cosmian-lib-sgx 0.2.1a3

Python library for code providers using Cosmian Secure Computation

Cosmian lib SGX

Overview

Python library for code providers using Cosmian Secure Computation.

Example

A simple structure of the code should look like:

$ tree .
.
├── secret_mod.py
└── run.py

where run.py is called the entrypoint and is the only file which can't be encrypted.

The goal of the entrypoint is to manage the I/O of you program whereas other module can hide the Python code you want to keep secret.

Here is an example of a statistical mean code where:

• Code Provider hides the whole module secret_mod which is decrypted in the SGX enclave
• Data Providers send bytes of an arbitrary long integer in big-endian as data input for the Code Provider
• Result Consumers receive bytes of the statistical mean of all input datas represented as IEEE 754 binary64

"""run module."""

from io import BytesIO
import struct
from typing import Iterator

from cosmian_lib_sgx import Enclave


def convert_input(datas: Iterator[BytesIO]) -> Iterator[int]:
    """Transform input data bytes to integer."""
    for data in datas:  # type: BytesIO
        yield int.from_bytes(data.read(), byteorder="big")


def main() -> int:
    """Entrypoint of your code."""
    with Enclave() as enclave:
        # import your ciphered module normally
        import secret_mod

        # convert input data bytes sent by Data Providers
        datas: Iterator[int] = convert_input(enclave.read())

        # apply function of your module on datas
        output: float = secret_mod.custom_mean(datas)

        # convert output result
        result: bytes = struct.pack("d", output)

        # write result for Result Consumers
        enclave.write(result)

    return 0


if __name__ == "__main__":
    main()

"""secret_mod module."""

from typing import Iterator


def custom_mean(input_datas: Iterator[int]) -> float:
    """Statistical mean of input datas."""
    n: int = 0
    mean: float = 0.0

    for x in input_datas:  # type: int
        n += 1
        mean += (x - mean) / n
    
    return mean if n > 0 else float("nan")