[tor-dev] Format-Transforming Encryption for Censorship Circumvention

[Kevin P Dyer]

Following my email to this list, dated 29/07/2012, I direct your
attention to the IACR eprint document http://eprint.iacr.org/2012/494,
"Format-Transforming Encryption: More than Meets the DPI". In it, we
present our framework for steganographic encoding of messages using
regular languages, along with initial findings for a Python/C++ based
implementation.

To briefly highlight some of the main things we deliver:

* A record-layer, powered by format-transforming encryption, that can
tunnel arbitrary SOCKS streams.

* Algorithms that allow efficient and invertible mapping between bit
strings and elements of a regular language L. These allow us to encode
traditional encryptions of plaintexts into strings from L in a way
that maximizes the number of bits that are encoded.

* A framework for the creation of good regular languages, "good" with
respect to performance and security. In particular, our
language-learning framework produces sets of regular expressions
(compact representations of the languages) that are learned from real
traffic.  Our system also allows you to use off-the-shelf regular
expressions, like those from appid [1] and l7-filter [2] that are
designed to detect HTTP traffic.

We consider this preliminary technical report on a work-in-progress.
In fact, we can already do more than what is reported (e.g. greater
variety of languages, better system performance), but we will save
these advances for a future release.

Cheers,
Kevin P Dyer (and his co-authors)

[1] http://code.google.com/p/appid/
[2] http://l7-filter.sourceforge.net/