Unpacking ERC-4337
🗃️

Unpacking ERC-4337

✍🏼
by Ankit Chiplunkar for Frontier Research
🗓️
1st December 2022
 
This is the first part of our three-part series on ERC-4337. In this article, we provide a simple overview of the ERC-4337 standard for account abstraction and compare it to previous approaches to user authentication by analyzing tradeoffs in feature support, censorship resistance, DOS protection, and gas efficiency. In future articles, we will dig into the extensions, and the costs induced by the ERC.
EOA Wallet
SC Wallet
4337 Wallet
Arbitrary execution logic (gas sponsoring, key recovery, new signature schemes, etc)
Censorship resistant
✅ (in theory)
✅ (in theory)
DOS protection
⚠️ (5.5x overhead vs EOA)
Gas overhead per user operation
21k gas
38k gas (best case)
39.8k gas (best case)
Compatible with current consensus rules

TLDR

  • Account Abstraction is the ability of a wallet to have arbitrary execution logic. This makes self custodying easy for retail users by enabling multiple options for key management and account recovery.
  • ERC-4337 in theory implements censorship-resistant Account Abstraction without making any changes to the consensus layer of Ethereum.
  • It achieves this by creating a new mempool, a smart contract wallet standard, and a shared entry point contract for managing User Operations.
  • 4337 does however introduce some tradeoffs, namely additional complexity which manifests itself as an increased chance of DOS and a loss in gas efficiency relative to EOA wallets.

Introduction

Self-custodying crypto is HARD. Both new and power users find private key management a nightmare. Moreover, with FTX mishandling customer funds, no retail user will be comfortable storing large sums on an unregulated exchange. Account Abstraction (AA) promises to solve the dilemma between key management UX and security for self-custodial funds.
AA achieves this by enabling arbitrary logic to determine if an account can move funds. This enables wallets to have multi-sigs, social recovery, hardware enclaves, and much more. Although this has been tried already via Smart Contract (SC) wallets, these experiments never gained widespread traction. Mainly, because users had to rely on a centralizing relay to emit their transactions (Authereum) or hold funds in 2 different wallets (Gnosis Safe), moreover each SC wallet had to develop its own infrastructure increasing development costs. Newer L2 chains such as zkSync and StarkNet identify the advantages of AA and come prebuilt with it even at the cost of losing compatibility with the EVM spec.
This article intends to explain how ERC-4337 implements AA while making NO changes to the EVM spec.

What does Account Abstraction even mean?

On the Ethereum network, there are currently two types of accounts, External, and SC accounts. AA reduces the two account types down to one, an SC account.
This means that after AA:
  1. Users can generate User Operations via SC wallets. The user's intent to interact with the blockchain is encoded in a User Operation and is signed by the user for authentication.
  1. The system validates if the User Operation is valid.
  1. The SC wallet interacts with dApps, this is to ensure that legacy dApps work with SC wallets using msg.sender opcode.
In the next section, we will look at the state of the art of a transaction's lifecycle in EVM chains and the constraints coming from the consensus layer.

State of the art

In EVM chains, transactions come from two types of wallets.
  1. Externally Owned Address (EOA) wallets: i.e. wallets generated via a private key and need a signature from the private key to move funds. Eg Metamask, Coinbase wallet or MPC wallet
  1. Smart contract (SC) wallets: i.e. wallets which have their own code and need successful execution of the code to move funds. Eg. Gnosis Safe, Argent, Biconomy, or Sequence

EOA wallet’s transaction lifecycle

EOA’s are the most commonly used wallets on EVM chains and consist of the “default” user experience. The transaction lifecycle of an EOA wallet follows several steps each of which has its own checks and constraints to ensure censorship-resistant and economically incentivized inclusion.
notion image
In the EVM semantics, EOA accounts interact with dApp smart contracts directly to perform their operations, eg swap on Uniswap or open loans on Compound. The EOA wallets which initiated the transaction are identified through tx.origin opcode which returns the first 20 bytes of the public key. While this opcode can be used to identify the EOA, it cannot be used to authenticate a user request as it is carried forward with call context. Smart contract developers typically use msg.sender for user authentication purposes and can insure the calling account is an EOA by requiring msg.sender == tx.origin.
notion image

Transaction lifecycle

  1. A user generates and signs the transaction payload using their private key.
  1. The transaction is then gossiped to the peer-to-peer (p2p) mempool via a node. Each node in the network checks the balance of the account and rejects the transaction if the user does not have enough balance to pay for the transaction fees. The p2p network provides both censorship resistance and DOS protection by providing multiple routes for valid transactions and efficiently removing invalid transactions.
  1. A miner/validator picks up the transaction from the mempool and includes it in a block, thus executing the transaction.

Constraints

  1. Consensus level: Base fees and priority fees are deducted from an EOA address that initiates the transaction. Nodes validate that the EOA has the sufficient native token balance to cover the maximum execution cost gasLimit * maxFee.
  1. Consensus level: No code is executed via the EOA address but it can call contracts that execute code.
  1. Consensus level: EOA accounts cannot change their associated private key. This means it is extremely difficult to implement key recovery, multi-factor authentication or granular access control.
  1. Legacy dApp: The user wallet directly calls the dApp contract.

SC wallet’s transaction lifecycle

As the usage of Ethereum grew, developers grew frustrated at the limitations of EOA wallets and started creating SC wallets with more advanced access control (social recovery, multi-signature, hardware enclave). The transaction lifecycle and the constraints explained above remain the same, but abstraction started happening in the transaction layer.
notion image

Transaction lifecycle

  1. The transaction is generated via a third-party relay or a secondary user address.
  1. In the transaction, the service address initiates the transaction by calling the SC wallet, which performs the intended User Operation by calling the dApp address.
notion image
Since User Operations cannot be directly submitted to the blockchain, a relayer or an external wallet is required to initiate the transaction. The constraints, imposed by the consensus layer and legacy dApps, are maintained in the above flow, however, it increases the transaction cost.

ERC 4337

Post ERC-4337 any EOA wallet will be able to initiate a transaction containing any User Operation, i.e. it guarantees that the initiator EOA wallet will be paid back for the transaction fees by the User Operation. This is achieved by splitting a User Operation into 2 sandboxed steps. A validation step, to check if the User Operation can move funds, and an execution step that performs the dApp interaction.
The whole User Operation lifecycle is split into 3 distinct parts:
  1. A higher level mempool to ensure that a User Operation is valid, and manage its inclusion in a transaction.
  1. An EOA wallet transaction lifecycle is the same as explained in the earlier section. The EOA wallet initiating the transaction is also called the Bundler.
  1. And finally, execution of the User Operation in a transaction.
Bundlers only need to verify if the validation step will succeed, to gain confidence that they will be paid for including this User Operation in their initiated transaction.

User Operation mempool

notion image
  1. The user generates a User Operation to perform an interaction using their SC wallet
  1. The validity (ability to compensate the bundler for execution fees, correct nonce, and correct authentication) is simulated via a Bundler. This validation logic is encoded in the validateUserOp function of the user's SC wallet.
  1. The validation logic should not depend on the blockchain state which can vary with time. This is to ensure that the results of the simulation and eventual blockchain execution are consistent. For e.g.:
    1. A User Operation’s validation step depending on signatures of required signers is a correct validation since it does not depend on any time-varying properties (eg. Gnosis Safe)
    2. A User Operation’s validation step depending on the balance of account X will be deemed incorrect validation since the balance of account X can change by the time User Operations lands in the blockchain.
  1. Once a Bundler is satisfied with the validity and fees earned from a User Operation it includes the User Operation in a transaction.

Censorship resistance and DOS protection

One of the key limitations of current SC wallet implementations of account abstraction is the lack of censorship resistance. While in theory, EOA wallet transactions are censorship-resistant due to the use of the gossip network for message routing, SC wallets generally use centralized message relays, thus being subject to censorship.
The key challenge to achieving censorship resistance is providing DOS protection to the servers forwarding messages. While the Ethereum transaction mempool is far from being considered perfect, the gossip network for EOA accounts is (in theory) protected from DOS by efficiently dropping invalid transactions from the mempool nodes. Each node checks that EOA transactions have:
  1. A valid signature,
  1. A valid nonce,
  1. A sufficient account balance to pay the maximum gas fees.
These checks cost the equivalent of 35k gas to perform on the EVM. You can find a solidity benchmark implemented here.
Since account abstraction enables arbitrary execution logic, work to be performed by mempool to identify invalid User Operations is now a function of the complexity of the validation step and therefore potentially unbounded. Typical SC wallet implementations of account abstractions are therefore forced to use centralized message relayers and achieve DOS protection through traditional means including IP allow / ban lists, API keys, rate limiting, and reputation systems.
ERC-4337 aims to provide the same level of censorship resistance and DOS protection as EOA accounts through the secondary gossip network. To mitigate the DOS vector caused by unbounded compute in the validation step, 4337 imposes a maximum gas limit of 200k gas on validateUserOp. Comparing this upper bound to the cost of validating an EOA transaction, we see that it is ~5.5 times more expensive to perform DOS protection on 4337 transactions than EOA transactions.
This additional overhead on transaction validation along with an unknown number and distribution of nodes in the 4337 gossip network means that there are valid concerns on the ability of 4337 to successfully match the censorship resistance and DOS protection of EOA accounts.
It is worth noting that this compares censorship resistance of 4337 relative to EOA accounts. Both are subject to block-building censorship as seen in MEV-Boost and L2 sequencers.

Generation and inclusion of transaction

Since we are not making changes to the consensus layer, this step remains the same.
notion image
  1. The bundler submits the transaction created by bundling several User Operations.
  1. Note that the fees for the bundled transaction come from the bundler's address and ERC-4337 has to guarantee that bundler is paid for its work.
  1. To minimize chances for lost fees/frontrunning, the optimal behavior for the validators is to run bundlers themselves or outsource its generation via Flashbots and quickly include them in the next block.

Execution of User Operation in a transaction

To ensure that the Bundler address gets paid and User Operation interacts with the dApp as intended, ERC-4337 orchestrates these operations via an EntryPoint contract. The following steps take place in such a transaction:
notion image
  1. The bundler EOA address sends a bundle of User Operations to the Entry Point contract.
  1. Each User Operation calls the function _validatePrePayment on the contract, this function makes sure that the operation is valid and takes the fees (PrePayment) from the SC wallet.
  1. _validatePrePayment calls the function validateUserOp on the SC wallet, this same function was simulated by the bundler in the User Operation mempool.
  1. If the User Operation is valid then Entry Point takes the fees for this transaction from the SC wallet.
  1. Now since User Operation is validated, the _executeUserOp function on Entry Point calls the SC wallet to execute the intended code.
  1. The User Operation can interact with a dApp to perform several operations on the dApp. Approve, swap, return, etc.
  1. SC wallet tells EntryPoint if the execution succeeded or failed.
  1. Finally, fees are paid back to the bundler wallet and any remaining fees are returned to the SC wallet.

Gas efficiency

While 4337 moves much of the transaction validation logic on chain, it aims to do so as efficiently as possible in order to provide a good user experience. Since 4337 allows for bundling multiple user operations together in a single EOA transaction, it has the opportunity of amortizing the 21k gas overhead the the EOA across these operations.
The benchmarks provided by the eth-infinitism implementation show that in a best case scenario of 10 user operations bundle with simple validation yields an overhead of 39.8k gas per user operation (~2x a typical EOA transaction overhead). This is not materially different to SC wallets.

Conclusion

In this article, we saw how ERC-4337 attempts to provide censorship-resistant account abstraction and how it compares to legacy transactions from EOA wallets and Smart Contract wallets.
The ERC proposes standards for:
  1. Generation of UserOperation from SC wallets
  1. Their incentivized and permissionless inclusion in the EVM layer
  1. Protecting bundlers from DOS of validating UserOperations
  1. Guarantees that bundlers are paid for their work and User Operations are executed as intended
We noted how the additional complexity brought by 4337 introduces some questions on gas efficiency and DOS protection which should be carefully evaluated by teams interested in implementing 4337 in their system.
In the upcoming articles, we will see how the aggregator and paymaster extensions work with ERC-4337 and take a deeper look into the costs associated with making these changes.
Special thanks to Sid Shekhar, and Jing Fan for reviewing the article.

References

  1. 4337 specs: https://github.com/ethereum/EIPs/blob/master/EIPS/eip-4337.md
  1. 4337 introduction: https://medium.com/infinitism/erc-4337-account-abstraction-without-ethereum-protocol-changes-d75c9d94dc4a
  1. Account abstraction introduction: https://docs.ethhub.io/ethereum-roadmap/ethereum-2.0/account-abstraction/
  1. 4337 gas cost analysis: https://github.com/eth-infinitism/account-abstraction/blob/develop/reports/gas-checker.txt#L17
  1. EOA and SC base gas analysis: https://github.com/ankitchiplunkar/erc4337#erc4337-gas-estimates----