Dynamic Binary Instrumentation

Manipulating programs without the source

Overview

Dynamic Binary Instrumentation (DBI) is a technique for observing, analyzing, or modifying a program while it is running, without needing its source code.

This may not sound that useful, but believe me when I say it's incredibly helpful! The classic example of this is fuzzing. In large binaries with ridiculously complex code bases, sometimes fuzzing is the best approach, but fuzzers like AFL++ generally like the binary to be compiled in a special fuzzer-approved way. This is because, for fuzzers to be efficient, we want to keep track of which code paths the fuzzer went down and causes a crash at. In order for this to happen, we compile the binary with a bunch of fuzzer "flags" that trigger a "message" when code execution reaches them, so fuzzers can keep track of exactly what bugs they've found - but this is only possible if we have the source code!

This is where DBI comes in. Despite not having the source code, we can inject extra instructions to log memory accesses, track control flow, count instructions and more. This lets us keep track of what actually runs, but of course at the cost of performance.

Common frameworks include the incredibly powerful Frida and the classic Valgrind, which is many projects' go-to tool for identifying memory leaks and corruption.