This framework lays out a homogenised foundation for homomorphic and semi-homomorphic development, greatly simplifying the implementation and use of new cryptosystems and oblivious protocols over server-client architecture.
Its heavy use of templating allows flexible combinations of cryptosystems and protocols and to build a full algebra on natural numbers (e.g. by combining an additive cryptosystem and a multiplication remote protocol). The resulting classes let you create and use encrypted values with regular C++ operators.
The framework files are meant to be used inside your own project and should compile without much effort on any modern compilers (C11).
The 'only' two non-standard requirements are:
- Boost library v.1.57 or later
- Gnu Multiple Precision Arithmetic Library v.6 or later
Additionally, to use the client-server protocol classes (RPCClient/RPCServer), you will need the C++ Msgpack-RPC library. If you have difficulties compiling this version, you can try my own fork, which fixed some issues the lib used to have with C11 support (may have been fixed by the time you read this). Similarly, this fork of MPIO (a requirement of msgpack-rpc) might be helpful.
A few usage examples (some of it commented out) are provided in the main.cpp file. Easiest way to compile it is with the Xcode project file (if you are an OS X user), but compiling directly shouldn't be particularly hard (sorry, no Makefile yet).
The code currently provides a fairly fast implementation of Paillier's additive cryptosystem, with a client-server protocol for oblivious multiplication and oblivious vector inner product.
Implementing a new cryptosystem is as easy as creating two subclasses (private key and public key versions) of Cryptosystem, specifying supported operations (addition, multiplication…) as template arguments. The only required methods to implement are enc and dec (for the private key class) and key management methods (getPublicKey and a public key init method). Depending on the other operators supported (specified as template arguments), additional methods such as AddTo or PlainMultiplyTo need to be implemented for the class to compile. While a default cipher format is provided (essentially a very long integer, of type mpz, as supported by GMP), it is possible to also provide your own cipher format.
The best way to implement a new cryptosystem is to take example on the PaillierCryptosystem class, which implements a full Paillier cryptosystem (some of the low-level mathematical code for encryption/decryption is re-used from the LibPaillier library) supporting homomorphic addition and plaintext multiplication.
The base class architecture provides a number of intelligent defaults for all additional operations based on the base operations (such as unary minus operator) and overrides C++ mathematical and stream operators to let you manipulate encrypted variables as if they were clear values (and therefore usable in any template-based libraries that might require such operator support).
To use the server/client architecture, look at the simple examples in main.cpp: little is needed beside an IP address and port number. Once a server/client is created with the desired cryptosystem, it will automatically handle key exchange and run supported oblivious procedures across the network.
You can subclass ppf::Procedure to create your own procedure. Take a look at existing procedures (such as ppf::ProcObliviousMult) for examples of how to do it. Each side can simultaneously run its own client and server in order to provide complex oblivious computaiton protocols.