README.md

Overview

This repository contains a dataset designed for binary code similarity detection, focusing on the semantic alignment between assembly instructions and programming language code snippets. The dataset is created to facilitate the training and evaluation of models in tasks related to binary function similarity detection, code reuse analysis, and software security.

Dataset Description

The dataset consists of pairs of assembly instructions and corresponding programming language code snippets extracted from various open-source projects. It is curated to ensure high-quality, semantically aligned pairs that can be used for supervised and unsupervised learning tasks.

Usage

Requirements

• Python 3

Generating Dataset

Use genDataset.py to process the .txt files in OriAsmFile.zip.

Loading the Dataset

• The dataset data structure is defined as:{'Function':"",'Code':[{'PLcode':"",'ASMcode':""}]}

import json
jsondir = './CMRL-paper/json-TestData.txt'
with open(jsondir, 'r+') as f:
    data = json.loads(f.read())
    for func in data:
        for pair in func['Code']:
            print(pair['PLcode'],pair['ASMcode'])
            print('===============================')

[{
    "Function": "<alpn2alpnid>",
    "Code": [{
        "PLcode": "{\n",
        "ASMcode": "\tpush   rbx\n\tmov    rbx,rdi\n"
    }, {
        "PLcode": "  if(strcasecompare(name, "h1"))\n",
        "ASMcode": "\tcall   fb40 <curl_strequal@plt>\n\ttest   eax,eax\n\tjne    fdf4 <alpn2alpnid+0x41>\n"
    }, {
        "PLcode": "  if(strcasecompare(name, "h2"))\n",
        "ASMcode": "\tmov    rdi,rbx\n\tcall   fb40 <curl_strequal@plt>\n\ttest   eax,eax\n\tjne    fdfb <alpn2alpnid+0x48>\n"
    }, {
        "PLcode": "  if(strcasecompare(name, H3VERSION))\n",
        "ASMcode": "\tmov    rdi,rbx\n\tcall   fb40 <curl_strequal@plt>\n\ttest   eax,eax\n\tje     fdf9 <alpn2alpnid+0x46>\n"
    }, {
        "PLcode": "    return ALPN_h3;\n",
        "ASMcode": "\tmov    eax,0x20\n\tjmp    fdf9 <alpn2alpnid+0x46>\n"
    }, {
        "PLcode": "    return ALPN_h1;\n",
        "ASMcode": "\tmov    eax,0x8\n"
    }, {
        "PLcode": "}\n",
        "ASMcode": "\tpop    rbx\n\tret    \n"
    }, {
        "PLcode": "    return ALPN_h2;\n",
        "ASMcode": "\tmov    eax,0x10\n\tjmp    fdf9 <alpn2alpnid+0x46>\n"
    }, {
        "PLcode": "",
        "ASMcode": ""
    }]
}, {
    "Function": "<altsvc_createid>",
    "Code": [{
        ..................................

• Load all datasets

import os
import json
jsondir = '/home/lucky/experiments/dataset/jsonDataSet/'
jsonfile = os.listdir(jsondir)
for file in jsonfile:
    pairnum = 0
    with open(jsondir+file,'r+') as f:
        data = f.read()
        data = json.loads(data)
        for func in data:
            pairnum += len(func['Code'])
        print(file,'    func_num: ',len(data),"     pair_num: ",pairnum)

output:
json-newopensslasm.txt     func_num:  15669      pair_num:  242968
json-newffmpegasm.txt     func_num:  28560      pair_num:  750129
json-newcoreutilsasm.txt     func_num:  20043      pair_num:  263348
json-testasm.txt     func_num:  2      pair_num:  35
json-newcurlasm.txt     func_num:  1887      pair_num:  41551

Contact

No contact information is available yet.