Checklist  \nYES  \nNO  \nDESCRIPTION  \n□  \n□  \nIs there a documented policy in place that describes how sensitive \ndata should be transmitted within a LAN? \n□  \n□  \nWhen using a sniffer application on each network segment (or \ndevice), is sensitive data being transmitted or received by any service \nrunning on the network in cleartext or in a format that can be easily \ndeciphered? \n□  \n□  \nAre the physical cables and sockets used to connect each of the \ncomponents on the network protected from an inside intruder directly \nattaching a network-sniffing device to the network? \n"
        },
        {
            "page_number": 70,
            "text": " \n63 \nWireless Segments \nIt is generally considered good practice to encrypt any nonpublic network traffic that may be \ntransmitted via wireless communications. However, due to the performance degradation \ncaused by using strong encryption and the possibility of a vulnerability existing within the \nencryption protocol itself, it would be prudent not to broadcast any wireless communication \nfurther than it absolutely needs to be. If wireless communications will be used anywhere on \nthe network under test, then the network's supporting documentation should specify the \nmaximum distance that these signals should be receivable. The larger the distance, the \nmore mobile-friendly the network will be, but the greater the risk that an eavesdropper may \nalso be able to listen to any communications. \nAlthough the wireless standard may specify certain distances that wireless devices should \nbe effective over, each individual implementation varies in the actual reception coverage. \nThe reasons why this coverage varies from network to network include the following: \nƒ \nTransmitter. The more power a transmitter devotes to broadcasting its signal, the \nfarther the signal is propagated. \nƒ \nReceiver. By using specialized (gain-enhancing and/or directional) antennas, a \nreceiving device can extend its effective range. \nƒ \nHeight. The higher the broadcasting (and receiving) device, the farther the signal \ncan travel. For example, a wireless router located on the third floor has a larger radius \nof coverage than one located in the basement. \nƒ \nBuilding composition. The construction materials and building design used to build \nthe facility where the broadcasting device is located will impede the signal's strength \nto varying degrees. For example, steel girders can create a dead zone in one \ndirection, while at the same time enhancing the signal in another direction. In addition, \na building's electrical wiring may inadvertently carrier a signal into other adjacent \nbuildings. \nƒ \nBackground noise. Electrical transmission pylons or other wireless networks \nlocated in the neighborhood generate background noise and thereby reduce the \neffective range of the broadcasting device. \nƒ \nWeather. Rain droplets on a facility's windows or moisture in the air can reduce the \neffective range of a broadcasting device. \nThe actual effective wireless range of wireless network segments should therefore be \nchecked to ensure that a particular wireless network implementation is not significantly \ngreater than the coverage called for by the network's design. \nDenial-of-Service (DoS) Attacks \nTechnically speaking, a denial-of-service (DoS) attack means the loss of any critical \nresource. Some examples of this attack include putting superglue into the server room's \ndoor lock, uploading to a Web server a Common Gateway Interface (CGI) script that's \ndesigned to run forever seeking the absolute value of π (slowing down the CPU), blocking \naccess to the Web site's credit-card service bureau (blocking new orders), or by creating \nhuge dummy data files (denying system log files free disk space, causing the system to \nhang). However, the most common DoS attack is an attempt to deny legitimate clients \naccess to a Web site by soaking up all the Web site's available network bandwidth or \nnetwork connections, typically by creating an inordinate number of phony Web site \n"
        },
        {
            "page_number": 71,
            "text": " \n64 \nrequests. Kelvinsky (2002) provides an extensive review of some of the most common \ntechniques and tools used to launch DoS attacks. \nA variation of a DoS attack is a distributed denial-of-service (DDoS) attack. Unlike a DoS \nattack, which is originated from a single source, a DDoS attack is launched from multiple \nsources (although it may still be orchestrated from a single point). This enables the amount \nof network traffic focused at the target network to be many times greater. \nMachiavellian DoS Attacks \nAn attacker might choose to employ a DoS attack for other less obvious reasons. DoS \nattacks are therefore not always what they seem. Here are some examples: \nƒ \nAn attacker could launch a small-scale DoS attack, but instead of using a bogus \nsource network address, he or she could use the network address of an important \nrouter/server on the Web site that is being attacked. A Web site that has already been \nattacked in a similar fashion may have installed an automated defense mechanism \nthat blocks all communication from the source of a DoS attack. Of course, if the \nnetwork address is the Web site's upstream router, the Web site may inadvertently \ncut itself off from the rest of the world! \nƒ \nDepending upon how a Web site is configured, a large DoS attack might actually \ncause a device to pause or starve to death some (or all) of the background processes \nthat are supposed to monitor and/or block intruder attacks. For instance, under \nnormal loads, a network-based intrusion detection system (IDS) may be able to detect \nemails containing viruses, but at higher network loads, the IDS may be unable to \nmonitor a sufficient number of network data packets to correctly match the virus \nagainst its virus signature database, enabling a virus to slip through unnoticed. \nƒ \nAn intruder may even use a DoS attack as a diversion measure, launching an \nobvious attack against one entry point while quietly (and hopefully unnoticed) \nattacking another entry point. Even if it is detected by an IDS, the IDS's \nwarnings/alarms may be ignored or lost due to the chaos being caused by the blatant \nDoS attack that is occurring simultaneously. \nƒ \nAn intruder that has successfully accessed a server may need to reboot the server \nbefore he or she can gain higher privileges. One way to trick a LAN administrator into \nrebooting the server (which is what the intruder wants) is to launch a DoS attack \nagainst the compromised server. (Therefore, it always pays to check that a server's \nstartup procedures have not been altered before rebooting a server, especially when \nrecovering from a DoS attack.) \nDoS Attack Countermeasures \nUnfortunately, many organizations are completely unprepared for a DoS attack, relying on \nthe get-lucky defense strategy. It may be infeasible to design a network to withstand every \npossible form of DoS attack. However, because many forms of DoS attack do have \ncorresponding countermeasures that can be put in place to avoid or reduce the severity of a \nDoS attack, it may therefore make sense to ensure that a network and its critical services \nare able to withstand the most common DoS attacks that intruders are currently using. \nDoS Attack Detection \nA DoS countermeasure may only work if the DoS attack can actually be detected. Some \nattackers may try to disguise their initial onslaught (for instance, by using multiple source \nnetwork addresses) so that the DoS attack either goes completely unnoticed by the on-duty \nsecurity staff or the deployment of any countermeasure is delayed. \n \n"
        },
        {
            "page_number": 72,
            "text": " \n65 \nEXAMPLE DOS COUNTERMEASURE \nAn example of a countermeasure that can be employed against an ICMP (ping) DoS attack \non a Web site is to have the Web site's ISP(s) throttle back the level of ICMP requests, \nreducing the amount of phony traffic that actually reaches the target Web server(s). Many \nhigh-end routers have throttling capabilities built into them. Therefore, an organization may \nwant to check with its ISP to see if the provider has this capability. If so, the organization \nshould find out what the procedures are for deploying this feature should a Web site \nbecome the subject of an ICMP DoS attack. \nTo help detect unusual rises in system utilizations (which are often the first observable \nsigns of a DoS attack), some organizations create a resource utilization baseline during a \nperiod of normal activity. Significant deviations from this norm (baseline) can be used to \nalert the system's support staff that a DoS attack may be occurring. \nWhatever DoS attack detection mechanisms have been deployed, they should be tested to \nensure that they are effective and that the on-duty security staff is promptly alerted when a \nDoS attack is initiated, especially a stealthy one. \nDoS Attack Emulation \nAlthough small-scale DoS attacks can be mimicked by simply running a DoS program from \na single machine connected to the target network, larger-scale tests that seek to mimic a \nDDoS attack may need to utilize many machines and large amounts of network bandwidth, \nand may therefore prove to be quite time consuming and resource intensive to set up and \nrun. As an alternative to using many generic servers to generate a DDoS attack, hardware \nappliances such as those listed in Table 3.15 can be used to create huge volumes of \nnetwork traffic (more than tens of thousands of network connection requests per second \nand millions of concurrent network connections) and even come with attack modules (which \nare updateable) designed to emulate the most common DoS attacks. \n \nTable 3.15: Sample List of DoS Emulation Tools and Services  \nNAME  \nASSOCIATED WEB SITE  \nFirewallStressor \nwww.antara.net  \nExodus \nwww.exodus.com  \nMercury Interactive \nwww.mercuryinteractive.com  \nSmartBits \nwww.spirentcom.com  \nWebAvalanche \nwww.caw.com  \nRather than having to set up an expensive test environment for only a few short DDoS \nattack tests (that will hopefully not have to be repeatable), another option is to use an online \nservice. (Table 3.15 lists some sample vendors that offer this service.) For a relatively small \nfee, these online vendors use their own site to generate the huge volumes of network traffic \nthat typically characterize a DDoS attack and direct this network traffic over the Internet to \nthe target network—an approach that can be much more cost effective than an organization \ntrying to build its own large-scale load generators. \nIn addition, some of the traditional load-testing tools listed in Table 3.16 can also be utilized \nto simulate DoS attacks that necessitate creating large volumes of Web site requests. \n"
        },
        {
            "page_number": 73,
            "text": " \n66 \n \nTable 3.16: Sample List of Traditional Load-Testing Tools  \nNAME  \nASSOCIATED WEB SITE  \nAstra LoadTest and LoadRunner \nwww.mercuryinteractive.com  \ne-Load \nwww.empirix.com  \nOpenSTA \nwww.sourceforge.net  \nPortent \nwww.loadtesting.com  \nQALoad \nwww.compuware.com  \nRemoteCog \nwww.fiveninesolutions.com  \nSilkPerformer \nwww.segue.com  \nTestStudio \nwww.rational.com  \nVeloMeter \nwww.velometer.com  \nWeb Application Stress Tool (WAST-\"Homer\") and \nWeb Capacity Analysis Tool (WCAT) \nwww.microsoft.com  \nWebLoad \nwww.radview.com  \nWeb Performance Trainer \nwww.webperfcenter.com  \nWebSizr \nwww.technovations.com  \nWebSpray \nwww.redhillnetworks.com  \nTable 3.17 summarizes the checks that the testing team can perform to help evaluate how \nprepared an organization is against a DoS (or DDoS) attack. \n \nTable 3.17: DoS Attack Checklist  \nYES  \nNO  \nDESCRIPTION  \n□  \n□  \nHas a documented strategy been developed to defend against DoS \nattacks? \n□  \n□  \nIs there a documented inventory of the specific DoS attacks for which \ncountermeasures have been put in place? \n□  \n□  \nHave the procedures that the on-duty security staff should follow \nwhen the network is under a DoS attack been documented? \n□  \n□  \nWhen emulating each of the defended DoS attacks, are the attacks \ndetected by the on-duty security staff? \n□  \n□  \nDoes the on-duty security staff always follow the procedures \ndocumented in the DoS policy documentation? \n□  \n□  \nIs the degradation suffered by the network and/or the services \nrunning on the network still acceptable while the network is \nexperiencing \na \nDoS \nattack \nthat \nhas \nan \nimplemented \ncountermeasure? \n \n"
        },
        {
            "page_number": 74,
            "text": " \n67 \nSummary \nThe network infrastructure that each network application resides on represents the \nelectronic foundation of the application. No matter how good an application's security \nprocedures are, the application can be undermined by vulnerabilities in the underlying \nnetwork that the application depends on for its network connectivity. This chapter has \noutlined a series of steps and techniques (summarized in Figure 3.5) that a security-testing \nteam can follow (or customize to their unique situation) to first define the scope and then \nconduct a network security-testing effort. \n \nFigure 3.5: Network security-testing approach summary.  \nOne final point is worth emphasizing: In order for the testing effort to be as comprehensive \nand systematic as possible, the security-testing team must be granted access to highly \nsensitive documentation such as a diagram depicting the network's topology. It goes \nwithout saying that the testing team should take every feasible precaution to prevent this \nsensitive information from being leaked to a potential attacker (external or internal) and that \nany test results (regardless of whether or not they demonstrate the existence of a security \nvulnerability) must be kept under lock and key and only distributed on a need-to-know \nbasis. Finding any of these artifacts could save an intruder a considerable amount of time \nand effort, and increase the possibility of a successful penetration. Additionally, the testing \nteam should be careful to clean up after themselves, making sure that once the testing is \ncomplete, any testing tools that could be utilized by an attacker are removed from all of the \ndevices they were installed upon. \n"
        },
        {
            "page_number": 75,
            "text": " \n68 \nChapter 4: System Software Security \nOverview \nThis book uses the term system software to refer to the group of commercial and open-\nsource software products that are developed and distributed by an external organization. \nThese include operating systems, database management systems, Java 2 Platform \nEnterprise Edition (J2EE) implementations, file-sharing utilities, and communication tools. \nTable 4.1 lists some specific examples of such products. \n \nTable 4.1: Sample List of System Software Products  \nNAME  \nASSOCIATED WEB SITE  \nApache \nwww.apache.org  \nLinux \nwww.redhat.com  \nNotes \nwww.lotus.com  \nMS SQL Server \nwww.microsoft.com  \npcAnywhere \nwww.symantec.com  \nWebLogic \nwww.bea.com  \nTypically, whatever Web application that an organization has deployed or will want to \ndeploy will depend upon a group of system software products. Before developing any \napplication software, an organization would be well advised to evaluate any system \nsoftware that the application is expected to utilize. Such an evaluation would ensure that \nthe planned system software and the specific installation configuration do not have any \nsignificant security issues. Determining security flaws or weaknesses early is important, as \ntrying to retrofit a set of patches or workarounds to mitigate these system software security \nvulnerabilities can cause significant reworking. For example, applications might have to be \nreconfigured, as the original ones were developed using different, often default, system \nsoftware configurations. Or perhaps, worse still, the applications would need to be ported to \na new platform, because the original platform was found to be inherently unsafe. \nThis chapter looks at the tests that should be considered to ensure that any system \nsoftware that is going to be deployed has been configured to remove or minimize any \nsecurity vulnerabilities associated with this group of software products and thereby provide \na firm foundation on which Web applications can be built. \n \nSecurity Certifications \nAlthough virtually every system software vendor will claim its product is secure, some \nproducts are designed to be more secure than others. For instance, some products \ndifferentiate the tasks that need to be performed by an administrator from those that are \ntypically only needed by a user of the system, thereby denying most users of the system \naccess to the more security-sensitive administrative functions. This is just one of the ways \nWindows NT/2000 is architected differently than Windows 9.x. \nWhen evaluating system software products for use on a Web site, an organization would \nideally want to review each proposed product's architecture to ensure that it has been \nsufficiently secure. Unfortunately, for all but the largest organizations (typically \n"
        },
        {
            "page_number": 76,
            "text": " \n69 \ngovernments), such an undertaking is likely to be cost prohibitive. To mitigate this problem, \nthe security industry has developed a common criteria for evaluating the inherent security of \nsoftware products. The goal of the common criteria is to allow certified testing labs to \nindependently evaluate, or certify, software products against an industry-standard criteria, \nthereby allowing potential users of the software to determine the level of security a product \nprovides, without each user having to individually evaluate each tool. \n \nSECURITY CERTIFICATION HISTORY \nCirca 1985, the U.S. Department of Defense (www.defenselink.mil) defined seven levels of \nsystem software security, A1, B1, B2, B3, C1, C2, and D1 (A1 being the highest level of \nsecurity), for the purpose of providing a common set of guidelines for evaluating the \nsecurity of software products from different vendors. The exact criteria used to assign \nproducts to different levels were spelled out in a document commonly referred to as the \nOrange book. During the early 1990s several European governments jointly developed a \nEuropean equivalent of the Orange book. These guidelines were called the European \nInformation Technology Security Evaluation and Certification Scheme, or ITsec (see \nwww.cesg.gov.uk/assurance/iacs/itsec/). \nBoth sets of guidelines have been superseded by a new set of general concepts and \nprinciples developed under the auspices of the International Organization for \nStandardization (www.iso.ch). This new standard (ISO 15408) is being referred to as the \nCommon Criteria for Information Technology Security Evaluation, or Common Criteria (CC) \nfor short. \nAlthough comparatively few products have currently completed the evaluation process, the \ncommon criteria is becoming more widely recognized, which in turn should lead to more \nproducts being submitted for testing. Additional information about the common criteria and \nthe status of the products that have been or are in the process of being evaluated can be \nfound at www.commoncriteria.org. \n \nPatching \nThe early versions of system software products used to support Web sites often contained \nobscure security holes that could potentially be exploited by a knowledgeable attacker. \nThanks to an army of testers that knowingly (or unknowingly) tested each new version of \nsoftware before and after its release, the later releases of these products have become \nmuch more secure. Unfortunately, more is a relative term; many of these products still have \nwell-documented security vulnerabilities that, if not patched, could be exploited by attackers \nwho have done their homework. \nIf a security issue with a particular version of a system software product exists, typically the \nproduct's end-users can't do much about it until the developers of the product (vendor or \nopen-source collaborators/distributor) are able to develop a patch or workaround. \nFortunately, most high-profile system software vendors are particularly sensitive to any \npotential security hole that their software might contain and typically develop a fix (patch) or \nworkaround very quickly. \nSystem software patches are only useful if they are actually installed. Therefore, rather than \nactually testing the product itself for as-yet-undiscovered security holes, the security-testing \nteam would be much better advised to review the work of others to determine what known \nsecurity issues relate to the products used on the Web site under testing. A common, \nmanual approach to researching known security issues is to view entries on online bug-\ntracking forums or incident response centers such as those listed in Table 4.2. In addition, \n"
        },
        {
            "page_number": 77,
            "text": " \n70 \nmany vendors also post up-to-date information on the status of known security defects \n(sometimes referred to by vendors as features) and what the appropriate fix or workaround \nis on their own Web site. \nTable 4.2: Web Sites of Bug- and Incident-Tracking Centers  \nWEB SITE NAME  \nWEB ADDRESS  \nCERT® Coordination Center \nwww.cert.org  \nComputer Incident Advisory Capability (CIAC) \nwww.ciac.org  \nComputer Security Resource Center (CSRC) \nhttp://csrc.nist.gov  \nCommon Vulnerabilities and Exposures (CVE) \nwww.cve.mitre.org  \nFederal Computer Incident Response Center \n(FedCIRC) \nwww.fedcirc.gov  \nInformation System Security \nwww.infosyssec.com  \nInternet Security Systems™ (ISS) \nwww.iss.net  \nNational \nInfrastructure \nProtection \nCenter \n(NIPC) \nwww.nipc.gov  \nNTBugtraq \nwww.ntbugtraq.com  \nPacket Storm \nhttp://packetstorm.decepticons.org  \nSystem \nAdministration, \nNetworking, \nand \nSecurity (SANS) \nwww.sans.org  \nSecurity Bugware \nwww.securitybugware.org  \nSecurityFocus (bugtraq) \nwww.securityfocus.com  \nSecurityTracker \nwww.securitytracker.com  \nVmyths.com \nwww.vmyths.com  \nWhitehats \nwww.whitehats.com  \nWindows and .NET Magazine Network \nwww.ntsecurity.net  \nIn addition, many vendors also post up-to-date information on the status of known security \ndefects (sometimes referred to by vendors as features) and what the appropriate fix or \nworkaround is on their Web site. \n \nOPEN-SOURCE VERSUS CLOSED-SOURCE DEBATE \nA debate still exists as to whether a proprietary (closed-source) product such as Windows is \nmore or less secure than an open-source product such as Linux or OpenBSD. Open-source \nadvocates claim that a product is much less likely to contain security holes when the source \ncode is tested and reviewed by hundreds of individuals from diverse backgrounds. \nHowever, proponents of the proprietary approach reason that an attacker is much less likely \nto find any security holes that might exist if they do not have access to the product's source \ncode. \n \nHOT FIXES, PATCHES, SERVICE PACKS, POINT RELEASES, AND BETA VERSIONS \n"
        },
        {
            "page_number": 78,
            "text": " \n71 \nDifferent vendors use different terms to describe their software upgrades. Often these \nnames are used to infer that different degrees of regression testing have been performed \nprior to the upgrade being released. The situation isn't helped when the vendor offers such \ngeneric advice as \"this upgrade should only be installed if necessary.\" Therefore, before \ninstalling any upgrade, try to determine what level of regression testing the vendor has \nperformed. An organization should consider running its own tests to verify that upgrading \nthe latest version will not cause more problems than it fixes. For example, an upgrade fixes \na minor security hole, but it impacts an application's performance and functionality. \nInstead of manually researching all the security holes and nuances of a system software \nproduct, the security-testing team could utilize an automated security assessment tool or \nonline service. Such a tool or service can be used to probe an individual machine or group \nof machines to determine what known security issues are present and remain unpatched. \nTable 4.3 lists some of the tools available for this task. \n \nTable 4.3: Sample List of System Software Assessment Tools  \nNAME  \nASSOCIATED WEB SITE  \nHFNetChk and Personal Security Advisor \nwww.microsoft.com  \nHotfix Reporter \nwww.maximized.com  \nInternet Scanner \nwww.iss.net  \nNessus \nwww.nessus.org  \nQuickinspector \nwww.shavlik.com  \nSecurity Analyzer \nwww.netiq.com  \nTitan \nwww.fish.com  \nWhichever approach is used, the goal is typically not to find new system software defects, \nbut to ascertain what (if any) security patches or workarounds need to be implemented in \norder to mitigate existing problems. \nFor some organizations, installing patches every couple of weeks on every machine in the \norganization may consume an unacceptable amount of resources. For instance, a risk-\nadverse organization may want to run a full set of regression tests to ensure that the \nfunctionality of any existing application isn't altered by the workaround, or that the Web \nsite's performance isn't noticeably degraded by a new patch. In some instances, a patch \nmay even turn on features that have previously been disabled or removed, or it may alter \nexisting security settings. Security policies should therefore be reviewed to ensure that they \ndescribe under what circumstances a patch or workaround should be implemented and \nwhat regression tests should be performed to ensure that any newly installed patch has not \nunknowingly changed any security configuration setting or otherwise depreciated the \ncapabilities of the Web site. \nTable 4.4 lists a series of checks that could be utilized to evaluate how well security \npatches are being implemented. \n \nTable 4.4: System Software Patching Checklist  \nYES  \nNO  \nDESCRIPTION  \n"
        },
        {
            "page_number": 79,
            "text": " \n72 \nTable 4.4: System Software Patching Checklist  \nYES  \nNO  \nDESCRIPTION  \n□  \n□  \nIs there a documented policy in place that describes under what \ncircumstances and how a security patch should be implemented? \n(This is especially important when multiple patches are to be applied, \nas the installation order may be critical.) \n□  \n□  \nHave all the known security issues for each system software product \nthat is or will be used by the Web site been researched and \ndocumented? \n(The \nresearch \nshould \ninclude \nevaluating \nany \nconsequences of installing the patch.) \n□  \n□  \nHave all the security patches deemed necessary by the documented \npolicy been obtained from a legitimate source? (It's not unheard of for \na supposed security patch to actually contain a Trojan horse.) \n□  \n□  \nHave tests been designed that can demonstrate the existence of the \nsecurity hole(s) that needs to be patched? (This is necessary if \nconfirmation is needed that the security hole has indeed been fixed \nby the correct application of the patch.) \n□  \n□  \nHave all the security patches and workarounds deemed necessary by \nthe policy been implemented on every affected machine? \n□  \n□  \nIn the event an issue is discovered with a newly installed patch, is a \nprocess in place that would enable the patch to be rolled back \n(uninstalled)? \n□  \n□  \nIs the person(s) responsible for monitoring new security issues aware \nof his or her responsibility and does he or she have the resources to \naccomplish this task? \n \nHardening \nHardening is a term used to describe a series of software configuration customizations that \ntypically remove functionality and/or reduce privileges, thereby making it harder for an \nintruder to compromise the security of a system. Unfortunately, the default installation \noptions for many system software products are usually not selected based on security \nconsiderations, but rather on ease of use, thus necessitating hardening customizations. \nTherefore, in addition to checking for known security holes, it also makes sense to \nsimultaneously check for known features that, if left unaltered, could be exploited by an \nintruder. \n \nROCKING THE BOAT \nOne of reasons that system administrators delay (or do not) apply system software patches \nis because they are afraid of \"rocking the boat.\" Because a system administrator may not \nhave time to thoroughly regression test a new patch, he or she probably fears that the new \npatch may destabilize the system. Holding off implementing the new patch until the next \nscheduled system upgrade (when the system will be rigorously tested) will allow an \norganization to find any unexpected consequences of installing the patch. Unfortunately, \nduring this period of time, the organization may be vulnerable to an attack through any of \nthe security holes that could have been filled by the uninstalled patch. \n"
        },
        {
            "page_number": 80,
            "text": " \n73 \n \nLOCKING DOWN THE OPERATING SYSTEM \nRather than requiring a system administrator to manually harden an operating system, \norganizations now offer products such as those listed in Table 4.5 that attempt to provide \nan extra level of protection to the operating system. These products often work by disabling \nor locking down all administrative-level services, which can then only be accessed using a \nsecure password administered by the protecting product. \n \nTable 4.5: Sample List of Operating System Protection Tools  \nNAME  \nASSOCIATED WEB SITE  \nBastille Linux \nwww.bastille-linux.org  \nEnGarde \nwww.engardelinux.org  \nIISLockdown \nwww.microsoft.com  \nImmunix \nwww.immunix.org  \nServerLock \nwww.watchguard.com  \nTable 4.6 is a generic list of tests that can be used to form the basis of a system software-\nhardening checklist, while Allen (2001) outlines processes for hardening several different \nplatforms. \n \nTable 4.6: System Software Hardening Checklist  \nYES  \nNO  \nDESCRIPTION  \n□  \n□  \nHave vendor- and industry-recommended hardening customizations \nbeen researched and documented for each system software product \nthat is or will be used by the Web site? \n□  \n□  \nHave all the procedures used to harden each system software \nproduct been documented? \n□  \n□  \nHave all the documented system software hardening procedures \nbeen implemented on every affected machine? \n \nMasking \nThe more information an intruder can obtain about the brand, version, and installation \noptions of any system software product installed upon the Web site (such as what brand \nand version of operating system is being used by the Web server), the easier it will be for \nthe intruder to exploit any known security holes for this particular version of the product. \nFor instance, buffer overflow attacks (discussed in more detail in Chapter 6) are typically \noperating system and architecture specific (for example, a buffer overflow attack that works \non an NT/Alpha platform is unlikely to work on a NT/Intel or UNIX/Alpha platform). \nTherefore, to exploit this kind of attack, the operating system and hardware platform must \nfirst be deduced or guessed. Additionally, when designing new exploits, authors often need \nto recreate an equivalent system software and hardware architecture environment in order \nto compile and/or test their newly discovered exploit(s). \n"
        },
        {
            "page_number": 81,
            "text": " \n74 \nGiven the usefulness of knowing this kind of information, it makes sense that an \norganization would want to minimize this knowledge. Unfortunately, many products give up \nthis kind of information all too easily. For instance, much of this information can be obtained \nvia hello (banner) or error messages that the product sends by default when somebody \ntries to initiate a connection with it. Intruders trying to deduce the brand and version of a \nproduct will often use a technique called banner grabbing to trick a machine into sending \ninformation that uniquely identifies the brand and version of the products being used by the \nWeb site. To reduce this information leakage, many organizations choose to mask their \nWeb sites, replacing these helpful default messages with legal warnings, blank or \nuninformative messages, or false banners that match the default response from a \ncompletely different brand of system software and therefore hopefully cause an intruder to \nwaste his time using an ineffective set of exploits. \nA security tester shouldn't have to rely on manual efforts (such as the ones illustrated in the \n\"Banner Grabbing\" sidebar). Rather, several tools now exist that will attempt to identify \n(fingerprint) a target by running a series of probes. Some even offer features designed to \nmake this activity less likely to be noticed by any intrusion-detection system (IDS) that might \nbe installed on the target Web site and tip off an organization that its Web site was being \nfingerprinted (also known as enumerated). Table 4.7 lists some sample fingerprinting tools \nand services, while Scambray 2001 provides more detailed information on the techniques \nused by intruders to fingerprint a target, and Klevinsky (2002) provides guidance on how to \nuse many of the fingerprinting tools used by penetration testers and intruders alike. \n \nTable 4.7: Sample List of Fingerprinting Tools and Services  \nNAME  \nASSOCIATED WEB SITE  \nCerberus Internet Scanner \nwww.cerberus-infosec.co.uk  \nCheops \nwww.marko.net  \nHackerShield \nwww.bindview.com  \nNetcat \nwww.atstake.com  \nNmap \nwww.insecure.org  \nSuper Scan/Fscan \nwww.foundstone.com  \nWhat's That Site Running? \nwww.netcraft.com  \nUnfortunately, it's not always possible to completely mask the identify of an operating \nsystem due to the Transmission Control Protocol/Internet Protocol (TCP/IP) being \nimplemented slightly differently by various operating system vendors. For example, TCP \nfeatures and packet sequence numbering may differ among vendors. However, many \nnovice attackers and some fingerprinting tools may still be fooled or at least delayed by a \nfalse banner. \nIf an organization has decided to use a nondescriptive legal warning or false banners on \nsome or all of their machines, then these machines should be checked to ensure that this \nrequirement has been implemented correctly. In the case of false banners designed to \ndeceive an intruder's fingerprinting effort, an assessment of the effectiveness of the \ndeception can be made by using several of the automated probing tools to see if they can \nsuccessfully see through this ploy. Table 4.8 summarizes these checks. \n \n"
        },
        {
            "page_number": 82,
            "text": " \n75 \nTable 4.8: System Software Masking Checklist  \nYES  \nNO  \nDESCRIPTION  \n□  \n□  \nIs there a documented policy in place that describes under what \ncircumstances a default banner should be replaced with a blank legal \nwarning or false banner? \n□  \n□  \nIf a legal warning is to be used, has it been approved by the legal \ndepartment? \n□  \n□  \nHave all the banner modifications deemed necessary by the policy \nbeen implemented on every affected machine? \n□  \n□  \nIf false banners are to be used, are they deceptive enough to trick a \nsignificant number of automated probing tools? \n \nBANNER GRABBING \nIt is very easy to identify the version of system software being used by a Web site and \nthereby hone in on known bugs or features with this specific product version. For example, \nthe following error messages were generated when an attempt was made to start a Telnet \nsession to two different Web sites using a port number not normally used by the Telnet \napplication (if installed, the Telnet application is normally configured to communicate on \nport 23). \nFrom a command-line prompt, enter telnet www.wiley.com 80 (80 is the port number used \nby HTTP) or telnet www.wileyeurope.com 25 (25 is the port number used by SMTP). \nExample 1 \nC:\\telnet www.wiley.com 80 \n \nHTTP/1.1 400 Bad Request \nServer: Netscape Enterprize/3.6 SP3 \n \nYour browser sent a message this server could not understand. \n \nExample 2 \nC:\\telnet www.wileyeurope.com 25 \n \n220 xysw31.hosting.wileyeurope.com ESMTP server (Post.Office v3.5.3 \nrelease 223 ID# 0-83542U500L100S0V35) ready Tue, 16 Jan 2002 \n17:58:18 - \n0500 \nBoth Telnet connections should subsequently fail (it may be necessary to hit Enter a few \ntimes), because Telnet is not configured to work on either of the requested port numbers, \nbut not before the target machine sends an error message that identifies the brand and \nversion of system software that is currently running. \n \n"
        },
        {
            "page_number": 83,
            "text": " \n76 \nServices \nThis book will use the term service to describe all the system software services, processes, \nor daemons (in UNIX terminology) installed on a machine that can communicate with the \nnetwork it is attached to. Before a service can communicate over the network, it must first \nbe bound to one or more network interface cards (NICs) and communication channels \n(ports). \nWhenever a service is started on a machine, the operating system will typically grant the \nservice the same security privileges the user account that initiated the service had. \nUnfortunately, if a service were to be tricked into executing malicious commands, these \nundesirable instructions would be executed using the same privileges that the service \ninherited from the account that owns this service. For example, if the Web server service \nwas running with administrative (or in UNIX lingo, root) privileges, an intruder could be able \nto trick the Web server into emailing the intruder the operating system's password file (a file \nthat normally only the administrator can access). Had the Web server service been running \nwith a lower privileged account, then chances are that the Web server service itself would \nhave been refused access to this system file by the operating system. It is therefore \nimportant to check that any service running on a machine is only granted the minimum \nprivileges needed to perform its legitimate functions and any unneeded services are \ndisabled (or ideally uninstalled). \nGenerally speaking, common network services such as the Hypertext Transfer Protocol \n(HTTP), Finger, or the Simple Mail Transfer Protocol (SMTP) use predefined (or well-\nknown) port numbers. These numbers are assigned by the Internet Assigned Numbers \nAuthority (IANA, www.iana.org), an independent organization with the aim of minimizing \nnetwork conflicts among different products and vendors. Table 4.9 lists some sample \nservices and the port numbers that the IANA has reserved for them. \n \nTable 4.9: Sample IP Services and Their Assigned Port Numbers  \nPORT NUMBER  \nSERVICE  \n7 \nEcho \n13 \nDayTime \n17 \nQuote of the Day (QOTD) \n20 and 21 \nFile Transfer Protocol (FTP) \n22 \nSecure Socket Shell (SSH) \n23 \nTelnet \n25 \nSMTP \n53 \nDomain Name System (DNS) \n63 \nWhois \n66 \nSQL*net (Oracle) \n70 \nGopher \n79 \nFinger \n80 \nHTTP \n"
        },
        {
            "page_number": 84,
            "text": " \n77 \nTable 4.9: Sample IP Services and Their Assigned Port Numbers  \nPORT NUMBER  \nSERVICE  \n88 \nKerberos \n101 \nHost Name Server \n109 \nPost Office Protocol 2 (POP2) \n110 \nPost Office Protocol 3 (POP3) \n113 \nIDENT \n115 \nSimple File Transfer Protocol (SFTP) \n137, \n138, \nand \n139 \nNetBIOS \n143 \nInternet Message Access Protocol (IMAP) \n161 and 162 \nSimple Network Management Protocol (SNMP) \n194 \nInternet Relay Chat (IRC) \n443 \nHypertext Transfer Protocol over Secure Socket Layer \n(HTTPS) \nFor an intruder to communicate with and/or try to compromise a machine via a network \nconnection, the intruder must utilize at least one port. Obviously, the fewer ports that are \nmade available to an intruder, the more likely it is that the intruder is going to be detected. \nIn just the same way, the fewer the number of doors and windows a bank has, the easier it \nis for the bank to monitor all of its entrances and the less likely it is that an intruder would be \nable to enter the bank unnoticed. Unfortunately, a single NIC could have up to 131,072 \ndifferent ports for a single IP address, 65,536 for TCP/IP, and another 65,536 for User \nDatagram Protocol (UDP)/IP. (Appendix A describes IP, TCP, and UDP in more detail.) \n \n \nPORT NUMBER ASSIGNMENTS \nPort numbers 0 through 1023 are typically only available to network services started by a \nuser with administrator-level privileges, and are therefore sometimes referred to as \nprivileged ports. An intruder who has only been able to acquire a nonadministrator account \non a machine may therefore be forced to utilize a nonprivileged port (1024 or higher) when \ntrying to communicate to the compromised machine. \nThe set of nonprivileged port numbers (1024 to 65535) has been divided into two groups: \nthe registered group (1024 to 49151) and the private (dynamic) group (49152 to 65535). \nThe registered ports differ from the well-known ports in that they are typically used by \nnetwork services that are being executed using nonadministrator level accounts. The \nprivate group of ports is unassigned and is often used by a network service that does not \nhave a registered (or well-known) port assigned to it, or by a registered network service that \ntemporarily needs additional ports to improve communication. In such circumstances the \nnetwork service must first listen on the candidate private port and determine if it is already \nin use by another network service. If the port is free, then the network service will \ntemporally acquire (dynamically assign) this port. \n \n"
        },
        {
            "page_number": 85,
            "text": " \n78 \nOnce a port is closed, any request made to a machine via the closed port will result in a \n\"this port is closed\" acknowledgment from the machine. A better defensive strategy is to \nmake the closed port a stealth port; a request to a stealth port will not generate any kind of \nacknowledgement from the target machine. This lack of acknowledgement will typically \ncause the requesting (attacker's) machine to have to wait until its own internal time-out \nmechanism gives up waiting for a reply. The advantage of a stealth port over a closed port \nis that the intruder's probing efforts are going to be slowed, possibly generating frustration, \nand potentially causing the intruder to go and look elsewhere for more accommodating \ntargets. \nWhile checking to see whether ports are open can be performed manually by logging on to \neach machine and reviewing the services running (as depicted in Figure 4.1), it's not always \nclear what some of the services are or which ports (if any) they are using, especially if more \nthan one NIC is installed. \n \nFigure 4.1: List of active processes running.  \nFortunately, a number of easy-to-use tools can automate this test; these tools are referred \nto as port scanners and often come as part of a suite of security-testing tools. Chirillo \n(2001) goes into considerable depth on securing some of the most commonly used ports \nand services, while Klevinsky (2002) provides an overview of many of the tools that can be \nused to automated a port scan. \n \n \nNote \nFoundstone (www.foundstone.com) provides a utility Fport, which can be \nused to map running services to the ports that they are actually using. \n \nREMAPPING SERVICES \n"
        },
        {
            "page_number": 86,
            "text": " \n79 \nIf a potentially dangerous service such as Telnet (port 23) absolutely needs to be made \navailable to external sources, the local area network (LAN) administrator may decide to try \nand hide the service by remapping it to a nonprivileged port (that is a port number above \n1023), where it is less likely (but still possible) for an intruder to discover this useful service. \nIf the LAN administrator is using such a technique, make sure that any port-scanning tool \nused for testing is able to detect these remapped services. \n \nDUAL NICS \nA machine that has two NICs can potentially have different services running on each card. \nFor instance, a LAN administrator may have enabled the NetBIOS service on the inward-\nfacing side of a Web server to make file uploads easier for the Webmaster, but disabled it \non the outward-facing side to stop intruders from uploading their graffiti or hacking tool of \nchoice. Although increasing the complexity of a network (which increases the risk of human \nerror), using multiple NICs for some of the servers can potentially improve security and \nperformance, and add additional configuration flexibility. \n \nEXAMPLE PORT SCAN \nIt appears from the following sample nmapNT port scan report that this machine is running \nquite a few services, several of which could potentially be used to compromise this \nmachine. nmap is a UNIX-based tool originally written by Fyodor. nmapNT is a version of \nnmap that was ported to the Windows platform by eEye Digital Security (www.eeye.com). \nInteresting ports on www.wiley.com (123.456.789.123): \n(The 1508 ports scanned but not shown below are in state: closed) \nPort State      Service \n21/tcp     open        ftp \n25/tcp     open        smtp \n79/tcp     open        finger \n80/tcp     open        http \n81/tcp     open        hosts2-ns \n106/tcp    open        pop3pw \n110/tcp    open        pop-3 \n135/tcp    open        loc-srv \n280/tcp    open        http-mgmt \n443/tcp    open        https \n1058/tcp   open        nim \n1083/tcp   open        ansoft-lm-1 \n1433/tcp   open        ms-sql-s \n4444/tcp   open        krb524 \n5631/tcp   open        pcanywheredata \nTables 4.10 and 4.11 list some sample port-scanning tools and services. \n \nTable 4.10: Sample List of Port-Scanning Tools  \n"
        },
        {
            "page_number": 87,
            "text": " \n80 \nNAME  \nASSOCIATED WEB SITE  \nCerberus Internet Scanner (formally NTInfoScan or \nNTIS) \nwww.cerberus-\ninfosec.co.uk  \nCyperCop Scanner \nwww.nai.com  \nFirewalk \nwww.packetfactory.net  \nHackerShield \nwww.bindview.com  \nHostscan \nwww.savant-software.com  \nInternet Scanner \nwww.iss.net  \nIpEye/WUPS \nwww.ntsecurity.nu  \nNessus \nwww.nessus.org  \nNetcat \nwww.atstake.com  \nNetcop \nwww.cotse.com  \nNetScan Tools \nwww.nwpsw.com  \nNmap \nwww.insecure.org  \nNmapNT \nwww.eeye.com  \nSAINT/SATAN \nwww.wwdsi.com  \nSARA \nwww.www-arc.com  \nScanport \nwww.dataset.fr  \nStrobe \nwww.freebsd.org  \nSuper Scan/Fscan \nwww.foundstone.com  \nTwwwscan \nwww.search.iland.co.kr  \nWhisker \nwww.wiretrip.net  \nWinscan \nwww.prosolve.com  \n \nTable 4.11: Sample List of Port-Scanning Services  \nNAME  \nASSOCIATED WEB SITE  \nShields Up \nwww.grc.com  \nMIS CDS \nwww.mis-cds.com  \nSecurityMetrics \nwww.securitymetrics.com  \nSymantec \nwww.norton.com  \nA port scan should be performed against each machine that forms part of the Web site. \nIdeally, this scan should be initiated from a machine on the inside of any installed firewall, \nsince an external port scan won't be able to tell if a service is disabled or if an intermediate \nfirewall is blocking the request. The results of the port scan should be compared to the \nservices that are absolutely required to be running on this machine and any \nadditional/unexpected services should be either disabled (and, if possible, uninstalled) or \njustified. \n"
        },
        {
            "page_number": 88,
            "text": " \n81 \nA more stringent check would be to ensure that any unused ports were not only closed, but \nalso configured to be stealthy and not provide any information on their status. \nUnfortunately, stealthy ports, while potentially slowing down an attacker's scan, may also \nslow down the testing team as they attempt to scan the machine for unnecessary services. \nTable 4.12 summarizes these checks. \n \nTable 4.12: System Software Services Checklist  \nYES  \nNO  \nDESCRIPTION  \n□  \n□  \nHas each machine been reviewed and have any unnecessary \nservices been stopped and, if possible, uninstalled? \n□  \n□  \nAre the services that are active run under the lowest privileged \naccount possible? \n□  \n□  \nHas each machine been configured to respond in a stealthy manner \nto requests directed at a closed port? \n \nREMOTE-ACCESS SERVICES \nPerhaps one of the most popular side-entry doors available to attackers, remote-access \nutilities (such as Symantec's PCAnywhere, Windows's RAS, or UNIX's SSL/RSH) provide \nlegitimate users with easy access to corporate resources from geographically remote \nlocations. (The system administrator no longer has to make 3 A.M. trips into the office to \nreboot a server.) \nUnfortunately, these convenient applications have two fundamental vulnerabilities. The first \nrisk is that the client machine (for instance, a traveling salesperson's laptop) could be \nstolen, depending upon how the client is authenticated. The theft could potentially use this \nstolen machine to gain direct access to the corporate network by simply turning the \nmachine on and double-clicking an icon on the desktop. The second risk is that the server-\nside component of the service running on the corporate network does a poor job of \nauthenticating the client. Here the requester of the service is wrongly identified as a \nlegitimate user and not an intruder trying to gain unauthorized access, especially when \naccess is via an unsanctioned modem installed on a machine behind the corporate firewall. \n \nDirectories and Files \nEach individual directory (or folder in Windows terminology) and file on a machine can have \ndifferent access privileges assigned to different user accounts. In theory, this means each \nuser can be assigned the minimum access privileges they need in order to perform their \nlegitimate tasks. Unfortunately, because maintaining Draconian directory and file privileges \nis often labor intensive, many machines typically enable users (and the services that they \nhave spawned) more generous access than they actually need. \nTo reduce the likelihood of human error when assigning directory and file access privileges, \nsome LAN administrators will group files together based on their access privilege needs. \nFor example, programs or scripts that only need to be granted execute access could be \nplaced in a directory that restricts write access, thereby inhibiting an intruder's ability to \nupload a file into a directory with execute privileges. \nMany products will by default install files that are not absolutely necessary; vendor demos \nand training examples are typical. Since some of these unneeded files may contain \n"
        },
        {
            "page_number": 89,
            "text": " \n82 \ncapabilities or security holes that could be exploited by an intruder, the safest approach is \nto either not install them or promptly remove them after the installation is complete. \nIntruders are particularly interested in directories that can be written to. Gaining write and/or \nexecute access to a directory on a target machine, even a temp directory that does not \ncontain any sensitive data, can be extremely useful. An intruder looking to escalate his or \nher limited privileges will often need such a resource to upload hacking tools (rootkits, \nTrojan horses, or backdoors) on to the target machine and then execute them. \nFor an intruder to gain access to a directory, two things must happen. First, the intruder \nmust determine the name and directory path of a legitimate directory, and second, the \nintruder must determine the password used to protect the directory (the topic of the next \nsection of this chapter). In order to reference the target directory, the intruder must figure \nout or guess the name and directory path of a candidate directory. This is an easy step if \ndefault directory names and structures are used, or the intruder is able to run a find utility \nagainst the target machine. \nIn an effort to supplement the built-in file security offered by an operating system, several \nproducts now provide an additional level of authorization security. Typically, these products \nprovide directory and file access using their own proprietary access mechanisms, thereby \npotentially mitigating any security hole or omission that could be exploited in the underlying \noperating system. Table 4.13 lists some of these products. \n \nTable 4.13: Sample List of Directory and File Access Control Tools  \nNAME  \nASSOCIATED WEB SITE  \nArmoredServer \nwww.armoredserver.com  \nAppLock \nwww.watchguard.com  \nAppShield \nwww.sanctuminc.com  \nAuthoriszor 2000 \nwww.authoriszor.com  \nEntercept \nwww.entercept.com  \nInterDo \nwww.kavado.com  \nPitBull LX \nwww.argus-systems.com  \nSecureEXE/SecureStack \nwww.securewave.com  \nStormWatch \nwww.okena.com  \nVirtualvault \nwww.hp.com  \n \nFILE-SEARCHING TECHNIQUE EXAMPLE \nA complete listing of all the files present in a Web server's directory can be easily obtained \nif the webmaster has not disabled a Web server's automatic directory service or redirected \nsuch requests to a default resource (such as a Web page named index.html). Simply \nadding a trailing forward slash (/) to the end of a URL entered via a browser's URL entry \nline (such as http://www.wiley.com/cgi-bin/) would display the entire contents of the \ndirectory. \n \nDIRECTORY-MAPPING TECHNIQUE EXAMPLE \n"
        },
        {
            "page_number": 90,
            "text": " \n83 \nThe following is a simple exploit intended to show how easy it could be for an Intruder to \nmap out a directory structure. A feature with early versions of Microsoft IIS can display the \ndirectory \nstructure \nof \na \nWeb \nsite \nas \npart \nof \nan \nerror \nmessage. \nEntering \nwww.wiley.com/index.idc from a browser's URL entry line would result in a response such \nas this one: \nError Performing Query \nThe query file F:\\wwwroot\\primary\\wiley\\index.idc could not be \nopened. \nThe file may not exist or you may have insufficient permission to \nopen \nthe file. \nRegardless of any third-party access control products that an organization may have \ndeployed, the directory and file permissions for critical machines should still be checked to \nensure that the permissions are no more liberal than they have to be. Fortunately, this \npotentially tedious manual task can to a large degree be automated using either \ncommercial tools designed to find permission omissions or tools originally designed for \nattackers with other intentions. Table 4.14 lists some sample file-share-checking tools and \nTable 4.15 offers a checklist for checking the security of directories and files. \n \nTable 4.14: Sample List of File-Share-Scanning Tools  \nNAME  \nASSOCIATED WEB SITE  \nLegion \nwww.hackersclub.com  \nNbtdump \nwww.atstake.com  \nNetBIOS Auditing Tool (NAT) \nwww.nmrc.org  \nIP Network Browser \nwww.solarwinds.net  \nWinfo \nwww.ntsecurity.nu  \n \nTable 4.15: System Software Directories and Files Checklist  \nYES  \nNO  \nDESCRIPTION  \n□  \n□  \nIs there a documented policy in place that describes how directory \nand file privileges are assigned? \n□  \n□  \nIs the directory structure on each machine appropriate? For example, \nare so many directories being used that administrative human errors \nare likely? Have read-only files been separated from execute-only \nfiles? \n□  \n□  \nHave only the minimum access privileges been assigned to each \nuser account? \n□  \n□  \nHave services running on each machine been reviewed to make sure \nthat any features that might give away information on the machine's \ndirectory structure have been disabled? An example would be a Web \nserver's automatic directory service. \n□  \n□  \nUsing a file share scanner located inside the firewall, can any \ninappropriate file shares be detected? \n"
        },
        {
            "page_number": 91,
            "text": " \n84 \n \nUserIDs and Passwords \nIn most organizations, the assets on a network that could be of any use to an intruder are \ntypically protected using combinations of userIDs and passwords. An intruder is therefore \ngoing to be forced to try and guess or deduce a useful combination. However, not all \nuserID/password combinations are created equal; some are more useful than others. An \nintruder would ideally like to capture the account of an administrator, allowing him or her to \ndo pretty much anything he or she wants on any machine that the captured account has \nadministrative rights on. If the administrator's account proves to be unobtainable but a lower \nprivileged account is susceptible, an experienced attacker may still be able to manipulate \nan improperly configured operating system into granting administrative privileges, a \ntechnique often referred to as account escalation. Care should therefore be taken to ensure \nthat not only are the administrators' userIDs and passwords sufficiently protected, but also \nany lower-level accounts. \nWEAK PASSWORD PROTECTION EXAMPLE \nSome Web servers provide the Webmaster with the ability to require client authentication \n(such as the .htaccess and .htgroup files that can be placed inside an Apache Web server \ndirectory) before displaying the contents of a directory. Upon requesting a protected \ndirectory, the Web server sends a request to the browser, which results in the browser \ndisplaying a simple userID/password pop-up window. Unfortunately, the data sent back to \nthe server using this method isn't encrypted (it uses a base 64 conversion algorithm to \nencode the data) and is therefore extremely easy for an eavesdropper to decode. Web \nserver client authentication should therefore not be relied upon to protect the contents of a \ndirectory. \nSome system software products use weak or no encryption to store and/or transmit their \nuserIDs and passwords from the client to the server component of the product, affording an \neavesdropper with the chance to capture unencrypted or easily decipherable passwords. If \nthe same password used for this service is the same as the password used for an \nadministrative-level account, learning these weak passwords may not only allow an intruder \nto manipulate the service, but also compromise additional resources. \nAlthough an attacker would like to compromise a single machine, compromising several \nmachines is definitely more desirable. This can happen relatively quickly if other machines \n(or even entire networks) have previously been configured to trust the compromised \nadministrator account. Alternatively, the LAN administrator may have used the same \npassword for several administrative accounts, thereby making network administration \neasier, but also increasing the probability that if the password on one machine is deduced, \nthe entire network may be compromised. \nEven if different userIDs and passwords are used for each local administrative account and \nthese accounts are granted limited trusts, an entire network may still be compromised if an \nintruder can get past the security of the machine used by the network for network security \nauthentication, that is, the network controller. Capturing the network controller (or backup \ncontroller) allows an intruder complete access to the entire network and possibly any other \nnetwork that trusts any of the accounts on the compromised network. Given the risk \nattached with compromising an administrator account on a network controller, many LAN \nadministrators choose to use exceptionally long userIDs and passwords for these critical \naccounts. \nOne of the leading causes of network compromises is the use of easily guessable or \ndecipherable passwords. It is therefore extremely important that an organization defines \n"
        },
        {
            "page_number": 92,
            "text": " \n85 \nand (where possible) enforces a password policy. When defining such a policy, an \norganization should consider the trade-off between the relative increase in security of using \na hard-to-crack password with the probable increase in inconvenience. Policies that are \ndifficult to follow can actually end up reducing security. For example, requiring users to use \na long, cryptic password may result in users writing down their passwords (sometimes even \nputting it on a post-it note on their monitor), making it readily available to a potential \nattacker walking by. Requiring users to frequently change their password may result in \nsome users using unimaginative (and therefore easily predictable) password sequences, \nsuch as password1, password2, and password3. Even if the access control system is smart \nenough to deduce blatant sequences, users may still be able to craft a sequence that is \neasy for them to remember but still acceptable to the access control system, such as \npassjanword, passfebword, and passmarword. As the following sections demonstrate, an \nintruder could acquire a userID/password combination in several distinct ways. \n \nSINGLE SIGN-ON \nSingle sign-on (SSO) is a user authentication technique that permits a user to access \nmultiple resources using only one name and password. Although SSO is certainly more \nuser-friendly than requiring users to remember multiple userID's and passwords, from a \nsecurity perspective it is a double-edged sword. Requiring users to only remember one \nuserID and password should mean that they are more willing to use a stronger and hence \nmore secure (albeit harder to remember) password, but on the other hand, should this \nsingle password be cracked, an intruder would have complete access to all the resources \navailable to the compromised user account, a potentially disastrous scenario for a highly \nprivileged account. \nManual Guessing of UserIDs and Passwords \nTypically easy to attempt, attackers simply guess at userID/password combinations until \nthey either get lucky or give up. This approach can be made much more successful if the \nintruder is able to first deduce a legitimate userID. \nObtaining a legitimate userID may not be as hard as you might think. When constructing a \nuserID, many organizations use a variation of an employee's first name and last name. A \nsample userID format can be obtained, for example, by viewing an email address posted on \nthe organization's Web site. Discovering the real name of a LAN administrator may be all \nthat is needed to construct a valid userID, and that information is easily obtained by \nacquiring a copy of the organization's internal telephone directory. Or perhaps an intruder \ncould look up the technical support contact posted by a domain name registrar to find a \ndomain name owned by the organization. \nMany system software products are initially configured with default userIDs and passwords; \nit goes without saying that these commonly known combinations should be changed \nimmediately (www.securityparadigm.com even maintains a database of such accounts). \nWhat is less well known is that some vendors include userIDs and passwords designed to \nenable the vendor to log in remotely in order to perform routine maintenance, or in some \ncases the organization's own testing team may have created test accounts intended to help \ndiagnose problems remotely. If any of the products installed at an organization, such as \nfirewalls, payroll packages, customer relationship management systems, and so on, use \nthis feature, the organization should consider whether or not this remote access feature \nshould be disabled, or at the very least the remote access password changed. \n \nEXAMPLE PASSWORD POLICY GUIDELINES \n"
        },
        {
            "page_number": 93,
            "text": " \n86 \nEnforceable Guidelines \nƒ Minimum length of x characters \nƒ Must not contain any part of userID or user's name \nƒ Must contain at least one character from at least four of the following categories: \no Uppercase \no Lowercase \no Numeral \no Nonalphanumeric, such as !@#$%^&*() \no Nonvisual, such as control characters like carriage return <CR> \nƒ Must be changed every X number of weeks \nƒ Must not be the same as a password used in the last X generations \nƒ Account is locked out for X minutes after Y failed password attempts within Z period \nof time \nHard-to-Enforce Guidelines \nƒ Do not use words found in an English (or another language) dictionary \nƒ Do not use names of family, friends, or pets (information often known by coworkers) \nƒ Do not use easy-to-obtain personal information such as parts of a \no Mailing address \no Social security number \no Telephone number \no Driving license number \no Car license plate number \no Cubical number \n \nUSERID ADVERTISEMENTS \nIn an effort to improve customer relations, many organizations have started to advertise the \nemail address of a senior employee, such as \"Please email the branch manager, Jon \nMedlin, at <jmedlin@wiley.com> with any complaints or suggestions for improvement.\" \nAlthough providing direct access to senior management may help improve customer \ncommunications, if the format used for the email address is the same as the format used for \nthe user's ID, the organization may also be inadvertently providing a potential attacker with \nuserIDs for accounts with significant privileges. \n \nIf an intruder is truly just guessing at passwords, then perhaps the easiest way to thwart this \napproach is to configure the system to lock out the account under attack after a small \nnumber of failed login attempts. Typically, lockout periods range from 30 minutes to several \nhours, or in some cases even require the password to be reset. \n \nTHE NULL PASSWORD \nSome organizations have adopted an easy-to-administer policy of not assigning a password \nto a user account until the first time it is used, the password being assigned by the first \nperson to log in to the account. \nObviously, accounts that have no (or null) password are going to be extremely easy for an \nattacker to guess and should therefore be discouraged. \n"
        },
        {
            "page_number": 94,
            "text": " \n87 \nAutomated Guessing of UserIDs and Passwords \nSeveral tools now exist that can be used to systematically guess passwords; these tools \ntypically employ one (or both) of two basic guessing strategies. The quickest strategy is to \nsimply try a list of commonly used passwords. Most of the tools come with lists that can be \nadded to or replaced (particularly useful if the passwords are expected to be non-English \nwords). Hackersclub (www.hackersclub.com) maintains a directory of alternative wordlists. \nThe second approach is to use a brute-force strategy. As the name implies, a brute-force \napproach does not try to get lucky by only trying a comparative handful of passwords; \ninstead, it attempts every single possible combination of permissible characters until it \ncracks the password. The biggest drawback with a brute-force approach is time. The better \nthe password, the longer it will take a brute-force algorithm to crack the password. Table \n4.16 lists some sample password-cracking tools. \n \nTable 4.16: Sample List of Password-Deciphering/Guessing/Assessment Tools  \nNAME  \nASSOCIATED WEB SITE  \nBrutus \nwww.antifork.org/hoobie.net  \nCerberus Internet Scanner \nwww.cerberus-infosec.co.uk  \nCrack \nwww.users.dircon.co.uk/~crypto  \nCyberCop Scanner[a]  \nwww.nai.com  \n(dot)Security \nwww.bindview.com  \nInactive Account Scanner \nwww.waveset.com  \nLegion and NetBIOS Auditing Tool (NAT) \nwww.hackersclub.com  \nLOphtcrack \nwww.securitysoftwaretech.com  \nJohn the Ripper, SAMDump, PWDump, \nPWDump2, PWDump3 \nwww.nmrc.org  \nSecurityAnalyst \nwww.intrusion.com  \nTeeNet \nwww.phenoelit.de  \nWebCrack \nwww.packetstorm.decepticons.org  \n[a]Effective July 1, 2002, Network Associates has transitioned the CyberCop product line \ninto maintenance mode. \nSuppose the intruder intends to use an automated password tool remotely—and this \napproach is thwarted by locking out the account after a small number of failed attempts. To \nget in, the intruder would need to obtain a copy of the password file. But once that was in \nhand, the intruder could then run a brute-force attack against the file. Using only modest \nhardware resources, some tools can crack weak passwords within a few hours, while \nstronger passwords will take much longer. Skoudis (2001) provides additional details on \nhow to attack password files for the purpose of determining just how secure sets of \npasswords are. \nIn theory, no matter how strong a password is, if the file that contains the password can be \nacquired, it can eventually be cracked by a brute-force attack. However, in practice, using \nlong passwords that utilize a wide variety of characters can require an intruder to spend \n"
        },
        {
            "page_number": 95,
            "text": " \n88 \nseveral weeks (or even months) trying to crack a file, a fruitless effort, if the passwords are \nroutinely changed every week. \nOne approach to evaluating the effectiveness of the passwords being selected is to ask the \nLAN administrator to provide you with legitimate copies of all the password files being used \non the production system. From a standalone PC (that is not left unattended), an attempt is \nmade to crack each of these passwords using a password-guessing tool that uses as input \na list of commonly used words (this file is obviously language specific). Any accounts that \nare guessed should be immediately changed. \nA second time-consuming test would be to run a brute-force attack against each of these \nfiles, since in theory any password could be deciphered given enough time; this test should \nbe time-boxed. For example, any password that can be cracked within 24 hours using \nmodest hardware resources should be deemed unacceptable. \n \nPASSWORD FILE NAMES \nPassword files on UNIX systems are generally named after some variation of the word \npassword, such as passwd. The Windows NT/2000 family of systems name their password \nfiles SAM (short for Service Account Manager). \nParticular care should be taken to destroy all traces of the copied password files, temporary \nfiles, and generated reports once the testing is complete, least these files fall into the wrong \nhands. \nEven if only strong passwords are used, it still makes sense to try and ensure that these \npassword files are not readily available to an intruder. An organization should therefore \nconsider designing a test to see if an unauthorized person can acquire a copy of these files. \nAlthough security may be quite tight on the production version of these files, it's quite \npossible that backup files either located on the machine itself or offsite are quite easily \naccessible. For example, the file used by Windows NT/2000 to store its passwords is \nprotected by the operating system while it is running. However, the operating system also \nautomatically creates a backup copy of this file, which may be accessible. A simple search \nof a machine's hard drive using *sam*.* should locate the production and backup version(s) \nof this file. \nA less obvious place to find clues to valid passwords is in the log files that some system \nsoftware products use to store failed login attempts. For example, knowing that user Tim \nWalker failed to login with a password of margarey may be all an intruder needs to know in \norder to deduce the valid password is margaret. Such log files, if used, should therefore be \nchecked to ensure that the failed password is also encrypted to stop an intruder from \nviewing these useful clues. \nGaining Information via Social Engineering \nCovered in more detail in Chapter 7, social engineering refers to techniques used by \nintruders to trick unsuspecting individuals into divulging useful information. A classic \nexample is that of an intruder calling an organization's help desk and asking them to reset \nthe password of the employee the intruder is pretending to be. \nDisgruntled Employees Committing Illicit Acts \nAlthough many organizations seriously consider the risk of a trusted employee taking \nadvantage of his or her privileged position to commit (or attempt to commit) an illicit act (a \n"
        },
        {
            "page_number": 96,
            "text": " \n89 \ntopic covered in more detail in Chapter 7), others chose to ignore this possibility. Obvious \nprecautions include ensuring that employees are only granted access to resources they \nabsolutely need, and accounts used by former employees are deactivated as soon as (or \nbefore) they leave. \nTable 4.17 summarizes some of the checks that should be considered when evaluating the \nprotection afforded to a system's userIDs and passwords. \n \nTable 4.17: System Software UserIDs and Passwords Checklist  \nYES  \nNO  \nDESCRIPTION  \n□  \n□  \nIs there a documented policy in place that describes how userIDs and \npasswords are assigned, maintained, and removed? \n□  \n□  \nWhen an employee leaves (voluntarily or involuntarily), are his or her \npersonal user accounts deactivated and are the passwords changed \nin a timely manner for any shared accounts that he or she had \nknowledge of? \n□  \n□  \nAre the procedures for handling forgotten and compromised \npasswords always followed? \n□  \n□  \nAre security access logs monitored for failed logins? For instance, \nhow long or how many tries does it take before someone responds to \na legitimate account using invalid passwords? \n□  \n□  \nDoes the system lock out an account for X minutes after Y failed \npassword attempts within Z period of time? \n□  \n□  \nAre different administrative userIDs and/or passwords used for each \nmachine? \n□  \n□  \nHave all default accounts been removed, disabled, or renamed, \nespecially any guest accounts? \n□  \n□  \nHave all remote access accounts been disabled? Or at least have \ntheir passwords been changed? \n□  \n□  \nAre variations of people's names not used when assigning userIDs? \n□  \n□  \nDo none of the critical accounts use common (and therefore easily \nguessable) words for passwords? \n□  \n□  \nAre hard-to-guess or decipher passwords (as defined by the \norganization's password policy) used for all critical accounts? \n□  \n□  \nAre the details of any failed login attempts sufficiently protected from \nunauthorized access? \n \nUser Groups \nMost system software products support the concept of user groups. Instead of (or in \naddition to) assigning individual user accounts system privileges, privileges are assigned to \nuser groups; each user account is then made a member of one (or more) user groups and \nthereby inherits all the privileges that have been bestowed upon the user group. Using user \ngroups can make security administration much easier, as a whole group of user accounts \n"
        },
        {
            "page_number": 97,
            "text": " \n90 \ncan be granted a new permission by simply adding the new privilege to a user group that \nthey are already a member of. \n \nSHADOWED PASSWORD FILES \nSome operating systems store their passwords in files that are hidden from all but \nadministrative-level accounts; such files are typically referred as shadowed password files \nand obviously afford greater protection than leaving the password file(s) in plain sight of an \nattack with nonadministrative privileges. \nThe danger with user groups is that sometimes, rather than creating a new user group, a \nsystem administrator will add a user account to an existing user group that has all the \nneeded privileges, plus a few unneeded ones. Thus, the system administrator grants the \nuser account (and any services running under this account) greater powers than it actually \nneeds. Of course, creating user groups that are so well defined that each user group only \nhas a single member defeats the whole purpose of defining security privileges by groups \ninstead of individual accounts. \n \nSEPARATION OF DUTIES \nA common practice for system administrators who need to access a machine as an \nadministrator but would also like to access the machine (or initiate services) with \nnonadministrative privileges is to create two accounts. The administrator would thus only \nlog into the administrator-level account to perform administrative-level tasks and use the \nless privileged account for all other work, reducing the possibility that he or she \ninadvertently initiates a service with administrative privileges. \nTable 4.18 summarizes some of the checks that should be considered when evaluating the \nappropriateness of user-group memberships. \n \nTable 4.18: System Software User-Group Checklist  \nYES  \nNO  \nDESCRIPTION  \n□  \n□  \nIs there a documented policy in place that describes how user groups \nare created, maintained, and removed? \n□  \n□  \nDo the user groups appear to have privileges that are too general, \nresulting in some user accounts being granted excessive privileges? \n□  \n□  \nDo the user groups appear to have privileges that are too specific, \nresulting in so many user groups that the system administrator is \nmore likely to make an error while assigning privileges? \n□  \n□  \nHas each user account be assigned to the appropriate user group(s)? \n \nSummary \nIt is not enough to review and test a Web site's network topology and configuration (the \nsubject of Chapter 3), since a poorly configured or unplugged security hole in a system \nsoftware product installed upon the Web site could provide an attacker with an easy entry \npoint. \n"
        },
        {
            "page_number": 98,
            "text": " \n91 \nAlthough few organizations have the resources to test another company's system software \nproducts for new vulnerabilities, it's not particularly desirable to discover that a known \nsecurity patch or workaround in a system software product has not been applied until after \nthe Web applications that will utilize this system software have been written, possibly \nnecessitating an unscheduled enhancement to the Web application. System software \nproducts should therefore always be evaluated from a security perspective before being \npressed into service. \n \n"
        },
        {
            "page_number": 99,
            "text": " \n92 \nChapter 5: Client-Side Application Security \nThink how convenient it would be if, once the security of the underlying infrastructure of a \nWeb site—the network devices and system software—had been tested and found to be \nsecure, anyone could host any Web application on this site and be confident that it would \nalso be secure. Unfortunately, this is an unrealistic scenario, as each Web application \nbrings with it the potential to introduce a whole new set of security vulnerabilities. For \nexample, the seemingly most secure of Web servers, with a perfectly configured firewall, \nwould provide no protection from an attacker who had been able to capture the userID and \npassword of a legitimate user of a Web application, possibly by simply sneaking a look at a \ncookie stored on the user's hard drive (see the Cookies section that follows for a more \ndetailed explanation of this potential vulnerability). Therefore, in addition to any testing done \nto ensure that a Web site's infrastructure is secure (the subject of Chapters 3 and 4), each \nand every Web application that is to be deployed on this infrastructure also needs to be \nchecked to make sure that the Web application does not contain any security \nvulnerabilities. \nApplication Attack Points \nMost Web applications are built on a variation of the multitier client-server model depicted in \nFigure 5.1. Unlike a standalone PC application, which runs entirely on a single machine, a \nWeb application is typically composed of numerous chunks of code (HTML, JavaScript, \nactive server page [ASP], stored procedures, and so on) that are distributed to many \nmachines for execution. \n \n \nFigure 5.1: Multitier client/server design.  \nEssentially, an attacker can try to compromise a Web application from three distinct entry \npoints. This chapter will focus on the first two potential attack points: the machine used to \nrun the client component of the application, and the communication that takes place \nbetween the client-side component of the application and its associated server-side \ncomponent(s). The next chapter will focus on the third potential entry point, the server-side \nportion of a Web application (for example, the application and database tiers of the \napplication). Ghosh (1998) discusses numerous e-commerce vulnerabilities, which he \ngroups into client, transmission, and server attack categories. \n \nClient Identification and Authentication \nFor many Web applications, users are first required to identify themselves before being \nallowed to use some (or any) of the features available via the application. For some \nsystems, this identification may be as simple as asking for an email address or a \npseudonym by which the user wishes to be known. In such circumstances, the Web \napplication may make no attempt to verify the identity of the user. For many Web \napplications, however, this laissez-faire level of identification is unacceptable, such \n"
        },
        {
            "page_number": 100,
            "text": " \n93 \napplications may not only ask users to identify themselves, but also to authenticate that the \nindividuals are who they claim to be. \nPerhaps one of the most challenging aspects of a Web application's design is devising a \nuser authentication mechanism that yields a high degree of accuracy while at the same \ntime not significantly impacting the usability of the application or its performance. Given that \nin the case of an Internet-based Web application, users could be thousands of miles away \nin foreign jurisdictions and using all manner of client-side hardware and system software, \nthis is far from a trivial task. Consequently, this is an area that will warrant comprehensive \ntesting. \nMost Web applications rely on one of three different techniques for establishing the \nauthenticity of a user: relying upon something users know (such as passwords), something \nthey have (like physical keys), or some physical attribute of themselves (for instance, \nfingerprints). These three strategies can all be implemented in a number of different ways, \neach with a different trade-off between cost, ease of use, and accuracy. Whichever \napproach is selected by the application's designers, it should be tested to ensure that the \napproach provides the anticipated level of accuracy. \nThe accuracy of an authentication mechanism can be measured in two ways: (1) the \npercentage of legitimate users who attempt to authenticate themselves but are rejected by \nthe system (sometimes referred to as the false rejection rate) and (2) the percentage of \nunauthorized users who are able to dupe the system into wrongly thinking they are \nauthorized to use the system (sometimes referred to as the false acceptance rate). \nUnfortunately, obtaining a low false acceptance rate (bad guys have a hard time getting in) \ntypically results in a high false rejection rate (good guys are frequently turned away). For \nexample, increasing the number of characters that users must use for their passwords may \nmake it harder for an attacker to guess (or crack) any, but it may also increase the \nfrequency with which legitimate users forget this information. \nThe risks associated with a false acceptance, compared to a false rejection, are quite \ndifferent. A false acceptance may allow intruders to run amok, exploiting whatever \napplication privileges the system had mistakenly granted them. Although a single false \nrejection may not be very significant, over time a large volume of false rejections can have \na noticeable effect on an application. For instance, a bank that requires multiple, hard-to-\nremember passwords may make its Web application so user-unfriendly that its adoption \nrate among its clients is much lower than its competitor's. This would result in a larger \npercentage of its clients visiting physical branches of the bank or using its telephone \nbanking system (both of which are more costly services to provide) and consequently giving \nits competitor a cost-saving advantage. \nFrom a testing perspective, the testing team should attempt to determine if a Web \napplication's false acceptance and false rejection rates are within the limits originally \nenvisaged by the application's designers (and users). Additionally, because there is no \nguarantee that a particular authentication technique has been implemented correctly, the \nmethod by which the authentication takes place should be evaluated to ensure the \nauthentication process itself can't be compromised (such as an attacker eavesdropping on \nan unencrypted application password being sent over the network). To this end, the \nfollowing sections outline some of the techniques that might be implemented in order to \nauthenticate a user of a Web application. Krutz (2001) and Smith (2001) provide additional \ninformation on user authentication strategies. \n"
        },
        {
            "page_number": 101,
            "text": " \n94 \nRelying upon What the User Knows: The Knows-Something Approach \nThe authenticity of the user is established by asking the user to provide some item of \ninformation that only the real user would be expected to know. The classic userID and \npassword combination is by far the most common implementation of this authentication \nstrategy. Variations of this method of authentication include asking the user to answer a \nsecret question (such as \"What's your mother's maiden name?\") or provide a valid credit \ncard number, together with corresponding billing address information. \nAlthough the issues associated with application-level userIDs and passwords are similar to \nthose affecting system software userIDs and passwords (see Chapter 4 for a discussion on \nthis topic), an organization that has developed its own applications may have more flexibility \nin its userID and password implementations. Specifically, the organization may choose to \nenforce more or less rigorous standards than those implemented in the products developed \nby outside vendors. For instance, the organization may check that any new password \nselected by a user is not to be found in a dictionary, or the organization may enforce a one-\nuser, one-machine policy (that is, no user can be logged on to more than one machine, and \nno machine can simultaneously support more than one user). \nTherefore, in addition to the userID checklist in Table 4.17, the testing team may also want \nto consider including some or all of the checks in Table 5.1, depending upon what was \nspecified in the applications security specifications. \n \nTable 5.1: Application UserID and Password Checklist  \nYES  \nNO  \nDESCRIPTION  \n□  \n□  \nDoes the application prohibit users from choosing common (and \ntherefore easily guessable) words for passwords? \n□  \n□  \nDoes the application prohibit users from choosing weak (and \ntherefore easily deciphered) passwords, such as ones that are only \nfour characters long? \n□  \n□  \nIf users are required to change their passwords every X number of \nweeks, does the application enforce this rule? \n□  \n□  \nIf users are required to use different passwords for Y generations of \npasswords, does the application enforce this rule? \n□  \n□  \nIs the application capable of allowing more than one user on the \nsame client machine to access the server-side component of the \napplication at the same time? \n□  \n□  \nIs the application capable of allowing one user to access the server-\nside component of the application from two or more client machines \nat the same time? \n□  \n□  \nIs the authentication method's false rejection rate acceptable? Is it \nmeasured by the number of calls made to the help desk for forgotten \npasswords? \n□  \n□  \nIs the authentication method's false acceptance rate acceptable? For \nexample, assuming no additional information, an attacker has a 1 in \n10,000 chance of correctly guessing a 4-digit numerical password. \n"
        },
        {
            "page_number": 102,
            "text": " \n95 \nRelying upon What the User Has: The Has-Something Approach \nInstead of relying on a user's memory, the Web application could require that users actually \nhave in their possession some artifact or token that is not easily reproducible. The token \ncould be physical (such as a key or smart card), software based (a personal certificate), or \nassigned to a piece of hardware by software (for instance, a media access control [MAC] \naddress could be assigned to an Ethernet card, a telephone number to a telephone, or an \nIP address to a network device). \nIf this has-something approach is used to authenticate a user to a Web application, the \ntesting team will need to test the enforcement of the technique. For instance, if the \nauthorized user Lee Copeland has a home telephone number of (123) 456-7890, then the \norganization may decide to allow any device to access the organization's intranet \napplications if accessed from this number. The testing team could verify that this \nauthentication method has been implemented correctly by first attempting to gain access to \nthe applications from Lee's home and then attempting access from an unauthorized \ntelephone number such as Lee's next-door neighbor's. \nSole reliance on an authentication method such as the telephone number or network \naddress of the requesting machine may not make for a very secure defense. For instance, \nLee's kids could access the organization's application while watching TV, or a \nknowledgeable intruder may be able to trick a system into thinking he is using an authorized \nnetwork address when in fact he isn't (a technique commonly referred to as spoofing). \nScenarios such as these illustrate why many of these has-something authentication \nmethods are used in conjunction with a knows-something method. Two independent \nauthentication methods provide more security (but perhaps less usability) than one. So in \naddition to the telephone number requirement, Lee would still need to authenticate himself \nwith his userID and password combination. \nThe following section describes some of the most common has-something authentication \ntechniques. \nPersonal Certificates \nA personal certificate is a small data file, typically obtained from an independent certification \nauthority (CA) for a fee; however, organizations or individuals can manufacture their own \ncertificates using a third-party product. (For more information on the use of open-source \nproducts to generate certificates, go to www.openca.org.) This data file, once loaded into a \nmachine, uses an encrypted ID embedded in the data file to allow the owner of the \ncertificate to send encrypted and digitally signed information over a network. Therefore, the \nrecipient of the information is assured that the message has not been forged or tampered \nwith en route. \nPersonal certificates can potentially be a more secure form of user authentication than the \nusual userID and password combination. However, personal certificates to date have not \nproven to be popular with the general public (perhaps because of privacy issues and their \nassociated costs). So keep in mind that any Web application aimed at this group of users \nand requiring the use of a personal certificate may find few people willing to participate. In \nthe case of an extranet and intranet application, where an organization may have more \npolitical leverage with the application's intended user base, personal certificates may be an \nacceptable method of authentication. Table 5.2 lists some firms that offer personal \ncertificates and related services (and therefore provide more detailed information on how \npersonal certificates work). \n \n"
        },
        {
            "page_number": 103,
            "text": " \n96 \nTable 5.2: Sample Providers of Personal Certificates and Related Services  \nNAME  \nASSOCIATED WEB SITE  \nBT Ignite \nwww.btignite.com  \nEntrust \nwww.entrust.com  \nGlobalSign \nwww.globalsign.net  \nSCM Microsystems \nwww.scmmicro.com  \nThawte Consulting \nwww.thawte.com  \nVeriSign \nwww.verisign.com  \nBrands (2000), Feghhi (1998), and Tiwana (1999) provide additional information on digital \ncertificates. \nSmart Cards \nSmart cards are physical devices that contain a unique ID embedded within them. With this \ndevice and its personal identification number (PIN), the identity of the person using the \ndevice can be inferred (although not all cards require a PIN). \nA SecurID smart card is an advanced smart card that provides continuous authentication by \nusing encryption technology to randomly generate new passwords every minute. It provides \nan extremely robust authorization mechanism between the smart card and synchronized \nserver. Table 5.3 lists some firms that offer smart cards and related services. Hendry \n(2001), Rankl (2001), and Wilson (2001) provide additional information on smart cards. \n \nTable 5.3: Sample Providers of Smart Cards and Related Services  \nNAME  \nASSOCIATED WEB SITE  \nActivcard \nwww.activcard.com  \nDallas Semiconductor Corporation \nwww.ibutton.com  \nDatakey \nwww.datakey.com  \nLabcal Technologies \nwww.labcal.com  \nMotus Technologies \nwww.motus.com  \nRSA Security \nwww.rsasecurity.com  \nSignify Solutions \nwww.signify.net  \nVASCO \nwww.vasco.com  \nMAC Addresses \nThe MAC address is intended to be a globally unique, 48-bit serial number (for instance, \n2A-53-04-5C-00-7C) and is embedded into every Ethernet network interface card (NIC) that \nhas ever been made. An Ethernet NIC used by a machine to connect it to a network can be \nprobed by another machine connected to the network directly or indirectly (for example, via \nthe Internet) to discover this unique number. It is therefore possible for a Web server to \nidentify the MAC address of any visitor using an Ethernet card to connect to the Web site. \n"
        },
        {
            "page_number": 104,
            "text": " \n97 \nFor example, the Windows commands winipcfg and ipconfig, or the UNIX command ifconfig \ncan be used to locally view an Ethernet card's MAC address. \nPrivacy concerns aside, the MAC address can be used to help authenticate the physical \ncomputer being used to communicate to a Web server. But because some Ethernet cards \nenable their MAC addresses to be altered, this authentication technique cannot be \nguaranteed to uniquely identify a client machine. Therefore, it should not be relied upon as \nthe sole means of authentication. \nIP Addresses \nEvery data packet that travels the Internet carries with it the IP network address of the \nmachine that originated the request. By examining this network source address, a receiver \nshould be able (in theory) to authenticate the sender of the data. Unfortunately, this form of \nauthentication suffers from two major problems; proxy servers hide the network address of \nthe sender, replacing the original sender's network address with their own. This means that \nwhen proxy servers are involved, IP address verification is typically only going to be able to \nverify the organization or the Internet service provider (ISP) that owns the proxy server, not \nthe individual machine that originated the request. The second problem with IP address \nauthentication is that the source IP address is relatively easy to alter, or spoof, and it \ntherefore should not be relied upon as the sole means of identifying a client. \nTelephone Numbers \nUsing a telephone number for authentication is a technique used by many credit card \nissuers to confirm delivery of a credit card to the card's legitimate owner. The credit card is \nactivated once a confirmation call has been received from the owner's home telephone \nnumber. \nRequiring users to access an organization's remote access server (RAS) from a specific \ntelephone number (often authenticated via a callback mechanism) is a common way of \nrestricting access to applications on an intranet or extranet. Unfortunately, an attacker can \nsubvert a callback mechanism by using call forwarding to forward the callback to an \nunintended destination. This would make this form of authentication undependable, if used \nas the sole method of authentication. \nTable 5.4 lists some tests that should be considered if the application uses some form of \ntoken to determine who a user is. \n \nTable 5.4: Secure Token Checklist  \nYES  \nNO  \nDESCRIPTION  \n□  \n□  \nIs there a documented policy in place that describes how application \nsecurity tokens are assigned and removed from circulation? \n□  \n□  \nAre the procedures for handling lost tokens adequate and always \nfollowed? \n□  \n□  \nCan the token be counterfeited? If technically feasible, how likely is it \nthat this would actually take place? \n□  \n□  \nDoes the application enable the same token to be simultaneously \nused more than once? \n□  \n□  \nIs the authentication method's false rejection rate acceptable? \n"
        },
        {
            "page_number": 105,
            "text": " \n98 \nTable 5.4: Secure Token Checklist  \nYES  \nNO  \nDESCRIPTION  \n  \n□  \nIs the authentication method's false acceptance rate acceptable? \nRelying upon What the User Is: The Biometrics Approach \nUndoubtedly the most secure of the three authentication approaches, a biometric device \nmeasures some unique physical property of a user that cannot easily be forged or altered, \nthereby providing an extremely accurate method of identifying an individual user. Biometric \nauthentication methods include fingerprints, hand geometry, face geometry, eye patterns \n(iris and/or retina), signature dynamics (speed, pressure, and outline), keyboard typing \npatterns, and voice. Ideally, a combination of two or more of these methods should be \nused, as advances in technology have made some of these measurements easier to fake. \nThis approach also has the most resistance to adoption by the general population. \nHowever, in situations where the use of a biometric device can be deployed without \nadoption concerns (such as in the military), it is often the method of choice for Web \napplications that need unequivocal confirmation of who the user is. One drawback of a \nbiometric measurement is what happens after an ID has been compromised. For example, \nsuppose the input data sent from the scanner has been compromised because of \neavesdropping or with the assistance of an infamous evil twin. Unfortunately, there is \ntypically no way to issue the user a new identifying characteristic. (Eye surgery would seem \na little drastic.) An additional drawback is the fact that some of the measurements are more \nsusceptible than others to false acceptances and rejections. Nanavati et al. (2002) provide \nadditional information on biometrics. \nTable 5.5 lists some firms that offer biometric devices and related services. (For more \ndetailed information on how they work, go to the individual Web sites listed.) \n \nTable 5.5: Sample Providers of Biometric Devices and Related Services  \nNAME  \nASSOCIATED WEB SITE  \nActivCard \nwww.activcard.com  \nCyber-SIGN \nwww.cybersign.com  \nDigitalPersona \nwww.digitalpersona.com  \nIdentix \nwww.identix.com  \nInterlink Electronics \nwww.interlinkelec.com  \nIridian Technologies \nwww.iridiantech.com  \nKeyware \nwww.keyware.com  \nSAFLINK \nwww.saflink.com  \nSecuGen \nwww.secugen.com  \nVisionics \nwww.visionics.com  \nTable 5.6 lists some tests that should be considered if the application is to use a biometric \ndevice to authenticate a user's identity. \n"
        },
        {
            "page_number": 106,
            "text": " \n99 \n \nTable 5.6: Biometric Device Checklist  \nYES  \nNO  \nDESCRIPTION  \n  \n  \nIs there a documented policy in place that describes how biometric \nmeasurements are originally captured and authenticated? \n□  \n□  \nAre the procedures for handling compromised measurements \nadequate and always followed? \n□  \n□  \nCan the biometric measurement be faked? If technically feasible, how \nlikely is it that this would actually take place? \n□  \n□  \nIs the false rejection rate too high? For example, the measuring \ndevice could be too sensitive. \n□  \n□  \nIs the false acceptance rate too high? For example, the measuring \ndevice could not be sensitive enough. \n \nUser Permissions \nIt would be convenient if all legitimate users of an application were granted the same \npermissions. Alas, this situation rarely occurs. (For example, many Web-based applications \noffer more extensive information to subscription-paying members than they do to \nnonpayers.) Permissions can be allocated to users in many ways, but generally speaking, \nrestrictions to privileges take one of three forms: functional restrictions, data restrictions, \nand cross-related restrictions. Barman (2001), Peltier (2001), and Wood (2001) all provide \nguidance on developing user security permissions. \nFunctional Restrictions \nUsers can be granted or denied access to an application's various functional capabilities. \nFor example, any registered user of a stock-trading Web application may get a free 15-\nminute-delayed stock quote, but only users who have opened a trading account with the \nstockbroker are granted access to the real-time stock quotes. \nOne of the usability decisions a Web application designer has to make is to decide whether \nor not features that a user is restricted from using should be displayed on any Web page \nthe user can access. For instance, in the previous example, should an ineligible user be \nable to see the real-time stock quote menu option, selecting this option will result in some \nsort of \"Access denied\" message being displayed. The following are some of the arguments \nfor displaying restricted features: \nƒ \nSeeing that the feature exists, users may be enticed into upgrading their status \n(through legitimate methods). \nƒ \nIn the case of an intranet application, when an employee is promoted to a more \nprivileged position, the amount of additional training that he or she needs in order to \nuse these new privileges may be reduced. The employee would already have a great \ndegree of familiarity with the additional capabilities that he or she has just been \ngranted. \nƒ \nHaving only one version of a user interface should reduce the effort needed to build \nan online or paper-based user manual. The manual would certainly be easy to follow, \nas any screen captures or directions in the manual should exactly match the screens \nthat each user will see, regardless of any restrictions. \nƒ \nHaving only one version of a user interface will probably reduce the amount and \ncomplexity of coding needed to build the application. \n"
        },
        {
            "page_number": 107,
            "text": " \n100 \nSome drawbacks also exist, however, among which are the following: \nƒ \nOne of the simplest forms of security is based on the need-to-know principal. Users \nare only granted access to the minimum amount of information they need to know in \norder to perform their expected tasks. Merely knowing the existence of additional \nfeatures may be more information than they need, as it could entice them into trying \nto acquire this capability through illegal channels. \nƒ \nSome legitimate users may find the error messages generated when they try to \naccess forbidden areas frustrating. They might think, \"If I can't access this feature, \nwhy offer me the option?\" or even assume the application is broken. \nƒ \nToo many inaccessible options may overcomplicate a user interface, thereby \nincreasing a user's learning curve and generating additional work for the help desk. \nWhichever approach is taken, the application's user interface should be tested to ensure \nthat the same style of interface is used consistently across the entire application, reducing \nthe probability of errors while at the same time improving usability. \nOne recurring problem with function-only restrictions is that these controls may be \ncircumvented if the user is able to get direct access to the data files or database. \n(Unfortunately, this is an all-too-common occurrence with the advent of easy-to-use \nreporting tools.) The database and data files should be checked to ensure that a \nknowledgeable insider couldn't circumvent an application's functional security measures by \naccessing the data directly. For instance, via an Open Database Connectivity (ODBC) \nconnection from an Excel spreadsheet. \nData Restrictions \nInstead of restricting access to data indirectly by denying a user access to some of an \napplication's functionality, the application could directly block access to any data the user is \nnot authorized to manipulate or even view. For example, sales representatives may be \nallowed to view any of the orders for their territory but not the orders for their peers. \nCorrespondingly, a regional sales manager may be able to run reports on any or all of the \nreps that report to him or her, but not for any other rep. \nFunctional and Data Cross-Related Restrictions \nMany applications use a combination of functional and data restrictions. For instance, in \naddition to only being able to see orders in their own territory, a rep may not be allowed to \nclose out a sales quarter, an action that can only be performed by the vice president of \nsales or the CFO. \nLess common are situations in which access to a particular function is based on the data \nthat the user is trying to manipulate with the function. For example, reps may be allowed to \nalter the details of an order up until it is shipped, after which they are denied this ability, a \nprivilege that is only available to regional managers. A more complicated example would be \na financial analyst who is restricted from trading in a stock for 72 hours after another analyst \nat the same firm changes their buy/sell recommendation. \nEach of these three forms of restrictions (functional, data, or a hybrid of both) can be \nenforced using one or more different implementations (for example, via application code, \nstored procedures, triggers, or database views). Regardless of the approach used to \nimplement the restrictions, the application should be tested to ensure that each category of \nuser is not granted too many or too few application permissions. Table 5.7 summaries \nthese checks. \n \n"
        },
        {
            "page_number": 108,
            "text": " \n101 \nTable 5.7: User Permissions Checklist  \nYES  \nNO  \nDESCRIPTION  \n□  \n□  \nIs there a documented policy in place that describes under what \ncircumstances users will be granted access to the application's \nfunctional capabilities and data? \n□  \n□  \nIs there a documented policy in place that describes how user \napplication privileges may be altered or removed? \n□  \n□  \nDoes the application's user interface take a consistent approach to \ndisplaying (or hiding) functions that the user is currently not \nauthorized to use? \n□  \n□  \nCan any of the users access a function that should be restricted? \n  \n  \nCan all the users access every function that they should be permitted \nto use? \n□  \n□  \nCan all the users access data that they should be permitted to use? \n \nTesting for Illicit Navigation \nOne of the features of the Internet is that users are able to jump around a Web application \nfrom page to page in potentially any order. Browser options such as Go, History, Favorites, \nBookmarks, Back, Forward, and Save pages only add to the flexibility. In an attempt to \nensure that a user deliberately attempting to access Web pages in an inappropriate \nsequence (such as trying to go to the ship to Web page without first going through the \npayment collection page) cannot compromise a site's navigational integrity and security, \ndesigners may have to utilize one or more techniques to curtail illicit activities. If such \nprecautions have been built into the application, they should be tested to ensure that they \nhave been implemented correctly. \nHTTP Header Analysis \nSome Web sites will use the information contained in a Hypertext Transfer Protocol (HTTP) \nheader (the Referer field) to ascertain the Web page that the client has just viewed and \nthereby determine if the client is making a valid navigational request. Although an attacker \ncould easily alter this field, many attackers may not suspect that this defense is being \nemployed and therefore will not consider altering this field. \nHTTP Header Expiration \nTo reduce the ease with which an attacker can try to navigate to a previously viewed page \n(instead of being forced to download a fresh copy), the HTTP header information for a Web \npage can be manipulated via HTTP-EQUIV meta tags Cache-control, Expires, or Pragma to \nforce the page to be promptly flushed from the requesting browser's memory. \nUnfortunately, only novice attackers are likely to be thwarted by this approach (as previous \nviewed Web pages can always be saved to disk). However, if this defense has been \ndesigned into the application, it should still be checked to ensure that it has been \nimplemented. \n"
        },
        {
            "page_number": 109,
            "text": " \n102 \nClient-Side Application Code \nSome Web applications rely on client-side mobile code to restrict access to sensitive pages \n(mobile code is discussed in more detail later in this chapter). For instance, before entering \na restricted Web page, a client-side script could be used to launch a login popup window. If \nthe Web application uses such a mechanism, it should be tested to ensure that a user \nturning off scripting, Java applets, or ActiveX controls in his or her browser before \nattempting to access the restricted page does not allow the user to circumvent this \nrestriction. \nSession IDs \nBy placing an item of unique data on the client (discussed in more detail in the Client-Side \nData section), a Web application can uniquely identify each visitor. Using this planted \nidentifier (sometimes referred to as session ID), a Web application can keep track of where \na user has been and thereby deduce where he or she may be permitted to go. \nThe effectiveness of this approach to a large degree depends on how and where this \nidentifier is stored on the client machine (as will be described in the Client-Side Data \nsection), with some methods being safer than others. \nNavigational Tools \nIf access to a large number of Web pages needs to be checked using several different user \nprivileges, it may make sense to create a test script using one of the link-checking tools \nlisted in Table 5.8. The script can then be played back via different userIDs. One can also \nproduce a report of all the Web pages that were previously accessible (when the test script \nwas created using a userID with full privileges) but have since become unobtainable due to \nthe reduced security privileges assigned to each of the userIDs. \n \nTable 5.8: Sample Link-Checking Tools  \nNAME  \nASSOCIATED WEB SITE  \nAstra SiteManager \nwww.mercuryinteractive.com  \ne-Test Suite \nwww.empirix.com  \nLinkbot \nwww.watchfire.com  \nSite Check \nwww.rational.com  \nSiteMapper \nwww.trellian.com  \nWebMaster \nwww.coast.com  \n \nHIDING CLIENT-SIDE CODE \nStoring code in a separate file (for instance, hiddencode.js) is a technique used by some \ndevelopers to avoid a user casually viewing client-side source code that controls security \nfunctions. The Web page that needs this code then references this file, thereby avoiding the \nneed to embed the code in the HTML used to construct the Web page (which would allow a \nviewer to easily view the code alongside the HTML code). Here's an example: \n<script language=\"JavaScript1.2\" src=\"hiddencode.js\" \ntype=\"text/javascript\"</script> \n"
        },
        {
            "page_number": 110,
            "text": " \n103 \nUnfortunately, even novice attackers are likely to realize that they could review the script by \nsearching the browser's cache on their hard drive and opening the supposedly hidden file \nwith their favorite text editor. \n \nTHE ANONYMOUS USER \nOne illicit navigation strategy that some Web application designers fail to consider \ndefending against (and therefore should be tested for) is the attacker who, after logging on \nlegitimately, saves a restricted Web page to his or her hard drive and then logs off. The \nattacker then opens this file (probably from an untraceable client), and using the links and \nclient-side scripts on this saved page (which may have been edited with the use of a Web \npage authoring tool), he or she attempts to reenter the restricted portion of the Web \napplication without going through a login procedure, thereby gaining complete anonymity. \nTable 5.9 lists some of the scenarios that should be considered when trying to ensure that a \nuser cannot inappropriately access any portion of the Web application. \n \nTable 5.9: Illicit Navigation Checklist  \nYES  \nNO  \nDESCRIPTION  \n□  \n□  \nIs there a documented policy in place that describes how clients will \nbe prohibited from accessing the Web application illegally? \n□  \n□  \nCan disabling client-side scripts or mobile code circumvent critical \nWeb pages or pop-up windows? \n□  \n□  \nBy using a browser's Go, History, Favorites, or Bookmark features, \ncan a restricted portion of the Web application be navigated to \nwithout first gaining authorization? \n□  \n□  \nBy using a browser's Go, History, Favorites, Bookmark, Forward, or \nBackward features, can a restricted portion of the Web application be \nnavigated to illegally? \n□  \n□  \nIs any HTTP header analysis or expiration option recommended by \nthe designers implemented correctly? \n□  \n□  \nCan a Web page previously saved to a local hard drive be used to \nnavigate to a restricted portion of a Web application and circumvent \nany login process? \n \nClient-Side Data \nToo often, sensitive application information (such as userIDs, passwords, authorization \nlevels, credit card numbers, and social security numbers) is stored in unencrypted or \nweakly encrypted formats on the client machine. Given the frequency with which employees \nshare their hard drives with one another (for instance, via Windows file-sharing capabilities), \nsnooping colleagues could view this information without even leaving their desks. \nAlternatively, users could review and alter the information stored on their own machines \nand thereby attempt to gain access to a portion of the Web application they are not \nauthorized to view. For example, after a successful login attempt, the user's security level is \nstored on the client machine and this level is then re-sent with every subsequent \ntransmission back to the Web site as a means of authenticating the user. The user could \nthen escalate his or her application privileges by simply editing the field used to store their \n"
        },
        {
            "page_number": 111,
            "text": " \n104 \nlevel. Possible values to try might include Admin, Anonymous, DBA, Dummy, Guest, \nMaster, Primary, QA, Root, Superuser, and, of course, Test.  \nIdeally, sensitive information about the client should be sent and stored in an encrypted \nformat. This endeavor will require a little extra central processing unit (CPU) effort and a \nlittle more network bandwidth, as encrypted data is typically longer than its unencrypted \nequivalent. But any information that must reside on the client machine, if only temporarily, is \nvulnerable to an intruder who could potentially use it to cause mischief. So, protection is in \norder through encryption or at least reformatting (hashing). An intruder who collects \nunencrypted client-side data files (possible from several different machines) and then \ncompares the data looking for a pattern may be able to crack any security-by-obscurity \napproach designed to protect the data. \nIf a Web application is going to store data on the client side, a designer may choose among \nseveral different places to locate this information. Although each location has its pros and \ncons, whichever approach is implemented, the testing team should check that the data has \nbeen sufficiently protected from a malicious user trying to take advantage of the \naccessibility of this client-side data. \nCookies \nA cookie is a little nugget of information that is sent to a browser from a Web server. This \nblock of data can be anything: a unique session ID generated by the Web server, the \ncurrent date and time, the network address of where the browser is accessing the Internet \nfrom, or any other chunk of data that might be useful to the Web application. Dustin et al. \n(2001) and St. Laurent (1998) provide additional information on cookie security issues. \nBrowsers manage cookies automatically. After receiving a cookie, they will by default send \nthis information back to the Web server that originated it every time the browser makes a \nrequest to that Web site (unless disabled by the browser's user). Basically, two types of \ncookies exist: persistent cookies and session cookies. \nPersistent Cookies \nPersistent cookies continue to be stored on the client machine after the browser has been \nclosed and the machine has powered down. This is accomplished by physically storing the \ncookie on the user's hard drive. (Microsoft's Internet Explorer uses a directory called \nCookies to store its persistent cookies, while Netscape uses a single file called Cookies.txt.) \nOrdinarily, a Web site will only use a single persistence cookie. Nevertheless, cookie \nimplementations do enable each individual Web page to have its own cookie. \nThe option exists for persistent cookies to be flagged to expire after a specified period of \ntime. This feature often is used if the cookie is being used to store the client's userID, \nthereby forcing users to periodically reidentify themselves. \nWeb applications can sometimes be tricked into giving away additional information by \ndeleting just some of the information contained in a persistent cookie. Rather than \ncompletely rejecting or resetting a corrupted cookie, a Web application may replace the \nmissing information with application defaults, which may enable the user to deduce more \nuseful values to try in this field and thereby gain access to resources they should have been \nrestricted from. \nTo reduce the possibility of cookie tampering (also known as cookie poisoning), some Web \napplications include a parity check in the cookie, rejecting the contents of any cookie that \nconflicts with its parity tag. In addition, some Web applications will embed the network \n"
        },
        {
            "page_number": 112,
            "text": " \n105 \naddress of the client machine into a cookie in an effort to hinder cookie-stealing (or \ncounterfeiting) activities. \nSession Cookies \nSession cookies reside in the browser's memory and only live as long as that instance of \nthe browser remains open. Each open browser instance will have its own session cookie for \na Web site. However, if the client's machine is short of memory, a session (memory \nresident) cookie may be swapped out to the client's hard drive (virtual memory). If the \nbrowser were to be terminated abnormally (crash), it would most likely not clean up its \nvirtual memory, and the session cookie would subsequently be visible to anyone interested \nin viewing the hard drive. \n \nMAGIC COOKIES \nThe name cookie was originally derived from the UNIX term magic cookies, which referred \nto objects that could be attached to a user or program, and change depending on the areas \nentered by the user or program. \nHidden Fields \nAs an alternative to cookies, some Web applications use hidden fields on an HTML form, or \nan Extensible Markup Language (XML) file, to store information on the client-side \ncomponent of the application. One of the advantages of this approach is that it works on \nbrowsers that have had their cookie capabilities disabled (or simply don't support cookies in \nthe first place). Unfortunately, these hidden fields are not hidden very well. All a user has to \ndo to view and alter the contents is to edit the Web page via a browser's built-in capabilities, \nor save it to a disk, and then edit it with his or her HTML or XML authoring tool of choice. \nFor example, suppose after a successful login, instead of storing the actual client userID in \na hidden field on an HTML form, the Web site designer decided to use a single character to \nindicate the appropriate level of user privileges (for example, A for administrator, R for read-\nonly, and W for read-write access). The theory behind this is that this information would \nnever be displayed by the browser, and even if users reviewed the source code used to \nbuild the Web page, they would be unlikely to figure out that this single character was being \nused to control their security level. Although this may have been a reasonable assumption \nfor some clients, it is certainly not true for members of the development and testing teams \nwho are aware of this design and could therefore easily exploit this design should they \nchose to do so at a later date. Dustin et al. (2001) provides additional information on \nsecurity issues related to hidden fields. \nURLs \nSome Web applications embed userIDs and other sensitive information into a URL, typically \nas parameters in the Query component of the URL (the fields that occur after the ? symbol \nin a URL). Unfortunately, this information is easily viewed by a passerby and also recorded \nin a Web site's log, where a corrupt Webmaster may view it at leisure. \nLocal Data Files \nDepending upon the privileges assigned by the client to download mobile code (mobile \ncode is discussed in the Mobile Application Code section), the mobile code may have free \nreign to create temporary or permanent data files on the machine it is being executed on or \nany network drive that the client is authorized to access. \n"
        },
        {
            "page_number": 113,
            "text": " \n106 \nAn area of particular concern is what happens if execution of the mobile code is abruptly \nterminated. Under such conditions, any temporary files might not be cleaned up as the \ndeveloper intended and may leave sensitive data lingering on the client's hard drive. \nWindows Registry \nIn the case of mobile code such as an ActiveX control, the designer may choose to store \napplication data in the client's registry. If unencrypted, this information could be viewed (and \noptionally modified) by simply running the built-in Windows utility regedit.  \nTable 5.10 lists some of the checks that should be considered when trying to ensure that \nany data stored on a client machine is protected from abuse. \n \nTable 5.10: Client-Side Data Checklist  \nYES  \nNO  \nDESCRIPTION  \n□  \n□  \nIs there a documented policy in place that describes what (if \nanything) will be stored permanently or temporarily on a client \nmachine? Where will it be stored? What precautions will be used to \nprotect the information from being tampered with? \n□  \n□  \nIs any sensitive client-side data not protected by encryption? \n□  \n□  \nAre there checks in place to prevent a client from tampering with \nclient-side data? \n□  \n□  \nAre there checks in place to detect tampered client-side data? \n  \n□  \nIs the client-side data easily visible to a casual passerby or via a \nnetwork share? If so, could this information be utilized from another \nmachine? \n□  \n□  \nDo any memory-resident data files contain any information, which, if \ntemporarily written to a hard drive (possible as the result of low \nmemory), might pose a potential security risk? \n \nSecure Client Transmissions \nSensitive data transmitted over the Internet should always be encrypted to avoid potential \nintruders from eavesdropping on the communication anywhere along the route the data \ntakes between the two machines. Wireless connections are particularly prone to this \nactivity, as the eavesdropper does not even need to acquire a physical connection to the \nnetwork being sniffed. Stories abound of would-be eavesdroppers driving around towns \nsearching for unprotected wireless communications. \nTo protect against eavesdropping on Internet traffic being eavesdropped, the industry has \ndeveloped several different encryption schemes. Examples include Secure-HTTP (S-\nHTTP), Secure Sockets Layer (SSL), and Internet Protocol Security (IPsec), each of which \nare described in more detail in Appendix A. In addition, several new encryption schemes \nare emerging, such as WAP's Wireless Transport Layer Security (WTLS) protocol, for \nsecuring the wireless hop(s) made by an Internet message. Burnett (2001), Feghhi et al. \n(1998), and Stallings (1998) provide additional information on secure network \ntransmissions. \n"
        },
        {
            "page_number": 114,
            "text": " \n107 \nDigital Certificates \nTo allow two parties to communicate with each other via encrypted messages, at least one \nof the parties must have a digital certificate installed. (Typically, this certificate is referred to \nas a server certificate if installed on a Web server, or a personal certificate if installed in a \nbrowser.) The digital certificate contains a key that enables the sender to encrypt the data \nin a way that ensures that only the intended receiver is able to decipher the data and also \nproves to the recipient that it was really the alleged sender who actually sent the message. \nCertificates are transported from issuer to user by means of a certificate file. Typically, \nthese files are themselves encrypted and protected by a user-selected (ideally strong) \npassword. If an attacker is able to guess the password used to protect the certificate, he or \nshe may be able to transfer the certificate to another machine (inside or outside of the \nowning organization). Then the attacker uses this second machine to try and trick another \nInternet user into thinking that he or she is communicating with the legitimate owner. \nTo reduce the threat of a certificate file being copied, thereby allowing an attacker to try and \ncrack the password protecting the certificate at his or her leisure, the file should be removed \nfrom any easily accessible location that an attacker may be able to get to. For example, a \nfloppy disk in a bank's safe deposit box makes for a much more secure home for the file \nthan the hard drive of the Web server on which the certificate was installed. (Note that, \nonce installed, the certificate file does not need to be stored on the machine that is using it.) \nEncryption Strength \nWhen using SSL to encrypt a Web page, the URL displayed by a browser is typically \npreceded by https (as opposed to http). Some browsers also use a visual cue to indicate \nwhether the communication is encrypted or not (with broken keys and padlock icons). What \nthe visual cue, or https prefix, does not indicate is the type of encryption being used for the \ncommunication. A number of different encryption algorithms exist (such as RC4, DES, and \nMAC), as do different encryption keys lengths (40, 56, 128, and 168 are currently the most \ncommon key-length implementations). \nThe ease with which a message can be deciphered using a brute-force attack (brute-force \nattacks are described in more detail in Chapter 4) is dependent upon the key size used to \nencrypt the message. (The larger the key, the longer it will take to break it.) So, a larger key \nsize will make the data being transmitted more secure. Unfortunately, a larger-sized key will \nalso make the encrypted data file larger, thereby utilizing additional network bandwidth, and \nit will place an additional strain on the CPU doing the encryption and deciphering. \nTherefore, a trade-off exists between improved security and improved performance. For \nmarginally confidential information, such as a financial analyst's report that only paid \nsubscribers are supposed to have access to, perhaps a weak and speedy encryption \nstrategy would work best. On the other hand, super-sensitive data, such as a Swiss \nnumbered bank account, ought to be encrypted using the strongest strategy available. \n \nCERTIFICATE CLASSES \nCertificates come in several different classes, the class reflecting the ease with which the \ncertificate can be obtained from its issuer, and therefore the implied authenticity of its user. \nClass 1 certificates are typically easy to obtain; a valid credit card number may be all that is \nneeded. Class 3 certificates typically require thorough background checks before they are \nreleased and therefore infer a higher level of trust that the person or organization \npurchasing the certificate is really the entity that they claim to be. \n"
        },
        {
            "page_number": 115,
            "text": " \n108 \nWhatever method of encryption an organization deems appropriate, the encryption method \nactually implemented by a Web site should be checked to ensure that it complies with the \nstrength of encryption specified by the Web application's designers. One way to test this is \nto communicate with the target Web site using a browser that can be configured to work \nwith different encryption settings, varying the settings to ensure that the data is not being \nunder- or overencrypted. For instance, Opera 6.0 and Netscape 6.2 enable users to select \nwhich SSL encryption algorithms and key lengths may be utilized by the browser. In \naddition, some browsers provide the user with the opportunity to view the details of a \ndownloaded certificate, such as its key length. \nMixing Encrypted and Nonencrypted Content \nBecause typically not every Web page on a Web site needs to be encrypted, most Web \nservers enable each page to be individually selected (or omitted) for encryption, thereby \nreducing the amount of data that actually needs to be encrypted. For instance, unlike the \nconfirm trade page, it's unlikely that the help page on a stockbroker's Web site would need \nto be encrypted. \nThis concept can be taken to a more granular level, with individual Web page components \nbeing selectively encrypted or left unencrypted. For instance, typically most of the time \nspent downloading a Web page is in fact spent waiting for the images (as opposed to the \ntext) on the Web page to be downloaded. Frequently, there is nothing confidential about \nthese images and consequently little point in encrypting these network bandwidth-intense \ngraphic files. (Once downloaded, the user could always save the unencrypted image to disk \nanyway.) \nMixing encrypted and unencrypted components on the same Web page has a drawback, as \ndoes, to a lesser extent, using encrypted and unencrypted Web pages on the same Web \nsite. This is that some browsers will generate numerous (well-intentioned, but still annoying) \nmessages, warning users that some of the components that make up the Web page are \nunencrypted, thereby creating a potential usability issue. \nJust because a Web page has been downloaded in an encrypted format does not \nnecessarily mean that the reply will be encrypted (or vice versa). For instance, a Web site's \nlogin page may be sent from the Web server to the browser unencrypted; there's no \nsecurity risk in an eavesdropper viewing an empty login form. But the corresponding reply \nfrom the user, with the associated userID and password, should be encrypted. \nUnfortunately, this approach also has usability issues, as a user may incorrectly assume \ntheir reply (containing their userID and password) will not be encrypted and therefore be \nhesitant to use. That deduction would be made because the download empty login form \nhad not been encrypted, causing the browser to display the unencrypted visual cue: an \nopen padlock or broken key. \n \nCOOKIE ENCRYPTION \nA Web site can use the HTTP Secure Cookie option to ensure that a cookie is only \ntransmitted back to it if the cookie is first encrypted. This is a useful option if the Web site \ncontains a mixture of encrypted and unencrypted Web pages. \nIf a mix-and-match approach to encryption is adopted, the testing team should check that \nall the Web pages (or page components) that should be encrypted are indeed encrypted, \nand any that should not be, aren't. This test can be accomplished by manipulating browser \nsecurity settings or by performing source code inspections. \n"
        },
        {
            "page_number": 116,
            "text": " \n109 \nAvoiding Encryption Bottlenecks \nFor many secure Web applications, encryption may be the first scalability bottleneck. Due \nto the heavy requirements of encryption, even a small number of simultaneous clients can \nsaturate a Web server. To mitigate this problem, Web sites have adopted a number of \ndifferent strategies, which are discussed in the following sections. \nDirecting Traffic to Dedicated Security Servers \nBy directing all encrypted network traffic to a dedicated Web server(s), the impact of a \nsudden surge in encrypted traffic can be localized to the secure portion of the Web \napplication, and not the entire Web site. In addition, the Web server's operating system and \nservices can be tuned (and hardened) to better handle copious amounts of encrypted \ntraffic. \nUsing Encryption Cards \nBecause encryption is so CPU intensive, the CPU is often the first component of a Web \nserver to suffer from an increase in encrypted traffic. For this reason, many companies \nmanufacture encryption accelerator cards to help alleviate the stress placed on a general-\npurpose CPU by large amounts of encryption and decryption. This is analogous to the way \nvideo cards are used to offload graphic manipulation from the CPU. For a list of companies \nmanufacturing such cards, see Table 5.11. \nTable 5.11: Sample Providers of Encryption Accelerators  \nNAME  \nASSOCIATED WEB SITE  \nAccelerated Encryption Processing \nwww.aep-crypto.com  \nAndes Networks \nwww.andesnetworks.com  \nCryptographic Appliances \nwww.cryptoapps.com  \nGlobal Technologies Group \nwww.powercrypt.com  \nIBM \nwww.ibm.com  \nnCipher \nwww.ncipher.co  \nRainbow Technologies \nwww.rainbow.com  \nSafeNet \nwww.safenet-inc.com  \nSonicWall \nwww.sonicwall.com  \nIf encryption is going to be used to any great extent on the Web site, the testing team \nshould consider running performance tests to ensure that a surge in encrypted traffic will \nnot bring the Web site to its knees. If another group is doing some performance testing, the \nteam should ensure that the load profiles used by the other group adequately reflect the \ndegree to which encryption will be used. \nEvaluating the security of client transmissions is unfortunately not simply a case of \ndetermining whether or not the Webmaster has enabled SSL on the Web server. It also \nincludes checks to make sure that not too much, or too little, is being encrypted, that the \nlevel of encryption is appropriate, and that the safety of any digital certificate files is \nensured. Table 5.12 lists some tests that should be considered when evaluating the \nsecurity of sensitive data being transmitted from or to the client-side component of a Web \napplication. \n"
        },
        {
            "page_number": 117,
            "text": " \n110 \n \nTable 5.12: Secure Client Transmission Checklist  \nYES  \nNO  \nDESCRIPTION  \n□  \n□  \nIs there a documented policy in place that describes what level of \ndata encryption should be used between the client-side and server-\nside components of a Web application? \n□  \n□  \nAre the administrative records maintained by the certification \nauthority (such as contact name, owner, and domain name) and kept \nup-to-date for each digital certificate used by the Web site? \n□  \n□  \nAre digital certificates renewed ahead of expiration? (Many \ncertificates must be renewed annually.) \n□  \n□  \nAre strong passwords used to protect each certificate file? With no \ntwo files using the same password? \n□  \n□  \nAre the certificate files stored under lock and key, away from the \nproduction Web site? \n□  \n□  \nIs the design of the Web application sufficiently robust to handle \nclients who may not wish to install a client-side (personal) certificate? \n□  \n□  \nIs the encryption algorithm and strength used by the Web site \nsupported by all the Web site's potential visitors? If not, is there a \nstrategy for handling those visitors that cannot communicate using \nthe preferred form of encryption? \n□  \n□  \nIs the encryption strength used by the Web site too strong (and \nresource intensive) or too weak (and potentially crackable)? \n□  \n□  \nHave all the Web site's servers been set to use the same strength of \nencryption? Using different strengths on the same Web site may \noptimize performance but may also cause additional compatibility \nproblems at the client. \n□  \n□  \nHave only the Web pages that need to be encrypted been encrypted? \n□  \n□  \nIf only portions of a Web page are to be encrypted, are only the \nappropriate components encrypted? For example, there is little \nbenefit in encrypting a Web banner advertisement. \n□  \n□  \nIs the Web site able to handle the expected volume of encrypted \nnetwork traffic now and in the foreseeable future? \n \nMobile Application Code \nMobile code refers to the chunks of code that are downloaded for execution from one \nmachine to another; in the Web world, this typically occurs as the result of a browser \nrequesting the mobile code from a Web server. \nCurrently, the majority of Web applications use one of two competing standards for \ndownloading executables over the Internet: Sun Microsystems's Java applets and \nMicrosoft's ActiveX controls. Unfortunately, these two technologies are currently \nincompatible, which means that most Web applications that use downloadable executables \ntend to standardize on one or the other of the technologies. Both standards offer similar \n"
        },
        {
            "page_number": 118,
            "text": " \n111 \nfunctionality but differ in their run-time environments; ActiveX controls can potentially be \nwritten in any programming language, but once compiled, they are primarily intended to be \nused only on Windows 32-bit platforms. In contrast, Java applets need to be written in Java \nbut may be executed on any platform that has an associated Java Virtual Machine (JVM) \n(described later in this section). In addition to executables, a browser may also download \ninterpretive code written in one (or more) scripting languages (such as EMCAScript, \nJavaScript, Jscript, or VBScript). \nWhichever technology a Web application implements, the organization needs to be \nconscious of the fact that it is using its Web site to distribute and install code to potentially \nall the visitors that frequent its Web site. An organization should therefore take every \nreasonable precaution to ensure that the code it is distributing is free of any unauthorized \nfeatures that could perform malicious activities or be subverted by another piece of code \nwith ill intentions. \nActiveX Controls \nTwo closely related security considerations only apply to ActiveX controls. At the \ndeveloper's discretion, ActiveX controls can be marked as safe for scripting, which means \nthat the developer believes that the ActiveX control is safe for any possible use of its \nproperties, methods, and events by any other program that might wish to utilize its \nfunctionality. For example, an ActiveX control downloaded from a bookkeeping Web site \ncould be instructed by a piece of JavaScript code downloaded by the browser from a tax \npreparation Web site to upload all of this tax year's income and expense records to the tax \npreparation Web site. \nThe second security consideration relates to the capability of an ActiveX control to be \ninitialized with local or remotely supplied data. The developer can optionally permit the input \nparameters of an ActiveX control to be reset (or initialized), thereby allowing a third party to \nnot only control execution of the ActiveX control but also modify the data used by the \ncontrol to perform the requested operation. For example, an ActiveX control downloaded \nfrom a computer manufacturer's Web site in order to install updates to the client machine's \noperating system could be utilized by a rogue script (inadvertently downloaded from a \nmalicious Web site) to install a Trojan horse variant of an operating system file on the client \nmachine. \nSince developers typically do not know where or how their ActiveX controls will be used, it \nis normally prudent to disable the scripting and initialization options for the control. If a \ncontrol really does need to be made available in an unprotected form, extensive testing \nshould be performed to ensure that an ill-intentioned script (possibly from another Web site) \ncould not cause the ActiveX control to do anything undesirable. This would include deleting \nfiles or emailing confidential information to another Web site. \nControl safety is ultimately a subjective judgment. However, Table 5.13 lists some Microsoft \nrecommendations for determining whether an ActiveX control can be marked as safe for \nscripting or initialization. As a general rule, none of the following undesirable effects should \nbe possible from any conceivable use of the control. \n \nTable 5.13: ActiveX Control Safety Checklist  \nYES  \nNO  \nDESCRIPTION  \n□  \n□  \nDoes the control access information about the local computer or \nuser? \n"
        },
        {
            "page_number": 119,
            "text": " \n112 \nTable 5.13: ActiveX Control Safety Checklist  \nYES  \nNO  \nDESCRIPTION  \n□  \n□  \nDoes the control expose private information on the local computer or \nnetwork? \n□  \n□  \nDoes the control modify or destroy information on the local computer \nor network? \n□  \n□  \nCan faulting the control (causing it to have an error) potentially cause \nthe browser to crash? \n□  \n□  \nDoes the control consume excessive time or resources such as \nmemory? \n□  \n□  \nDoes the control make potentially damaging system calls, such as \nexecuting another program? \n□  \n□  \nCan the control be used in a deceptive manner and thereby cause \nunexpected results? \nSource: http://msdn.microsoft.com/workshop/components/activex/security.asp  \nJava Applets \nJava was defined by Sun Microsystems and originally aimed at the set-top boxes that cable \noperators use (Java was also called Oak back then). When the Web started to take off, Sun \ndecided to redeploy the language for use over the Internet. Unlike many computer \nlanguages, Java is not defined by a standards committee but is still owned by Sun. One \nadvantage to this model is that new features, or extensions, can be added to the language \ncomparatively quickly. Of course, from a testing perspective, this can make the testing \nharder; as the number of versions increases, so does the number of potential Java versions \nthat will need to be tested. \nOne of Java's most touted benefits \"write once, run anywhere\" is accomplished by using a \nJVM installed on the machine where the program is to be run. Each platform needs a \nseparate JVM customized and optimized for the underlying platform. Each JVM converts \nthe Java code into platform-specific machine code that, although different, appears to the \nuser to execute identically on any of the JVM-supported platforms (as depicted in Figure \n5.2). Unfortunately, because the various JVMs are written by different organizations, and \nthe underlying operating systems and hardware architecture have different capabilities, \neach JVM implementation may behave slightly differently and may therefore necessitate \ntesting with multiple JVMs. \n \n"
        },
        {
            "page_number": 120,
            "text": " \n113 \n \nFigure 5.2: JVM implementation.  \nThe early versions of Java (circa 1.0) used a security mechanism called sandboxing. \nSandboxing makes untrusted code (such as a Java applet downloaded from a Web site of \nunknown trustworthiness) run inside a secure area, and it limits the applet's capability to \naccess any resource outside of the secure area. This sandbox principle is enforced on the \nmachine executing the code by three components built into most browsers: the byte code \nverifier, the class loader, and the security manager. \nByte Code Verifier \nThe verifier performs a number of static checks before the code is executed. Checks \ninclude making sure that all variables are initialized before being used and that arguments \nuse correctly typed parameters. \nClass Loader \nThe loader ensures that all classes needed by the applet are loaded and it remembers \nwhere each class was loaded from. Downloaded classes are typically regarded as \nuntrusted, while classes loaded from the local system are trusted. In addition, the class \nloader keeps classes from different applets separated. \nSecurity Manager \nThe security manager provides run-time verification that the resource requests being made \nby an applet are acceptable. Together, these three components restrict an applet's access \nto local data files and executables, network services, and a browser's internal services. One \nparticular challenge with testing Java applets is that the applet being tested may not be able \nto print to a file on the host machine. Instead, this information must be sent back to a Web \nserver for later analysis and review. \nIn hindsight, the original Java 1.0 sandbox implementation proved to be too restrictive for \nmany application environments (such as an intranet application where the applet could be \ncompletely trusted). With the release of Java 1.1, the sandbox became porous. Java 1.1 \nenabled code to be marked as trusted and subsequently allowed to run outside of the \nsandbox. Java 1.2 took the concept of trusted code a step further, using more granular \nsecurity policies established by the user or system administrator, a more flexible solution \nbut one that has the disadvantage of making security administration more complex to \nadminister. McGraw (1999) provides additional information on security issues related to \nmobile Java code. \n"
        },
        {
            "page_number": 121,
            "text": " \n114 \nClient-Side Scripts \nClient-side scripts are used by many Web applications to perform simply client-side \nprocessing, such as recalculating a shopping cart after the method of shipping has been \nchanged. In theory, the browser should ensure that any downloaded client-side scripts \ncannot do anything malicious to the machine that the browser is installed upon. \nUnfortunately, security vulnerabilities have been discovered in numerous browsers. \nTypically, the vulnerability allows a script downloaded from a rogue Web site to read some \nprivate information on the client machine (such as cached URLs, specific files stored on a \nhard drive, cookies, or keystrokes). This information is then relayed back to the script's \nauthor for him or her to digest and exploit. In addition to the client-side script advisors \nposted on the Web sites listed in Table 4.2, Georgi Guninski maintains a Web site \n(www.guninski.com) that demonstrates many of these vulnerabilities. \nDetecting Trojan Horse Mobile Code \nBecause most Web site visitors will unknowingly automatically download any mobile code \nattached to the Web page they are viewing, an organization should consider protecting \nitself against a supposedly trusted employee embedding a small piece of rouge code inside \na legitimate program, thereby creating a Trojan horse out of the organization's legitimate \nmobile code (Trojan horses are discussed in more detail in Chapter 8). \nHow can an organization ensure that a disgruntled employee (contractor or permanent) is \nnot building a Trojan horse or other form of rogue code into the mobile code hosted on its \nWeb site? Possible solutions are discussed in the following sections. \nInspections and Reviews \nAssuming a conspiracy does not exist within an entire team of developers, standard manual \ncode review and inspection practices may detect undesirable code. Even if the possibility of \nrogue code slipping through a review exists, the likelihood of it being detected may be \nsufficient to deter all but the most determined of mischief-makers. Craig (2002), Myers \n(1979), and Perry (2000) provide additional information on the technique of conducting \nsoftware inspections and reviews. \nUnfortunately, a loophole to this process exists. If configuration management practices \nallow the developer to rework their own code after the review or inspection has taken place \n(a common practice), it would be quite easy for the disgruntled developer to insert the \nmalicious code into his or her work after the source code approval process has been \ncompleted. \n \n \nTHE DISGRUNTLED DEVELOPER \nImagine a contract developer being told that his contract is being terminated early due to \ncorporate cutbacks. The developer, being upset at this development, decides to seek \nrevenge by adding a couple of lines of code to the ActiveX control he is currently working \non. This additional code is designed to delete as many files as possible on the client \nmachine's hard drive, but only if the system date is at least three months later than today. \nThe developer then recompiles the code and forwards the binary file to the testing team for \nfunctional testing. The control passes and is then uploaded to the production Web site. \nSeveral weeks after the developer leaves, the organization's customer support desk is hit \nwith a tidal wave of confused and angry customers. \n"
        },
        {
            "page_number": 122,
            "text": " \n115 \nCode Coverage \nUsing a code coverage tool such as McCabe and Associate's IQ product \n(www.mccabe.com), functional testing can be evaluated to determine what percentage of \nthe source code has been exercised with the current set of tests. Code that for some \nreason had not been exercised may then be scrutinized to ensure that the reason the code \nwas excluded was justifiable, and not part of a ploy by the code's author to prevent the \nundesirable code from being detected during testing. \nOf course, if the testing effort were only able to exercise a small portion of the code (often \nthe case in large projects, which are short on time), the amount of code that would need a \nmanual review would be so large as to be prohibitively resource intense. \nScanning Mobile Code \nMobile code-scanning tools (such as those listed in Table 8.10) can be used to exercise the \ncode in a monitored environment. The testing tools look for suspicious behavior that might \nbe indicative of rogue code being present. \nAn organization may also wish to ensure that its good code does not give off false positives \nwhen scanned by any commonly deployed firewall, intrusion detection systems (IDS), \napplication security server, or antivirus program (Chapter 8 discusses these products in \nmore detail). It's quite possible that many of their clients have installed such products and \nwould discard any mobile code that the tool flagged as being potentially harmful (regardless \nof whether the code is harmful or just appears to be). \nCode Signing \nAssuming the ActiveX controls and Java applets originally loaded on the production Web \nsite were found to be clear of malicious code, how can an organization ensure that this \ngood code is not subsequently modified and Trojanized either on the hosting Web site or in \ntransit from the Web site to the requesting client? \nCode signing is a technique that seeks to ensure that any program (ActiveX control, Java \napplet, or other downloaded executable) has not been altered since the original developer \ncompiled and packaged it, a process that is outlined in Table 5.14. \n \nTable 5.14: Code-Signing Process  \nSTEP  \nDESCRIPTION  \n1. \nThe developer writes the code. \n2. \nThe developer signs the code using a digital certificate obtained from a \ncertification authority. \n3. \nThe code is uploaded to the Web site. \n4. \nThe viewer requests the Web page containing the code and thereby \ndownloads the code. \n5. \nThe viewer is then prompted to decide whether or not to trust the \ndownloaded code (based upon how trustworthy the viewer regards the \nowner of the digital certificate). Note that depending upon the viewer's \nbrowser setting, the browser may not prompt the viewer, automatically \naccepting or rejecting the code instead. \n"
        },
        {
            "page_number": 123,
            "text": " \n116 \nTable 5.14: Code-Signing Process  \nSTEP  \nDESCRIPTION  \n6. \nIf the viewer decides to trust the code, then the code is allowed to execute; \notherwise, the code is discarded. \nUnfortunately, code-signing techniques such as Microsoft's Authenticode or Netscape's \nObject Signing do not provide any assurance that the code will not perform an undesirable \naction; they only ensure that the code has not been tampered with since it was packaged \nby the certificate's owner. It goes without saying that any code-signing certificate file should \nbe closely guarded, in order to ensure that an unauthorized person can't trick viewers into \ndownloading malicious code onto their machines, because it appears that the code was \nwritten by a trusted organization. \nConfiguration Management \nRather than embedding malicious functionality in their own code, employees with less \ndesire to be caught would probably try to plant their Trojan horse code into somebody else's \nwork. The best defense against such attempts is good configuration management \nprocedures, typically employing some form of configuration management or code \ncomparison tool (such as those listed in Table 2.2 and Table 8.7). Dart (2000), Haug \n(1999), and Leon (2000) provide additional information on configuration management \npractices. \nTable 5.15 summarizes the checks and balances that an organization could consider \nimplementing to reduce the possibility of a rogue employee (or someone else who has \naccess to the Web application's source code) inserting some malicious lines of code into \nthe mobile code portion of a Web application. \nTable 5.15: Detecting Trojan Horse Mobile Code Checklist  \nYES  \nNO  \nDESCRIPTION  \n□  \n□  \nIs there a documented policy in place that describes the security \nprocedures that should be used to ensure that any mobile code used \nby the Web site is free of any malicious code? \n□  \n□  \nAre any of the ActiveX controls used on the Web site marked safe for \nscripting or initialization? If so, are they safe? \n□  \n□  \nAre any of the Java applets used on the Web site able to function in a \nsandbox environment? If not, are the permissions that they need \nreasonable and likely to be granted by potential users of the applet? \n□  \n□  \nHas the mobile code been reviewed and inspected for malicious code \n(perhaps focusing on portions of the code that were not executed \nduring normal functional testing)? \n□  \n□  \nIs the mobile code flagged as suspicious when scrutinized by any \ncommonly used mobile code-scanning product? \n□  \n□  \nHas the mobile code been signed with an authorized digital \ncertificate? \n□  \n□  \nAre processes in place to ensure that any digital certificates used to \nsign mobile code are renewed with the certification authority before it \nexpires? \n"
        },
        {
            "page_number": 124,
            "text": " \n117 \nTable 5.15: Detecting Trojan Horse Mobile Code Checklist  \nYES  \nNO  \nDESCRIPTION  \n□  \n□  \nAre any digital certificate files used by the Web application stored in a \nsafe place? \n□  \n□  \nHave configuration management procedures been implemented that \nwould inhibit a rogue employee tampering with another employee's \ncode? \n \nClient Security \nAfter a user has been identified and authenticated as legitimate, the Web application runs \nthe risk that this (now-trusted) client machine will be successfully compromised by an \nattacker, allowing the assailant to access the application using the privileges intended to be \nbestowed upon a legitimate user. \nAlthough the onus may be upon the users (or their technical support personnel) to ensure \nthat any session that they initiate with a Web application is not hijacked via an intruder \ngaining access to their client machine, the precautions that users may seek to employ in \norder to protect their machines can have a direct impact on the functionality of a Web \napplication. A Web application should therefore be tested to ensure that it is still usable, \neven when accessed by clients with reasonable client-side security measures in place. For \nexample, an application that is completely dependent upon a user granting a Java applet \n(downloaded from the application's Web site) unfettered access to the client machine, may \nfind that a significant proportion of their users refuse to grant this code such access, and \nare consequently unable to use the Web application. \nThe following sections look at some of the precautions a user could be expected to take \nand therefore what client-side environments the testing team may wish to consider \nemulating in order to ensure that the Web application functions with these fortified-minded \nclients. \nFirewalls \nMany organizations use proxy servers to manage their employees' Internet access, improve \nperformance, and also provide their employees with a degree of anonymity. Unfortunately, \nthis level of anonymity can prove problematic to Web applications trying to authenticate a \nuser based on some piece of information that the proxy server is masking. An example of \nthis would be the proxy server replacing the network address of the client machine it serves \nwith its own. Allen (2001), Rubin (2001), and Zwicky et al. (2000) provide additional \ninformation on securing firewalls. \nAnother consideration is the filtering that a network or personnel firewall may perform. An \norganization's perimeter firewall is typically used to provide some level of protection from an \nattack originating outside the organization. A PC-based (or personal) firewall (often used in \nconjunction with an antivirus program) is often deployed to stop internal colleagues (or, in \nthe case of home users with cable modems, their neighbors) from attempting to \ncompromise machines via local network connections. Either way, communication between \nthe client-side and server-side components of a Web application may be inhibited by the \nfirewall if the Web application is designed to use unusual methods of communication (such \nas via a port other than port 80 or 443 or using a network protocol other than HTTP or \nHTTPS). Table 5.16 lists some sample personal firewalls, and Table A.3 lists some sample \nnetwork firewalls and proxy servers. \n"
        },
        {
            "page_number": 125,
            "text": " \n118 \nTable 5.16: Sample List of Personal Firewalls  \nNAME  \nASSOCIATED WEB SITE  \nBlackICE \nwww.networkice.com  \nCyberGatekeeper \nwww.infoexpress.com  \nCyberwallPLUS \nwww.network-1.com  \neSafe Desktop \nwww.esafe.com/ealaddin.com  \nFireWall \nwww.mcafee.com  \nF-Secure Distributed Firewall \nwww.f-secure.com  \nInternet Connection Firewall (ICF) \nwww.microsoft.com  \nNeoWatch \nwww.neoworx.com  \nNetfilter/iptables \nwww.netfilter.samba.org  \nNorton Personal Firewall \nwww.symantec.com  \nSeattle Firewall (ipchains) \nwww.seawall.sourceforge.net  \nZoneAlarm \nwww.zonelabs.com  \nBrowser Security Settings \nA number of browser-based technologies have the potential to do harm to a client machine \nusing a browser to access a Web application. Recognizing the concerns of their users, \nmost browser manufactures allow users of their products to disable some or all of these \npotentially dangerous capabilities (although it should be noted that these options vary from \nbrowser to browser). \nHaving a Web application that works fine on a developer's desktop where liberal security \nsettings have been used (thereby making it easier for the developer to get the software to \nwork) does not guarantee that the application will function correctly for all the clients. This is \ndue to users disabling or limiting some of their browser's capabilities for security or privacy \nreasons. \nIn the case of an intranet or extranet, users may not have any control over their choice of \nbrowser or browser security settings. This makes for a homogenous set of client machines \nthat can be easily replicated by the testing team to ensure that the Web application being \ntested will work correctly in this tightly controlled environment. However, this does not take \ninto account the issue of what happens when the organization decides to upgrade their \nversion of browser or operating system. \nUnfortunately, the same is not true for Internet-based Web applications. Although the vast \nmajority of Internet users never change any of their browser's (typically liberal) default \nsecurity settings, the larger the number of these suspect technologies that a Web \napplication uses, the more likely it is that one or more of these technologies will be (rightly \nor wrongly) disabled by a cautious user. For instance, suppose 5 percent of a Web site's \npotential audience rejects cookies, 15 percent won't let ActiveX controls run, and 20 percent \ndon't have a browser with strong encryption capabilities. Combined, these groups could \nrepresent a sizable portion of a Web site's client base, large enough that a marketing \ndepartment might be rather disconcerted to hear that such a large percentage of its \npotential customers won't be able to fully utilize the Web application. \n"
        },
        {
            "page_number": 126,
            "text": " \n119 \nEven if a particular technology is specified in the Web application's requirements (and any \ncorresponding end-user agreement), a user-friendly Web application should degrade \ngracefully (if only to issue a warning message), rather than simply not function and thereby \nconfuse and frustrate the user. \nWhichever technology is used, the Web application should be tested using different \ncombinations of browsers and operating systems with varying client-side security settings. \nTechniques such as orthogonal arrays (or all-pairs) can be used to help mitigate the \npossible combinatory explosion that such exhaustive compatibility testing may cause. \nHedayat (1999) and Kaner et al. (2001) provides more information on orthogonal arrays. \nTable 5.17 lists some tools that can be used to help pair down the number of possible \nenvironmental permutations. \n \nTable 5.17: Sample List of Orthogonal Array Manipulation Tools  \nNAME  \nASSOCIATED WEB SITE  \nAETGWEB \nwww.argreenhouse.com  \nRdExpert \nwww.phadkeassociates.com  \nStatLib (oa.c and owen) \nwww.lib.stat.cmu.edu  \nThe following are some of the most common client-side security settings that should be \nconsidered when evaluating whether or not a Web application will function correctly on a \nclient machine due to a user's security concerns. The higher (or more restrictive) the \nsecurity settings that the Web application is able to still functional correctly with, the more \nlikely the application will actually be used by its intended audience. At a minimum, consider \nchecking that the Web application can still provide expectable functionality with the default \nsecurity settings of the most commonly used versions of browsers. \nCookies \nPrimarily because of privacy concerns, many Internet users choose to block cookie use \npartially or completely. This could be a problem if the Web application's security depends \nupon this feature being enabled. \nEncryption Capabilities \nUnfortunately, not all Web servers and browsers support the same encryption algorithms \nand key lengths, in some cases because of legal considerations (for example, for many \nyears the U.S. government restricted the export of software able to support 128 or higher \nkey lengths) while other versions of Web servers and browsers support different encryption \nalgorithms and key lengths due to software vendors choosing to charge different prices for \ntheir products based on the level of encryption supported by the tool. Therefore, a highly \nsecure encryption setting may result in a substantial number of potential clients being \nunable to utilize the encrypted portion of a Web application. \nClient-Side Mobile Code \nAs previously discussed in this chapter, a great many security issues are associated with \nrunning client-side mobile code, which is one of the reasons why some browsers offer quite \nelaborate options for deciding when and with what restrictions (if any) a piece of mobile \ncode will be allowed to execute. Although the default settings of most browsers are quite \n"
        },
        {
            "page_number": 127,
            "text": " \n120 \nliberal, many organizations have instigated corporate policies that limit or completely \nprohibit the use of mobile code from an untrusted source. \nTiny Windows \nBecause very small windows or floating frames (HTML IFRAME tags) have been used to \nhide illicit activities, some browsers either prohibit them completely or enable a user to \ndisable them. \nMultiple Domains and Redirection \nIn order to improve performance and/or make content configuration management easier, \nmany Web sites actually distribute their content over multiple domain names and/or redirect \na request for one domain to another. For instance, a national newspaper may host a news \narticle that forms the main component of a page on their own site, but they use Web banner \nadvertisements from a separate online advertiser's site (such as www.aol.com or \nwww.doubleclick.net). Unfortunately, because exploits have been developed that can allow \nan intruder to trick a user into unknowingly communicating with a secondary Web site using \nredirection, some browsers enable a user to prohibit a Web page from downloading any \npage component that does not reside on the same Web site that the original resource \nrequest was made to. \nAutomatic Updates \nTechnologies such as Netscape's SmartUpdate and MS-IE's Install on Demand enable \nbrowsers to download and install missing or more up-to-date software without any user \ninvolvement. Although such technologies may be very user friendly, they also have the \npotential to be abused by a rogue Web site that would like to download a Trojaned version \nof the software. Therefore, most browsers enable this feature to be disabled by the client. \nClient Adaptive Code \nRather than trying to develop a single instance of a Web application that will work in every \nperceivable client environment, some organizations choose to develop multiple variations of \nthe same application, or a portion of an application. An example would be one version of \nthe application being designed for presentation by a Netscape browser (using proprietary \nextensions only supported by Netscape) and another especially for MS-IE (perhaps using \nfeatures only found in MS-IE). The Web application then uses information freely supplied by \nthe client (typically as part of the client's HTTP header request) to identify what platform the \nuser is using and then serves up the most appropriate variant of the application. \nTaking this concept a step further, instead of basing the adaptive code on the brand and \nversion of the browser installed on the client, the Web application could base its decision on \nwhich technologies have been enabled (or disabled) by the user. For example, the login \npage of a Web site may be used to send a cookie to the client. Upon receiving the \ncorresponding reply, the Web application can check for the existence of a returned cookie. \nThe absence of a returned cookie would imply that the user has either disabled this \ntechnology or that the browser simply does not support it. Either way, the Web application \nwill now have to employ a noncookie method of communicating with the user or issue a \nwarning to the effect that access has been denied due to this capability being disabled. \nIf a Web application uses client adaptive code that is dependent on any client-side security \nsetting or is used to implement any client-side security measure, then the testing team \nshould ensure that all the client environments that have been specifically targeted by the \n"
        },
        {
            "page_number": 128,
            "text": " \n121 \nadaptive components of the Web application are represented in the test configurations used \nto test the application. \n \nJAVA FALLBACK \nOne approach to handling Java being disabled on the client is to place an HTML-based \nwarning message between the <Applet> tags on a Web page. The message explains to the \nuser why the Java applet's functionality is not available; this HTML will be ignored if Java is \nenabled in the browser but is executed if it is turned off. \nClient Sniffing \nUnwilling to trust that a client has taken reasonable security measures to protect their \nmachine from intruders, some extremely cautious Web sites will download mobile code to \nreconnoiter (or sniff) a prospective client machine. For example, the mobile code may \ncheck for the existence of the latest operating system security patch or the most recent \nantivirus signature file (.dat file). \nOnce sniffed and assessed, the mobile code reports back to the Web application (and \noptionally the user as well) what it has found. Aside from the huge privacy issues involved \nwith such an approach, this proactive defensive measure runs the risk that a \nknowledgeable attacker could either prohibit the mobile code from being executed or even \nmodify the code to report back incorrect information. \nIf mobile code is used by a Web application for any security-related reconnaissance or \nenforcement, then the mobile code should be tested using as many of the client \nenvironments as it may be expected to be executed in. At a minimum, the mobile code \nshould be tested with browsers that have had their mobile code execution capability \ndisabled or highly restricted, or tested with a browser that simply does not support that \nparticular technology. \nTable 5.18 summarizes the checks that a testing team may wish to consider performing to \nensure that a Web application will function acceptably and that its security will not be \ncompromised when used by clients with widely varying security settings. \n \nTable 5.18: Client Security Checklist  \nYES  \nNO  \nDESCRIPTION  \n□  \n□  \nIs there a documented policy in place that describes which client-side \ntechnologies are necessary for the Web application security to \nfunction correctly? And is there a policy for how the Web application \nshould handle a client that does not support or permit one of these \nrequired technologies? \n□  \n□  \nIs the design of the Web application sufficiently robust to handle \nsome clients accessing the Web site from behind a proxy server? \n□  \n□  \nIs the design of the Web application sufficiently robust to handle \nsome clients accessing the Web site from behind a network and/or \npersonal firewall? \n□  \n□  \nIs the design of the Web application sufficiently robust to handle \nsome clients disabling cookies? \n"
        },
        {
            "page_number": 129,
            "text": " \n122 \nTable 5.18: Client Security Checklist  \nYES  \nNO  \nDESCRIPTION  \n□  \n□  \nIs the design of the Web application sufficiently robust to handle \nsome clients who do not have encryption capabilities or only have a \nbrowser capable of weak encryption? \n□  \n□  \nIs the design of the Web application sufficiently robust to handle \nsome clients disabling mobile code? \n□  \n□  \nIs the design of the Web application sufficiently robust to handle \nsome clients disabling tiny windows or floating frames? \n□  \n□  \nIs the design of the Web application sufficiently robust to handle \nsome clients not wanting to work across multiple domains or enable \nredirects? \n□  \n□  \nIs the design of the Web application sufficiently robust to handle \nsome clients not wanting to allow automatic software updates? \n□  \n□  \nIf client-adaptive code is used by the Web application, do the \napplication's security capabilities still work with all client-side \nenvironments? \n□  \n□  \nIf client-sniffing is used by the Web application for any security \nprecautions, can the application's security be compromised by \ndisabling or restricting mobile code execution on the client? \n□  \n□  \nIf client-sniffing is used by the Web application for any security \nprecautions, can the mobile code be easily reverse engineered or \nmodified in such a way as to send bogus security information back to \nthe server-side component of the application? \nIf the Web application is found to frequently only work in an extremely liberal security \nenvironment, then the organization may wish to reconsider deploying the application in its \ncurrent state, potentially reworking the application to make it work with less demanding \nclient-side environmental requirements. \n \nSummary \nAt a simplistic level, the client-side application security tests described in this chapter can \nbe grouped into two categories. The first consists of those that are designed to ensure that \nthe Web application's security capabilities function correctly for the good guys, providing \nacceptable usability, performance, and compatibility. The second set of tests (that are often \nnot as frequently employed) attempt to establish that the application's security \nimplementation is robust enough to stop the bad guys from trying to gain unauthorized \naccess to the application's resources (data, functionality, or connectivity). \nWhat percentage of its resources the testing team should spend on positive testing (making \nsure it works) for the good guys, versus negative testing (making sure it can't be broken) for \nthe bad guys, will vary from project to project. To a large degree, it will be influenced by \nhow important it is to try and keep the good guys happy, versus keeping the bad guys \nunhappy. \n"
        },
        {
            "page_number": 130,
            "text": " \n123 \nChapter 6: Server-Side Application Security \nOverview \nWouldn't it be nice if a Web site's perimeter defenses completely protected it from all \nexternal intruders? And wouldn't we all rest easier if we knew that a Web application's \nclient-side components would always ensure that users only submit nonmalicious input to \nthe application's server-side components? The reality, of course, is not quite as idyllic. \nUnfortunately, a supposedly well-configured perimeter firewall may have an as-yet-\nundiscovered hole or be completely circumvented by an employee attacking the Web site \nfrom a machine located behind the perimeter firewall. And since few organizations have any \ncontrol over the machines that Internet users will use to access the Web application, there \ncan be no guarantee that any client-side checks have actually taken place or been done the \nway they were intended to be performed. For these reasons alone, a Web application \ndesigner would be well advised to include additional security precautions on the server-side \nas well. These precautions should be checked by the testing team to make sure that they \nhave been implemented correctly. \nBreaking through a multilayer defense that employs safeguards at each and every juncture \nis much tougher than breaking through a single layer (such as a Web site that solely relies \nupon a perimeter firewall). The previous chapter focused on testing the security of the \nclient-side component of a Web application. This chapter will focus on what needs to be \nchecked on the server side by looking at the different technologies that may have been \nused. It will then examine the vulnerabilities associated with each technology and \nconsequently what features the testing team should check to determine if a specific \nimplementation is vulnerable. \nThe sections in this chapter can be conceptually grouped into three topics. The first group \nfocuses on the vulnerabilities associated with programming technologies that are specific to \nWeb implementations, such as Common Gateway Interfaces (CGI), Server Side Includes \n(SSI), Active Server Pages (ASPs), and Java Server Pages (JSP), and then general \nvulnerabilities that are applicable to all server-side application code. The second grouping \nlooks at what can be done to guard against invalid input data being received by the Web \napplication and then subsequently what defenses can be put in place to protect legitimated \ndata stored on the Web site. Finally, the last section in this chapter looks at the \ncontroversial topic of using client behavioral patterns to detect compromised application \nsecurity. \n \nCommon Gateway Interface (CGI) \nCGI is a protocol (not a programming language). Many Web applications use the CGI \nprotocol to enable a Web server to communicate (pass data) with other programs running \non the Web server or another server such as an application or database server. Figure 6.1 \ndepicts this situation. \n"
        },
        {
            "page_number": 131,
            "text": " \n124 \n \nFigure 6.1: CGI implementation example.  \nUnfortunately, \nmany \nknown \nsecurity \nvulnerabilities \nare \nassociated \nwith \nCGI \nimplementations, and therefore vulnerabilities that the testing team should check for, should \nthe Web application's designers decide to use CGI in their architecture. \nLanguage Options \nA CGI implementation may be written in a scripting language such as Perl or less \ncommonly in a compiled language such as C. Table 6.1 lists some of the languages that \nmay be used to write CGI applications. \nTable 6.1: Possible CGI Application Languages  \nSCRIPTING  \nCOMPILED/PSUEDOCOMPILED  \nAppleScript \nC \nAWK \nC++ \nJavaScript \nC# \nPerl \nCOBOL \nPHP \nJava \nPython \nJ++ \nRuby \nJ# \nTcl \nPowerBuilder \nVBScript \nVisual Basic \nVarious shells[a]  \n  \n[a]Shells are a group of operating-system-interpretive languages (such as bash, bourne, \ncsh, sh, and DOS) that give a user the ability to invoke basic operating system utilities \nfrom a command line or via a batch script file. \nThe following points seek to compare and contrast some of the typical differences between \ncompiled CGI implementations and interpretive ones: \nƒ \nA compiled version is likely to execute faster and consume fewer resources than an \nequivalent script being parsed and then executed by an interpreter. \nƒ \nA script must have a resource-consuming interpreter installed on the server that is \ngoing to execute it, with each interpreter requiring additional resources. \n"
        },
        {
            "page_number": 132,
            "text": " \n125 \nƒ \nThe object code (binary executable) of a compiled application provides an attacker \nwith fewer opportunities (even using a code reverse-engineering tool) to understand a \nprogram's logic and thereby discover a security hole than the source code of an \nequivalent script. Note that this assumes that the source code used by the compiler to \ngenerate the executable is removed from any machine to which the attacker might be \nable to gain access. \nƒ \nRunning a program through a compiler will result in the entire program's source code \nbeing syntax-checked before being executed, as opposed to interpreters that typically \nperform their syntax checks on-the-fly as the script is being executed (which is one of \nthe reasons why compiled code normally runs faster than interpretive code). Although \nsome interpreters have the capability to parse a script ahead of time, the optionality of \nthis check leads some developers to skip this step when deploying interpretative \nscripts, possibly allowing errors that would have been caught by a compiler to make \ntheir way into production. \nƒ \nThe strict data-typing syntax often used by compiled languages may make it harder \nfor an attacker to pass off malicious input (such as system commands) as legitimate \nuser input. \nƒ \nThe syntax of a scripting language is often simpler and easier to learn than a \ncompileable language. \nƒ \nA script need not first be compiled and linked before it is ready to run, potentially \nreducing development time. \nƒ \nConfiguration management is simpler for scripting code because only one version of \nthe code exists, as opposed to compiled code where two versions are used (object \ncode and source code). \nƒ \nOnce a program is compiled, it becomes platform dependent; the resulting \nexecutable only runs on the platform for which it was compiled. In contrast, a script \nshould (in theory) be easily portable from one platform to another with few or no \nmodifications. \nFrom a security perspective, an inherent advantage exists in standardizing on a single \nlanguage when implementing a Web application's CGI needs. For example, the more CGI \ninterpreters that are loaded on to a Web server (an interpreter is needed in order to execute \na script), the greater the chance for misconfiguration and the subsequent possibility of an \nattacker exploiting this installation error. Also, standardizing on a single language improves \nthe probability that a CGI programming error will be caught during a program review (as all \nthe developers speak the same language). It also increases the usefulness of safe common \nroutines (which have been extensively examined and tested for any security vulnerability). \nInput Data \nA CGI program can receive input from a Web server through a variety of channels (for \ninstance, via command-line arguments, by requesting environmental information such as \nthe date and time, or being piped information from another program). Unfortunately, CGI \nprograms are all too often not set up to handle input error conditions very well. \nBad input to a CGI program is a particular problem when the input data is used as a \nparameter in a subsequent CGI system call, the CGI program misinterpreting the input data \nas a system command and subsequently executing the command instead of treating it as \ndata. Therefore, before being used, the input data should always be scanned to ensure that \nit only contains legitimate characters (easier in some languages than others) and, if \npossible, that it is never actually used as a parameter in a subsequent CGI system \ncommand. \n"
        },
        {
            "page_number": 133,
            "text": " \n126 \nTo mitigate the potential problem of a CGI program inadvertently treating input data as a \nsystem command, an organization may wish to consider using a combination of code \ninspections and destructive testing to ensure each CGI program is robust enough to handle \ndata input errors correctly. (These techniques are described in more detail later in the Input \nData section of this chapter.) The Web Design Group (www.htmlhelp.com) provides a CGI \ntest harness (cg-eye) that can be used to assist the destructive testing. \n \nTAINTED DATA \nPerl interpreters come with a useful data tainting option for making sure that any user input \ndata is first inspected before being used in a potentially dangerous situation. When this \nenvironmental variable is set, the interpreter will not enable the tainted data to be used as a \nparameter for any high-risk system command (such as a path name, file name, or directory \nname parameter). To confirm that this feature is being used, inspect the beginning of the \nscript file to look for the line that invokes the Perl interpreter. The -T parameter must be \npresent in order to force the interpreter to use its tainted data functionality. An example \nwould be # ! /usr/bin/perl -T. \nPermissions and Directories \nA CGI program inherits the security permissions of the process that invokes it, because of \nan attacker's potential to trick a CGI program into executing a malicious command; the \nowning process should be assigned a minimal set of privileges, thereby limiting the \npotential impact of an attack. Since the privileges that the owning process is granted are \ndetermined by the user account it is run under, this process should be assigned to a lowly \nuser account (and definitely not a system administrator account such as root, superuser, or \nadministrator). \nUnder normal circumstances, there should be no reason to permit a Web application's user \nto upload (or download) a CGI program over the Internet (and should therefore be blocked). \nIf, by chance, this capability were to be unintentionally available (for instance, because the \nFTP service was unintentionally left running on the Web server), an intruder could use this \ncapability to upload a rogue CGI program (possibly designed to email the attacker sensitive \ndata files). Then, once installed into an appropriately privileged directory, the rogue CGI \nprogram could be executed by simply requesting it from a browser. Allowing a CGI program \nto be downloaded would make it easier for an intruder to examine the code and potentially \nfind a security hole (such as unscreened input data being directly used in a system call). \nRunning the tests outlined in Table 4.12 should detect the presence of any such potentially \nruinous service running on the Web server, and thereby deny an intruder the opportunity to \ncommit this easy exploit. \n \nHTTP HEADERS \nExamining the HTTP Referrer field in a CGI request is one method that some Web sites \nuse to detect suspicious CGI invocation attempts, as this environmental variable can be \nused to check that the CGI program in question was requested from a legitimate Web page \n(and not an attacker's editor). Although not foolproof (due to the ease with which the HTTP \nReferrer field can be faked), this technique does add one more layer of protection for the \nWeb application. \nIf employed, simply cutting and pasting a legitimate invocation into a test harness Web \npage and then attempting to invoke the CGI script from this illegitimate starting point can \nverify the correct implementation of this defense. \n"
        },
        {
            "page_number": 134,
            "text": " \n127 \nSome Web servers can be configured to use multiple directories to store their CGI \nprograms, while others require files with an extension of .cgi to be placed in a directory \ncalled cgi-bin. Since it is much easier to administer a single directory than a distributed set, \nthe chances of a system administrator making the mistake of assigning the wrong security \nprivilege to a directory is much lower, thereby making a single directory the preferred choice \nof most webmasters. Of course, keeping all the CGI programs in a single directory can also \nmake it easier for an attacker to find these scripts, especially if the directory name is cgi-bin. \nEither way, the security permissions should be checked to ensure that the CGI programs \nare granted minimum access privileges. \nIf an interpreter has been installed on the Web server, installing the interpreter into its own \ndirectory and not the directory containing the CGI programs that it will execute permits a \nWebmaster to grant different (and thereby more restrictive) directory and execution \npermissions to the interpreter and its programs. \nOne way to test the collection of CGI programs to see if the security permissions that they \nhave inherited are too generous is to create a short CGI program (for example, called \ntestcase.cgi) and upload it to the Web server to be tested. The CGI program should contain \none or more commands (written in the appropriate language for the Web site) that attempt \nto perform an operation that (in theory) should be blocked by the server's security settings. \nFor example, adding a new subdirectory to the root directory of the Web server's operating \nsystem. \nIf, after requesting this CGI program via a browser (for instance, by entering the URL \nwww.wiley.com/cgi-bin/testcase.cgi), the program is successfully executed and the new \ndirectory is created, then a further inspection of the Web server's security settings would be \nwarranted. \nScalability \nTypically, each time a legitimate user requests a CGI program to be run, a unique copy of \nthe script is loaded into the Web server's memory. Because of the additional resources \nused by each CGI invocation, CGI programs have been known to have scalability issues \nand are therefore potential candidates for denial-of-service attacks. A CGI-enabled Web \nserver should therefore be performance tested to ensure that it has sufficient resources to \nhandle (in an acceptable manner) an unexpectedly large number of CGI requests. FastCGI \nis a more advanced implementation that reduces some of CGI's inherent scalability issues. \nFor more information on how FastCGI works, visit www.fastcgi.com. \nTable 6.2 lists the checks that a testing team may wish to consider utilizing if the Web site \nthey are testing supports CGI. Traxler (2001) provides additional information on CGI \nsecurity issues. \n \nTable 6.2: CGI Checklist  \nYES  \nNO  \nDESCRIPTION  \n□  \n□  \nIs there a documented policy in place that describes when and how a \nCGI program may be used? \n□  \n□  \nIf compiled CGI programs are used, are the original source code \n(often located in the cgi-src directory) and any backup copies \nremoved from the production Web server? \n□  \n□  \nAre old CGI programs from previous releases removed from the \n"
        },
        {
            "page_number": 135,
            "text": " \n128 \nTable 6.2: CGI Checklist  \nYES  \nNO  \nDESCRIPTION  \nproduction Web server as soon as they are no longer needed? \n□  \n□  \nAre all CGI inputs first scanned for inappropriate input characters? \n□  \n□  \nIf programming standards call for HTTP header analysis, is it always \nimplemented? \n□  \n□  \nHave only the bare minimum of system privileges been assigned to \nthe process that will run the CGI programs? \n□  \n□  \nDoes the Web site degrade gracefully when an exceptionally large \nnumber of CGI program invocations are requested? \n \nCGI ALTERNATIVES \nAs an alternative to CGI, Microsoft has developed the Internet Server Application \nProgramming Interface (ISAPI). In contrast to a CGI implementation, an ISAPI application is \nonly loaded once. This is possible because ISAPI is designed to enable multiple threads to \nrun within a single instance of the application and therefore does not have the same \nscalability issues that CGI programs have. The Netscape Server API (NSAPI) and Apache \nAPI provide similar alternatives to CGI on Netscape/iPlanet and Apache Web servers. \n \nThird-Party CGI Scripts \nSome CGI programs that come with third-party software (such as Web servers) have been \nfound to contain security holes. For example, the PHF script was a standard CGI script that \ncame with some early versions of the NCSA, Apache, and Netscape Web servers (and has \nsince been fixed). Unfortunately, this script did not parse its input data thoroughly \n(permitting the line feed control character to remain in any input data). Once this exploit \nbecame known, attackers were then able to exploit this feature by tagging an operating \nsystem command to the end of a request to the phf.cgi script. For example, if the PHF script \nwere located in the cgi-bin directory, the following string would instruct the operating system \nto send the contains of the password file back to the requester: \n  http://wiley.com/cgi-bin/phf?Qalais=x%OA/bin/cat%20/etc/passwd \n  Note: The space character is represented as 20 In Hexadecimal \nASCII \n  And line feed is represented as 0A \nPHF has become such a classic exploit that tools have been specifically designed to scan \nfor it (such as phfscan), and drop-in replacements act as tripwires for intruder detection \nsystems to detect when it's requested (tripwires and intrusion detection systems are \ndiscussed in Chapter 8). \nRather than searching for a single CGI program, CGI scanners can specifically scan for the \npresence of numerous flawed CGI programs. The attacker can then subsequently \ncustomize his or her attack based on the CGI vulnerability detected at the target Web site. \nKlevinsky et al. (2002) and Skoudis (2001) provide additional information on how to use \nCGI scanners. Table 6.3 lists some sample CGI scanners. \n \n"
        },
        {
            "page_number": 136,
            "text": " \n129 \nTable 6.3: Sample List of CGI Scanners  \nNAME  \nASSOCIATED WEB SITE  \nAdvanced Administrative Tools \nwww.glocksoft.com  \nCgicheck \nhttp://packetstorm.decepticons.org  \nCgichk \nwww.sourceforge.net  \nCgiscan \nwww.nebunu.home.ro  \nNessus \nwww.nessus.org  \nWebscan \nwww.atstake.com  \nWhisker \nwww.wiretrap.net  \nZcgiscan \nwww.point2click.de  \nNote that many of the security assessment tools listed in Table 9.4 include a CGI \nscan in their assessment. \nBecause of known and (as yet) undiscovered security defects with third-party CGI \nprograms, it's generally considered a good idea to remove (or, better still, not install) any \nvendor demo scripts, training scripts, or utility scripts that the Web application does not \nabsolutely need. If third-party CGI programs are to be used, each CGI program found as \nthe result of a wildcard file search on each Web server should be thoroughly researched to \nensure that it doesn't have any known vulnerabilities. An example would be searching for \ncgi files from a server's root directory through all subdirectories with the wildcard string \n*.cgi. (Table 4.2 provides a list of alert centers that track such vulnerabilities.) In addition, \nthe testing team should consider running at least one CGI scanner against each Web \nserver to ensure that no vulnerable CGI program inadvertently got installed. \nAdding a custom HTTP 404 error page (resource not found) to a Web site can cause many \nof the simple CGI scanning tools to return false positives for all of their probes, as these \nscanners typically determine the presence of a flawed CGI program based on receiving any \npositive response (such as a custom \"Requested resource does not exist\" Web page) from \nthe Web server. The scanning tool's report is effectively made useless, as every possible \nexploit will be falsely reported as being present. Table 6.4 summaries the third-party CGI \nchecks that should be considered. \n \nTable 6.4: Third-Party CGI Script Checklist  \nYES  \nNO  \nDESCRIPTION  \n□  \n□  \nIs a documented policy in place that describes which third-party \nscripts (if any) may be installed on a Web server? \n□  \n□  \nHave all third-party CGI scripts that will be installed been researched \nto make sure that no known security issues exist with the CGI script? \n□  \n□  \nCan any of the widely available CGI scanners detect the presence of \na flawed CGI script? Note: This may involve temporally disabling any \ncustom error pages used by the Web site. \n \nHELPFUL SEARCH ENGINES \n"
        },
        {
            "page_number": 137,
            "text": " \n130 \nOne technique that attackers have used to locate Web sites with vulnerable CGI programs \ninstalled is to use the services on a regular Internet search engine. These search engines \nsometimes come across these CGI programs as they index (spider) the Web site, and they \nmay provide a list of these identified sites if the CGI program is used as the input query to \nthe search engine. \n \nServer Side Includes (SSIs) \nServer Side Includes (SSI) are placeholders (or markers) in an HTML document that the \nWeb server will dynamically replace with data just before sending the requested document \nto a browser. Should viewers review the resulting HTML from their browsers, they will \nprobably not be able to tell if the HTML was static (fixed) or had been dynamically built by \nthe Web server as it parsed the HTML source code. This is because the SSI commands \nwould have been replaced with the actual contents of the SSI placeholder. \nHTML source code files containing SSI commands are often saved with the .shtml suffix \ninstead of the usual .html (or .htm) suffix to ensure the Web server parses the HTML code \nwith an SSI interpreter that looks for SSI commands and replaces these markers with the \ncorresponding data referenced by these commands. Different Web servers support slightly \ndifferent implementations of SSI. However, Table 6.5 lists six of the SSI commands that are \ntypically available. \n \nTable 6.5: Common SSI Commands  \nNAME  \nDESCRIPTION  \nConfig \nSpecifies the format of data sent to the browser. \nEcho \nRetrieves the value of an environmental variable. \nExec \nExecutes a shell command or CGI application. \nFlastmod \nRetrieves the last modified date for a specified file. \nFsize \nRetrieves the file size for a specified file. \nInclude \nIncludes the contents of another file into the current file. \nThe Include command can be particularly useful to a developer wanting to reuse standard \ncomponents in multiple Web pages; the common code is dynamically cut and pasted into \nthe requested Web page at the point indicated by the Include statement. For example, let's \nsay a copyright.inc file contains an organization's standard copyright notice that appears at \nthe bottom of every page on its Web site. This enables the organization to change the \nwording of its copyright notice across every Web page on the site by simply updating this \nsingle file. Thus, the following HTML code \n \n  <HTML> \n  <HEAD><TITLE>Show SSI at work</TITLE></HEAD> \n  <BODY> \n  <P>Lots of really Interesting stuff to read</P> \n  <!--#Include file = \"copywrite.Inc\"--> \n  </BODY> \n  </HTML> \n"
        },
        {
            "page_number": 138,
            "text": " \n131 \nwould be parsed by the Web server before it is sent to a requester and cause the following \nHTML to actually be sent to a requesting browser: \n \n  <HTML> \n  <HEAD><TITLE>Show SSI at work</TITLE></HEAD> \n  <BODY> \n  <P>Lots of really Interesting stuff to read</P> \n  <P>Copyright 2002 wiley.com all rights reservered</P> \n  </BODY> \n  </HTML> \nThe danger with an Include command comes when an intruder is able to manipulate a Web \npage into including a file that would otherwise not be available. For example, if an intruder \nis able to gain write access to a directory on a Unix Web server (possibly a .temp directory \nthat didn't have any sensitive information stored in it and was therefore not locked down), \nthe intruder could upload a .shtml Web page containing the following include statement: \n  <!--- #exec cmd=\"/bin/cat /etc/passwd\" ---> \nBy subsequently requesting the uploaded .shtml file from a browser, the intruder causes the \nWeb server to parse the file and attempt to substitute the include statement with the \npassword file using its authority to read the requested file. \nAnother security risk associated with the Include command occurs when executable code is \nplaced inside a file that will be included in another file. An attacker could download the \nsource code for an included file by directly entering its URL in a browser and then download \nthe file without the source code first being parsed by the SSI interpreter. For example, the \ninclude file getHITcounter.inc on the wiley.com Web site could be downloaded by entering \nthe following URL via a browser: www.wiley.com/common/getHITcounter.inc. Being able to \nreview this source code may allow an attacker to discover some internal secret about the \napplication that the developer thought no one would ever see. For example, a database \nconnection string, together with userIDs and passwords, is a common use for include files \n(a problem expended on later in this chapter). \nThe Exec command has even more issues associated with it, instead of simply being used \nto surreptitiously download privileged information. The Exec command can be used to \ninvoke existing (or covertly uploaded) CGI programs or any valid shell command. For \nexample, DOS shell commands such as Ver (which displays the version of the operating \nsystem), Dir (which lists the contents of a directory), or Ipconfig (which displays network \naddress information) would all provide useful information to an intruder. \nIf the SSI Exec command capability is enabled on a Web server, then the only challenge \nthat an intruder faces is trying to figure out a way to trick the Web server into executing a \nmalicious command. The following are two possible ways an intruder could accomplish this: \nƒ \nUploading an .shtml Web page containing the malicious Exec command to the Web \nserver, and then requesting it via a regular browser (in the same way the Include \ncommand could be abused). \nƒ \nEntering the SSI command into a regular data entry field on a legitimate Web page \nand hoping that the input from this Web page is reused in a subsequent Web page, \nthereby causing the command to be executed. An example would be the mailing \naddress field on a Ship to Web page being redisplayed on the Order Confirmation \nWeb page. \n"
        },
        {
            "page_number": 139,
            "text": " \n132 \nWhen programs referenced by an Exec command are executed, they run with the system \nprivileges of the Web service. A Web site could incur significant damage if an intruder were \nable to install a virus, or Trojan, and then invoke it via an SSI Exec command running with \nadministrative privileges. Therefore, if the Exec command is enabled, the Web service \nshould be checked to make sure it is running with the fewest privileges possible (as \ndescribed in the sidebar \"SSI Configuration Test\"). \nBecause the security risks may outweigh the programming benefits of enabling SSI \n(especially the Exec command), many Web sites simply choose not to enable the SSI \ninterpreter at all. If this is the case, then each Web server should be inspected to ensure \nthat this feature has indeed been disabled. In the case of an Apache Web server, this can \nbe achieved by using the includesnoexec option in the server's configuration file \n(httpd.conf). To block this feature on an IIS Web server, the SSIEnableCmdDirective entry \nmust not be present in the Web server's Windows registry. \n \nSSI CONFIGURATION TEST \nTo check if SSI has been disabled, create an HTML/SSI file called testssi.shtml (the.shtml \nsuffix is used to ensure that the Web server knows the HTML file needs to be parsed by the \nSSI interpreter) with the following lines of code: \n<HTML> \n<HEAD><TITLE>Test SSI Configuration</TITLE></HEAD> \n<BODY> \n \n<H1>Test 1 - Is SSI Enabled?</H1> \n<P>This file was last modified on \n<!--#flastmod file = \"testssi.shtml\" --></P> \n \n<H1>Test 2a - Is SSI Exec Command Enabled on an \nApache/Unix server?</H1> \n<P>Today's date is <!--#exec cmd = \"/bin/date\" --></P> \n \n<H1>Test 2b - Is SSI Exec Command Enabled on an \nIIS/Windows server?</H1> \n<P>Today's date is <!--#exec cmd = \"c:\\testssi.bat\" --></P> \n \n<!--Note: For the IIS test to be successful, the \ntestssi.bat file must contain a valid shell \ncommand like \"date /t\" and the .bat and .cmd file \nextensions must be mapped to the cmd.exe for this \nWeb server. \n--> \n \n</BODY> \n</HTML> \n"
        },
        {
            "page_number": 140,
            "text": " \n133 \nIf SSI is enabled, the Web server will attempt to parse any text that starts with <!--# (and \nends with --->). The flastmod SSI command will retrieve the date when the specified file (in \nthis case the test file itself) that was last modified can be used to determine if SSI is \nenabled. \nThe SSI Exec command will attempt to run the specified shell command (in this case the \nUNIX date function or the Windows .bat file) and can be used to determine if the Web \nserver has been configured to permit SSI Exec commands: \nƒ If SSI is disabled or is unavailable, both commands will fail. \nƒ If SSI is enabled, but the Exec command has been selectively disabled, then only \nthe flastmod command will work. \nƒ If SSI is enabled and the Exec command has not been restricted, then both \ncommands should work. \n \nTable 6.6 summaries the SSI configuration checks that a testing team may wish to consider \nrunning. \n \nTable 6.6: SSI Configuration Checklist  \nYES  \nNO  \nDESCRIPTION  \n□  \n□  \nIs there a documented policy in place that specifies which SSI \ncommands (if any) will be enabled on the Web server? \n□  \n□  \nIf SSI is not to be enabled, has each Web server been checked to \nensure that it has not actually been enabled? \n□  \n□  \nIf any SSI feature is to be enabled, has the Web service on each \nserver been checked to ensure that it is running with minimal \nprivileges? (This is a prudent measure to take even if SSI is not \nused.) \n□  \n□  \nIf only the SSI Exec command is to be disabled, has each Web \nserver been checked to ensure that this feature has been disabled? \n \n \nDynamic Code \nInstead of using an SSI statement to include a small section of code into a static Web page, \ndynamic code can be used to build an entire Web page on the fly (dynamically). Dynamic \ncode can be crudely thought of as programming code used to dynamically construct other \nprogram code (such as using some VBScript on a server to create the HTML that will be \nsent to a browser for rendering). \nDynamic code is particularly useful for Web pages whose structure will change based upon \nuser interactions or frequently changing content. For example, a shipping confirmation Web \npage may contain a line entry for each item the viewer has purchased. Using traditional \nstatic Web pages, a developer may have to code a separate Web page to handle a visitor \npurchasing one item, two items, three items, and so on, or have a single, fixed-length Web \npage long enough to handle the largest of orders. With dynamic code, the developer can \nsimply write a template that loops around, building the shipping confirmation Web page by \nrepeatedly inserting the HTML code for a single line item until the order is complete. \n"
        },
        {
            "page_number": 141,
            "text": " \n134 \nWhen a Web server receives a request for a resource that is marked as a dynamic code \ntemplate (such as a file with a .jsp suffix), the Web server automatically knows to use the \nresource as a template for building the Web page that will ultimately be sent to the \nrequestor (rather than actually sending the template itself). The Web server parses the \ntemplate with the appropriate dynamic code interpreter, executing any dynamic code it \nfinds, before sending the template's output to the requestor. \nAlthough dynamic code makes developing Web sites with nonstatic content much more \nfeasible, it does have some security issues. These issues are typically environment \nspecific. For example, an ASP security issue is unlikely to be a problem in a PHP \nenvironment. Table 6.7 lists some common dynamic code environments. \n \nTable 6.7: Sample Dynamic Code Environments  \nNAME  \nASSOCIATED WEB SITE  \nActive Server Pages (ASP) and ASP.NET (ASP+) \nwww.microsoft.com  \nColdFusion \nwww.macromedia.com  \nContentSuite \nwww.vignette.com  \nInstant ASP \nwww.halcyonsoft.com  \nSun Chili!soft ASP \nwww.chilisoft.com/sun.com  \nJava Server Page (JSP) \nwww.sun.com  \nPersonal Home Page (PHP) \nwww.php.net  \nThe details of the most recent security issues that apply to a specific dynamic code \nenvironment can be identified by visiting the incident-tracking Web sites listed in Table 4.2. \nHowever, a large portion of these incidents can be considered to belong to one of the \nfollowing general categories of vulnerability. \nViewing the Template \nSome of the dynamic code environments have been found to contain defects that, unless \npatched, allow an intruder to trick the Web server into not processing the template with the \nassociated interpreter. This causes the template's source code to be downloaded, rather \nthan the code that should have been generated by executing the template. For instance, \nearly versions of IIS could be persuaded to cough up ASP source code by merely adding a \nperiod to the end of a URL (for example, wiley.com/resource.asp.). The danger with \nallowing intruders to view the source code is that they may be able to find sensitive \ninformation (such as in the following example ASP code) or a weakness in the template's \ndesign that they could then exploit: \n  <% \n  ' Use this when you need to establish a connection to the \n  database. \n  data_source  = \"primedb\" \n  data_user    = \"maintrader\" \n  data_password        = \"m15A384$G\" \n \n  DSN = \"Data Source=\" & data_source & \";User ID=\" & data_user \n"
        },
        {
            "page_number": 142,
            "text": " \n135 \n  & \";Password=\" & data_password & \";\" \n  %> \nTemplates should therefore be checked to ensure that sensitive information that could be \nuseful to an attacker is not being hard-coded into the templates (such as userIDs, \npasswords, directory and file structures, and database schemas) and could be potentially \nviewable to an intruder via an unpatched security hole. \nSingle Point of Failure \nUnlike CGI programs that often run as separate processes, a dynamic code environment \nmay have the template interpreter running as a single process. If the entire Web site is \nbeing processed by this single (multithreaded) process, then it may be possible for an \nattacker to bring down the Web site by submitting input data that, when processed by the \ninterpreter, causes the interpreter to crash, effectively creating a denial-of-service attack on \nthe entire Web site. An example of this would be a sufficiently large input string that causes \na buffer overflow error. Hence, the need exists to test any dynamic code to ensure that it is \nrobust enough to handle any erroneous input that could possibly cause it to fail in a \ncatastrophic manner (a topic discussed in detail in the Input Data section of this chapter). \nSystem Commands \nIn the same way that an intruder could trick an SSI interpreter into executing SSI \ncommands, so can dynamic code interpreters be tricked into executing the system \ncommands that are passed to the Web site via a data input field on a legitimate Web page. \nTherefore, any input data received from the client should always be parsed to ensure that it \ndoes not contain any covert system commands that might be inadvertently executed by a \ndynamic code interpreter (a topic covered in more detail later in this chapter). \nDemonstration Scripts \nVendors of dynamic code environments will often include example dynamic code in their \ndefault installations. Unfortunately, as mentioned previously in regard to CGI programs, \nsome of these scripts may not have been exhaustively tested and may therefore contain \nfeatures that an intruder could exploit. For this reason, many Web sites remove any \nnonessential demo, training, and utility dynamic code that may have been installed during \ninstallation. \nHelpful Error Messages \nEnvironments should be checked to ensure that if the dynamic code interpreter encounters \nan error, they don't give up too much detailed information about the internal workings of the \nsystem. For example, IIS/ASP can be configured by the Webmaster to provide different \nlevels of error messages when it encounters an error, the most detailed of which is useful \nduring development, but has a higher security exposure in a production environment. Table \n6.8 summaries the dynamic code security checks that a testing team should consider \nconducting. Traxler (2001) provides additional information on testing some of the more \ncommon dynamic code environments. \n \nTable 6.8: Dynamic Code Checklist  \nYES  \nNO  \nDESCRIPTION  \n□  \n□  \nIs there a documented policy in place that specifies which dynamic \n"
        },
        {
            "page_number": 143,
            "text": " \n136 \nTable 6.8: Dynamic Code Checklist  \nYES  \nNO  \nDESCRIPTION  \ncode interpreters (if any) are to be installed on the Web server? \n□  \n□  \nHave all the known security issues been researched and documented \nfor the specific dynamic code environment that will be installed? \n□  \n□  \nHave any security patches and workarounds necessitated by a \nknown security issue been implemented on every Web server? \n□  \n□  \nHave all templates been reviewed for the presence of hard-coded \nsensitive information? \n□  \n□  \nHave all unneeded dynamic code scripts been removed from the \nproduction environment? \n□  \n□  \nHas the dynamic code environment been checked to ensure that it \ndoes not provide too detailed information about the application, in the \nevent that an application error occurs? \n \nApplication Code \nClient-side source code, such as HTML, can be expected to be seen by a Web site's \nvisitors. Simply click View, Page Source (Netscape), or View, Source (MS-IE). However, \noften the amount of client-side code needed to build a Web page is so voluminous that a \ndeveloper might doubt that anyone other than him- or herself could be bothered to read this \nsource code. Unfortunately, this assumption can lull some developers into a false sense of \nsecurity, allowing them to include sensitive information in the source code that they would \nnot normally do, had they thought that an intruder would read this information. \nThe same is true for server-side interpretive code (CGI scripts, SSI includes, ASP code, \nand so on), where the developer assumes no one other than the webmaster would see this \ncode. But as can be seen from some of the exploits previously discussed, it should not be \nassumed that any source code (client-side or server-side) will not be reviewed by an \nintruder. There are, however, a few precautions that can be practiced to minimize the \namount of useful information an intruder can glean, should he or she gain access to an \napplication's code base. \nCompileable Source Code \nShould a Web server become compromised, needlessly storing source code used to \ncompile executables on the Web server may provide an intruder looking for sensitive \ninformation (such as userIDs and passwords), with the opportunity to review this code. Yet \nthis opportunity can easily be denied by only storing the executable (object code) on the \nproduction server. Note that although code reengineering tools can reconstruct source code \nfrom object code, this computer-generated source code is often devoid of comments and \nmeaningful naming conventions, typically making the code extremely hard to read and \ncomprehend. \nNoncompileable Source Code \nWell-commented source code is often a sign of a quality-minded developer and an \nadmirable characteristic, because these comments provide assistance to other developers \n(or even the original author) trying to understand how the code works. Unfortunately, \ncomments left in the production version of any interpretive (noncompileable) code (client-\n"
        },
        {
            "page_number": 144,
            "text": " \n137 \nside or server-side) may also prove useful to an intruder trying to figure out how an \napplication works, and thereby deduce a potential flaw in the application that could be \nexploited. \nSource code comments for interpretive code should ideally be removed before the code is \nplaced into production. Removal could be done by hand or with the assistance of a tool \n(such as Imagiware's HTML Squisher (www.imagiware.com), which removes superfluous \nHTML). Table 6.9 lists some Web sites that have software tools that can be used to scan \nsource code for specific character strings, such as a comment header. Note that removing \ncomments should also have the beneficial side effect of slightly speeding up the download \nof any client-side code. \nTable 6.9: Sample Software Tool Libraries Web Sites  \nNAME  \nASSOCIATED WEB SITE  \nACME Laboratories \nwww.acme.com  \nCNET Networks \nwww.download.com/shareware.com/zdnet.com  \nTucows \nwww.tucows.com  \nUltimate Search \nwww.freeware32.com  \nVA Software \nwww.davecentral.com/osdn.com/sourceforge.net  \nNote that the Unix Grep command can also be used to search source code and the \nstrings command can search object code. \nIf testing occurs against nonfinal code (such as commented code that has yet to be stripped \nof its comments), then another iteration of testing against the final (comment-stripped) \nversion would be recommended. Otherwise, the application runs the risk that an error will \nbe accidentally introduced as a by-product of removing the comments (especially if the \ncomments are deleted by hand). \nIn the case of client-side code, an intruder may not even need to compromise the Web site. \nCertain tools can enable an intruder to download and then search an entire Web site for \nsuch useful keywords as passwords or userIDs (see Table 6.10). \n \nTable 6.10: Sample Web Site Crawlers/Mirroring Tools  \nNAME  \nASSOCIATED WEB SITE  \nCrawl \nwww.monkey.org  \nSam Spade \nwww.samspade.org  \nTeleport Pro \nwww.tenmax.com  \nWget \nwww.wget.sunsite.dk  \nCopyrights \nOne exception to the no production comments rule is the inclusion of a copyright statement \nin the source code. Depending upon the jurisdiction, adding such protection may increase \nthe number of offenses intruders would commit if they attempted to access a Web site by \naltering the Web site's code, potentially increasing the penalty they would face should they \nbe caught. \n"
        },
        {
            "page_number": 145,
            "text": " \n138 \nHelpful Error Messages \nIn the same way that dynamic code environment should be checked (described previously \nin this chapter), applications should also be checked to ensure that if an error occurs in the \nproduction environment the application doesn't give up too much detailed information about \nthe internal workings of the application to an external user. For example, returning the \nsource code of a failed SQL statement to the requesting browser may provide an intruder \nwith information on the application's database schema, which he or she may be able to \nutilize, should the intruder gain limited access to the database server. \nAnother common error message that is often more helpful than it needs to be is that of the \nfailed login attempt. Some applications will inadvertently let attackers know when they have \nguessed the right userID. For instance, if entering an invalid userID and password \ncombination results in an \"Invalid login attempt\" error message, but guessing the right \nuserID (while still using an invalid password) results in an \"Invalid password\" error \nmessage, observant attackers will notice the change in error messages. They will thus be \nable to deduce that they have stumbled across a valid userID and thereby focus their effort \non cracking the password (a much easier task to accomplish now that a valid userID is \nknown). \nOld Versions \nAny components of a Web application that are no longer needed or that have been \nsuperseded by a more recent version should be removed from the production server as \nsoon as configuration management procedures permit. Should an intruder discover these \ncomponents, they might be able to use these legacy components to interfere with (or crash) \na legitimate process or even corrupt the application's database, in effect launching a form of \na denial-of-service attack. Table 6.11 summarizes the application code security checks that \nshould be considered. Viega (2001) provides additional information on auditing application \nsource code for security vulnerabilities. \n \nTable 6.11: Application Code Checklist  \nYES  \nNO  \nDESCRIPTION  \n□  \n□  \nDo the coding standards used to build the application include \nguidelines on the use/nonuse of comments, copyright notices, error \nmessages, and hard-coding sensitive data into an application? \n□  \n□  \nWith the exception of a copyright notice, have all comments been \nremoved from the production version of any noncompileable \n(interpretive) application code? \n□  \n□  \nHave the application's error handlers been reviewed to ensure they \ndo not divulge too much information to the client when invoked in the \nproduction environment? \n□  \n□  \nHas all unneeded application code, compiled source code, old \nversions of code, test harnesses, and so on been removed from the \nproduction environment? \n \nInput Data \nA Web application can receive input data in a number of ways (for example, from visible \nand hidden HTML form fields, cookies, and HTTP headers/URLs). Because an attacker \n"
        },
        {
            "page_number": 146,
            "text": " \n139 \ncould disable or modify any client-side validation routine, none of this input data should be \nprocessed until it has first been reexamined by a server-side validation routine. There are \nseveral different data input scenarios that the server-side validators should be able to deal \nwith, and should therefore be tested to ensure that they are indeed robust enough to handle \nthese situations. Whittaker (2002) offers extensive guidance on the categorization and \nselection of test input data that may be used by a testing team to break software. \nInvalid Data Types \nMost programming languages do not take kindly to receiving input data in a data format \ndifferent from the one specified by a program's input parameters. Such an occurrence may \nresult in data truncation, incorrect conversations, or even the demise of the program itself. \nFor example, a developer may have used a drop-down HTML control for a user to rate a \nnew product, and expect to receive a rating between 1 and 10 (inclusive) from the resulting \nHTML form submittal. Unfortunately, a HTML-savvy attacker could quite easily edit the \nHTML form (or HTTP message) and replace the expected numeric input with a value such \nas astalavistababy. If this erroneous data is not caught by the data validation routine, the \nrecipient application may do any number of things with the bogus input, none of which are \nlikely to be desirable. \nEach data validation routine should therefore be tested to ensure that it is able to \nappropriately handle input data of the wrong data type. This testing can be accomplished by \nediting the HTML used to build the data input Web page, writing a custom test harness, or \nusing the scripting capabilities of a functional testing tool (such as one of those listed in \nTable 6.12). \n \nTable 6.12: Sample Web-Testing-Based Scripting Tools  \nNAME  \nASSOCIATED WEB SITE  \nAutoTester ONE \nwww.autotester.com  \ne-Test Suite \nwww.empirix.com  \nEValid \nwww.soft.com  \nFunction Checker \nwww.atesto.com  \nSilkTest \nwww.segue.com  \nTeamTest and Test Studio \nwww.rational.com  \nTestPartner \nwww.compuware.com  \nWebART \nwww.oclc.org  \nWebFT \nwww.radview.com  \nWinRunner, XRunner, and Astra QuickTest \nwww.mercuryinteractive.com  \nWinTask \nwww.wintask.com  \nInvalid Ranges \nA Web developer may decide to use some of the built-in validation capabilities of a client-\nside language (such as HTML, JavaScript, or VBScript) to ensure that an input value is no \nlonger (or shorter) than expected. For example, the developer may have used an HTML \nform field with a maxlength of 2, combined with some client-side JavaScript, to ensure that \n"
        },
        {
            "page_number": 147,
            "text": " \n140 \nwhen the user is asked to \"enter the month you were born,\" he or she enters a value \nbetween 1 and 12. However, if the user has disabled client-side scripting (perhaps as a \nsecurity precaution of his or her own), the user would be able to enter values such as 0, 35, \n58, or 93. If he or she were also willing to edit the HTML form itself, values such as 123, \n4567, 89012, and so on, become possible. This would potentially cause problems for an \nunsuspecting server-side process. \nRather than relying on manual tests to input numerous combinations of test input data, \nscripting tools (such as those listed in Table 6.12) can be used to automate this often-\ntedious process and also potentially reduce the effort needed to run a regression test of \nthese features in the future. \nInstead of testing each input field with a random selection of input values or a large range of \nnumbers traditional testing techniques such as equivalence partitioning and boundary value \nanalysis can be used to help identify optimal test data for invalid range testing as it is \ntypically impossible to test every conceivable input value, because the possible number of \ninput values are often infinite. \nBuffer Overflows \nA variation on the invalid input attack is to submit huge volumes of data in the attempt to \ncause a buffer overflow (a technique described in great detail by Aleph One in his white \npaper; \n\"Smashing \nthe \nStack \nfor \nFun \nand \nProfit,\" \navailable \nfrom \nwww.insecure.org/stf/smashstack.txt). A buffer overflow occurs when the size of the input \ndata exceeds the room reserved for it in a program's memory, causing the program that \nwas processing the data to fail. An attacker may use this type of attack to bring down the \nWeb site, effectively creating a denial-of-service attack. He or she could also use a buffer \noverflow as a means of inserting a command into the portion of memory that the server \nuses to store the data and commands that it is currently processing (for instance, a stack or \nheap). This is done in the hope that the server will inadvertently execute this covert \ncommand, a sequence of events described in greater detail by Skoudis (2001) and Viega \n(2001). If the program that is compromised is running with administrator (or root) privileges, \nthen the rogue command will be executed with this inherited power. All input data should \ntherefore be bounds-checked to avoid buffer overflows (a technique expanded on later in \nthis chapter). \n \nA CRASH COURSE IN TWO TESTING TECHNIQUES \nEquivalence partitioning attempts to group the entire set of possible values for an input field \ninto two classes: valid and invalid (a task that may necessitate gaining knowledge of how \nthe application will process the data). A minimal set of test cases would necessitate using at \nleast one value from each class. \nBoundary value analysis (BVA) complements equivalence partitioning. Rather than \nselecting any element in an equivalence class, those values at the edge (or boundary) of \nthe class are selected. The assumption being that errors tend to occur at the boundaries of \nvalid input data, and using the values that are closest to the boundary will most likely \nexpose any errors. Typically, this means selecting the boundary value, the boundary plus \nthe smallest possible increment, and the boundary value minus the smallest possible \nincrement \nFor instance, let's perform a BVA using the month field example previously described in this \nsection. An appropriate set of input values for the lower boundary would be 0, 1, and 2 \n"
        },
        {
            "page_number": 148,
            "text": " \n141 \n(basically the boundary value plus and minus the next possible input value). For the upper \nboundary, the values 11, 12, and 13 would be appropriate. \nJorgensen (1995) and Kaner et al. (1999) provide much more detailed explanations of \nthese two testing techniques, while Beizer (1995) explains an associated testing technique: \ndomain testing. \nHatching an Egg \nApplication-level buffer overflow attacks are application, operating system, and hardware \narchitecture specific; some platforms and programming languages enforce better memory \nmanagement on their applications than others. Therefore, these attacks typically \nnecessitate an intruder gaining access to the application's source code as well as an \nequivalent system software and hardware environment. This is done in order to construct \nan appropriately formatted input stream (sometimes referred to as an egg) that can result in \none of the attacker's own system commands being executed. \nDenying an attacker the knowledge to design an application-specific buffer overflow is just \none reason why access to the application's source code, and knowledge of which system \nsoftware and hardware is being used by the Web site, should be restricted. Without this \ninformation, an attacker trying to author a new exploit is forced to rely on a brute-force or \nget-lucky strategy, which can be quite time consuming and therefore less likely to be \npursued. \nExecution Denial \nAlthough not foolproof, one method of preventing malicious system commands being \nexecuted via a buffer overflow is to configure the system software to prohibit command \nexecution directly from memory. This is an option that (if supported by the system software) \nwill defeat many buffer overflow attacks but may cause issues with some legitimate \napplications that require this functionality to be enabled in order to run correctly. For \nexample, Sun Microsystems's Solaris operating system has a built-in option to disable stack \nexecution, while similar protection can be added to other platforms using third-party plug-\nins, such as SecureStack (www.securewave.com) for Windows NT/2000 or StackGuard \n(www.immunix.org) for Linux. \nCode Reviews and Inspections \nOne of the most effective ways of detecting application-level buffer overflows is via peer-\nlevel code reviews and inspections (a technique described in more detail in Chapter 3). \nUnfortunately, this approach has at least two drawbacks: The sheer volume of code that \nneeds to be reviewed may make it resource prohibitive for many organizations to conduct a \npeer review for every component of a Web application. In such circumstances, reviews may \nhave to be restricted to components that have been identified as being particularly at risk \n(such as brand new code, code written by a novice programmer, or code responsible for \nbounds-checking data from an untrusted source like the client-side portion of a Web \napplication). \nThe second issue that relates to an application code review is that the application code may \nbe devoid of errors, but the system software or third-party utilities that the application \ndepends on may contain a buffer overflow. Few organizations have the resources to review \nthird-party source code before utilizing it in their own applications (assuming the source \ncode for the third-party utility is even available for review; Chapter 4 discusses the pros and \ncons of closed-source versus open-source code). However, if an application fails because it \nuses a system utility that has a buffer overflow vulnerability and an attacker can \n"
        },
        {
            "page_number": 149,
            "text": " \n142 \ncompromise the application because of this, any Web site using this application may be \nvulnerable. This is despite the fact that no overflow exists in the lines of code written by the \napplication's developer. Traxler (2001) provides additional information on reviewing a Web \napplication for buffer overflows. \nAs an alternative to conducting manual reviews, several source-code-critiquing tools \nattempt to automatically detect problematic code, using proprietary heuristics to look for \nsuspicious code, calls to specific utilities known to have vulnerability issues, or a \ncombination of both (see Table 6.13). For example, Illuma from Reasoning Solutions \n(www.reasoning.com) is designed to check for memory management issues, uninitialized \nvariables, and poor pointer management, while ITS4 from Cigital (www.cigital.com) \nexamines function calls, specifically looking for security issues, such as potential buffer \noverflows. \n \nTable 6.13: Sample Source-Code-Scanning Tools  \nNAME  \nASSOCIATED WEB SITE  \nCodeWizard \nwww.parasoft.com  \nIlluma \nwww.reasoning.com  \nITS4 \nwww.cigital.com  \nLDRA Testbed \nwww.ldra.co.uk  \nPC-lint \nwww.gimpel.com  \nQA C/C++ \nwww.programmingresearch.com  \nRATS \nwww.securesw.com  \nSplint \nwww.splint.cs.virginia.edu  \nWebInspect \nwww.spidynamics.com  \nBuffer Overflow Testing \nIf application source code reviews and inspections are not an option, or nonfailure of the \napplication is so critical that additional testing is warranted, then an application (together \nwith any third-party and system utilities that it utilizes) can be further tested. This can be \ndone by executing destructive tests intentionally designed to detect the existence of a buffer \noverflow vulnerability. Dustin et al. (2001) and Whittaker (2002) provide additional \ninformation on testing applications for buffer overflow vulnerabilities. \nUnlike an attacker trying to create a new exploit, a tester should not have to refine an \napplication crash (or any observable weird behavior) to the point where he or she would \ninsert a malicious system command into memory to prove that the application is vulnerable \nto a buffer overflow. Any application crash or uncontrolled behavior is a potential denial-of-\nservice issue. \nTesting Tools \nA number of tools could be used to create and submit the huge volumes of data typically \nneeded to simulate a buffer overflow attack to the target application: \nƒ \nPrograms specifically built for sending large amounts of input data, such as NTOMax \nat www.foundstone.com or Hailstrom (www.cenzic.com). \n"
        },
        {
            "page_number": 150,
            "text": " \n143 \nƒ \nNetwork packet manipulation tools such as SendIP (www.earth.li/projectpurple). \nƒ \nHTML authoring tools, which can be used to edit an application's Web pages to \npermit \nhuge \ndata \ninputs \nvia \na \nregular \nbrowser, \nsuch \nas \nFrontPage \n(www.microsoft.com.) \nƒ \nScripting languages, such as Perl (www.perl.com) \nƒ \nTesting tools which record and play back HTTP transactions at the network layer, \nsuch as Webload (www.radview.com). \nƒ \nTesting tools that record and play back browser interactions at the browser layer, \nsuch as e-Test Suite (www.empirix.com). \nWhichever data submission tool is used, it should be evaluated with the rest of the test \nenvironment to ensure that it does not prematurely truncate the test data that it is being \nasked to submit to the target application. \nCalibrating the Test Environment \nBefore being able to test any application for a buffer overflow, the testing environment must \nfirst be checked out to determine the maximum size of test input data that will be permitted \nby the test environment. This is due to the fact that many system software products will \ntruncate extremely long input data. This truncation not only protects any recipient of the \ndata, but also ironically inhibits tests designed to probe an application beyond this system \nlimitation. Unfortunately, these system software truncations can't always be counted on to \nprotect an application, as the point at which a truncation may occur may vary from module \nto module. For example, the system software code used to handle an HTTP Get command \nmay have a different limitation than that used to process an HTTP Post request. \nIn an ideal test environment, it should be possible for the testing team to use input data of \nan infinite size. Unfortunately, sooner or later the test environment itself will truncate the \ndata, thereby prohibiting further testing. Since a truncation could occur at any system \nsoftware layer, the fewer layers that the test input data has to pass through, the more likely \nit is that it will make it to the intended target application without being curtailed. At a \nminimum, the test environment should permit an input string of 64K characters to be \npassed to the target application without being truncated. Figure 6.2 illustrates these \ndifferent layers. \n \n \nFigure 6.2: Test environment layers.  \nTest Logs \nUsing an automated tool makes data submission a lot easier (no need to hit the X key \nseveral thousand times). Unless the tool generates some sort of test log, however \n(recording the input used for each test and the associated test results), identifying the input \n"
        },
        {
            "page_number": 151,
            "text": " \n144 \nstream that first caused an application to fail is likely to necessitate several reruns and \nprove rather tedious. \nIdentifying the specific input field that caused the failure, together with the length of data \nbeing used at the time, should prove invaluable for the developer charged with diagnosing \nthe application's overflow problem. \nTest-Entry Criteria \nIf not already done as part of functional testing, the initial series of tests should establish the \npositive functionality of the application (for example, does the application process valid \ninput sizes correctly, never mind invalid lengths), effectively creating an entry criteria for the \ndestructive testing that will be performed to search for the existence of a buffer overflow. \nThe exact size of the input data should be the maximum size allowed by application's user \ninterface, program specification, application requirements, or the size of the database field \nused to store the data (which theoretically should all be the same). \nA supplemental functional check would involve reviewing the final destination of the input \ndata (typically a database field) to ensure that the largest permissible input has not been \ninadvertently truncated somewhere on its way through the application. This is useful when \nconducting internationalization testing using non-Latin character sets that utilize double-\ncharacter byte encodings. \n \n \nSmall-Scale Overflows \nThe next set of tests should focus on testing the functional boundary of the application. By \nusing input values just a few characters larger than those used in the previous set of tests, \nthe application can be tested to make sure that the input data does not solely rely on any \nclient-side checks (which can be easily circumvented) to prohibit invalid data from being \nsubmitted to the server-side component of the application. \nA small-scale test may be easier to monitor if the final destination of the input data can be \nmodified to accommodate input data slightly larger than what it would normally be expected \nto receive. For example, if the final destination were a database field defined as 256 \ncharacters, temporally expanding the field to accommodate 300 characters would facilitate \nthe testing team by demonstrating whether or not any truncation took place. If the input data \nis not truncated, then a high probability exists that the input field is not being bounds-\nchecked correctly and is therefore a good candidate for a buffer overflow scenario. \nUnfortunately, an appropriately truncated file being deposited in the corresponding \ndatabase file does not necessarily mean the input field is immune from a buffer overflow \nattack. At this point, there is no telling where the truncation took place, and a buffer \noverflow could occur before the point at which the data is truncated. \nLarge-Scale Overflows \nIf a small-scale overflow has not resulted in any noticeable impact on the application, then \nit's time for the testing team to pump up the volume. Progressively increase the size of the \ninput data until the application starts to behave abnormally, or completely fails. \nRather than repeating the test time after time by incrementing the input stream by a single \ncharacter, the testing team may wish to focus on the input sizes that are more prone to \nbuffer overflows. Designers, developers, and database administrators (DBAs) often select \n"
        },
        {
            "page_number": 152,
            "text": " \n145 \n(implicitly or explicitly) the size of a data field based on its data type declaration (such as \ninteger, varchar, smallint, or text). Therefore, selecting input stream lengths that are \nclustered around the common maximum lengths for data types is likely to drastically reduce \nthe number of tests used, without significantly reducing the quality of the testing. (This is an \nexample of boundary value analysis at work.) \nFor example, for a text-based input field, increasing the data input size in the increments of \n256, 512, 1K, 2K, 4K, 8K, 16K, 32K, 64K, and so on, is likely to be a much more efficient \ntesting strategy than each time simply incrementing the input data by single character. The \nlatter would result in 65,536 tests before reaching a test that uses a data input stream of \n64K characters. \nSince some languages or platforms add or subtract a few characters from the logical length \nof a data declaration (for instance, a varchar(256) declaration may actually consume 258 \nbytes), unless the testing team completely understands the inner workings of the platform \nbeing used, it may pay to simply add a few additional characters to each input stream. An \nexample would be using data sizes of 264 (256 + 8) and 520 (512 + 8), instead of 256 and \n512. \nTest Optimization \nAssuming no observable failure occurs, sooner or later the input steam will reach the \nmaximum size that the test environment can handle without the input data being truncated. \nIf, based on previous experience, the application should be able to handle the maximum \ninput that the test environment can punish it with, a quicker way to prove this hypothesis is \nto start with the maximum input size that the test environment can submit. If the application \ncan withstand this worst-case scenario, then there is no need to perform the smaller-sized \ntests. If, on the other hand, this extreme test fails, additional smaller tests will probably be \nneeded to assist with diagnosing the approximate location of the problem. \nThe Return Leg \nIf a database field has been defined to be larger than what would normally be needed to \nstore input data from a legitimate source (for example, a varchar(256) field that under \nnormal circumstances never exceeds 50 characters), the testing team may want to consider \ninserting data directly into the database to fill up such fields. Once populated, these fields \ncan be requested via the application, causing these unexpectedly long values to travel back \nthrough the application to the clients, thereby allowing the testing team to check for buffer \noverflows on the return leg of an application. \nTest Observations \nOne of the challenges of testing an application for buffer overflows is that in many instances \na small buffer overflow does not produce any observable symptoms. For example, a 257-\ncharacter input stream may actually cause a buffer overflow to occur, but if the application \ndoes not crash because the extra character overflowed into a portion of memory that by \nchance isn't referenced, the defect may be indistinguishable from an application that \ncorrectly truncates the input string to 256 characters. \nEven if a buffer overflow has occurred, a significantly larger input may be required before \nthe situation can be detected, and then the event may not be as dramatic as a program \ncrash but much more subtle. Examples would be a degradation in system performance or \nthe gradual loss of allocated memory (a memory leak). These are symptoms that may not \nhave much of an impact on the system under light processing loads but could prove \ndisastrous under higher loads. \n"
        },
        {
            "page_number": 153,
            "text": " \n146 \nThe ease with which a buffer overflow can be observed is also dependent upon the \nresources available to the testing team. For example, testing teams that have access to \nmemory monitors which can dynamically report on an application's memory allocation, are \nat a distinct advantage to those who have to rely on symptoms observable to the naked \neye. \nDiagnostics \nOnce a program crashes, or some sort of weird behavior is detected, the testing team may \nbe able to refine their input data to determine the approximate point of the failure, thereby \nspeeding up debugging. Although executing additional destructive tests may identify the \nexact circumstances that cause a program to fail, it is likely that a code review will prove to \nbe the most efficient way of identifying the exact program location of the defect that results \nin the buffer overflow. \nSince all input data should first be bounds-checked before being processed by any other \ncomponent of the application, the first candidate for a code review is likely to be the server-\nside routine that initially receives the input data (and therefore theoretically performs the \nbounds check). Often the cause can be as simple as a developer forgetting to bounds-\ncheck just one of the input fields, or using logic that only permits bounds-checking the data \nfor reasonable out-of-bounds values. \nEscape Characters \nSome operating systems will execute system-level commands if they are embedded in an \napplication's data input stream. This can occur when the system command is hidden in \ninput data that is prefixed by special control (escape) characters, such as $$. The \napplication may then permit the command to escape up to the process that is currently \nrunning the application. The receiving process then attempts to execute the system \ncommand using its own system privileges. \nData input streams should therefore always be scanned for suspicious characters as soon \nas they arrive. Although the specific escape sequence will vary from platform to platform, \nit's generally considered a safer programming practice to check for the inclusion of only \nlegal characters than to check for and attempt to discard illegal ones. A Justification for this \nviewpoint includes the possibility that a developer might have inadvertently forgotten to \ncheck for one illegal character (not checking for one legal character would not pose a \nsecurity risk). Another possibility is that the application was being ported to a platform the \ndeveloper had not considered (thereby offering the opportunity for a new set of escape \ncharacters). \nOne temporal usability problem with having extremely tight input data validation rules is that \nsome legitimate input may get rejected, stripped, or replaced, because it was wrongly \nidentified as illegal input. For example, a validation routine that checks the input data for the \nsurname field may only permit characters A through Z (lower-and uppercase) and space. \nUnfortunately, this routine fails to consider the situation of a double-barreled name (Baxter \nSmith-Crow), stripping the hyphen and offending a small number of individuals. This \nproblem is usually temporal in nature, because it is typically identified and fixed relatively \nquickly. \nWhichever route is taken, excluding illegal characters or only including the legal ones, each \ninput data stream should be checked to ensure the code has been implemented according \nto the application's design specification. If the specification doesn't document exactly what \nis acceptable and what is not (and assuming the specification is unlikely to be clarified \nbefore testing), then it would be prudent for the security-testing team to assume that any \n"
        },
        {
            "page_number": 154,
            "text": " \n147 \ninput other than a through z, A through Z, or 0 through 9, including spaces, is suspicious. \nThe team should thus report on any input fields that do not discard this input. For \nconvenience, Table 6.14 lists all possible 128 (7-bit) ASCII input characters. Ideally, the \napplication's specification should explicitly document which of these characters are \nacceptable (and consequently which ones are not). \n \nTable 6.14: ASCII Data Input Characters  \nCHARACTER  \nASCII HEXDEC CODE  \nNull \n00 \nStart of heading \n01 \nStart of text \n02 \nEnd of text \n03 \nEnd of transmission \n04 \nEnquiry \n05 \nAcknowledge \n06 \nBell \n07 \nBackspace \n08 \nCharacter tabulation \n09 \nLine fed \n0A \nLine tabulation \n0B \nForm fed \n0C \nCarriage return \n0D \nShift out \n0E \nShift in \n0F \nDatalink escape \n10 \nDevice control 1 \n11 \nDevice control 2 \n12 \nDevice control 3 \n13 \nDevice control 4 \n14 \nNegative acknowledgement \n15 \nSynchronous idle \n16 \nEnd of transmission block \n17 \nCancel \n18 \nEnd of medium \n19 \nSubstitute \n1A \nEscape \n1B \n"
        },
        {
            "page_number": 155,
            "text": " \n148 \nTable 6.14: ASCII Data Input Characters  \nCHARACTER  \nASCII HEXDEC CODE  \nFile separator \n1C \nGroup separator \n1D \nRecord separator \n1E \nUnit separator \n1F \nSpace \n20 \nExclamation \n21 \nQuote \n22 \nNumber sign \n23 \nDollar sign \n24 \nPercent sign \n25 \nAmpersand \n26 \nApostrophe \n27 \nLeft parenthesis \n28 \nRight parenthesis \n29 \nAsterisk \n2A \nPlus sign \n2B \nComma \n2C \nHyphen/minus sign \n2D \nFull stop \n2E \nForward slash \n2F \nDigits 0 through 9 \n30-39 \nColon \n3A \nSemicolon \n3B \nLess than sign \n3C \nEquals sign \n3D \nGreater than sign \n3E \nQuestion mark \n3F \n@ \n40 \nUppercase A through Z \n41-5A \nLeft square bracket \n5B \nBackslash \n5C \nRight square bracket \n5D \n"
        },
        {
            "page_number": 156,
            "text": " \n149 \nTable 6.14: ASCII Data Input Characters  \nCHARACTER  \nASCII HEXDEC CODE  \nCircumflex \n5E \nLow line \n5F \nGrave \n60 \nLowercase a through z \n61-7A \nLeft curly bracket \n7B \nVertical line \n7C \nRight curly bracket \n7D \nTilde \n7E \nDelete \n7F \nSource: www.ansi.org. \n \nPOISONING DATABASE INPUT DATA \nOne technique that attackers may use to execute illicit commands against a database is to \ninsert a database expression where the developer was expecting to receive a single input \nparameter. For instance, suppose a login Web page consistent of two input fields (userID \nand password). In addition to entering a bogus userID and password, an attacker might \nappend to each input field the following string: \n\" or \"123\" <> \"1234\" \nIf the input fields are fed directly into a database request, then it's possible that a database \nmight not treat this input string as a single parameter, but instead attempt to evaluate the \nexpression and then find that 123 is indeed not equal to 1234, and thereby permit the \nattacker to successfully log in. Of course the probability of such an attack being successful \nis increased if the attacker is first able to view the source code that he or she intends to \nmanipulate (or poison). \nUnfortunately, writing individual, customized data validation routines for each data input \nstream may result in more coding errors making it into production than if a common (and \ntherefore probably slightly more liberal) set of well-tested data valuation routines were \nreused. Therefore, a trade-off exists between using a large collection of tight validation \nroutines and using a small set of more liberal validation routines. \nWhile certainly not a replacement for good data input (and output) validation routines, there \nare some tools (examples of which are listed in Table 6.15) that attempt to validate all the \ninput data sent to a Web site, intercepting and discarding any suspicious input data before it \nis able to do any harm. \n \nTable 6.15: Input Data Validation Tools  \nNAME  \nASSOCIATED WEB SITE  \nAPS \nwww.stratum8.com  \nG-Server \nwww.gilian.com  \n"
        },
        {
            "page_number": 157,
            "text": " \n150 \nTable 6.15: Input Data Validation Tools  \nNAME  \nASSOCIATED WEB SITE  \niBroker SecureWeb \nwww.elitesecureweb.com  \nURLScan \nwww.microsoft.com  \nWhichever approach is used, all data input options should be tested to ensure that their \ncorresponding data input validation routines (third-party or otherwise) are robust enough to \nwithstand the worst possible scenarios. Table 6.16 summarizes these scenarios. \n \nTable 6.16: Valuation Routine Checklist  \nYES  \nNO  \nDESCRIPTION  \n□  \n□  \nDo the coding standards used to build the application include \nguidelines on the use of input data validation routines? \n□  \n□  \nHave all data input validation routines been inspected and/or tested \nto ensure they are able to handle invalid data types? \n□  \n□  \nHave all data input validation routines been inspected and/or tested \nto ensure they are able to handle invalid data ranges? \n□  \n□  \nHave all data input validation routines been inspected and/or tested \nto ensure they are able to handle buffer overflow attempts? \n□  \n□  \nHave all data input validation routines been inspected and/or tested \nto ensure they are able to detect system command escape \ncharacters? \n \nServer-Side Data \nIf an organization's firewall(s) were to be breached or circumvented, sensitive information \nstored in obviously named files or database tables (as depicted in Figure 6.3 as a means of \nimproving design readability) may allow an intruder to locate desirable information much \nmore easily. This is especially true if the information is conveniently located on a Web \nserver (a practice not recommended). \n \nFigure 6.3: Database naming standards.  \n"
        },
        {
            "page_number": 158,
            "text": " \n151 \nTo mitigate this threat, many organizations have been known to implement one or more \ndatacentric defenses, such as the examples in the following sections. \nData Filenames \nIf sensitive application data (such as userIDs and passwords) is to be stored in a database \naccessible via the Web site, the database design may have called for such fields and files \nto be named in a manner that does not give away their true usage. If such a strategy is \nemployed, the database design should be reviewed at an early stage to ensure that \nsensitive information is not stored in such obviously named files as users, userIDs, \npasswords, security, accounts, and so on. \nUnfortunately, if blatantly obvious names do make it into production, the cost of reworking \n(and testing) the application code just to use alternate filenames may be unjustifiably \nexpensive, hence the need to decide early on in the application's design whether or not this \nsecurity by obscurity approach should be adopted by the application's developers. \nData Tripwires \nAn exception to this security by obscurity rule would be when these easily identifiable files \nare used to store bogus userIDs and passwords. An attempt to log in with a userID from \none of these bogus files may then be used as a tripwire to alert support staff to the fact that \nsomeone has already gained unauthorized access to the application's database and is \nattempting to penetrate further. (Tripwires are discussed in more detail in Chapter 8.) \nUnfortunately, tripwires are generally only effective if someone takes notice of their alerts. \nTherefore, if a database tripwire is deployed, when it is triggered, the on-duty staff should \nmonitor and react appropriately if triggered. \nData Vaults \nVulnerable data may be placed inside a data vault for further protection. A data vault is a \ndatabase (or data repository) that has had additional security measures applied to it to \nensure that the data is protected from unauthorized access. Such controls improve the \nsecurity around the data but add to the administrative workload and may impact data I/O \nspeeds. If a data vault is implemented, the testing team should check that all the sensitive \ndata that should be placed inside the vault does actually reside there, and that no \nperformance-enhancing copies have been replicated outside the vault. Examples of data \nvaults are listed in Table 6.17. \n \nTable 6.17: Sample Data Vaults  \nNAME  \nASSOCIATED WEB SITE  \nBorderWare \nwww.borderware.com  \nCyber-Ark \nwww.cyber-ark.com  \nWORMs \nWrite-once read-many (WORM) devices may impact performance, but using devices such \nas CD writers to store transaction logs or network audit trails would most likely thwart any \nattempt by an intruder to cover their tracks by deleting or modifying these logs. \nIn the case of read-only data files and executables, fast CD readers (with no writing \ncapabilities) and ample memory may make for a more secure storage solution than storing \n"
        },
        {
            "page_number": 159,
            "text": " \n152 \nthese files on hard drives that could be compromised. The amount of memory needed \nshould be sufficient to keep the read-only files in memory long enough to avoid the \nconsiderable performance hit that repeatedly reading this data from the CD would cause. \nIf a Web application's design calls for hardware-based write restrictions, a quick visual \ninspection of the hardware may not be able to detect the difference between a write-once \nCD writer and a CD writer with rewriting capabilities. In such cases, a software-based test \nmay be warranted to see if files on the CD can be altered or deleted. To ensure that any \nrestriction is due to the hardware device, as opposed to a security setting, the file access \nrequest should be done with an administrative account, which is not limited by any \nsoftware-based security restriction. \nData Encryption \nAlthough stringent precautions may be taken to protect data files used to permanently store \ndata, less attention may be paid to transient data files such as temporary files, Web logs, \ndatabase logs, application logs, third-party data files, .ini files, and obscure storage \nlocations such as Window's registry entries. Unfortunately, these files are often used to \ntemporarily store sensitive data (credit card numbers, social security numbers, and so on). \nThey should therefore be checked to ensure that if such data is stored, it is only stored in \nan encrypted format, lest an intruder stumble across these forgotten nuggets of information. \nIn addition, or as an alternative to any file encryption capabilities that the host operating \nsystem may provide (for instance, Windows 2000's Encrypting File System [EFS]), products \nsuch as those listed in Table 6.18 can be used to encrypt data stored on a server. \n \nTable 6.18: Sample File Encryption Tools  \nNAME  \nASSOCIATED WEB SITE  \nAgProtect \nwww.hallogram.com  \nBlowfish and Crypto \nwww.gregorybraun.com  \nCipherPack \nwww.cipherpack.com  \nData Vault \nwww.reflex-magnetics.com  \nDataSAFE \nwww.data-encryption.com  \nEasyCrypt \nwww.easycrypt.co.uk  \nEncryption Plus \nwww.pcguardian.com  \nFileCrypto \nwww.f-secure.com  \nGNU Privacy Guard \nwww.gnupg.org  \nKryptel \nwww.bestcrypto.com  \nRSA BSAFE \nwww.rsasecurity.com  \nSteganos Security Suite \nwww.steganos.com  \nData Deception \nSince even the most strongly encrypted files can theoretically be cracked given enough \ntime and resources, deception may be a more viable alternative to using heavy encryption. \nIn other words, by making encrypted data not look like it's actually encrypted, an intruder \n"
        },
        {
            "page_number": 160,
            "text": " \n153 \nmay never suspect that the data has been encrypted and therefore not even attempt to \ndecipher it, thereby protecting the data ad infinitum. For example, instead of encrypting a \nclient's PIN into a 128-bit binary format (which would make the resulting data longer), a \nhashing algorithm could be used that creates in another value that is the same length. \nUnsuspecting intruders may never figure out that the data is actually encrypted, and when \nthey try using these values, they'll become frustrated because none of these stolen \nnumbers work. \nA fundamental weakness of a deception strategy is that for it to be effective, the deception \nmust be kept secret. If such a strategy is to be employed, knowledge of its use should be \nrestricted to a need-to-know basis, and certainly not documented in a requirements \nspecification that any employee of the organization could easily gain access to. (Chapter 8 \ndiscusses the implications of testing defenses based on deception in more detail.) \nData Islands \nMany organizations choose not to expose their master database to the Internet. One \nstrategy for doing so is to replicate the master data on a database server that is part of the \nWeb application and then to remove all physical network connections between the Web \napplication and this master database, effectively creating a data island. \nIf a Web application's design calls for a data island implementation, then the testing team \nshould determine if the implementation has left open any network path from the Web site to \nthe master database. For instance, via an internal firewall that has not been tested, or via a \ntemporary unsecured network connection that only exists while the two databases are \nbeing synchronized. \nDistributed Copies \nIn an effort to improve scalability, some Web sites use a distributed database design to \nstore data across multiple servers. In addition to any capacity flexibility that this design \nenables, such an approach may make a Web application harder for an intruder to corrupt, \nas the intruder would need to gain access to several servers before being able to corrupt \nevery single copy of a data file. \nWhere feasible, these distributed copies should ideally have different security permissions, \nmeaning that if an intruder is able to crack a userID and password protecting one copy, the \nsame combination could not be used to access the others. Therefore, the testing team \nshould check that as few accounts as possible are granted access to more than one copy \nof the data. \nFragmented Data \nIn contrast to storing multiple copies of the same data, extremely sensitive data may be \nbroken up and distributed across multiple geographic locations, making the information \nuseless unless the intruder is able to compromise all the storage sites and reassemble this \nfragmented data. For instance, some credit-card-processing Web sites store a client's \nuserID, password, and account information in three separate databases, each residing at a \ndifferent physical location. \nOne danger with this approach is that in order to improve database performance, a \ndeveloper may reassemble the fragmented data into a single temporary file, which, if \ndiscovered by an intruder, would circumvent the need to attack all the dispersed locations. \nThis situation is more likely to be discovered via design reviews than through test \nexecution. \n"
        },
        {
            "page_number": 161,
            "text": " \n154 \nDatabase Management System (DBMS) Enforced Constraints \nMost Database Management Systems (DBMSs) offer several different ways in which the \ndata inside a DBMS can be protected from unauthorized access. The following are some of \nthe more common approaches: \nƒ \nUser groups. Rather than assigning security privileges to individual userIDs, most \nDBMSs enable the DBA to assign users to user groups, and then grant access to the \ngroup, which in turn is passed on to all the members of the group. Because user \ngroups typically make security administration easier and therefore less prone to \nhuman error, they are generally considered a more secure approach than assigning \nprivileges directly to individual users. \nƒ \nViews. Instead of assigning access directly to the data, many DBAs create an \nintermediate database object-a view. Instead of accessing a table (or group of tables) \ndirectly, users reference the data indirectly via a view. This extra level of indirection \ncan allow a DBA to hide some of the database's physical attributes, such as the \ntable's real name and location, or limit the user to only viewing some of the table's \ncolumns and/or rows. \nƒ \nStored procedures. Rather than actually assigning read and write access to users \n(or user groups), some DBAs will only grant direct access to a database's tables (or \nviews) to a program stored within the database itself (a stored procedure). Users are \nthen limited to executing these predefined programs, thereby removing the data \nmanipulation flexibility intruders would have if they were able to gain direct access to \nthe tables (or even views) themselves. \nƒ \nReferential integrity (RI). Features such as foreign keys, triggers, user-defined data \ntypes, rules, and constraints that are intended to ensure that the data stored within \nthe database remains consistent may also frustrate an intruder that is trying to insert, \nupdate, or delete illicit data in an ad hoc fashion. For instance, trying to delete a \ncustomer account may first necessitate purging all the client's transactions, which in \nitself may require tampering with the application's audit trail. Or as another example, \ntrying to add a fictitious stock purchase transaction could require the intruder to \nsimultaneously add an associated stock sale transaction. However, many DBAs \ndisable RI on production databases in order to improve database performance (if no \nRI exists for the DBMS to enforce, then these resource-consuming operations will not \nbe performed). Therefore, a tradeoff takes place between performance and \nsecurity/data integrity priorities. \nWhichever approach the database design calls for, the database schema that has actually \nbeen created in the production environment should be reviewed to ensure that all the \nintended methods of protection have been implemented correctly. In addition, tools such as \nAppDetective from Application Security (www.appsecinc.com) and Database Scanner from \nInternet Security Systems (www.iss.net) can be used to scan database implementations \nlooking for DBMS-specific security vulnerabilities. Castano et al. (1994), Dustin et al. \n(2001), and Heney et al. (1998) provide additional information on database security \nmeasures. \nFiltered Indexes \nInternal search engines can be useful to legitimate users trying to locate hard-to-find \ncontent. Unfortunately, a malicious user could also use this common Web site feature to \nlocate confidential information using search words such as password, security, fraud, \nsecret, or even the name of a competitor. To mitigate this possibility, some internal search \n"
        },
        {
            "page_number": 162,
            "text": " \n155 \nengines enable a Webmaster to list words that should not be indexed. Either way, the \ninternal search engine should be tested to ensure that it doesn't provide any sensitive \ninformation. \nThe defenses designed to protect an application's data are often an application's last line of \ndefense. They should therefore be thoroughly exercised by the testing team, especially \nwhen you consider that unlike some defenses that are under constant attack (such as a \nperimeter firewall), an organization may never have had this final line of defense put to the \ntest by a real attacker. Table 6.19 summarizes the checks that a testing team may want to \nconsider performing as a means of determining if the data residing on the server-side \nportion of a Web application is adequately protected. \n \nTable 6.19: Server-Side Data Checklist  \nYES  \nNO  \nDESCRIPTION  \n□  \n□  \nDoes the Web application's design specify which defensives should \nbe put in place to protect data stored on the server side (with the \npossible exception of a deception-based defense)? \n□  \n□  \nHave sensitive data files been given names that do not indicate their \ntrue purpose? \n□  \n□  \nIf the application has a database tripwire, does the on-duty support \nstaff react to an alert appropriately, or are the alerts ignored? \n□  \n□  \nAre any devices (or media) that are intended to be read-only or write-\nonce actually write inhibited? \n□  \n□  \nAre all data files protected from being corrupted, altered, or \nprematurely destroyed by an intruder using data vaults, WORM \ndevices, data islands, distributed copies, DBMS-enforced constraints, \nand so on? \n□  \n□  \nAre all sensitive data files protected from being read by an intruder \nusing data vaults, encryption, deception, fragmented copies, and so \non? \n□  \n□  \nHas the desired database schema been implemented correctly in the \nproduction environment? \n□  \n□  \nHas any internal search engine been checked to ensure that it does \nnot return any confidential information? \n \nApplication-Level Intruder Detection \nFinancial institutions such as credit card issuers have long used a client's behavior to help \ndetect when another person has gained unauthorized access to a client's account. For \ninstance, many jewelers expect a credit card's verification department to want to talk to the \npurchaser. This is to verify that their client is the one who is actually performing this \nunusually large transaction, not someone who has stolen a card and is using it before the \nrightful owner has noticed that it is being used inappropriately. Another sign that some \ncredit card issuers use to indicate a potential problem is the sudden use of the card at \nseveral locations that do not require signature confirmation (such as gas station pumps). \nThieves will often use these low-risk purchases to determine whether a card is still active \nbefore attempting a higher-risk transaction. \n"
        },
        {
            "page_number": 163,
            "text": " \n156 \nIf a Web application attempts to identify unauthorized transactions based on a client's past \nbehavior, then the testing team should consider including tests to determine how sensitive \nthe monitoring is. For example, let's look at a buy-and-hold investor, who normally buys \nshares in blocks of $500 once a month and generally holds on to them for several years. \nThis client would probably appreciate a phone call to confirm that the 20 daytrades that \nwere placed yesterday were actually done by him or her and not an unauthorized person \n(such as an intruder or one of the client's own children). On the other hand, the \norganization's customer service department may not be able to handle calling 1 percent of \ntheir customer base every day, just because the behavioral model being used is too \nsensitive (and additionally notifying their customers that behavior data on them is being \ncollected and analyzed, possibly generating privacy concerns amongst them). Table 6.20 \nsummarizes these scenarios. Garfinkel (2002) and Ghosh (2001) provide additional \ninformation on the topic of Web privacy. \n \nTable 6.20: Application Intruder Detection Checklist  \nYES  \nNO  \nDESCRIPTION  \n□  \n□  \nDoes the application's specification document under which user \nbehavioral circumstances an alert should be raised? \n□  \n□  \nAre the detection rules too sensitive, resulting in too many false \nalarms? \n□  \n□  \nAre the detection rules too lax, allowing blatant changes in user \nbehavioral patterns to go unnoticed? \n□  \n□  \nIf an application-level intruder detection system is used, does the on-\nduty support staff react to an alert appropriately, or are the alerts \nignored? \n \nSummary \nAs can be seen from the preceding sections, having the most perfectly secured Web site \nnetwork and system software infrastructure is little protection if the Web application being \nhosted on the site is not designed and developed with security in mind. By thoroughly \ntesting both the client-side and server-side components of a Web application, an \norganization can endeavor to make the Web application as secure as the infrastructure that \nthe application itself depends upon. \n"
        },
        {
            "page_number": 164,
            "text": " \n157 \nChapter 7: Sneak Attacks: Guarding Against the Less-\nThought-of Security Threats \nAlthough many people imagine attackers slaving away over a hot keyboard hundreds of \nmiles away from their intended target, assailants could use any of several alternative \napproaches to meet their objectives. This chapter looks at several of these sneaky attacks \nand offers suggestions on how the security policies that an organization has decided to \nadopt can be tested to ensure that they provide an acceptable level of protection from these \noften overlooked security risks. Harris (2001), Krutz et al. (2001), Parker (1998), and Peltier \n(2002) provide additional information on this problem domain. \nCombating Social Engineers \nThe term social engineering is used to describe the various tricks used to fool innocent \npeople (employees, business partners, or customers) into voluntarily giving away \ninformation that would not normally be known to the general public. These techniques were \nexpanded upon by Winkler (1999). Examples include names and contact information for \nkey personnel, system userIDs and passwords, or proprietary operating procedures. Armed \nwith this newfound knowledge, the attacker may then be able to penetrate even the \nsecurest of Web sites. \nAs the following scenarios seek to demonstrate, an attacker could seek to employ an \nalmost infinite number of tricks, using any one of several different communication channels. \nTricks by Telephone \nAn attacker could call an organization's help desk and try any one of the following plausible \nstories: \nƒ \n\"Hi, this is David, the VP of sales. I'm at the Chicago branch today and I can't \nremember my password. The machine in my home office has that 'Remember \npassword' set, so it's been months since I actually had to enter it. Can you tell \nme what it is, or reset it or something? I really need to access this month's \nsales reports ASAP.\"  \nƒ \n\"Hi, this is Melissa at the Salt Lake City branch. I'm the new LAN administrator \nand my boss wants this done before he gets back from Chicago. Do you know \nhow I can:  \no \nConfigure our firewall to have the same policies as corporate?  \no \nDownload the latest DNS entries from the corporate DNS server to our \nlocal server?  \no \nRun a transaction on a remote file and print server using a Shell \ncommand?  \no \nBack up the database to our off-site disaster recovery location?  \no \nLocate the IP address of the main DNS server?  \no \nSet up a backup dial-up connection to the corporate LAN?  \no \nConnect this new network segment to the corporate intranet?\"  \nTricks by Email \nAlternatively, attackers could utilize emails to communicate to their victims: \nƒ \nTo: An unsuspecting employee  \nƒ \nHi Shawn,  \nƒ \nThis is Brian from network support. We are currently testing our new corporate \nWeb application at testwiley.com and are having intermittent problems getting \n"
        },
        {
            "page_number": 165,
            "text": " \n158 \nanyone from your location signed in correctly. As soon as you read this, can \nyou please try to log into the main application using your existing userlD and \npassword?  \nƒ \nThX  \nOf course, the wiley.com Web site doesn't contain a real application, but rather a single \nHyperText Markup Language (HTML) form designed to capture and relay to the attacker \nany login attempts after it has informed the user that the login has failed. \nƒ \nTo: The registrar for the organization's domain name  \nƒ \nHi Registrar,  \nƒ \nI'm Rajan, the admin contact for the wiley.com Web site. We're moving to a \ncheaper ISP, effective immediately. Can you please set the IP address for our \nWeb site to 123.456.789.123?  \nƒ \nThanks, Rajan  \nIf the only security precaution the registrar requires is that the From: field of the email match \nthe email address on file (which may be obtainable from the registrar's own Web site), then \nthe Web site's traffic could be redirected to a new (less friendly) machine. \nƒ \nTo: The ISP hosting the organization's Web site  \nƒ \nHi tech support,  \nƒ \nI'm Carrie, the new contact point for the wiley.com Web site. The old guy \nSundari quit last week to go work for some dot.com startup and didn't stick \naround to do any kind of handover, so can you update your contact records to \nshow me as the new contact point for our Web site?  \nƒ \nThanks, Carrie  \nA couple of days later, after the contact records have been altered, the original email is \nfollowed up with a second email: \nƒ \nHi tech support,  \nƒ \nI'm Carrie, the contact point for the wiley.com Web site. Can you set up a \nsubnet (test.wiley.com) on the same Web server that wiley.com is hosted on \nand email me the FTP userID and password so we can upload and test the new \nversion of our Web application?  \nƒ \nThanks, Carrie  \nOf course, instead of uploading a new application, the attacker uploads a toolkit (otherwise \nknown as a rootkit) for compromising the server's system administrator account. \nTricks by Traditional Mail \nAn attacker could even use a mass-mailing strategy, sending letters in the mail to hundreds \nof organizations, in the hope that several of them will fall for the scam. Here's an example: \nƒ \nAttn: Network Engineer  \nƒ \nCongratulations, you've been preapproved for a 1-year free subscription to \nNetworkMag. To claim your gift, please complete the enclosed circulation \nsurvey and return it to us at Bogus Publishing.  \nƒ \nSTOP PRESS.  \nƒ \nIf you accept this offer by <insert date two weeks from today>, we will also \nsend you a complementary round of golf at <insert name of exclusive local golf \ncourse>.  \nOf course, in addition to asking for personal information such as one's first name, last \nname, mailing address, work telephone number, and email address, the survey also asks \nwhat type of hardware and system software the Web site uses. \nƒ \nAttn <insert name of product with a detected security vulnerability> Engineer  \n"
        },
        {
            "page_number": 166,
            "text": " \n159 \nƒ \nBogus Training Corporation would like to offer you a complimentary wall chart \ngraphically depicting all of <insert product name's> most common commands. \nTo receive this free gift, please fill out the attached coupon and return it to us \nhere at...  \nNeedless to say, there is no wall chart; all the attacker is trying to do is find potential \ncandidates who may not have gotten around to installing the latest security patch for a \nvulnerable product. \nTricks in Person \nDepending upon how confident attackers are and how much they are willing to risk being \ncaught to acquire the information they seek, they may attempt to gain entry to a secure \nfacility by blatantly walking in through the front door and utilizing stories like the following: \nƒ \nAt the IT nightshift leader's office: \n\"Hi, I'm Thomas Copeland. I'm with the external auditors Arthur Waterhouse. \nWe've been told by corporate to do a surprise inspection of your disaster \nrecovery procedures. Your department has 10 minutes to show me how you \nwould recover from a Web site crash.\"  \nƒ \nAt an ISP Web-hosting facility: \n\"Hi I'm Cathleen, I'm a sales rep out of the New York office. I know this is short \nnotice, but I have a group of perspective clients out in the car that I've been trying \nfor months to get to outsource their Web-hosting needs to us. They're located \njust a few miles away and I think that if I can give them a quick tour of our \nfacilities, it should be enough to push them over the edge and get them to sign \nup. Oh yeah, they are particularly interested in what security precautions we've \nadopted. Seems someone hacked into their Web site a while back, which is one \nof the reasons they're considering the outsourcing route.\"  \nIn reality, the sales rep and clients are a group of intruders who in addition to learning \nmore information about their next potential target, hope to outnumber the facility staff \ngiving the tour, thereby allowing at least one member of the group to, unnoticed, grab a \nbackup tape or even try to gain access to a server long enough to download a \npassword file onto a floppy disk. \nƒ \nDressed in appropriate engineerial attire (complete with clipboard): \n\"Hi, I'm with Cooler and Sons Air Conditioning. We received a call that the \ncomputer room was getting too warm and need to check your HVAC system.\" \nUsing professional-sounding terms like HVAC (Heating, Ventilation, and Air \nConditioning) may add just enough credibility to an intruder's masquerade to \nallow him or her to gain access to the targeted secured resource.  \nDue to the sheer number of imaginative stories an attacker could come up with, it's not \npractical to describe every possible scenario in an organization's security policies, although \ndocumenting the techniques used (together with preventive countermeasures) for some of \nthe most common methods might merit inclusion. Instead, the best defense against the \nmyriad social engineering attacks is employee awareness. To this end, an organization \nshould have a training scheme in place to ensure that employees (and any temporary staff) \nare able to recognize a con artist at work and promptly notify the appropriate security \npersonal. A training scheme could be evaluated by the security-testing team attempting \nsome of the most common social engineering exploits on unsuspecting employees who \nhave been through the training program and should therefore know better. Table 7.1 \nsummarizes these social engineering checks. \n"
        },
        {
            "page_number": 167,
            "text": " \n160 \n \n \nTable 7.1: Social Engineering Checklist  \nYES  \nNO  \nDESCRIPTION  \n□  \n□  \nDoes the organization have security policies in place that provide \nguidance (preventive measures and procedures for confirming the \nlegitimacy of a request) on the most common forms of social \nengineering? For example, how can a member of the help desk \ndifferentiate between a legitimate user who has forgotten his or her \npassword and an imposter trying to work the system? \n□  \n□  \nHas personal contact information for critical individuals such as the \nwebmaster, LAN administrator, and database administrator (DBA) \nbeen removed from publicly available documentation? Examples \nwould be domain name registrations and corporate phone directories. \n□  \n□  \nDoes the new employee orientation process teach employees the \norganization's security procedures, and, in particular, social \nengineering awareness? \n□  \n□  \nAre employees continually reminded to be vigilant against social \nengineering? For example, are they required to reread the \norganization's security policies on a regular basis? \n□  \n□  \nAre all social engineering attempts made by the security assessment \nteam thwarted by observant employees? \n \nTwarting Dumpster Divers \nThe term dumpster diving is used to describe searching disposal areas for information that \nhas not been properly destroyed. Since this information may be stored on a number of \ndifferent mediums and vary in degrees of its sensitivity, an organization may use a number \nof alternate methods to dispose of this information. The security-testing team may therefore \nneed to evaluate whether the appropriate disposal methods are actually being employed for \nall an organization's waste items. \nProper Disposal of Paper \nAn organization should ensure that any paper-based report, audit trail, transaction log, fax, \nletter, or post-it note containing confidential information is destroyed beyond recognition \nbefore it is discarded. For the most sensitive documents, regular shredding may not be \nenough; instead, crosscutting shredders or furnaces may be required. \n \nCLASSIFYING A DOCUMENT \nMany organizations use multiple categories for specifying how sensitive a particular \ndocument is. Terms such as public, unclassified, sensitive, secret, company confidential, \ntop secret, internal use only, private, and for your eyes only are just some of the names \nused to categorize different levels of sensitivity. Security policies should therefore clearly \nstate the differences between each category, process, or criteria for assigning a document \nto the most appropriate category and the procedures utilized to handle these documents. \n"
        },
        {
            "page_number": 168,
            "text": " \n161 \nCleaning Up Brainstorms \nMany organizations utilize hotel conference rooms or other unsecured facilities (internal or \noff-site) to conduct brainstorming sessions. Unfortunately, it is often the case that once the \nsession is complete, no one considers wiping down the whiteboard(s) used to record the \noutput of the meeting, potentially leaving sensitive information (such as a network topology) \nup on the whiteboard for days, weeks, or basically until somebody else needs to use the \nwhiteboard. \nA quick tour of an organization's meeting rooms should provide the security-testing team \nwith an indication of how adept the organization's culture is at cleaning up sensitive \ninformation after group meetings. \nProper Disposal of Electronic Hardware \nOld hard drives (especially if leased machines are to be returned to a leasing company), \ntapes, Zip and Jaz disks, floppy disks, and CDs that at one time were either used in the \nproduction environment or to back up production data should (where possible) be \ndegaussed (a process described in the sidebar \"Degaussers\"), or overwritten multiple times \nwith dummy data files before being recycled and then physically crushed before being \ndisposed of. \nContrary to popular opinion, overwriting the data once may still leave a magnetic residue \nthat could be used to recreate the erased data. In the case of a reformatted disk, the \noperating system may not even delete the data files but simply mark the sectors on the disk \nused to store this information as free space, again making it possible to recover the data \nthat had appeared to have been destroyed. \nIf, instead of being destroyed on-site, media are transported to a secondary site for disposal \nwhile still in a readable format, adequate precautions should be taken to ensure that an \nattacker cannot intercept these items en route and acquire this information before it reaches \nits final destination. Table 7.3 summaries these checks. \n \nDEGAUSSERS \nDegaussing is the process of passing magnetic media (such as tapes and floppy disks) \nthrough a powerful magnet field to rearrange the polarity of the particles used to store \nelectronic data, thereby completely removing any trace of the previously recorded data. \nDegaussers are sometimes referred to as bulk erasers because whole packs of media \n(such as a carton of 100 floppy disks) can be erased in one operation. Table 7.2 lists some \nsample vendors who manufacture or sell these devices. \n \nTable 7.2: Sample List of Vendors Who Manufacture or Sell Degaussing Devices  \nNAME  \nASSOCIATED WEB SITE  \nBenjamin \nwww.benjaminsweb.com  \nData Devices International \nwww.Datadev.com/degausser.com  \nData-Link Associates \nwww.datalinksales.com  \nData Security \nwww.datasecurityinc.com  \nGarner Products \nwww.garner-products.com  \n"
        },
        {
            "page_number": 169,
            "text": " \n162 \nTable 7.2: Sample List of Vendors Who Manufacture or Sell Degaussing Devices  \nNAME  \nASSOCIATED WEB SITE  \nProton Engineering \nwww.proton-eng.com  \nVerity Systems \nwww.veritysystems.com/degaussers.net  \nWeircliffe \nwww.weircliffe.co.uk  \n \n \n \n \nTable 7.3: Data Disposal Checklist  \nYES  \nNO  \nDESCRIPTION  \n□  \n□  \nDo the organization security policies define what sensitive data is, \nand if multiple categories are used, do they explain how a particular \nitem of data would be assigned to the most appropriate category? \n□  \n□  \nDoes the organization have security policies in place that provide \nguidance on how media that contains (or previously contained) \nsensitive data should be recycled or destroyed? \n□  \n□  \nDoes the policy include all mediums used to store sensitive data such \nas tapes, floppy disks, hard drives, and paper? \n□  \n□  \nDoes the policy provide guidance on what constitutes an acceptable \nlocation for the storage or disposal of sensitive data? For example, \nare DBAs allowed to take database backups home, and are computer \nroom wastepaper bins acceptable receptacles for the disposal of \nnetwork management reports? \n□  \n□  \nDoes the policy include protection measures for the transportation of \nsensitive data to remote locations? \n□  \n□  \nDoes the policy include protection measures for sensitive data at off-\nsite/backup locations or disposal destinations (including third \nparties)? \n□  \n□  \nDo all employees adhere to these data disposal policies? \n \nDefending against Inside Accomplices \nLife is a lot easier for external attackers if someone on the inside is willing to help them (or if \nthe attackers are insiders themselves). This help could take any number of different forms: \nƒ \nA developer building a Trojan horse or time bomb into a Web application (especially \nif the developer is a short-term consultant) \nƒ \nA DBA adding some attacker-friendly stored procedures to the production database \nƒ \nA Webmaster installing a backdoor or rootkit on the Web server \nƒ \nA network engineer making an illicit copy on a production server's entire hard drive \nƒ \nA system administrator forgetting to install an operating system security patch \nƒ \nA value-added retailer (VAR) installing a firewall and leaving the manufacturer's \ndefault maintenance account active \nƒ \nAn engineer from the local utility company attaching a network sniffer to a LAN \n"
        },
        {
            "page_number": 170,
            "text": " \n163 \nƒ \nA member of the cleaning crew retrieving any useful-looking post-its from the \ncomputer room's wastepaper bin \nFortunately, many techniques are available for an organization to protect itself against \nsomeone on the inside abusing his or her privileged position. These techniques generally \nfall in one (or more) of the categories mentioned in the following section. \nPreventative Measures and Deterrents \nThe following sample security measures can either actually prevent an insider from \ncommitting a breach of confidence or, by acting as a deterrent, result in the insider deciding \nnot to attempt the act: \nƒ \nEstablishing a continuous security awareness program. \nƒ \nLocating critical servers in a secure room with the doors locked at all times (not just \nwhen the last person leaves at night). \nƒ \nEnsuring authorized personnel supervise any janitorial staff while they clean a \nsecured area. \nƒ \nIf local employment practices permit, running preemployment (and/or ongoing) \nbackground checks on employees to detect employees who may be predisposed to \ncommitting a network intrusion. Such checks might include calling references, \nconfirming resume details, checking for criminal records, testing for drug use, and \nrunning a credit check. \nƒ \nAppropriately using authorization badges and \"Authorized Personnel Only\" signs. \nFor example, authorized employees should be clearly identifiable and all \nunauthorized personnel (especially visitors) should be escorted at all times when they \nare present inside a secure area. \nƒ \nTraining employees to appropriately challenge someone who does not appear to be \nauthorized for the area they are in. \nƒ \nConfiguring each machine to use a different BIOS password. \nƒ \nConfiguring each machine to use a password-protected screensaver. \nƒ \nMounting video cameras in clearly visible locations. \nƒ \nMonitoring all electronic media that is brought into or out of secure areas. An \nexample would be checking to see that only blank floppies are brought into a secure \narea and only crushed floppies are removed. \nƒ \nEstablishing revolving duties. For example, each week reassigning the machines \nthat each LAN administrator is responsible for. Unless there is a conspiracy, corrupt \nLAN administrators will know that they run the risk that a vigilant peer will detect one \nor more of their wrongdoings. \nƒ \nDefining procedures that separate duties or require dual controls and consequently \nprevent any single employee from accomplishing anything significant without the help \nof a second employee. An example would be only allowing the librarian to promote \nnew protection code into the staging area and then restricting the webmaster from \nuploading files on to a production Web server from anywhere else other than the \nstaging area. Or perhaps network administrators would be required to get a \nsupervisor's approval before creating a new user account. \nƒ \nRequiring all employees who are leaving the organization (both voluntarily and \ninvoluntarily) to return all hardware owned by the organization, changing any \npasswords that the employees might have known and disabling their personal login \nIDs (especially any dial-up accounts). \nFrom a security perspective, requesting the return of software may be a moot point, due \nto the ease with which software can be duplicated (CD burners have become extremely \ninexpensive). However, an organization may still want to explicitly request the return of \n"
        },
        {
            "page_number": 171,
            "text": " \n164 \nsoftware, so that any software licenses can be recycled and to prohibit a former \nemployee from legally using the software. \nƒ \nUsing objective third parties to regularly perform security assessments. \nDetective Measures \nDetective measures are intended to alert an organization to the fact that a security breach \nhas, is, or will take place (discussed further in Chapter 8). Some examples of measures that \ncan be used to detect an inside accomplice include the following: \nƒ \nInstalling anomaly-based intrusion detection systems that monitor for unusual \nbehavior \nƒ \nUsing tripwires that check for any activity on theoretically static files \nƒ \nPlacing motion detectors in supposedly uninhabited rooms \nƒ \nMonitoring outbound emails for suspicious content (assuming local emplacement \npractices permit this activity) \nƒ \nInstalling keyboard-logging devices (such as the ones listed in Table 7.4) on critical \nconsoles (assuming local emplacement practices permit this activity) \n \nTable 7.4: Sample List of Keyboard-Logging and Screen-Recording Tools  \nNAME  \nASSOCIATED WEB SITE  \nBlack Box \nwww.enfiltrator.com  \nComputer Spy \nwww.computerspy.com  \neBlaster/Spector \nwww.spectorsoft.com  \nKEYKatcher \nwww.keykatcher.com  \nOnline Recorder \nwww.internet-monitoring-software.com  \nSnapshot Spy \nwww.snapshotspy.com  \nWinGuardian \nwww.webroot.com  \nCorrective and Prosecutive Measures \nOnce a wrongdoing has been detected, an organization will need to attempt to recover from \nthis disturbance (a topic discussed in more detail in Chapter 8). The corrective actions will \nobviously be heavily dependent on what and how much damage has been done. An \norganization is also likely to want to identify beyond repudiation which individual(s) was \nresponsible for the security breach and take some form of action to ensure that the event \ndoes not happen again. If formal disciplinary actions or criminal prosecutions are sought, \nthen the organization will need evidence indicating the suspected employee's guilt. Such \nevidence is typically much easier to collect if the systems and computer facilities have been \ndesigned with this requirement in mind. Examples of such measures include the following: \nƒ \nCreating transaction logs, network audit trails, file access (including read) logs, and \nWeb logs \nƒ \nLogging keyboard stokes from critical terminals \nƒ \nVideotaping everyone entering and leaving a restricted area \nAs can be seen from these preceding scenarios, an insider could consciously assist an \nexternal attacker or commit an intrusion him- or herself in many different ways. Although \ntrying to protect against every conceivable scenario is unlikely to be feasible, it's not \nunreasonable to attempt to ensure that at least one preventive, detective, and corrective \nmeasure exists for each category of employee that has access to some critical or sensitive \n"
        },
        {
            "page_number": 172,
            "text": " \n165 \nresource. (Table 7.5 summarizes these checks.) This may not be as hard to implement as it \nmight first appear, as some security measures may actually fulfill the needs of more than \none category. For example, a monitored video camera in plain sight that is backed up to \ntape may act as a deterrent, allow an on-duty guard to detect unauthorized access, and \nalso provide evidence that can be used in a prosecution. \n \n \n \nTable 7.5: Inside Accomplice Checklist  \nYES  \nNO  \nDESCRIPTION  \n□  \n□  \nHas the organization clearly defined the unacceptable activities, and \nare employees made aware of these activities and the ramifications \nshould they pursue them? \n□  \n□  \nDoes the organization have security policies in place that include \nmeasures to protect against an inside (as well as an outside) \nassailant? Note that some of these measures may be more effective \nif left undisclosed to anyone not involved in implementing the policy, \nwhile other policies should be published if they are intended to act as \na deterrent. \n□  \n□  \nSpecifically, are there preventive measures in place to dissuade any \nemployee with access to sensitive data or a critical component of the \nWeb site not to behave inappropriately? \n□  \n□  \nSpecifically, are there detective measures in place that should detect \ninappropriate employee behavior? \n□  \n□  \nAre any inappropriate actions performed by the security assessment \nteam as part of the security assessment detected? \n□  \n□  \nOnce an internal security breach has been detected, are procedures \nin place that would allow the organization to correct this situation? \n□  \n□  \nIf the organization would like to have the option of prosecuting or \ntaking formal disciplinary action against an employee suspected of \ninappropriate behavior, is the system designed to collect sufficient \nevidence to allow this to happen? \nAn organization may wish to internally publicize an attack initiated from the inside, helping \nto raise employee awareness (and perhaps act as a deterrent). Alternatively, an \norganization may wish to keep knowledge of such an attack confidential, possibly because \ncriminal charges may be brought, or to reduce the likelihood that news of the attack makes \nit into the public domain and subsequently causes the organization to lose credibility. Either \nway, the organization should clearly document what its standing policy is toward this matter. \n \nPreventing Physical Attacks \nEven the best firewalls provide no protection against an unauthorized person gaining \nphysical access to a machine. Intruders don't have to be able to pick up a machine and \nwalk out of the building. Typically, all they need is an unguarded console or an easily \nguessed screensaver password and a few minutes to change a few security settings, or to \n"
        },
        {
            "page_number": 173,
            "text": " \n166 \nbe able to copy all important password files to a floppy disk. For example, even if nobody is \nlogged in or the screensaver password is unguessable, an intruder could potentially \ncircumvent this precaution. This could be achieved by simply turning the power off, \nrestarting the machine by booting to DOS (via a floppy disk or CD), and then copying the \nfile onto the floppy before finally executing a normal reboot. \nNo security assessment is complete without addressing the physical security used to \nprotect the hardware that the Web site will utilize and the facility that it is to be housed in. \nDisregarding this topic is analogous to only crash-testing the front half of a car, ignoring the \nless frequent but potentially more dangerous side and rear impacts. Therefore, tests should \nbe designed to ensure that each component of a Web site (the servers, the cabling, the \nmedium used to store software and data, and the facilities) is adequately protected from an \nintruder willing to employ some form of physical attack. \nSecuring a Facility \nPut simply, if attackers don't know where the physical l A great many precautions can be \nimplemented when trying to ensure that no one is able to gain unauthorized physical \naccess to a secure area, such as the computer room used to store the servers that host the \nWeb site being assessed. The best form of defense is a layer approach, requiring an \nintruder to successfully breach several security measures before being able to access a \ntarget machine. \nShielding Knowledge of a Facility's Proximity \nocation of the Web site is, they are not going to be able to break into it. Therefore, to hinder \nany attempt by an attacker to acquire this information, all nonessential references to a \nsecure location should be removed. For example, corporate telephone and mailing \ndirectories should be scrutinized to determine if the location of a secure site could be \ndeduced. Removing driving directions from the organization's Web site and road signs from \nprivate roads may not only frustrate and delay a FedEx delivery guy, but also a potential \nintruder trying to reconnoiter a target facility. \nProtecting the Facility Perimeter \nMany secure sites use a fence or wall as an outer perimeter, creating a buffer zone that an \nintruder must cross before reaching a building's exterior. Ideally, some form of detection \nmechanism such as monitored video cameras, motion detectors, or even guard dogs \nshould be used to detect entry into this buffer zone. With the possible exception of guard \ndogs, the security-testing team may want to consider running unannounced mock intrusions \nat different times of the day to ascertain whether or not an intruder might be able to evade \ndetection while traversing this outer defense. Having an accomplice create a diversion such \nas delivering pizza to the guards charged with monitoring the buffer zone may provide a \nmore strenuous variation of the test. \n \nLOCATION DECEPTION \nDuring the Cold War, the Soviet Union was particularly adept at naming sensitive facilities \nafter towns located hundreds of kilometers away and ensuring that these facilities (and \neven their supporting townships) never showed up on any maps. Although it's probably a \nlittle James Bond-esque to use a functioning bakery as a facade for a secure computer \nroom, trying to pass the room off as a document archive might not be unreasonable. \n"
        },
        {
            "page_number": 174,
            "text": " \n167 \nIn the case of video surveillance, also check that tapes are not recycled too quickly. In the \nevent of an attempted break-in, tapes may need to be reviewed to help determine when the \nevent occurred and identify who the assailant was. This task can be made near impossible \nif the tape has already been overwritten with more recent surveillance footage. \nProtecting Facility Entrances \nIt goes without saying that all external entry points such as doors and windows should be \nchecked to ensure that they are always locked and that any attempt to break in through any \nof these openings would result in an alarm (silent or otherwise) being raised. Less obvious \nis checking that the means by which the alarm is to be raised is functional and tamper-\nresistant. This would include ensuring that any telephone line used by an alarm has not \nbeen disconnected by the telephone company due to no one paying the standing charge for \nthe line (yes, this happens). Another example would be making sure an intruder can't nullify \nthe alarm by simply cutting the wire running from the alarm box to the telephone poll (to \nmitigate this threat, some alarms employ a line supervision technique to protect their \nconnection to the outside world from being cut without anyone being notified). \nAlthough safety considerations may necessitate multiple exits from a secure facility, from an \nideal security perspective, there should only be one entryway into the facility (and no \nwindows present). This entrance should be tested to ensure that it has a reliable method of \nensuring that only authorized personal are allowed in. Traditional methods of restricting \nentry to a facility include smart cards, photo badges, and good old-fashioned keys. \nMaking Rooms Secure \nThe entrance to the room(s) storing the computers themselves may be protected by \nanother authentication method, ideally different in nature from the one used to gain access \nto the building. The method could be a biometric device such as a fingerprint scanner or a \nvoice recognition unit. Once again, whatever form of authentication is used, it should be \nchecked to ensure that it cannot easily be fooled or circumvented. For example, many \ncomputer rooms are housed in facilities that make use of artificial ceiling or raised floors. If \nused, these structures should be checked to ensure that an intruder can't circumvent a \nlocked door by simply crawling under or over the door. \nIf video surveillance is used within a computer room, the positioning of the cameras should \nallow a computer's keyboard and/or screen to be clearly viewable or angled so that these \ndetails can definitely not be captured. If the former placement is adopted, then precautions \nshould be taken to ensure that userIDs, passwords, and other sensitive information are not \ninadvertently leaked by an unauthorized person gaining access to the surveillance videos. \nProtecting Cabling \nAlthough locked doors protect many computer rooms, the cables and telephone wires that \nenter and leave these rooms are often left unprotected. Access to cable and telephone \nwiring closets or local exchanges should be as secure as access to the computer room. \nOtherwise, an eavesdropper armed with a network sniffer may be able to obtain all the \ninformation he or she needs without ever setting foot inside the computer room itself. \nSecuring Hardware \nIf an intruder is able to gain physical access to the inner sanctum, all may not be lost. If the \nintruder is interested in acquiring some piece of data on one (or more) of the servers, as \nopposed to destroying or stealing the hardware, then hardening the server may either \n"
        },
        {
            "page_number": 175,
            "text": " \n168 \nthwart or at least delay intruders long enough for their illicit activity to hopefully be detected. \nServer-hardening options include the following: \nƒ \nEnabling screensaver password protection. \nƒ \nLocking down any network-accessible tape unit or CD jukebox. \nƒ \nRemoving all floppy drives and CD (or other external media) drives. Note: This \nmeasure may be circumvented if external ports are available for an intruder to plug \nhis or her own media devices into and the operating system is able to auto detect the \ndevice (plug-and-play style). \nƒ \nRemoving all nonessential external ports: serial, parallel, IDE, SCSI, USB, and \nfirewire. Note: This measure may be circumvented if an intruder is able to gain \nphysical access to the machine's motherboard and is able to install his or her own \ninterface cards. \nƒ \nUsing a locked shield to restrict access to the server's on/off switch or reset button \n(and thereby inhibiting an intruder from rebooting the machine). Note: This measure \nmay be circumvented if the power cable is easily accessible. \nƒ \nConfiguring machines to boot from their hard drive first (just in case intruders try to \nboot from their own floppies or CDs). Note: This measure may be circumvented if an \nintruder is able reset the machines' BIOS configuration. \nƒ \nEnabling BIOS-level password protection. \nƒ \nAutomatically sending an alert to the LAN administrator every time the machine is \nrebooted. \nSecuring Software \nOften-overlooked physical assets that should be secured against thieves are CDs and other \nmedia used to store the system software, third-party software, and application software \nused by the Web site. Although it may prove easy to obtain the most recent version of an \noperating system, it may be much more troublesome to try and locate an older version that \nis not being distributed by the vendor but is still being used by the Web site. In some \ninstances, the vendor may have stopped supporting the product or even gone out of \nbusiness. Conversely, the challenge with application software is making sure that the \nversion used for a restoration is up-to-date with the most recent fixes. No one wants to \nrestore a server back to a previous version and then have to manually reapply the last 20 \nfixes before it can go back into production. \n \nSAMPLE SECURE ROOM AND SERVER ACCESS TESTS \nTo test the physical security of a Web site, consider asking an unauthorized fellow \nemployee to attempt to gain physical access to a target machine located within a \nsupposedly secure room. Then once inside, attempt to copy the machine's password file \n(such as the SAM file on a Windows NT/2000 system or the passwd file on many UNIX \nsystems) to a floppy disk. Once captured, this file could be cracked at leisure by a real \nattacker. Some tactics that the unauthorized employee could use include the following: \nƒ Working late and attempting to follow the cleaning crew into the locked computer \nroom \nƒ Standing by the locked computer room door with two very full cups of coffee, waiting \nfor somebody to open the door \nƒ Taking advantage of a scheduled fire alarm drill to slip unnoticed into an unlocked \ncomputer room (Even if an alarm were to go off, who would hear it?) \nƒ Adding the words \"IT Auditor\" (ideally in big red letters) to a standard identification \nbadge and calmly following someone else into the secure area \nƒ Carrying a large box that appears to contain a new server and waiting for somebody \nto open a locked door \n"
        },
        {
            "page_number": 176,
            "text": " \n169 \nƒ Carrying a clipboard and toolbag with a utility company's name on it and then \nrequesting to be allowed access to the utilities junction box \nƒ Offering to pick up lunch (pizzas, sandwiches, and so on) for someone in the \ncomputer room or even pretending to be the pizza delivery guy \nIf the fellow employee is successful, consider repeating the test, but this time using the \nassistance of a contractor or somebody off the street. \nSoftware libraries (ideally located off-site) and configuration management systems should \nalso be audited to ensure that in the event of a disaster this software vault contains a copy \nof all the software needed to recreate the Web site and its associated application(s). In \naddition, the mechanism for checking out any physical copies should be checked to ensure \nthat an intruder could not easily pilfer a critical disk or review an application's source code \nto look for an as-yet-undiscovered vulnerability. \nSecuring Data \nMany organizations offload historical data from production servers onto backup/ archival \nmediums such as tapes, Zip disks, and CDs. For a Web site that does so, the storage of \nthese tapes and disks should be checked to ensure that they are stored as securely as the \ncurrent production data. For instance, a DBA backing up the production database onto tape \nand then storing it in a home office may sound like a cheap off-site backup strategy, but it \nmay also be easily exploitable by an intruder knowledgeable about this practice and willing \nto break into a residence while no one is home. That's much easier than having to deal with \nthose pesky guard dogs and heavyset guards protecting the facility housing the Web site. \nIn addition to checking the security of any off-site location, the transportation of any \nsensitive data to or from the site should also be considered, especially if a predictable \npattern is used to transfer this information. Although carjacking is perhaps a little farfetched, \nwalking off with the floppies sitting in the network engineer's briefcase is not. Table 7.6 \nsummarizes this list of physical security checks. \n \nTable 7.6: Physical Security Checklist  \nYES  \nNO  \nDESCRIPTION  \n□  \n□  \nDoes the organization have security policies in place that provide \nguidance on what physical security measures should be implemented \nat each secure facility? \n□  \n□  \nAre all recommendations on obscuring the actual location of a facility \nfollowed? \n□  \n□  \nAre all recommendations on securing the facility's perimeter \nfollowed? \n□  \n□  \nAre all recommendations on securing the entrance to a facility \nfollowed? \n□  \n□  \nAre all recommendations on securing the room(s) used to house the \nWeb site's hardware infrastructure followed? \n□  \n□  \nAre all recommendations on securing any cabling that the Web site \ndepends on followed? \n□  \n□  \nAre all recommendations on securing each server used by the Web \n"
        },
        {
            "page_number": 177,
            "text": " \n170 \nTable 7.6: Physical Security Checklist  \nYES  \nNO  \nDESCRIPTION  \nsite followed? \n□  \n□  \nAre all recommendations on protecting the media that stores software \nused by the Web site followed? \n□  \n□  \nAre all recommendations on protecting the media that stores current \nor archived data used by the Web site followed? \n□  \n□  \nAre all unannounced attempts by the security assessment team to \ngain physical access to the hardware, cabling, or media used to store \nsoftware or data thwarted? \n \nPlanning against Mother Nature \nPlanning against and recovering from natural disasters is in itself an entire discipline and, \nas such, developing a disaster recovery plan and an associated business continuity plan is \nin all probability outside the scope of any security-testing effort. That said, during a security \nassessment, it would be prudent to check that these plans do indeed exist and cover such \nmajor eventualities as hurricanes, tornadoes, floods, earthquakes, subsidence, and \nlightning strikes. Maiwald et al. (2002) and Toigo et al. (1999) provide additional information \non disaster recovery. \nOne area that may not be covered by a disaster recovery plan is that of environmental \ndegradation caused by nature. Examples include dust, mold, bacteria, condensation, \nhumidity, ultraviolet light, cosmic radiation, and static electricity slowly corroding away \nhardware components. Fortunately, such slow-acting deteriorations should ensure that only \none hardware component will fail at a time and should therefore not seriously impact the \nWeb site. Two exceptions do exist, however; if the failing component is a single point of \nfailure, or if the failure of one component can cause a chain reaction that results in \nadditional components failing, then the entire Web site may be affected. Hopefully, any \nsingle points of failure and potential chain reactions were evaluated as part of the Web \nsite's reliability and availability test plan, and they can therefore be regarded as outside the \nscope of the security-testing effort. Of course, this assumption should be confirmed. Table \n7.7 summarizes this list of checks. \n \nTable 7.7: Protection from Nature Checklist  \nYES  \nNO  \nDESCRIPTION  \n□  \n□  \nHas a disaster recovery and business continuity plan been developed \nfor the Web site? \n□  \n□  \nDo these plans cover all conceivable natural disasters that could in \nany way affect the Web site? \n□  \n□  \nHas the Web site's design been reviewed and tested to determine if \nany single point of failure exists? \n□  \n□  \nIf one or more single points of failure exist, is the organization willing \nto accept the risk associated with these critical components failing? \n□  \n□  \nHas the Web site's design been reviewed and tested to determine if a \nfailure by a single component could result in a chain reaction that \n"
        },
        {
            "page_number": 178,
            "text": " \n171 \nTable 7.7: Protection from Nature Checklist  \nYES  \nNO  \nDESCRIPTION  \ncould significantly impact the Web site? \n□  \n□  \nIf a chain reaction could occur, is the organization willing to accept \nthe risk associated with this event occurring? \n□  \n□  \nHave the disaster recovery and business continuity plans been fully \ntested? \n□  \n□  \nHas environmental compliance been covered by another testing \neffort? \n \nGuarding against Sabotage \nWhat if an attacker is not interested in stealing data or hardware, but instead wants to \nlaunch a denial-of-service (DoS) attack using physical means? This attack could be as \nobvious as putting a shovel through the telecommunication lines connecting the Web site to \nits ISP or swinging a sledgehammer in a computer room. Or it could be as subtle as placing \npowerful magnets next to unshielded cables or removing nearly all the memory cards from \nthe Web server-attacks that might not completely bring down the Web site but could \nsignificantly impact its capacity. \nSome attacks don't even require the assailant to be present. Telephoning in a bomb threat \nor sending hazardous material in the mail can cause disruption not only to the computer \nroom, but also to the entire facility for an extended period of time. \nIdeally, the plans and procedures put in place to handle natural disasters should also \nprovide equivalent protection against these human-made disasters as well. However, it \nwould be a wise precaution to review these plans to ascertain whether or not they do \nindeed cover an attacker deliberately sabotaging any of the critical components on which \nthe Web site depends. \n \nSummary \nAs can be seen from the topics covered in this chapter, ensuring the security of a Web site \nencompasses much more than simply testing software for security holes. Failing to consider \nthese other potential exposures during a security assessment is likely to leave an \norganization open to some form of alternate attack, especially when the physical security of \na Web site is distributed across several departments that operate independently. For \nexample, building security may monitor video cameras, while the IT department keeps track \nof who has a key to the computer room. But who keeps an eye on those critical utility \njunction boxes? \n"
        },
        {
            "page_number": 179,
            "text": " \n172 \nChapter 8: Intruder Confusion, Detection, and Response \nOverview \nThis chapter looks at the defenses and processes an organization may have put in place to \nrespond to attacks by accomplished intruders. The first section, Intruder Confusion, focuses \non tactics that an organization may have adopted to confuse and distract intruders, thereby \nfrustrating them to the point that it is hoped they voluntarily give up their pursuit, or at the \nvery least hinder them long enough for the organization to detect them at work. Testing \ndefenses that rely upon confusion differs from testing traditional security precautions in that \nactually knowing the desired outcome (a state of confusion) may influence the testing and \nconsequently affect the validity of the test results. Consequently, orchestrating a test of a \nconfusion-based defense typically necessitates a more unorthodox testing strategy, an \napproach expanded upon by this chapter. \nOf course, slowing down a dedicated intruder isn't likely to help much unless intrusion-\ndetection mechanisms are in place to detect the intruder's activities. The second section of \nthis chapter, Intrusion Detection, reviews the various methods an organization could employ \nto detect an intruder and how these methods could be tested to ensure their effectiveness. \nThe final section, Intrusion Response, discusses the need for an organization to plan ahead \nof time how they might react to an intruder probing a system looking for an exploitable \nsecurity hole or, the unthinkable, an intruder being detected in an unauthorized area. Once \ndocumented, this plan can be reviewed to ensure its comprehensiveness and be exercised \nto ensure that, when needed, the on-duty staff will implement it correctly. \n \nIntruder Confusion \nSome people in the security industry believe that, given enough time and resources, any \nWeb site can be cracked. This belief appears to be born out with the seemingly ever-\ngrowing list of high-profile (and therefore assumed to be well-fortified) Web sites that have \nbeen successfully compromised. The theory behind this belief is that dedicated attackers \nwill eventually always find a way through or around a fixed, or static, defense. It is merely a \nmatter of time before they find some little crack that they can then exploit into something \nmore threatening. Applying this theory to American football, even the worst offense in the \nleague will sooner or later figure out a way to beat any defense that never changes its \nformation or strategy. \nSo how can an organization avoid this apparent inevitability? By implementing better and \nmore up-to-date static defenses an organization can certainly make the task much more \ndifficult and therefore delay the attackers, perhaps long enough that they give up trying. \nHowever, there is an alternate strategy that could be deployed and, if successful, might \nthwart an attacker indefinitely. If techniques designed to confuse an attacker are used, the \nattacker may never get a clear picture of the Web site being attacked and therefore never \nbe able formulate a strategy that will ultimately lead to its demise. \nDynamic Defenses \nOne approach that can prove to be a formidable defense is that of a moving target. Even \nthe most tenacious of attackers will become confused and frustrated if the target they are \nso painstakingly mapping out keeps changing before they are able to utilize any information \nthey have been able to acquire. Options for varying a Web site's configuration include the \nfollowing: \nƒ \nFrequently changing all the system administrator userIDs and passwords. \n"
        },
        {
            "page_number": 180,
            "text": " \n173 \nƒ \nRegularly reassigning new IP addresses to each of the Web site's servers. This \nsituation may already be occurring if the network uses a Dynamic Host Configuration \nProtocol (DHCP) to assign IP addresses to its servers. \nƒ \nInstalling and configuring multiple operating systems on a server and occasionally \nrebooting the machine into a different operating system. This strategy is much easier \nto implement for a relatively simple application like a Web server than a more \ncomplicated environment like an application server being used to support a Java 2 \nPlatform Enterprise Edition (J2EE) environment. A side benefit of this approach is that \nif a new vulnerability is discovered in one of the operating systems, the other \noperating system can be utilized until a patch or workaround is developed for the \nnewly discovered hole. \nUnfortunately, continuously reconfiguring a Web site is likely to place too heavy a burden \non all but the highest funded IT department. Also, the risk is run that one of these frequent \nchanges will inadvertently introduce a security hole that might not be discovered by the \norganization until after an intruder has found and exploited it. \nDeceptive Defenses \nFor thousands of years, military commanders have realized that often it is not necessarily \nthe strongest army that wins a battle but rather the army that is best informed. \nConsequently, a general attaches great value to discovering an opposing army's size, \nstrength, location, and objectives. Also highly valued is a means to deceive the foe with as \nmuch disinformation as possible in order to hide the general's own true disposition and \nintentions. \nIn the IT security world, this strategy of deception is most easily implemented through the \nuse of decoys (or red herrings), incorrect or useless information that has been purposely \nplanted for attackers to find and thereby causes them to waste their time and energy on a \ncompletely useless endeavor. The following are some examples of decoys that could be \nused to frustrate and slow down an intruder who is able to compromise a network's outer \ndefenses: \nƒ \nPlace some files called userids.txt and passwords.txt on a Web server, and populate \nthese files with completely bogus account information. \nƒ \nAdd a table to the production database called CreditCards and then fill this table with \ninvalid credit card numbers. \nƒ \nInstall the operating system and any other software packages in nondefault \ndirectories and then manually create empty directories using the names the default \ninstallation would have used. For example, install Windows 2000 into a directory \ncalled Acrobat and then create an empty Windows directory so that attackers waste \ntime trying to figure out why they can't see any of these system files. Alternatively, \ninstead of leaving this decoy directory empty, a second nonexecuting version of the \noperating system with different configuration settings could be copied into the \ndirectory. \nƒ \nSome Unix environments support the chroot command, which may be used to limit \nthe visibility of a system's real directories and files, and instead enable access to a \nbogus (decoy) file structure. \nOf course, as with dynamic defenses, the added complexity of a deceptive defense may \nconfuse not only a would-be intruder, but also the legitimate administrators of a Web site, \npotentially causing them to introduce more errors than they would under a more \nstraightforward security strategy. An organization should therefore carefully weigh these \npros and cons before deciding to implement supplemental defenses based on deception. \n"
        },
        {
            "page_number": 181,
            "text": " \n174 \nHoney Pots \nA honey pot is a variation of the decoy theme. In this case, a resource is purposely left \npoorly defended. Unbeknownst to the attacker, this resource has no data or privileges that \ncould be used against the Web site and has been set up with a tripwire to automatically set \noff an intruder alert as soon as the resource is accessed. The Honeynet Project et al. \n(2001) provides additional information on the concept and employment of honey pots. \n \nANCIENT ADVICE \n'All warfare is based on deception. \n\"It is often possible by adopting all kinds of measures of deception to drive the enemy into \nthe plight of making erroneous judgments and taking erroneous actions, thus depriving him \nof his superiority and initiative.\" \n-Sun Tzu, The Art of War, circa 500 B.C., Griffith Translation, Oxford University Press, 1963 \nOne of the reasons honey pots should always have some sort of tripwire implemented is \nthe risk that the easily compromiseable honey pot machine may be used as a staging point \nfor not just an attack on the organization's Web site, but other people's sites as well (a \npotentially libelous situation). \nThe honey pot could be used passively, with the security personal simply watching the \nattackers at work and learning from their behavior the types and frequency of the exploits \nthat might be employed against the real Web site. A more offensive strategy would be to \ngather information that could be used to identify and prosecute the attackers. Indeed, honey \npots come in all shapes and sizes: \nƒ \nA standalone Web server posted outside an organization's perimeter firewall could \nbe used by an organization as an early warning system. \nƒ \nIn the case of Web sites that support extremely sensitive applications (such as the \nmilitary), an entire network of honey pots may be deployed, complete with fictitious \nusers requesting data and services just to make the network look real. In reality, the \nonly real users are the security personnel watching and tracing the attackers who try \nto compromise the network of honey pots. \nƒ \nLaw enforcement agencies have been known to recreate fully functional e-commerce \nWeb sites that appear to be storefronts for small naive businesses that, not too \nsurprisingly, have not protected their clients' credit card information too well. Of \ncourse, these customer records contain bogus credit card numbers whose only useful \npurpose is to help expose the identity of the attackers, should they try to use these \ncard numbers. \n \nDECEPTION AT WORK \nƒ During World War II, the Allies spent considerable effort trying to convince the \nGerman army that their intended invasion of Western Europe would take place in \nthe vicinity of Calais. By many accounts, when the invasion got underway in \nNormandy, Hitler refused to commit the main body of the German army to this \nlanding, as he believed the Normandy invasion was a diversion for the real \ninvasion, which he still expected to occur at Calais. By the time the German high \ncommand realized that there would be no Calais landing, the Allies had secured \ntheir beachheads and were making good progress into the French hinterland. \n \n"
        },
        {
            "page_number": 182,
            "text": " \n175 \nCommercial organizations who market intrusion-detection systems and alert centers who \nreport on the latest exploits may deploy numerous honey pots scattered around the world \nin order to discover what new tactics are being developed and refined by attackers. In \neffect, they are creating a global intruder weather map.  \nA more cost-effective approach to deploying an entire network of honey pots is to configure \na single machine to appear as if it is several distinct servers, creating in effect a virtual \nhoney pot network. This reduces the cost of the hardware and system software needed to \nrecreate a network, making the honey pot more complex. \nOne more consideration when weighting whether or not to deploy a honey pot (or pots) is \nthat adding an enticing, poorly defended resource to a Web site may actually draw attention \nto the site. It may also consequently attract more assailants than if no honey pot were \ndeployed, which is hardly a desirable situation. Table 8.1 lists some sample honey pots. \n \nTable 8.1: Sample List of Honey Pots  \nNAME  \nASSOCIATED WEB SITE  \nCyberCop Sting \nwww.nai.com  \nDragnet \nwww.snetcorp.com  \nHoneyNet Project \nwww.project.honeynet.org  \nManTrap \nwww.recourse.com  \nNetFacade \nwww.netsecureinfo.com  \nEvaluating Intruder Confusion \nFor a confusion strategy to be effective, its very existence must be kept secret, meaning \nthat this is one of the few features of a Web site that should not be documented in the Web \nsite's standard requirements documentation. Instead this aspect of the Web site's design \nshould be secretly documented elsewhere, and only made available on a need-to-know \nbasis. \nAssuming one or more confusion strategies have been employed, any serious evaluation of \nthe deceptiveness (and hence effectiveness) of these strategies must be done by testers \nwho are not in the know. Obvious choices are the firms that specialize in security testing \nand have had no prior knowledge of the Web site to be tested; if such firms are to be used, \nthen their activities should be closely monitored to see if they can be misdirected or delayed \nby the site's confusion strategy. Even the most honest of security testing firms is unlikely to \nwant to admit they were duped, and if it is mentioned at all in the firm's test summary report, \nit is likely to be underreported. It somehow just doesn't sound right to be telling a client that \nyou wasted an extended amount of expensive time being sidetracked. \nOf course, the acid test for a confusion strategy is not determining if consultants at a \nsecurity-testing firm can be fooled, but whether real-life attackers can be tricked. In this \nsense, the best way to gauge the effectiveness of these diversions is to monitor their use in \na production environment. Compare how many attackers are suckered into spending at \nleast some measurable amount of time on the decoys, as opposed to the attackers that do \nnot stumble across the decoys or seem to ignore their existence (perhaps because they \ncan see them for what they really are). \n"
        },
        {
            "page_number": 183,
            "text": " \n176 \nUnfortunately (from a test metric's perspective), if the decoys are deployed within an \norganization's perimeter defenses and (as yet) no known intruder has compromised this line \nof defense, the test sample is going to consist of zero occurrences, hardly an effective \nmeans of measuring decoy effectiveness. In such circumstances, an organization may wish \nto create a bogus Web site with sufficiently weak perimeter defenses to permit a statistically \nsignificant number of attackers (test samples) to encounter the decoys. This is a rather \nexpensive exercise but ultimately is the only realistic way of testing this deceptive line of \ndefense. Table 8.2 summarizes these confusion strategies in the form of a checklist. \n \nTable 8.2: Confusion Checklist  \nYES  \nNO  \nDESCRIPTION  \n□  \n□  \nHave dynamic defenses been considered for the Web site? \n□  \n□  \nHave deception defenses been considered for the Web site? \n□  \n□  \nHave honey-pot deployments been considered for the Web site? \n□  \n□  \nHave all the specific details on the implementation of any deception \ndefense been removed from the Web site standard requirements \ndocumentation and placed in a separate document with extremely \nrestricted access? \n  \n  \nHave any defenses that rely on confusion been evaluated to \ndetermine just how confusing they actually are to a would-be intruder \n(and optionally for a system administrator to manage)? \n \nIntrusion Detection \nHaving an intruder compromise your Web site is a situation that's not supposed to happen \nbecause \"the perimeter firewall will protect us.\" But what if the unthinkable happens and an \nintruder makes it past a Web site's perimeter defenses, perhaps because the intrusion is \ninitiated from the inside? It would be nice to know that there is still a chance that the \nintruder will be detected and stopped before he or she has a chance to do any significant \ndamage, a topic expanded upon by Allen (2001), Northcutt (2000), and Proctor (2000). \nSecurity personnel monitoring a Web site may be alerted to the presence of an unwanted \nguest in many ways. \nThe following are example tale-tale signs of an intruder at work: \nƒ \nShrinking log files \nƒ \nChanges to file attributes of supposedly static files (date, size, user, and so on) or \ndirectories (possibly caused by an intruder planting a toolkit for later use) \nƒ \nUnexplained performance degradation \nƒ \nThe creation of unauthorized new user accounts \nIntrusion Detection Systems (IDSs) \nRelying on the vigilance of an overworked LAN administrator to continuously monitor raw \nWeb logs and other eyeball-unfriendly files is likely to result in the administrator missing the \ntell-tale signs of an intruder at work. For this reason, several companies have developed \ntools that can be used to automate this tedious task and thereby detect an intruder \nrummaging around. These tools are commonly referred to as intrusion detection systems \n(IDSs). Table 8.3 lists some sample IDSs. \n"
        },
        {
            "page_number": 184,
            "text": " \n177 \n \nTable 8.3: Sample IDS List  \nNAME  \nASSOCIATED WEB SITE  \nAttacker \nwww.foundstone.com  \nBlackICE and RealSecure \nwww.iss.net  \nCisco IDS (NetRanger) \nwww.cisco.com  \nDragon \nwww.enterasys.com  \ne-Sentinel \nwww.esecurityinc.com  \nEtrust \nwww.ca.com  \nIDS/9000 \nwww.hp.com  \nINDESYS \nwww.infinity-its.com  \nIntruder Alert/NetProwler \nwww.symantec.com  \nISA Server \nwww.microsoft.com  \nManHunt/ManTrap \nwww.recourse.com  \nNFR HID/NID \nwww.nfr.com  \nOneSecure IDP \nwww.onesecure.com  \nScanlogd \nwww.openwall.com  \nSnort \nwww.snort.org  \nSTAT \nwww.statonline.com/harris.com  \nStormWatch \nwww.kena.com  \nTriSentry \nwww.psionic.com  \nIDSs can be crudely categorized into four groups, although some IDSs may exhibit \nproperties of more than one group: \nƒ \nSignature detection via network traffic \nƒ \nSignature detection via host activities \nƒ \nAnomaly detection via network traffic \nƒ \nAnomaly detection via host activities \nSignature detection works by matching the observed parameters of network traffic or host \nactivities against a database of known attack characteristics (signatures). Unfortunately, \nthis database must be updated regularly to keep up with newly developed attack methods \nand consequently is poor at detecting new types of attacks. In comparison, anomaly \ndetection works by comparing current network traffic and/or host activities with a previously \ndefined baseline of normal behavior, looking for suspicious deviations from this norm. \nOf course, a network-based IDS will only scan network traffic that it is able to detect; a \nsegmented network that does not broadcast network traffic to every device on the network \nis likely to need several network-based IDSs in order to listen to all of the traffic being sent \nacross its network. In addition, skilled attackers have been known to deceive network-\nbased IDSs by breaking up, or fragmenting, their attacks (thereby making their attack \nsignatures hard to detect) as they are sent over the network to their intended target. \nTherefore, a host-based IDS solution may need to be considered as an alternative (or in \n"
        },
        {
            "page_number": 185,
            "text": " \n178 \naddition), as it may prove to be more secure. The downside would be that it might \nnecessitate installing and maintaining an IDS on every machine that needs to be protected. \nIt is important to understand that IDSs are often passive in nature. They attempt to detect \nattacks or attack preparations by passively monitoring network traffic or the activities of the \nhost machines they are installed on in real time (or a delayed mode). Once an attack is \ndetected, the IDS is typically configured to raise an alarm and enables the on-duty security \npersonnel to decide what to do next. However, some IDSs can be configured to \nautomatically react to a perceived attack, attempting to trace the source of the attack or \nshutting down external network connections. The most aggressive of these reacting IDSs \nare sometimes referred to as intrusion detection and response systems (IDRSs). \nUnfortunately, the possibility of false positives tends to limit the range of reflex reactions \nthat an IDRS might be entrusted to perform. \nIDSs can also provide information on what was attacked and how, thereby assisting \ndamage assessment and hopefully assisting in the development of a patch to prevent the \nsame thing from happening again. In addition, this forensic evidence may also prove useful \nin identifying and potentially prosecuting attackers. \nAn IDS may be the most likely way an organization learns that an intruder has successfully \nbreached their perimeter defenses, and the information captured by an IDS could be used \nagainst an attacker at a later date. It is for these reasons that attackers often target IDSs, \nattempting to disable the IDS in the same way a burglar cuts the telephone wires running \nfrom a building's security alarm. Attempts to disable, isolate, or overwhelm an IDS should \ntherefore be taken very seriously, as it may be the last thing that the IDS ever gets to \nreport. Indeed, it may pay to install a secondary service that regularly checks the health of \nthe IDS, immediately reporting if an IDS becomes unavailable, either because the IDS itself \nis disabled, or because the means of communication from the IDS has been interrupted \n(perhaps by the attacker launching a denial of service attack against the mail server used to \nforward emails from the IDS alerting the organization of an attack). \nSince most IDSs have sensitivity settings, the security testing team should check that the \ndetection level is set appropriately. If an IDS is too sensitive, frequent false alarms may \ncause the security personal to ignore all alarms. Alternatively, if the IDS sensitivity is turned \ndown too low, an attacker may be able to compromise a Web site without being noticed. To \ndetermine how sensitive an IDS is and whether anyone reacts to an IDS alert, the security \ntesting team should consider running tests similar to those outlined in Table 8.4. \n \nTable 8.4: IDS Sensitivity Checklist  \nYES  \nNO  \nDESCRIPTION  \n□  \n□  \nDoes the IDS detect a blatant attack, such as sequentially port \nscanning all 65K ports in a matter of minutes from a single IP source \naddress? Table 4.10 lists sample tools for conducting a port scan. \n□  \n□  \nDoes the IDS detect a stealthy attack, such as a partial scan of all \npossible 65K ports over a period of days (or even longer), randomly \nusing different IP source addresses? \n□  \n□  \nIf network-based IDSs are to be used, are they located on the \nnetwork such that all network traffic can be monitored? \n□  \n□  \nIf host-based IDSs are to be used, is an IDS loaded on to every \ncritical server that needs to be monitored? \n"
        },
        {
            "page_number": 186,
            "text": " \n179 \nTable 8.4: IDS Sensitivity Checklist  \nYES  \nNO  \nDESCRIPTION  \n□  \n□  \nDo the on-duty security personnel notice and respond appropriately \nto a failed (or even successful) attempt to disable or isolate the IDS? \n□  \n□  \nDo the on-duty security personnel notice and respond appropriately \nto every IDS alert, or have they become desensitized? \nAudit Trails \nUnless a Web site keeps audit trails (or system logs), an organization is less likely to be \nable to identify and successfully prosecute an attacker. Audit trails can therefore also act as \na deterrent if their existence is known or suspected (a situation that is quite possibly the \ncase for an internal assailant). Any organization that intends to prosecute an attacker \nshould check with their legal counsel to make sure that the data being captured in the \nsystem's logs is sufficient to be legally useful, and that this information is being archived \nappropriately. There is no point collecting this information if it's deleted before it can be \nused. Retracing the work of an intruder who has been undetected for quite a while may \nnecessitate conducting forensic work on log files several months old. The last thing a \ncybercrime prosecutor wants to hear is that the log files that were to be used as evidence \nhave been purged during a routine tape-recycling process. Unfortunately, enabling auditing \nwill slow down the system's performance. However, selective audits have less of a \nperformance impact and may be an acceptable compromise. \nSince more experienced attackers will often try to either purge an audit trail or manipulate it \nto hide their tracks, unauthorized attempts to access an audit trail are often an indication \nthat an intruder is at work. Because system logs are so useful-acting as deterrents and \nproviding prosecutorial evidence, assistance in detecting vulnerability, and recovery from an \nintrusion-these critical files should be protected with additional security measures to prohibit \nan intruder from tampering with them. One tactic that can be employed to reduce the \npossibility of tampering is to set the file attributes of the log files to append only, which is a \npotentially trivial defense to implement but one that an intruder may find troublesome to \ncircumvent. Other options include writing the files to a heavily fortified server or even \nburning the logs onto a write-once CD.  \nOf course, if auditing isn't turned on, or nobody ever reviews the logs, then there's nothing \nfor the attacker to worry about! This is why several companies now offer tools that will \nautomatically create or review existing system logs, reporting on suspicious activities in a \nreal or delayed timeframe. Table 8.5 provides a list of such tools. \n \nTable 8.5: Sample List of System Event Logging and Reviewing Tools  \nNAME  \nASSOCIATED WEB SITE  \nConsul/eAudit \nwww.consul.com  \nETrust Audit \nwww.ca.com  \nLANguard \nwww.gfisoftware.com  \nLogAlert \nwww.spidynamics.com  \nLT Auditor + \nwww.bluelance.com  \nnetForensics \nwww.netforensics.com  \n"
        },
        {
            "page_number": 187,
            "text": " \n180 \nTable 8.5: Sample List of System Event Logging and Reviewing Tools  \nNAME  \nASSOCIATED WEB SITE  \nPrivate 1 \nwww.opensystems.com  \nTo evaluate whether sufficient system logging is enabled and that someone takes notice of \nan attack, the security testing should consider running an unannounced test to see if \nanyone notices an obvious or a less obvious attack. Maybe no one is paying any attention \nto the logs, perhaps there are too many low-grade alarms occurring, or possibly the LAN \nadministrator has forgotten to pull the logs because he or she has too many \"more \nimportant\" things to do first. If the logs are continually ignored, then an organization may \nwant to consider whether or not creating these logs is worth the performance hit that the \nsystem is taking to create them. The organization may also need to determine whether the \nexisting log reviewers need additional assistance, perhaps in the form of training, review \nautomation, or a larger staff. \nIf the simulated attack by the testing team is detected, the next phase of the evaluation \nwould be to assess the data that could be collected by the forensics team (see the Damage \nAssessment and Forensics section later in this chapter). That data has to be sufficient to \nmeet the organization's needs, which will to a large degree depend upon how aggressive \nan organization wants to be in the prosecution of an offender. Table 8.6 summarizes these \nauditing issues in a checklist. \n \nTable 8.6: Audit Trail Checklist  \nYES  \nNO  \nDESCRIPTION  \n□  \n□  \nIs there a documented policy in place that describes which events are \nto be recorded in an audit trail or system log? \n□  \n□  \nAre all audit trails specified in the audit trail policy that has been \nimplemented? Or have some or all of the trails been disabled to \nimprove system performance? \n□  \n□  \nIf the organization desires to have the ability to prosecute intruders, \nhas the documented audit trail policy been reviewed by the \norganization's legal counsel to ensure that the information being \ncollected is sufficiently comprehensive to allow an organization to \npursue a prosecution? \n□  \n□  \nAre all audit trails protected to prevent them from being deleted or \ntampered with? \n□  \n□  \nAre all auditing trails monitored vigilantly (manually or via an \nautomated tool)? \n□  \n□  \nAre all audit trails archived for a sufficiently long enough period of \ntime? \nTripwires and Checksums \nA tripwire is a mechanism that alerts a Web site to the presence of an intruder. A common \nimplementation of a tripwire is to create checksums for directories and/or files that are \nsupposed to be static. In the event that a static file is altered, deleted, or replaced by an \nintruder, the checksum should differ and thereby signal the presence of an intruder at work. \n"
        },
        {
            "page_number": 188,
            "text": " \n181 \nThis would happen, for example, after an intruder successfully uploads a toolkit into a \nsystem directory or replaces a legitimate system file with a Trojan horse variant. Table 8.7 \nlists some sample tripwire and checksum tools. \n \nTable 8.7: Sample List of Tripwire and Checksum Tools  \nNAME  \nASSOCIATED WEB SITE  \nIntact \nwww.pedestalsoftware.com  \nIntergrit \nwww.sourceforge.net  \nMD5sum \nwww.redhat.com  \nTripwire \nwww.tripwire.com/tripwire.org  \nViperDB \nwww.resentment.org  \nWebAgain \nwww.lockstep.com  \nWhen used judiciously, checksums typically place only a small additional load on their host \nsystem. A factor that influences the load is how often the checksum algorithm runs. The \nmore frequent the check, the more system resources will be consumed by this activity. \nHowever, the longer the period between checks, the longer an intruder has for mischief \n(such as trying to disable the tripwire) before potentially being detected. Additionally, the \ntime it takes each iteration of the checksum to run is affected by the amount of directories \nand files included in the checksum, and the frequency with which these static files are \naltered (such as monthly updates, instead of hourly). \n \nCONFIGURATION MANAGEMENT CHECKSUMS \nChecksum tools have uses other than protecting a system against a malicious attacker. In \nthe event that an organization has not invested in a full-blown configuration management \nsystem, these tools can be pressed into service to act as a lightweight configuration \nmanagement tool for application source code, test scripts, and test data sets, detecting any \nunauthorized modifications to these files. \nChecksum-scanning tools can even be enhanced to create a self-healing system. In the \nevent that the checksum utility finds a system file has been modified, it automatically installs \na fresh version of the file from a secure source (such as a read-only CD drive), an \nextremely annoying obstacle for an intruder. \nAs with any audit trail, a tripwire implementation should be tested to ensure that it has been \nimplemented correctly, and that if a tripwire is triggered, somebody notices. Table 8.8 lists \nsome of the checks that can be used to evaluate the implementation of a tripwire-style \ndefense. \n \nTable 8.8: Tripwire and Checksum Checklist  \nYES  \nNO  \nDESCRIPTION  \n□  \n□  \nIs there a documented policy in place that describes which machines, \ndirectories, and files should be protected by a tripwire? \n□  \n□  \nDoes the documented policy also specify the frequency with which \n"
        },
        {
            "page_number": 189,
            "text": " \n182 \nTable 8.8: Tripwire and Checksum Checklist  \nYES  \nNO  \nDESCRIPTION  \nthe protected machines should be scanned? \n□  \n□  \nAre all the resources that are supposed to be protected by a tripwire \nactually protected? \n□  \n□  \nWhen modifying a directory that is protected by a checksum defense, \ndo \nthe \non-duty \nsecurity \npersonnel \ntake \nnotice \nand \nreact \nappropriately? \nMalware \nNot all intrusions are instigated manually. In addition to an attacker sitting at a remote \nterminal trying to compromise a Web site, a whole host of software programs can be \nutilized to attack Web sites. These programs are collectively known as malware. They have \nevolved over the years from slow-spreading programs that require a human to physically \ncarry a floppy disk from one infected machine to a new victim into virulent forms that can \nspread themselves around the world in a matter of hours, causing billions of dollars' worth \nof damage. \nDepending upon the method used to replicate them, each of these troublesome programs \ncan be grouped into one of three different subcategories of malware: viruses, Trojan \nhorses, and worms. \n \nVIRUSES, TROJAN HORSES, AND WORMS \nSymantec (www.symantec.com) defines these different flavors of malware in the following \nways: \nƒ A computer virus is a computer program written by an ill-intentioned programmer. A \ncomputer can catch a virus from disks, a local network, or the Internet. Just as a \ncold virus attaches itself to a human host, a computer virus attaches itself to a \nprogram. And just like a cold, it is contagious. \nƒ A Trojan horse, while not technically a virus, has the potential to cause the same \nkinds of problems that viruses do. Many Trojan horses are designed to steal login \nIDs and passwords and then email them to someone else who can make use of \nthe account. Other Trojan horses display obscene messages or delete the \ncontents of hard drives. A Trojan horse is typically acquired by downloading a \nprogram that seems safe or promises something like free online time. Once it is \ndownloaded and executed, the malicious code begins to work. The difference \nbetween Trojan horses and viruses is that Trojan horses do not replicate or \nspread on their own. They can only be transmitted intentionally via email, disk, or \na download directly onto a computer. \nƒ Like viruses, worms replicate themselves. However, instead of spreading from file to \nfile, they spread from computer to computer, infecting an entire network. Worms \ncopy themselves from one computer to another over a network (using email, for \nexample). Because worms don't require human interaction to replicate, they can \nspread much more rapidly than computer viruses. \n \nTo guard against this horde of malicious software that now wanders the electronic planet, \nmany commercial organizations have developed software products that attempt to detect \nthese malware programs before they have a chance to do any harm or replicate \n"
        },
        {
            "page_number": 190,
            "text": " \n183 \nthemselves. Commonly referred to as antivirus software (even through they are designed to \nsearch for more than just viruses), these software products should be used to scan all \nincoming and outgoing mobile code, file downloads, and emails (including their \nattachments). In addition, regularly scheduled full scans of every machine directly \nconnected to the Web site should also be conducted to ensure that any malware that \nescaped initial detection (perhaps with the aid of an inside accomplice) can still be caught \nbefore further damage can occur. Table 8.9 is a sample list of commercial antivirus \nsoftware products. \n \nTable 8.9: Sample List of Antivirus Products  \nNAME  \nASSOCIATED WEB SITE  \nCommand AntiVirus \nwww.commandcom.com  \neTrust Antivirus \nwww.ca.com  \nF-Secure Anti-Virus \nwww.f-secure.com  \nInterScan/PC-cillin/ServerProtect \nwww.antivirus.com  \nLockDown \nwww.lockdowncorp.com  \nNorton Antivirus \nwww.norton.com/symantec.com  \nPestPatrol \nwww.pestpatrol.com  \nSophos Anti-Virus \nwww.sophos.com  \nTauscan \nwww.agnitum.com  \nVirusScan \nwww.mcafee.com/nai.com/drsolomon.com  \nNote that www.vmyths.com maintains a list of computer virus hoaxes and myths. \nAccording to several of the antivirus tool vendors, by the summer of 2002, there were over \n60,000 malware programs in circulation. Fortunately, the majority of these programs are \nclones of existing programs or are built from standard virus-generation kits, making it easier \nfor an antivirus-signature-based product to detect it due to common code usage. However, \nas with signature-based IDSs, antivirus programs that rely on signatures to detect a \nmalware program must continuously update their signature databases in order to stay \nabreast of the most recent mutations. Unfortunately, even the most recent and extensive \nsignature file still provides less than 100 percent protection. Therefore, some antivirus tools \nalso provide functionality to monitor the machine they are installed on for suspicious or \nanomalous activities (such as writing to the boot sector of a hard drive) that do not typically \noccur under normal circumstances. Such monitoring would thereby indicate the potential \npresence of a malware program that would not be detected by the standard signature-\ndetection algorithm. \nTo prevent malicious mobile code (such as a Java applet or ActiveX control attached to a \nWeb page) from reaching its intended destination, some organizations scan all inbound \n(and optionally outbound) mobile code. The scan can take place at one or more of the \nfollowing locations: \nƒ \nAt the organization's perimeter firewall, using security policies defined in the firewall's \nnetwork traffic filtering rules. \n \nƒ TESTING ANTIVIRUS PROTECTION \n"
        },
        {
            "page_number": 191,
            "text": " \n184 \nƒ To ensure that an antivirus product has been installed correctly, a security testing \nteam may be tempted to install a real live virus onto a machine or send an infected \nattachment over a network that is theoretically protected by the antivirus software. \nƒ Unfortunately, using a real virus for testing purposes (even if it is a comparatively \nharmless one) could result in the virus inadvertently escaping the controlled test \nenvironment and making it into the real world, contributing to the problem that the \nvirus detection solution is trying to mitigate. \nƒ Fortunately, there exists an industry-standard dummy virus definition that, when \ndetected by a virus detection product, will be reported as a virus and can be used \nto test that the antivirus product has been installed correctly. This inert test file \n(known as the eicar test file/string) and advice on how to use it can be obtained \nfrom www.eicar.org. \n \nƒ \nEn route across the organization's LAN via a network-based IDS. \nƒ \nVia a special-purpose application server residing on the organization's LAN, which \nintercepts all mobile code requests and attempts to evaluate the potential for mischief \nby exercising the code in a contained environment. The server then forwards the \nmobile code only if it appears not to request any suspicious resources or performs \nany questionable activity. Table 8.10 provides a list of mobile code-scanning tools. \nƒ \n \nTable 8.10: Sample List of Mobile Code-Scanning Tools  \nNAME  \nASSOCIATED WEB SITE  \nAppletTrap \nwww.trendmicro.com/antivirus.com  \neSafe Enterprise \nwww.esafe.com  \neTrust Content Inspection \nwww.ca.com  \nSurfinGate \nwww.finjan.com  \n \nƒ \nAt the client machine requesting the mobile code. Many desktop-based antivirus \ntools now also perform security checks on Java applets and ActiveX controls. \nTable 8.11 summarizes the checks that may be performed to determine if an organization is \ntaking reasonable precautions to prevent malicious code from infiltrating a Web site. \n \nTable 8.11: Malware Prevention Checklist  \nYES  \nNO  \nDESCRIPTION  \n□  \n□  \nIs a documented policy in place that describes how antivirus products \nare to be deployed? \n□  \n□  \nAre all the resources that are supposed to be protected by antivirus \nsoftware actually protected? Or have they been partially or \ncompletely disabled to improve performance or cut down on those \nannoying alerts that keep popping up onscreen? \n□  \n□  \nIf the deployed antivirus products use a signature database to detect \nviruses, is there a process in place that ensures that the signature \ndatabase is regularly updated? \n"
        },
        {
            "page_number": 192,
            "text": " \n185 \nTable 8.11: Malware Prevention Checklist  \nYES  \nNO  \nDESCRIPTION  \n□  \n□  \nWhen an antivirus program detects what it believes to be a virus, are \nthe on-duty security personnel notified? And if so, do they react \nappropriately? \nMonitoring \nIf any kind of automated intrusion detection mechanism is to be deployed, whether it's a \ntripwire, IDS, or log analyzer, someone needs to be alerted if a suspicious event occurs. \nUnfortunately, these alerts don't always indicate the presence of an intruder. The activity \ncould be legitimate and just so happen to have a profile similar to that of a known security \nproblem. Therefore, before any drastic response is initiated, the security personnel will \ntypically evaluate the alert and decide if there really is an intruder at large. \nEvaluating suspected threats requires a great deal of expertise. Rather than set up the \nintrusion detection mechanism to send alerts to different points of contact, many \norganizations choose to send these system alerts to the same person(s) or to a central \nmonitoring system that filters and then relays all the alerts to a single point of contact. \nEnsuring that any deployed central monitoring system has been configured correctly may \nnecessitate emulating an attempted breach on every device and service that has been tied \ninto the monitoring system. The central monitoring console would then need to be checked \nto ensure that each event was reported in a timely manner. Table 8.12 lists some of the \ntools that act as central clearinghouses for security alerts from other products such as \nnetwork-based IDSs, host-based IDSs, honey pots, firewalls, and performance monitors (a \nperformance degradation is often a sign that an intruder is at work). \n \nTable 8.12: Sample List of Centralized Security-Monitoring Tools and Services  \nNAME  \n \nASSOCIATED WEB SITE  \n1stWatch \nwww.redsiren.com  \nAffinity \nwww.mis-cds.com  \nBrinks \nwww.brinksinternetsecurity.com  \ne-Sentinel \nwww.esecurityinc.com  \nEtrust \nwww.ca.com  \nExodus \nwww.exodus.net  \nGuardent \nwww.guardent.com  \nIntelliSecurity \nwww.verio.com  \nIsensor \nwww.secureworks.com  \nISS \nwww.iss.net  \nHarvester \nwww.farm9.com  \nMETA Security \nwww.metases.com  \nNetwork Security Manager \nwww.itactics.com  \n"
        },
        {
            "page_number": 193,
            "text": " \n186 \nTable 8.12: Sample List of Centralized Security-Monitoring Tools and Services  \nNAME  \n \nASSOCIATED WEB SITE  \nNeuSECURE \nwww.guarded.net  \nNSEC \nwww.nsec.net  \nOnlineGuardian \nwww.ubizen.com  \nRiptech \nwww.riptech.com  \nSAVVISecure \nwww.sawis.com  \nSecurity Manager \nwww.netiq.com  \nSentry \nwww.counterpane.com  \nSpectrum \nwww.aprisma.com  \nTruSecure \nwww.trusecure.com  \nVeriSign \nwww.verisign.com  \nVeritect \nwww.veritect.com  \nVigilEnt Security Manager \nwww.pentasafe.com  \nVigilinx \nwww.vigilinx.com  \nFor some organizations, having the single point of contact be an outsourced vendor, \ntypically located at a remote location, would make more economic sense. Once the vendor \ninstalls one or more software agents on each system to be monitored, each agent then \nmakes an initial assessment of any detected activity and reports back suspicious events to \nthe vendor's control center for further assessment and potential escalation. The agent could \nalso simply forward the raw log files to the vendor's control center where they are typically \nscanned by proprietary forensic tools. The results would then be manually reviewed to \ndetermine what, if any, action should then be pursued. \nIf a remote control center is implemented, then the response time of the control center \nshould be evaluated to ensure that the control center can detect and report any attempted \nintrusion within the period of time specified in the accompanying service level agreement \n(SLA). If the SLA states that some activities will be immediately reported and others which \nhave been determined by the control center to be of no immediate threat will be logged and \nreported on a weekly (or other frequency) basis, an organization should confirm exactly \nwhat activities will not be immediately reported and then assess whether or not this is \nacceptable. For example, every firewall that is pinged or each email that is stripped of its \nvirus-infected attachment would probably not warrant a phone call from the central control \ncenter to the local security personal. It would instead be included on the regular status \nreport, whereas the disappearance (or reduction in size) of a system log should warrant \nimmediate notification. Table 8.13 summarizes some of the tests that could be run to \ndetermine if the security defenses that have been deployed are being monitored \nadequately. \n \nTable 8.13: Intrusion-Monitoring Checklist  \nYES  \nNO  \nDESCRIPTION  \n□  \n□  \nIs there a documented policy in place that describes how each \n"
        },
        {
            "page_number": 194,
            "text": " \n187 \nTable 8.13: Intrusion-Monitoring Checklist  \nYES  \nNO  \nDESCRIPTION  \nsecurity defense will be monitored to detect attempted breaches? \n□  \n□  \nIf manual monitoring is to be used, is each and every defensive \nmeasure actually monitored with the frequency specified in the \ndocumented policy guidelines? \n□  \n□  \nIf automated monitoring is to be used, is each and every defensive \nmeasure connected to the central monitoring application? \n□  \n□  \nIf a remote control center is used, does the control center detect and \nreport on each suspected intrusion attempt within the period defined \nby the control center's SLA? \n \nIntrusion Response \nPerhaps because so many organizations believe that being hacked is something that \nhappens to other people, few of these organizations take sufficient time to plan their \nresponse to such an unpalatable event happening. For a CIO, being woken up at 3:00 A.M., \nto be told that an intruder may be at large on the organization's Web site is bad news, but \nthen being asked, \"What do we do now?\" is truly a nightmare. To avoid making an \nundesirable situation worse than it already is, an organization should have a clearly defined \nprocedure in place for handling the unenviable situation of an intruder successfully \ncompromising some part of the organization's internal network. \nThe specific activities that each organization will want to undertake in response to the \ndetection of an intruder will vary from organization to organization. For example, the laws \ngoverning what an organization can and cannot do vary significantly from country to \ncountry. If fact, within a single country, the laws may differ, depending upon whether the \nsystem being attacked is used for regular commerce or national security. Krutz et al. \n(2001), Lierley (2001), and Matsuura (2001) provide additional information on the legal \naspects of computer crimes. An organization's response policy should therefore be \nreviewed to ensure that it is complete and appropriate within the legal and resource \nconstraints that it will be utilized under. The following sections focus on the main \ncomponents that typically make up an organization's intrusion response policy. \nConfirmation of Intrusion \nDepending upon the source and severity of the initial alert, it may be desirable to obtain a \nsecond independent assessment of the situation (assuming that this assessment can be \ndone within an appropriate timeframe), thereby attempting to rule out any false alarms \nbefore excessive actions are implemented. For example, suppose a network-based IDS \nbelieves it has detected suspicious communications between two servers. The host-based \nIDSs installed on either of the servers in question could be reviewed to see if some event \noccurred that, although unusual, wasn't deemed odd enough for the host-based system to \nraise an alert. \nDamage Containment \nOnce an organization has made the assessment that an intruder is or has recently been \nactive in a restricted area, the organization then needs to decide whether to attempt to cut \noff the attack or to allow the intrusion to temporarily continue. Although a kneejerk reaction \nmight be to pull the plug on the Web site, if the method of entry has not yet been identified, \n"
        },
        {
            "page_number": 195,
            "text": " \n188 \nmonitoring the intruder's activities until his or her method of entry has been deduced may \nprove to be the most secure approach in the long term. Prematurely sealing the Web site \nmay cause the site's outage to be longer, effectively creating a more severe denial-of-\nservice (DoS) attack (even if this was not the intruder's intention). Of course, the longer that \nintruders have access to the Web site, the more opportunity they have to cause damage, \nand subsequently the more costly the cleanup. For example, if the intruder appears to be \njust moments away from downloading the swiss_number_bank_account file, then pulling \nthe plug is quite probably the right thing to do. \nHopefully, the scenarios under which an intruder would be allowed to continue to roam or \nbe cut off won't be a split-second decision made by a barely awake CIO at 3:00 A.M. or by \nthe unfortunate LAN administrator who happened to be carrying the site pager that night. \nInstead, guidelines on how to manage some of the probable intrusion scenarios should \nhave been thought out, documented, and tested ahead of time. For instance, one major \nconsideration that affects whether an intruder should be put under surveillance is whether \nor not the organization has sufficient evidence to prosecute the offender. If the organization \nhas decided that it has no intention of prosecuting an intruder (perhaps because of the \nadverse publicity it might attract), then allowing the intruder continued access to the \ncompromised system while efforts are made to trace the attack back to its source may not \nbe a particularly productive endeavor. Another consideration is whether or not the portion of \nthe Web site that has been compromised can be isolated and quarantined, which is more \nlikely if internal firewalls have been built into the site's design. If it can, the organization's \nimmediate exposure can be limited. \nDamage Assessment and Forensics \n\"So just how bad is it?\" is possibly the first question the CIO is going to ask upon hearing \nthat the organization's Web site has just been hacked. It's at this point that the site's \ndesigners start to wish that they had enabled the various logging features that had been \nturned off to improve the site's performance. \nIdeally, the damage assessment will be able to recreate the exact movements and activities \nof the intruder(s) and thereby precisely deduce what has been compromised. Questions \nthat this assessment will seek to answer include the following: \nƒ \nHave any executables been added, altered, or deleted? This could possibly have \nhappened as the intruder uploaded a rootkit, Trojan horse, backdoor, or some other \nkind of executable that could be used later to reenter the site. \nƒ \nHas any sensitive data been compromised? For instance, a credit card file could \nhave been viewed or downloaded. \nƒ \nHas the systems environment been altered in any way? For example, new user \naccounts or directories could have been added. \nThe ease with which a damage assessment can be carried out will not only depend upon \nthe amount of raw data available for examination, but also the tools available to automate \nthis research. For instance, utilizing a Web log analyzer makes reviewing Web logs much \neasier than eyeballing the raw files. (The University of Uppsala in Sweden maintains an \nextensive listing of log analysis tools at www.uu.se/software/ana-lyzers.) \nInstead of trying to identify exactly which files were tampered with, it might be more \nexpedient to simply assume that every file has been compromised and reinstall the \noperating system on a reformatted (or, better still, brand-new) hard drive on every \nsuspected machine. In the case of sensitive data files, instead of trying to figure out which \nspecific records may have been viewed or downloaded by an intruder, assume that all the \nrecords have been compromised and notify all the possibly affected data owners \naccordingly. \n"
        },
        {
            "page_number": 196,
            "text": " \n189 \nThe primary purpose of doing damage assessment is to assist an organization in \nrecovering from the intrusion and restoring normal service as quickly as possible. The \nclosely related field of forensics is concerned with collecting evidence that can \nsubsequently be used in legal proceedings. (Typically, this endeavor requires a much \nhigher standard of data collection than would necessarily be needed by an organization for \nits own internal purposes.) Kruse et al. (2001), Marcella et al. (2002), and Vacca et al. \n(2002) provide additional information on computer forensics. Unfortunately, an expedient \ndamage assessment may negatively impact forensic data collection. For example, a LAN \nadministrator quickly reviewing a system directory to see if any files have recently been \nchanged may inadvertently alter the last accessed information for the directory, potentially \ndestroying evidence that the forensic team would like to have had the opportunity to collect. \n \nDAMAGE ASSESSMENT FOLKLORE \nA classic story is of a Webmaster stating that he would rather his Web site be completely \npurged by an attacker than have an attacker deface a single Web page and the Webmaster \nnot find out about it for several hours. Why? Because the Webmaster could completely \nrestore the purged site from backups, but the damage done to the company's credibility by \nthe defaced page could take months or even years to restore. \nBecause the task of collecting evidence in a manner that permits it to be used in a court of \nlaw may delay or hinder the completion of an internal damage assessment, an organization \nshould decide ahead of time whether formal forensics collection will be done. If so, the \norganization should ensure that all employees likely to be involved in a damage \nassessment are made aware of the precautions that they will need to follow in order to \navoid contaminating forensic evidence. Since many manual data collection efforts, or the \ntools not specifically intended for the task of collecting forensic evidence, may inadvertently \ncorrupt the very evidence being collected (breaking the chain of evidence), the forensics \nteam may wish to use one or more tools especially designed for this task. Table 8.14 lists \nsome tools designed to assist with forensic data collection. \n \nTable 8.14: Sample List of Forensic Data Collection Tools  \nNAME  \nASSOCIATED WEB SITE  \nDIBS Analyzer \nwww.dibsusa.com  \nEnCase \nwww.encase.com  \nForensic Toolkit/NTLast \nwww.foundstone.com  \nFRED \nwww.digitalintel.com  \nFTK \nwww.accessdata.com  \nIlook \nwww.ilook-forensics.org  \nIntelligent Computer Solutions \nwww.ics-iq.com  \nVarious \nwww.dmares.com  \nVarious \nwww.forensics-intl.com  \nVarious \nwww.tucofs.com  \nIt is perhaps the damage assessment and forensics phase of an intrusion response that \ncould benefit the most from a simulated (or mock) intrusion. The intrusion may show just \n"
        },
        {
            "page_number": 197,
            "text": " \n190 \nhow hard it is to do dependable forensic research, especially when key system logs have \nbeen disabled, tampered with, or were never even implemented. Indeed, one of the action \npoints that may come out of running the simulated intrusion is that the organization decides \nto evaluate and select a firm that specializes in computer forensics. The firm may even be \nplaced on a retainer, so that if a real intrusion occurs, a team of preapproved experts can \nimmediately be summoned to assist the local staff in conducting a damage assessment and \nforensic collection. Table 8.15 lists some firms that provide forensic consulting services. \n \nTable 8.15: Sample List of Forensic Consulting Firms  \nNAME  \nASSOCIATED WEB SITE  \nASR Data \nwww.asrdata.com  \nComputer Forensics \nwww.forensics.com  \nCrypTEC Forensic \nwww.cryptec-forensic.com  \nDIBS USA \nwww.dibsusa.com  \nFred Cohen and Associates \nwww.all.net  \nLee and Allen \nwww.lee-and-allen.com  \nNew Technologies \nwww.forensics-intl.com  \nScience.org \nwww.forensics.org  \nNote that the inclusion (or exclusion) of a specific firm in this list should not be \nconstrued as any form of recommendation by this book. \nDamage Control and Recovery \nThe damage control and recovery phase is concerned with minimizing the impact of the \nintrusion on the organization. For example, based on the findings of the initial damage \nassessment, an organization may not be comfortable with allowing the compromised \ncomponents of a system to stay online, possibly even deciding to lock down the entire \nsystem. In the event that this outage is on a mission-critical system, the organization may \nnot be able to postpone using this functionality until after this valuable resource has been \nthoroughly examined by forensic experts, cleaned up, and restored to a state that would \nprevent the same intrusion from occurring again. This would necessitate some sort of \ninterterm recovery measure to control the amount of damage done to the business of the \norganization. \nRather than waiting for a system to be taken offline due to an intrusion before considering \nhow to implement some temporary solution, an organization should develop damage \ncontrol and recovery plans for their most critical system ahead of time. Maiwald et al. (2002) \nand Toigo et al. (1999) provide additional information on disaster-recovery planning. \nExample recovery strategies include the following: \nƒ \nA Web site is distributed across multiple geographic locations and the damage \nassessment has concluded that only one site has been compromised. No evidence \nsuggests that the remaining sites are susceptible to the same kind of attack that \naffected the infected location. (For example, it appears that the breach was a result of \na local system administrator's password being mistakenly reset by the help desk, and \nthis account does not have any privileges at the other geographic sites.) One possible \nrecovery strategy would be to take the affected site offline. Temporarily disabling \nsome of the resource-intensive (but less critical) components at the remaining \n"
        },
        {
            "page_number": 198,
            "text": " \n191 \nuncontaminated locations would enable the Web site to return to an acceptable, albeit \nless than optimal level of performance until the compromised location can be safely \nbrought back online. \nƒ \nIf the organization has invested in a backup site for protection against natural \ndisasters, it may be possible to press this site into service to help the organization \nalso recover from an unnatural disaster, such as an intruder compromising the \nprimary site. \nƒ \nA Web site purely being used to process a small number of sales orders has been \ncompromised. It may be feasible to replace the site's normal home page with a simple \nWeb page, hosted on an uncompromised machine, explaining that the Web site is \nexperiencing technical difficulties and anyone wishing to place an order should call \nthe following toll-free telephone number. \nWhatever damage control and recovery strategies an organization decides to adopt, once \ndocumented in the form of a plan, these strategies should be tested to make sure that these \npaper plans can actually be implemented within the desired timeframe. Unfortunately, all \ntoo often, what sounds like a perfectly easy procedure to follow on paper turns out to take \nsignificantly longer to implement than originally envisaged. Merely trying to document this \nprocess diagrammatically may indicate just how complex the process is and/or may identify \nholes in the perceived process. \nBy performing a test run of these recovery procedures, it should be possible to identify \nwhich steps need improvement, thereby placing an organization in better shape should the \nunthinkable happen and it actually have to utilize these plans for real. \nSystem Salvage and Restoration \nOnce the damage has been controlled and a temporal fix put into place (or, if sufficient staff \nare available, to initiate in parallel with the damage-control effort), the organization can turn \nits attention to restoring the system back to full strength. The first step is for the \norganization to decide which parts of the system should be salvaged and which parts \nshould be written off as a total loss. \nHardware and Software \nWith the exception of physical sabotage, it's rare for a piece of hardware to be so badly \ndamaged from an intrusion that it can't be salvaged and returned to service. (An exception \nto this rule would be a microprocessor that is the victim of a DoS attack designed to \noverheat the CPU by performing excessive numerical calculations.) However, it may be \nmore expedient to swap out compromised hardware devices such as hard drives rather \nthan trying to recycle them, especially if the original hard drives are going to be needed by \nforensics. \nIn the case of system and application software, rather than trying to figure out exactly which \nservices may have been compromised, it's probably safer to assume the worst and erase \nand reformat any hard drives (and possibly backup tapes and disks) that may have been \naccessed by the intruder. This would entail reinstalling the system and application software \nfrom a trusted medium such as a nonrewriteable CD. Unfortunately, all too often weak \nconfiguration management procedures may have allowed developers or Webmasters to \nmake changes to the production Web site without updating the original source code or test \nenvironments, thereby resulting in recent fixes and optimizations being lost as the result of \nthe restoration process. Restoration procedures should therefore be checked to ensure that \nthey are comprehensive and can be implemented in a timely manner. \n"
        },
        {
            "page_number": 199,
            "text": " \n192 \nData \nAlthough restoring the system and application software that provides a Web site's \nfunctionality may be tedious, a much more challenging situation exists when a Web site's \nnonstatic data needs to be restored. Assuming the damage assessment is able to \ndetermine at what point in time an intruder gained unauthorized access to the data, it may \nbe possible to restore the data to a precompromised state. Unfortunately, rolling the data \nforward from the last backup made before the intrusion may not be possible if transaction \nlogs have not been kept (or cannot be trusted), leaving the organization with an unenviable \ndilemma of implementing a less-than-ideal recovery process, such as one of the following \nstrategies: \nƒ \nRestore the data back to its last known uncompromised state and then attempt to \nreenact all the transactions that are believed to have occurred after this point in time. \nFor example, the first step would be to restore last week's version of a database used \nto support an intranet application that processes all the invoices a company receives. \nThe database could then be brought up-to-date by rekeying all the paper invoices and \nresubmitting any electronic invoices that the company had received within the last \nweek. \nƒ \nRestore the database back to the last known uncompromised state and then void \nany transaction that occurred after that point in time. Such a course of action may be \nchosen because noncorrupted versions of the voided transaction cannot be found (or \ncan't be trusted) or the business cost of reapplying these transactions outweighs the \nbenefit of a full restoration. For example, the administrator for a fantasy football \nleague Web site that did not maintain any transaction logs may decide to simply void \nall the games that took place after the Web site was defaced. He would do so rather \nthan rely upon the recollections (and honesty) of the individual players to resubmit \ntheir game selections. Play would then be resumed for all future games. \nƒ \nAttempt to patch the compromised data files or databases by trying to identify and fix \nthe individual records that have been tampered with, typically using information from \nan independent trusted source. For example, suppose it appears that the only data \nthat was tampered with was a few employees' salary records. It may be possible to fix \nthese records by reconciling them with records held in a separate uncompromised \nsystem, such as the human resource department's paper files, or the payroll run that \ntook place immediately before the intrusion occurred. \nRecovering corrupted data becomes increasing more difficult the longer the corruption goes \nunnoticed. So, an organization wanting to test the capabilities of its data restoration \nprocedures might wish to challenge the team responsible for recovery efforts. It could see \nhow long the recovery team takes to salvage a tampered database that has subsequently \nhad several days' (or weeks') worth of legitimate, nonvoidable transactions applied to it \nsince the corruption occurred. \nOnce the primary system has been salvaged and restored, one additional step may need to \ntake place before it can reenter service: synchronizing, or reapplying the transactions that \nhave occurred on the temporal system while the primary system was offline. This task \nshould be feasible if the temporal system has been maintaining adequate transaction logs; \nnevertheless, the synchronizing process should be tested to ensure that it can be \ncompleted in a timely manner. \nNotification \nOne more item that should be specified in an organization's intrusion response policy is the \npeople who should be notified of a successful (or unsuccessful) intrusion. The policy should \n"
        },
        {
            "page_number": 200,
            "text": " \n193 \nalso spell out when these parties should be informed. Possible candidates would include \nthe following: \nƒ \nAn organization's legal counsel. \nƒ \nLaw enforcement agencies such as the FBI (www.fbi.gov). \nƒ \nCoordination centers such as the Computer Emergency Response Team \ncoordination center (www.cert.org) or the System Administration, Networking, and \nSecurity (SANS) institute (www.sans.org). \nƒ \nThe unwitting owners of the machines from which the attack was launched. \nƒ \nBusiness partners who might be affected, such as credit card issuers, banks, \ncustomers, and suppliers. \nƒ \nThe media and general public. If the news of the attack is likely to make it into the \npublic domain, an organization's public relations department may want to be the \nsource that announces the incident, thereby being given the opportunity to put a more \npalatable spin on the event. \nƒ \nIf the organization has a cybercrime insurance policy, then the organization may \nneed to put the insurance carrier on notice that the organization may be filing a claim. \nTable 9.2 provides a sample list of insurance carriers that offer cybercrime insurance \npolicies. \nRetaliation and Prosecution \nOne critical decision that should have been made long before any serious attack is detected \nis whether or not an organization, given the opportunity, would consider trying to prosecute \nan offender. This decision is critical, as the collection and preservation of evidence for use \nin a possible prosecution will most likely slow down the recovery effort and potentially \nincrease the recovery cost. Possibly too, an organization may decide to adopt a different \npolicy for internal attacks than for an externally launched attack. \nSome organizations may choose to attempt to acquire information on the attacker by trying \nto trace the attack to the original source. Unfortunately, many sophisticated attackers \nchoose not to launch an attack from their own machine but instead employ a previously \ncompromised (or drone) machine to do their bidding. So, any counterattack launched by the \norganization that suffered from the original attack may only end up affecting another of the \nintruder's victims. Although attempting to trace the source of the attack may yield useful \ninformation (if only to inform another organization that their system has been \ncommandeered by an intruder), more aggressive action is probably best left to local, nation, \nor international law enforcement. \nIn situations when successfully compromising the Web site under attack would be \nconsidered a breach of national security, the attack may be considered an act of cyberwar. \nIt could therefore warrant much more aggressive action by the nation's government than \nactions likely to be taken by national or international law enforcement agencies. \nPolicy Review \nEven when no successful intrusion attempt has been detected, the rapidly changing nature \nof Web security and the relative inexperience of some of those charged with protecting an \norganization's Web sites and applications can still make holding a regular policy review \nextremely beneficial. In addition, it is considered good practice to always review the current \nsecurity policies and infrastructure after a successful break-in has occurred to make sure \nthat a similar attack would not be successful in the future. Table 8.16 lists the questions to \nask when creating an intrusion response policy. \n \n"
        },
        {
            "page_number": 201,
            "text": " \n194 \nTable 8.16: Intrusion Response Checklist  \nYES  \nNO  \nDESCRIPTION  \n□  \n□  \nIs there a documented policy in place that describes how the \norganization should respond to a concerted external attack that is in \nprogress but appears not to have been successful so far? \n□  \n□  \nIs there a documented policy in place that describes how the \norganization should respond to an external intruder who has (at least) \nbreached the Web site's perimeter defense? \n□  \n□  \nIs there a documented policy in place that describes how the \norganization should respond to an attack that appears to have been \nlaunched from within the organization? \n□  \n□  \nIf a policy does exist, does it adequately address all the following \nconsiderations: confirmation of intrusion, damage containment, \ndamage assessment, forensic collection, damage control and \nrecovery, system salvaging and restoration, notification, retaliation \nand prosecution, and policy review? \n□  \n□  \nUnder simulated intruder attacks, does the on-duty security staff \nfollow the procedures documented in the policy? And if followed \ncorrectly, are they completed within the anticipated timeframe? \n \nSummary \nTesting the topics covered in this chapter has primarily two benefits. First, by testing the \nsuccess of confusion tactics, the accuracy of detection devices, and the reliability of \nresponse procedures, it becomes possible to identify deficiencies and thereby improve the \nsystem's defenses. The second benefit is the experience the security staff gains by running \nthrough simulated attacks, training that should hopefully allow them to execute these \nprocedures more efficiently when placed under the emotional stress of a real attack. \n"
        },
        {
            "page_number": 202,
            "text": " \n195 \nPart IV: Test Implementation \nChapter List \nChapter 9: Assessment and Penetration Options  \nChapter 10: Risk Analysis  \n"
        },
        {
            "page_number": 203,
            "text": " \n196 \nChapter 9: Assessment and Penetration Options \nOverview \nMany different terms are used within the security industry to describe the tests associated \nwith legitimate security testing. However, fundamentally these activities can be grouped into \ntwo categories of testing: \nƒ \nSecurity assessment or security audit. The tester will endeavor to verify that no \nknown security vulnerability is present on the target system. This type of testing is \ntypically implemented by exercising the target system against an extensive \nproprietary checklist, in effect conducting a form of software inspection. This is a \nstructured approach that lends itself to automation due to its high degree of \npredictability and need for repeatability. Myers (1979) and Perry (2001) provide \nadditional information on the general concept of software inspections, while Allen \n(2001) and Skoudis (2001) describe security specific software inspections. \nƒ \nPenetration testing or ethical hacking. In an attempt to recreate the trickery and \ncreativity that a real-live attacker would seek to employ, the tester uses creative \ntechniques, which are modified and honed as the tester learns more about the system \nbeing interrogated. In some respects, this resembles an exploratory testing approach. \nKaner et al. (2001) provide additional information on the exploratory testing approach, \nwhile Klevinsky et al. (2002) describes the approach used in penetration testing. \nResources permitting, a security-testing project will include tests from both of these \ncategories, with exploratory penetration tests complementing the structured testing of \nknown security vulnerabilities. This chapter looks at the different ways these categories of \nsecurity testing could be conducted, specifically focusing on who will conduct the tests and \nwhat tools they will use to accomplish these tasks. \n \nStaffing Options \nPerhaps one of the toughest decisions an information services (IS) director has to make is \nto decide who will actually carry out a security assessment or penetration test of the \norganization's Web site and associated applications. Fundamentally, the choice boils down \nto adopting a do-it-yourself (DIY) approach, outsourcing the work to a firm that specializes \nin this area, or some hybrid of the two. Obviously, the organization's size and potential \nexposure will be a significant input into this decision, a multinational telecommunications \ncompany, for example, having a far greater need for on-staff security testing personnel than \nthat of a small-town newspaper. However, for many organizations, the choice is less clear, \nso the following sections examine the pros and cons to each of these approaches. Black \n(2002), Craig et al. (2002), and Kaner et al. (2001) provide additional guidance on the \nchallenges of managing a team of testers (internal or external). \nDo It Yourself (DIY) \nIf new staff are to be hired, a major consideration will be whether or not qualified individuals \nwill be attracted to the compensation package that the organization can afford to offer. \nPerhaps this would not be a problem for a cash-rich bank located in New York, but not \nnecessarily as easy for a financially strapped rural hospital. Conversely, if instead of hiring \nexperienced staff, the organization makes a commitment to train some of its existing \nemployees, this process could (depending upon the employees' starting point) take months \nor even years before the employees reach the required level of proficiency. In the \nmeantime, the organization is left potentially exposed. \n"
        },
        {
            "page_number": 204,
            "text": " \n197 \nA DIY approach may well be the best path to follow if the IS director is able to obtain a \nbudget to allow the organization to hire highly skilled security personal and/or retrain the \nexisting staff in the knowledge needed to conduct rigorous security testing. The long-term \nbenefit of having competent security professionals on staff may outweigh any short-term \ncosts associated with bringing the organization up to speed. Bragg (2002), Endorf (2001), \nHarris (2001), Krutz et al. (2001, 2002), and Peltier (2002) provide additional information on \nthe knowledge domain that a security tester needs to comprehend in order to pass the \ncertified information systems security professional (CISSP) exam (www.cissp.com and \nwww.isc2.org), the security industry's widely recognized security certification (not tied to a \nspecific product). \nOne of the major variables that will affect the skill level and time needed to run a security \nassessment is the degree to which the testing will be automated (as the Tools section in \nthis chapter describes in more detail). The increasing sophistication of these tools makes it \neasier for less knowledgeable individuals to actually perform an assessment. Also, the \nautomation of many of the mundane tasks speeds up the work of even the most skilled \nconsultant. Therefore, when trying to derive the total cost of a security implementation, the \ndegree of test automation must be taken into account, as the extra cost of purchasing a \ntool, and the time needed to get up to speed with it, may in fact be less than the additional \ncost associated with a longer manual testing period, facilitated by more knowledgeable (and \nhence expensive) testers. \nOutsourcing \nDue to the complexity and depth of knowledge needed to thoroughly test the security of a \nWeb site and its associated applications, many organizations are now outsourcing their \nsecurity testing needs to firms that specialize in this service. Advantages to this approach \ninclude the following: \nƒ \nDue to a faster implementation time, testing can potentially be scheduled and \nexecuted within hours. \nƒ \nNo need exists to purchase expensive security-testing tools, and no maintenance \nfees are needed to keep the tools up-to-date. Typically, the outsourcing firm's charges \ninclude any fees they may have to pay to the vendor whose tools they are using. \nAlso, by not purchasing a tool, the organization removes the risk that it might buy an \ninappropriate or hard to use tool, which results in the tool becoming shelfware. \nFinally, from an accounting perspective, the entire fee can often be absorbed in the \ncurrent tax year rather than having to depreciate an expensive tool over a number of \nyears. \nƒ \nBecause these firms specialize in this activity, they can be expected to be up-to-date \nwith the latest versions of tools and are aware of the most recent penetration \nstrategies that are being employed by attackers. \nƒ \nNo need exists to hire expensive (and potentially hard to find) full-time security \nexperts (although such staff may still be needed to fix any problems that the \noutsourced testing discovers). For smaller organizations that cannot justify full-time \ndedicated security personnel, this may be a particularly compelling consideration. \nƒ \nThe need to train internal staff to use the various security assessment tools (or their \nmanual equivalents) is reduced. \nƒ \nThe organization may be able to explicitly establish that it has shown due diligence \ntowards securing the assets it is responsible for, a standard that may be harder to \nprove if the testing is done internally. For many organizations, demonstrating due \ndiligence is particularly important, as it may potentially reduce the liability an \norganization faces in the event of a successful intrusion occurring (assuming the \norganization showed due diligence in the selection process used to hire the firm that \nthe security testing was outsourced to). \n"
        },
        {
            "page_number": 205,
            "text": " \n198 \nLeaving aside the issue of evaluating and selecting a competent testing firm for a moment, \nthe main disadvantage with using an outsourced firm is the fee that the firm charges. A one-\ntime fee may be more cost effective than building up the necessary skills in-house, but the \ncosts associated with retesting several Web sites every few months, or having an outside \nfirm continuously monitor a Web site, will soon add up. For some organizations, this may \nultimately be a more expensive approach than conducting the same level of testing in-\nhouse. \nDeciding upon a particular security-testing firm (or firms) is unlikely to be as easy as it might \nfirst appear. As can be seen from Table 9.1, which lists just a few of the firms offering this \ntype of service, quite a few firms exist from which to choose. \n \nTable 9.1: Sample List of Firms Offering Security Testing Services  \nNAME OF FIRM  \nASSOCIATED WEB SITE  \nAlphaNet Solutions \nwww.alphanetsolutions.com  \n@Stake \nwww.atstake.com  \nBaltimore \nwww.baltimore.com  \nBoran Consulting \nwww.boran.com  \nCigital \nwww.cigital.com  \nConQWest \nwww.conqwest.com  \nControl Risks Group \nwww.crg.com  \nCryptek \nwww.cryptek.com  \nDefcom \nwww.defcom.com  \nEmprise Technologies \nwww.emprisetech.com  \neSMART \nwww.esmartcorp.com  \nExodus \nwww.exodus.net  \nErnst & Young \nwww.ey.com  \nFarm9 \nwww.farm9.com  \nFoundstone \nwww.foundstone.com  \nGrayhat Security \nwww.grayhatsecurity.com  \nGRC International \nwww.grci.com  \nGuardent \nwww.guardent.com  \nHyperon Consulting \nwww.hyperon.com  \nIBM \nwww.ibm.com  \nInfidel \nwww.infidel.net  \nInfoScreen \nwww.infoscreen.com  \nInternet Security Systems \nwww.iss.net  \niXsecurity \nwww.ixsecurity.com  \n"
        },
        {
            "page_number": 206,
            "text": " \n199 \nTable 9.1: Sample List of Firms Offering Security Testing Services  \nNAME OF FIRM  \nASSOCIATED WEB SITE  \nMaven Security Consulting \nwww.mavensecurity.com  \nMETA Security Group \nwww.metasecuritygroup.com  \nMIS Corporate Defence Solutions \nwww.mis-cds.com  \nNetwork Security \nwww.nsec.net  \nPredictive Systems \nwww.predictive.com  \nProCheckUp \nwww.procheckup.com  \nRedSiren Technologies \nwww.redsiren.com  \nRiptech \nwww.riptech.com  \nSecureInfo \nwww.secureinfo.com  \nSecurity Automation \nwww.securityautomation.com  \nSword & Shield Enterprise Security \nwww.sses.net  \nSysinct \nwww.sysinct.com  \nSystem Experts \nwww.systemexperts.com  \nTesCom \nwww.tescom-usa.com  \nTestPros \nwww.testpros.com  \nTiger Testing \nwww.tigertesting.com  \nTritonic \nwww.tritonic.com  \nTruSecure \nwww.trusecure.com  \nTrustAsia \nwww.trustasia.com  \nVeriSign \nwww.verisign.com  \nVeritect \nwww.veritect.com  \nVigilinx \nwww.vigilinx.com  \nNote: The inclusion (or exclusion) of a specific firm in this list should not be \nconstrued as any form of recommendation by this book. \nAside from the usual considerations associated with outsourcing work to another party \n(such as availability, consulting rates, and scheduling), security testing has some additional \nconsiderations that do not necessarily apply to the same degree to other outsourced \nactivities. The following sections will expand upon these considerations. \nTrustworthiness of Firms Doing Outsourced Work \nBy hiring another firm to attempt to break into one or more of its systems, an organization is \nimplicitly trusting that the individuals doing the testing will not take advantage of this \nopportunity to do something untoward. For example, a tester could fail to report one or \nmore detected security vulnerabilities that could be exploited at a later point in time. \nAnother example would be uploading software (such as a rootkit) on to a client's machine \nthat could be used later to enable the tester to anonymously take control of the machine for \n"
        },
        {
            "page_number": 207,
            "text": " \n200 \nmalicious purposes, such as a load generator for a distributed denial-of-service attack. After \nall, if the rootkit were to be detected, the tester could simply claim to have forgotten to \nremove it after the testing was complete. \nMany testing firms that provide functional and performance-testing services also offer \nsecurity testing, but due to the specialist nature of this form of testing they out-source any \nsuch work to a subcontractor with expertise in this area, a scenario that the end-client may \nor may not be aware of. Given the risk an organization assumes when it requests another \nparty to attempt to penetrate or assess its defenses, care should be taken to ensure that \nany firm hired to undertake this task is reputable and will only delegate this work out to \nsubcontractors that the client is comfortable with. Any security-testing contract should \ntherefore state whether or not a subcontractor may be used, and if so, who they are or the \ncriteria for selecting one (such as requiring x number of satisfactory third-party references). \nPerhaps the most controversial debate surrounding the use of outsourcing security testing \nis the use or nonuse of reformed criminals. The basic premise is that only someone who \nhas had real experience trying to illegally break into numerous systems would be able to \nthink like a real intruder and thereby create a realistic probe of the client's defenses. \nSeveral problems exist with this reasoning, for instance, just because someone alleges that \nthey were able to break into someone else's system does not necessarily mean that he or \nshe is an expert. Many Web sites out there have negligible defenses, and trying to use a \ncriminal conviction as a form of certification only proves that they were not devious enough \nto cover their own tracks. Another problem is how do you know they are really reformed? \nEven a positive reference from a probation or parole office can't guarantee that while testing \na Web site the reformed criminal won't do a little extracurricular reconnaissance work. \nSome firms actively advertise the fact that some of their staff are convicted criminals, while \nothers make a specific point of telling their perspective clients that they run extensive \nbackground checks on all their employees in order to avoid such individuals. It appears that \nthe marketplace has yet to decide which solution is the most desirable. It is therefore a \ngood idea to explicitly ask a potential security-testing firm what viewpoint they prescribe to, \nand make sure that it does not run counter to the organization's own view on this subject. \nUse of Insurance Underwriters \nMany insurance companies offer policies designed to help compensate an organization that \nhas been penetrated by an intruder, as part of a general business continuity policy, or as \npart of a special purpose cybercrime policy. Table 9.2 lists some examples of carriers that \noffer this type of insurance. \n \nTable 9.2: Sample List of Insurance Carriers Offering Cybercrime Policies  \nNAME  \nASSOCIATED WEB SITE  \nINSUREtrust.com \nwww.insuretrust.com  \nLloyds of London \nwww.lloydsoflondon.co.uk  \nSafeonline \nwww.safeonline.com  \nThese insurance carriers will often require any organization applying for such a policy to \nfirst be audited by an approved security assessment firm. This is done in order to obtain a \nsecurity certification that may be used as proof of due diligence and subsequently qualify \nthe organization for the policy or a reduced insurance premium. Although passing such an \nassessment will not guarantee that the organization will not be successfully attacked, in the \n"
        },
        {
            "page_number": 208,
            "text": " \n201 \neyes of the insurance company's underwriter at least some level of minimum protection has \nbeen implemented. The possibility of an intrusion is therefore reduced to the point that the \nunderwriter finds it acceptable to transfer the risk of such an event from the applicant to the \ninsurance company. \nIf an organization intends to purchase insurance coverage to mitigate the ramifications of \ntheir Web site's security being compromised, then the number of firms from which they can \nchoose to perform a security assessment may be limited to those firms that have been \npreapproved by the potential insurance carrier. \nExplication of Terms of Engagement \nBefore an organization sanctions a penetration test against any of its production systems, it \nshould ensure that the contract agreed upon with the security-testing firm explicitly states \nthe terms of engagement (or terms of reference) under which the testing firm can interact \nwith the target of the testing effort. Examples of such terms include questions like can social \nengineering be exploited, or can the testers attempt to overload an intrusion detection \nsystem by launching a denial-of-service attack (which may have the unpleasant side effect \nof denying legitimate users access to the target Web site)? \nOffers of Compensation \nDoes the security-testing firm offer any kind of compensation (possibly specified in the form \nof a service level agreement [SLA]) should they inadvertently bring down or severely impact \nthe system they are testing? What if an intruder compromises a Web site that has already \nbeen tested, using an exploit that the security-testing firm should have been expected to \nfind, but the testing firm either failed to check for such an opening or wrongly interpreted \ntheir own test results? \nComprehensiveness of Coverage \nMany companies offering security-testing services do not in fact offer a complete range of \nservices. For example, some firms (particularly those that have performance testing as one \nof their many offerings) may offer to recreate one or more forms of denial-of-service attacks \nby generating huge volumes of network traffic (from the comfort of their own facilities), but \ndo not know the first thing about testing an application for potential buffer overflows. \nSimilarly, some firms are willing to perform an external port scan of a Web site remotely, but \nare unwilling to fly out to a client's site to conduct an internal port scan from inside of the \nclient's perimeter firewall. \nAnother way to differentiate testing firms is whether or not they include recommendations \non how to fix all the issues they detected or merely document the issues. Some may even \ngo a step further and offer a consulting service that can be hired to actually implement \nthese recommendations. \nAs with most things in life, you often get what you pay for. The cheapest firm may simply be \nthe cheapest because it doesn't do much. This is perhaps an acceptable situation if the \nclient doesn't have much of a budget, but it is an extremely dangerous situation if the client \nwrongly believes they are getting a more comprehensive checkup. Before signing on the \ndotted line, an organization should make sure that the contract documents exactly what \ntypes of tests will be covered (and not covered) and that they are comfortable with this level \nof test coverage. Beizer (1995) provides additional information on the concept of test \ncoverage. \n"
        },
        {
            "page_number": 209,
            "text": " \n202 \nWhich Tools Are Selected \nBecause some tool vendors charge security-testing firms a per-use fee, many firms opt to \ncreate their own tools or use freeware tools. A probing question to consider asking a \ncandidate security-testing firm is what tools do they use. Although using an expensive \ncommercial tool will not necessarily detect more holes than a bunch of freeware tools in the \nhands of an expert, a leading commercial tool can provide the client with a higher level of \ncomfort that the testing will provide a minimum level of comprehensiveness. \nAlthough far from a perfect measure, expensive commercial tools may also indirectly gauge \nthe financial strength of the testing company. A fly-by-night or start-up testing firm is not as \nlikely to have invested in top-notch, expensive security assessment tools as a well-funded \nfirm that has been around for a while. Conversely, using tools developed by the firm's own \nstaff may be indicative of highly skilled employees who don't necessarily need the help of \nexpensive tools. \nStarting Point of Testing \nOne consideration that will have an immediate effect on the effort needed (and hence cost) \nof running a penetration test is whether or not the security-testing team should be given a \ncomprehensive set of network addresses to be tested. (Another consideration is specifying \ntelephone numbers if remote access is included in the scope.) Providing these \npseudoattackers with such a head start will save them a considerable amount of time (and \nhence expense) that they would otherwise have had to spend discovering this information \nthemselves (a technique commonly referred to as enumerating the target). Of course, the \nonus is now on the client to make sure that this list is indeed complete and not missing \nsome overlooked entry point, such as a recently configured test system for another testing \ngroup to use. \nProviding a penetration-testing team with additional information such as the exact version of \nsystem software used by the target system will also speed up the testing, but it may provide \nthe testers with an unrealistic advantage. The same goes for providing highly sensitive \ninformation, such as the network-filtering rules that were supposed to have been used to \nconfigure a perimeter firewall, or the names and userIDs of system administrators. \nTherefore, a trade-off must be made between helping the testing team conduct their test \nmore quickly and making the testing environment more realistic by keeping them in the \ndark. \nIf the objective of the penetration effort is to find as many vulnerabilities as possible, \nproviding the testing team with as much information as possible will help decloak any \nsecurity by obscurity defense, possibly showing areas that were solely protected by this \nunreliable form of defense. On the other hand, if the purpose of the penetration test is to \nevaluate the effectiveness of the security defenses as a whole (including any confusion or \nobscurity defenses), then withholding this information will make the results more realistic. \nFor some organizations, it may even make sense to perform the penetration test twice. The \ntesting team conducting the first test would be provided the bare minimum of information \nneeded to make sure the testing is applied to the right systems, while the second team \n(which may in fact be the same team that did the first test) is provided with as much \ninformation as the organization is comfortable providing. \nEnding Point of Testing \nOne consideration that should be agreed upon before testing starts is under what \nconditions should the testing be considered completed or temporally suspended. This is an \neasy situation to define in the case of a straightforward scan by a commercial security \n"
        },
        {
            "page_number": 210,
            "text": " \n203 \nassessment tool, but it is less obvious if a manual penetration test is being contemplated. In \nthe case of a penetration team attempting to compromise a Web site by using creative \nmeans instead of merely running through a predefined list of checks, it may be more \nappropriate to time-box the testing effort to a maximum of X elapsed days for a team of Y, \nor alternatively Z person-hours of testing effort. The amount of time allocated would depend \non the size of the task and the business risk that the organization faces should the Web site \nbe compromised. For instance, the time needed to exhaustively test a single-server, \nbrochureware Web site would be much less than that needed to sufficiently probe a large e-\ncommerce Web site. As a rule of thumb, the effort expended on a penetration attempt \nshould not be less than the amount of time a real intruder (or team of intruders) would be \nreasonably expected to expend trying to compromise the target Web site. \nIn the event that the Web site is easily cracked, it may be more prudent to suspend further \ntesting until these obvious flaws have been fixed. To facilitate detecting this condition, some \norganizations will strategically plant trophies for the penetration-testing team to try and \nacquire (a variation on the capture the flag theme) and may even award bonuses for the \nsuccessful capture of such trophies before the allotted testing time has expired. \nReconfirmation of Test Results \nThe longer the period of time taken to complete a security-testing effort, the higher the \nprobability that a change to the testing environment will occur, resulting in the earlier test \nresults being called into question. For example, such a change could be a new, must-\nimplement hotfix security patch that has been released for the operating system used by \nthe majority of the system's servers. \n \nTEST EFFECTIVENESS \nBy deliberately adding known defects to the target of a penetration effort, some \norganizations hope to measure the effectiveness of the penetration-testing effort, in effect \nimplementing a variation of a defect-seeding testing technique. Since defect seeding has \nbecome largely discredited in the broader testing community, using such a technique to \nform the basis for evaluating the effectiveness of the testing effort (or the effectiveness of a \nsystem's defenses) is likely to lead to an inaccurate assessment. A significant problem with \nthis approach is the inability to plant vulnerabilities in exactly the same frequency and \nmanner that real vulnerabilities will occur, not to mention the possibility of a planted \nvulnerability being lost or forgotten, and subsequently proving a real intruder with an \nadditional opportunity. Craig et al. (2002) and Musa (1998) provide additional perspectives \non the concept of defect seeding. \nThis problem is not only related to changes brought about by external stimuli, but an early \npenetration test may have detected a serious misconfiguration in a perimeter firewall that \nwarrants immediate attention. In such circumstances, tests may need to be rerun to confirm \nthat any fix has indeed fixed the problem and has not created a new vulnerability \nsomewhere else. Unfortunately, retesting takes time and is rarely free. Therefore, the \norganization should agree up front, with any testing team (outsourced or internal), under \nwhat circumstances a change may be made to the system being tested while testing is in \nprogress, and if such an event should occur, how much retesting will be done. Possible \nstrategies include rerunning only the high-priority tests, only rerunning tests that have \npreviously failed, or performing a complete retest of the entire system. \n"
        },
        {
            "page_number": 211,
            "text": " \n204 \nLocation Where the Testing Is Conducted \nSome security-testing firms may only be willing to conduct their tests from an offsite or \nremote location, typically utilizing the Internet to reach the target Web site. Others may be \nwilling to send their consultants to a client's site but may charge an additional fee that \nmakes this scenario prohibitively expensive, especially if the consulting firm is based \noverseas. Although many may argue that testing over the Internet is a more realistic test \nenvironment, a well-configured perimeter firewall may mean that a remotely located testing \nteam is only able to confirm that a Web site's outer network and Web application defenses \nare working correctly. The team may not be able to test any additional inner network \ndefenses put in place to guard against an internal intruder or as a second line of defense in \ncase an as-yet-undiscovered flaw exists with a perimeter firewall. \nA possible solution includes opening up the firewall to provide the remote testing team with \naccess to the inner workings of the Web site, a potentially risky endeavor since the firewall \nwould be opened up for the rest of the world as well. Another option is to package up the \nWeb application(s) and ship it to the testing firm for them to install in a testing lab that has \nbeen configured to closely match the client's own production environment. Of course, the \naccuracy of these test results will largely depend upon how close the test lab's configuration \nmatches that of the clients. Table 9.3 attempts to summarize the points that should be \nconsidered when evaluating a perspective outside security-testing firm. \n \nTable 9.3: Outside Security Testing Firm Consideration Checklist  \nYES  \nNO  \nDESCRIPTION  \n□  \n□  \nHas the testing firm taken reasonable precautions to ensure that its \nown employees or the staff of any subcontractor will not take \nadvantage of the opportunity afforded them by the testing assignment \nto later initiate an unsanctioned attack against the client? \n□  \n□  \nIs the testing firm acceptable to the organization's insurance \nunderwriter (if any)? \n□  \n□  \nDoes the testing firm agree to be bound by clearly defined (and \ndocumented) terms of engagement? \n□  \n□  \nDoes the testing firm provide any kind of compensation if they, as a \nresult of their testing activities, significantly disrupt the normal \noperation of the system being tested? \n□  \n□  \nDoes the testing firm provide any kind of compensation (or \nguarantee) if they fail to detect a security hole that is later \nsuccessfully exploited by an intruder? \n□  \n□  \nDoes the contract with the testing firm specify what types of tests will \nbe conducted? Examples would be ping sweeps, port scans, \nsimulated distributed denial-of-service attacks, file share scans, \napplication source code reviews, submitting system commands via \napplication input data, and so on. \n□  \n□  \nDoes the contract with the testing firm specify whether the testing firm \nwill include recommendations on how to fix any detected \nvulnerability? \n□  \n□  \nDoes the testing firm also offer a consulting service for implementing \n"
        },
        {
            "page_number": 212,
            "text": " \n205 \nTable 9.3: Outside Security Testing Firm Consideration Checklist  \nYES  \nNO  \nDESCRIPTION  \nany recommendations they might make? \n□  \n□  \nIs the testing firm willing to divulge what testing tools (and versions) \nthey will use to conduct their tests? \n□  \n□  \nDoes the contract specify what head-start information (if any) will be \nprovided to the testing firm prior to commencement of the \nassessment? Examples include a complete list of network addresses \nused by the target site, specific version numbers of the system \nsoftware installed on the target site, or a list of services running on \nthe servers located closest to the perimeter firewall. \n□  \n□  \nDoes the contract with the testing firm specify the duration of the \ntesting effort and under what circumstances may testing be \nterminated, suspended, or extended? \n□  \n□  \nIf the testing is to be done remotely, have additional tests been \nscheduled that will test the security of the Web site from an internal \nattacker? \nCombination of In-House and Outsourced Testing \nPerhaps the most cost-effective approach to conducting a comprehensive security \nassessment and an associated penetration test of a system that is already built is to use a \ncombination of in-house and outsourced testing. An in-house testing team using free or low-\ncost tools that require only a moderate degree of training might conduct an initial security \nassessment, finding obvious and easily detected vulnerabilities without the need for \nexpensive outside consultants. Once all the detected holes have been plugged, one or \nmore testing firms that specialize in penetration testing may be hired to attempt to \ncompromise the system externally and, if resources permit, internally too. This approach \nnot only provides a second set of eyes and thereby reduces the chance that an unplugged \nhole exists in the production environment, but also provides a crude means of gauging the \neffectiveness of the testing process used by the in-house team. This is a useful gauge to \nhave when deciding whether or not future assessments or penetration tests should be \noutsourced or conducted by in-house staff. \nIn the case of a system being created, bringing in outside security experts early on may \nprove to be a cost-effective form of prevention. Not only would the system's design be more \nsecure, but the in-house staff can incorporate the expert's recommendations into their \nsecurity assessment checklists. \n \nTools for Testing \nAlthough it would be possible to manually conduct the vast majority of the tests mentioned \nin the previous chapters of this book, it often makes sense to try and automate as many of \nthese tests as feasibly possible. To this end, each section of this book has, where possible, \nincluded an accompanying list of sample tools that can be used to automate some or all of \nthe tests discussed within its section. Since all these lists contain more than one tool, a \nrecurring task is to evaluate each group of tools to determine which tool (or subset of tools) \nwould be the most appropriate for a specific security-testing assignment. The following \nsections summarize some of the main differences between a predominately manual \napproach and a predominately automated approach to security testing. If an automated \n"
        },
        {
            "page_number": 213,
            "text": " \n206 \napproach is desired, considerations for selecting the most appropriate tool are also \nincluded. \nManual Approach \nA manual approach may be cheaper to initially implement than an automated approach, but \nthe ongoing costs are likely to be higher. In addition, manual approaches typically do not \nscale as well as automated tests. For instance, a vulnerability assessment tool barely takes \nany longer to scan two Web sites than it does to scan one. In contrast, a manual effort may \ntake considerably longer. \nHighly creative security professionals performing a penetration test on a Web site may be \nable to figure out an ingenious way around a Web site's defenses and thereby reveal \nvulnerabilities that a less imaginative automated test might not have found. Unfortunately, \nunless these tests are well documented, such tactics may not prove to be as repeatable as \nan automated script. Such tests are more vulnerable to staff turnover, with the expertise \nliterally walking out of the door, something that is not likely to happen to an assessment tool \nowned by the organization. \nAutomated Approach \nRather than relying on an army of security experts, many organizations and security-testing \nfirms are now automating their Web site security assessments. In such a scenario, an \norganization acquires a security tool that is run against a target Web site in an attempt to \nreplicate the attacks that intruders have been known to use. Based on the success or \nfailure of these attacks, the tool attempts to assess and report which security vulnerabilities \nmay be present. Although these systems are fast, easily repeatable, and a possibly \ncheaper way of probing a Web site, they do have several potential drawbacks: \nƒ \nBecause the tools rely upon developers to create probes that look for each specific \nsecurity hole, the tool vendors are always running a little behind the latest tricks that \nattackers have discovered. Just like antivirus programs that rely on signature .dat \nfiles, it always takes a while before a new exploit is discovered by the good guys and \nsubsequently added to the list of automated probes. \nƒ \nAlthough quick, many automated tools are not as smart or as flexible as an \nexperienced security professional. A situation that often results is the automated tool \ngenerating a test summary report that contains many false positives (the testing tool \nwrongly detecting nonexistent problems). \nƒ \nAutomated tools rarely do a good job of prioritizing the legitimate issues, potentially \nburying must-fix-immediately problems in a sea of trivial warnings. For example, \ninstalling a custom HTTP 404 error on a Web site will cause some security \nassessment tools to generate false positives for every Common Gateway Interface \n(CGI) script vulnerability it tests for. The security assessment tool mistakenly \ninterprets the Web server returning the custom error page as an indication that the \nflawed CGI script it requested is present on the target Web server. \nƒ \nIf the tool has a high learning curve (possibly learning which automated warnings \nshould be taken seriously and which ones are false positives), the tool may be \nvulnerable to staff turnover. Klevinsky et al. (2002) and Scambray et al. (2001) \nprovide an introduction to using many of the tools commonly used for security testing \n(such as those listed in Table 9.4). \n \nTable 9.4: Sample List of Security Assessment Tools  \nNAME  \nASSOCIATED WEB SITE  \nAppScan \nwww.sanctuminc.com  \n"
        },
        {
            "page_number": 214,
            "text": " \n207 \nTable 9.4: Sample List of Security Assessment Tools  \nNAME  \nASSOCIATED WEB SITE  \nbv-Control for Internet Security/HackerShield \nwww.bindview.com  \nCerberus Internet Scanner \nwww.cerberus-infosec.co.uk  \nCybercop Scanner[a]  \nwww.nai.com  \nFoundScan \nwww.foundstone.com  \nNessus \nwww.nessus.org  \nNetRecon \nwww.symantec.com  \nRetina \nwww.eeye.com  \nSAINT \nwww.wwdsi.com  \nSANS Top 20 Scanner \nwww.cisecurity.org  \nScanner Database/Internet/System/Wireless \nwww.iss.net  \nSecureNet \nwww.intrusion.com  \nSecureScan \nwww.vigilante.com  \nSecurity \nAdministrator \nTool \nfor \nAnalyzing \nNetworks (SATAN) \nwww.fish.com  \nSecurity Analyzer \nwww.netiq.com  \nSecurity Auditor's Research Assistant (SARA) \nwww.www-arc.com [b]  \nSTAT Analyzer/Scanner \nwww.statonline.com/harris.com  \nTwwwscan \nwww.search.iland.co.kr  \nVigilEnt \nwww.pentasafe.com  \nWebInspect \nwww.spidynamics.com  \n[a]Effective July 1, 2002, Network Associates has transitioned the CyberCop product line \ninto maintenance mode. \n[b]This is not a typographical error-the Web site address is www.www-arc.com. \nThe pros and cons of manual versus automated testing are not restricted to security testing. \nMany of these same issues affect other types of testing such as functional and performance \ntesting. Buwalda et al. (2001), Dustin et al. (1999), Graham et al. (1999), and Hayes (1995) \nprovide additional information on test automation. Table 9.4 lists some security assessment \ntools that can be used to help automate a security-testing effort. \nTool Evaluation \nBecause the needs and resources of each organization and each project within a single \norganization will differ, it is not possible to specifically recommend a best tool (or set of \ntools) for each organization, especially when many of these tools are continually being \nupgraded and ported, making any judgment only valid for a snapshot in time and \nsubsequently rapidly out-of-date. Instead, this section discusses the recurring criteria that \nshould be considered when evaluating a group of similar tools. \n"
        },
        {
            "page_number": 215,
            "text": " \n208 \nDustin et al. (2001), Graham et al. (1999), and Stottlemyer (2001) provide additional \nsuggestions for evaluating testing tools. \nPlatform on Which Tools Will Be Used \nThe vast majority of security-testing tools available today were either initially intended to be \nexecuted from the Windows 9.x (95, 98, or ME), Windows NT (NT, 2000, or XP), or UNIX \nfamilies of operating systems. Although some of the most popular tools have been ported \nfrom their original platform (such as eEye Digital Security's porting of the nmap \nfingerprinting tool from a UNIX to a Windows NT platform), many tools are still only \navailable on one platform. Interestingly, some client/server security tools have been partially \nported, typically the client-side being usable from many platforms, but the server-side being \ninstallable on a much smaller number of platforms-for example, Nessus (www.nessus.com). \nThe choice of tools available to a tester may therefore be immediately restricted based \nupon the platforms available from which to conduct the tests. Unfortunately, for some \ntesters, their hardware budget may be nonexistent, restricting them to a single machine and \nconsequently forcing them to either install a boot manager (Table 9.5 lists some example \ntools for managing multiple operating systems on the same machine) that enables them to \nboot into two or more operating systems (a solution that typically degrades the available \ndisk space, virtual memory, and processing speed of the host machine) or restricting them \nto the tools from a single family. Although there typically exists a tool from each category \nthat will run on the Windows NT and UNIX families of platforms, the Windows 9.x family is \nnot supported as well. \n \nTable 9.5: Multiple O/S Management Tools  \nNAME  \nASSOCIATED WEB SITE  \nBoot Manager (System Selector) \nwww.bootmanager.com  \nOS/2 boot manager \nwww.ibm.com  \nPartitionMagic \nwww.powerquest.com  \nSystem Commander \nwww.v-com.com  \nVmware \nwww.vmware.com  \nWindows NT/2000 boot manager \nwww.microsoft.com  \nNote: In contrast to the software-based solutions listed above, Romtec \n(www.romtecusa.com) offers a hardware product (TriOS) that can be used to run \nseveral operating systems via multiple hard drives. \nSince no platform has a monopoly on the best tool for every category, many security testers \nregard having a sufficiently powered Windows NT family machine and a separate machine \nrunning some flavor of UNIX (often a Linux or BSD variant) as a bare minimum for \nconducting efficient testing. This, of course, assumes that the person(s) conducting the \ntests is familiar with both operating systems, a consideration that may play a significant \nfactor in the selection of an appropriate set of tools. Ideally, an organization may wish to \ninvest in a small test lab that contains the various operating systems needed to support the \ntools used to not only probe the target site, but also to build a replica of the site. This \nreduces the number of tests that need to be actually executed in the production \nenvironment and consequently reduces the risk that one of these tests might inadvertently \naffect the production site. Extreme care should be taken to exactly replicate the production \n"
        },
        {
            "page_number": 216,
            "text": " \n209 \nenvironment in the test lab, and even then, some testing would still be warranted in the \nproduction environment to confirm that it does indeed match the test lab and to execute \ntests against parts of the production environment known to differ from the test lab (for \nexample, ensuring the passwords on the production environment are sufficiently encrypted, \nas presumably these passwords are likely to be different from the ones used in the test lab). \nAn additional advantage of having an isolated testing lab is that the tools used to assess \nthe replica environment do not themselves pose the same security risk. Having these tools \ninstalled on a machine that has direct access to the production environment could make life \na lot easier for an attacker should they be able to compromise such a machine. Short of \nbeing given system administration userIDs and passwords on a silver platter, an attacker \ncould not wish for a better find than to stumble across a trusted machine with all their \nfavorite fingerprinting and cracking tools already installed, possibly along with test summary \nreports (saving the attacker the trouble of even having to run these tools). For this reason, if \nan isolated environment is not to be used, care should be taken to uninstall or disable any \nsecurity-testing tool present on a machine that might become accessible to an attacker. \nCost \nNo matter how big or small an organization, cost will always be a factor when comparing \ntwo or more tools, frequently being the main factor. Something that is often not considered \nso readily is the total cost of ownership. This cost not only includes the purchase price of a \ntool (if any), but also the cost associated with training people to use the tool, any hardware \nand system software licenses needed to use the tool, and the cost of keeping the tool up-to-\ndate (either in terms of a maintenance fee or the time spent by internal staff upgrading the \ntool). From a purchase cost perspective, tools can be grouped into one of the following \ncategories. \nFreeware \nUpon manually discovering a new vulnerability, many exploit authors will invest some time \nconverting their sequence of manual steps into an automated tool and then freely make it \navailable to the population at large. This is sometimes done as a means of encouraging the \nowner of the compromised product to quickly bring out a patch or merely as a means of \ndemonstrating the tester's own prowess. \nAdditionally, some security testers who are tired of running tedious and lengthy checks for \nknown exploits have developed useful utilities that significantly reduce the amount of time \nneeded to probe and potentially penetrate a target site. The net result is that a vast array of \nsecurity tools are available that can be downloaded from numerous Web sites free of \ncharge. Table 9.6 lists some sample Web sites that contain libraries of such tools. \n \nTable 9.6: Sample List of Tool Libraries  \nNAME  \nASSOCIATED WEB SITE  \nACME Laboratories \nwww.acme.com  \n@stake \nwww.atstake.com  \nChurch of the Swimming Elephant \nwww.cotse.com  \nCNET Networks \nwww.download.com  \n  \n  \nwww.shareware.com  \n  \n  \nwww.zdnet.com  \n"
        },
        {
            "page_number": 217,
            "text": " \n210 \nTable 9.6: Sample List of Tool Libraries  \nNAME  \nASSOCIATED WEB SITE  \nDaveCentral \nwww.davecentral.com  \nHackingExposed \nwww.hackingexposed.com  \nIdeahamster Organization \nwww.ideahamster.org  \nInsecure \nwww.insecure.org  \nNomad Mobile Research Centre (NMRC) \nwww.nmrc.org  \nNtsecurity \nwww.ntsecurity.nu  \nSourceForge \nwww.sourceforge.net  \nTucows \nwww.tucows.com  \nUltimate Search \nwww.freeware32.com  \n  \n  \n  \nWhen trying to locate and download testing tools via the Internet, it would be prudent to \ntemporarily disable any mobile code capability (such as ActiveX controls or Java applets) \nthat the browser uses. Unfortunately, some of the Web sites professing to offer free \nsecurity-testing tools may have also placed a hidden surprise on their Web site-a piece of \nmalicious mobile code that the visitor potentially downloads and executes (the promise of \nthe free tool being used as bait to lure unsuspecting victims to the Web site). \nAn even safer strategy for downloading files would be to use the services of an anonymous \nWeb server service (such as www.anonymizer.com), which hides personnel information \n(such as the network address being used by the browser) from a Web site of questionable \nintegrity. \nSome Trojan horses masquerade as security tools, performing their intended function as \nwell as some undesirable activity. For example, a well-known and trusted security-scanning \ntool could be enhanced to send a copy of its output report to an email account located in a \nforeign country, from which the author of the Trojan horse could review this sensitive \ninformation at his or her convenience. Therefore, after downloading any tool over the \nInternet, the file should be thoroughly scanned for viruses and Trojan horses, and then only \ninitially used in a safe, controlled environment where it can be monitored (such as via a \nnetwork or a host-based intrusion detection system [IDS]). \nAlthough a downloaded executable may be easier to initially get working, acquiring and \nreviewing the source code for an equivalent tool may allow the tool to be enhanced and \ncustomized (such as porting the tool to another platform). This also provides a developer a \ngreater learning opportunity. Additionally, having access to a tool's source code affords the \norganization the opportunity to examine the algorithms used by the tool to ensure that no \nTrojan horses are present. The organization can also determine if the tool contains any \nbugs (or features) that might result in inaccurate test results being produced or \nmisinterpreted. \nInterestingly, many of the freeware tools intended to run on a Windows platform are \ntypically distributed as binary files (exes, dlls, and so on). Tools destined for the UNIX \nenvironment are often made available as source code, requiring the perspective user to first \ninstall an appropriate compiler or interpreter before being able to execute the program. This \n"
        },
        {
            "page_number": 218,
            "text": " \n211 \ntrend is in part due to the relative homogeneity of the Windows platforms in comparison to \nthe diversity of UNIX. \nUnfortunately, many freeware tools come with minimum documentation, which may not be \nsufficient for anyone who is not highly conversant in the operating environment that the tool \nwas originally intended to be deployed in or will be installed into. Nor are there likely to be \nvery many training courses or books that provide detailed instructions on how to install, run, \nand interpret the tool's findings. This is not necessarily a huge concern for simple tools such \nas a basic ping-sweeping tool that comes with a near self-explanatory graphic user \ninterface (GUI). However, it is more of an issue with a tool that uses cryptic attribute \nsettings via a command-line interface and is only available as source code intended to be \ncompiled with obscure compiler options. \nShareware \nShareware differs from freeware in that the tool may be downloaded free of charge, but the \ntool's author expects anyone who continues to use the product after an evaluation period to \nremit some sort of payment. What makes shareware different from other commercially \navailable tools is that the payment is requested, not demanded by the author, relying on the \nhonor of the user rather than some built-in, time-elapsed control that disables the tool after \na short evaluation period. \nBuild It Yourself \nGiven the breadth of existing freeware security-testing tools currently available, it is unlikely \nthat an organization developing its own set of testing tools from scratch will ever get a \nsufficient return on investment (ROI) to make this approach feasible. This situation may not \nbe the case, however, if the organization intends to market its security-testing expertise to \nother firms or produces a tool with a unique feature not readily available anywhere else. An \nacceptable ROI may also be achieved if, instead of starting from scratch, a tool's existing \nsource code is modified to provide additional functionality, such as the capability to read \nnetwork addresses from an input file rather than a GUI. A tool's source code could also be \ncustomized to meet the precise needs of an organization; for instance, the tool could be \nported to a new environment, such as a different flavor of UNIX. \nLow-Cost/Budget Software \nMany security-testing products can now be purchased for less than a few hundred dollars. \nBoutique software companies, often composed of only a handful of full-time developers, \nhave created many of these products, while other products may have started life as a \nhobby before being turned into a part-time business by the product's author. Needless to \nsay, product support, reliability, compatibility, and the frequency with which upgrades are \nmade available vary greatly from product to product. \nSome tools are dependent on the functionality provided by products from other software \nvendors (for example, many tools have been designed to reuse components of MS-IE). \nAlthough using services from another vendor may often save the tool vendor considerable \nexpense and improve its time to market, dependency on another vendor's product can often \ncause compatibility issues. For example, tool X works fine with MS-IE v5.0, but not with \nv6.0, while the latest version of tool Y expects MS-IE 6.0 or higher to be installed. This \ncauses an issue, because only one version of MS-IE can typically be installed on a single \nmachine (boot managers and so on aside). The situation is further compounded if all the \nmembers of the testing team needs to install MS-IE 5.5 sp1, because this is the corporate \nstandard that all the in-house-developed intranet applications have been designed for and \ntherefore need to be tested with. \n"
        },
        {
            "page_number": 219,
            "text": " \n212 \nFreeware and shareware executables could also fall into this category if the product \nrequires a large amount of effort to install. Due to the typical lack of comprehensive \ndocumentation and nonexistent customer support, it may take a significant amount of time, \nand time is rarely free. \nHigh-End Software \nSome of the best tools on the market cost thousands of dollars, so much that actually \npurchasing them might not be the most cost-effective way of acquiring them. Leasing an \nexpensive tool for a short period of time or hiring another firm that already owns the \nsoftware to test the target site might be a cheaper way of conducting a one-time security \nassessment. Of course, over the long run, a single purchase may prove to be cheaper than \na recurring expense. \nTypically, these tools are executed remotely from the tool vendor's location or are \ndistributed as a platform-specific executable that is relatively easy to install. Thus, you have \nno need to worry about trying to build a new executable using dozens of source code files \nand a free compiler downloaded from a university's Web site. \nThe tool vendor may also offer a consulting service or be partnered with numerous value-\nadded retailers (VARs) who, for an additional fee, will install, configure, run, and even \ninterpret the tool's results. This provides a low-learning-curve entry into security testing, \nalbeit an expensive one. If such services are not easily available, then before investing in \nthe tool, an organization should weight the additional risk of not being able to find or train \nstaff who can effectively use the tool, causing the tool to either be underutilized or, worse \nstill, abandoned. \nHigh-end products have the potential to generate sufficient revenue to support frequent \nproduct updates (something a low-cost tool vendor may not be able to cost-justify), an \nimportant consideration given the rate with which new exploits are being discovered. Before \npurchasing an expensive security-testing tool, an organization should determine how \nfrequent the tool vendor releases updates and what the cost of obtaining these updates is \nexpected to be (an expense that is often quoted as a percentage of the original purchase \nprice). \nFinally, before investing a considerable sum into acquiring and learning how to use a \nspecific tool, the tool's vendor should be evaluated to gauge how likely it is that they will still \nbe in business in the foreseeable future. This is especially true if the lion's share of the \ncosts is to be expended up front. Table 9.7 lists the points to consider when evaluating \ndifferent security-testing tools. \n \nTable 9.7: Security-Testing Tool Selection Checklist  \nYES  \nNO  \nDESCRIPTION  \n□  \n□  \nDoes the organization have the resources to create a test lab for \nsecurity-testing purposes? \n□  \n□  \nHave all the ancillary costs such as training, setup time, future \nsoftware upgrades, consultant fees, and so on been included into the \ntotal cost of ownership of the tool? \n□  \n□  \nIs the proposed tool affordable? \n□  \n□  \nAre all tools downloaded from the Internet scanned for viruses and \n"
        },
        {
            "page_number": 220,
            "text": " \n213 \nTable 9.7: Security-Testing Tool Selection Checklist  \nYES  \nNO  \nDESCRIPTION  \nTrojan horses before being installed? \n□  \n□  \nAre all recently installed tools initially monitored for suspicious \nbehavior in a quarantined area before being deployed against the \nproduction environment? \n□  \n□  \nAre all tools that potentially have access to the production \nenvironment uninstalled or disabled when not in use? \n□  \n□  \nWill a proposed tool run on the organization's existing infrastructure? \n□  \n□  \nWill a proposed tool require extensive training? This includes any \ntime needed to learn a new operating system or set up a custom \nenvironment needed by the tool. An example would be installing a \nnew interpreter. \n□  \n□  \nIs the proposed tool available as an executable for the desired \nplatform? \n□  \n□  \nCan the proposed tool be used without any customization? \n□  \n□  \nIs the proposed tool available as source code? \n□  \n□  \nCan the proposed tool be used independently of any other product? \n□  \n□  \nIs the proposed tool intuitive and easy to use, or does it come with \ncomprehensive documentation? \n \nSummary \nUnless the organization has been the victim of a recent attack, the decision on who will \nconduct the testing and what tools they will use often comes down to how quickly and \ncheaply an initial assessment can be done in order to make it appear that the organization \nis being duly diligent.  \nActivities such as ongoing monitoring and reassessments may be given a lower priority \nthan a single initial assessment and consequently be underfunded. An approach that may \nwork well for accountants looking for a quantifiable one-time cost, as opposed to a recurring \ncost that will never go away, but may ultimately result in the organization being lulled into a \nfalse sense of security, and thereby exposed to any unknown exploit that may be present in \nan existing (or as-yet-unwritten) Web application or system software product installed on \nthe Web site. \n"
        },
        {
            "page_number": 221,
            "text": " \n214 \nChapter 10: Risk Analysis \nOverview \nYou may know what you are supposed to test, but you realize you don't have the time or \nresources to test everything. So which tests should you perform first (thereby ensuring that \nthese tests are executed), and which tests should you leave to last (and thereby risk never \ngetting to)? This chapter seeks to help the security-testing team answer these questions by \nlooking at how a risk analysis can be used to help the team decide which tests will provide \nthe best return on their testing investment and the order in which these tests should be run \n(test priority) just in case an unexpected event causes the testing effort to be curtailed after \ntesting has commenced. For example, it might turn out that the testing schedule was overly \noptimistic or that the quality of the system being tested was much poorer than anticipated \nand thereby slow the testing effort. \nIt may well be the case that the original designers of the system conducted a security risk \nanalysis prior to deciding upon what security measures would be incorporated into the \nsystem. Since it may be possible to reuse this analysis for the testing effort, the first portion \nof this chapter provides a brief overview of three different techniques that may have been \nperformed. If no such analysis exists, then the second section of this chapter should prove \nuseful, as it describes a simple approach that can be used to quickly help a security-testing \nteam decide where to focus their testing effort. The final section of this chapter outlines a \nmore rigorous approach, that of a Failure Mode, Effects, and Criticality Analysis (FMECA), \nwhich some practitioners may wish to use for highly critical systems. Gerrard et al. (2002) \nand Peltier (2001) both provide additional information on security risk analysis. \n \nSOME DEFINITIONS \n\"A risk is a measure of a probability and consequence of some undesirable event or \noutcome.\" Gerrard et al. (2002) \n\"Risk analysis is the process of identifying, estimating, and evaluating risk.\" Craig et al. \n(2002) \n \nRecycling \nIdeally, the architects of the system being tested will have chosen the security measures to \nimplement based upon a risk analysis of the security threats posed against the system. If \nthis is the case and the results of the risk analysis are still available, one approach for \ndetermining what should be tested and in what order is to simply review the results of the \noriginal analysis. Then tests can be designed to specifically check that each safeguard \nspecified by the design (as a means of mitigating each identified threat) has been \nimplemented correctly or if it has been implemented at all. \nOf course, the danger with recycling an earlier risk analysis is that any previous \nassumptions are no longer valid. For instance, at the time a new technology was selected \nto be used in the implementation of a Web site, there may not have been any known \nsecurity issues associated with the technology. With the passage of time, problems may \nhave surfaced, affecting the assumptions made during the original risk analysis. With the \nbenefit of hindsight, it would be prudent for the security-testing team to review the original \nanalysis before reusing it to make sure any previously made assumptions are still valid. \n"
        },
        {
            "page_number": 222,
            "text": " \n215 \nThe following are brief explanations of three different techniques that may have been used \nby the system's architects to help them assess what safeguards to implement: asset audits, \nfault and attack trees, and gap analysis. \nAsset Audit \nOne strategy used by some practitioners is that of an asset audit. This approach focuses on \nwhich assets (typically confidential information, but potentially including physical assets as \nwell) the organization possesses and then attempts to determine if they are being \nsufficiently protected. Krutz et al. (2001) provides additional information on asset audits. \nThe following is an outline of this approach: \nƒ \nIdentify all the data that the system being tested stores or has access to. Although \ncustomer records and bank accounts should obviously be included, less obvious \ncandidates for inclusion are files containing program source code, photographic \nimages, backup tapes, or other intellectual property. Care should therefore be taken \nto ensure that the list of identified assets is truly comprehensive. \nƒ \nFor each identified data asset, the means by which the data arrives and leaves the \nsystem should be determined (possibly using good old-fashioned dataflow \ndiagramming techniques), because each entry and exit point also poses a security \nrisk. \nƒ \nDetermine which mechanisms (threats) could be employed by an attacker to acquire \nthis information as the data enters the system, is stored on the system, or leaves the \nsystem. For example, an intruder could walk into a computer room and steal a backup \ntape or use a service running on a server (such as a Telnet) to view data files stored \non the machine. \nƒ \nOnce the threats have been identified, an approximation should be made of how \nlikely it is that each threat will be realized. \nƒ \nAssign a monetary value to the impact of data being destroyed, unavailable for a \ncertain period of time, stolen, or corrupted. This may depend to a large extent on how \nlong the organization could continue to operate with the problem. \nƒ \nDevelop a security policy (or modify an existing policy) that specifies the safeguards \nthat need to be implemented in order to protect all of the organization's critical data. \nƒ \nAlthough not necessarily a step in the asset audit that developed the security policy, \nthe measures specified in the policy should be checked to ensure that they have been \nimplemented correctly. \nOne issue that this technique quickly raises is the additional security risks associated with \ndistributing confidential information across multiple locations instead of in a heavily \ncontrolled central location. For example, when temporary copies of confidential data are \ncached on multiple hosts to improve performance, a speed-versus-security dilemma occurs, \nwhich the designer of a system must respond to by making trade-offs. \nFault Trees and Attack Trees \nAn asset audit takes the approach of focusing on what an organization needs to protect. An \nalternative approach, or one that can be used in combination with an asset audit, is to think \nabout what a potential assailant would want to acquire and then consider all the ways in \nwhich the attacker could go about trying to obtain the sought-after information. \nTo help document the different tactics that an attacker could use, some practitioners have \nadopted the fault-tree analysis or failure-tree analysis (FTA) technique, which is commonly \nused by the manufacturing industry. FTA is a deductive, top-down method for analyzing a \nsystem's design. It involves specifying a root event to analyze (such as a physical break-in \n"
        },
        {
            "page_number": 223,
            "text": " \n216 \nto a computer room), followed by identifying all the associated events (or second-tier \nevents) that could cause the root event to occur. \nFault trees are generally depicted graphically using a logical structure that consists of \nand/or decision boxes. Sometimes more than one second-tier event needs to occur before \nthe root event is triggered. In this case, these second-tier events would be arranged under \nan and box, meaning that all the second-tier events connected by the and would need to \nhappen in order to cause the root event to occur. All single second-tier events that would \ntrigger the root event on their own would be grouped under an or box. Leveson (1995) \nprovides \nadditional \ninformation \non \nfault \ntrees, \nwhile \nRelex \nSoftware \n(www.relexsoftware.com) offers a tool (Relex) designed to support a fault-tree analysis. \nFigure 10.1 shows a fault tree for an attacker trying to acquire a system administrator's \npassword. \n \nFigure 10.1: Fault tree example.  \nAttack trees are a variation of fault trees. An attack tree provides a formal, methodical way \nof describing who, when, why, how, and with what probability an intruder might attack a \nsystem. It thereby helps identify potential gaps in the current set of security policies. \nPotential attacks against a system are represented using the same tree-like structure used \nto illustrate a fault tree. The root node of the tree represents the ultimate goal of the \nattacker, and the branch and leaf nodes illustrate the different ways of achieving that goal. \nThe following steps outline how an attack tree can be built: \n1. Identify all the different types of intruders that might want to attack the system. This \nwould include script kiddies, accomplished attackers, dishonest employees, \norganized criminals, competitors, foreign governments, and so on. \n2. For each type of adversary, consider what their ultimate goal(s) might be. Each goal \nwill then be used as the root node for a separate attack tree. Although many of the \nattackers may share common goals, if their respective resources differ significantly, \n"
        },
        {
            "page_number": 224,
            "text": " \n217 \nit may make more sense to model their respective capabilities using separate trees \nthan to show a composite view of their combined capabilities in a single tree \n(although two or more trees may share common subtrees). \n3. Identify all the possible ways in which an attacker could hope to achieve each goal; \nthese tactics then become the second-tier goals that hang off the root node (goal). \n4. For each second-tier goal, consider whether this sub-goal could be accomplished in \nseveral different ways. For each strategy that could be employed to obtain a \nsubgoal, create a third-tier goal connected to the second-tier goal that it supports. \n5. This process should be repeated until each of the leaf nodes on the tree are \nspecified as a single, clearly defined approach. \n6. Evaluate each attack path using criteria such as the likelihood of this approach being \nattempted, the impact to the business should the goal be reached, and the ease \n(cost) with which safeguards can be put in place to block the attack. \n7. Modify the existing security policy (or if none exists, specify a new one) that specifies \nthe safeguards that need to be implemented in order to block all the critical attack \npaths. \n8. Once the measures specified in the policy have been applied, they should be \nchecked to ensure that they have been implemented correctly. \nOne drawback with this approach is that it requires the analyst to be sufficiently educated in \nthe ways of all the different types of attackers in order to accurately predict which strategies \nthey may choose to use. For this reason, this approach is probably not an ideal technique \nfor a novice security tester who may not have the insight needed to use this procedure (and \nmay even prove challenging to a seasoned veteran). Schneier (2000), Viega et al. (2001), \nCigital \nLabs \n(www.cigitallabs.com), \nand \nCounterpane \nInternet \nSecurity \n(www.counterpane.com) provide additional information on attack trees. \nGap Analysis \nA gap analysis is another strategy that can be used to determine how complete a system's \nsecurity measures are. The purpose of a gap analysis is to evaluate the discrepancies (or \ngaps) between an organization's vision of where it wants to be and its current reality. In the \nrealm of security testing, the analysis is typically accomplished by establishing the extent to \nwhich the system meets the requirements of a specific internal or external standard (or \nchecklist). Examples of external security standards include BS 7799 (for more information, \ngo to www.bsi-global.com) and ISO 17799 and 15408 (for details, go to www.iso.ch). The \nanalysis may also optionally make recommendations on which gaps should be plugged and \nhow. \nIf the gap analysis is to be conducted by comparing the system undergoing a test against a \nstandard list of items that need to be checked, the rigor of the analysis will depend to a \nlarge extent on the comprehensiveness and detail of the specified checks. A loosely \nworded checklist intended for a wide range of applications is more likely to be open to \ninterpretation than a technology- and application-specific standard, and therefore less likely \nto provide the same degree of assurance. Of course, the downside to using a platform-\nspecific standard is the time and skill needed to initially put one together and then keep it \nup-to-date as the technology evolves and as new weaknesses are discovered and \nsubsequently patched. Greenbridge Management (www.greenbridge.com), the Praxiom \nResearch Group (www.praxiom.com), and the Victoria Group (www.victoriagroup.com) \nprovide additional information on the gap analysis technique. \n \n"
        },
        {
            "page_number": 225,
            "text": " \n218 \nTest Priority \nThe security-testing team's work is simplified if it is able to inherit and reuse a \ncomprehensive risk analysis that has already been performed on the target of the testing \neffort. Unfortunately, this scenario seldom occurs. If the testing team is unable to inherit a \nprevious risk analysis, the team would be well advised to conduct its own risk analysis for \nthe purpose of deciding which tests should be given the highest priority, and test design \nand execution should be scheduled accordingly. Any testing team that does not perform \nsome form of risk analysis runs the risk that some of the most critical risks to the system will \nnot be tested. Such tests may be scheduled to be done at the end of the process, and due \nto a time crunch, they would quite probably be omitted, leaving the system dangerously \nexposed. \nThe following sections describe a risk analysis that can be done relatively quickly for the \npurpose of assisting the security-testing team as they try to decide which areas of the \nsystem should be tested first and most comprehensively. This process does not purport to \nbe precise or necessarily the appropriate approach for selecting security defenses. It is \nmerely a quick and simple way of systematically assigning priorities to the collection of tests \nthat the testing team would ideally want to perform, but either know for certain or suspect \nthat they will not have sufficient time to actually execute. \nDevice Inventory \nThe first step in the risk analysis process is to build an inventory of all the devices that are \nto be tested. Hopefully, this should already have been completed (at least at a high level) \nwhen the scope of the testing effort was defined (refer back to Chapter 3 on identifying the \nscope of the security-testing project). The amount of descriptive detail to include in the \ndevice inventory will depend upon how large and complicated the system to be tested is. \nFor instance, for larger installations, using static IP addresses or hostnames may be a more \naccurate way of identifying a specific hardware device than using generic role names, such \nas Web Server 1. Each of the identified devices may be further described by specifying the \nkey software and data components that are present on the device. The software installed \non a particular device will, to a large degree, determine which vulnerabilities may be \npresent for that device, while the data stored on the device will influence the impact on the \nbusiness should a particular threat be realized. Each entry in the device inventory \n(sometimes referred to as a test objective inventory) will subsequently be evaluated to \ndetermine what degree of testing (if any) will be applied to it. Table 10.1 depicts an example \nlayout structure for a device inventory. \n \nTable 10.1: Layout Structure Example for a Device Inventory  \nDEVICE \nID  \nDEVICE \nDESCRIPTION  \nKEY \nSOFTWARE \nCOMPONENTS  \nKEY \nDATA \nCOMPONENTS  \nInternal  \n1 \nPerimeter firewall \nEmbedded \noperating \nsystem \nFirewall rules. \n2 \nWeb server 1 \nWindows \n2000, \nIIS \n(HTTP \nonly), \nASP, \nSSI, CGI, host-based \nIDS, and Web-server-\ntier \nportion \nof \nthe \nWeb server and IDS \nlogs. Non-sensitive data \nsuch as image files and \nsource code for the \nportion of the application \n"
        },
        {
            "page_number": 226,
            "text": " \n219 \nTable 10.1: Layout Structure Example for a Device Inventory  \nDEVICE \nID  \nDEVICE \nDESCRIPTION  \nKEY \nSOFTWARE \nCOMPONENTS  \nKEY \nDATA \nCOMPONENTS  \napplication \nnot requiring a user to \nlogin. \n3 \nWeb server 2 \nWindows \n2000, \nIIS \n(HTTP and HTTPS), \nASP, SSI, CGI, host-\nbased IDS, and Web-\nserver-tier portion of \nthe application \nWeb server and IDS \nlogs. Source code for \nthe \nportion \nof \nthe \napplication requiring a \nuser to login. \n4 \nWeb server 3 \nWindows \n2000 \nand \nFTP \nPublicly \navailable \nreports. \n5 \nLoad balancer \nEmbedded \noperating \nsystem \nRouting restrictions (if \nany). \n6 \nLocal \narea \nnetwork (LAN) \nTCP/IP and Ethernet \nLAN \ndata \ncommunication between \ndifferent tiers of the \napplication. \n7 \nDemilitarized \nZone \n(DMZ) \nfirewall \nEmbedded \noperating \nsystem \nFirewall rules. \n8 \nInternal switch \nEmbedded \noperating \nsystem \nRouting tables. \n9 \nNetwork sniffer \nWindows \n2000 \nand \nNetwork-based IDS \nIDS logs. \n10 \nApplication server \nUNIX, \napplication-\nserver-tier portion of \nthe application, and \nhost-based IDS \nIDS \nlogs, \nlots \nof \napplication .tmp files, \nand an application audit \ntrail. \n11 \nDatabase server \nUNIX, Oracle DBMS, \nhost-based IDS, and \ndatabase-server-tier \nportion \nof \nthe \napplication \nbeing \ntested \nDBMS and IDS logs, \napplication data. \n12 \nNetwork \ncontroller \nand \nDomain \nName \nSystem \n(DNS) \nserver \nWindows 2000 \nDNS lookup tables as \nwell as network user IDs \nand passwords. \n13 \nBackup \nnetwork \ncontroller \nand \nDNS server \nWindows 2000 \nDNS lookup tables as \nwell as network user IDs \nand passwords. \n14 \nComputer \nroom \nWindows 2000 \nLocal password file. \n"
        },
        {
            "page_number": 227,
            "text": " \n220 \nTable 10.1: Layout Structure Example for a Device Inventory  \nDEVICE \nID  \nDEVICE \nDESCRIPTION  \nKEY \nSOFTWARE \nCOMPONENTS  \nKEY \nDATA \nCOMPONENTS  \ndesktop \n15 \nGateway firewall \n(connection \nto \nlegacy system) \nEmbedded \noperating \nsystem \nFirewall rules. \n16 \nComputer \nroom \nprinter \nEmbedded \noperating \nsystem \nN/A \n17 \nPhysical security \nof all hosts in the \ncomputer room \nN/A \nAll of the above. \n18 \nPhysical security \nof all cabling at \nthe host site \nN/A \nLAN \ndata \ncommunication \ncomponent \nof \nthe \napplication \nand \nany \nsystem \nsoftware \ncommunication. \n19 \nAnd so on ... \n  \n  \nExternal  \n32 \nISP router \nUnknown \nRouting restrictions (if \nany). \n33 \nInternet \nUnknown \nWAN \ndata \ncommunication \ncomponent \nof \nthe \napplication. \n34 \nClient machine \nAny \nhardware, \noperating \nsystem, \nbrowser, and plug-in \ncombination. \nClient-\ntier \nportion \nof \nthe \napplication \nBrowser \ncached \nfiles \nand Web site cookie. \nNote: The Device ID column may be populated with an organization's existing \ninventory-tracking ID, vendor serial number, or with a new ID assigned by the \nsecurity-testing team, whichever is easiest to implement. \n \nTECHNIQUE HISTORY \nThe approach outlined in this section is derived from a test-prioritizing technique that Rick \nCraig devised while conducting a risk analysis of U.S. military computer systems for the \npurpose of determining security-testing priorities. A more detailed explanation of the \napproach as it applies to testing all software features (usability, performance, functionality, \nand so on) can be found in his book, Systematic Software Testing (Craig et al. [2002]). \n"
        },
        {
            "page_number": 228,
            "text": " \n221 \nThreats \nOnce a device inventory has been compiled, the next step in this process is to list the \ndifferent security threats (or failure modes) that each hardware device and software \ncomponent faces. Table 10.2 illustrates a worksheet for recording the threats that a Web \nsite could face. \n \nTable 10.2: Example Worksheet for Recording Threats  \nDEVICE/THREAT/EXPLOIT \nID  \nDEVICE \nDESCRIPTIO\nN  \nTHREAT  \nPOSSIBLE \nCAUSES \n(EXPLOITS)  \n1.1.1 \nPerimeter \nfirewall \nLegitimate \nnetwork traffic \nis unable to \npass through \nfirewall. \nDenial-of-\nservice attack. \n1.1.2 \n  \n  \nPhysical thief or \ndamage to the \nfirewall. \n1.1.3 \n  \n  \nAn \nintruder \nchanges \nthe \nfirewall's \nrules \nto \nblock \neverything \nor \nmore than it did \nbefore. \n1.2.1 \n  \nUnauthorized \nnetwork traffic \nis permitted to \npass through \nthe firewall. \nWrongly \nconfigured \nfirewall rules. \n1.2.2 \n  \n  \nAn \nintruder \nchanges \nthe \nfirewall \nrules \n(possibly using \nvendor default \nuser \nID/password). \n2.1.1 \nWeb server 1 \nAn \nunauthorized \nprocess \nmay \nbe run on the \nserver \nusing \nsystem-\nadministrator-\nlevel \nprivileges. \nExploitable \nthird-party CGI \nscripts \nare \npresent. \n2.1.2 \n  \n  \nEasily \n"
        },
        {
            "page_number": 229,
            "text": " \n222 \nTable 10.2: Example Worksheet for Recording Threats  \nDEVICE/THREAT/EXPLOIT \nID  \nDEVICE \nDESCRIPTIO\nN  \nTHREAT  \nPOSSIBLE \nCAUSES \n(EXPLOITS)  \nguessable \nsystem \nadministrator \nuser \nID \nand \npassword. \n2.2.1 \n  \nConfidential \ninformation \n(source code, \ndata \nfiles, \nsystem \nconfiguration \ninformation, \nand \nso \non) \nstored on the \nserver can be \naltered \n(appended, \nchanged, \nor \ndeleted). \nFile \nand \ndirectory \nshares are not \nadequately \nprotected. \n2.2.2 \n  \n  \nUnneeded \nservices are left \nenabled on the \nWeb server. An \nexample would \nbe NetBIOS on \nports \n135 \nthrough 139. \n2.3.1 \n  \nConfidential \ninformation \nstored on the \nserver can be \ndeduced, \nviewed online, \nor \ndownloaded. \nPerimeter \nfirewall \ndoes \nnot \nblock \ninappropriate \ncommunication. \nDNS \nzone \ntransfers are an \nexample. \n2.3.2 \n  \n  \nUser \naccount \nwith \nnull \npassword \nis \npresent. \n2.3.3 \n  \n  \nUnneeded \nservices are left \nenabled on the \nWeb \nserver. \nFTP \nis \nan \nexample. \n"
        },
        {
            "page_number": 230,
            "text": " \n223 \nTable 10.2: Example Worksheet for Recording Threats  \nDEVICE/THREAT/EXPLOIT \nID  \nDEVICE \nDESCRIPTIO\nN  \nTHREAT  \nPOSSIBLE \nCAUSES \n(EXPLOITS)  \n2.4.1 \n  \nAn \nunauthorized \nprocess \nmay \nbe run on the \nserver \nusing \nlimited \nprivileges. \nCanned buffer \noverflow exists \nfor \ninstalled \nsystem \nsoftware. \n2.4.2 \n  \n  \nServer wrongly \ntrusts \nan \ninsecure \nserver. \n2.4.3 \n  \n  \nServer \nSide \nInclude \n(SSI) \noptions \nare \nmisconfigured. \n2.5.1 \n  \nAccess to the \nserver's \nfunctionality is \nlost. \nDenial-of-\nservice attack. \n2.5.2 \n  \n  \nPhysical \ntheft \nor damage to \nthe server. \n2.5.3 \n  \n  \nAn \nintruder \nacquires control \nof the server \nand \ndisables \nthe \nfunctionality. \n3.1 \nWeb server 2 \n... and so on. \n... and so on. \nNote: A device, threat, and exploit ID column can be added to make referencing \neach of the device, threat, and exploit combinations easier. \nThe possible threats could be determined by identifying the specific exploits that could \ncause such threats to occur. \"A group of drone servers infected with the Stachel-draht tool \nmay be used to launch a tribal-flood network style denial-of-service attack.\" Although going \ninto such detail may make designing a test case that can establish whether or not a system \nis susceptible to this sort of attack easier, for the purposes of assigning a testing priority it is \nprobably overkill. Instead, describing in more generic terms groups of similar exploits that \npose substantially the same threat should help to keep the list of exploits to a more \nmanageable size. For example, \"Exploitable third-party CGI scripts are present\" may be \nmore useful at this stage than listing each individual CGI script that's known to be \nexploitable. \n"
        },
        {
            "page_number": 231,
            "text": " \n224 \nNote a device, threat, and exploit ID column can be added to make referencing each of the \ndevice, threat, and exploit combinations easier. A common way by which a list of threats \nmay be compiled is to host a brainstorming workshop. Ideally, the workshop candidates \nwould have different skill sets and knowledge domains. For instance, candidates would \ninclude network administrators and engineers, application developers and testers, security \nofficers and consultants, auditors, and so on. \nTo make the workshop more productive, the security-testing team may wish to put together \nan initial candidate list of threats to prompt the workshop discussions. This candidate list \nmay be based on an inventory produced in a previous workshop or the team's own \nresearch and experiences. The participants in the workshop can then initially focus on \nadding threats that have not yet been identified, later removing threats that do not appear to \nbe worth testing, and merging or splitting threats into more meaningful and manageable \ngroupings. \nSome threats may be removed during the course of the workshop because they are \nconsidered extremely unlikely, would have a negligible effect on the business if they were to \nhappen, or are too impractical to reliably test for. Their removal and the associated rationale \nfor their removal should be documented, thereby allowing future risk analysis to reexamine \nany assumptions made and subsequently reevaluate the threat's removal. \nThe outcome of the workshop should be a worksheet that may not list every conceivable \nthreat, but it should be comprehensive enough to have hopefully identified all the system's \nsignificant threats. If time permits, this threat list can be validated by researching the \nvarious exploit-tracking databases to ensure that the list is comprehensive and up-to-date \n(see Table 4.2 for a sample list of such databases.) Figure 10.2 summarizes the steps \nperformed in this research process. \n \n \nFigure 10.2: Risk analysis process overview.  \nBusiness Impact \nWhat would be the business impact on the organization if a threat were to be realized? \nUnfortunately, every security breach has the potential to escalate, so the typical gut \nreaction of a security tester is to rate the severity of every potential breach as critical. \nAlthough assigning a value of critical to each threat may help to stress the importance of \nsecurity testing to senior management, it does not help a security-testing team prioritize \ntheir testing effort (not every test case can be run first). While technicians such as network \nengineers and security analysts are often best suited to determine the likelihood of a \nparticular exploit occurring, the users or owners (or their proxies) of the system are usually \nin the best position to judge how great an impact each security failing might be on the \n"
        },
        {
            "page_number": 232,
            "text": " \n225 \nbusiness. Of course, these users may require the help of a knowledgeable security expert \nin order for them to convert a security threat into a business impact that they can \ncomprehend and thereby accurately assess. \nIt would be nice to specify the business impact in terms of dollars (a quantitative \nassessment), but trying to put an exact monetary figure on the cost of an intruder being \nable to execute a shell command on a Web server is hard to do. Instead, it may prove \neasier to apply an approach that uses a qualitative assessment. For example, one can \nsimply rank the threats that have been identified, placing the most undesirable threats near \nthe top of the order and the threats that have the least relative impact at the bottom. \nA relative severity can then be assigned to each threat. Although some risk analysis \ntechniques identify many different degrees of severity, for the purposes of identifying test \npriority, three categories typically prove sufficient. The three levels of severity are as \nfollows: \nƒ \nHigh. Of all the threats identified for this system, these have the greatest potential \nimpact on the business. \nƒ \nMedium. These threats would still have a significant impact on the business, but in \ncomparison to the ones listed in the high category, they would be relatively moderate. \nƒ \nLow. These would have a small or negligible impact when compared to the previous \nthreats. \nNote that the key word here is relative. Because this process is trying to determine the \ntesting priority, the ultimate goal is to determine which test to run first, which means that \neven the most trivial of systems will have some threats that are relatively high compared to \nother threats. Table 10.3 illustrates the relative business impact of some of the threats that \nwere identified in Table 10.2. \n \nTable 10.3: Assigning a Relative Business Impact to Threats  \nDEVICE/THREAT \nID  \nDEVICE \nDESCRIPTIO\nN  \nTHREAT  \nRELATIVE \nBUSINES\nS IMPACT  \nCOMMENT  \n1.1 \nPerimeter \nfirewall \nLegitimate \nnetwork \ntraffic \nis \nunable \nto \npass through \nfirewall. \nM \nSome loss of \nrevenue and \nincreased \nhelp \ndesk \ncalls. Impact \nto \nbusiness \nwould \nincrease the \nlonger \nthe \noutage \nlasted. \n1.2 \nPerimeter \nfirewall \nUnauthorized \nnetwork \ntraffic \nis \npermitted to \npass through \nthe firewall. \nH \nFailure \nof \nperimeter \nfirewall would \npotentially \nallow \nintruders \nto \n"
        },
        {
            "page_number": 233,
            "text": " \n226 \nTable 10.3: Assigning a Relative Business Impact to Threats  \nDEVICE/THREAT \nID  \nDEVICE \nDESCRIPTIO\nN  \nTHREAT  \nRELATIVE \nBUSINES\nS IMPACT  \nCOMMENT  \ntry \nand \ncompromise \nany (or all) of \nthe machines \nin the DMZ. \nThe \ninterior \nfirewall \nmay \nstill \nprovide \nsecurity \nto \nthe \ninternal \nLAN. \n2.1 \nWeb server 1 \nAn \nunauthorized \nprocess may \nbe run on the \nserver using \nsystem-\nadministrator\n-level \nprivileges. \nH \nGives \nintruder \naccess to all \nconfidential \ninformation \nstored \non \nmachine. \nA \nconcern \nexists that the \nmachine \ncould \nbe \nused \nas \na \ndrone \nin \na \ndistributed \ndenial-of-\nservice attack \nagainst \nanother Web \nsite \n(the \nhardware and \nbandwidth \nhas a high \ncapacity) \nor \nas a host for \nlaunching \nintrusion \nattempts \nagainst other \nnetworks. \n2.2 \nWeb server 1 \nConfidential \ninformation \n(source \ncode, \ndata \nfiles, system \nconfiguration \ninformation, \nH \nA \nworse \nsituation than \nbeing able to \nread \nthe \ninformation, \nthis will likely \nmake \nthe \n"
        },
        {
            "page_number": 234,
            "text": " \n227 \nTable 10.3: Assigning a Relative Business Impact to Threats  \nDEVICE/THREAT \nID  \nDEVICE \nDESCRIPTIO\nN  \nTHREAT  \nRELATIVE \nBUSINES\nS IMPACT  \nCOMMENT  \nand so on) \nstored on the \nserver \ncan \nbe \naltered \n(appended, \nchanged, or \ndeleted). \ncleanup \nprocess \nmessier. \nIf \naccess allows \nfor \nthe \nuploading of \nan attacker's \ntoolkit, \nthis \nbreech \nmay \nallow \nthe \nattacker \nto \nescalate their \nprivileges, \nultimately \ntaking \nfull \ncontrol of the \nserver. \n2.3 \nWeb server 1 \nConfidential \ninformation \nstored on the \nserver \ncan \nbe deduced, \nviewed \nonline, \nor \ndownloaded. \nM \nIt \nis \nsuspected \nthat \ndevelopers \nmay \nhave \nhard-coded \ndatabase \nuserIDs and \npasswords \ninto \nthe \napplication \nsource code \nresiding \non \nthis \nserver. \nAlso, \nthe \ncurrent \n(questionable\n) \nsystem \ndesign \ncalls \nfor \ncaching \nin-progress \nfinancial \ntransactions \non the Web \nserver. \n2.4 \nWeb server 1 \nAn \nunauthorized \nprocess may \nbe run on the \nserver \nwith \nM \nThe \nWeb \nserver is in a \nDMZ \nand \nshould \nonly \nhave \n"
        },
        {
            "page_number": 235,
            "text": " \n228 \nTable 10.3: Assigning a Relative Business Impact to Threats  \nDEVICE/THREAT \nID  \nDEVICE \nDESCRIPTIO\nN  \nTHREAT  \nRELATIVE \nBUSINES\nS IMPACT  \nCOMMENT  \nlimited \nprivileges. \nmarginally \nmore access \nprivileges \nto \nthe \ninner \nnetwork \n(application \nserver and so \non) than any \nother \nmachine \nof \nthe Internet. \nAn \nintruder \nmay be able \nto \naccess \nconfidential \ninformation \nand/or \nescalate \nhis \nor \nher \nauthority to a \nsystem \nadministrator'\ns level. \n2.5 \nWeb server 1 \nAccess \nto \nthe \nserver's \nfunctionality \nis lost. \nL \nIn the short \nterm, \nspare \ncapacity \nin \nother servers \nshould \nbe \nable to pick \nup the slack, \nassuming \nother servers \nare \nstill \nfunctional. If \nall servers go \ndown, \nthe \napplication is \nnot \nmission \ncritical \nand \nthe \norganization \ncan \nsurvive \nuntil \nthe \nbackup \nsite \ncomes online. \nNote: The business impact assessment and comments in this table are intended to \nillustrate this process, not provide an assessment for every organization's Web site, as \n"
        },
        {
            "page_number": 236,
            "text": " \n229 \nTable 10.3: Assigning a Relative Business Impact to Threats  \nDEVICE/THREAT \nID  \nDEVICE \nDESCRIPTIO\nN  \nTHREAT  \nRELATIVE \nBUSINES\nS IMPACT  \nCOMMENT  \nthe consequences of a Web site being compromised will vary significantly from site to \nsite. \nFrom a test priority perspective, it would be convenient if a third of all the threats could be \nassigned to each of the three categories. In reality, the dividing point between a high impact \nand a medium impact may not happen exactly one-third of the way down the list, which is \nfine. But an analysis that results in 90 percent of the threats being assigned a relative \nseverity of high will diminish the usefulness of the analysis for test priority purposes. \n \nRATING VARIANCE \nA rating system that is predominately subjective or judgmental in nature (such as assigning \nlow, medium, or high values to the business impact in a risk analysis) and demonstrates a \nwide variation in the values assigned to a specific entry by the raters may indicate that the \nraters do not possess enough information to make a valid assessment. For example, if two \nof the four raters feel a particular threat is extremely improbable (low), but the other two feel \nthat it very likely (high), then the average rating is going to be in the middle (medium), a \nrating that no one currently concurs with. Under such circumstances, the group would be \nwell advised to postpone making a final assignment until they have had a chance to debate \nwhy they have such differing opinions and, if deemed necessary, to conduct further \nresearch on the topic. \nRisk Likelihood \nSome exploits are far easier to accomplish than others. For example, a remote port scan is \na lot easier to run than trying to find and exploit a new application buffer overflow. The \nreward for some successfully executed techniques is much greater than for others. For \nexample, correctly guessing a system administrator's userID and password combination is \nlikely to be much more satisfying than reviewing the results of a ping sweep. In addition, in \norder to be successful, some exploits may depend on the existence of more than one \nsecurity failing (such as a perimeter firewall permitting a DNS transfer to take place and the \nDNS server fulfilling this request). It is not surprising that intruders favor some exploits over \nothers; therefore, the probability of different exploits being performed successfully will vary \nand this should influence the priorities assigned to the tests designed to detect the \nexistence of these vulnerabilities. \nUp until this point, the relative likelihood of each possible security failure occurring has not \nbeen considered (with the exception of ruling out threats that are so unlikely to happen that \nit would be a poor use of time to continue to include them in the risk analysis). \nUnfortunately, trying to accurately gauge the likelihood of one type of exploit over another is \nnot an easy thing to do, especially when exploit popularity changes by the month. \nFundamentally, this problem can be resolved in one of two ways: using metrics specific to \nthe organization or using generic industry metrics. \nInternal Metrics \nThe benefit of using information that is directly derived from an organization's own \nexperiences is that it is theoretically a better gauge of the exploits that are likely to be \n"
        },
        {
            "page_number": 237,
            "text": " \n230 \nemployed by potential attackers against this specific system. Possible methods of collecting \nexploit likelihood metrics for a specific organization include the following: \nƒ \nReviewing an organization's own records of failed (and successful) exploit attempts. \nAn example would be reviewing IDS reports. \nƒ \nReviewing the results of a honey pot project (see Chapter 8 for an explanation of \nhoney pots). \nƒ \nExtending the workshop used to identify security threats (described earlier in this \nchapter) to also include an estimation of exploit likelihood. \nExternal Metrics \nUnfortunately, gathering organization-specific metrics can be time consuming or impossible \nto calculate due to a complete lack of raw data or data that is too small a sample to be \nstatistically valid. The latter would be the case if an organization were connecting a legacy \nsystem to the Internet for the first time and there had therefore never been an opportunity \nfor an external intruder to attack the system. It is prudent to not only take into account any \nmetrics collected internally, but also to consider externally gathered metrics. Possible \navenues for externally researching exploit popularity include the following: \nƒ \nReviewing the exploit lists on Web sites that report on security incidents. Appendix B \nlists the top-20 critical Internet security vulnerabilities as identified by the System \nAdministration, Networking, and Security (SANS) Institute (www.sans.org). \nƒ \nSeeking the recommendations of external security consultants. \nƒ \nReviewing publications (such as the latest edition of Hacking Exposed by Scambray \net al. (2001), which rates specific exploits by popularity). \nWhether one (or more) of the previously mentioned methods of research is employed, or \nthe security-testing team decides to rely on their own gut feeling, each identified exploit \nshould be assigned a relative probability of being successful. This is assuming no additional \nsecurity tests were to be done to determine the existence or nonexistence of the \nvulnerability. \nSince assigning the relative likelihood is a subjective matter, using scales with many \noptions, such as 1 to 100, may result in more time being spent debating whether or not a \nparticular exploit has a relative probability of 66 or 67 than it helps with prioritizing the \ntesting effort. (It may also lead the casual reviewer to believe that the rating is more precise \nthan it actually is.) Instead, a simple categorization of high, medium, and low may prove to \nbe sufficient. Table 10.4 depicts a portion of the inventory used to document the relative \nlikelihood of specific exploits being successfully employed against some of the devices \nfound in Table 10.2. \n \n \nTable 10.4: Assigning a Relative Likelihood to Potential Exploits  \nDEVICE/T\nHREAT/EX\nPLOIT ID  \nDEVICE \nDESCRIPTION  \nTHREAT  \nEXPLOIT  \nRELATIV\nE \nLIKELIH\nOOD  \nCOMMENT  \n1.1.1 \nPerimeter \nfirewall \nLegitimate \nnetwork \ntraffic \nis \nunable \nto \npass \nDenial-of-\nservice \nattack. \nL \nLarge \nsurplus \nnetwork \ncapacity \nmeans \n"
        },
        {
            "page_number": 238,
            "text": " \n231 \nTable 10.4: Assigning a Relative Likelihood to Potential Exploits  \nDEVICE/T\nHREAT/EX\nPLOIT ID  \nDEVICE \nDESCRIPTION  \nTHREAT  \nEXPLOIT  \nRELATIV\nE \nLIKELIH\nOOD  \nCOMMENT  \nthrough \nfirewall. \nfirewall \nnot \nlikely to be a \nbottleneck in \na \ndenial-of-\nservice \nattack. \n1.1.2 \n  \n  \nPhysical \ntheft \nor \ndamage to \nthe firewall. \nL \nBox is in a \nsecured \nroom. \n1.1.3 \n  \n  \nAn intruder \nchanges \nthe firewall \nrules \nto \nblock \neverything. \nM \nFirewall may \nstill \nhave \ndefault \naccount \nactive. \n1.2.1 \n  \nUnauthoriz\ned network \ntraffic \nis \npermitted \nto \npass \nthrough the \nfirewall. \nWrongly \nconfigured \nfirewall \nrules. \nH \nPrevious \ninspection of \nrules \nfound \nseveral \nerrors. \n1.2.2 \n  \n  \nAn intruder \nchanges \nthe firewall \nrules. \nM \nFirewall may \nstill \nhave \ndefault \naccount \nactive. \n2.1.1 \nWeb server 1 \nAn \nunauthoriz\ned process \nmay be run \non \nthe \nserver \nusing \nsystem-\nadministrat\nor-level \nprivileges. \nVulnerable \nCGI script \ninstalled on \nWeb \nserver. \nL \nCGI \ncapability \n(theoretically\n) disabled. \nCalculating Relative Criticality \nOnce high, medium, and low values have been assigned to the likelihood of an exploit \nbeing successful, and the impact to the business should the event occur, it then becomes \npossible to combine these values into a single assessment of the criticality of this potential \n"
        },
        {
            "page_number": 239,
            "text": " \n232 \nvulnerability. For example, assigning a numeric value of 3 to a high, 2 to a medium, 1 to a \nlow, and then adding the two numeric values together will result in a criticality between 2 \nand 6, as depicted in Figure 10.3. \n \n \nFigure 10.3: Calculating relative criticality. Note that multiplying instead of adding the two \nvariables will result in a scale of 1 to 9 and marginally increase the criticality of \nmedium/medium pairs over low/high pairs.  \nRegardless of whether addition or multiplication is used to combine the two variables, the \nresulting threat/exploit matrix can be sorted using the calculated criticality. Table 10.5 \nillustrates this for some of the threats and exploits mentioned previously. \n \nTable 10.5: Calculating Threat/Exploit Criticality  \nDEVICE/T\nHREAT/E\nXPLOIT \nID  \nDEVICE \nDESCRI\nPTION  \nTHREA\nT  \nRELATI\nVE \nBUSIN\nESS \nIMPAC\nT  \nEXPLOI\nT  \nRELATIVE \nLIKELIHO\nOD  \nRELATIVE \nCRITICALIT\nY  \n1.1.1 \nPerimet\ner \nfirewall \nLegitima\nte \nnetwork \ntraffic is \nunable \nto pass \nthrough \nfirewall. \nM (2) \nDistribut\ned \ndenial-\nof-\nservice \nattack. \nL (1) \n2(1 + 1) \n1.1.2 \n  \n  \nM (2) \nPhysical \ntheft \nof \nor \ndamage \nto \nthe \nfirewall. \nL (1) \n3(2+1) \n"
        },
        {
            "page_number": 240,
            "text": " \n233 \nTable 10.5: Calculating Threat/Exploit Criticality  \nDEVICE/T\nHREAT/E\nXPLOIT \nID  \nDEVICE \nDESCRI\nPTION  \nTHREA\nT  \nRELATI\nVE \nBUSIN\nESS \nIMPAC\nT  \nEXPLOI\nT  \nRELATIVE \nLIKELIHO\nOD  \nRELATIVE \nCRITICALIT\nY  \n1.1.3 \n  \n  \nM (2) \nAn \nintruder \nchanges \nthe \nfirewall \nrules to \nblock \neverythi\nng. \nM (2) \n4 (2 + 2) \n1.2.1 \n  \nUnautho\nrized \nnetwork \ntraffic is \npermitte\nd \nto \npass \nthrough \nthe \nfirewall. \nH (3) \nWrongly \nconfigur\ned \nfirewall \nrules. \nH (3) \n6 (3 + 3) \n1.2.2 \n  \n  \nH (3) \nAn \nintruder \nchanges \nthe \nfirewall \nrules. \nM (2) \n5 (3 + 2) \n2.1.1 \nWeb \nserver 1 \nAn \nunautho\nrized \nprocess \nmay be \nrun \non \nthe \nserver \nusing \nsystem-\nadminist\nrator-\nlevel \nprivilege\ns. \nH (3) \nVulnera\nble CGI \nscript \ninstalled \non Web \nserver. \nL (1) \n4(3+1) \nThis risk analysis approach has now attempted to sort into a pecking order the various \nthreats that the system being tested may foreseeably be expected to face. This enables the \n"
        },
        {
            "page_number": 241,
            "text": " \n234 \nsecurity-testing team to identify the system's greatest exposures and therefore the areas \nthat the testing effort should focus on. \nIf created before a system is built, this list may also be used by the system's designers, to \nreview their planned implementation, paying particular attention to ensuring that any highly \ncritical threats are not solely protected by a single security measure (or single point of \nfailure). \nIdentify and Assign Candidate Tests \nOnce an objective assessment of the relative exposures that threaten a system's security \nhas been determined, it becomes much easier to decide which vulnerabilities should ideally \nbe tested for first. Tests can be designed to check for specific vulnerabilities and scheduled \nfor execution depending upon the criticality of the vulnerability (or vulnerabilities) that the \ntest is designed to expose. \nIt is possible that some threats will be ranked so low that they may not warrant any tests \nbeing assigned at all. Even if this is the case, the threat has still been identified, evaluated, \nand documented, enabling a future risk analysis to revisit any assumptions made and \npossibly change them in light of new information. \nPriority Modifiers \nThe preceding steps are intended to help the security-testing team draft an outline of their \nsecurity-testing schedule, the key word being draft. The output from this approach should \nnot be regarded as the final say on which tests should be executed first, but rather a \nstarting point from which other criteria can be applied. Some of the other factors that should \nbe taken into account before finalizing the testing schedule (and will vary in importance \nfrom project to project) include the following: \nƒ \nTest cost and ease of implementation. Some tests are extremely easy and cheap \nto run-for example, externally launched port scans using a freeware tool-while other \ntests may be quite expensive in terms of monetary cost, elapsed time, staff \nresources, or political capital-for example, purchasing some of the most sophisticated \nsecurity assessment tools cost tens of thousands of dollars, and manually identifying \nand validating every service running on the hosts being assessed can be quite time \nconsuming. \nƒ \nTest dependencies. Some tests only make sense to run if a previous test has \nalready passed and should therefore be scheduled to occur after the former test is \nexecuted. For instance, there is little point in trying to run a brute-force, password-\ncracking algorithm against a product's password file if the vendor's default userIDs \nand passwords are found to be still active. \nƒ \nPretesting dependencies. It's quite probable that some components of the system \nwill be available for testing before others, typically a reflection of development \npriorities and dependencies. It would be nice if the development team factored in the \ntesting team's priorities when finalizing the development schedule, but unfortunately \nthis is not always the case. \nƒ \nPost-testing dependencies. If other (nontesting) project activities are dependent \nupon the results of a specific test, it may be more important to schedule that test early \nin the testing effort in order to help the project as a whole proceed more quickly. For \ninstance, finding out that an organization's standard operating system install is \nmissing a critical security patch is a lot more helpful before the organization deploys \n200 machines to 3 different locations than afterwards. \n"
        },
        {
            "page_number": 242,
            "text": " \n235 \nƒ \nTest coverage. Some tests can be used to mitigate multiple exploits. For example, \nrunning a security assessment tool may check for the existence of numerous \nvulnerabilities. \nƒ \nScheduling conflicts. It could be that the DBA is going on vacation next week. \nExecuting tests that focus on the database during this week may be less productive \nthan waiting for the DBA to return. \nƒ \nFragile application code. Some parts of the application may be noticeably more \ncomplex, have more interfaces, have recently been significantly modified, or have a \nhistory of excessive defects. All these traits typically indicate that the application code \nis likely to be prone to future defects (functional or security related) and should \ntherefore also be considered for more extensive testing. \nƒ \nRegression tests. These tests consist of reviewing the test logs of earlier security \nassessments for security holes that were previously identified and may not have been \nadequately fixed. \nƒ \nPrevious history. If this consideration wasn't factored in earlier in the analysis, then \nmemories of recent attacks, especially accidental intrusions by employees, should be \nconsidered. \nƒ \nSenior management's preferences. For many security-testing teams, it's an \nunfortunate fact of life that testing priorities will be influenced to some degree by \nsenior management's willingness to invest in certain types of testing more than \nothers. For example, the CIO may be quite willing to write a one-time check for an \nexternal-penetration-testing firm to try and break into the organization's internal \nnetwork, but be less willing to sanction tests designed to find vulnerabilities potentially \navailable to internal employees, because \"We trust everybody here.\" Although a well-\ndocumented risk analysis may help educate senior management and thereby reduce \nany irrational interference, this factor may not be completely eliminated. \nƒ \nCommon sense. This process should be a helpful guideline, not an inflexible \nmandate etched in stone. \nTest Schedule \nOnce this set of factors has been evaluated and the test schedule has been updated to \nreflect these considerations, it becomes possible to determine the point where this phase of \nsecurity testing will theoretically come to a close. This is typically due to funding, project \ndeadlines, or some other project constraint, rather than the exhaustion of possible tests. \nUsed wisely, a prioritized test schedule with a cutoff point showing where the resources will \nrun out and which low-priority tests will therefore not be run can make a powerful ally when \nnegotiating with senior management for additional funding. \nIn addition, if the test schedule has been front-loaded to perform the most critical tests first, \nan unplanned reduction in the time or resources available to conduct the testing can be \nhandled by removing the less critical tests from the test schedule. This would not be \npossible if the tests' criticality was unknown or if the testing done to date had focused on the \neasy but less critical tests, which perhaps would be done in order to make a manager's \n\"tests completed\" metric look artificially better. Figure 10.4 summarizes all the steps \nperformed in this process. \n \n"
        },
        {
            "page_number": 243,
            "text": " \n236 \nFigure 10.4: Risk analysis process overview.  \nIf warranted, this style of risk analysis can be used as a starting point for conducting a more \nrigorous analysis, such as a full-fledged FMECA. \nFailure Mode, Effects, and Criticality Analysis (FMECA) \nThe Failure Mode, Effects, and Criticality Analysis (FMECA), or its abbreviated version, \nFailure Mode and Effects Analysis (FMEA), is a rigorous approach to quantifying and \nprioritizing the risks associated with a product. FMECA is extensively used by organizations \nwith extremely low risk tolerances (such as the military, space agencies, and manufactures \nof medical devices) as a systematic process for identifying potential design and process \nfailures before they occur, with the intent to eliminate them or minimize the risk. This \napproach can be adopted for analyzing the potential failure of a system's security defenses. \nTo quickly (if rather crudely) convert the information gathered by the risk analysis described \nin the previous section of this chapter to the format used by an FMECA study, you would \nperform the following steps: \n1. Replace the term likelihood with occurrence and use a scale of 1 to 10 (1 meaning \nnot likely to happen) instead of high, medium, and low. \n2. Replace the term business impact with severity and use a scale of 1 to 10 (1 \nmeaning negligible consequence) instead of high, medium, and low. \n3. Add an additional variable, detection (described later), to the matrix, using a scale of \n1 to 10 (1 meaning it is obvious and is therefore extremely likely to be detected and \n10 indicating that the current detection mechanisms are extremely unlikely to detect \nthis problem). \n4. Multiply the three variables together to produce a risk priority number (RPN). \nDetection (perhaps easier to think of as detectability) is an assessment of the likelihood that \nthe current set of preventive measures (design reviews, unit testing, and so on) will detect \n"
        },
        {
            "page_number": 244,
            "text": " \n237 \nthe vulnerability, thus preventing it from reaching the production environment. The theory \nhere is that it is more important to do additional testing for potential failures that are not \nlikely to be caught using existing detection measures than for those that will probably be \ndetected without further scrutiny. Unfortunately, in a security analysis, unless reliable \nmetrics have been collected previously, relying upon a gut feeling to estimate this value is \nunlikely to be very accurate. For instance, most security testers would be hard-pressed to \nestimate the probability of a misconfigured firewall rule being detected before it is placed \ninto service, or that a network-based intrusion detection system will not notice an intruder \ncommunicating to a compromised server. \nIf an objective measure of detection can be assigned to each vulnerability, then multiplying \nall three variables (occurrence, severity, and detection) together will generate a RPN, which \nmay range from 1 (low significance) to 1,000 (fire alarm). The vulnerabilities with the \nhighest score represent the greatest threats to the system and should therefore be given \nthe highest priority for risk mitigation or reduction. \nTo reduce the risk posed by each of these threats, an organization could implement one (or \nmore) of the following actions: \nƒ \nRework the design of the system to remove the threat. Unfortunately, it's not always \npossible to completely remove one threat without creating another. \n \nRISK PRIORITY NUMBERS (RPNs) \nƒ Interestingly, the RPN equation can generate only 120 of the numbers in the range \nof 1 through 1,000. For example, 10 × 10 × 9 would yield an RPN of 900, and \nbecause the next riskiest category of 10 × 10 × 10 would generate a RPN of \n1,000, there is no combination of values that could result in an RPN number \nbetween 901 and 999 (inclusive). Fortunately, 120 different priority categorizations \nshould prove sufficient for any risk analysis (Source: www.fmeca.com). \n \n \nƒ \nReduce the likelihood of a failure (exploit) occurring. For instance, encrypting data \nbeing transmitted over the Internet or a LAN would reduce the probability of an \nexternal or internal eavesdropper acquiring sensitive information. \nƒ \nReduce the impact on the business should the failure occur. For instance, moving a \ndatabase off the Web server would reduce the business impact of the Web server \nbeing compromised, or developing and testing a contingency plan that utilized a \nbackup site could reduce the amount of time a compromised Web site was \nunavailable. A variation of this mitigation strategy is for an organization to transfer the \nrisk to another entity. For instance, purchasing a cybercrime insurance policy would \nprovide an organization with compensation to help reduce the impact of a successful \nattack. \nƒ \nImprove the organization or system's detection capabilities. An example of this would \nbe designing and executing additional security tests, or installing an intruder detection \nsystem on a critical server. \nTypically, a combination of these solutions will be employed. Once the remedies have been \nsuccessfully implemented, the RPN equation can be recalculated to identify the new riskiest \nthreats and thereby enable continuous system and process improvement. McDermott et al. \n(1996), Stamatis (1995), the Haviland Consulting Group (www.fmeca.com), and Relex \nSoftware (www.relexsoftware.com) all provide additional information on the FMECA and \nFMEA techniques, while Relex Software (www.relexsoftware.com) and SkyMark \n(www.skymark.com) both offer tools (Relex and PathMaker, respectively) designed to \nsupport FMECA and FMEA deliverables. \n"
        },
        {
            "page_number": 245,
            "text": " \n238 \n \nSummary \nIn a well-funded security-testing project, without the prospect of truncating the period of \ntime available for testing and consequently guaranteeing enough time to perform all the \nplanned tests, the need to do a risk analysis for the purpose of scheduling tests will be less \ncritical. Unfortunately, this is an unlikely scenario for the majority of security-testing teams, \nwho must instead prudently decide how to best expend their limited resources in order to \nmaximize the reduction in an organization's risk exposure. A risk analysis can go a long \nway toward facilitating a smart test schedule that gives testing priority to tests that are \nintended to detect vulnerabilities that pose the greatest risk to the organization early on in \nthe testing effort and thereby ensure that the high-priority tests stand the best chance of \nbeing executed. \nAdditionally, using a systematic approach, such as a risk analysis, to put together a test \nschedule has the added benefit of providing the testing team with test traceability \n(answering questions such as \"Why was this specific test needed?\", \"How did you test for X \nvulnerability?\" and \"Why didn't you check for Y?\"). Such an approach may allow for better \ntest coverage estimates, which are useful things to have if the testing team is required to \nshow due diligence. \n"
        },
        {
            "page_number": 246,
            "text": " \n239 \nEpilogue \nOne thing that is certain about the future of Web security is that it will always be changing. If \nhistory is any indicator of the future, as soon as a technology matures enough for its \nweaknesses to become fully understood and avoidable, the technology will either mutate or \nbe replaced by another newer technology. This metamorphosis can potentially create a \nwhole new set of vulnerabilities waiting to be discovered and subsequently plugged. \nAt the time of this writing, the next anticipated technology shift is that of Web services. It still \nremains to be seen how secure Microsoft's .NET or Sun Microsystems's open network \nenvironment (Sun ONE) implementations will ultimately be, especially when implemented \nover a wireless network. When you consider how open and accessible this technology is \ndesigned to be, it doesn't take much creativity to imagine how an attacker could try to utilize \nit for some malicious purpose. \nFor example, many organizations are considering swapping their electronic data \ninterchange (EDI) connections for simple object address protocol (SOAP) implementations. \nEffectively, this will replace cryptic and proprietary data formats transmitted over private \nnetworks with, self-describing industry-standard data formats, such as Extensible Markup \nLanguage (XML) and XML schemas, over the public Internet. Although one of the greatest \nbenefits of using a standard format is that it simplifies the exchange of data between \nbusiness partners, one of the greatest disadvantages is that an eavesdropper can easily \nintercept and (unless encrypted) interpret the data. Such an exploit is also made \nincreasingly easier to perform with the continued, widespread adoption of wireless \nnetworks. \nThe intent of Universal Description, Discovery, and Integration (UDDI) online registries is to \nenable organizations to advertise the Web services they offer and to publish the technical \nspecifications needed to connect to the applications that provide these services. However, it \nremains to be seen whether or not attackers will be able to utilize information specified in an \napplication's associated UDDI registry entry to locate and gain unauthorized access to \nthese applications. \nAs Web-service-related vulnerabilities begin to surface (such as CERT notification \nVU#736923, which relates to Oracles9iAS's implementation of SOAP), relying solely upon \nthe good old firewall to protect an organization's Web services would be a precarious \ndecision. One of the reasons why SOAP was designed to work using Hypertext Transfer \nProtocol (HTTP) is because HTTP network traffic is comparatively unfettered by many \ncurrent firewall implementations, thereby making its deployment much easier. Instead, an \norganization would be well advised to consider a defense-in-depth approach for protecting \nits electronic assets and execute an associated testing strategy designed to validate that all \nthese defenses have been correctly implemented (and maintained) using this new \ntechnology. \n"
        },
        {
            "page_number": 247,
            "text": " \n240 \nPart V: Appendixes \nAppendix List \nAppendix A: An Overview of Network Protocols, Addresses, and Devices  \nAppendix B: SANS Institute Top 20 Critical Internet Security Vulnerabilities  \nAppendix C: Test-Deliverable Templates  \n"
        },
        {
            "page_number": 248,
            "text": " \n241 \nAppendix A: An Overview of Network Protocols, \nAddresses, and Devices \nFor the benefit of readers who may not be too familiar with the networking terms used in \nthis book, this appendix provides additional background on the network protocols (and \nassociated network addressing issues) commonly used by Web applications, and the \nnetwork devices typically found on networks used to host a Web site. Lierley (2001), \nNorthcutt et al. (2000), and Skoudis (2001) provide additional introductory explanations of \nnetworking concepts geared towards the security tester. \nNetwork Protocols \nTo reduce network design complexity, the vast majority of modern networks use not one, \nbut several network protocols to facilitate communication between two network devices. \nThese protocols are organized as a series of layers (or levels), each layer building upon the \nfunctionality and capabilities offered by the previous layer. Conceptually, this collection of \nlayers is often referred to as a stack of network protocols (or simply the stack). An analogy \nfrom the desktop PC world would be the way in which a user develops MS Office macros. \nHe or she utilizes the functionality of MS Office, which in turn makes system calls to an \noperating system, which itself utilizes the desktop's basic input/output system (BIOS) to \naccess hardware devices, such as a hard drive. \nThe number, names, and purposes of each network layer varies from network architecture \nto network architecture, which can make interfacing networks that use different protocols \nproblematic. In 1980, in an effort to try and make network protocols more standard and \nthereby easier to integrate, the International Organization for Standardization (ISO, \nwww.iso.ch) proposed the Open Systems Interconnection (OSI) model. This model \nattempted to group the numerous tasks that needed to be performed on a network into \nseven reference layers, each layer being responsible for a set of specific functions. Ideally, \neach network protocol would be designed to perform all the functions that belong to a single \nlayer (and no more). Protocols at the same layer would then (theoretically) become \ninterchangeable, but they would not repeat any of the functionality being handled by \nanother layer of this network stack, thereby improving network performance and reducing \nthe complexity of each layer. \nUnfortunately, some of the network protocols in use today date back to before 1980 and \nwere therefore not originally designed with the OSI model in mind. In some cases, a \nprotocol's owner (a proprietary vendor or industry committee) has tried to retrofit the \nprotocol into one of the seven OSI layers. As can be seen from the comparison in Table \nA.1, the network model used by the Internet differs from the OSI model in that it \nconceptually only defines four separate network protocol layers, instead of the seven \ndefined by the OSI model. \n \nTable A.1: Internet-to-OSI Network Model Comparison  \nINTERNET \nNAME  \nLAYER  \nOSI NAME  \nEXAMPLE PROTOCOLS  \nApplication \n7 \nApplication \nBOOTP, FTP, HTTP, DNS, LPD, NFS, S-\nHTTP, SET, SMTP, SNMP, TelNet, and \nTFTP \n  \n6 \nPresentation \nASCII, AVI, EDCDIC, GIF, JPEG, and \n"
        },
        {
            "page_number": 249,
            "text": " \n242 \nTable A.1: Internet-to-OSI Network Model Comparison  \nINTERNET \nNAME  \nLAYER  \nOSI NAME  \nEXAMPLE PROTOCOLS  \nMPEG \n  \n5 \nSession \nSecure Socket Layer (SSL) \nHost-to-\nhost \n4 \nTransport \nTransmission Control Protocol (TCP) or \nUser Datagram Protocol (UDP) \nInternet \n3 \nNetwork \nInternet Protocol (IP), Internet Protocol \nSecurity \n(IPsec), \nInternet \nControl \nMessage \nProtocol \n(ICMP), \nInternet \nGroup Membership Protocol (IGMP), \nAddress Resolution Protocol (ARP), and \nReverse Address Resolution Protocol \n(RARP) \nNetwork \nAccess \n2 \nData Link \nARCNET, \nAsynchronous \nTransfer \nMethod (ATM), Compressed Serial Line \nInternet \nProtocol \n(CSLIP), \nDigital \nSubscriber Line, (DSL), Ethernet, Frame \nRelay, \nIntegrated \nServices \nDigital \nNetwork (ISDN), Point-to-Point Protocol \n(PPP), Serial Line Internet Protocol \n(SLIP), Token Bus, and Token Ring \n  \n1 \nPhysical \nIEEE 802.x, RS232, V.32, V.34, and V.90 \nThe following sections provide a brief summary of what each OSI layer is intended to \naccomplish. \nApplication Layer \nThe application layer is where all the programs that actually use the network typically \nreside, the Web being just one of the network applications that functions at this layer. Since \nthis layer is at the top of the stack, its functionality is open-ended, permitting applications to \noffer services that range from supporting distributed databases to broadcasting live \ntelevision shows. \nPresentation Layer \nThe presentation layer is charged with transforming data into a format that an application \nlayer program can understand. In effect, the presentation layer provides a translation \nservice when the two machines that are trying to communicate with each other are using \ndifferent data formats, such as EBCDIC and ASCII. Additionally, this layer may be enabled \nto compress large volumes of data and/or encrypt sensitive information. \n \nHTTP \nThe hypertext transfer protocol (HTTP) is the application-layer network protocol that is used \nto transfer Web content between a Web server and its client (typically a browser). \ninterestingly, a client can send information in more than one way to a Web server using \nHTTP. The following describes the two most common methods that are used for this \npurpose. \n"
        },
        {
            "page_number": 250,
            "text": " \n243 \nƒ Get. Initially conceived as a method to get information from a Web server, in the Get \nmethod, any associated data is added to the end of the URL and is transmitted as \npart of the Query component (note the ? in the following example). Aside from the \nfact that this information may be truncated when transmitted, it may also be readily \nseen by anyone viewing the client screen, reviewing their browser's history, or \nexamining the target Web site's logs. Additionally, when the visitor leaves the Web \nsite, this URL (with its embedded data) may be carried over to the next Web site \nwhere it can be viewed via the new Web site's logs (this is true regardless of \nwhether or not the communication is encrypted). For example, the following URL's \nblatant exposure of the userID and password would make it easy for someone \nelse to acquire this particular userID and password combination. \nƒ https://www.wiley.com/cgi/login.cgi?uid=888753369&pass=888424749  \nƒ Post. Initially conceived as a method for posting information to a Web server, Post is \nsimilar to Get but embeds any input information (such as data from a HTML form) \ninto the HTTP body, rather than into the URL. Post thereby avoids the Web log \nand screen-viewing issues associated with the Get command, which is in part why \nthe Web's governing body, the World Wide Web Consortium (W3C, \nwww.w3c.org), recommends using Post over Get whenever possible. \nSession Layer \nThe session layer opens a dialog (or session) between the sending and receiving \nmachines. It accomplishes this by using three steps: connection establishment, data \ntransfer, and connection release. Once the session has been established and data transfer \nhas begun, the data can be passed to the presentation layer, where it can potentially be \nreformatted to make it ready for the application that is waiting to receive it. \nThe secure sockets layer (SSL) is typically considered to be a session layer in the OSI \nmodel, as it utilizes the functionality provided by the transport layer protocols to enhance \nthe service available to the higher-level application layer protocols. \nTransport Layer \nThe transport layer accepts data from the session layer and then chops it up into small \nchunks of information (typically known as datagrams) that can easily be sent over the \nnetwork. Conversely, the transport layer is responsible for reassembling the datagrams that \nit receives, requesting replacements for any that appear to have gotten lost somewhere on \ntheir journey across the network. \nThe Internet uses two layer 4 protocols: the Transmission Control Protocol (TCP) and the \nUser Datagram Protocol (UDP). The TCP protocol attempts to guarantee that any data sent \nover a network via TCP will ultimately get to its intended destination. It does this at a \nsimplistic level by requesting that the recipient of the data send a confirmation that it has \nactually received the data (analogous to requesting a return receipt from the postal service \nwhen it delivers an item of mail). If no confirmation is received after a predefined period of \ntime, the sender assumes the data was lost and resends it. In comparison, UDP is regarded \nas a connectionless protocol, which means that it effectively sends and forgets, offering no \nguarantee that the data will actually arrive at its intended destination. Such an approach \ntypically allows data to be sent faster, since no confirmation messages are being sent or \nwaited upon. UDP is typically used for streaming video and audio, where losing a few \npieces of data here and there is a small price to pay for increased network speed. Web \npages, on the other hand, typically use TCP as a means of ensuring that the entire page is \nreceived correctly. \n"
        },
        {
            "page_number": 251,
            "text": " \n244 \nNetwork Layer \nThe network layer handles the task of actually conveying network packets across the \nnetwork, resolving issues such as converting logical network addresses (that are easy for \nhumans to remember) into physical addresses, or determining the route a data packet \nshould take across the network. \nThe layer 3 protocol used by the Internet is called the Internet Protocol (IP). Its specification \ncan be traced back to a 1960s U.S. Department of Defense network research project called \nARPANET. For a computer to be directly connected to the Internet, it must use IP as its \nlayer 3 (network) protocol. However, it may potentially utilize the services of any of the \nlower-layer protocols (data link and physical) in order to support the various application \nlayer protocols that it in turn must support. Figure A.1 illustrates how an Internet data \npacket might be composed of data from each of the various layers of a network protocol \nstack. \n \n \nFigure A.1: Internet data packet composition.  \nData Link Layer \nThe data link layer organizes the data packets into a series of data frames, transmitting \neach frame sequentially between two network devices. It is this layer of the network model \nthat is expected to correct errors caused by network noise (interference that causes some \nor the entire network frame to be lost) and negotiate a mutually acceptable transmission \nspeed that both network devices can handle. \nPhysical Layer \nAt the physical layer, data is represented as electronic bits in the form of zeros and ones. \nLayer 1 protocols specify such fundamental communication requirements as when a zero is \na zero and when a one is a one (or even what the difference is between two ones back to \nback, and a single one). This layer of the model to a large degree will be molded by the \nphysical medium used to convey the data signal (for example, copper wire versus \nairwaves). \nSecurity-Minded Network Protocols \nSeveral different network protocols have been specifically designed to provide secure \ncommunication over the Internet (or intranet) via encryption. Secure Hypertext Transfer \nProtocol (S-HTTP), SSL, and IP Security (IPsec) are three such protocols. The primary \ndifference between these three protocols is the network layer at which that they work. \n"
        },
        {
            "page_number": 252,
            "text": " \n245 \nS-HTTP is a superset of HTTP and is designed to send individual messages securely by \nencrypting and decrypting the message at the OSI application layer (layer 7). SSL is \ndesigned to establish a secure connection between two machines, thereby protecting all \ncommunications (not just HTTP-based messages). It achieves this by working down the \nnetwork stack, nearer to the bits and bytes, encrypting and decrypting the data at the OSI \nsession layer (layer 5). IPsec is a standard for security at the network packet-processing \nlayer or network communication layer (layer 3 of the OSI model). IPsec actually provides \ntwo choices of security service: authentication header (AH), which essentially enables \nauthentication of the sender and integrity (but not encryption) of the data, and \nencapsulating security payload (ESP), which supports both authentication of the sender \nand encryption of the data as well. \nSince each encryption method works at a different network layer and has slightly different \ngoals, in theory it will be possible to simultaneously use all three to provide extremely \nsecure transmissions. However, in practice the three methods are not equally well \nsupported. First introduced by Netscape in its 2.x-generation browser, SSL has since been \nimplemented in all the leading browsers and Web servers, as well as many other Web-\nenabled supporting tools. In contrast, S-HTTP is scarcely supported, and the Internet \nEngineering Task Force (IETF, www.ietg.org) is still working on IPsec. When you also take \ninto consideration that each encryption and decryption increases network latency and \nplaces a heavier resource requirement on a CPU, it's understandable that a Web site \ndesigner would not want to use an over-engineered, multitiered encryption strategy that has \nissues communicating with some of its clients. This is why most Web sites today that want \nto transmit data securely rely upon SSL to do their encryption work directly, or via an \napplication-layer secure protocol that builds upon the functionality of SSL. An example of \nthis would be Secure Electronic Transaction (SET), a protocol designed to ensure the \nsecurity of financial transactions over the Internet. Additional information on SET can be \nfound at www.setco.org. Table A.2 lists some vendors that provide more background \ninformation on encrypting network traffic. \n \nTable A.2: Vendors Offering Encryption Solutions  \nNAME  \nASSOCIATED WEB SITE  \nBaltimore Technologies \nwww.baltimore.com  \nEntrust \nwww.entrust.com  \nMicrosoft \nwww.microsoft.com  \nRSA Security \nwww.rsa.com  \nValiCert \nwww.valicert.com  \nVeriSign \nwww.verisign.com  \n \nNetwork Addresses \nSurprising as it may seem to many, each network device will typically have more than one \nnetwork address (the reasons for this situation are many and to a large extent historical). At \nthe most basic network layer (or bottom of the network stack), the media access control \n(MAC) address (such as aa-bb-cc-dd-ee-ff) enables devices to find and communicate to \neach other (over relatively short distances) using the Ethernet protocol. The IP network \naddress (such as 123.456.789.123) permits messages to be reliably sent across multiple \nnetworks (including, of course, the Internet) using IP. Finally, a device's host name (such as \n"
        },
        {
            "page_number": 253,
            "text": " \n246 \nweb1.tampa) enables the machine to be more easily identifiable to humans and hence \neasier to remember and administrate. \nBinding is a term used to describe the process of converting one network address into \nanother. In the case of IP-to-MAC address conversions, this binding is performed using a \nnetwork utility called the Address Resolution Protocol (ARP), while an IP address to host \nname conversion is typically performed using a request to a DNS server. \nThe following sections describe some of the network addressing scenarios that a testing \nteam may encounter (and could possibly cause them confusion) while trying to build (or \nverify) an inventory of network addresses used by a Web site (often an initial step in \nconducting a security assessment). \nDynamic IP Addresses \nIn a dynamically assigned IP configuration (or a Dynamic Host Configuration Protocol \n[DHCP] implementation), as each network device is powered up, it requests and receives \nthe IP address that it is expected to use while it is up and running. This address may or may \nnot be the same IP address that it had the last time it was assigned an address and may \ntherefore confuse an unsuspecting member of the testing team. \nOne of the reasons an organization may choose to use DHCP is to reduce the amount of \nadministration needed to maintain a network, because it allows each DHCP device to \nautomatically configure itself based on the values set in the DHCP controller. Although it's \nquite common for network administrators to assign IP addresses to desktop machines \ndynamically, it's much less common for servers, routers, and other server-orientated \nnetwork devices (which are typically the focus of a security assessment) to have their IP \naddresses dynamically assigned. \nPrivate IP Addresses \nRather than assign each network device an IP address that is universally unique (that is, no \nother device anywhere in the world has this IP address), many network administrators will \nconfigure their network devices which do not have direct access to the outside world to use \nIP addresses that are only unique on the network they reside upon. The Internet Assigned \nNumbers Authority (www.iana.org), the governing body that administers IP addresses, has \nset aside specific ranges of IP addresses for this purpose (such as the IP address range \n10.0.0.0 to 10.255.255.255). This technique works because each of the gateways to the \noutside world can be configured to dynamically translate these nonunique (or private) IP \naddresses into globally unique ones, a technique normally referred to as network address \ntranslation (NAT). \nMultiple IP Addresses \nAlthough most network devices will typically only have a single IP address, some network \nadministrators will assign multiple IP addresses to the same device. This can be done by \neither assigning multiple IP addresses to a single network interface card (NIC) or by \ninstalling multiple NICs and assigning each NIC one or more IP addresses. \nIP-less Devices \nIt's possible that one or more of the devices connected to the network will not have an IP \naddress. For example, a network-based intrusion detection system (IDS) appliance may not \nuse an IP address. It may be impossible for an intruder to compromise these machines due \nto his or her inability to communicate with them remotely. However, these machines could \n"
        },
        {
            "page_number": 254,
            "text": " \n247 \nstill be compromised by an intruder walking up to the device and using its physical \nconnection to the network for eavesdropping or other subversive purposes. \nMisdirecting Host Names \nDeducing the physical location of a network device based on its host name will be easier in \nsome organizations than in others. For instance, a network address naming standard that \nembeds the physical location of a device (together with its primary function) into its host \nname provides an easy way of identifying exactly where the device is located and its \nprimary purpose. For example, the Apache Web server in room 525, which has been \nassigned a host name of apache525, will prove easier to track down than an ad hoc \nstandard. An example of such a standard would be a collection of servers named \nncc1701a, ncc1701b, and ncc1701c (the network administrator who installed these servers \nis a Star Trek fan and named each server after the registration of a different version of the \nstarship Enterprise). \nOf course, making the network easier to comprehend for a legitimate network user may \nalso make life easier for an intruder trying to figure out what each machine does. Therefore, \na trade-off takes place between the ease of administration and an additional level of \nsecurity by obscurity. One strategy that tries to take advantage of both approaches is to use \na standard naming scheme but employ a misdirection strategy. For example, an Apache \nWeb server would have the text IIS embedded within its host name, and an MS SQL Server \ndatabase server would be named something like oracleprime. \n \nNetwork Devices \nThe modern network is often compromised of many faceless boxes. To the casual user of a \nnetwork, these boxes may seem to be rather mysterious in nature and their purpose may \nnot necessarily be obvious. This section will endeavor to provide a short explanation of \nwhat each of these boxes does. It is intended to be a convenient reference, not a definitive, \nall-encompassing set of definitions. For such detailed descriptions, the reader should \nconsider references such as one or more of the Web-based computer technology \nencyclopedias (such as www.webopedia.com or www.whatis.com). \nRepeater \nA repeater is a network device that can be used to extend the effective distance that data \ncan be sent over a network connection. The repeater being used to boost a signal that has \nbecome weak because of the physical distance it has already traveled before reaching the \nrepeater. \nHub \nHubs can be thought of as multiport repeaters. Instead of receiving and resending data \nbetween just two other network devices, a hub can be used to broadcast a message to \nseveral network devices that have been connected to a communal hub. Hubs are simple to \nimplement but tend not to scale well because as data communication rates increase, so do \ndata collisions and retransmit rates. \nBridge \nA bridge is a network device that is typically used to connect two network segments \ntogether. The bridge blocks network traffic that is internal to one network segment from \nbeing forwarded (repeated) into the other network segment that the bridge is connected to. \n"
        },
        {
            "page_number": 255,
            "text": " \n248 \nHoweever, a bridge will permit network traffic destined for a device located on another \nnetwork segment to pass through the bridge to the other network segment. \nBridges can be used to segment a large network into several smaller ones, reducing the \namount of network traffic collisions due to high utilization, while still enabling each network \ndevice to communicate to any other network device on any of the segments. \nGateway \nA gateway is a network device that acts as an entrance to another network. Gateways are \noften used to restrict unauthorized traffic from passing through and/or for translating data \nfrom one network protocol to another (for example, from Ethernet to Token Bus). \n \n \nSNIFFING SWITCHED NETWORKS \nAlthough some advanced mechanisms can be used for sniffing switched networks, they are \nmuch harder for an attacker to use than eavesdropping on a simple hub-based network. \nTypical drawbacks to a switched network are the additional network administration \ncomplexity needed to implement the network and a potentially higher network hardware \nexpense. \nSwitch (Switching Hub) \nSwitches differ from hubs in that they only resend (pass on) data to the data packets' \nultimate destination. Other machines connected to the network are unlikely to hear the data \nbeing passed back and forth by two other network devices connected to the same network \n(a switch can be crudely thought of as a combination of a hub and a bridge). In addition to \noffering better performance, this approach also makes network eavesdropping (sniffing) \nless feasible, as the amount of data available for the eavesdropper to sniff is greatly \nreduced. \nRouter \nA router is a network device that is used to connect two or more networks. Unlike many \nother network devices, the router tries to make intelligent decisions about which way to \nforward network traffic that it has received (rather than broadcasting it to every possible \ndestination). The forwarding information is maintained in a routing table and is often \nupdated dynamically to include the latest information on the health of the various networks \nthe router is connected to as well as the other distant networks connected to the ones the \nrouter is immediately adjacent to. \nBrouter \nA brouter is a network device that combines the features of a bridge with that of a router. \nNetwork Controller \nNetwork controller is the name given to the device (or devices) that performs network \nadministration tasks, such as detecting the presence of a new network device being added \nto the network or validating that a userID and password combination entered from any of \nthe network devices it controls is valid. \n"
        },
        {
            "page_number": 256,
            "text": " \n249 \nLoad Balancer \nA network device is designed to spread the work of processing large amounts of network \nrequests across two or more machines. Numerous strategies can be used to determine \nwhich request is sent to which specific destination (such as round-robin, IP multiplexing, \nand Domain Name System [DNS] redirection), each with its own set of pros and cons. \nServers \nServers typically run general-purpose operating systems (such as some variation of \nWindows or UNIX) and are loaded with one or more network-aware applications (although \ntechnically speaking, a server refers to any device that provides a service of some kind to \nanother device, which is the client). Although such a device may be able to simultaneously \nrun numerous different applications, only using a server to support a single application \nallows the underlying operating system to be tuned to provide better performance and \ntighter security. Examples of servers include the following. \nApplication Server \nAn application server is dedicated to running applications in a batch-like manner, typically \nany user-interaction component of the application is handled by another component of the \napplication. Some application servers utilize special environmental software designed to \nshare resources more effectively across numerous, small applications such as Enterprise \nJavaBean (EJB) implementations. \nDatabase Server \nFor many Web applications, the database access component of a Web transaction is often \nthe most resource-intensive and time-consuming portion of the request to fulfill. For this \nreason alone, many Web applications are architected with the database installed on a \ndedicated machine, a server that has been specifically tuned to handle database \nprocessing. \nDNS Server \nDNS is a network address translation program that converts domain names (such as \nwww.wiley.com) into IP addresses (such as 123.456.789.123) and vice versa. This program \nmay be installed on the device used as the network controller or on a separate dedicated \nmachine. \nFile and Print Server \nFile servers provide network storage facilities that can be easily shared by all the network \nusers, while print servers enable groups of users to share a scarce resource, such as a \nhigh-end color printer. \nFile Transfer Protocol (FTP) Server \nThe File Transfer Protocol (FTP) was designed and optimized to transfer large files across \na wide area network (WAN) as quickly as possible without losing any data on the way. \nStandalone FTP servers are often used to isolate irregular network bandwidth utilizations \n(due to the large files typically associated with this application). They also are used to \nmitigate any security hole inadvertently left open that an attacker might be able to exploit by \nuploading a toolkit or downloading a sensitive file using FTP. \n"
        },
        {
            "page_number": 257,
            "text": " \n250 \nWeb Server \nA Web server uses the HTTP network protocol to serve up content to requesting users \n(typically via a browser). \n \nFirewalls \nFrom the security perspective, firewalls are perhaps the most exposed device on a network. \nFor this reason, this specific network device will be expanded upon in much greater detail. \nA firewall is a software program or hardware appliance that typically resides at the edge of \na network or network segment and is used to protect resources located on the network. Not \nall firewalls are created equal; a firewall can be implemented in several different ways (the \nvendors of each approach typically claim their tool's approach is superior to their \ncompetitors). \nFirewall Types \nFirewalls can be grouped into two main categories: those that examine network traffic at the \napplication layer of the OSI network model (layer 7) and those that work at the lower \nnetwork layer (layer 3). \nApplication-Layer Firewalls \nApplication-layer firewalls (also known as application proxies, application forwarders, \napplication gateways, or circuit-layer gateways) run on top of a general-purpose operating \nsystem such as UNIX or Windows. These firewalls base their decision of whether or not to \npermit or deny an entire group of network packets (which together comprise a single \nnetwork message) to pass based on predefined security policies (typically established via a \nuser-friendly software application). Application-layer firewalls can also maintain elaborate \nlogging and auditing information on the traffic passing through them, and they are \nconsidered by many to be the easiest firewall to maintain, which in part is why their \nadvocates consider them to be the most secure. \nNetwork-Layer Firewalls \nNetwork-layer firewalls (also known as packet filters) work at a more granular level than \napplication-layer firewalls, typically examining each network packet in isolation. Each \nindividual network packet is typically dropped, rejected, or approved based on comparing \nthe source network address, destination network address, requested port number, or \nnetwork protocol used, with a predefined set of security rules. \nDynamic packet filters (also known as stateful filters) are a more recent variation of network \nlayer firewalls. Rather than relying on fixed (static) firewall openings, openings are \ndynamically created and closed for groups of packets based on the header information \ncontained in the data packets. Once a series of packets has passed through the opening to \nits destination, the firewall closes the opening. \nNetwork-layer firewall vendors claim that their approach allows for faster communication \nthan an application-layer firewall, because the firewall's software is running at a lower \nnetwork layer and therefore consumes a lower overhead. In the case of static packet \nfiltering, the firewall can make its own decision of whether or not to forward a packet as \nsoon as the packet has been received. It doesn't have to wait until an entire sequence of \npackets has been received (as is the case with application-layer firewalls) and therefore \nreduces the network latency for this network hop. \n"
        },
        {
            "page_number": 258,
            "text": " \n251 \nUnfortunately (from a categorization perspective), vendors are starting to incorporate \nseveral different strategies into the same product, blurring the lines between the capabilities \nof application-layer and network-layer firewalls. \nWhen you factor in the cost of purchasing, installing, and maintaining these firewalls, it's not \nobvious which approach is best. Best may be based on purchase price (some software \nfirewalls are available free of charge), performance, ease of use, and, of course, robustness \nto attack. Because there is not an outright winner, some organizations choose to implement \none or more different types or brands of firewall on the same network. Table A.3 lists some \nsample firewalls. \n \nTable A.3: Sample List of Firewalls  \nNAME  \nASSOCIATED WEB SITE  \nBorderManager \nwww.novell.com  \nDI-701 \nwww.dlink.com  \ne-Gap \nwww.whalecommunications.com  \nFirebox \nwww.watchguard.com  \nFirewall \nwww.netmax.com  \nFirewall-1 \nwww.checkpoint.com  \nFirewall Server \nwww.borderware.com  \nFortiGate \nwww.fortinet.com  \nGNATBox \nwww.gnatbox.com  \nGuardian \nwww.netguard.com  \nLinux Netfilter \nwww.netfilter.org  \nMS ISA Server (Proxy Server) \nwww.microsoft.com  \nNetGap \nwww.sphd.com  \nNetScreen \nwww.netscreen.com  \nPIX \nwww.cisco.com  \nSecureWay Firewall \nwww.ibm.com  \nSidewinder \nwww.securecomputing.com  \nSonicWALL \nwww.sonicwall.com  \nStoneGate \nwww.stonesoft.com  \nSuperStack 3 \nwww.3com.com  \nVelociRaptor \nwww.symantec.com  \nVPN Firewall \nwww.lucent.com  \nZoneAlarm \nwww.zonealarm.com  \nNote: A more extensive list of firewalls can be found at www.firewall.com. \n"
        },
        {
            "page_number": 259,
            "text": " \n252 \nUpstream Firewalls \nIt may be possible to work with an organization's Internet service provider (ISP) to \nimplement some network-filtering (firewall) rules, not just at the organization's own \nperimeter firewall(s), but also duplicated at the ISP's network routers that relay network \ntraffic to the organization, effectively creating an additional upstream firewall. For example, \nthe ISP router could be configured to drop all ICMP requests (pings) originating from the \noutside world (external ICMP requests are frequently used by attackers, but rarely used for \nlegitimate purposes). \nDownstream Firewalls \nA machine such as a Web server located just behind a perimeter firewall could be \nconfigured to act as a downstream firewall. This could be achieved by configuring the Web \nserver to perform a reverse DNS lookup. A reverse DNS lookup is a process where the \nsource IP address of any incoming data is compared to the Web site's own DNS entries to \nsee if the domain name resolved by the Web site matches the one contained in the \nincoming message. A discrepancy between the packet's alleged domain name and the \nlocally derived one may indicate a spoofer at work, resulting in the message being \ndiscarded. Unfortunately, DNS lookups take time and slow down the performance of the \nWeb server, once again creating a trade-off between being faster and being slightly more \nsecure. \nFirewall Configurations \nThe simplest firewall configuration is to place everything behind a single-perimeter firewall. \nUnfortunately, a network's security is only as good as its weakest link, so many network \nadministrators will place applications particularly prone to compromise or subversion on a \nquarantined segment of the network. Web servers are considered by many to be the most \nvulnerable of all network devices, in part because they are designed with the intention of \ndispersing information to anyone and everyone. For this reason, Web servers are \nsometimes placed on the outward side of a firewall, isolated from the more valuable \nresources, consequently making these resources more secure, but making the Web server \nmuch more exposed. Figure A.2 depicts this sacrificial lamb approach. \n \n \nFigure A.2: Sacrificial lamb configuration.  \nThe theory behind the sacrificial lamb approach is that the core network is made safer by \nplacing the Web server outside the firewall, and the rules that the firewall can now impose \nare much stricter and therefore more secure. In the event that the Web server is \ncompromised by an intruder (perhaps through a recently discovered vulnerability in the \n"
        },
        {
            "page_number": 260,
            "text": " \n253 \nWeb service that the webmaster has not had a chance to fix yet), the amount of damage \nthat the intruder could inflict on the organization would be relatively minor compared to what \nthe intruder could do if he or she had compromised a Web server sitting behind the firewall. \nFor example, if the Web server is only displaying static information, the hard drive(s) of a \ncompromised Web server could be swapped out for a clean set and then restarted, a much \neasier task than having to try to recover a corrupted database server or retrieving a stolen \ndocument. Unfortunately, the word relative, is just that, relative. The intruder could use the \ncaptured Web server to launch an attack against another Web site or upload some \nembarrassing graffiti-style graphics that the organization's legitimate visitors see and \nremember for a long time to come. \n \nDUAL NICS \nAn additional step that some network administrators use to make the core network a little \nmore secure is to add a second network interface card (NIC) to the servers inside the DMZ. \nThis gives each DMZ machine one network address for its outward bound (and therefore \nvery public) network communications and a second (more private) network address for \ncommunicating to machines residing on the internal network, creating one more hurdle for \nan intruder to overcome. \nDual-Firewall Demilitarized Zone \nA more sophisticated (and popular) approach to protecting the core network and the weak \nlink(s) is to create what is commonly referred to as a demilitarized zone (DMZ). Rather than \nleaving the weakest links completely open to attack (as in the case of the sacrificial lamb \napproach), they are sheltered behind a liberal (or thin) firewall. The core network is then \nfurther protected by a more rigorous (or strict) firewall, imposing restrictions that could not \nbe imposed on the servers in the DMZ without making them unreachable by their intended \naudience. In effect, the weak links are sandwiched between two firewalls (as depicted in \nFigure A.3). It is the area between the two firewalls that is considered to be a DMZ \n \n \nFigure A.3: Dual-firewall DMZ configuration.  \nUnfortunately, using two (or more) firewalls has some drawbacks. The extra hardware, \nsystem software licenses, and increased administrative maintenance will consume \nresources that could have been used elsewhere. For example, implementing a dual-firewall \nDMZ will require two sets of firewall rules to be maintained, one for the liberal firewall and \none for the strict firewall, potentially increasing the chance of human error. In addition, \nadding an extra firewall will add one more hop to each network request, increasing network \nlatency and slowing the overall performance of the network. On the plus side, using two \ndifferent brands of firewall, one for the liberal and the one for the strict, means that an \nintruder who discovers a security hole in one firewall is unlikely to be able to use the same \nflaw to circumvent the second firewall. \n"
        },
        {
            "page_number": 261,
            "text": " \n254 \nSingle-Firewall DMZ \nUsing a single firewall to implement a DMZ is often a much more economical option. \nInstead of using two physical firewalls, a single firewall is configured to act as two virtual \nfirewalls (see Figure A.4). \n \n \nFigure A.4: Single-firewall DMZ configuration.  \nThe single firewall uses a different set of firewall rules for network traffic attempting to \naccess the machines in the DMZ than for the traffic destined for the core network. If the \ninbound traffic originating from the DMZ machines is subjected to the same set of rules that \ntraffic from the outside world is subjected to, then compromising a DMZ machine will not \nincrease the vulnerability of any of the machines inside the firewall. \nThe downside to a single-firewall DMZ configuration is that the firewall has to deal with \nincreased network traffic, as the single physical firewall has to handle external-to-DMZ \nrequests across the same hardware that is monitoring the DMZ-to-internal requests. This is \nnot a problem if the firewall has plenty of spare capacity, but it is a potential performance \nbottleneck if the firewall is already exceeding 50 percent of its throughput capacity just \nmonitoring one of these routes. \nCombination DMZ \nPerhaps the most secure configuration is to combine the concepts of both the single-firewall \nand dual-firewall DMZ configurations. Here the outer firewall supporting the devices located \nin the DMZ is backed up by a second stricter firewall, which in all likelihood is not only a \ndifferent brand of firewall, but quite possibly working at a different OSI network layer. For \nexample, the outer firewall might be a Cisco PIX router working at the network layer, while \nthe stricter firewall may be a Microsoft ISA server checking network traffic at the application \nlayer (see Figure A.5). \n \n"
        },
        {
            "page_number": 262,
            "text": " \n255 \n \nFigure A.5: Combination DMZ configuration.  \n"
        },
        {
            "page_number": 263,
            "text": " \n256 \nAppendix B: SANS Institute Top 20 Critical Internet \nSecurity Vulnerabilities \nThe System Administration, Networking and Security Institute (SANS, www.sans.org) was \nestablished in 1989 as a cooperative research and education organization for security \nprofessionals, auditors, system administrators, and network administrators to share their \nknowledge. \nIn October 2001, the SANS Institute in conjunction with the FBI released a list summarizing \nthe 20 most critical Internet security vulnerabilities (summarized in Table B.1). This new list \nupdates and expands the Top 10 list released by the SANS Institute and the National \nInfrastructure Protection Center (NIPC, www.nipc.gov) in 2000. Thousands of organizations \nhave used these lists to prioritize their testing efforts so they could detect and close the \nmost dangerous security holes first. The majority of successful attacks on computer \nsystems via the Internet can be traced to the exploitation of security vulnerabilities on this \nlist. \n \nTable B.1: The 20 Most Critical Internet Security Vulnerabilities  \nPLATFORM  \nVULNERABILITY  \nAll platforms \nDefault installs of operating systems and applications \n  \nAccounts with no passwords or weak passwords \n  \nNonexistent or incomplete backups \n  \nLarge number of open ports \n  \nNot filtering packets for correct incoming and outgoing addresses \n  \nNonexistent or incomplete logging \n  \nVulnerable Common Gateway Interface (CGI) programs \nWindows-\nspecific \nUnicode vulnerability-Web server folder traversal \n  \nInternet server application programming interface (ISAPI) \nextension buffer overflows \n  \nIIS Remote Data Services (RDS) exploit \n  \nNetwork Basic Input Output System (NetBIOS), unprotected \nWindows networking shares \n  \nInformation leakage via null session connections \n  \nWeak hashing in SAM (Security Accounts Manager)-LanManager \nhash \nUNIX-specific \nBuffer overflows in Remote Procedure Call (RPC) services \n  \nSendmail vulnerabilities \n  \nBind weaknesses \n  \nRemote system command (such as rcp, rlogin, and rsh) \nvulnerabilities \n"
        },
        {
            "page_number": 264,
            "text": " \n257 \nTable B.1: The 20 Most Critical Internet Security Vulnerabilities  \nPLATFORM  \nVULNERABILITY  \n  \nLine Printer Daemons (LPD) vulnerabilities \n  \nSadmind and mountd exploits \n  \nDefault Simple Network Management Protocol (SNMP) strings \nReprinted with permission: www.sans.org/top20.htm, last updated October 2001. \n \n"
        },
        {
            "page_number": 265,
            "text": " \n258 \nAppendix C: Test-Deliverable Templates \nThis appendix illustrates some example documentation layouts for the test deliverables \npreviously described in Chapter 2, and may be used by the reader as templates for \nconstructing their own customized testing deliverables. These templates are themselves \nloosely based on the structure defined by the Institute of Electrical and Electronics \nEngineers (IEEE, www.ieee.org) in their standard for Software Test Documentation (IEEE \n829-1998). Nguyen (2000), Perry (2000), Stottlemyer (2001), and the Rational Unified \nProcess (www.rational.com) provide additional test deliverable templates. \nTemplate Test Status/Summary Report \nA test status report is normally produced on a regular basis (weekly, for example) to \nsummarize the progress of the testing effort to date. The very last test status report may \nbecome the test summary report for the entire testing effort (perhaps after a few cosmetic \nchanges, such as changing the title of the document to test summary report, the names of \nthe persons who will formally accept the report, and so on). The following sections describe \nsome of the components that are typically found in a test status/summary report. \nƒ \nTest Status Report Identifier. For example, Application XYZ, build n.n test status \nreport mm/dd/ccyy. \nƒ \nStatus. Identifies the test items that have been tested (and optionally the items that \nhave yet to be tested), indicating their versions (or build). This section may also \nhighlight any major project management metrics that have been captured and that \nneed to be reported, such as the number of tests completed, the percentage of tests \nfailing, the total number of effort hours expended, and so on. \nƒ \nThis section may also document any major issues that have impacted (or are \nexpected to impact) the testing effort, supplemented with a brief description of how \nthe closed issues were resolved and how the open issues could potentially be dealt \nwith. \nƒ \nVariances. Reports any variances of the test items to their (explicit or implicit) \ndesign specifications. Examples of variances include vulnerabilities in system \nsoftware that have not been patched, buffer overflows in application code, on-duty \nsecurity staff not responding appropriately to an intrusion detection system (IDS) \nalert, doors to computer rooms being left unlocked, and so on. \nƒ \nComprehensive Assessment. Documents the comprehensiveness of the testing \neffort compared to the testing objectives listed in the test plans. Identifies and justifies \nany aspects of the testing effort that have not yet been (or will not be) performed to \nthe degree originally envisioned by the original version of the test plan. \nƒ \nSummary of Results (Incidents). Summarizes the results of the testing done to \ndate, itemizing all resolved and unresolved incidents, their resolutions (if resolved), \nand impact (if any) on the testing effort. \nƒ \nEvaluation. Provides an overall evaluation of the items that have been tested, \nhighlighting each of the vulnerabilities that have been detected and potentially rating \ntheir severity. These evaluations should be based on the test results and not any \nexpectations that the test team may have based on hunches or gut feelings. If \nunsubstantiated suspicious are to be included (perhaps because time limitations \nprohibited sufficient investigation), then their inclusion should clearly indicate that the \nexistence of these suspected vulnerabilities has yet to be established by impartial \ntesting. \nƒ \nThe evaluation may also include an estimate of the likelihood of a detected \nvulnerability being exploited by an attacker and the impact to the business should he \nor she be successful. This evaluation may then be used by the Web site/application's \nowner (or their proxy, such as a project manager) to decide what vulnerabilities need \n"
        },
        {
            "page_number": 266,
            "text": " \n259 \nto be plugged before the product can be placed into production or, if already in \nproduction, the priority with which the vulnerabilities should be fixed. \nƒ \nSummary of Activities. Summarizes the major testing activities and events that \nhave occurred to date, providing an appropriate level of breakdown to justify the \namount of resources that have so far been consumed by the testing effort. \nƒ \nApproval. Specifies the names and titles of persons who must approve (accept) this \nstatus report. \n \nTemplate Test Incident Report \nMany organizations choose to skip writing a formal test incident report and instead enter \nequivalent information directly into a incident/defect tracking system, a system that typically \nprovides a reporting feature that can, if so desired, produce actual test incident reports. The \nfollowing sections describe some of the components that are typically found in a test \nincident report. \nƒ \nTest Incident Identifier. For example, Application XYZ, incident nnnn. \nƒ \nSummary. Provides a brief description of the incident that was observed. This would \ninclude pertinent information such as the item that was being tested when the incident \noccurred (including the exact version or build of the item), the test(s) that was being \nexecuted when the incident occurred (ideally cross-referenced to the appropriate test \nlog entry, test case, and test plan), and the name of the person who observed the \nincident. \nƒ \nIncident Description. The following is a list of some of the additional attributes that \nan organization may wish to consider capturing at the moment (or shortly after) an \nincident was observed: \nƒ \nThe date and time when the incident occurred \nƒ \nContact information for the person who observed the incident (and, if different, \nthe person who wrote up the incident), such as an email address or telephone \nnumber \nƒ \nThe phase of testing that the incident occurred in: unit level, system level, \noutsourced penetration test, production, and so on \nƒ \nA description of the steps taken to create the incident (cross-referenced to \nany relevant test automation script) \nƒ \nWhether or not the incident is repeatable \nƒ \nWhether or not the incident will prevent, or inhabit further testing \nƒ \nIf detected by a tool, the name and version of the tool and, if relevant, any \nspecial configuration settings \nƒ \nThe initial estimate of the relative severity of the incident to the business \nƒ \nThe initial estimate of the relative likelihood (or frequency) of the incident \nreoccurring in production \nƒ \nThe initial (and subsequently refined) estimation of the cause of the incident, \nsuch as a defect in the testing tool, application software, system software \nconfiguration, testing procedure, and so on \nƒ \nSupporting information, such as screen capture attachments or test output \nreports \nƒ \nA detailed description of the incident \nƒ \nAdditional attributes that may be added at a later date to facilitate better defect \ntracking include the following: \nƒ \nA status, such as: new (unassigned), open (assigned), closed, unable to \nrecreate, rejected, deferred, duplicate, and so on \nƒ \nThe date and name of the person closing out the incident (and, if authorized \nby a different person, their name and title) \nƒ \nThe priority given to the investigation effort \n"
        },
        {
            "page_number": 267,
            "text": " \n260 \nƒ \nWho the investigation is assigned to (and optionally the date by which they \nneed to complete their assessment) \nƒ \nThe priority given to fix the defect \nƒ \nAn estimate of the effort needed to fix the defect \nƒ \nWho the fix is assigned to (and optionally the date or version number by which \nhe or she needs to complete the fix) \nƒ \nWhether or not retesting will be required (and if not, why not) \nƒ \nWho the retesting is assigned to (and optionally the date by which he or she \nneeds to complete the retest) \nƒ \nThe root cause of the defect, such as: requirements, design, code, reference \ndata, version control, environment, test tool, system software configuration, and \nso on \nƒ \nRecommendations on how to avoid a reoccurrence of the same (or similar) \nincident \nIn addition, changes to any of this data may also warrant tracking. \n \nTemplate Test Log \nA test log could be as informal as the back of an envelope or as comprehensive as a series \nof screen captures that records every single interaction that the tester has with the system \nbeing tested. The degree of detail deemed appropriate for a particular set of tests will, to a \nlarge extent, depend on what this documentation will actually be used for. For instance, an \noutsourced testing firm conducting a security assessment may need to provide a very \ndetailed test log to prove that they actually performed the work they were hired to do. On \nthe other hand, an in-house security group conducting some exploratory penetration tests \nmay just want to jot down some pointers that will be used to design more structured tests at \na later point in the testing effort. \nThe following template should therefore not be regarded as the way to document a test log, \nbut merely an example of how such information could be formally recorded for posterity. \nƒ \nTest Log Identifier. For example, Application XYZ, build n.n test log report \nmm/dd/ccyy (and to mm/dd/ccyy if spanning multiple days). \nƒ \nDescription. This section often includes information that applies to all the entries in \nthe test log, such as the hardware and system software environment where the \ntesting took place, or the test items that were the subject of the tests covered by this \ntest log. \nƒ \nActivity and Event Entries. Documents the actual results of each test run, \ninformation that may be more easily documented using a landscape format such as \nthe example layout depicted by Table C.1. \n \nTable C.1: Example Test Log Layout  \nTESTING \nSTART \nTIME: \n10.15 \nAM \nTESTING END \nTIME: 2.30 PM  \n  \nLOG AUTHOR: \nTESTER(S): \nOBSERVER(S):  \nMARC \nSILVER \nMARY \nCATCHALL  \n"
        },
        {
            "page_number": 268,
            "text": " \n261 \nTEST \nIDENTIFIER \nAND \nTEST \nNAME  \nPASS/FAIL  \nINCIDENT \nIDENTIFIER  \nBRIEF DESCRIPTION OF \nINCIDENT  \nSysNet25: Port \nscan \nWeb \nserver \n#1 \nfor \nunauthorized \nopen ports \nFail \n91 \nPort #139 (NetBIOS) found \nopen \nSysNet26: Port \nscan \nWeb \nserver \n#2 \nfor \nunauthorized \nopen ports \nPass \n  \n  \nSysNet27: Ping \ninternal (private) \nnetwork \naddresses from \noutside \nperimeter \nfirewall \nFail \n92 \nResponse detected from IP \naddress 123.456.789.123 \nSysNet28: DNS \nzone transfer \nPass \n  \n  \nSysNet29: \nStatic \nnetwork \naddress entries \nin Web server \n#1's hosts file \nPass \n  \n  \nSysNet30: \nStatic \nnetwork \naddress entries \nin Web server \n#2's hosts file \nPass \n  \n  \nSysNet31: \nPerimeter \nfirewall \nblocks \nall \noutbound \nUDP \nnetwork \ntraffic \nPass \n  \n  \nSysNet32: \nStress \ntest \nperimeter \nfirewall \nPass \n  \n  \nSysNet33: \nWireless \nsegment \neffective range \nFail \n93 \nConnection to the corporate \nLAN (which has unfiltered \nnetwork access to the Web \napplication) \nwas \n"
        },
        {
            "page_number": 269,
            "text": " \n262 \nTable C.1: Example Test Log Layout  \nTESTING \nSTART \nTIME: \n10.15 \nAM \nTESTING END \nTIME: 2.30 PM  \n  \nLOG AUTHOR: \nTESTER(S): \nOBSERVER(S):  \nMARC \nSILVER \nMARY \nCATCHALL  \nTEST \nIDENTIFIER \nAND \nTEST \nNAME  \nPASS/FAIL  \nINCIDENT \nIDENTIFIER  \nBRIEF DESCRIPTION OF \nINCIDENT  \nestablished \nfrom \na \nhamburger \nrestaurant's \nparking lot located half a \nmile away from the Web \nsite's physical location. \n \n"
        },
        {
            "page_number": 270,
            "text": " \n263 \nAdditional Resources \nThe subject matter for testing Web security is so vast that one book cannot hope to cover \nevery facet of this knowledge domain in great detail. This section has therefore been \nincluded in this book to provide the reader with a consolidated list of book and Web sites \nthat may be used to acquire additional information of the topic of testing Web security. \nBooks \nThis list of books not only serves as a bibliography of all the books referenced in this book, \nbut also includes other books related to this book's subject matter and may therefore prove \nto be an additional resource for the reader. \nComputer Forensics \nCaloyannides, Michael A. Computer Forensics and Privacy. Artech House, 2001. \nCasey, Eoghan. Digital Evidence and Computer Crime. Academic Press, 2000. \nKruse, Warren G., II, and Jay G. Heiser. Computer Forensics: Incident Response \nEssentials. Addison-Wesley, 2001. \nMarcella, Albert J., Jr., and Robert S. Greenfield. Cyber Forensics: A Field Manual for \nCollecting, Examining, and Preserving Evidence of Computer Crimes. Auerbach, 2002. \nProsise, Chris, and Kevin Mandia. Incident Response: Investigating Computer Crime. \nOsborne McGraw-Hill, 2001. \nShinder, Debra Littlejohn. Scene of the Cybercrime: Computer Forensics Handbook. \nSyngress Media, 2002. \nVacca, John R., and Michael Erbschloe. Computer Forensics: Computer Crime Scene \nInvestigation. Charles River Media, 2002. \nConfiguration Management \nBrown, William J.,Hays W. McCormick, and Scott W. Thomas. Anti-Patterns and Patterns in \nSoftware Configuration Management. John Wiley & Sons, 1999. \nDart, Susan. Configuration Management: The Missing Link in Web Engineering. Artech \nHouse, 2000. \nHaug, Michael, Eric W. Olsen, and Gonzalo Cuevas (eds.). Managing the Change: \nSoftware Configuration and Change Management. Springer Verlag, 1999. \nLeon, Alexis. A Guide to Software Configuration Management. Artech House, 2000. \nLyon, David D. Practical CM: Best Configuration Management Practices. Butterworth \nArchitecture, 2000. \nWhite, Brian A., and Geoffrey M. Clemm. Software Configuration Management Strategies \nand Rational ClearCase. Addison Wesley Professional, 2000. \n"
        },
        {
            "page_number": 271,
            "text": " \n264 \nDisaster Recovery \nMaiwald, Eric, William Sieglein, and Michael Mueller. Security Planning and Disaster \nRecovery. Osborne McGraw-Hill, 2002. \nToigo, Jon William, and Margaret Romano Toigo. Disaster Recovery Planning: Strategies \nfor Protecting Critical Information Assets, Second Edition. Prentice Hall, 1999. \nInternet Law \nBrinson, J. Dianne, and Mark F. Radcliffe. Internet Law and Business Handbook: A \nPractical Guide. Ladera Press, 2000. \nMatsuura, Jeffrey H. Security, Rights, & Liabilities in E-Commerce. Artech House, 2001. \nSaunders, Kurt M. Practical Internet Law for Business. Artech House, 2001. \nSingleton, Susan Ecommerce: A Practical Guide to the Law. Ashgate, 2001. \nPoindexter, Carl J., and David L. Baumer. Cyberlaw and E-Commerce. Irwin/McGraw-Hill, \n2001. \nWright, Benjamin, and Jane K. Winn. The Law of Electronic Commerce. Aspen, 2001. \nMiscellaneous \nHedayat, A.S.,Neil J. A. Sloane, and John Stufken. Orthogonal Arrays: Theory and \nApplications. Springer Verlag, 1999. \nSt. Laurent, Simon. Cookies. McGraw-Hill, 1998. \nNetwork Design \nBrooks, Kari. Networking Complete. Sybex, 2001. \nDimarzio, J. F. Network Architecture & Design: A Field Guide for IT Consultants. Sams, \n2001. \nOppenheimer, Priscilla. Top-Down Network Design. Cisco Press, 1999. \nRisk Analysis \nChapman, C. B., and Stephen Ward. Project Risk Management: Processes, Techniques \nand Insights. John Wiley & Sons, 1996. \nFriedman, M., and J. Voas. Software Assessment: Reliability, Safety, and Testability. Wiley-\nInterscience, 1995. \nGrey, Stephen. Practical Risk Assessment for Project Management. John Wiley & Sons, \n1995. \n"
        },
        {
            "page_number": 272,
            "text": " \n265 \nKoller, Glenn Robert. Risk Assessment and Decision Making in Business and Industry: A \nPractical Guide. CRC Press, 1999. \nLeveson, Nancy. Safeware: System Safety and Computers. Addison-Wesley, 1995. \nMcDermott, Robin E.,Raymond J. Mikulak, and Michael R. Beauregard. The Basics of \nFMEA. Productivity Inc., 1996. \nOuld, Martyn. Managing Software Quality and Business Risk. John Wiley & Sons, 1999. \nPalady, Paul. FMEA Author's Edition. PAL Publishing, 1997. \nPeltier, Thomas R. Information Security Risk Analysis. Auerbach Publications, 2001. \nProject Management Institute. A Guide to the Project Management Body of Knowledge. \nProject Management Institute, 2001. \nStamatis, D. H. Failure Mode and Effect Analysis: FMEA from Theory to Execution. \nAmerican Society for Quality, 1995. \nWideman, R. Max (ed.), and Rodney J. Dawson. Project and Program Risk Management: A \nGuide to Managing Project Risks and Opportunities. Project Management Institute \nPublications, 1998.  \nSecurity \nAllen, Julia H. The CERT® Guide to System and Network Security Practices. Addison-\nWesley, 2001. \nAnderson, Ross J. Security Engineering: A Guide to Building Dependable Distributed \nSystems. John Wiley & Sons, 2001. \nBarman, Scott. Writing Information Security Policies. New Riders Publishing, 2001. \nBragg, Roberta. CISSP Training Guide. Que, 2002. \nBrands, Stefan A. Rethinking Public Key Infrastructures and Digital Certificates: Building in \nPrivacy. MIT Press, 2000. \nBrenton, Chris. Mastering Network Security. Sybex, 1998. \nBrenton, Chris, and Cameron Hunt. Active Defense: A Comprehensive Guide to Network \nSecurity. Sybex, 2001. \nBurnett, Steve, and Stephen Paine. RSA Security's Official Guide to Cryptography. \nOsborne McGraw-Hill, 2001. \nCanavan, John E. The Fundamentals of Network Security. Artech House, 2001. \nCastano, Silvano,Maria Grazia Fugini,Giancarlo Martella, and Silvana Castano. Database \nSecurity. Addison-Wesley, 1994. \n"
        },
        {
            "page_number": 273,
            "text": " \n266 \nChirillo, John. Hack Attacks Denied: Complete Guide to Network LockDown. John Wiley & \nSons, 2001. \n------. Hack Attacks Revealed: A Complete Reference with Custom Security Hacking \nToolkit. John Wiley & Sons, 2001. \nCole, Eric, Hackers Beware. New Riders Publishing, 2001. \nCronkhite, Cathy, and Jack McCullough. Access Denied: The Complete Guide to Protecting \nYour Business Online. Osborne McGraw-Hill, 2001. \nEndorf, Carl F. Secured Computing: A Cissp Study Guide. Trafford, 2001. \nErbschloe, Michael. Information Warfare: How to Survive Cyber Attacks. McGraw-Hill, \n2001. \nFeghhi, Jalal,Peter Williams, and Jalil Feghhi. Digital Certificates: Applied Internet Security. \nAddison-Wesley, 1998. \nForno, Richard,Kenneth R.Van Wyk, and Rick Forno. Incident Response. O'Reilly & \nAssociates, 2001. \nGarfinkel, Simson,Gene Spafford, and Debby Russell. Web Security, Privacy and \nCommerce, Third Edition. O'Reilly & Associates, 2002. \nGhosh, Anup K. E-Commerce Security: Weak Links, Best Defenses. John Wiley & Sons, \n1998. \n------. Security and Privacy for E-Business. John Wiley & Sons, 2001. \nHarris, Shon. CISSP All-in-One Exam Guide. Osborne McGraw-Hill, 2001. \nHassler, Vesna. Security Fundamentals for E-Commerce. Artech House, 2000. \nHendry, Mike. Smart Card Security and Applications. Artech House, 2001. \nHeney, William,Marlene L. Theriault, and Debby Russell. Oracle Security, 1998. \nHerrmann, Debra S. A Practical Guide to Security Engineering and Information Assurance. \nCRC Press, 2001. \nHoneynet Project (ed.), Lance Spitzner (Preface), and Bruce Schneier. Know Your Enemy: \nRevealing the Security Tools, Tactics, and Motives of the Blackhat Community. Addison-\nWesley, 2001. \nHuston, L. Brent (ed.), Teri Bidwell, Ryan Ruyssell, Robin Walshaw, and Oliver Steudler. \nHack Proofing Your Ecommerce Site. Syngress Media Inc., 2001. \nKlevinsky T.J.,Scott Laliberte, and Ajay Gupta. Hack I.T.-Security Through Penetration \nTesting. Addison-Wesley, 2002. \nKing, Christopher (ed.), Ertem Osmanoglu, and Curtis Dalton. Security Architecture: \nDesign, Deployment and Operations. McGraw-Hill, 2001. \n"
        },
        {
            "page_number": 274,
            "text": " \n267 \nKrutz, Ronald L.,Russell Dean Vines: Advanced CISSP Prep Guide: Exam Q&A. John \nWiley & Sons, 2002. \nKrutz, Ronald L.,Russell Dean Vines, and Edward M. Stroz. The CISSP Prep Guide: \nMastering the Ten Domains of Computer Security. John Wiley & Sons, 2001. \nLarson, Eric, and Brian Stephens. Administrating Web Servers, Security, and Maintenance \nInteractive Workbook. Prentice Hall, 1999. \nLierley, Mark (ed.). Security Complete. Sybex, 2001. \nLoeb, Larry. Secure Electronic Transactions: Introduction and Technical Reference. Artech \nHouse, 1998. \nMcClure, Stuart, Saumil Shah, and Shreeraj Shah. Web Hacking: Attacks and Defense. \nAddison-Wesley, 2002. \nMcGraw, Gary, and Ed Felten. Securing Java: Getting Down to Business with Mobile Code. \nJohn Wiley & Sons, 1999. \nMerkow, Mark S.,Jim Breithaupt, and James Breithaupt. The Complete Guide to Internet \nSecurity. AMACOM, 2000. \nNanavati, Samir,Michael Thieme, and Raj Nanavati. Biometrics: Identity Verification in a \nNetworked World. John Wiley & Sons, 2002. \nNichols, Randall K.,Julie J. C.H. Ryan, and William E. Baugh, Jr. Defending Your Digital \nAssets: Against Hackers, Crackers, Spies and Thieves. Osborne McGraw-Hill, 1999. \nNorthcutt, Stephen,Donald McLachlan, and Judy Novak. Network Intrusion Detection: An \nAnalyst's Handbook. New Riders Publishing, 2000. \nOppliger, Rolf. Internet and Intranet Security. Artech House, 2002. \n------. Security Technologies for the World Wide Web. Artech House, 1999. \nParker, Donn B. Fighting Computer Crime: A New Framework for Protecting Information. \nJohn Wiley & Sons, 1998. \nPeltier, Thomas, and Patrick D. Howard. The Total CISSP Exam Prep Book: Practice \nQuestions, Answers, and Test Taking Tips and Techniques. Auerbach, 2002. \nPeltier, Thomas R. Information Security Policies, Procedures, and Standards: Guidelines \nfor Effective Information Security Management. CRC Press, 2001. \nPhaltankar, Kaustubh M., and Vinton G. Cerf. Practical Guide for Implementing Secure \nIntranets and Extranets. Artech House, 1999. \nPhoha, Vir V. Internet Security Dictionary. Springer Verlag, 2002. \nPieprzyk, Josef,Thomas Hardjono, and Jennifer Seberry. Fundamentals of Computer \nSecurity. Springer Verlag, 2002. \n"
        },
        {
            "page_number": 275,
            "text": " \n268 \nProctor, Paul E. Practical Intrusion Detection Handbook. Prentice Hall PTR/Sun \nMicrosystems Press, 2000. \nRankl, W., and W. Effing. Smart Card Handbook. John Wiley & Sons, 2000. \nRubin, Aviel D. White-Hat Security Arsenal: Tackling the Threats. Addison-Wesley, 2001. \nScambray, Joel,Stuart McClure, and George Kurtz. Hacking Exposed, Third Edition. \nMcGraw-Hill, 2001. \nScambray, Joel, and Mike Shema. Hacking Exposed: Web Applications. Osborne McGraw-\nHill, 2002. \nSchneier, Bruce. Secrets and Lies: Digital Security in a Networked World. John Wiley & \nSons, 2000. \nSherif, Mostafa Hashem, and Ahmed Sehrouchni. Protocols for Secure Electronic \nCommerce. CRC Press, 2000. \nShumway, Russell, and E. Eugene Schultz. Incident Response: A Strategic Guide to \nHandling System and Network Security Breaches. New Riders Publishing, 2002. \nSkoudis, Ed. Counter Hack: A Step-by-Step Guide to Computer Attacks and Effective \nDefenses. Prentice Hall PTR, 2001. \nSmith, Richard E. Authentication: From Passwords to Public Keys. Addison-Wesley, 2001. \nStallings, William. Cryptography and Network Security: Principles and Practice. Prentice \nHall, 1998. \nStein, Lincoln D. Web Security: A Step-by-Step Reference Guide. Addison-Wesley, 1998. \nSyngress Media (ed.), Ryan Russell, and Stace Cunningham. Hack Proofing Your Network: \nInternet Tradecraft. Syngress Media Inc., 2000. \nTiwana, Amrit. Web Security. Digital Press, 1999. \nTraxler, Julie, and Jeff Forristal. Hack Proofing Your Web Applications. Syngress Media \nInc., 2001. \nTudor, Jan Killmeyer. Information Security Architecture: An Integrated Approach to Security \nin the Organization. CRC Press, 2000. \nViega, John, and Gary McGraw. Building Secure Software: How to Avoid Security Problems \nthe Right Way. Addison-Wesley Professional, 2001. \nWilson, Chuck, and Daniel Schillaci. Get Smart: The Emergence of Smart Cards in the \nUnited States and their Pivotal Role in Internet Commerce. Mullaney Corporation, 2001. \nWinkler, Ira. Corporate Espionage: What It Is, Why It Is Happening in Your Company, What \nYou Must Do About It. Prima Publishing, 1999. \n"
        },
        {
            "page_number": 276,
            "text": " \n269 \nWood, Charles Cresson. Information Security Policies Made Easy Version 8. Baseline \nSoftware, 2001. \nZwicky, Elizabeth D.,Simon Cooper, D.Brent Chapman, and Deborah Russell. Building \nInternet Firewalls, Second Edition. O'Reilly & Associates, 2000. \nSoftware Engineering \nDunn, Robert H. Software Defect Removal. McGraw Hill, 1984. \nKan, Stephen H. Metrics and Models in Software Quality Engineering. Addison-Wesley, \n1995. \nKolawa, Adam,Cynthia Dunlop, and Wendell Hicken. Bulletproofing Web Applications. John \nWiley & Sons, 2001. \nMoller, Karl-Heinrich, and Daniel J. Paulish. Software Metrics: A Practitioner's Guide to \nImproved Product Development. IEEE Computer Society, 1993. \nMusa, \nJohn \nD.,Anthony \nIannino, \nand \nKazuhira \nOkumoto. \nSoftware \nReliability: \nMeasurement, Prediction, Application. McGraw Hill, 1987. \nPfleeger, Shari Lawrence. Software Engineering: Theory and Practice. Prentice Hall, 2001. \nTesting (General) \nBeizer, Boris. Black-Box Testing: Techniques for Functional Testing of Software and \nSystems. John Wiley & Sons, 1995. \n------. Software System Testing and Quality Assurance. International Thomson Publishing, \n1984. \nBlack, Rex. Critical Testing Processes. Addison-Wesley, 2003. \n------. Managing the Testing Process: Practical Tools and Techniques for Managing \nHardware and Software Testing. John Wiley and Sons, 2002. \nBuchanan, Robert W., Jr., The Art of Testing Network Systems. John Wiley & Sons, 1996. \nBuwalda, Hans,Dennis Janssen,Iris Pinkster, and Paul A. Watters. Integrated Test Design \nand Automation: Using the Testframe Method. Addison-Wesley, 2001. \nCraig, Rick, and Stefan Jaskiel. Systematic Software Testing. Artech House, 2002. \nCulbertson, Robert,Chris Brown, and Gary Cobb. Rapid Testing. Prentice Hall PTR, 2001. \nDustin, Elfriede,Jeff Rashka, and John Paul. Automated Software Testing: Introduction, \nManagement, and Performance. Addison-Wesley, 1999. \nGraham, Dorothy,Mark Fewster (Preface), and Brian Marick. Software Test Automation: \nEffective Use of Test Execution. Addison-Wesley, 1999. \n"
        },
        {
            "page_number": 277,
            "text": " \n270 \nHayes, Linda G. Automated Testing Handbook. Software Testing Institute, 1995. \nHetzel, William. The Complete Guide to Software Testing. John Wiley & Sons, 1993. \nJorgensen, Paul C. Software Testing: A Craftsman's Approach, Second Edition. CRC \nPress, 2002. \nKaner, Cem,Hung Quoc Nguyen, and Jack Falk. Testing Computer Software. John Wiley & \nSons, 1999. \nKaner, Cem,James Bach, and Bret Pettichord. Lessons Learned in Software Testing. John \nWiley & Sons, 2001. \nKit, Edward, and Susannah Finzi (ed.). Testing in the Real World: Improving the Process. \nAddison-Wesley, 1995. \nKoomen, Tim, and Martin Pol. Test Process Improvement. Addison-Wesley, 1999. \nLewis, William E. Software Testing and Continuous Quality Improvement. CRC Press, \n2000. \nMusa, John D. Software Reliability Engineered Testing. McGraw-Hill, 1998. \nMyers, Glenford J. The Art of Software Testing. John Wiley & Sons, 1979. \nPatton, Ron. Software Testing. Sams, 2000. \nPerry, William E. Effective Methods for Software Testing, Second Edition. John Wiley & \nSons, 2000. \nPerry, William E., and Randall W. Rice. Surviving the Top Ten Challenges of Software \nTesting: A People-Oriented Approach. Dorset House, 1997. \nRakitin, Steven R. Software Verification and Validation for Practitioners and Managers. \nArtech House, 2001. \nSweeney, Mary Romero. Visual Basic for Testers. Apress, 2001. \nTamres, Louise. Introducing Software Testing. Addison-Wesley, 2002. \nWatkins, John. Testing IT: An Off-the-Shelf Software Testing Process. Cambridge \nUniversity Press, 2001. \nWhittaker, James A. How to Break Software: A Practical Guide to Testing. Addison-Wesley, \n2002. \nWieczorek, Martin, Dirk Meyerhoff, and B. Baltus (eds.). Software Quality: State of the Art in \nManagement, Testing, and Tools. Springer Verlag, 2001.  \n"
        },
        {
            "page_number": 278,
            "text": " \n271 \nTesting (Web) \nDustin, Elfriede A.,Jeff Rashka, and Douglas McDiarmid. Quality Web Systems: \nPerformance, Security, and Usability. Addison-Wesley, 2001. \nGerrard, Paul, and Neil Thompson. Risk Based E-Business Testing. Artech House, 2002. \nMarshall, Steve,Ryszard Szarkowski, and Billie Shea. Making E-Business Work: A Guide to \nSoftware Testing in the Internet Age. Newport Press Publications, 2000. \nNguyen, Hung Quoc. Testing Applications for the Web: Testing Planning for Internet-Based \nSystems. John Wiley & Sons, 2000. \nSplaine, Steven, and Stefan Jaskiel. The Web Testing Handbook. STQE Publishing, 2001. \nStottlemyer, Diane. Automated Web Testing Toolkit: Expert Methods for Testing and \nManaging Web Applications. John Wiley & Sons, 2001. \n \nWeb Sites \nThe following tables summarize in a single consolidated section all the Web sites that have \nbeen referenced by this book. The chapter that references the site provides a more detailed \nexplanation of the reason that the Web site was referenced. Note that, while these Web \nsites were accurate at the time this book was written, due to the constantly changing nature \nof the Web, some of these links may have since become broken. \nChapter 1: Introduction \nPURPOSE  \nREFERENCED WEB SITES  \nSecurity vulnerability statistics and incidents \nwww.cert.org  \nwww.gocsi.com  \nwww.money.cnn.com  \nwww.reuters.co.uk  \nTesting terms and definitions \nwww.ieee.org  \nwww.pmi.org  \nwww.rational.com  \nCISSP Certification \nwww.cissp.com  \nwww.isc2.org  \nChapter 2: Test Planning \nPURPOSE  \nREFERENCED WEB SITE!  \nTesting documentation or processes \nwww.ideahamster.org  \nwww.osstmm.org  \nwww.rational.com  \nwww.standards.ieee.org  \n"
        },
        {
            "page_number": 279,
            "text": " \n272 \nPURPOSE  \nREFERENCED WEB SITE!  \nDefect-tracking tools \nwww.avensoft.com  \nwww.elementool.com  \nwww.elsitech.com  \nwww.empirix.com  \nwww.fesoft.com  \nwww.hstech.com.au  \nwww.mercuryinteractive.com  \nwww.mozilla.org  \nwww.nesbitt.com  \nwww.novosys.de  \nwww.pandawave.com  \nwww.prostyle.com  \nwww.rational.com  \nwww.remotedebugger.com  \nwww.seapine.com  \nwww.skyeytech.com  \nwww.softwarewithbrains.com  \nwww.threerock.com  \nwww.visible.com  \nConfiguration management tools \nwww.ca.com  \nwww.mccabe.com  \nwww.merant.com  \nwww.microsoft.com  \nwww.mks.com  \nwww.rational.com  \nwww.serena.com  \nwww.starbase.com  \nwww.telelogic.com  \nwww.visible.com  \nDisk replication tools \nwww.alohabob.com  \nwww.altiris.com  \nwww.highergroundsoftware.com  \nwww.ics-iq.com  \nwww.logicube.com  \nwww.storagesoftsolutions.com  \nwww.symantec.com  \nProject-scheduling tools \nwww.aecsoft.com  \n"
        },
        {
            "page_number": 280,
            "text": " \n273 \nPURPOSE  \nREFERENCED WEB SITE!  \nwww.axista.com  \nwww.gigaplan.com  \nwww.itgroupusa.com  \nwww.microsoft.com  \nwww.niku.com  \nwww.openair.com  \nwww.pacificedge.com  \nwww.patrena.com  \nwww.performancesolutionstech.com  \nwww.planview.com  \nwww.primavera.com  \nwww.projectkickstart.com  \nwww.rationalconcepts.com  \nwww.timedisciple.com  \nChapter 3: Network Security \nREASON  \nREFERENCED WEB SITES  \nIP-address-sweeping tools and services \nwww.deter.com  \nwww.eeye.com  \nwww.insecure.org  \nwww.ipswitch.com  \nwww.kyuzz.org/antirez  \nwww.nmrc.org  \nwww.nwpsw.com  \nwww.procheckup.com  \nwww.solarwinds.net  \nwww.trulan.com  \nwww.twpm.com  \nNetwork-auditing tools \nwww.ca.com  \nwww.centennial.co.uk  \nwww.dell.com  \nwww.eurotek.co.uk  \nwww.hp.com  \nwww.ibm.com  \nwww.lanauditor.com  \nwww.mfxr.com  \nwww.microsoft.com  \nwww.nai.com/magicsolutions.com  \n"
        },
        {
            "page_number": 281,
            "text": " \n274 \nREASON  \nREFERENCED WEB SITES  \nwww.trackbird.com  \nwww.vislogic.org  \nDomain name service (DNS) zone transfer \ntools and services \nwww.domtools.com  \nftp.cerias.purdue.edu  \nftp.technotronic.com  \nftp.uu.net/networking/ip/dns  \nwww.samspade.org  \nwww.solarwinds.net  \nwww.visi.com/~barr  \nTrace route tools and services \nwww.marko.net  \nwww.mcafee.com  \nwww.netiq.com  \nwww.network-tools.com  \nwww.packetfactory.net  \nwww.procheckup.com  \nwww.solarwinds.net  \nwww.trulan.com  \nwww.visualroute.com  \nNetwork-sniffing tools \nwww.distinct.com  \nwww.ee.lbl.gov  \nwww.eeye.com  \nwww.fbi.gov  \nwww.fte.com  \nwww.hackersclub.com  \nwww.microsoft.com  \nwww.ndgssoftware.com  \nwww.netgroup-serv.polito.it  \nwww.onenetworks.comms.agilent.com  \nwww.rootshell.com  \nwww.sniff-em.com  \nwww.sniffer.com  \nwww.tamos.com  \nwww.tcpdump.org  \nwww.zing.org  \nDenial of service emulation tools and services \nwww.antara.net  \nwww.caw.com  \nwww.exodus.com  \n"
        },
        {
            "page_number": 282,
            "text": " \n275 \nREASON  \nREFERENCED WEB SITES  \nwww.mercuryinteractive.com  \nwww.spirentcom.com  \nTraditional load testing tools \nwww.compuware.com  \nwww.empirix.com  \nwww.fiveninesolutions.com  \nwww.loadtesting.com  \nwww.mercuryinteractive.com  \nwww.microsoft.com  \nwww.radview.com  \nwww.rational.com  \nwww.redhillnetworks.com  \nwww.segue.com  \nwww.sourceforge.net  \nwww.technovations.com  \nwww.velometer.com  \nwww.webperfcenter.com  \nChapter 4: System Software Security \nPURPOSE  \nREFERENCED WEB SITES  \nSystem software products \nwww.apache.org  \nwww.bea.com  \nwww.lotus.com  \nwww.microsoft.com  \nwww.redhat.com  \nwww.symantec.com  \nSecurity certifications \nwww.commoncriteria.org  \nwww.defenselink.mil  \nwww.iso.ch  \nwww.itsec.gov uk \nBug- and incident-tracking centers \nwww.bugnet.com  \nwww.cert.org  \nwww.ciac.org  \nwww.csrc.nist.gov  \nwww.cve.mitre.org  \nwww.fedcirc.gov  \nwww.infosyssec.com  \nwww.iss.net  \n"
        },
        {
            "page_number": 283,
            "text": " \n276 \nPURPOSE  \nREFERENCED WEB SITES  \nwww.nipc.gov  \nwww.ntbugtraq.com  \nwww.ntsecurity.net  \nhttp://packetstorm.decepticons.org  \nwww.sans.org  \nwww.securitybugware.org  \nwww.securityfocus.com  \nwww.securitytracker.com  \nwww.vmyths.com  \nwww.whitehats.com  \nSystem software assessment tools \nwww.fish.com  \nwww.iss.net  \nwww.maximized.com  \nwww.microsoft.com  \nwww.nessus.org  \nwww.netiq.com  \nwww.shavlik.com  \nOperating system protection tools \nwww.bastille-linux.org  \nwww.engardelinux.org  \nwww.immunix.org  \nwww.watchguard.com  \nFingerprinting tools and services \nwww.atstake.com  \nwww.bindview.com  \nwww.cerberus-infosec.co.uk  \nwww.foundstone.com  \nwww.insecure.org  \nwww.marko.net  \nwww.netcraft.com  \nMapping services to ports \nwww.foundstone.com  \nwww.iana.org  \nPort-scanning tools \nwww.atstake.com  \nwww.bindview.com  \nwww.cerberus-infosec.co.uk  \nwww.cotse.com  \nwww.dataset.fr  \nwww.eeye.com  \nwww.foundstone.com  \n"
        },
        {
            "page_number": 284,
            "text": " \n277 \nPURPOSE  \nREFERENCED WEB SITES  \nwww.freebsd.org  \nwww.insecure.org  \nwww.iss.net  \nwww.nai.com  \nwww.nessus.org  \nwww.ntsecurity.nu  \nwww.nwpsw.com  \nwww.packetfactory.net  \nwww.prosolve.com  \nwww.savant-software.com  \nwww.search.iland.co.kr  \nwww.wiretrip.net  \nwww.wwdsi.com  \nwww.www-arc.com  \nPort-scanning services \nwww.grc.com  \nwww.mis-cds.com  \nwww.norton.com  \nwww.securitymetrics.com  \nDirectory and file access control tools \nwww.argus-systems.com  \nwww.armoredserver.com  \nwww.authoriszor.com  \nwww.entercept.com  \nwww.hp.com  \nwww.kavado.com  \nwww.okena.com  \nwww.sanctuminc.com  \nwww.securewave.com  \nwww.watchguard.com  \nFile share scanning tools \nwww.atstake.com  \nwww.hackersclub.com  \nwww.nmrc.org  \nwww.ntsecurity.nu  \nwww.solarwinds.net  \nPassword directories \nwww.hackersclub.com  \nwww.securityparadigm.com  \nPassword-deciphering/guessing/assessment tools \nwww.antifork.org  \nwww.bindview.com  \n"
        },
        {
            "page_number": 285,
            "text": " \n278 \nPURPOSE  \nREFERENCED WEB SITES  \nwww.cerberus-infosec.co.uk  \nwww.hackersclub.com  \nwww.hoobie.net  \nwww.intrusion.com  \nwww.nai.com  \nwww.nmrc.org  \nhttp://packetstorm.decepticons.org  \nwww.phenoelit.de  \nwww.securitysoftwaretech.com  \nwww.users.dircon.co.uk/~crypto  \nwww.waveset.com  \nChapter 5: Client-Side Application Security \nPURPOSE  \nREFERENCED WEB SITES  \nProviders of personal certificates and related services \nwww.btignite.com  \nwww.entrust.com  \nwww.globalsign.net  \nwww.openca.org  \nwww.scmmicro.com  \nwww.thawte.com  \nwww.verisign.com  \nProviders of smart cards and related services \nwww.activcard.com  \nwww.datakey.com  \nwww.ibutton.com  \nwww.labcal.com  \nwww.motus.com  \nwww.rsasecurity.com  \nwww.signify.net  \nwww.vasco.com  \nProviders of biometric devices and related services \nwww.activcard.com  \nwww.cybersign.com  \nwww.digitalpersona.com  \nwww.identix.com  \nwww.interlinkelec.com  \nwww.iridiantech.com  \nwww.keyware.com  \nwww.saflink.com  \nwww.secugen.com  \n"
        },
        {
            "page_number": 286,
            "text": " \n279 \nPURPOSE  \nREFERENCED WEB SITES  \nwww.visionics.com  \nLink-checking tools \nwww.coast.com  \nwww.empirix.com  \nwww.mercuryinteractive.com  \nwww.rational.com  \nwww.trellian.com \nwww.watchfire.com  \nProviders of encryption accelerators \nwww.aep-crypto.com  \nwww.andesnetworks.com  \nwww.cryptoapps.com  \nwww.ibm.com  \nwww.ncipher.co  \nwww.powercrypt.com  \nwww.rainbow.com  \nwww.safenet-inc.com  \nwww.sonicwall.com  \nActiveX Control Safety Checklist \nwww.msdn.microsoft.com  \nClient-side script vulnerability demonstrations \nwww.guninski.com  \nCode coverage tool \nwww.mccabe.com  \nPersonal firewalls \nwww.esafe.com  \nwww.f-secure.com  \nwww.infoexpress.com  \nwww.mcafee.com  \nwww.microsoft.com  \nwww.neoworx.com  \nwww.netfilter.samba.org  \nwww.network-1.com  \nwww.networkice.com  \nwww.seawall.sourceforge.net  \nwww.symantec.com  \nwww.zonelabs.com  \nOrthogonal array manipulation tools \nwww.argreenhouse.com  \nwww.lib.stat.cmu.edu  \nwww.phadkeassociates.com  \nOnline advertisers \nwww.aol.com  \nwww.doubleclick.net  \n"
        },
        {
            "page_number": 287,
            "text": " \n280 \nChapter 6: Server-Side Application Security \nPURPOSE  \nREFERENCED WEB SITES  \nCGI test harness tool \nwww.htmlhelp.com  \nCGI scalability \nwww.fastcgi.com  \nSample CGI scanners \nwww.atstake.com  \nwww.glocksoft.com  \nwww.nebunu.home.ro  \nwww.nessus.org  \nhttp://packetstorm.decepticons.org  \nwww.point2click.de  \nwww.sourceforge.net  \nwww.wiretrap.net  \nSample dynamic code environments \nwww.chilisoft.com  \nwww.halcyonsoft.com  \nwww.macromedia.com  \nwww.microsoft.com  \nwww.php.net  \nwww.sun.com  \nwww.vignette.com  \nHTML squishing tool \nwww.imagiware.com  \nTool libraries \nwww.acme.com  \nwww.davecentral.com  \nwww.download.com  \nwww.freeware32.com  \nwww.osdn.com  \nwww.shareware.com  \nwww.sourceforge.net  \nwww.tucows.com  \nwww.zdnet.com  \nWeb site crawlers/mirroring tools \nwww.monkey.org  \nwww.samspade.org  \nwww.tenmax.com  \nwww.wget.sunsite.dk  \nWeb-testing-based scripting tools \nwww.atesto.com  \nwww.autotester.com  \nwww.compuware.com  \nwww.empirix.com  \nwww.mercuryinteractive.com  \n"
        },
        {
            "page_number": 288,
            "text": " \n281 \nPURPOSE  \nREFERENCED WEB SITES  \nwww.oclc.org  \nwww.radview.com  \nwww.rational.com  \nwww.segue.com  \nwww.soft.com  \nwww.wintask.com  \nBuffer overflow explanation \nwww.insecure.org  \nBuffer overflow protection tools \nwww.immunix.org  \nwww.securewave.com  \nSource-code-scanning tools \nwww.cigital.com  \nwww.gimpel.com  \nwww.ldra.co.uk  \nwww.parasoft.com  \nwww.programmingresearch.com  \nwww.reasoning.com  \nwww.securesw.com  \nwww.splint.cs.virginia.edu  \nwww.spidynamics.com  \nBuffer overflow generation tools \nwww.cenzic.com  \nwww.earth.li/projectpurple  \nwww.empirix.com  \nwww.foundstone.com  \nwww.microsoft.com  \nwww.perl.com  \nwww.radview.com  \nASCII data input characters \nwww.ansi.org  \nInput data validation tools \nwww.elitesecureweb.com  \nwww.gilian.com  \nwww.microsoft.com  \nwww.stratum8.com  \nData vaults \nwww.borderware.com  \nwww.cyber-ark.com  \nFile encryption tools \nwww.bestcrypto.com  \nwww.cipherpack.com  \nwww.data-encryption.com  \nwww.easycrypt.co.uk  \nwww.f-secure.com  \n"
        },
        {
            "page_number": 289,
            "text": " \n282 \nPURPOSE  \nREFERENCED WEB SITES  \nwww.gnupg.org  \nwww.gregorybraun.com  \nwww.hallogram.com  \nwww.pcguardian.com  \nwww.reflex-magnetics.com  \nwww.rsasecurity.com  \nwww.steganos.com  \nDatabase assessment tool \nwww.appsecinc.com  \nwww.iss.net  \nChapter 7: Sneak Attacks: Guarding against the Less-Thought-of Security Threats \nPURPOSE  \nREFERENCED WEB SITES  \nManufactures \nor \nresellers \nof \ndegassing \ndevices \nwww.benjaminsweb.com  \nwww.datadev.com  \nwww.datalinksales.com  \nwww.datasecurityinc.com  \nwww.garner-products.com  \nwww.proton-eng.com  \nwww.veritysystems.com  \nwww.weircliffe.co.uk  \nKeyboard logging and screen reordering tools \nwww.computerspy.com  \nwww.enfiltrator.com  \nwww.internet-monitoring-\nsoftware.com  \nwww.keykatcher.com  \nwww.snapshotspy.com  \nwww.spectorsoft.com  \nwww.webroot.com  \nChapter 8: Intruder Confusion, Detection, and Response \nPURPOSE  \nREFERENCED WEB SITES  \nHoney pot examples \nwww.nai.com  \nwww.netsecureinfo.com  \nwww.project.honeynet.org  \nwww.recourse.com  \nwww.snetcorp.com  \nIntrusion detection systems (IDSs) \nwww.ca.com  \n"
        },
        {
            "page_number": 290,
            "text": " \n283 \nPURPOSE  \nREFERENCED WEB SITES  \nwww.cisco.com  \nwww.enterasys.com  \nwww.esecurityinc.com  \nwww.foundstone.com  \nwww.hp.com  \nwww.infinity-its.com  \nwww.iss.net  \nwww.microsoft.com  \nwww.nfr.com  \nwww.okena.com  \nwww.onesecure.com  \nwww.openwall.com  \nwww.psionic.com  \nwww.recourse.com  \nwww.snort.org  \nwww.statonline.com  \nwww.symantec.com  \nSystem event logging and reviewing tools \nwww.bluelance.com  \nwww.ca.com  \nwww.consul.com  \nwww.gfisoftware.com  \nwww.netforensics.com  \nwww.opensystems.com  \nwww.spidynamics.com  \nTripwire and checksum tools \nwww.lockstep.com  \nwww.pedestalsoftware.com  \nwww.redhat.com  \nwww.resentment.org  \nwww.sourceforge.net  \nwww.tripwire.com  \nwww.tripwire.org  \nVirus, Trojan horse, and worm definitions \nwww.symantec.com  \nAntivirus products \nwww.agnitum.com  \nwww.antivirus.com  \nwww.ca.com  \nwww.commandcom.com  \nwww.f-secure.com  \n"
        },
        {
            "page_number": 291,
            "text": " \n284 \nPURPOSE  \nREFERENCED WEB SITES  \nwww.lockdowncorp.com  \nwww.mcafee.com  \nwww.pestpatrol.com  \nwww.sophos.com  \nwww.symantec.com  \nComputer virus hoaxes and myths \nwww.vmyths.com  \nMobile-code-scanning tools \nwww.ca.com  \nwww.esafe.com  \nwww.finjan.com  \nwww.trendmicro.com  \nAntivirus test file \nwww.eicar.org  \nCentralized security-monitoring tools and services \nwww.aprisma.com  \nwww.brinksinternetsecurity.com  \nwww.ca.com  \nwww.counterpane.com  \nwww.esecurityinc.com  \nwww.exodus.net  \nwww.farm9.com  \nwww.guarded.net  \nwww.guardent.com  \nwww.iss.net  \nwww.itactics.com  \nwww.metases.com  \nwww.mis-cds.com  \nwww.netiq.com  \nwww.nsec.net  \nwww.pentasafe.com  \nwww.redsiren.com  \nwww.riptech.com  \nwww.savvis.com  \nwww.secureworks.com  \nwww.trusecure.com  \nwww.ubizen.com  \nwww.verio.com  \nwww.verisign.com  \nwww.veritect.com  \nwww.vigilinx.com  \n"
        },
        {
            "page_number": 292,
            "text": " \n285 \nPURPOSE  \nREFERENCED WEB SITES  \nWeb log analysis tools \nwww.uu.se/software/analyze  \nForensic data collection tools \nwww.accessdata.com  \nwww.dibsusa.com  \nwww.digitalintel.com  \nwww.dmares.com  \nwww.encase.com  \nwww.forensics-intl.com  \nwww.foundstone.com  \nwww.ics-iq.com  \nwww.ilook-forensics.org  \nwww.tucofs.com  \nForensic consulting firms \nwww.all.net  \nwww.asrdata.com  \nwww.cryptec-forensic.com  \nwww.dibsusa.com  \nwww.forensics.com  \nwww.forensics.org  \nwww.forensics-intl.com  \nwww.lee-and-allen.com  \nIntruder coordination centers \nwww.cert.org  \nwww.sans.org  \nChapter 9: Assessment and Penetration Options \nPURPOSE  \nREFERENCED WEB SITES  \nSecurity testing firms \nwww.alphanetsolutions.com  \nwww.atstake.com  \nwww.baltimore.com  \nwww.boran.com  \nwww.cigital.com  \nwww.conqwest.com  \nwww.crg.com  \nwww.cryptek.com  \nwww.defcom.com  \nwww.emprisetech.com  \nwww.esmartcorp.com  \nwww.exodus.net  \nwww.ey.com  \n"
        },
        {
            "page_number": 293,
            "text": " \n286 \nPURPOSE  \nREFERENCED WEB SITES  \nwww.farm9.com  \nwww.foundstone.com  \nwww.grayhatsecurity.com  \nwww.grci.com  \nwww.guardent.com  \nwww.hyperon.com  \nwww.ibm.com  \nwww.infidel.net  \nwww.infoscreen.com  \nwww.iss.net  \nwww.ixsecurity.com  \nwww.mavensecurity.com  \nwww.metases.com  \nwww.mis-cds.com  \nwww.nsec.net  \nwww.predictive.com  \nwww.procheckup.com  \nwww.rawtenla.com  \nwww.redsiren.com  \nwww.riptech.com  \nwww.secureinfo.com  \nwww.securityautomation.com  \nwww.sses.net  \nwww.sysinct.com  \nwww.systemexperts.com  \nwww.tescom-usa.com  \nwww.testpros.com  \nwww.tigertesting.com  \nwww.tritonic.com  \nwww.trusecure.com  \nwww.trustasia.com  \nwww.verisign.com  \nwww.veritect.com  \nwww.vigilinx.com  \nInsurance carriers offering cybercrime policies \nwww.insuretrust.com  \nwww.lloydsoflondon.co.uk  \nwww.safeonline.com  \nSecurity assessment tools \nwww.bindview.com  \n"
        },
        {
            "page_number": 294,
            "text": " \n287 \nPURPOSE  \nREFERENCED WEB SITES  \nwww.cerberus-infosec.co.uk  \nwww.cisecurity.org  \nwww.eeye.com  \nwww.fish.com  \nwww.foundstone.com  \nwww.intrusion.com  \nwww.iss.net  \nwww.nai.com  \nwww.nessus.org  \nwww.netiq.com  \nwww.pentasafe.com  \nwww.sanctuminc.com  \nwww.search.iland.co.kr  \nwww.spidynamics.com  \nwww.statonline.com  \nwww.symantec.com  \nwww.vigilante.com  \nwww.wwdsi.com  \nwww.www-arc.com  \nMultiple O/S management tools \nwww.bootmanager.com  \nwww.ibm.com  \nwww.microsoft.com  \nwww.powerquest.com  \nwww.romtecusa.com  \nwww.v-com.com  \nwww.vmware.com  \nTool libraries \nwww.acme.com  \nwww.atstake.com  \nwww.cotse.com  \nwww.davecentral.com  \nwww.download.com  \nwww.freeware32.com  \nwww.hackingexposed.com  \nwww.ideahamster.org  \nwww.insecure.org  \nwww.nmrc.org  \nwww.ntsecurity.nu  \nwww.shareware.com  \n"
        },
        {
            "page_number": 295,
            "text": " \n288 \nPURPOSE  \nREFERENCED WEB SITES  \nwww.sourceforge.net  \nwww.tucows.com  \nwww.zdnet.com  \nAnonymous Web server service \nwww.anonymizer.com  \nChapter 10: Risk Analysis \nPURPOSE  \nREFERENCED \nWEB \nSITES  \nInformation on fault trees \nwww.relexsoftware.com  \nInformation on attack trees \nwww.cigitallabs.com  \nwww.counterpane.com  \nSecurity standards \nwww.bsi-global.com  \nwww.iso.ch  \nInformation on gap analysis \nwww.greenbridge.com  \nwww.praxiom.com  \nwww.victoriagroup.com  \nTop 20 critical Internet security vulnerabilities \nwww.sans.org  \nInformation on Failure Mode, Effects and Criticality \nAnalysis (FMECA) \nwww.fmea.net  \nwww.fmeca.com  \nwww.relexsoftware.com  \nwww.skymark.com  \nAppendix A: An Overview of Network Protocols, Addresses, and Devices \nPURPOSE  \nREFERENCED WEB SITES  \nOpen Systems Interconnection model \nwww.iso.ch  \nHTTP usage recommendation \nwww.w3c.org  \nIPsec \nwww.ietg.org  \nSecure electronic transaction \nwww.setco.org  \nVendors offering encryption solutions \nwww.baltimore.com  \nwww.entrust.com  \nwww.microsoft.com  \nwww.rsa.com  \nwww.valicert.com  \nwww.verisign.com  \nInternet number assignment \nwww.iana.org  \nComputer technology encyclopedias \nwww.webopedia.com  \n"
        },
        {
            "page_number": 296,
            "text": " \n289 \nPURPOSE  \nREFERENCED WEB SITES  \nwww.whatis.com  \nFirewalls \nwww.3com.com  \nwww.borderware.com  \nwww.checkpoint.com  \nwww.cisco.com  \nwww.dlink.com  \nwww.firewall.com  \nwww.fortinet.com  \nwww.gnatbox.com  \nwww.ibm.com  \nwww.lucent.com  \nwww.microsoft.com  \nwww.netfilter.org  \nwww.netguard.com  \nwww.netmax.com  \nwww.netscreen.com  \nwww.novell.com  \nwww.pgp.com  \nwww.securecomputing.com  \nwww.sonicwall.com  \nwww.sphd.com  \nwww.stonesoft.com  \nwww.symantec.com  \nwww.watchguard.com  \nwww.whalecommunications.com  \nwww.zonealarm.com  \nAppendix B: SANS Institute Top 20 Critical Internet Security Vulnerabilities \nPURPOSE  \nREFERENCED WEB SITES  \nTop 20 critical Internet security vulnerabilities \nwww.nipc.gov  \nwww.sans.org  \nAppendix C: Test-Deliverable Templates \nPURPOSE  \nREFERENCED WEB SITES  \nTesting documentation standards \nwww.ieee.org  \nwww.rational.com  \n"
        },
        {
            "page_number": 297,
            "text": " \n290 \nNews and Information \nThe following Web sites have been included for those readers who want to stay abreast of \nthe latest news and information in the field of Web security testing. \nPURPOSE  \nREFERENCED WEB SITES  \nNews and information \nwww.attrition.org  \nwww.astalavista.com  \nwww.esecurityonline.com  \nwww.hackersdigest.com  \nwww.incidents.org  \nwww.infosecuritymag.com  \nwww.infowar.com  \nwww.isalliance.org  \nwww.itsecurity.com  \nwww.iwar.org.uk  \nwww.phrack.com  \nwww.scmagazine.com  \nwww.searchsecurity.techtarget.com  \nwww.smarthack.com  \nwww.slashdot.org  \nwww.stickyminds.com  \n \n"
        }
    ]
}