Account's stETH balance getting lower on 1 or 2 wei due to rounding down integer math

[TheDZhon]

We had been got already familiar with the 1 wei corner case before a protocol-wide regression test coverage was introduced. Since the integration tests adoption moment we have been discovering the possibility to face the 2 wei corner case. It happens due to the fact that stETH balance calculation depends on two integer divisions (each one has 1 wei "loss" at most).

It should be addressed somehow in the future to prevent dealing with such peculiarities, especially for the upcoming Withdrawals support (Capella hardfork). Thus, I mark the issue with the "next-upgade" label here to catch up during a development stage.

[TheDZhon]

Possible explanation (made on the basis of the original statement by @skozin):

Let's have the following designations:

• x is the number of tokens passed to the transfer function;
• x' is the actual balance increase for the destination addr;
• rate is the total pooled ether / total stETH shares (changes as time goes by, according to the actual APR)
• floor is the usual integer rounding down

Given that:

shares_to_transfer = floor(x / rate); // stETH transfers implemented as shares transfers internally
x' = floor(shares_to_transfer * rate); // balance is calculated by multiplying shares number on rate

and

x' ∈ {[shares_to_transfer * rate - 1, shares_to_transfer * rate] ⋂ ℕ}
shares_to_transfer ∈ {[x / rate - 1; x / rate] ⋂ ℕ}

we may conclude:

x' ∈ {[(x / rate - 1) * rate - 1, (x / rate) * rate] ⋂ ℕ}

or, simplifying finally:

x' ∈ {[x - rate - 1, x] ⋂ ℕ}

thus, transferring x is guaranteed to actually transfer (i.e. balance increase x') an integer number between (x - rate - 1) and x.

Since rate currently is ≈1.08, max error is limited to 2 wei.

While the current APR is 4%, after the Merge it might increase to as high as 8%,
that gives rate ≈ 1.47 after 5 years (max error 2 wei)
after 10 years rate ≈ 2.17 (max error 3 wei)
after 20 years — rate ≈ 4.7 (max error 5 wei).

[sakulstra]

This issue is currently breaking collateralSwap and repayWithCollateral features on the aave protocol which have now been disabled on the decentralized interface for this very reason.
Also it's hard to reason about flashloans as it's hard to repay exact amounts when there's a +-2 wei margin.

Would love to see this resolved 👍

[TheDZhon]

https://www.arbitraryexecution.com/blog/shared-vulnerabilities-between-erc-4626-vaults-and-vault-like-contracts-deep-dive-part-3
https://docs.openzeppelin.com/contracts/4.x/erc4626

[Ekzer]

Hi, we just got the case :

https://etherscan.io/token/0xae7ab96520de3a18e5e111b5eaab095312d7fe84?a=0xf570746d04915f4df7816d18438da11da244d60c
in event logs, we can see the transfer event of 395.97 stETH. However, from smart-contract perspective, there is this 1 wei delta unfortunately