We recently updated our solution to MVC 2, and this has updated the way that the AntiForgeryToken works. Unfortunately this does not fit with our AJAX framework any more.
The problem is that MVC 2 now uses symmetric encryption to encode some properties about the user, including the user's Name property (from IPrincipal). We are able to securely register a new user using AJAX, after which subsequent AJAX calls will be invalid as the anti forgery token will change when the user has been granted a new principal. There are also other cases when this may happen, such as a user updating their name etc.
My main question is why does MVC 2 even bother using symmetric encryption? And then why does it care about the user name property on the principal?
If my understanding is correct then any random shared secret will do. The basic principle is that the user will be sent a cookie with some specific data (HttpOnly!). This cookie is then required to match a form variable sent back with each request that may have side effects (POST's usually). Since this is only meant to protect from cross site attacks it is easy to craft up a response that would easily pass the test, but only if you had full access to the cookie. Since a cross site attacker is not going to have access to your user cookies you are protected.
By using symmetric encryption, what is the advantage in checking the contents of the cookie? That is, if I already have sent an HttpOnly cookie the attacker cannot override it (unless a browser has a major security issue), so why do I then need to check it again?
After having a think about it it appears to be one of those 'added layer of security' cases - but if your first line of defence has fallen (HttpOnly) then the attacker is going to get past the second layer anyway as they have full access to the users cookie collection, and could just impersonate them directly, instead of using an indirect XSS/CSRF attack.
Of course I could be missing a major issue, but I haven't found it yet. If there are some obvious or subtle issues at play here then I would like to be aware of them.
It was added to offer greater protection in the case where you have one subdomain trying to attack another - bad.example.com trying to attack good.example.com. Adding the username makes it more difficult for bad.example.com to contact good.example.com behind the scenes and try to get it to generate a token on your behalf.
Going forward, it's possible that the cookie will be removed as it's not strictly necessary for the proper functioning of the system. (For example, if you're using Forms Authentication, that cookie could serve as the anti-XSRF cookie instead of requiring the system to generate a second cookie.) The cookie might only be issued in the case of anonymous users, for example.
Besides the "evil subdomain"-scenario outlined by Levi, consider an attacker that has an account on the targeted site. If the CSRF-token does not encode user-specific information, the server can not verify that the token has been generated exclusively for the logged-in user. The attacker could then use one of his own legitimately acquired CSRF-tokens when building a forged request.
That being said, anonymous tokens are during certain circumstances accepted by ASP.NET MVC. See Why does ValidateAntiForgeryTokenAttribute allow anonymous tokens?
来源：https://stackoverflow.com/questions/2697772/mvc-2-antiforgerytoken-why-symmetric-encryption-iprinciple