Feature-CompProv

• Feature-CompProv is a simple compiler provenance project using human-interpretable, intuitive features.
• Features from elf binaries are extracted to be fed into a basic Tensorflow Decision Tree model as training data.
• Please refer to the PDF file in slides directory for further details.

Requirements

• Please see requirements.txt for all required packages.

Preprocess

• IMPORTANT: Every ELF file for preprocessing must include its ground-truth compiler name in its file name. ex) binutils-2.31.1_x64_clang_nopie_o0_addr2line, coreutils-8.30_x64_icc_pie_os_sleep

• Run the preprocess script preprocess.py:

  python preprocess.fsx {{dir_path}} {{csv_name}}

• Features are stored as .csv files in features directory with name {{csv_name}}.csv.

Train

• Run the train script train.py:

  python train.py {{features_path}}.

  where {{features_path}} is the path to the .csv file obtained from preprocess.py.

Test

• Run the test script test.py:

  python test.py {{model_path}} {{test_path}}

  where
  • {{model_path}} is the path of the model stored by train.py,
  • {{test_path}} is the path of the .csv file created by preprocess.py for testing the model.

Credits

• This project was guided and advised by SoftSec Lab, KAIST. https://github.com/SoftSec-KAIST