Submitted by hyh, also found by 0xsomeone
An attacker can become the first depositor for a recently created gALCX contract, providing a tiny amount of ALCX tokens by calling stake(1) (raw values here, 1 is 1 wei, 1e18 is 1 ALCX). Then the attacker can directly transfer, for example, 10^6 * 1e18 - 1 of ALCX to the gALCX contract and run bumpExchangeRate(), effectively setting the cost of 1 gALCX to be 10^6 * 1e18 of ALCX. The attacker will still own 100% of the gALCXs ALCX pool being the only depositor.
All subsequent depositors will have their ALCX token investments rounded to 10^6 * 1e18, due to the lack of precision which initial tiny deposit caused, with the remainder divided between all current depositors, i.e. the subsequent depositors lose value to the attacker.
For example, if the second depositor brings in 1.9*10^6 * 1e18 of ALCX, only 1 of new vault to be issued as 1.9*10^6 * 1e18 divided by 10^6 * 1e18 will yield just 1, which means that 2.9*10^6 * 1e18 total ALCX pool will be divided 50/50 between the second depositor and the attacker, as each will have 1 wei of the total 2 wei of vault tokens, i.e. the depositor lost and the attacker gained 0.45 * 10^6 * 1e18 of ALCX tokens.
As there are no penalties to exit with gALCX.unstake(), the attacker can remain staked for an arbitrary time, gathering the share of all new deposits remainder amounts.
Placing severity to be medium as this is principal funds loss scenario for many users (most of depositors), easily executable, but only new gALCX contract instances are vulnerable.
gAmount of gALCX to be minted is determined as a quotient of amount provided and exchangeRate:
gALCX.sol#L93-L94
gALCX.sol#L15
exchangeRate accumulates balance increments relative to total gALCX supply:
gALCX.sol#L69-L76
When gALCX contract is new, the very first stake -> bumpExchangeRate() yields nothing as the balance is empty, i.e. exchangeRate is still 1e18 and gAmount == amount:
gALCX.sol#L85-L93
This way, as there is no minimum amount or special treatment for the first deposit, the very first gAmount can be made 1 wei with stake(1) call.
Then, a combination of direct ALCX transfer and bumpExchangeRate() will make exchangeRate equal to the total amount provided by the attacker, say 10^6 * 1e18 * 1e18, as totalSupply is 1 wei.
When a second depositor enters, the amount of gALCX to be minted is calculated as amount * exchangeRatePrecision / exchangeRate, which is amount / (10^6 * 1e18), which will trunk the amount to the nearest divisor of 10^6 * 1e18, effectively dividing the remainder between the depositor and the attacker.
For example, if the second depositor brings in 1.9*10^6 ALCX, only 1 (1 wei) of gALCX to be issued as 1.9*10^6 * 1e18 * 1e18 / (10^6 * 1e18 * 1e18) = 1.
As the attacker and depositor both have 1 of gALCX, each owns (2.9 / 2)*10^6 * 1e18 = 1.45*10^6 * 1e18, so the attacker effectively stole 0.45*10^6 * 1e18 from the depositor.
Any deposit lower than total attackers stake, 10^6 * 1e18, will be fully stolen from the depositor as 0 gALCX tokens will be issued in this case.
The issue is similar to the TOB-YEARN-003 one of the Trail of Bits audit of Yearn Finance:
https://github.com/yearn/yearn-security/tree/master/audits/20210719_ToB_yearn_vaultsv2
A minimum for deposit value can drastically reduce the economic viability of the attack. I.e. stake() can require each amount to surpass the threshold, and then an attacker would have to provide too big direct investment to capture any meaningful share of the subsequent deposits.
An alternative is to require only the first depositor to freeze big enough initial amount of liquidity. This approach has been used long enough by various projects, for example in Uniswap V2:
Uniswap/UniswapV2Pair.sol#L119-L121
0xfoobar (Alchemix) acknowledged, disagreed with severity and commented:
0xleastwood (judge) commented:
