Chainlink CCIP Owner Contracts

This repo contains a set of contracts used for administering Chainlink contracts, most notably CCIP.

The contracts in this repo's main branch are considered production-grade and have been reviewed as part of a code4rena contest.

Development

We use foundry. The tests rely on some ffi code written in Go 1.18 (see testCommands/). See the official Go docs for installation instructions. Once you have Go running, forge test --ffi should do the trick.

Format code with forge fmt.

Generate a code coverage report by running ./coverage.sh.

Design Considerations

The CallProxy, ManyChainMultiSig, RBACTimelock contracts are all part of a system of owner contracts that is supposed to administer other contracts (henceforth referred to as OWNED). OWNED contracts represent any system of contracts that (1) have an owner or similar role (e.g. using OpenZeppelin's OwnableInterface) and that (2) are potentially deployed across multiple chains.

Here is a diagram of how we envision these contracts to interact:

graph LR;
    owned[OWNED contracts];
    prop[ManyChainMultiSig for proposers];
    cancel[ManyChainMultiSig for cancellers];
    forwarder[CallProxy];
    timelock[RBACTimelock];
    emerg[ManyChainMultiSig for bypassers];
    prop --> |PROPOSER| timelock;
    prop --> |CANCELLER| timelock;
    cancel --> |CANCELLER| timelock;
    forwarder --> |EXECUTOR| timelock;
    timelock --> |ADMIN| timelock;
    timelock --> |OWNER| owned;
    timelock --> |OWNER| emerg;
    timelock --> |OWNER| cancel;
    timelock --> |OWNER| prop;
    emerg --> |BYPASSER| timelock;

Loading

Regular administration of the OWNED contracts is expected to happen through the RBACTimelock's Proposer/Executor/Canceller roles. The Bypasser role is expected to only become active in "break-glass" type emergency scenarios where waiting for RBACTimelock.minDelay would be harmful.

Proposers can also cancel so that they may "undo" proposals with mistakes in them.

Gas cost isn't particularly important for these contracts because they're not expected to be called often. Correctness matters much more.

RBACTimelock Considerations

We expect to set RBACTimelock.minDelay and delay to ~ 24 hours, but in general values between 1 hour and 1 month should be supported. This enables anyone to inspect configuration changes to the OWNED contracts before they take effect. For example, a user that disagrees with a configuration change might choose to withdraw funds stored in OWNED contracts before they can be executed.

We may use RBACTimelock.blockFunctionSelector to prevent specific functions on the OWNED contracts from being called through the regular propose-execute flow.

RBACTimelock is based on an OpenZeppelin contract. We intentionally use the old require syntax (and some other old techniques) in RBACTimelock to keep the diff vs the original OZ contract smaller.

CallProxy Considerations

The CallProxy is intentionally callable by anyone. Offchain tooling used for generating configuration changes will make appropriate use of the RBACTimelock's support for predecessors to ensure that configuration changes are sequenced properly even if an adversary is executing them. Since the adversary can control the gas amount and gas price, callees are expected to not have gas-dependent behavior other than reverting if insufficient gas is supplied.

The CallProxy is not expected to be used with contracts that could SELFDESTRUCT. It thus has no EXTCODESIZE-check prior to making a call. We expect it to be configured correctly (i.e. pointing to a real RBACTimelock) on deployment.

ManyChainMultiSig Considerations

Unlike standard multi-sig contracts, ManyChainMultiSig supports signing many transactions targeting many chains with a single set of signatures. (We currently only target EVM chains and all EVM chains support the same ECDSA secp256k1 standard.) This is useful for administering systems of contracts spanning many chains without increasing signing overhead linearly with the number of supported chains. We expect to use the same set of EOA signers across many chains. Consequently, ManyChainMultiSig only supports EOAs as signers, not other smart contracts. Similar to the rest of the system, anyone who can furnish a correct Merkle proof is allowed to execute authorized calls on the ManyChainMultiSig, including a potential adversary. The adversary will be able to control the gas price and gas amount for the execution.

The proposer and canceller ManyChainMultiSig contracts are expected to be configured with a group structure like this, with different sets of signers for each (exact k-of-n parameters might differ):

          ┌──────────┐
          │Root Group│
      ┌──►│  6-of-8  │◄─────────┐
      │   └──────────┘          │
      │         ▲               │
      │         │               │
 ┌────┴───┐ ┌───┴────┐     ┌────┴───┐
 │signer 1│ │signer 2│ ... │signer 8│
 └────────┘ └────────┘     └────────┘

The bypasser ManyChainMultiSig contract is expected to be configured with a more complex group structure like this (exact structure might differ):

graph TD;

    root[root group<br>2-of-2];
    sub1[subgroup 1<br>6-of-8];
    sub2[subgroup 2<br>2-of-3];
    sub21[subgroup 2.1<br>6-of-8];
    sub22[subgroup 2.2<br>1-of-3];
    sub23[subgroup 2.3<br>6-of-8];
    sigs1to8[signers 1 ... 8];
    sigs9to16[signers 9 ... 16];
    sigs17to19[signers 17 ... 19];
    sigs20to27[signers 20 ... 27];

    root --- sub1;
    root --- sub2;
    sub2 --- sub21;
    sub2 --- sub22;
    sub2 --- sub23;
    sub1 --- sigs1to8;
    sub1 --- sigs1to8;
    sub1 --- sigs1to8;
    sub1 --- sigs1to8;
    sub1 --- sigs1to8;
    sub1 --- sigs1to8;
    sub1 --- sigs1to8;
    sub1 --- sigs1to8;
    sub21 --- sigs9to16;
    sub21 --- sigs9to16;
    sub21 --- sigs9to16;
    sub21 --- sigs9to16;
    sub21 --- sigs9to16;
    sub21 --- sigs9to16;
    sub21 --- sigs9to16;
    sub21 --- sigs9to16;
    sub22 --- sigs17to19;
    sub22 --- sigs17to19;
    sub22 --- sigs17to19;
    sub23 --- sigs20to27;
    sub23 --- sigs20to27;
    sub23 --- sigs20to27;
    sub23 --- sigs20to27;
    sub23 --- sigs20to27;
    sub23 --- sigs20to27;
    sub23 --- sigs20to27;
    sub23 --- sigs20to27;

Loading

Subgroup 1 has the same signers as the canceller ManyChainMultiSig. No change can ever be enacted without approval of this group.

In practice, we expect the k-of-n configurations of groups to typically have 1<=k<=32 and 1<=n<=32 (where k<=n and we tolerate the overall limits on groups/signers set in ManyChainMultiSig code).

We intentionally store chain ids in uint256, foregoing some storage savings. We want to minimize the likelihood of having to change the contract later to support larger chain ids and the cost savings aren't very significant since we don't envision setting new roots all that frequently. We are aware of a proposal to bound chain ids to 64 bits, but it is still unresolved.

We choose the tried-and-true OpenZeppelin Merkle tree. We remain conservative and don't make use of multiproofs for the sake of simplicity. The same KISS approach also leads us to not make use of Merkle tries with support for storing key-value-pairs (e.g. sparse Merkle trees). Such tries would enable us to compute a unique key for each Op, preventing e.g. two Ops with the same nonce and multisig contract from being included in the trie. However, the practical benefit of doing so seems limited: a faulty set of signers can take far more damaging actions that putting two conflicting transactions in the trie.

Propose-and-Execute Flow

The following steps need to be performed for a set of onchain maintenance operations on the OWNED contracts:

• [offchain, out of scope] Merkle tree generation & signing: A Merkle tree containing all the required ManyChainMultiSig ops (containing RBACTimelock.scheduleBatch calls) for the desired maintenance operations is generated by the proposers. A quorum of signers from the proposer ManyChainMultiSig must sign (offchain) the Merkle root.
• setRoot call on all relevant ManyChainMultiSig contracts across chains: The signed Merkle root is then sent to ManyChainMultiSigs. Anyone who has been given the root and the signatures offchain can send it to ManyChainMultiSigs.
• execute on ManyChainMultiSig: To propose an action to the RBACTimelock, a multi-sig op is executed by providing a Merkle proof for that specific op. Anyone who has been given the full Merkle tree offchain can propose the action.
• executeBatch on RBACTimelock: After the timelock wait period expires, the proposed actions in TimeLock can be executed. This assumes that the cancellers have not cancelled them in the meantime. Anyone can execute the actions because all the required information is available on the blockchain through event logs.

Canceller Flow

This can be thought of as an optional step of the propose-and-execute flow. If a quorum of cancellers disapproves of an action pending on the RBACTimelock, they can create a set of ManyChainMultiSig.Ops that calls RBACTimelock.cancel on all relevant RBACTimelocks.

Bypasser Flow

This is completely independent of the propose-and-execute flow. Bypassers create a set of ManyChainMultiSig.Ops that calls RBACTimelock.bypasserExecuteBatch on all relevant RBACTimelocks.

Porting

Developers porting ManyChainMultiSig (MCMS) to new target chains should keep the following considerations in mind.

• If the target chain supports keccak256 and secp256k1, as well as sufficient programmability to mirror the Solidity contract's Merkle tree computations, then the same set of signers as on the Solidity contract can be used. We expect that most target chains will fall in this category as they aim to be compatible with existing Ethereum signing infrastructure. This option is recommended, so that signers only have to sign once, and due to the prevalence of eth_sign support in hardware wallets, which may be used by the multisig signers.
• Use distinct domain separators for the Merkle tree leaves for each target chain family to avoid ambiguity.
  • Use distinct domain separators for metadata, for each target chain family (e.g., keccak256("MANY_CHAIN_MULTI_SIG_DOMAIN_SEPARATOR_METADATA_SOLANA")).
  • Use distinct domain separators for ops, for each target chain family (e.g., keccak256("MANY_CHAIN_MULTI_SIG_DOMAIN_SEPARATOR_OP_SOLANA")).
  • This prevents ops and metadata for a chain family from being replayable on another.
• Preimages of Merkle tree leaves must conform to the following rules:
  • The domain separator must always be the first word (32 bytes) of any leaf preimage.
  • The rest of the leaf preimage can be encoded in any way that makes sense for the target chain and language, i.e., does not need to follow the same abi encoding of the Solidity contract.
  • The encoding must be canonical:
    • It must not be possible for two distinct metadata to encode to the same value.
    • It must not be possible for two distinct ops to encode to the same value.
  • Any leaf preimage must always be of length greater than 64 bytes, to avoid ambiguity with internal nodes.
• Computation of internal nodes of the Merkle tree must happen identically to the Solidity contract, in order to reuse the Solidity Merkle tree. In particular, the internal node hash should be computed as the commutative keccak256 hash of its two subtree root hashes. See this snippet from OpenZeppelin for the exact implementation depended upon by the Solidity contract.
• For chains that have some notion similar to the EVM chain id, and where the chain id is available in the target language environment (similar to block.chainid), the chain id must be used for both metadata and ops. If such a chain id does not exist, please contact the authors for guidance. We prefer use of chain ids to chain selectors for MCMS because:
  • Chain ids gracefully handle permanent forks. If an entity decides to fork a chain (e.g., as in the case of Ethereum & Ethereum Classic), it is considered good practice for them to assign a different chain id to their fork to avoid replayability of transactions across forks. In such a scenario, use of chain ids in MCMS avoids replayability of MCMS metadata and ops across forks.
  • Chain selectors must be configured during deployment, and if misconfigured can cause MCMS metadata and ops to be executable on an unintended chain.
• If unsure whether a target chain warrants a new chain family, reach out to the authors. Even if two target chains use the same language, they might not belong to the same family (e.g., Aptos & Sui both support Move as a language, but they belong to different chain families). Always ensure that there can never be two chains with the same chain id in the same chain family.