LPAR '18: extended abstract

April 04, 2019

Arrays Made Simpler: An Efficient, Scalable and Thorough Preprocessing


General topic

Constraint preprocessing for efficient Symbolic Execution

Motivation

Automatic decision procedures for Satisfiability Modulo Theory are at the heart of almost all recent formal verification methods, including Symbolic Execution (SE). Especially, the theory of arrays is key as it allows to model memory or essential data structures such as maps, vectors and hash tables.

However, this theory is known to be hard to solve in both theory and practice. Even more so in the case of very long formulas coming from binary-level SE. Standard simplification techniques à la Read-over-Write (check the paper for more details) have 2 main drawbacks:

  1. They do not scale on very long sequences of stores because of a quadratic worst-case complexity. This is a major issue in practice: for example, Symbolic Execution over malware or obfuscated programs may need to consider execution traces of several millions of instructions.

  2. They miss many simplification opportunities because of crude approximate equality checks (typically, syntactic term equality). With such checks, index equality may be sometimes proven, but disequality can never be — except in the very restricted case of constant-value indexes.

The theory of arrays can then quickly become a bottleneck of constraint solving and Symbolic Execution.

Contributions

This papers presents a novel efficient and thorough approach to array simplification based upon 3 key components:

  • dedicated data structure;
  • base normalization;
  • domain reasoning.

Experimental results in BINSEC demonstrate that this approach consistently improves the resolution times of the leading SMT solvers (Boolector, Yices, Z3). Moreover, it also scales over very large formulas (several hundreds of thousands of array accesses), with drastic gains in terms of runtime — passing from hours to seconds on the reverse engineering of a code protected by AsPack.

Integrated in BINSEC 0.2