Security by Design

The technologies of computer security are based on logic. There is no universal standard notion of what secure behavior is. "Security" is a concept that is unique to each situation. Security is extraneous to the function of a computer application, rather than ancillary to it, thus security necessarily imposes restrictions on the application's behavior.

There are several approaches to security in computing, sometimes a combination of approaches is valid:

1. Trust all the software to abide by a security policy but the software is not trustworthy (this is computer insecurity).
2. Trust all the software to abide by a security policy and the software is validated as trustworthy (by tedious branch and path analysis for example).
3. Trust no software but enforce a security policy with mechanisms that are not trustworthy (again this is computer insecurity).
4. Trust no software but enforce a security policy with trustworthy mechanisms.

Many systems unintentionally result in the first possibility. Approaches one and three lead to failure. Since approach two is expensive and non-deterministic, its use is very limited. Because approach number four is often based on hardware mechanisms and avoid abstractions and a multiplicity of degrees of freedom, it is more practical. Combinations of approaches two and four are often used in a layered architecture with thin layers of two and thick layers of four.

There are myriad strategies and techniques used to design security systems. There are few, if any, effective strategies to enhance security after design.

One technique enforces the principle of least privilege to great extent, where an entity has only the privileges that are needed for its function. That way even if an attacker gains access to one part of the system, fine-grained security ensures that it is just as difficult for them to access the rest.

Furthermore, by breaking the system up into smaller components, the complexity of individual components is reduced, opening up the possibility of using techniques such as automated theorem proving to prove the correctness of crucial software subsystems. This enables a closed form solution to security that works well when only a single well-characterized property can be isolated as critical, and that property is also assessable to math. Not surprisingly, it is impractical for generalized correctness, which probably cannot even be defined, much less proven. Where formal correctness proofs are not possible, rigorous use of code review and unit testing represent a best-effort approach to make modules secure.

The design should use "defense in depth", where more than one subsystem needs to be violated to compromise the integrity of the system and the information it holds. Defense in depth works when the breaching of one security measure does not provide a platform to facilitate subverting another. Also, the cascading principle acknowledges that several low hurdles does not make a high hurdle. So cascading several weak mechanisms does not provide the safety of a single stronger mechanism.

Subsystems should default to secure settings, and wherever possible should be designed to "fail secure" rather than "fail insecure" (see fail safe for the equivalent in safety engineering). Ideally, a secure system should require a deliberate, conscious, knowledgeable and free decision on the part of legitimate authorities in order to make it insecure.

In addition, security should not be an all or nothing issue. The designers and operators of systems should assume that security breaches are inevitable. Full audit trails should be kept of system activity, so that when a security breach occurs, the mechanism and extent of the breach can be determined. Storing audit trails remotely, where they can only be appended to, can keep intruders from covering their tracks. Finally, full disclosure helps to ensure that when bugs are found the "window of vulnerability" is kept as short as possible.

Early history of security by Design

The early Multics operating system was notable for its early emphasis on computer security by design, and Multics was possibly the very first operating system to be designed as a secure system from the ground up. In spite of this, Multics' security was broken, not once, but repeatedly. The strategy was known as 'penetrate and test' and has become widely known as a non-terminating process that fails to produce computer security. This led to further work on computer security that prefigured modern security engineering techniques producing closed form processes that terminate.

Secure Coding

If the operating environment is not based on a secure operating system capable of maintaining a domain for its own execution, and capable of protecting application code from malicious subversion, and capable of protecting the system from subverted code, then high degrees of security are understandably not possible. While such secure operating systems are possible and have been implemented, most commercial systems fall in a 'low security' category because they rely on features not supported by secure operating systems (like portability, et al.). In low security operating environments, applications must be relied on to participate in their own protection. There are 'best effort' secure coding practices that can be followed to make an application more resistant to malicious subversion.

In commercial environments, the majority of software subversion vulnerabilities result from a few known kinds of coding defects. Common software defects include buffer overflows, format string vulnerabilities, integer overflow, and code/command injection.

Some common languages such as C and C++ are vulnerable to all of these defects (see Seacord, "Secure Coding in C and C++"). Other languages, such as Java, are more resistant to some of these defects, but are still prone to code/command injection and other software defects which facilitate subversion.

Recently another bad coding practise has come under scrutiny; dangling pointers. The first known exploit for this particular problem was presented in July 2007. Before this publication the problem was known but considered to be academic and not practically exploitable.

In summary, 'secure coding' can provide significant payback in low security operating environments, and therefore worth the effort. Still there is no known way to provide a reliable degree of subversion resistance with any degree or combination of 'secure coding.'