search

GaloisInc / swanky

A suite of rust libraries for secure multi-party computation
MIT License
273 stars 57 forks source link

Tests do not pass with alternative AES circuits. #14

Open themighty1 opened 2 years ago

themighty1 commented 2 years ago

Hi, I tried to plug in the aes_128.txt circuit from https://homes.esat.kuleuven.be/~nsmart/MPC/ by first "downgrading" its header into the old format:

36663 36919
128 128  128

and also changing key,msg input ordering and reversing the output bits.

It passed the first 2 tests in parser.rs but failed on the 3rd: https://github.com/GaloisInc/swanky/blob/4455754abadee07f168079ac45ef33535b0df27d/fancy-garbling/src/parser.rs#L191-L196

Then I took another AES circuit from here https://github.com/n-for-1-auth/circuits/tree/main/aes (aes128_full.txt) , adjusted input/output ordering and set its header to

36548 36804
128 128 128

this time the test failed on test2

How can such non-deterministic behaviour happen? Does swanky choose different code paths based on the gate ordering of the circuit? Either way, you probably want to look at this as some corner case is triggering a bug.

amaloz commented 2 years ago

Interesting... is the only format difference between the two formats the header, or could there be something else in the new format that's causing problems when downgrading to the old format?