UMA Audit – L2 Bridges

 

UMA is a platform that allows users to enter trust-minimized financial contracts on the Ethereum blockchain. We previously audited the decentralized oracle, a particular financial contract template, some ad hoc pull requests, the Perpetual Multiparty template and various incremental pull requests over a longer engagement. In this audit we reviewed a new mechanism to quickly send tokens from a layer 2 chain to the Ethereum mainnet, and related changes to the Optimistic Oracle. The review was completed by 2 auditors over 3 weeks.

Scope

The audited commit is f24ad501c8e813cf685f72217e7f13c8f3c366df and the scope includes the following contracts:

  • contracts/insured-bridge/* (excluding test contracts)
  • contracts-ovm/insured-bridge/implementation/*
  • contracts/common/implementation/AncillaryData.sol
  • contracts/oracle/implementation/SkinnyOptimisticOracle.sol

We also reviewed the changes to solidity files in Pull Request 3445.

All external code and contract dependencies were assumed to work as documented.

System overview

The supported layer 2 (L2) chains, Optimism and Arbitrum, provide a mechanism to transfer funds to the Ethereum mainnet (L1). However, for security reasons, there is a significant delay before those transfers are finalized. To address this, L2 token holders can deposit funds in an UMA contract, the Deposit Box, knowing the tokens will eventually be transferred (as a batch) to an L1 UMA contract, the Bridge Pool. There is a separate Bridge Pool for each token to be transferred.

Following an L2 deposit, anyone can relay the details to the L1 Bridge Pool, which waits a short period in case someone wants to dispute the relayed information. All disputes are handled by the Skinny Optimistic Oracle (described below). Before accepting the relay, liquidity providers must pre-fund the Bridge Pool contract in exchange for LP tokens. Undisputed relays are assumed valid, and the Bridge Pool completes the transfer using its own reserves, where a fraction of the transfer is diverted to the relayer and a fraction is retained as liquidity fees. The funds will eventually be replenished when the L2 deposit is finalized, and the liquidity fees are assigned to the LP token holders.

The Bridge Pool also allows anyone to individually fund a transfer (without the Bridge Pool reserves) before the dispute period expires, in exchange for a fraction of the transfer amount. Since the relay can still be disputed, these funds would be lost if the relay was deemed to be incorrect. It is expected that in most cases, this mechanism will allow users to experience immediate L2-to-L1 token transfers.

The Skinny Optimistic Oracle is conceptually very similar to the existing Optimistic Oracle. It provides an incentive mechanism for users to simply assert the result of an oracle request, which is assumed to be accurate if it is not disputed. Disputes are relegated to the slower DVM mechanism described in our previous audit reports. The main difference is that the new version requires users to provide all relevant information when executing function calls, so the values do not need to be saved or retrieved from storage. It also removes the ability for requesters to change configuration parameters in active requests.

We have previously reviewed the Long-Short-Pair contract, which provides a generic mechanism to create various financial instruments. These contracts can be resolved when their expiration time is reached and the settlement price is known. The changes introduced by Pull Request 3445 introduce the possibility of resolving the contracts early if the settlement price is known before the expiration time.

Privileged Roles

The L2 deposit boxes have several configuration parameters, including the supported tokens and the maximum rate for sending batched tokens over the bridge to L1. They must also be configured to ensure consistency between the L1 token contract, the L2 token contract, and the corresponding Bridge Pool. These parameters are set by an administrator contract on L1, which also parameterizes the dispute resolution process of the bridge pools. The contract is expected to be controlled by the UMA governance mechanism, so users must trust this process to manage the system sensibly and fairly.

Ecosystem dependencies

All reviewed components use time-based logic, which means they depend on Ethereum availability. In particular, if dispute transactions are delayed significantly then invalid relays or price proposals may be incorrectly confirmed.

Additionally, the token bridge implicitly assumes that all funds sent to L2 deposit boxes will eventually be transferred to the corresponding L1 bridge pool. This relies on the correct and continued functioning of the Optimism and Arbitrum bridges, and their dispute resolution mechanisms.

Lastly, tokens sent to the L2 deposit box are assigned to the bridge pool in L1, not the intended recipient. To retrieve the funds from the pool, L1 token holders must first match them with additional tokens. Therefore, the mechanism relies on a sufficiently deep market of L1 tokens to ensure there is always liquidity.

Client-reported issues

During the audit, the UMA team independently identified a number of issues and behaviors worth highlighting:

If the Optimistic Oracle or Bridge Admin parameters change during a relay’s challenge period, then disputing the relay deletes the relay with no additional recourse for either the proposer or the disputer. For example, imagine that the relay is sent for the collateral token TOKEN_A, but in the middle of the relay TOKEN_A is removed from the collateral whitelist. A dispute will now revert since you cannot submit any price requests to the OO or DVM for unwhitelisted collateral. Since we don’t want to block valid dispute requests, the BridgePool will delete the pending relay for TOKEN_A in the case of a dispute. The consequences of this design decision are:
1. An increase in the final fee will cause: Any outstanding relay on that token to be “cancellable” through dispute whether right or wrong. A cancellation doesn’t advantage either party, so it assumes the existence of honest disputers who are willing to kill the (rare) bad request that happens to exist during a final fee change execution. This also means that it’s possible for a griefer to spend gas to cancel relays and force them to be re-relayed.
2. A de-whitelisting of the identifier or token, which shouldn’t happen unless something goes very wrong would cause:
3. An extended period of undisputable requests where any request can be canceled and there exist no economic incentives for disputing. This seems better than the alternative of blocking disputes completely, but it’s admittedly quite bad, as any griefer can block relays indefinitely by paying gas or send bad relays without punishment (other than gas fees).

Note: this is an alternative to having the OO that “freeze” parameters such as the final fee or collateral whitelist for some time, but this would require additional calls to the OO, which would be costly for the happy path.

Relays can be sped up via speedUpRelay() after they pass liveness. While we don’t see any risk with this, it does open up the possibility for flash loans + speed ups + settle after liveness, giving the flash borrower the instant relay fee for “free”. We prevent this in this proposed PR.

On settle, if the BridgePool is a WETH pool and the recipient is a contract that is not payable (cannot accept ETH), then settle will fail. We plan to fix this and fallback on sending WETH, but there is no outstanding work done yet on this.

In relayDeposit, we check that the BridgePool‘s balance is greater than the amount to relay PLUS the proposer bond. This is an outdated check and too conservative since the proposer bond is pulled from the user after the check. We address this in this proposed PR.

I just caught a bug where chainId in BridgePool, included as part of the Deposit struct and as a function input to all of the relay-related functions (i.e. relayDeposit, speedUpRelay, settle) is type uint8. This is too small of a type to handle Arbitrum for example whose ID is 421611. We actually caught this bug and fixed it on the L2 side: BridgeDeposit has set its chainId type to uint256. This PR will make the chainId on BridgePool match the type on BridgeDepositBox: UMAprotocol/protocol#3463

Previously, the dispute function did not subtract off the deposit amount from pendingReserves (this is the variable that tracks how much of the reserve pool is locked up due to relays that have not settled yet). The result was that each dispute would indefinitely lock up the relay amount in the pool. It cannot be withdrawn by LPs or used by future relays. Fix is here: UMAprotocol/protocol#3473.

We found a bug in BridgeDepositBox where hasEnoughTimeElapsedToBridge does not check if a uint256 value is equal to 0 by default: Fixed in PR 3484

The exchange rate method (which is state modifying) is called between tokens being transferred in and LP tokens being minted in the addLiquidity method of the bridge pool contract. This computation needs to be moved up to the top of the method. This causes very strange state values. See PR here to fix.

The view method liquidityUtilizationPostRelay (which is only used offchain), reports an incorrect utilization number. The denominator on this line shouldn’t be just liquidReserves, it should instead be a representation of unutilized and utilized reserves. Fixed here.

Update

In addition to the issue fixes, we also reviewed the following incremental changes:

  • PR3500 removes the redundant token parameter from BridgePool events.
  • PR3478 adds the DVM final fee to the list of locally cached variables.
  • PR3460 accounts for a possible negative liquidity utilization edge case (in addition to addressing N04)
  • PR3482 removes redundant files and updates OVM constants in accordance with OVM 2.0 changes.
  • PR3585 updates the BridgeDepositBox interface for consistency and uses the OpenZeppelin SafeERC20 library.

While reviewing the fixes we identified another issue. When determining the value of BridgePool LP tokens, there is an intermediate calculation that could unexpectedly negative overflow, which would temporarily disable adding and removing liquidity. The calculation should be reordered to add the utilized reserves before subtracting the undistributed fees.

Critical severity

[C01] Trapped Proposer Reward

The LongShortPair contract retrieves a proposer reward from whichever address triggers the expiration, which is used to incentivize price proposals in the Optimistic Oracle. However, the LongShortPairCreator contract also retrieves and forwards the funds from the deployer address. These additional funds are not passed to the Optimistic Oracle, and instead remain trapped within the LongShortPair contract.

Consider removing the duplicate transfer.

Update: Fixed as of commit 9bab1ff353a417952ba8c96a098773f340d9da17 in PR3523.

High severity

[H01] Concurrent relays deplete reserves

The relayDeposit function of the BridgePool contract ensures the contract has sufficient funds to execute the transfer. However, it does not account for the pending reserves, which tracks funds that are earmarked for active relays. Therefore, multiple simultaneous relays may rely on the same funds, and they may not all be settleable immediately. In particular, with a steady stream of transfers, instant relayer returns may be delayed indefinitely.

Consider preventing relays that would cause the pending reserves to exceed the liquid reserves.

Update: Fixed as of commit 6290f3facbca8d878605a1d390ed59d4b6b6db02 in PR3501.

[H02] Bridging parameter bounds don’t match

The deposit function of the BridgeDepositBox contract, deployed on layer 2 chains, is used to bridge funds between the L2 and L1. In particular, relayers are incentivized to relay the transaction details on the associated L1 BridgePool. However, the deposit box uses inclusive bounds to restrict the relay fees, while the bridge pool uses exclusive bounds. This means that some deposits (with 25% relay fees) cannot be relayed, and the funds will be inaccessible on both layers.

Consider synchronizing the validations on both layers to ensure all valid deposits can be relayed.

Update: Fixed in commit 2345966b3a2ace0159379b3a13256cc1a4c5d52f of PR3494. This was originally classified as Critical severity but was downgraded when the UMA team pointed out the funds would not be strictly trapped and could be released if the DVM voters agreed to accept a modified relay description for affected deposits.

Medium severity

[M01] Callbacks to wrong address

The SkinnyOptimisticOracle invokes callback functions on the price requester, if they exist, so the requester can respond to significant state changes. However, the callback is incorrectly invoked on the price proposer instead of the price requester in the proposePriceFor function. This means the price requester is unable to respond to price proposals.

Fortunately, this feature is not used in the current code base. Nevertheless, consider invoking the priceProposed callback on the requester.

Update: Fixed at commit 7bd3faeb6f3706132f77b9ba2dce192d1a151e74 in PR3531.

[M02] Faulty append function

The appendKeyValueBytes32 function should combine its inputs into a formatted bytes array. However, the currentAncillaryData is incorrectly discarded.

Since the ancillary data influences the oracle resolution process, an incorrect value could undermine the oracle results. Fortunately, there is only one call to appendKeyValueBytes32 in the codebase, and it uses an empty currentAncillaryData buffer, so the bug does not affect this case.

Consider updating the appendKeyValueBytes32 function so that the currentAncillaryData is included in the returned bytes array.

Update: Fixed in commit 5609433c154f47e8ee9c52f9b6d7c787fbe3e455 of PR3532.

[M03] Incomplete ancillary data validation

The LongShortPair constructor confirms that the customAncillaryData is small enough. However, it doesn’t account for the early expiration field. This means the Optimistic Oracle may unexpectedly reject early expiration price requests, which would disable this feature.

Consider updating the validation to account for the additional field.

Update: Fixed as of commit 4a56e66492f40e20254cebb145c2d91304f7cb43 in PR3524.

[M04] Mishandled zero timestamp

In the LongShortPair contract, a zero early expiration timestamp is used as a flag to indicate that nobody has triggered the early expiration mechanism. However, it is possible to trigger that mechanism with a zero timestamp. In this scenario, the Optimistic Oracle will be be invoked but the protection against subsequent price requests will not be effective. Fortunately, once the settlement price is chosen, it won’t be overriden, so this won’t lead to inconsistent settlements. Nevertheless, a subsequent price request could change the recorded early expiration timestamp, even if the zero timestamp is used to determine the settlement price. It could also emit a misleading event.

Consider preventing early expiration using the zero timestamp.

Update: Fixed as of commit 11d287c07c93c04f534b2ef3c869966d9f18ac60 in PR3526.

[M05] Possible zero bond

The requestPrice function of the SkinnyOptimisticOracle contract uses the final fee as the bond if the bond is not specified. However, the requestAndProposePriceFor function may use a zero bond, which contradicts its @notice and @param comments. A zero bond weakens the incentive against invalid proposal or disputes.

Fortunately, the only call to this function in the code base sets a proposer bond. Nevertheless, consider using the final fee if the bond is not specified.

Update: Fixed as of commit daaabfc342ba1395a577159b6eb26adb20fcd232 in PR3534.

[M06] Unnecessary administrator privileges

The BridgePool contract inherits from ExpandedERC20 so that it can issue LP tokens to liquidity providers. This inherits the functionality of OpenZeppelin’s ERC20 contract and also provides administrator privileges to the contract deployer, which allows them to mint and burn LP tokens. However, this power is not required and, if exercised, could unfairly penalize liquidity providers.

Consider modifying BridgePool to inherit directly from ERC20 instead of ExpandedERC20.

Update: Fixed in commit 370e8b21b660543eadbd764fed984a5bdeddce24 in PR3492.

Low severity

[L01] Cannot settle at expiration

The settle function of the LongShortPair contract considers settlement conditions when the current time is strictly before or after the expiration timestamp. However, it incorrectly reverts when the current time matches the expiration timestamp.

Consider using an inclusive bound to match the postExpiration modifier.

Update: Fixed in commit f03cdaa50b16d29e8f42f000bf7cd50a042cf616 in PR3527.

[L02] Missing error message in require statement

There is a require statement in the BridgePool contract without an error message.

Consider including specific and informative error messages in all require statements.

Update: Fixed as of commit 67e60faa3a44c842c37211d2e903a983ff192e57 in PR3536.

[L03] Missing docstrings

There are some instances throughout the codebase where the Ethereum Natural Specification is missing or incomplete. Examples include:

Consider thoroughly documenting all functions (and their parameters) that are part of the contracts’ public API.

Update: The highlighted comments were fixed in commit e943e85a7dae60acd17a6d6aa027fbb1017c95ee of PR3533. We did not validate NatSpec completeness in the rest of the code base.

Notes & Additional Information

[N01] Call return value not checked

In the deposit function of the L2 BridgeDepositBox contract there is a low-level call to the l2Token when the l1Token is l1Weth. This low-level call is to the deposit() function, which belongs to the WETH interface. If this l2Token behaves exactly like WETH it should never fail. But in the case the l2Token behaves differently and does fail, there would be no revert since the success flag of this low-level call is never checked.

Consider checking and reacting appropriately to the return values of all low-level calls.

[N02] Lack of indexed parameters in events

Many of the events defined in this codebase have parameters that should be indexed:

Consider indexing event parameters to avoid hindering the task of off-chain services searching and filtering for specific events.

Update: Partially fixed in commit d156b40b2ddb109806336c4d169dbdea91ed1c3e of PR3535. The chainId parameter of WhitelistToken was not updated.

[N03] Implicit casting inconsistency

The LongShortPair contract generally treats timestamps as uint64 values, which are implicitly cast to uint256 values when passed to the Optimistic Oracle. However, the requestTimestamp parameter of the _requestOraclePrice function is prematurely cast to a uint256. This has no functional consequences.

Nevertheless, in the interest of consistency, consider using a uint64 for this parameter and allowing it to be implicitly cast to a uint256 when passed to the Optimistic Oracle.

Update: Fixed in commit 1c3c5c000ef450f5e2da056e41caff468c3fcdcb of PR3528. The timestamp is now explicitly cast.

[N04] Incorrect type

The sendMessage function of the iOptimism_CrossDomainMessenger interface uses a uint256 gas limit while Optimism’s OVM_CrossDomainEnabled uses a uint32 gas limit.

For consistency and predictability, consider updating the iOptimisim_CrossDomainMessenger sendMessage function to use a uint32 gas limit.

Update: Fixed as of commit 381951aad988bbba6b2ef1b136ed5c48df50aa88 in PR3460.

[N05] Lack of validation

All functions in BridgeAdmin that call _relayMessage assume the transaction value matches the l1CallValue parameter, but this is not enforced.

Consider ensuring the correct msg.value is set.

Update: Fixed as of commit f19b8d04c2343051ff2a8145abd41c39bd025063 in PR3537.

[N06] Readability

The _getDepositHash function of the BridgePool contract unrolls the depositData struct to interstice the l1Token as an argument in the composition of keccak256 with the abi encoding. This makes the operation unnecessarily verbose and can lead to bugs when reimplemented on other layers.

Consider simplifying the arguments to simply be the ordered pair depositData and l1Token.

Update: Fixed as of commit 31754be4a818109fa12131f854c3f70d6c72dba7 in PR3538.

[N07] Reentrant function

The requestAndProposePriceFor function of the SkinnyOptimisticOracle contract makes a call to an untrusted msg.sender but is not guarded by a nonReentrant modifier. While, in this instance, this does not seem to be a security concern, this can introduce unexpected behavior.

Consider adding the nonReentrant modifier to all functions which make calls to possibly untrusted contracts.

Update: Fixed in commit b744d24e7579b7afa2c778f4dd680f26117b3990 of PR3539.

[N08] seqNum not logged

The relayMessage function of the Arbitrum_Messenger contract does not emit a relevant event after executing a sensitive action. The relayMessage function calls as a subroutine sentTxToL2NoAliasing which itself returns the uint256 value seqNum, but this return value is not logged in the relayMessage function.

Consider emitting events after sensitive changes take place, to facilitate tracking and notify off-chain clients following the contract’s activity.

Update: Fixed as of commit 30343f33532a6c255dc4cc18c3b497d9b2767a7c in PR3541.

[N09] Typographical errors

The codebase contains the following typos:

Consider correcting these typos to improve code readability.

Update: Fixed as of commit 2dccbe1c2c82fe2a21c179ac06c2d4f0d911a2ca in PR3540.

[N10] Undocumented ERC20 approval requirement

The requestEarlyExpiration and expire functions of the LongShortPair contract each assume that the caller has granted the contract an allowance to pull the proposer reward.

For the sake of predictability, consider documenting this requirement in the function comments.

Update: Fixed in commit da3754f50284480df57b90b80002da06a1ce0d02 in PR3529.

[N11] Unused modifier

In the BridgePool contract, the onlyFromOptimisticOracle modifier is defined but is never used in the codebase and should therefore be removed.

Update: Fixed in commit 7abece6377637e8c4cd3bd07ab9adcfa051d4e94 in PR3542.

Conclusions

2 critical and 1 high severity issues were found. Some changes were proposed to follow best practices and reduce potential attack surface.