Abstract
Computer intrusions can occur in various ways. Many of them occur by exploiting program flaws and
system configuration errors. Existing solutions that detects specific kinds of flaws are substantially different from each other, so aggregate use of them may be incompatible and require substantial changes in the current system and computing practice. Intrusion detection systems may not be the answer either, because they are inherently inaccurate and susceptible to false positives/negatives. This dissertation presents a taxonomy of security flaws that classifies program vulnerabilities into finite number of error categories, and presents a security mechanism that can produce accurate solutions for many of these error categories in a modular fashion. To be accurate, a solution should
closely match the characteristic of the target error category. To ensure this, we focus only on error categories whose characteristics can be defined in terms of a violation of process integrity.
The thesis of this work is that the proposed approach produces accurate solutions for many error categories. To prove the accuracy of produced solutions, we define the process integrity checking approach and analyze its properties. To prove that this approach can cover many error categories, we develop a classification of program security flaws and find error characteristics (in terms of a
process integrity) from many of these categories. We implement proof-of-concept solutions for two most prevalent error categories, the buffer overflow and the race condition, and analyze their accuracy and performance.
