Exploiting Heterogeneous Program Knowledge Graph for Register-related Instruction Vulnerability Prediction (VP-HPKG)

VP-HPKG explores the random propagation of instruction-stream errors by representing the source program as a multi-layer heterogeneous program knowledge graph, and accurately locates the instructions prone to errors.

Result on WebAssembly programs(C++)

methods	generate_Fib	float-mm
ours	A:0.89 P:0.98 F1:0.80	A:0.85 P:0.90 F1:0.82
PrograML	A:0.85 P:0.93 F1:0.80	A:0.82 P:0.85 F1:0.78
MVD	A:0.67 P:0.50 F1:0.40	A:0.63 P:0.56 F1:0.67

Overview

Each folder is a benchmark and contains the program data. The most important files in each folder are as follow:

• data_test.py: Data acquisition and processing module.
  • def get_node : Identifying entities and relations from the information obtained by LLVM.
  • def get_Ins_SDC : Getting the attribute information of each entity.
• main.py: Data division, model building, and training module.
  • def train_model : Building a knowledge graph. Model building and training.
  • G : Heterogeneous Graph of instrcution layer.
  • BB_G : Graph of basic block layer.

Setup

To run the code, you need the following dependencies:

• Pytorch 1.10.2
• DGL 0.9.0
• sklearn
• numpy 1.18.5

DataSet

Our current experiments are conducted on data obtained by LLVM and LLFI.

• Ins_g.dot : The text and structure information of instructions.
• F_B_I.dot : The position number of the instruction in the program.
• cycle_result.txt, result_other.txt : The result of fault injection.

Usage

Execute the following scripts to train on node classification task:

python main.py