Testing for Logout and Browser Cache Management (OWASP-AT-007)
In this phase we check that the logout function is properly implemented, and that it is not possible to “reuse” a session after such logout. We also check that the application automatically logs out a user when that user has been idle for a certain amount of time, and that no sensitive data remains stored in the browser cache.
Description of the Issue
The end a web session is usually triggered by one of the following two events:
- The user logs out
- The user remains idle for a certain amount of time and the application automatically logs him/her out
Both cases must be implemented carefully, in order to avoid to introduce weaknesses that could be exploited by an attacker to gain unauthorized access. More specifically, the logout function must ensure that all session tokens (e.g.: cookies) are properly destroyed or made unusable, and that proper controls are enforced at the server side to forbid them to be used again. If such actions are not properly carried out, an attacker could replay these session tokens in order to “resurrect” the session of a legitimate user and virtually impersonate him/her (this attack is usually known as 'cookie replay'). Of course, a mitigating factor is that the attacker needs to be able to access those tokens (that are stored on the victim PC), but in a variety of cases it might not be too difficult. The most common scenario for this kind of attack is a public computer that is used to access some private information (e.g.: webmail, online bank account, ...): when the user has finished using the application and logs out, if the logout process is not properly enforced the following user could access the same account, for instance by simply pressing the “back” button of the browser. Another scenario can result from a Cross Site Scripting vulnerability or a connection that is not 100% protected by SSL: a flawed logout function would make stolen cookies useful for a much longer time, making life for the attacker much easier. The third test of this chapter is aimed to check that the application forbids the browser to cache sensitive data, which again would pose a danger to an user accessing the application from a public computer.
Black Box testing and examples
The first step is to test the presence of the logout function. Check that the application provides a logout button and that this button is present and well visible on all pages that require authentication to be accessed. A logout button that is not clearly visible, or that is present only on certain pages, poses a security risk, as the user might forget to use it at the end of his/her session.
The second step consists in checking what happens to the session tokens when the logout function is invoked. For instance, when cookies are used a proper behavior is to erase all session cookies, by issuing a new Set-Cookie directive that sets their value to a non-valid one (e.g.: “NULL” or some equivalent value) and, if the cookie is persistent, setting its expiration date in the past, which tells the browser to discard the cookie. So, if the authentication page originally sets a cookie in the following way:
Set-Cookie: SessionID=sjdhqwoy938eh1q; expires=Sun, 29-Oct-2006 12:20:00 GMT; path=/; domain=victim.com
the logout function should trigger a response somewhat resembling the following:
Set-Cookie: SessionID=noauth; expires=Sat, 01-Jan-2000 00:00:00 GMT; path=/; domain=victim.com
If by pressing the 'back' button we can access previous pages but not access new ones then we are simply accessing the browser cache. If these pages contain sensitive data, it means that the application did not forbid the browser to cache it (by not setting the Cache-Control header, a different kind of problem that we will analyze later).
After the “back button” technique has been tried, it's time for something a little more sophisticated: we can re-set the cookie to the original value and check whether we can still access the application in an authenticated fashion. If we can, it means that there is not a server-side mechanism that keeps track of active and non active cookies, but that the correctness of the information stored in the cookie is enough to grant access. To set a cookie to a determined value we can use WebScarab and, intercepting one response of the application, insert a Set-Cookie header with our desired values:
Alternatively, we can install a cookie editor in our browser (e.g.: Add N Edit Cookies in Firefox):
A notable example of a design where there is no control at server side about cookies that belong to users that have already logged out is ASP.NET FormsAuthentication class, where the cookie is basically an encrypted and authenticated version of the user details that are decrypted and checked by the server side. While this is very effective in preventing cookie tampering, the fact that the server does not maintain an internal record of the session status means that it is possible to launch a cookie replay attack after the legitimate user has logged out, provided that the cookie has not expired yet (see the references for further detail).
It should be noted that this test only applies to session cookies, and that a persistent cookie that only stores data about some minor user preferences (e.g.: site appearance) and that is not deleted when the user logs out is not to be considered a security risk.
The same approach that we have seen in the previous section can be applied when measuring the timeout logout. The most appropriate logout time should be a right balance between security (shorter logout time) and usability (longer logout time) and heavily depends on the criticality of the data handled by the application. A 60 minutes logout time for a public forum can be acceptable, but such a long time would be way too much in a home banking application. In any case, any application that does not enforce a timeout-based logout should be considered not secure, unless such a behavior is addressing a specific functional requirement. The testing methodology is very similar to the one outlined in the previous paragraph. First we have to check whether a timeout exists, for instance logging in and then killing some time reading some other Testing Guide chapter, waiting for the timeout logout to be triggered. As in the logout function, after the timeout has passed all session tokens should be destroyed or be unusable. We also need to understand whether the timeout is enforced by the client or by the server (or both). Getting back to our cookie example, if the session cookie is non-persistent (or, more in general, the session token does not store any data about the time) we can be sure that the timeout is enforced by the server. If the session token contains some time related data (e.g.: login time, or last access time, or expiration date for a persistent cookie), then we know that the client is involved in the timeout enforcing. In this case, we need to modify the token (if it's not cryptographically protected) and see what happens to our session. For instance, we can set the cookie expiration date far in the future and see whether our session can be prolonged. As a general rule, everything should be checked server-side and it should not be possible, re-setting the session cookies to previous values, to be able to access the application again.
Logging out from an application obviously does not clear the browser cache of any sensitive information that might have been stored. Therefore, another test that is to be performed is to check that our application does not leak any critical data into the browser cache. In order to do that, we can use WebScarab and search through the server responses that belong to our session, checking that for every page that contains sensitive information the server instructed the browser not to cache any data. Such a directive can be issued in the HTTP response headers:
HTTP/1.1: Cache-Control: no-cache
HTTP/1.0: Pragma: no-cache Expires: <past date or illegal value (e.g.: 0)>
Alternatively, the same effect can be obtained directly at the HTML level, including in each page that contains sensitive data the following code:
HTTP/1.1: <META HTTP-EQUIV="Cache-Control" CONTENT="no-cache">
HTTP/1.0: <META HTTP-EQUIV="Pragma" CONTENT="no-cache"> <META HTTP-EQUIV=”Expires” CONTENT=”Sat, 01-Jan-2000 00:00:00 GMT”>
For instance, if we are testing an e-commerce application, we should look for all pages that contain a credit card number or some other financial information, and check that all those pages enforce the no-cache directive. On the other hand, if we find pages that contain critical information but that fail to instruct the browser not to cache their content, we know that sensitive information will be stored on the disk, and we can double-check that simply by looking for it in the browser cache. The exact location where that information is stored depends on the client operating system and on the browser that has been used, but here are some examples:
- Mozilla Firefox:
- Unix/Linux: ~/.mozilla/firefox/<profile-id>/Cache/
- Windows: C:\Documents and Settings\<user_name>\Local Settings\Application Data\Mozilla\Firefox\Profiles\<profile-id>\Cache>
- Internet Explorer:
- C:\Documents and Settings\<user_name>\Local Settings\Temporary Internet Files>
Gray Box testing and example
The gray box testing is not much different from the black box one. In a gray box testing we can assume we have some partial knowledge about the session management of our application, and that should help us in understanding whether the logout and timeout functions are properly secured. As a general rule, we need to check that:
- The logout function effectively destroys all session token, or at least render them unusable
- The server performs proper checks on the session state, disallowing an attacker to replay some previous token
- A timeout is enforced and it is properly checked by the server. If the server uses an expiration time that is read from a session token that is sent by the client, the token must be cryptographically protected
For the secure cache test, the methodology is equivalent to the black box case, as in both scenarios we have full access to the server response headers and to the HTML code.
- ASP.NET Forms Authentication - Best Practices for Software Developers: http://www.foundstone.com/resources/whitepapers/ASPNETFormsAuthentication.pdf
- The FormsAuthentication.SignOut method does not prevent cookie reply attacks in ASP.NET applications: http://support.microsoft.com/default.aspx?scid=kb;en-us;900111
OWASP Testing Guide v2