Cross Site Request Forgery in Tumblr "Ask a Question" feature

This might be the first nerd-blog post in a long time, so my apologies if I startle anyone. As a quick update, I wrote a book on security in Drupal and then founded a company focused on Drupal security services and then sold that company to Acquia where I currently work.

So....Tumblr is a big deal. They apparently have 120million users and are totally awesome.

I was trying Tumblr out for a site and noticed that many of their interactions relied 100% on Javascript for a "delete/cancel" confirmation. I fired up my handy-dandy browser tools and inspected the http headers associated witih deleting a message. Turns out that it's vulnerable to a cross-site-request-forgery. In general Tumblr uses the token-synchronizer csrf prevention (as documented on OWASP). I'm not necessarily saying they copiedOWASP or were inspired, just that it follows the pattern of using a second form token that is sent on all requests for a session. They do not use a different token per form/action: once you get the anti-CSRF-nonce it's the same for multiple different operations. Their token is called the form_key.

Want to see the problem in action? I even made this handy dandy movie of the problem:

It's also worth noting that it was relatively difficult to figure out how to report a security issue to their engineering team. In the end I just sent a support message to tumblr via the normal support channel which got an "auto-deflect" response. I had to re-open it by responding to get a human to look at it. IMO, if you have a web property with more than a few hundred users, you should have a policy around submitting security issues that is really dang easy for people to follow.