PolyGlot: a fuzzing framework for language processors

PolyGlot is a coverage-guided language fuzzer. It allows you to generate semantically correct test cases easily.

Features:

• Support antlr4 grammar.
• Implemented as aflpp custom mutator.

The original version that uses bison can be found here.

Build

1. Install dependencies:

   apt install -y cmake ninja-build build-essential
   pip install conan

2. Clone the repo and build aflpp:

   git clone https://github.com/OMH4ck/PolyGlot && cd PolyGlot && git submodule update --init && cd AFLplusplus && make -j

3. Build PolyGlot:

Put the following code in your grammar.g4:

options {
  contextSuperClass = PolyGlotRuleContext;
}

@parser::header {
#include "polyglot_rule_context.h"
}

If you use separate lexer and parser file, you should include polyglot_rule_context.h in both files.

Build with cmake: If you have a single g4 file, you should set -DGRAMMAR_FILE=path/to/grammar.g4. For example:

cmake -DCMAKE_BUILD_TYPE=Release -Bbuild -G Ninja -DBUILD_TESTING=OFF -DGRAMMAR_FILE=path/to/grammar.g4

If you have seperate grammar files for parser and lexer, you need to specify PARSER_FILE and LEXER_FILE instead. For example:

cmake -DCMAKE_BUILD_TYPE=Release -Bbuild -G Ninja -DBUILD_TESTING=OFF -DPARSER_FILE=path/to/parser.g4 -DLEXER_FILE=path/to/lexer.g4

If you have additional helper cpp files for antlr, put them all in a dir and specify -DGRAMMAR_HELPER_DIR=/path/to/dir. PolyGlot will use the cpp/cc files when it compiles a parser for the language.

Then build with Ninja:

ninja -C build

Configuration

1. Mutation corpus: PolyGlot uses a corpus of seeds as the source of mutation. Such corpus should cover every rule in the grammar. You can run build/corpus_evaluate --corpus_dir your_corpus to see what rules your corpus is not covering. The larger the corpus the better.

2. If you want to filter out unparsable inputs, run build/corpus_evaluate --corpus_dir your_corpus --output_dir sanitized_output, which will copy all the parsable test cases to sanitized_output.

3. A yaml semantic configuration file. This specifies the semantics of the language. This part is still in migration. For now, you can just copy the following content to a file named semantic.yml:

   ---
   InitFileDir: abs_path/to/mutation_corpus
   IsWeakType: true
   BasicTypes:
   - X

   You need to set InitFileDir to be the path of your mutation corpus.

4. Set the environment variable

   export POLYGLOT_CONFIG=abs_path/to/semantic.yml
   export AFL_CUSTOM_MUTATOR_LIBRARY=abs_path/to/build/libpolyglot_mutator.so
   export AFL_CUSTOM_MUTATOR_ONLY=1
   export AFL_DISABLE_TRIM=1 # We haven't implemented trimming yet.

Run

You can now run it with aflpp: afl-fuzz -i seed_corpus -o out -- ./your_target @@. The seed_corpus should be fully parsable by your grammar.

Currently tested languages