Repository files navigation

A vulnerable Node.js demo application, based on theDreamers Lab tutorial.

This vulnerable app includes the following capabilities to experiment with:

• Exploitable packages with known vulnerabilities
• Docker Image Scanning for base images with known vulnerabilities in system libraries
• Runtime alerts for detecting an invocation of vulnerable functions in open source dependencies

mongod&git clone https://github.com/snyk-labs/nodejs-goofnpm installnpm start

This will run Goof locally, using a local mongo on the default port and listening on port 3001 (http://localhost:3001)

Note: Youhave to use an old version of MongoDB version due to some of these old libraries' database server APIs. MongoDB 3 is known to work ok.

You can also run the MongoDB server individually via Docker, such as:

docker run --rm -p 27017:27017 mongo:3

docker-compose up --builddocker-compose down

Goof requires attaching a MongoLab service to be deployed as a Heroku app.That sets up the MONGOLAB_URI env var so everything after should just work.

Goof requires attaching a MongoLab service and naming it "goof-mongo" to be deployed on CloudFoundry.The code explicitly looks for credentials to that service.

To bulk delete the current list of TODO items from the DB run:

npm run cleanup

This app uses npm dependencies holding known vulnerabilities,as well as insecure code that introduces code-level vulnerabilities.

Theexploits/ directory includes a series of steps to demonstrate each one.

Here are the exploitable vulnerable packages:

• Mongoose - Buffer Memory Exposure - requires a version <= Node.js 8. For the exploit demo purposes, one can update the Dockerfilenode base image to useFROM node:6-stretch.
• st - Directory Traversal
• ms - ReDoS
• marked - XSS

• Open Redirect
• NoSQL Injection
• Code Injection
• Command execution
• Cross-site Scripting (XSS)
• Information exposure via Hardcoded values in code
• Security misconfiguration exposes server information
• Insecure protocol (HTTP) communication

The page at/account_details is rendered as an Handlebars view.

The same view is used for both the GET request which shows the account details, as well as the form itself for a POST request which updates the account details. A so-called Server-side Rendering.

The form is completely functional. The way it works is, it receives the profile information from thereq.body and passes it, as-is to the template. This however means, that the attacker is able to control a variable that flows directly from the request into the view template library.

You'd think that what's the worst that can happen because we use a validation to confirm the expected input, however the validation doesn't take into account a new field that can be added to the object, such aslayout, which when passed to a template language, could lead to Local File Inclusion (Path Traversal) vulnerabilities. Here is a proof-of-concept showing it:

curl -X'POST' --cookie c.txt --cookie-jar c.txt -H'Content-Type: application/json' --data-binary'{"username": "admin@snyk.io", "password": "SuperSecretPassword"}''http://localhost:3001/login'

curl -X'POST' --cookie c.txt --cookie-jar c.txt -H'Content-Type: application/json' --data-binary'{"email": "admin@snyk.io", "firstname": "admin", "lastname": "admin", "country": "IL", "phone": "+972551234123",  "layout": "./../package.json"}''http://localhost:3001/account_details'

Actually, there's even another vulnerability in this code.Thevalidator library that we use has several known regular expression denial of service vulnerabilities. One of them, is associated with the email regex, which if validated with the{allow_display_name: true} option then we can trigger a denial of service for this route:

curl -X'POST' -H'Content-Type: application/json' --data-binary"{\"email\":\"`seq -s"" -f"<" 100000`\"}"'http://localhost:3001/account_details'

Thevalidator.rtrim() sanitizer is also vulnerable, and we can use this to create a similar denial of service attack:

curl -X'POST' -H'Content-Type: application/json' --data-binary"{\"email\":\"someone@example.com\",\"country\":\"nop\",\"phone\":\"0501234123\",\"lastname\":\"nop\",\"firstname\":\"`node -e'console.log(" ".repeat(100000) + "!")'`\"}"'http://localhost:3001/account_details'

A POST request to/login will allow for authentication and signing-in to the system as an administrator user.It works by exposingloginHandler as a controller inroutes/index.js and uses a MongoDB database and theUser.find() query to look up the user's details (email as a username and password). One issue is that it indeed stores passwords in plaintext and not hashing them. However, there are other issues in play here.

We can send a request with an incorrect password to see that we get a failed attempt

echo'{"username":"admin@snyk.io", "password":"WrongPassword"}'| http --json$GOOF_HOST/login -v

And another request, as denoted with the following JSON request to sign-in as the admin user works as expected:

echo'{"username":"admin@snyk.io", "password":"SuperSecretPassword"}'| http --json$GOOF_HOST/login -v

However, what if the password wasn't a string? what if it was an object? Why would an object be harmful or even considered an issue?Consider the following request:

echo'{"username": "admin@snyk.io", "password": {"$gt": ""}}'| http --json$GOOF_HOST/login -v

We know the username, and we pass on what seems to be an object of some sort.That object structure is passed as-is to thepassword property and has a specific meaning to MongoDB - it uses the$gt operation which stands forgreater than. So, we in essence tell MongoDB to match that username with any record that has a password that is greater thanempty string which is bound to hit a record. This introduces the NoSQL Injection vector.

The/admin view introduces aredirectPage query path, as follows in the admin view:

<input type="hidden" name="redirectPage" value="<%- redirectPage %>" />

One fault here is that theredirectPage is rendered as raw HTML and not properly escaped, because it uses<%- > instead of<%= >. That itself, introduces a Cross-site Scripting (XSS) vulnerability via:

http://localhost:3001/login?redirectPage="><script>alert(1)</script>

To exploit the open redirect, simply provide a URL such asredirectPage=https://google.com which exploits the fact that the code doesn't enforce local URLs inindex.js:72.

The application initializes a cookie-based session onapp.js:40 as follows:

app.use(session({secret:'keyboard cat',name:'connect.sid',cookie:{secure:true}}))

As you can see, the sessionsecret used to sign the session is a hardcoded sensitive information inside the code.

First attempt to fix it, can be to move it out to a config file such as:

module.exports={cookieSecret:`keyboard cat`}

And then require the configuration file and use it to initialize the session.However, that still maintains the secret information inside another file, and Snyk Code will warn you about it.

Another case we can discuss here in session management, is that the cookie setting is initialized withsecure: true which means it will only be transmitted over HTTPS connections. However, there's nohttpOnly flag set to true, which means that the default false value of it makes the cookie accessible via JavaScript. Snyk Code highlights this potential security misconfiguration so we can fix it. We can note that Snyk Code shows this as a quality information, and not as a security error.

Snyk Code will also find hardcoded secrets in source code that isn't part of the application logic, such astests/ orexamples/ folders. We have a case of that in this application with thetests/authentication.component.spec.js file. In the finding, Snyk Code will tag it asInTest,Tests, orMock, which help us easily triage it and indeed ignore this finding as it isn't actually a case of information exposure.

TheDockerfile makes use of a base image (node:6-stretch) that is known to have system libraries with vulnerabilities.

To scan the image for vulnerabilities, run:

snyktest --docker node:6-stretch --file=Dockerfile

To monitor this image and receive alerts with Snyk:

snyk monitor --docker node:6-stretch

Snyk provides the ability to monitor application runtime behavior and detect an invocation of a function is known to be vulnerable and used within open source dependencies that the application makes use of.

The agent is installed and initialized inapp.js.

For the agent to report back to your snyk account on the vulnerabilities it detected it needs to know which project on Snyk to associate with the monitoring. Due to that, we need to provide it with the project id through an environment variableSNYK_PROJECT_ID

To run the Node.js app with runtime monitoring:

SNYK_PROJECT_ID=<PROJECT_ID> npm start

** The app will continue to work normally even if it's not provided a project id

To find these flaws in this application (and in your own apps), run:

npm install -g snyksnyk wizard

In this application, the defaultsnyk wizard answers will fix all the issues.When the wizard is done, restart the application and run the exploits again to confirm they are fixed.