Link

Link is a general RL framework to find reflected XSS vulnerabilities in a black-box and fully automatic manner. It implemented on top of Wapiti a popular open source web scanner. And reinforcement learning components are implemeted based on OpenAI gym and Stable baselines. The details of Link is in our paper, "Link: Black-Box Detection of Cross-Site Scripting Vulnerabilities Using Reinforcement Learning" which appeared in The Web Conference 2022.

Requirements

Recommend to use Anaconda3
Tensorflow==1.14
gym
stable-baselines

Instruction

Training Session

$ python3 train.py -u <training application url> -t <timesteps>
$ python3 train.py -u 'http://localhost:8080' -t 200000

XSS detection phase using trained agent

$ python3 attack_A2C.py -u <target url> -n <model name>
$ python3 attack_A2C.py -u 'http://localhost:8080' -n sample_agent.pkl

Training visulization (Tensorboard)

$ tensorboard --logdir [log directory name]
$ tensorboard --logdir ./tensorboard_log/

Test Suite Installation

Google Firing Range

sudo apt-get install git ant
Download Google AppEngine SDK file in test suite dependency folder and unzip it
git clone https://github.com/google/firing-range.git
cd firing-range
Modify build.xml, appengine.sdk should be your own path of extracted folder
Add below code on line 70 in build.xml

<get src="https://repo1.maven.org/maven2/servletapi/servlet-api/2.4/servlet-api-2.4.jar" dest="${war.dir}/WEB-INF/lib"/>
ant runserver
Test Suite will run on localhost:8080
You should kill process before restart

    $ sudo netstat -lpn |grep :8080
    $ kill process_id

OWASP Benchmark

$ git clone https://github.com/OWASP/benchmark 
$ cd benchmark
$ mvn compile   (This compiles it)
$ sudo runBenchmark.sh/.bat - This compiles and runs it.

Access on https://localhost:8443/benchmark/

WAVSEP

$ docker pull owaspvwad/wavsep
$ docker run -p 127.0.0.1:8090:8080 owaspvwad/wavsep

Access on http://localhost:8090/wavsep/active/index-xss.jsp

Authors

Soyoung Lee
Seongil Wi
Sooel Son

Citing Link

To cite our paper:

@inproceedings{lee:www:2022,
    author = {Lee, Soyoung and Wi, Seongil and Son, Sooel},
    title = {Link: Black-Box Detection of Cross-Site Scripting Vulnerabilities Using Reinforcement Learning},
    year = 2022,
    booktitle = {Proceedings of the {ACM} Web Conference},
    pages = {743--754}
}

Mr-nightskycoder / Link