This repository contains all resources and instructions for the Muse tutorial presented at SecDev 2020.
GitHub Account (sign up at: https://github.com)
MuseDev Account (sign up at: https://console.muse.dev)
Muse GitHub App (install from: https://github.com/apps/muse-dev)
Jupyter
pip3 install notebook
Install Python GitHub Library, and Git Library
pip3 install pygithub gitpython
Install Python Pandas and Matplotlib Libraries
pip3 installl pandas matplotlib
Install Muscle (MUse Command Line Environment)
pip3 install muscle-musedev
Produce a GitHub Token
Produce a Muse Authentication Token
- Email [email protected] or DM @MDTom on Twitter
Docker Image: docker pull musedev/build-test
jq command line utility
-
On Mac
brew install jq -
On Linux
apt-get install jq
Go to: https://github.com/muse-dev/secdev20-tutorial and fork the repository.
Log into the Muse Console https://console.muse.dev
Find the forked repository and click analyze. About five minutes later you should see a list of results, including one result returned by FindSecBugs that points out an unsafe use of deserialization.
Using the GitHub edit UI, navigate to the file at:
vulnerability-java-samples/src/main/java/io/meterian/samples/jackson/ProductsDatabase.java
Click the pencil icon to edit the file.
Add the lines marked with a plus below (they will go right after line 29). Do not copy the '+' character:
+ synchronized
public Product add(Product newProduct) {
Click 'save' and select 'Create a new branch for this commit and start a pull request.' (you can accept the default branch name or change it)
Click 'Propose changes'. On the next page, click 'Create pull request'.
You should now see a message from musedev saying 'Pending — Analyzing. Musebot needs a minute'.
In about five minutes you should see the check complete and comments appear in the pull request to indicating that variables in the class are now inconsistently synchronized. In particular, the 'add' method on line 33 calls 'products.put' under synchronization, but 'findById' on line 26 calls 'products.get' without synchronization and so a multi-threaded client may add a product in one thread and have other threads not be able to find that product even if they call 'findById' after the add occurred. These results are provided by Infer, one of the Java static analysis tools included in Muse.
While Infer finds five instances of potential data races, only one result is reported in the pull request. This is because GitHub only allows comments on lines of code that are close to lines that were changed in the pull request. To see the rest of the errors, click on 'Details' in the Muse status line at the bottom of the pull request.
Since we are done analyzing the vulnerability-java-samples directory, we can remove the Muse config for that sub-project to help future analyses go faster. Remove the file vulnerability-java-samples/.muse.toml.
Running Infer and the other tools built into Muse is great, but what if you have your own tool that you want to include? Muse supports a plugin interface that allows you to add custom tools. We will demonstrate this by adding support for Go by including StaticCheck as a custom static analysis tool.
The documentation for the Muse plugin API is here: https://docs.muse.dev/docs/extending-muse/
A 'hello world' tool is given at the bottom of that page and stored in
the secdev20-tutorial repository as hello_world_tool.sh.
Modify secdev20-tutorial/.muse/config.toml to say:
customTools = ["hello_world_tool.sh"]
Go to console.muse.dev and analyze the secdev20-tutorial repo again. After a few minutes you should see a "Hello World" message among the tool results.
To see a 'Hello World' example with more results, try changing secdev20-tutorial/.muse/config.toml to say:
customTools = ["hello-muse.sh"]
Save and again click the Analyze button on the secdev20-tutorial repo on console.muse.dev. After a few minutes you should see several results that list the files over 1337 lines in the repository with messages stating who checked those files in.
The file secdev20-tutorial/staticcheck contains a 64-bit Linux binary for the Staticcheck tool.
The file secdev20-tutorial/run_staticcheck.sh contains a script that downloads this binary as well as its dependencies and runs it
using the Muse Plugin API.
We will use this tool to analyze the "Gen" project. Go to:
https://github.com/clipperhouse/gen
Fork the repository. In your fork of the repo, add a .muse/ directory containing two files:
run_staticcheck.sh(from thesecdev20-tutorialrepo)config.toml(containing the single linecustomTools = [".muse/run_staticcheck.sh"])
Go to console.muse.dev), find the Gen repo, and click "Analyze". In a few minutes you should see several results.
First we will set up your Muse API Token. Assuming a Bash shell, this can be done with the following.
export MUSEDEV_TOKEN="<token>"
where <token> is the string you obtained when you requested a Muse API Token.
To trigger analsis of Gen from the command line, type:
muscle analyze <gh_username> gen
where <gh_username> is your GitHub username. You should see a job ID printed to the console.
To check status, type:
muscle status <job_id>
When the job has completed (status is "JobCompletedSuccess"), type:
muscle results <job_id>
to get the results in a JSON format.
Load the Jupyter notebook we'll use for our experiments:
cd notebooks
jupyter notebook
A web browser should open with a list of notebooks. Open the "Experiments" notebook. The rest of the tutorial takes place in the "Experiments" notebook and is explained in comments in that notebook.
You can use the check-muse-api.sh script to check that your custom script conforms to the Muse API. Usage is:
./check-muse-api.sh <script_name> <directory>
where <script_name> is the name of your custom script and <directory> is the directory to run it in (whatever directory contains code for it to analyze).
To debug issues with a custom tool, it can be helpful to run the tool in a docker container that matches Muse's analysis environment. To do this, type:
docker run --rm -it -v `pwd`:/code musedev/build-test bash
This will start a docker container that matches Muse's run-time environment and will mount the current directory as /code within that container.
You will get a bash terminal and can now try checking out code, running your custom tool, looking at results, and running the check-muse-api.sh script.