Channel Bound Cookies
The following Proposed RFC comes from a Google employee and describes a way for Clients use a self-signed Browser Certificate (thus requiring no confusing "pop-up" for the end user) which can also address the cookie security issue known as "Related Domain Cookies"
What follows below is an extract of http://www.browserauth.net/ , a section of the RFC, some interesting commentary, and some criticism on this extension.
Overview of Channel Bound Cookies
Once the underlying TLS channel uses TLS client authentication (with the TLS-OBC extension), the server can bind its cookies to the TLS channel by associating them with the client's public key, and ensuring that the cookies are only ever used over TLS channels authenticated with that public (client) key.
This means that if such a channel-bound cookie is ever stolen off a client's machine, that cookie won't be able to authenticate an HTTP session to the server from other machines. This includes man-in-the-middle attackers that inject themselves into the connection between client and server, perhaps by tricking users into clicking through certificate-mismatch warnings: such a man-in-the-middle will have to generate its own TLS session with the server, which won't match the channel that the cookie is bound it.
Channel Binding
It's up to the server to decide whether to bind cookies to TLS channels. If the client doesn't support TLS-OBC, or if the cookie it's about to set will be used across different origins, then the server will not channel-bind the cookie. If it does decide to channel-bind the cookie, it should associate the cookie with the client's public key. This is similar to RFC 5929, but instead of the client binding data to the server's public key, in this case the server would be binding data (the cookie) to the client's public key. The server can do this either by simply storing, in a backend database, the fact that a certain HTTP session is expected to be authenticated with a certain client public key, or it can use suitable cryptography to encode in the cookie itself which TLS client public key that cookie is bound to.
In the figure above, the server includes the client's public key into a cryptographically signed datastructure that also includes the authenticated user's id. When the server receives the cookie back from the client, it can verify that it indeed issued the cookie (by checking the signature on the cookie), and verify that the cookie was sent over the correct channel (by matching the TLS client key with the key mentioned in the cookie).
To be continued here.... http://www.browserauth.net/channel-bound-cookies
RFC Snip
TLS Origin-Bound Certificates RFC Draft
(Excerpt)
4.3. Cookie Hardening
One way TLS-OBC can be used to strengthen cookie-based
authentication is by "binding" cookies to an origin-bound
certificate. The server, when issuing a cookie for an HTTP
session, would associate the client's origin-bound certificate with
the session (either by encoding information about the certificate
unforgeably in the cookie, or by associating the certificate with
the cookie's session through some other means). That way, if and
when a cookie gets stolen from a client, it cannot be used over a
TLS connection initiated by a different client - the cookie thief
would also have to steal the private key associated with the
client's origin-bound certificate, a task considerably harder
especially when we assume the existence of a Trusted Platform
Module or other Secure Element that can store the
origin-bound-certificate's private key.
Additional Commentary from [email protected]
Also, note that somewhat counter-intuitively, channel-bound cookies protect against many related-domain attacks even if the client cert that they are bound to has broader scope than a web origin.
Imagine, for a moment, that a user-agent creates a single self-signed certificate that it uses as a TLS client cert for all connections to all servers (not a good idea in terms of privacy, but follow me along for this thought experiment). The servers then set cookies on their respective top-level domains, but channel-bind them to the user-agent's one-and-only client cert.
So, let's say that an app app.heroku.com sets a (channel-bound) cookie on my browser for domain .heroku.com, and that there is an attacker on attacker.heroku.com. One attack we might be concerned about is that the attacker simply harvests the .heroku.com cookie from my browser by luring me to attacker.heroku.com. They won't be able to actually use the cookie, however, because the cookie is channel-bound to my browser's client cert, not to the attacker's client cert.
Another attack we might be concerned about is that attacker.heroku.com sets an .heroku.com cookie on my user agent in order to make me log into app.heroku.com as himself. Again, assuming that the only way the attacker can obtain the cookies is by getting them from app.heroku.com, this means that the cookies he has at his disposal will be channel-bound to his client cert, not to my client cert - thus when my browser sends them to app.heroku.com they won't be valid.
The TLS-OBC proposal, of course, assumes more fine-grained "scopes" for the client certificates. The reason for that, however, is purely to prevent tracking across unrelated domains. Related-domain attacks are already mitigated even if we used coarse-grained client certificates and coarse-grained (i.e., domain) cookies. I, at least, found this a little counter-intuitive at first, since the other proposed defense it to forbid coarse-grained cookies altogether and use origin cookies instead.
Criticism from [email protected]
There are a number of issues that need to be considered for TLS-OBC; I'll highlight a couple here that I'm aware of.
Some SSL handshake logic may need to be modified slightly; see https://bugzilla.mozilla.org/show_bug.cgi?id=681839 for technical discussion.
There are potential privacy considerations; in particular if the unique client certificate is sent in cleartext before the negotiation of the master secret, a passive network observer may be able to uniquely identify a client machine. The attacker would already have the client's IP address, so this isn't a huge problem if the certificate is regenerated on an IP address change, but that would nullify much of the authentication benefit. A proposal to allow a client certificate to be sent after the master secret negotiation has been made. (can't find the bug right now, sorry)
One proposal how #2 could be addressed is here: https://datatracker.ietf.org/doc/html/draft-agl-tls-encryptedclientcerts
There are tricky interactions with SPDY. There will be updates on browserauth.net for this.
