Angora is a mutation-based coverage guided fuzzer. The main goal of Angora is to increase branch coverage by solving path constraints without symbolic execution.
Angora consists of a fuzzer, instrumenting compilers and runtime libraries. Target programs should be compiled with instrumentation in order to collect runtime information.
Two copies of the target program should be prepared, specifically one with taint tracking instrumentation and the other with branch and constraint instrumentation. This ensures a reasonable amount of efficiency when fuzzing due to taint tracking being resource demanding.
Similar to AFL, Angora mutates a set of seeds to increase program coverage. Inputs that trigger new explored branches will be appended to the queue. Angora implements a wide selection of strategies to solve branch constraints. For each new seed, taint tracking will be applied to learn which part of the input will affect which branch constraint. Then mutations will be applied to the input with the tainted parts in consideration. This allows for efficient and precise input generation which significantly increases input coverage.
More details are available in the published works
- build: Scripts for building Angora components.
- common: Common constants and data structures.
- fuzzer: Contains the source code for the fuzzer. The fuzzer runs the target program and repeatedly mutates the input attempting to increase its code coverage statistics.
- src/bin: Source files for the executable binaries.
- src/depot: Depot module for input/output file management.
- src/executor: Executor module for managing target program runs.
- src/search: Exploration strategies. You are free to implement and integrate your own strategy with Angora.
- llvm_mode: Includes source code for instrumenting compilers and DFSan, the taint tracking framework.
- runtime: Taint tracking runtime library for target program.
- runtime_fast: Branch and constraint information collection library for the target program.
- tests: Sample tests to evaluate fuzzer performance.
Copyright (C) 2018 Peng Chen, Hao Chen