November 15th, 2022
This security assessment was prepared by OpenZeppelin.
We audited the pods-finance/yield-contracts repository at the 9389ab46e9ecdd1ea1fd7228c9d9c6821c00f057 commit.
In scope were the following contracts:
contracts/
├── configuration
│ └── ConfigurationManager.sol
├── interfaces
│ ├── IConfigurationManager.sol
│ ├── ICurvePool.sol
│ ├── IERC4626.sol
│ └── IVault.sol
├── libs
│ ├── AuxMath.sol
│ ├── CastUint.sol
│ └── DepositQueueLib.sol
├── mixins
│ └── Capped.sol
├── proxy
│ ├── ETHAdapter.sol
│ └── Migration.sol
└── vaults
├── BaseVault.sol
└── STETHVault.sol
The Pods Yield Vault is a one-click deposit investment product that features an options strategy to generate principal-protected returns when the ETH market is volatile. The underlying technique uses a portion of the yield from stETH assets to invest in the same number of call and put ETH options in an off-chain market. Returns from these investments will be given to the participants who provided their yield to the investor.
The following diagram gives an overview of the system:
The Ethereum Volatility Vault, stETHvv for short, is a tokenized vault on the Ethereum mainnet, conforming to the EIP-4626 standard. It integrates with the Lido stETH token as the single underlying asset. Users deposit the underlying asset into the vault in return for the stETHvv vault shares to participate in the investment strategy.
The investment strategy is executed in rounds. After a new round starts, users can join the deposit queue, where the users’ underlying assets will be transferred to the vault, thus starting to generate yield to the vault. However, stETHvv vault shares will not be minted to the user until the queued deposit is processed after the round ends.
The user can withdraw their queued deposit before it is processed, but any stETH yield generated by the queued deposit will not be accessible. Before the round ends, users with stETHvv vault shares can withdraw their share of assets for a fee.
When the round ends, only a portion of the stETH yield, not the principal, will be transferred to a centralized investor account, where the yield is manually used to participate in the off-chain option strategy. In the same transaction, the strategy profit denominated in stETH from the previous round will be transferred to the vault from the investor account.
After the round ends, deposits to the queue and withdrawals from the vault are temporarily paused while processing of queued deposits is enabled. Anyone can process any queued deposit by specifying a valid start and end index of the queue. Each processed deposit will mint new stETHvv vault shares, as computed based on the total assets in the vault. Any unprocessed deposit will remain in the queue, which can be either processed or withdrawn in the next round.
When the next round starts, new deposits and withdrawals to the queue may resume, while queued deposits can no longer be processed.
Parameters such as the vault controller, the percentage of commission for withdrawals, and the cap of shares that can be issued are established in this contract for the different vaults of the protocol. One particularity of this contract is that these values are set through a function that allows storing any name-value pair for a specific target on a nested mapping. This allows more values to be added in the future without the need to replace the contract but adds more complexity to managing contract storage.
It is worth mentioning that there is an exception where an exclusive function was defined to establish if a vault can be used as a migration destination. It has its own setter and getter.
In order to allow vault deposits to be made directly with ETH, this contract acts as an intermediary so that users can participate in the protocol without needing to perform the swap themselves. This contract swaps ETH for stETH through a Curve Finance pool. After the exchange, the ETH adapter deposits the resulting tokens to the vault on behalf of the user within the same transaction.
In the case of an upgrade, a stETHvv vault shareholders can call migrate() on the current vault contract to redeem their shares from this vault and join the deposit queue in the new vault. Also, users can leverage the migration contract, an implementation that has already set the origin and destination vaults, to join the deposit queue in a new vault. In addition to the migrate() available on the current vault, the migration contract also has a permit feature that allows users to perform the migration in one transaction.
In order to fulfill its value proposition of being a one-click deposit investment product, Pods Finance team implements permit functionalities on deposits and other key vault functions. With this, users can sign a permit off-chain and then use the functions with permit integrations to perform the approval in the same transaction.
Some privileged roles exercise powers over the vault and periphery contracts:
Owner of the ConfigurationManager
Controller of the vault
controller can start a new round anytime.controller can end a round.Investor of the vault
Here we present our findings.
When users attempt to join the STETHVault, they first join a DepositQueue. While being part of the DepositQueue, users can elect to either refund their deposit or wait until the round ends for their deposit to be processed. Both of these actions will then call the underlying remove function in order to restructure the DepositQueue.
The issue is that the remove function performs gas-intensive operations that drastically limit the upper bound on the queue size, which we conservatively estimate to be in the magnitude of a few thousands. With a large enough queue, any transaction attempting to refund or process even a single deposit will exceed the block gas limit of 30,000,000 gas. This could occur due to organic user growth, or a malicious actor may choose to send dust amounts of stETH to the vault from different addresses to take up positions in the queue. Without the ability to process or refund users, all current and future deposits made to the vault would be locked in the DepositQueue.
Consider accounting for the gas usage of all queue operations and ensuring they cannot reasonably exceed the block gas limit. Some recommendations to consider:
DepositQueue so that operations such as remove can be performed in significantly less (and preferably constant) gasDepositQueue size to guarantee it can be processed each roundDepositQueue, such as an EnumerableMap, to reduce code maintenance overhead and ensure necessary operations can be performed in constant gasUpdate: Fixed in commit e05f165.
When processing queued deposits, the amount of assets that have been processed are accounted for in the processedDeposits variable and used to calculate the amount of shares to be minted. This variable is reset to 0 at the beginning of each round.
However, each time the processQueuedDeposits function concludes, the processed amount will be removed from totalIdleAssets, hence increasing the totalAssets. When the queue is processed the next time, the totalAssets value already contains the amount of assets processed during the round. By adding the processedDeposits to the totalAssets again, the previously processed amount is double counted, which leads to an incorrect calculation for the amount of shares minted to the next depositor in the queue.
This penalizes depositors by minting fewer shares if they are not included in the first call of the processQueuedDeposits function. This issue is particularly severe if each deposit is processed individually. Hence it breaks the system invariant that the number of splits should not modify the result when processing the deposits.
Additionally, this means users not processed in the first round may receive fewer assets than they deposited when they redeem their shares. This breaks another important system invariant that 100% withdrawal in stETH should always be equal to or higher than the initial deposit amount.
Consider correcting the total asset calculation to ensure the processed amount is accounted for accurately.
Update: Fixed in PR#54, with commit 3ad9b76 being the last one added.
During the withdrawal process, users can specify the amount of assets to withdraw, which is then rounded down to shares.
When the asset amount specified by the user is less than the minimum amount that can be converted to a unit share, the shares argument is zero in the internal withdraw function but the assets argument is not. Hence, with zero shares, all internal calls can succeed and a non-zero amount of asset token will be transferred out to the receiver without burning any shares. This process can be repeated many times to drain the entire vault. The attack can also be executed with any asset amount by burning a rounded-down amount of shares and extracting the excess assets.
Since the vault is expected to become more valuable over time due to its yield strategy, this could lead to a profitable attack when one share is worth more than the cost.
Consider rounding up the shares for a given amount of assets during withdrawal.
Update: Fixed in PR#46, with commit 5ac5e3c being the last one added.
withdraw can disable endRoundThe controller should be able to call the endRound() function any time after a round starts to remove a portion of the accrued interest from the vault to the investor account for the execution of investment strategies. In the _afterRoundEnd hook, the totalAssets() is compared to the lastRoundAssets to compute the accrued interest. When a user withdraws assets before the endRound, the lastRoundAssets is updated in the _beforeWithdraw() hook, where the withdrawn shares are rounded down to the corresponding asset value, which is subsequently subtracted from the lastRoundAssets.
This may cause the lastRoundAssets to be more than the totalAssets() when a user calls the withdraw function. This can result in a revert due to underflow. In such a case, the controller cannot end a round and must wait for the accrued interest to accumulate enough value.
Consider keeping track of the withdrawn asset directly. Also, consider making sure the key system functionality is not impeded by users’ actions.
Update: Fixed in PR#73, with commit b258ca9 being the last one added.
The internal _deposit function handles user deposits, transferring a specified amount of stETH from msg.sender to the vault. Before moving the funds, it adds the deposit to the queue, which is processed later by the processQueuedDeposits function.
As the underlying token could have hooks that allow the token sender to execute code before the transfer (e.g., ERC777 standard), a malicious user could use those hooks to re-enter the deposit function multiple times.
This re-entrancy will result in an increment in the receiver balance on the queue, even though this balance will not correspond to the actual amount deposited into the vault.
In the current implementation, the _deposit function in the BaseVault contract is overridden by the implementation in the STETHVault, which has the correct order of operation. However, the BaseVault is likely to be inherited by future vaults, so it is crucial to have the correct _deposit implementation in this contract in case it is not overridden.
Consider reordering the calls, doing the transfer first, and then adding the receiver to the queue to prevent this reentrancy scenario. Also, consider using OpenZeppelin’s ERC4626 implementation, which already has this solution implemented.
Update: Fixed in PR#41, with commit 2ffcb1e being the last one added.
When processing the queued deposits, shares are minted to the receiver in the queue according to the amount of assets deposited. However, the amount of shares minted is always rounded up. This means that one can always receive 1 vault share with a 1-wei deposit.
As the vault is expected to be increasing in value from yield rewards, 1 vault share will be worth more than 1 wei asset eventually. A malicious user can spam the deposit queue with 1-wei deposit from many accounts to get 1 share each and then redeem them for more assets when each share is worth more.
Consider rounding down when minting vault shares.
Update: Fixed in PR#46, with commit 5ac5e3c being the last one added.
The vault implements a mintWithPermit and depositWithPermit function intended to allow users to transfer assets to the vault in a single transaction. However, the vault’s underlying asset is intended to be stETH which does not have a permit function. Currently, any user who tries to perform a mintWithPermit or depositWithPermit will have their transaction reverted due to the stETH contract’s fallback function.
Consider removing the mintWithPermit and depositWithPermit functions. We note that wstETH does have a permit function for future considerations.
Update: Acknowledged, and will not fix. Pods Finance team’s statement for this issue:
We decided not to fix this issue because we may implement assets similar to LIDO as the yield source where they may have permit functionality (aTokens, for instance).
The ETHAdapter contract is used as a proxy to allow users to interact with the vault through sending and receiving ETH instead of stETH. The adapter achieves this by converting ETH and stETH through a curve pool and then forwarding interactions to and from the vault. In the course of a normal withdrawal or redemption transaction, the ETHAdapter will pull the funds out of the vault before passing them on to the designated receiver. During the moment the ETHAdapter is holding the funds, it first converts all of its stETH to ETH, and then sends its entire ETH balance to the receiving address.
The issue is that the ETHAdapter sends its full balance to the receiver each time, meaning any ETH or stETH that is mistakenly sent to it can be drained by any user who performs a withdrawal or redemption on the ETHAdapter. This is exacerbated by the fact that the vault is passed in as a parameter, potentially allowing a user to perform withdrawals and redemptions without interacting with the actual stETH vault.
Consider transferring out the exchanged balance from the curve pool to the receiver instead of the entire balance of the ETHAdapter. Also consider implementing a rescue or sweep function to allow the recovery of funds that are accidentally sent to the ETHAdapter.
Update: Acknowledged, will not fix. Pods Finance team’s statement:
For now, we do not want to take action in case of funds sends by mistake to our contract. We see this as a low-priority issue.
We found the following instances of incorrect calculations in view functions that are not currently called internally:
previewWithdraw function, the DENOMINATOR over invertedFee is always bigger than 1 when the fee is non-zero. Hence, the final returned shares are always an overestimate. Further, the withdrawFeeRatio is multiplied to shares instead of assets as in other instances such as _getFee. Consider correcting the withdrawal fee calculation.assetsOf function over-estimates the actual commitment by an additional committedAssets. Consider removing the extra component.Update: Partially fixed in commit 6d37029 in PR#75. previewWithdraw does not include fees as stated in the EIP.
The AuxMath library is a custom auxiliary math library that performs multiplication and division with rounding specifications.
The implementation of mulDivUp and mulDivDown first compute the multiplication of a uint256 with a uint256 without taking into account the possibility of an overflow in the product. The product could overflow into a uint512 in the multiplication step even if the result were to fit into a uint256 after the subsequent division. Hence the current implementation will not give the right result in such a case.
Consider using OpenZeppelin’s Math library which implements the mulDiv function that is developed especially considering these scenarios and is widely accepted in the ecosystem.
Update: Fixed in commit 2a8c58a.
The amount of vault shares available for minting is limited by the availableCap. Similarly, the availableCap also limits the amount of assets that may be deposited and then later converted into vault shares for the depositor. The maxMint and maxDeposit functions do not account for the availableCap and instead return the max uint256 value when called. In order to conform to the EIP4626 standard, these functions must return the real amount that can be minted or deposited.
Consider changing the maxMint and maxDeposit functions to return a value that accounts for the availableCap.
Update: Fixed in PR#42, with commit 3159de5 being the last one added.
When a user deposits into the vault by joining the deposit queue, the corresponding shares, yet to be minted, are deducted from the spending cap immediately. The spending cap is restored only at withdrawal when the shares are burned.
If one user decides to use the refund function to leave the queue before the round ends, the spending cap will not be restored. When the cap is not zero, a malicious user with a sufficiently large amount of assets could repeatedly deposit and refund to reach the cap limit and stop other eligible users from joining the queue.
Note that the spendCap variable cannot be manually restored, but the owner can reset the cap to a higher value to unlock the deposit.
Consider accounting for the available cap during refunds from the queue.
Update: Fixed in commit 4a2e475.
Deposits added to the queue are point in time stETH balance amounts. The stETH token balance rebases regularly to account for yield, and in the event of slashing, may be subject to a negative yield. In the event that a stETH token rebase is negative between the time a user deposits and calls for a refund, the vault will over credit the user by the rebase difference.
Consider handling the deposits in the queue in shares instead of balances to account for rebase changes on refunds.
Update: Acknowledged, will not fix. Pods Finance team’s statement:
Although we agree with the issue, we won’t prioritize it right now because of timing. In case of a slashing event, we can refund the vault, transferring funds directly to the contract.
The configuration manager determines the maximum vault shares available for minting. When setting a new cap through the setCap function, the target contract address is checked for the zero address, and a SetCap event is emitted.
However, the underlying setParameter function can be called directly to change the maximum vault shares available. This will bypass the zero address check and emit a ParameterSet event instead of the expected SetCap event.
Consider ensuring consistency between the two mechanisms in setting the share cap. Depending on the desired outcome, this could involve removing the setParameter function and adding individual get and set functions for all necessary configuration parameters.
Update: Fixed in commit cbe61d0.
There are many instances throughout the codebase where changes can be made to improve gas consumption. For example:
public functions might consume more gas than external functions. This is because with public functions, the EVM copies inputs (especially dynamic-sized arrays) into memory, while it reads from calldata if the function is external, which is cheaper. Throughout the codebase there are public functions like mintWithPermit, convertToSTETH, convertToETH, decimals, symbol, name and assetsOf that might be declared as external. Consider reviewing the entire codebase for more instances.toAddress has a conditional statement that returns the zero address if the value passed is equal to zero. This additional check increases the gas consumption and does not add any benefit even if the provided value is equal to zero.processedDeposits, isProcessingDeposits, roundAccruedInterest, endSharePrice and the while loop counter inside the remove function.for loop inside the refund function does not cache the result of depositQueue.size() and perform unnecessary operations on each iteration. The same issue occurs in the last while loop of the remove function.name, symbol and the operation 10**sharePriceDecimals.When performing these changes, aim to reach an optimal tradeoff between gas optimization and readability. Having a codebase that is easy to understand reduces the chance of future errors and improves community transparency.
Update: Partially fixed in PR#81, with commit 6f1bb8e being the last one added.
The name attribute in the constructor of the BaseVault, passed to the draft-ERC20Permit contract does not match the name returned from the STETHVault implementation. All vaults that use the same underlying asset and inherit from BaseVault will have the same hashedName when calculating the domain separator by the EIP712 standard. Although this does not represent a risk, it is recommended that the name sent to the ERC20Permit match the ERC20 token name.
Also, the name and symbol defined in the BaseVault constructor are inconsistently overridden in the STETHVault implementation.
To avoid complications and improve code clarity, consider passing both the name and symbol from the STETHVault contract as parameters to maintain consistency. This also saves calculating the name and symbol at every call, thus saving more gas.
Update: Fixed in commit 61a9996.
Comments are meant to be helpful and describe the intent of the code block. However, in multiple parts of the codebase, comments are not consistent with the code that is written or do not add any value to the reader. These are some examples:
_deposit in both the STETHVault and BaseVault contracts does not create any shares even though the comments say so.setParameter, getParameter and getGlobalParameter does not add any value to the reader.target parameter on setParameter has a wrong description._spendCap has the same description as availableCap.Consider correcting the comments to not mislead users or developers and write comments that add value to the reader. This will improve code clarity and consistency between docstrings and contract implementations.
Update: Fixed in commit 3ad59d0.
Throughout the codebase, there are several files that do not have docstrings. For instance:
ConfigurationManager contract has no docstrings for any of its state variables.IConfigurationManager interface does not have any docstrings.ICurvePool interface is for an external contract, docstrings need to be added to explain the interface purpose and why each function is used within the codebase.IERC4626 interface has no docstrings for the interface description and also for any of its events.IVault interface has no docstrings for the interface description and also for any of its events and errors.AuxMath library does not have any docstrings.CastUint library has no docstrings for the library description.DepositQueueLib library just has some inline comments inside the remove function.Capped contract has no docstrings for the contract description, state variables, constructor, and its custom error.ETHAdapter contract just has some inline comments for the receive function.Migration contract does not have any docstrings.BaseVault contract has no docstrings for some of its state variables (configuration, _asset, currentRoundId, isProcessingDeposits, EMERGENCY_INTERVAL, processedDeposits, _lastEndRound and depositQueue), the constructor, and the functions depositWithPermit and mintWithPermit.STETHVault contract has no docstrings for some of its state variables (sharePriceDecimals, lastRoundAssets, lastSharePrice, and investor) and the constructor. Adding comments in the internal function _afterRoundEnd can be beneficial for easy understanding of its purpose.This hinders reviewers’ understanding of the code’s intention, which is fundamental to correctly assess not only security but also correctness. Additionally, docstrings improve readability and easy maintenance.
Consider thoroughly inserting documentation above each contract, interface, library, function, state variable, event, and custom errors. Functions implementing sensitive functionality, even if not public, should be clearly documented as well. When writing docstrings, consider following the Ethereum Natural Specification Format (NatSpec).
Update: Acknowledged, will not fix. Pods Finance team’s statement:
We didn’t prioritize this because of time constraints.
Any user may process deposits in the queue after a round ends. After deposits are processed, the section of the queue they were held in is removed. This means that any user who processes deposits will be changing the size and indexes of the queue for all others. This may lead to reverted transactions if an index that was previously thought to be in range no longer exists in the queue. It may also lead to some processing of unintended indexes and others missing their targeted indexes in the queue.
Consider having additional validations on the queue size and indexes to ensure the correct transactions are executed without going out of bounds.
Update: Fixed in commit e05f165.
Refunds are processed by searching the entire queue front to back for a user’s position. Currently, the gas price to refund increases linearly based on the user’s position in the queue. Users that have joined the queue later in the round will incur a more significant gas penalty and, in an extreme case, could be prevented from refunding.
Consider keeping track of a user’s queue position alongside other attributes for easier retrieval and update.
Update: Fixed in commit e05f165.
startIndex is greater than or equal to endIndexThe remove function is responsible for removing records from the queue. When called, it checks if the endIndex is greater than the startIndex to execute the function’s implementation. The issue is that if the condition is not met, the transaction is not reverted, and the code execution continues. This could cause processQueuedDeposits to fail silently when the indexes are not valid.
Although it does not currently cause any errors other than spending gas unnecessarily, consider rolling back the transaction if the condition is met to avoid unwanted behaviors.
Update: Fixed in commit e05f165.
The system allows users to migrate their shares to a new vault with the same underlying asset. There are a few risks regarding the current migration mechanism design.
migrate function in the old vault or through the exclusive Migration contract. There is no clarity on which route is preferred as both essentially perform the same functionality. Consider removing duplicate actions that can lead to user confusion.Migration contract sets both the old vault and the new vault addresses as immutable variables in the constructor. This limits its use case to one single migration. It would not be reusable across multiple vaults with the same underlying assets and different investment strategies.Migration contract due to direct transfers or mistaken withdrawals, the entire balance of the contract will be deposited to the new vault by the next user. This design is not resistant to genuine user mistakes and prohibits recovering user funds. Consider using the returned value from the redeem function to migrate the correct amount of assets and establishing a sweep or rescue function for funds locked in the contract.Update: Partially fixed in commit d97672d.
The investmentYield and the investmentAmount are transferred back and forth from the vault to the same investor account. Instead of making two separate transactions, one can compare the two values and reach the final balance in one transaction.
Consider simplifying the code for clarity and readability.
Update: Acknowledged, will not fix. Pods Finance team’s statement:
We actively preferred to leave the code in the back-and-forth way, so it’s easier to understand the flow of funds of the operations.
Although constants are generally used correctly throughout the codebase, there are a few occurrences of literal values being used with unexplained meaning. For example, the following blocks use hardcoded values:
ETHAdapter contract, the pool’s coins method is called with arguments 0 and 1, with no explanation.convertToETH and convertToSTETH, the pool’s get_dy method is called with arguments 0 and 1 with no explanation.deposit function, the pool’s exchange method is called with arguments 0 and 1, with no explanation.To improve the code’s readability and facilitate refactoring, consider defining a constant for every magic number, giving it a clear and self-explanatory name. Consider adding an inline comment explaining how they are calculated or why they are chosen for complex values.
Update: Acknowledged, will not fix. Pods Finance team’s statement:
We didn’t prioritize this because of time constraints.
Throughout the codebase there are require statements and custom error blocks that need multiple conditions to be satisfied. For instance:
ETHAdapter contract there is a require statement that checks if coin 0 from the pool is equal to the ETH address and if coin 1 is equal to stETH address. If one of the conditions is not met, there is no error message to indicate why the transaction was reverted.migrate function in the BaseVault contract has a conditional block that checks if the vault is not allowed and if the destination vault does not have the same underlying asset as the current vault. Although the custom error that is thrown is not wrong, it might be better if these two validations are separated to return more specific error messages.onlyRoundStarter modifier has a custom error block that checks if less than a week has passed since the last end round and if the caller is different from the controller. If the conditions are met, the error returned is that the caller is not the controller. Consider separating both conditions in order to throw a more specific error message.To simplify the codebase and to raise the most helpful error messages for failing blocks, consider having a single require statement with an appropriate error message or a single custom error block per condition.
Update: Acknowledged, will not fix. Pods Finance team’s statement:
We didn’t prioritize this because of time constraints.
To favor explicitness and readability, several parts of the contracts may benefit from better naming. Our suggestions are:
withdrawFeeRatio can be changed to getWithdrawFeeRatio. The current name suggests that the action is the withdrawal of fee ratio but what it actually does is to query the WITHDRAW_FEE_RATIO value.ConfigurationManager__InvalidCapTarget can be changed to ConfigurationManager__TargetCannotBeTheZeroAddress or something similar, since the only invalid target is the zero address.Update: Fixed in commit 88ce562.
Non-explicit imports are used throughout the codebase, which reduces code readability and could lead to conflicts between names defined locally and the ones imported. This is especially important if many contracts are defined within the same Solidity files or the inheritance chains are long. For instance, but not limited to:
BaseVault.solEthAdapter.solFollowing the principle that clearer code is better code, consider using named import syntax (import {A, B, C} from "X") to explicitly declare which contracts are being imported.
Update: Acknowledged, will not fix. Pods Finance team’s statement:
We didn’t prioritize this because of time constraints.
All contracts within the codebase use compiler version 0.8.9, an old version with various bugs that have been fixed in later releases.
Consider taking advantage of the latest Solidity version to improve the overall quality and security of the codebase.
Update: Fixed in commit 93956a9 in PR#66.
Throughout the codebase, there are state variables that lack explicitly declared visibility. For instance:
CAP in ConfigurationManager.solETH_ADDRESS in ETHAdapter.solSTETH_ADDRESS in ETHAdapter.solfrom in Migration.solto in Migration.solFor clarity, consider always explicitly declaring the visibility of variables, even when the default visibility matches the intended visibility.
Update: Fixed in PR#64, with commit a3eb939 being the last one added.
Style inconsistencies can be seen throughout the codebase. For example:
require statements for error handling and others use custom errors. In general custom errors are preferred over require statements.StartRound, the function’s name with the first letter capitalized, and others, like FeeCollected, describe the action that was executed. Consider following only one convention on the entire codebase.investorRatio does not use all capital letters with underscores separating words but all other constants in the codebase do. Consider changing it to INVESTOR_RATIO.Update: Fixed in commit cf6ae4d.
Across the codebase, there are some typographical errors in the comments. Some examples are:
ConfiguratorManager should be ConfigurationManager.MAX_WITDRAW_FEE should be MAX_WITHDRAW_FEE.split.Consider correcting the above and any further typos in favor of correctness and readability.
Update: Fixed in PR#41 commit a515964 and PR#69 commit 4ec70cb.
We found some internal functions with unused arguments in the implementation, hence are unnecessary to be defined in the base contract. For example:
_afterRoundStart(uint256) function does not require an argument in the implementation but is defined with an argument in the base contract._beforeWithdraw(uint256 shares, uint256) function does not require a second argument but is defined with a second argument in the base contract.Consider removing unnecessary arguments for clarity and readability.
Update: Fixed in commit f100925.
The following instances of internal virtual functions are implemented in BaseVault.sol, which are later overridden by a different implementation in STETHVault.sol.
_withdraw function is overridden by the same implementation._deposit function is overridden by a different implementation.Consider removing unnecessary implementations for readability and maintenance.
Update: Acknowledged, will not fix. Pods Finance team’s statement:
We didn’t prioritize this because of time constraints.
The global address and the getGlobalParameter() function of the ConfigurationManager contract are supposed to manage global configurations that do not depend on a specific vault. However, the purpose of those entities is unclear, as they are not used anywhere in the codebase.
To avoid confusion and favor explicitness, consider documenting the purpose of these functions and variables. If they are not expected to be used, consider removing them from the codebase to reduce the code’s size and attack surface.
Update: Acknowledged, will not fix. Pods Finance team’s statement:
We didn’t prioritize this because of time constraints.
Throughout the codebase imports on the following lines are unused and could be removed:
Consider removing unused imports to avoid confusion and improve the overall readability of the codebase.
Update: Fixed in commits b702189 and 89869ad.
Throughout the codebase, there are several occasions where named return variables are declared but not used. For instance:
newShares on migrate and migrateWithPermit functionsstETHAmount on convertToSTETHethAmount on convertToETHdepositedAssets on _deposit on both STETHVault and BaseVaultroundId on startRoundshares on maxMint, maxRedeem, deposit, convertToShares, and previewDepositassets on maxWithdraw, maxDeposit, convertToAssets, and previewMinteffectiveAmount on _stETHTransferFromConsider adding those variables to function implementations or remove them.
Update: Fixed in commit 1228184.
Three critical and three high-severity issues are found among various lower-severity issues. The system documentation and diagrams provided by the Pods Finance team made it easier for the auditors to assess and comprehend the code. Throughout the audit, the team has been responsive and active in responding to questions and suggestions from the auditors.
Given the number of important issues raised, we believe there could be more undiscovered issues, and we anticipate a significant refactor to the codebase. If this turns out to be the case, we highly recommend another round of audits on the new code.
Update as of November 15th:
The Pods Finance team considerably refactored the codebase to address many findings that were initially reported; however, another critical issue was discovered during the fix-review process. OpenZeppelin worked with the Pods Finance team to address this issue, though we noted that the fix-review process was limited in assessing security issues within the significant changes made to the codebase. OpenZeppelin strongly recommended the Pods Finance team obtain a full code re-audit and the Pods Finance team is proceeding with that.
With the goal of providing a complete security solution, we have identified several actions that might benefit from on-chain monitoring to provide useful alerts and insights about on-chain protocol activities. Here are some example scenarios that would benefit from monitoring:
Deposits to the vault with dust or large amounts. This could represent a potential threat to manipulate vault share calculations.Owner of the ConfigurationManager, the Controller of the vault, and the Investor of the vault. Consider monitoring events related, but not limited to, ownership changes, vault configuration modifications, and transfers initiated by the Investor to somewhere other than the vault as these may signal key compromise of a privileged role.Critical Issue: Malicious user can manipulate shares calculation and steal deposits
As part of the fixes developed and provided during the fix-review process, it was noted that the current queue setup could potentially be exploited by Maximum Extractable Value (MEV) searchers if the endRound transaction is sent through the public mempool.
In the scenario when the MEV searcher has sufficient liquidity, the following strategy can be viable.
endRound.endRound transaction with a transaction that
In this way, the MEV searcher gets one share while everyone else gets zero shares due to the rounding in the division. Since no deposit or withdrawal is possible during the processing period, the MEV searcher waits until either the next startRound or one week has passed. Then the searcher can redeem its shares and remove the entire liquidity from the vault. We note that the searcher may need to back-run the startRound transaction to ensure that its one share is not diluted by further large sum deposits. Now, the vault is back to zero shares, zero liquidity and the same attack can be repeated.
In case the MEV searcher does not have enough liquidity, a milder version of the same strategy can be carried out by processing only part of the queue with the correct ordering.