This ERC provides standard interfaces and processes for flash lenders and borrowers, allowing for flash loan integration without a need to consider each particular implementation.
## Motivation
Flash loans allow smart contracts to lend an amount of tokens without a requirement for collateral, with the condition that they must be returned within the same transaction.
Early adopters of the flash loan pattern, such as [Aave](https://github.com/aave/aave-protocol/blob/e8d020e9752fbd4807a3b55f9cf98a88dcfb674d/contracts/flashloan), [DxDy](https://help.dydx.exchange/en/articles/3724602-flash-loans), [Uniswap](https://uniswap.org/docs/v2/core-concepts/flash-swaps/) and the [Yield Protocol](https://github.com/yieldprotocol/fyDai/blob/master/contracts/FYDai.sol) have produced different interfaces and different use patterns. The diversification is expected to intensify, and with it the technical debt required to integrate with diverse flash lending patterns.
Some of the high level diferences in the approaches across the protocols include:
- Repayment approaches at the end of the transaction, where Aave V2 pulls the flash loaned amount plus the flash fee off the flash smart contract, compared to other protocols where the contract needs to explicitly calculate the debt+fee amount and manually return it to the lending pool.
- Uniswap's Flash Swaps offer the ability to repay the flash transaction using a token that is different to what was originally flash borrowed, which can reduce the overall complexity of the flash transaction and gas fees, depending on the purpose of the flash swap (i.e. the second last step in flash self liquidation to swap back into the repayment token).
- DyDx offering a single entry point into the protocol regardless of whether you're buying, selling, depositing or chaining them together as a flash loan, whereas other protocols offer discrete entry points (e.g. Uniswap V2's swap() and Aave V2's flashLoan() methods).
- The Yield Protocol allows to flash mint any amount of its native token without charging a fee, effectively allowing flash loans bounded by computational constraints instead of asset ownership constraints.
## Specification
A flash lending feature integrates two smart contracts using a callback pattern. These are called the LENDER and the RECEIVER in this EIP.
The `maxFlashAmount` function MUST return the maximum loan possible for `token`. If a `token` is not currently supported `maxFlashAmount` MUST return 0, instead of reverting.
For each `token` in `tokens`, the `batchFlashLoan` function MUST transfer `amounts[i]` of `tokens[i]` to `receiver` before the callback to the borrower.
The `batchFlashLoan` function MUST include a `fees` argument to `onBatchFlashLoan` with the fee to pay for each individual `token` and `amount` lent, ensuring that `fees[i] = flashFee(tokens[i], amounts[i])`.
After the callback, for each `token` in `tokens`, the `batchFlashLoan` function MUST take the `amounts[i] + fees[i]` of `tokens[i]` from the `receiver`, or revert if this is not successful.
For all functions above, including both mandatory and optional sections, address(1) is used as a sentinel value for Ether. If the token parameter is address(1) then the function should be processed as defined except using Ether instead of a token.
For the transaction to not revert, for each `token` in `tokens`, `receiver` MUST approve `amounts[i] + fees[i]` of `tokens[i]` to be taken by `msg.sender` before the end of `onBatchFlashLoan`.
If successful, `onBatchFlashLoan` MUST return `true`.
For all functions above, including both mandatory and optional sections, address(1) is used as a sentinel value for Ether. If the token parameter is address(1) then the function should be processed as defined except using Ether instead of a token.
The interfaces described in this ERC have been chosen as to cover the known flash lending use cases, while allowing for safe and gas efficient implementations.
`flashLoan` has been chosen as a function name as descriptive enough, unlikely to clash with other functions in the lender, and including both the use cases in which the tokens lended are held or minted by the lender.
Existing flash lenders (Aave, dYdX and Uniswap) all provide flash loans of several token types from the same contract (LendingPool, SoloMargin and UniswapV2Pair). Providing a `token` parameter in both the `flashLoan` and `onFlashLoan` functions matches closely the observed functionality.
A `bytes calldata data` parameter is included for the caller to pass arbitrary information to the `receiver`, without impacting the utility of the `flashLoan` standard.
`onFlashLoan` has been chosen as a function name as descriptive enough, unlikely to clash with other functions in the `receiver`, and following the `onAction` naming pattern used as well in EIP-667.
A `sender` will often be required in the `onFlashLoan` function, which the lender knows as `msg.sender`. An alternative implementation which would embed the `sender` in the `data` parameter by the caller would require an additional mechanism for the receiver to verify its accuracy, and is not advisable.
The `amount` will be required in the `onFlashLoan` function, which the lender took as a parameter. An alternative implementation which would embed the `amount` in the `data` parameter by the caller would require an additional mechanism for the receiver to verify its accuracy, and is not advisable.
A `fee` will often be calculated in the `flashLoan` function, which the `receiver` must be aware of for repayment. Passing the `fee` as a parameter instead of appended to `data` is simple and effective.
The `amount + fee` are pulled from the `receiver` to allow the `lender` to implement other features that depend on using `transferFrom`, without having to lock them for the duration of a flash loan. An alternative implementation where the repayment is transferred to the `lender` is also possible, but would need all other features in the lender to be also based in using `transfer` instead of `transferFrom`. Given the lower complexity and prevalence of a "pull" architecture over a "push" architecture, "pull" was chosen.
*@param receiver The contract receiving the tokens, needs to implement the `onFlashLoan(address user, uint256 amount, uint256 fee, bytes calldata)` interface.
*@param receiver The contract receiving the tokens, needs to implement the `onFlashLoan(address user, uint256 amount, uint256 fee, bytes calldata)` interface.
The arguments of `onFlashLoan` are expected to reflect the conditions of the flash loan, but cannot be trusted unconditionally. They can be divided in two groups, that require different checks before they can be trusted to be genuine.
0. No arguments can be assumed to be genuine without some kind of verification. `sender`, `token` and `amount` refer to a past transaction that might not have happened if the caller of `onFlashLoan` decides to lie. `fee` might be false or calculated incorrectly. `data` might have been manipulated by the caller.
1. To trust that the value of `sender`, `token`, `amount` and `fee` are genuine a reasonable pattern is to verify that the `onFlashLoan` caller is in a whitelist of verified flash lenders. Since often the caller of `flashLoan` will also be receiving the `onFlashLoan` callback this will be trivial. In all other cases flash lenders will need to be approved if the arguments in `onFlashLoan` are to be trusted.
2. To trust that the value of `data` is genuine, in addition to the check in point 1, it is recommended to implement the `flashLoan` caller to be also the `onFlashLoan` receiver. With this pattern, checking in `onFlashLoan` that `sender` is the current contract is enough to trust that the contents of `data` are genuine.
The safest approach is to implement an approval for `amount+fee` before the `flashLoan` is executed.
Including in `onFlashLoan` the approval for the `lender` to take the `amount + fee` needs to be combined with a mechanism to verify that the borrower is trusted, such as those described above.
If an unsuspecting contract with a non-reverting fallback function, or an EOA, would approve a `lender` implementing ERC3156, and not immediately use the approval, and if the `lender` would not verify the return value of `onFlashLoan`, then the unsuspecting contract or EOA could be drained of funds up to their allowance or balance limit. This would be executed by a `borrower` calling `flashLoan` on the victim. The flash loan would be executed and repaid, plus any fees, which would be accumulated by the `lender`. For this reason, it is important that the `lender` implements the specification in full and reverts if `onFlashLoan` doesn't return the keccak256 hash for "ERC3156FlashBorrower.onFlashLoan".
The typical quantum of tokens involved in flash mint transactions will give rise to new innovative attack vectors.
#### Example 1 - interest rate attack
If there exists a lending protocol that offers stable interests rates, but it does not have floor/ceiling rate limits and it does not rebalance the fixed rate based on flash-induced liquidity changes, then it could be susceptible to the following scenario:
FreeLoanAttack.sol
1. Flash mint 1 quintillion DAI
2. Deposit the 1 quintillion DAI + $1.5 million worth of ETH collateral
3. The quantum of your total deposit now pushes the stable interest rate down to 0.00001% stable interest rate
4. Borrow 1 million DAI on 0.00001% stable interest rate based on the 1.5M ETH collateral
5. Withdraw and burn the 1 quint DAI to close the original flash mint
6. You now have a 1 million DAI loan that is practically interest free for perpetuity ($0.10 / year in interest)
The key takeaway being the obvious need to implement a flat floor/ceiling rate limit and to rebalance the rate based on short term liquidity changes.
#### Example 2 - arithmetic overflow and underflow
If the flash mint provider does not place any limits on the amount of flash mintable tokens in a transaction, then anyone can flash mint 2^256-1 amount of tokens.
The protocols on the receiving end of the flash mints will need to ensure their contracts can handle this. One obvious way is to leverage OpenZeppelin's SafeMath libraries as a catch-all safety net, however consideration should be given to when it is or isn't used given the gas tradeoffs.
If you recall there was a series of incidents in 2018 where exchanges such as OKEx, Poloniex, HitBTC and Huobi had to shutdown deposits and withdrawls of ERC20 tokens due to integer overflows within the ERC20 token contracts.
The coupling of flash minting with business specific features in the same platform can easily lead to unintended consequences.
#### Example - Treasury draining
In early implementations of the Yield Protocol flash loaned fyDai could be redeemed for Dai, which could be used to liquidate the Yield Protocol CDP vault in MakerDAO:
1. Flash mint a very large amount of fyDai.
2. Redeem for Dai as much fyDai as the Yield Protocol collateral would allow.
3. Trigger a stability rate increase with a call to `jug.drip` which would make the Yield Protocol uncollateralized.
4. Liquidate the Yield Protocol CDP vault in MakerDAO.
## Copyright
Copyright and related rights waived via [CC0](https://creativecommons.org/publicdomain/zero/1.0/).