I have been working quite a bit recently on my BitTorrent implementation. My project is faring reasonably well so far - it can successfully download large real-world torrents at decent speed. In the course of development, I have been focusing solely on the core specification, as documented here. Having gotten support for peer connections - outgoing and incoming - working, I started to notice something. Some peers would send ostensibly random data to my client. Investigation of this led me to conclude these were peers using the Message Stream Encryption protocol.
I am not hugely interested in avoiding traffic shaping, but the encryption protocol looks like fun to implement and I’d like my software to be able to communicate with these funky peers (instead of just killing them, as it does now). I know very little about cryptography, but I know a little more now that I have begun implementing MSE. MSE really has two components - a Diffie-Hellman (D-H) key exchange to negotiate a secret, and RC4 stream cipher to encrypt the data. Both these functions are used in Secure Sockets Layer. However it should be pointed out that MSE is more about obfuscating the data to confuse traffic shapers than hiding its contents from prying eyes.
Fortunately for me, the OpenSSL library contains nice implementations of both D-H key exchange and RC4 . Before I can get to the nitty gritty of actually negotiating obfuscated streams, I had to make considerable changes to my software’s peer handling logic. I had written the peer response handler with quite hard assumptions about the nature of the incoming peer handshakes - for example, assuming the first byte would be the length of the protocol string. Furthermore I needed to add some timekeeping logic for each peer, because MSE handshake depends on timing for synchronisation. The initial peer handshake message is variably padded and so time must be considered to know when the message is complete. For example a MSE handshake from peer A may be considered invalid by peer B a) if A sent less than 96 Bytes within 30 seconds or b) if A sent more than 608 bytes.
My first step has been to re-work my peer handshake handling so that MSE handshakes can be distinguished and separately handled by the software, while traditional handshakes continue to work. Code to do this has been committed. The next step is to actually support crypto peers. For anyone curious about where the source code for my software is, I am not yet ready to release it. I have distributed it to a small group of testers to help iron out some of the bugs, but I don’t feel its ready for public release. I am being somewhat selfish and sitting on my personal project until I have polished it to where I want it to be. If you are really interested though, shoot me an email (niallo (*) niallo dot net).
Tags: BitTorrent, C, OpenBSD, UNIX
Related posts: