Submitted by LessDupes, also found by blutorque, ilchovski, 0x73696d616f, zzykxx, kennedy1030, and KupiaSec
https://github.com/code-423n4/2024-04-renzo/blob/519e518f2d8dec9acf6482b84a181e403070d22d/contracts/Delegation/OperatorDelegator.sol#L269
https://github.com/code-423n4/2024-04-renzo/blob/519e518f2d8dec9acf6482b84a181e403070d22d/contracts/Delegation/OperatorDelegator.sol#L501
The OperatorDelegator.completeQueuedWithdrawal() function is used by admins to finalize previously initiated withdraws of shares from EigenLayer.
We note that both this and the OperatorDelegators receive() functions are nonReentrant:
However, the receive() function is normally called by the EigenPod in the call stack originated by the L274 completeQueuedWithdrawal() when receiveAsTokens == true like in this case. This particular instance of reentrancy is not only acceptable but also required to allow ETH redemptions from EigenLayer. However, the nonReentrant modifier prevents it.
All withdrawals that include any amount of ETH will be permanently stuck in EigenLayer and wont be redeemable. Only amounts coming from new deposits can be redeemed and the team will have no way to fill the withdrawal queues. To unblock them, the team will necessarily have to upgrade OperatorDelegator.
To prove the concept, its sufficient to upgrade OperatorDelegator on a mainnet fork and initiate a withdrawal that has ETH among the withdrawn strategies.
While it would be too bulky to provide a coded PoC, you can find in this GH Gist the Foundry traces of such failed call on a mainnet fork.
Foundry
Consider removing nonReentrant from OperatorDelegators receive, or applying the modifier only in case msg.sender != eigenPod.
Reentrancy
jatinj615 (Renzo) confirmed via duplicate Issue #571
Renzo mitigated
Status: Mitigation confirmed. Full details in reports from 0xCiphky, grearlake, Fassi_Security, Bauchibred, and LessDupes.
