Date of Award
Department or Program
Sean W. Smith
Any computer program processing input from the user or network must validate the input. Input-handling vulnerabilities occur in programs when the software component responsible for filtering malicious input---the parser---does not perform validation adequately. Consequently, parsers are among the most targeted components since they defend the rest of the program from malicious input. This thesis adopts the Language-Theoretic Security (LangSec) principle to understand what tools and research are needed to prevent exploits that target parsers. LangSec proposes specifying the syntactic structure of the input format as a formal grammar. We then build a recognizer for this formal grammar to validate any input before the rest of the program acts on it. To ensure that these recognizers represent the data format, programmers often rely on parser generators or parser combinators tools to build the parsers. This thesis propels several sub-fields in LangSec by proposing new techniques to find bugs in implementations, novel categorizations of vulnerabilities, and new parsing algorithms and tools to handle practical data formats. To this end, this thesis comprises five parts that tackle various tenets of LangSec. First, I categorize various input-handling vulnerabilities and exploits using two frameworks. First, I use the mismorphisms framework to reason about vulnerabilities. This framework helps us reason about the root causes leading to various vulnerabilities. Next, we built a categorization framework using various LangSec anti-patterns, such as parser differentials and insufficient input validation. Finally, we built a catalog of more than 30 popular vulnerabilities to demonstrate the categorization frameworks. Second, I built parsers for various Internet of Things and power grid network protocols and the iccMAX file format using parser combinator libraries. The parsers I built for power grid protocols were deployed and tested on power grid substation networks as an intrusion detection tool. The parser I built for the iccMAX file format led to several corrections and modifications to the iccMAX specifications and reference implementations. Third, I present SPARTA, a novel tool I built that generates Rust code that type checks Portable Data Format (PDF) files. The type checker I helped build strictly enforces the constraints in the PDF specification to find deviations. Our checker has contributed to at least four significant clarifications and corrections to the PDF 2.0 specification and various open-source PDF tools. In addition to our checker, we also built a practical tool, PDFFixer, to dynamically patch type errors in PDF files. Fourth, I present ParseSmith, a tool to build verified parsers for real-world data formats. Most parsing tools available for data formats are insufficient to handle practical formats or have not been verified for their correctness. I built a verified parsing tool in Dafny that builds on ideas from attribute grammars, data-dependent grammars, and parsing expression grammars to tackle various constructs commonly seen in network formats. I prove that our parsers run in linear time and always terminate for well-formed grammars. Finally, I provide the earliest systematic comparison of various data description languages (DDLs) and their parser generation tools. DDLs are used to describe and parse commonly used data formats, such as image formats. Next, I conducted an expert elicitation qualitative study to derive various metrics that I use to compare the DDLs. I also systematically compare these DDLs based on sample data descriptions available with the DDLs---checking for correctness and resilience.
Anantharaman, Prashant, "Protecting Systems From Exploits Using Language-Theoretic Security" (2022). Dartmouth College Ph.D Dissertations. 80.