Date: Fri Jan 13 09:05:30 2006
From: alla at scanit.be (Alla Bezroutchko)
Subject: Session data pollution vulnerabilities in	web
	applications

Frank Knobbe wrote:
> The proposed fix is -- besides being only specific to this example --
> equally flawed. The underlying issue is that you trust user supplied
> data. When a user supplies a user name for login purposes, you should
> only use that input to perform a search in your database. If a match has
> been found, take a *trusted* value from your database, for example an
> index ID, and carry that in the session object to identify the user.

Suppose in my example resetpw3.php instead of doing

$_SESSION['login'] = $_POST['login'];

did

$_SESSION['login'] = $db->getOne("SELECT login FROM users WHERE login=? 
AND secret_answer=?", array($_POST['login'], $_POST['secret_answer']));

As you suggest it takes a trusted value from the database. It is still 
does not prevent using register2.php to initialize $_SESSION['login'] 
and then jump to resetpw4.php and have it use the value to change password.

> In other words, don't accept any user input (even after proper input
> validation) and carry it as trusted data in your session object. Don't
> base further decisions on this data. Since it is user supplied it can
> not be trusted.

Are you saying that user supplied data should never ever be stored in 
the session? Where do you suggest to keep the data that needs to be 
passed between different pages in the application?

Say you have a form that is filled in in several steps, each step being 
a separate page. Where would you store the data that the user entered to 
have it all available on the last page? I can only think of carrying it 
around in hidden form fields, and this is not always possible.

> Your example is further flawed in that it allows the change of a
> password without being properly authenticated. Just having a valid
> 'login' session object present doesn't indicate that the user is
> authenticated. 

I think it does, if 'login' session object is only ever set by the 
authentication procedure after verifying user's credentials. I believe 
the usual approach to authentication in web applications is to check 
user's credentials and then store something in the session that 
indicates that the credentials were correct. Usually it is something 
like a user ID or an object or JavaBean storing user information. Then 
every pafge that needs to check auth just checks if the session object 
is present and valid.

How do applications that you encounter check if the user is authenticated?

> I really, really hope this was just an example you made
> up, and not something you actually saw being used. If so, go back to
> that web site with a clue-by-four and give it a few whacks.

I work for a company that does penetration testing, vulnerability 
assessments, and stuff like that. The contracts we usually sign do not 
authorise us to apply blunt instruments to customers' web sites. :)

This particular example is made up. Giving real ones would violate 
non-diclosure agrements. It is similar to the real ones in the way that 
it has the same problem - one storage location is used to keep both 
trusted and untrusted data. I used the password reset example because 
the security implications are clear and one don't need application 
specific context to understand it. The bug does not have to do anything 
with authentication as such.

And, in case you are wondering, the real ones were just as bad. :)

Alla.

Powered by blists - more mailing lists