Saturday, September 18, 2010

Asp.Net critical security vulnerability


Edit: Some information in the original version of this post was not correct. I have fixed it.

A few hours ago Microsoft had publicly disclosed a critical Asp.Net security vulnerability. In short, a hacker can get sensitive data from an Asp.Net application in a few minutes. This information includes any file on the server (e.g. web.config) and any viewstate data even if it is encrypted. To block this attack, site owners should immediately direct all errors to a single page (custom errors redirect).

If you have a production Asp.Net system read this now and go fix it.

How does the vulnerability work?

Without making this a walkthrough to my hacker readers, this is the general flow:

Some common encryption algorithms (among them TripleDes and AES) divide the encrypted data to blocks of equal size, where the last block is padded with additional data to make all blocks equal. A few years ago a security researcher had published an article explaining how to hack such an algorithm provided that a "padding oracle" exists. What is a padding oracle? It is an application (or some abstract... oracle) which tels us if the padding of an encrypted data is in the correct format.

Well, it turns out that Asp.Net uses TripleDes, and that some specific feature of Asp.Net turns it into a "padding oracle", you get the idea...

How to avoid this?

1. The single most important rule is this:

When there is an error in your application, never let the end user know what it was. In particular never allow an end user to distinguish between 404 and 500 status codes.

The way to do it is to redirect all errors to one page.
ScottGu's post explains how to do it.

2. The hacker that found this hole wrote his in this twitter account:

error message setting is irrelevant. no error? there's always HTTP status. always the same HTTP status? there's always big timing different

He means that even though we hide from him the real Http status (by redirecting to a single error page) the real status still "leaks" by computing the response time.

This means that adding a random wait before sending the error page can help us. ScottGu's post has the details too.

If you are really interested in the bits & bytes (&blocks) of this kind of attacks this is a nice introduction.


What's next? get this blog rss updates or register for mail updates!