Anonymous Smart Contracts

Every full node is a verifier.

Prover is the person executing the smart contract function on their secret witness data. They are also verifiers in our model.

Lets take a pseudocode smart contract:

contract Dao {
    # 1: the DAO's global state
    dao_bullas = DaoBulla[]
    proposal_bullas = ProposalBulla[]
    proposal_nulls = ProposalNull[]

    # 2. a public smart contract function
    #    there can be many of these
    fn mint(...) {
        ...
    }

    ...
}

Important Invariants

  1. The state of a contract (the contract member values) is globally readable but only writable by that contract's functions.
  2. Transactions are atomic. If a subsequent contract function call fails then the earlier ones are also invalid. The entire tx will be rolled back.
  3. foo_contract::bar_func::validate::state_transition() is able to access the entire transaction to perform validation on its structure. It might need to enforce requirements on the calldata of other function calls within the same tx. See DAO::exec().

Global Smart Contract State

Internally we represent this smart contract like this:

mod dao_contract {
    // Corresponds to 1. above, the global state
    struct State {
        dao_bullas: Vec<DaoBulla>,
        proposal_bullas: Vec<ProposalBulla>,
        proposal_nulls: Vec<ProposalNull>
    }

    // Corresponds to 2. mint()
    // Prover specific
    struct MintCall {
        ...
        // secret witness values for prover
        ...
    }

    impl MintCall {
        fn new(...) -> Self {
            ...
        }

        fn make() -> FuncCall {
            ...
        }
    }

    // Verifier code
    struct MintParams {
        ...
        // contains the function call data
        ...
    }
}

There is a pipeline where the prover runs MintCall::make() to create the MintParams object that is then broadcast to the verifiers through the p2p network.

The CallData usually is the public values exported from a ZK proof. Essentially it is the data used by the verifier to check the function call for DAO::mint().

Atomic Transactions

Transactions represent several function call invocations that are atomic. If any function call fails, the entire tx is rejected. Additionally some smart contracts might impose additional conditions on the transaction's structure or other function calls (such as their call data).

/// A Transaction contains an arbitrary number of `ContractCall` objects,
/// along with corresponding ZK proofs and Schnorr signatures.
#[derive(Debug, Clone, Eq, PartialEq, SerialEncodable, SerialDecodable)]
pub struct Transaction {
    /// Calls executed in this transaction
    pub calls: Vec<ContractCall>,
    /// Attached ZK proofs
    pub proofs: Vec<Vec<Proof>>,
    /// Attached Schnorr signatures
    pub signatures: Vec<Vec<Signature>>,
}

Function calls represent mutations of the current active state to a new state.

/// A ContractCall is the part of a transaction that executes a certain
/// `contract_id` with `data` as the call's payload.
#[derive(Debug, Clone, Eq, PartialEq, SerialEncodable, SerialDecodable)]
pub struct ContractCall {
    /// ID of the contract invoked
    pub contract_id: ContractId,
    /// Call data passed to the contract
    pub data: Vec<u8>,
}

The contract_id corresponds to the top level module for the contract which includes the global State.

The func_id of a function call corresponds to predefined objects in the submodules:

  • Builder creates the anonymized CallData. Ran by the prover.
  • CallData is the parameters used by the anonymized function call invocation. Verifiers have this.
  • state_transition() that runs the function call on the current state using the CallData.
  • apply() commits the update to the current state taking it to the next state.

An example of a contract_id could represent DAO or Money. Examples of func_id could represent DAO::mint() or Money::transfer().

Each function call invocation is ran using its own state_transition() function.

mod dao_contract {
    ...

    // DAO::mint() in the smart contract pseudocode
    mod mint {
        ...

        fn state_transition(states: &StateRegistry, func_call_index: usize, parent_tx: &Transaction) -> Result<Update> {
            // we could also change the state_transition() function signature
            // so we pass the func_call itself in
            let func_call = parent_tx.func_calls[func_call_index];
            let call_data = func_call.call_data;
            // It's useful to have the func_call_index within parent_tx because
            // we might want to enforce that it appears at a certain index exactly.
            // So we know the tx is well formed.

            ...
        }
    }
}

The state_transition() has access to the entire atomic transaction to enforce correctness. For example chaining of function calls is used by the DAO::exec() smart contract function to execute moving money out of the treasury using Money::transfer() within the same transaction.

Additionally StateRegistry gives smart contracts access to the global states of all smart contracts on the network, which is needed for some contracts.

Note that during this step, the state is not modified. Modification happens after the state_transition() is run for all function call invocations within the transaction. Assuming they all pass successfully, the updates are then applied at the end. This ensures atomicity property of transactions.

mod dao_contract {
    ...

    // DAO::mint() in the smart contract pseudocode
    mod mint {
        ...

        // StateRegistry is mutable
        fn apply(states: &mut StateRegistry, update: Update) {
            ...
        }
    }
}

The transaction verification pipeline roughly looks like this:

  1. Loop through all function call invocations within the transaction:
    1. Lookup their respective state_transition() function based off their contract_id and func_id. The contract_id and func_id corresponds to the contract and specific function, such as DAO::mint().
    2. Call the state_transition() function and store the update. Halt if this function fails.
  2. Loop through all updates
    1. Lookup specific apply() function based off the contract_id and func_id.
    2. Call apply(update) to finalize the change.

ZK Proofs and Signatures

Lets review again the format of transactions.

/// A Transaction contains an arbitrary number of `ContractCall` objects,
/// along with corresponding ZK proofs and Schnorr signatures.
#[derive(Debug, Clone, Eq, PartialEq, SerialEncodable, SerialDecodable)]
pub struct Transaction {
    /// Calls executed in this transaction
    pub calls: Vec<ContractCall>,
    /// Attached ZK proofs
    pub proofs: Vec<Vec<Proof>>,
    /// Attached Schnorr signatures
    pub signatures: Vec<Vec<Signature>>,
}

And corresponding function calls.

/// A ContractCall is the part of a transaction that executes a certain
/// `contract_id` with `data` as the call's payload.
#[derive(Debug, Clone, Eq, PartialEq, SerialEncodable, SerialDecodable)]
pub struct ContractCall {
    /// ID of the contract invoked
    pub contract_id: ContractId,
    /// Call data passed to the contract
    pub data: Vec<u8>,
}

As we can see the ZK proofs and signatures are separate from the actuall call_data interpreted by state_transition(). They are both automatically verified by the VM.

However for verification to work, the ZK proofs also need corresponding public values, and the signatures need the public keys. We do this by exporting these values. (TODO: link the code where this happens)

These methods export the required values needed for the ZK proofs and signature verification from the actual call data itself.

For signature verification, the data we are verifying is simply the entire transactions minus the actual signatures. That's why the signatures are a separate top level field in the transaction.

Parallelisation Techniques

Since verification is done through state_transition() which returns an update that is then committed to the state using apply(), we can verify all transactions in a block in parallel.

To enable calling another transaction within the same block (such as flashloans), we can add a special depends field within the tx that makes a tx wait on another tx before being allowed to verify. This causes a small deanonymization to occur but brings a massive scalability benefit to the entire system.

ZK proof verification should be done automatically by the system. Any proof that fails marks the entire tx as invalid, and the tx is discarded. This should also be parallelized.