wang-mengying/asteria

Asteria

This is the official repository for Asteria.

Files description

• train.py: python file for model training
• Tree.py: The Tree class defination
• datahelper.py: function set for accessing sqlite database

dirs description

• application: prototype for Asteria
• data: sqlite database files of ASTs
• diaphora_test: the implement of BCSD based AST of Diaphora
• ASTExtraction: scripts for AST extraction.

Functions

1. AST extraction and preprocessing
2. model training
3. vulnerability search

Usage

use as ida plugin

see readme

Vulnerability Search

python application/main_app.py "database of vulnerable functions" "database of firmware functions

For example:

python application/main_app.py data/vul.sqlite data/NetGear_Small.sqlite

vul.sqlite saves the ASTs of vulnerable functions. NetGear_Small.sqlite saves the ASTs of part of firmware functions from NetGear manufacture.

Then the search results are saved into "VulSearch.result".

The format of results is as follows:

[VULFUNC]:   ftp_retrieve_glob  VULELF:/home/ubuntu/disk/hdd_1/ysg/binary_pool/vulnerable_set/CVE-2014-4877/wget
       |Sim:1.000      |Func:   ftp_retrieve_glob      |ELFPath:/home/ubuntu/disk/hdd_2/iie/acfg/firmwareExtracted/NetGear/_R8000-V1.0.3.32_1.1.21.zip.extracted/_R8000-V1.0.3.32_1.1.21.chk.extracted/squashfs-root/bin/wget

       |Sim:1.000      |Func:   ftp_retrieve_glob      |ELFPath:/home/ubuntu/disk/hdd_2/iie/acfg/firmwareExtracted/NetGear/_R8000-V1.0.2.44_1.0.96.chk.extracted/squashfs-root/bin/wget

The VULFUNC denotes the name of vulnerable function. The Sim denotes the similarity score. The Func denotes the candidate function name. The ELFPath denotes the path of binary where candidate function come from.

Model Training

Since the buildroot dataset we used is too large(28G), we construct a demo training dataset for demonstrating.

python train.py

We have the trained model in "data/saved_model.pt".