- It's All About the Software
- Hackers, Crackers, and Attackers
- Dealing with Widespread Security Failures
- Technical Trends Affecting Software Security
- The 'ilities
- Penetrate and Patch Is Bad
- On Art and Engineering
- Security Goals
- Know Your Enemy: Common Software Security Pitfalls
- Software Project Goals
What does it mean for a software system to be secure? Does it even make sense to make claims like "Java is secure"? How can a program be secure?
Security is not a static feature on which everyone agrees. It's not something than can be conveniently defined away in a reductionist move. For most developers and architects, security is like pornography is to the United States Supreme Court: They may not be able to define it, but they think they know it when they see it.
The problem is, security is relative. Not only is there no such thing as 100% security, even figuring out what "secure" means differs according to context. A key insight about security is to realize that any given system, no matter how "secure," can probably be broken. In the end, security must be understood in terms of a simple question: Secure against what and from whom?
Understanding security is best understood by thinking about goals. What is it we are trying to protect? From whom are we protecting it? How can we get what we want?
As in today's criminal justice system, much more attention is paid to security after something bad happens than before. In both cases, an ounce of prevention is probably worth a pound of punishment.
Internet time compresses not only the software development life cycle (making software risk management a real challenge), it also directly affects the propagation of attacks. Once a successful attack on a vulnerability is found, the attack spreads like wildfire on the Internet. Often, the attack is embedded in a simple script, so that an attacker requires no more skill than the ability to hit return in order to carry it out.
Internet time is the enemy of software security. Automated Internet-based attacks on software are a serious threat that must be factored into the risk management equation. This makes prevention more important than ever.
Traceability and Auditing
Because there is no such thing as 100% security, attacks will happen. One of the keys to recovering from an attack is to know who did what, and when they did it. Although auditing is not a direct prevention technology, knowing that there is a system for accountability may in some cases dissuade potential attackers.
Auditing is well understood by accountants, who have practiced double-entry bookkeeping for more than 500 years. Banks and other financial institutions have entire divisions devoted to auditing. Most businesses audit their inventories. Every public company has its books audited by a designated accounting firm to meet Security Exchange Commission regulations. Any system in which security is important should seriously consider including auditing.
Good auditing and traceability measures are essential for forensics. They help detect, dissect, and demonstrate an attack. They show who did what when, and provide critical evidence in court proceedings.
Software auditing is a technological challenge. Bits stored on disk as an audit log are themselves susceptible to attack. Verification of an audit log is thus tricky. Nevertheless, auditing is an essential part of software security.
Monitoring is real-time auditing. Intrusion detection systems based on watching network traffic or poring over log files are simple kinds of monitoring systems. These systems are relative newcomers to the commercial security world, and getting shrink-wrapped products to be useful is not easy because of the alarming number of false alarms.
Monitoring a program is possible on many levels, and is an idea rarely practiced today. Simple approaches can watch for known signatures, such as dangerous patterns of low-level system calls that identify an attack in progress. More complex approaches place monitors in the code itself in the form of assertions.
Often, simple burglar alarms and trip wires can catch an attack in progress and can help prevent serious damage.
Privacy and Confidentiality
Privacy and confidentiality are deeply intertwined. There are clear reasons for business, individuals, and governments to keep secrets. Businesses must protect trade secrets from competitors. Web users often want to protect their on-line activities from the invasive marketing machines of AOL, Amazon.com, Yahoo!, and DoubleClick. Governments have classified military secrets to protect.
Of these three groups, individuals probably least understand how important privacy can be. But they are beginning to clue in. Privacy groups such as the Privacy Foundation (http://www.privacyfoundation.org) and the Electronic Privacy Information Center (http://www.epic.org) are beginning to improve the awareness of the general public. Interestingly, the European Union has a large head start over the United States in terms of privacy laws.
In any case, there are often lots of reasons for software to keep secrets and to ensure privacy. The problem is, software is not really designed to do this. Software is designed to run on a machine and accomplish some useful work. This means that the machine on which a program is running can pry out every secret a piece of software may be trying to hide.
One very simple, useful piece of advice is to avoid storing secrets like passwords in your code, especially if that code is likely to be mobile.
Some kinds of information are more secret than others. Most governments have multiple levels of information classification, ranging from Unclassified and merely For Official Use Only, through Secret and Top Secret, all the way to Top Secret/Special Compartmentalized Intelligence.
Most corporations have data to protect toosometimes from their own employees. Having a list of salaries floating around rarely makes for a happy company. Some business partners will be trusted more than others. Technologies to support these kinds of differences are not as mature as we wish.
Different levels of protection are afforded different levels of information. Getting software to interact cleanly with a multilevel security system is tricky.
Anonymity is a double-edge sword. Often there are good social reasons for some kinds of anonymous speech (think of AIDS patients discussing their malady on the Internet), but just as often there are good social reasons not to allow anonymity (think of hate speech, terrorist threats, and so on). Today's software often makes inherent and unanticipated decisions about anonymity. Together with privacy, decisions about anonymity are important aspects of software security.
Microsoft's Global Identifier tracks which particular copy of Microsoft Office originated a document. Sound like Big Brother? Consider that this identifier was used to help tie David L. Smith, the system administrator who created and released the Melissa virus, to his malicious code. 2 What hurts us can help us too.
Often, technology that severely degrades anonymity and privacy turns out to be useful for law enforcement. The FBI's notorious Carnivore system is set up to track who sends e-mail to whom by placing a traffic monitoring system at an Internet service provider (ISP) like AOL. But why should we trust the FBI to stop at the headers? Especially when we know that the supposedly secret Echelon system (also run by the US government) regularly scans international communications for keywords and patterns of activity!
Cookies are used with regularity by e-commerce sites that want to learn more about the habits of their customers. Cookies make buying airline
tickets on-line faster and easier, and they remember your Amazon.com identity so you don't have to type in your username and password every time you want to buy a book. But cookies can cross the line when a single collection point is set up to link cross-Web surfing patterns. DoubleClick collects reams of surfing data about individuals across hundreds of popular Web sites, and they have publicly announced their intention to link surfing data with other demographic data. This is a direct marketer's dream, and a privacy advocate's nightmare.
Software architects and developers, along with their managers, should think carefully about what may happen to data they collect in their programs. Can the data be misused? How? Does convenience outweigh potential privacy issues?
Authentication is held up as one of the big three security goals (the other two being confidentiality and integrity). Authentication is crucial to security because it is essential to know who to trust and who not to trust. Enforcing a security policy of almost any sort requires knowing who it is that is trying to do something to the what we want to protect.
Software security almost always includes authentication issues. Most security-critical systems require users to log in with a password before they can do anything. Then, based on the user's role in the system, he or she may be disallowed from doing certain things. Not too many years ago, PCs did very little in the way of authentication. Physical presence was good enough for the machine to let you do anything. This simplistic approach fails to work in a networked world.
Authentication on the Web is in a sorry state these days. Users tend to trust that a universal resource locator (URL) displayed on the status line means they are looking at a Web site owned by a particular business or person. But a URL is no way to foster trust! Who's to say that yourfriendlybank.com is really a bank, and is really friendly?
People falsely believe that when the little lock icon on their browser lights up that they have a "secure connection." Secure socket layer (SSL) technology uses cryptography to protect the data stream between the browser and the server to which it is connected. But from an authentication standpoint, the real question to ponder is to whom are you connected? Clicking on the little key may reveal a surprise.
When you buy an airline ticket from ual.com (the real United Airlines site as far as we can tell), a secure SSL connection is used. Presumably this secures your credit card information on its travels across the Internet. But, click on the lock and you'll see that the site with which you have a secure channel is not ual.com, it's itn.net (and the certificate belongs to GetThere.com of Menlo Park, CA). Who the heck are they? Can they be trusted with your credit card data?
To abuse SSL in a Web spoofing attack (as described in Web Spoofing [Felten, 1997]), a bad guy must establish a secure connection with the victim's browser at the right times. Most users never bother to click the lock (and sophisticated attackers can make the resulting screens appear to say whatever they want anyway).
Authentication in software is a critical software security problem to take seriously. And there will be literally hundreds of different ways to solve it (see Chapter 3). Stored-value systems and other financial transaction systems require very strong approaches. Loyalty programs like frequent flyer programs don't. Some authentication schemes require anonymity, and others require strict and detailed auditing. Some schemes involve sessions (logging in to your computer in the morning) whereas others are geared toward transactions (payment systems).
Last but certainly not least comes integrity. When used in a security context, integrity refers to staying the same. By contrast to authentication, which is all about who, when, and how, integrity is about whether something has been modified since its creation.
There are many kinds of data people rely on to be correct. Stock prices are a great example. The move toward Internet-enabled devices like WAP (Wireless Application Protocol) phones or iMode devices is often advertised with reference to real-time trading decisions made by a busy person watching the stock market on her phone as she walks through the airport or hears a stockholder's presentation. What if the data are tampered with between the stock exchange (where the information originates) and the receiver?
Stock price manipulation via misinformation is more common than you may think. Famous cases include Pairgain Technologies, whose stock was intentionally run up in 1999 by a dishonest employee, and Emulex Corporation, a fiber channel company whose stock price was similarly manipulated with fake wire stories by a junior college student in California.
Digital information is particularly easy to fake. Sometimes it can be harder to print counterfeit money than to hack a stored-value smart card with differential power analysis (DPA) and to add some electronic blips (see the book Web site for links on DPA). The more the new economy comes to rely on information, the more critical information integrity will become.