I’m quite pleased to announce that TLS 1.0 is now available. It took 9 months of brewing in its own branch, to get the last missing protocol bits implemented and much more.
Here is the list of features in a short changelog
- Client certificates, courtesy of Martin Grabmuller.
- NPN extension support, courtesy of Lennart Kolmodin.
- ServerName extension.
- New Certificate Store mechanism (certificate-1.3).
- Major refactoring of code.
Most if not all of the implementation is thanks to Martin here; He diligently implemented the missing bits, so that now the client can provides certificates, and that the server can ask for it. It also complete the last important handshake message that was still missing in TLS 0.9. Client certificates provides a way to authenticate using X509 certificate, so that a TLS stream can be mutually authenticated.
The usual way is just to authenticate the server side, to be sure who you are talking to, and have client authentication done at a different level (for example imaps sending a user and password). Client certificate allows the client to bring the “something i have” security dimension (a certificate), along or instead of the “something i know” (e.g. a password).
Client certificate use strong public key cryptography to allows authentications, commonly using RSA or DSA signing.
Next Protocol Negotiation Extension
Next protocol negotiation (NPN) is a google extension to TLS, that allow a client to tell the server to which service this stream is suppose to connect.
It has been designed as the basis of SPDY (a fast web protocol to fix some issue with HTTP), to connect to a normal HTTPS server, but negotiate the new SPDY protocol dynamically at the TLS handshake level.
The extension itself is completely generic; you could design a single point of entry for all your services on a server, hidden in a generic TLS stream, on a generic port (for example HTTPS). This is quite likely to send some corporate network admins chills through their spines, as no-one except the client and the server at the destination can dictates what is allowed or denied. Filtering on destination ports (imap = 143, etc), with this kind of capability, will be thereafter utterly pointless.
The certificate store is a new feature in the certificate 1.3 package. The end goal is to make certification better and faster; For now the implementation is quite simple, it preload all trusted certificates from the system and indexed them with how we look for them.
The previous approch was, at verification time, we’ll list all the trusted certificates, parse them one by one, until we find (or not) the one we’re suppose to match. This is a very costy process and also very wasteful as this is akin to searching in a list (with the list element coming from the disk), which is o(n). The certificate store even with the simplest implementation is able to do it in o(1) with, a constant and one-time o(n) preload.
The benefits is also in the future, as the design allows combining implementations together (it’s a Monoid !), so it will be easy to provides intelligent caches or better finding algorithm mechanism.
One such thing is using the algorithm openssl uses, which use sha1 shorten hashes symlinks. The hash mechanism is already implemented in certificate 1.3 for a better (future) certificate store implementation.
Server Name Extension
One minor extension has been implemented that provides the ability to specify a server name in the handshake. This is useful for the HTTP service, so that each virtual host can provides their own certificates with the name matching.
It’s not very used in general unfortunately, so it’s likely not going to make a difference to anyone. For more information Server Name Indication
The major restructuration allows the stack to be more flexible, better documented, and better structured. There is now more files, however they are generally smaller, and organized in folders. For example everything handshake related, is to be found in the Network/TLS/Handshake/ folder. Unfortunately this restructuration is also bringing some interface changes. It should make things better in the long run, and I tried to bundle all the breaking changes in this version, so transition to next versions would be hopefully smoother.
One last thing is some repositories of the tls stack have been merged together, so there’s now only one repository for tls, tls-extra and tls-debug.
That’s all for now, and now it’s time to look at the next version already. Enjoy, and do report bugs, improvements, and suggestions to the usual place.