Source: https://github.com/sherlock-audit/2022-10-astaria-judging/issues/272
Jeiwan
The implied value of a public vault can be impaired, liquidity providers can lose funds
Borrowers can partially repay their liens, which is handled by the _payment function (LienToken.sol#L594).
When repaying a part of a lien, lien.amount is updated to include currently accrued debt (LienToken.sol#L605-L617):
Lien storage lien = lienData[lienId];
lien.amount = _getOwed(lien); // @audit current debt, including accrued interest; saved to storage!Notice that lien.amount is updated in storage, and lien.last wasn't updated.
Then, lien's slope is subtracted from vault's slope accumulator to be re-calculated after the repayment (LienToken.sol#L620-L630):
if (isPublicVault) {
// @audit calculates and subtracts lien's slope from vault's slope
IPublicVault(lienOwner).beforePayment(lienId, paymentAmount);
}
if (lien.amount > paymentAmount) {
lien.amount -= paymentAmount;
// @audit lien.last is updated only after payment amount subtraction
lien.last = block.timestamp.safeCastTo32();
// slope does not need to be updated if paying off the rest, since we neutralize slope in beforePayment()
if (isPublicVault) {
// @audit re-calculates and re-applies lien's slope after the repayment
IPublicVault(lienOwner).afterPayment(lienId);
}
}In the beforePayment function, LIEN_TOKEN().calculateSlope(lienId) is called to calculate lien's current slope (PublicVault.sol#L433-L442):
function beforePayment(uint256 lienId, uint256 amount) public onlyLienToken {
_handleStrategistInterestReward(lienId, amount);
uint256 lienSlope = LIEN_TOKEN().calculateSlope(lienId);
if (lienSlope > slope) {
slope = 0;
} else {
slope -= lienSlope;
}
last = block.timestamp;
}The calculateSlope function reads a lien from storage and calls _getOwed again (LienToken.sol#L440-L445):
function calculateSlope(uint256 lienId) public view returns (uint256) {
// @audit lien.amount includes interest accrued so far
Lien memory lien = lienData[lienId];
uint256 end = (lien.start + lien.duration);
uint256 owedAtEnd = _getOwed(lien, end);
// @audit lien.last wasn't updated in `_payment`, it's an older timestamp
return (owedAtEnd - lien.amount).mulDivDown(1, end - lien.last);
}This is where double counting of accrued interest happens. Recall that lien's amount already includes the interest that was accrued by this moment (in the _payment function). Now, interest is calculated again and is applied to the amount that already includes (a portion) it (LienToken.sol#L544-L550):
function _getOwed(Lien memory lien, uint256 timestamp)
internal
view
returns (uint256)
{
// @audit lien.amount already includes interest accrued so far
return lien.amount + _getInterest(lien, timestamp);
}function _getInterest(Lien memory lien, uint256 timestamp)
internal
view
returns (uint256)
{
if (!lien.active) {
return uint256(0);
}
uint256 delta_t;
if (block.timestamp >= lien.start + lien.duration) {
delta_t = uint256(lien.start + lien.duration - lien.last);
} else {
// @audit lien.last wasn't updated in `_payment`, so the `delta_t` is bigger here
delta_t = uint256(timestamp.safeCastTo32() - lien.last);
}
return
// @audit rate applied to a longer delta_t and multiplied by a bigger amount than expected
delta_t.mulDivDown(lien.rate, 1).mulDivDown(
lien.amount,
INTEREST_DENOMINATOR
);
}Double counting of interest will result in a wrong lien slope, which will affect the vault's slope accumulator. This will result in an invalid implied value of a vault (PublicVault.sol#L406-L413):
beforePayment), and vault's implied value will also be smaller. Liquidity providers will lose money because they won't be able to redeem the whole liquidity (vault's implied value, totalAssets, is used in the conversion of LP shares, ERC4626-Cloned.sol#L392-L412)See Vulnerability Detail
Manual Review
In the _payment function, consider updating lien.amount after the beforePayment call:
--- a/src/LienToken.sol
+++ b/src/LienToken.sol
@@ -614,12 +614,13 @@ contract LienToken is ERC721, ILienToken, Auth, TransferAgent {
type(IPublicVault).interfaceId
);
- lien.amount = _getOwed(lien);
-
address payee = getPayee(lienId);
if (isPublicVault) {
IPublicVault(lienOwner).beforePayment(lienId, paymentAmount);
}
+
+ lien.amount = _getOwed(lien);
+
if (lien.amount > paymentAmount) {
lien.amount -= paymentAmount;
lien.last = block.timestamp.safeCastTo32();In this case, lien's slope calculation won't be affected in the beforePayment call and the correct slope will be removed from the slope accumulator.
LIEN_TOKEN.ownerOf(i) should be LIEN_TOKEN.ownerOf(liensRemaining[i])Source: https://github.com/sherlock-audit/2022-10-astaria-judging/issues/259
0xRajeev, __141345__
In endAuction(), the check for public vault owner is referred to the wrong lien token id. And the actual vault lien amount is not properly recorded.
The lien token id should be queried is liensRemaining[i] instead of i.
YIntercept will not be correctly recorded. The accounting for LienToken amounts will be wrong. Hence the totalAssets on book will be wrong, eventually the contract and users could lose fund due to the wrong accounting.
Manual Review
Change LIEN_TOKEN.ownerOf(i) to LIEN_TOKEN.ownerOf(liensRemaining[i]).
Source: https://github.com/sherlock-audit/2022-10-astaria-judging/issues/245
bin2chen
LienToken#buyoutLien() did not reduce vault#liensOpenForEpoch when vault#processEpoch()will check vault#liensOpenForEpoch[currentEpoch] == uint256(0) so processEpoch() will fail
when create LienToken , vault#liensOpenForEpoch[currentEpoch] will ++ when repay or liquidate , vault#liensOpenForEpoch[currentEpoch] will -- and LienToken#buyoutLien() will transfer from vault to to other receiver,so liensOpenForEpoch need reduce
function buyoutLien(ILienToken.LienActionBuyout calldata params) external {
....
/**** tranfer but not liensOpenForEpoch-- *****/
_transfer(ownerOf(lienId), address(params.receiver), lienId);
}processEpoch() maybe fail
https://github.com/sherlock-audit/2022-10-astaria/blob/main/src/LienToken.sol#L121
Manual Review
function buyoutLien(ILienToken.LienActionBuyout calldata params) external {
....
+ //do decreaseEpochLienCount()
+ address lienOwner = ownerOf(lienId);
+ bool isPublicVault = IPublicVault(lienOwner).supportsInterface(
+ type(IPublicVault).interfaceId
+ );
+ if (isPublicVault && !AUCTION_HOUSE.auctionExists(collateralId)) {
+ IPublicVault(lienOwner).decreaseEpochLienCount(
+ IPublicVault(lienOwner).getLienEpoch(lienData[lienId].start + lienData[lienId].duration)
+ );
+ }
lienData[lienId].last = block.timestamp.safeCastTo32();
lienData[lienId].start = block.timestamp.safeCastTo32();
lienData[lienId].rate = ld.rate.safeCastTo240();
lienData[lienId].duration = ld.duration.safeCastTo32();
_transfer(ownerOf(lienId), address(params.receiver), lienId);
}IAmTurnipBoy
Escalate for 1 USDC
Duplicate of #194, two sides of the same coin. One points out it that buyout doesn't decrement correctly on one side and the other points out it doesn't increment correctly on the other side
sherlock-admin
Escalate for 1 USDC
Duplicate of #194, two sides of the same coin. One points out it that buyout doesn't decrement correctly on one side and the other points out it doesn't increment correctly on the other side
You've created a valid escalation for 1 USDC!
To remove the escalation from consideration: Delete your comment. To change the amount you've staked on this escalation: Edit your comment (do not create a new comment).
You may delete or edit your escalation comment anytime before the 48-hour escalation window closes. After that, the escalation becomes final.
Evert0x
Escalation rejected.
The argument is not giving us full conviction this should be tagged as a duplicate.
sherlock-admin
Escalation rejected.
The argument is not giving us full conviction this should be tagged as a duplicate.
This issue's escalations have been rejected!
Watsons who escalated this issue will have their escalation amount deducted from their next payout.
Source: https://github.com/sherlock-audit/2022-10-astaria-judging/issues/233
TurnipBoy, 0x0, obront, tives, yixxas, ctf_sec, zzykxx
_deleteLienPosition is a public function that doesn't check the caller. This allows anyone to call it an remove whatever lien they wish from whatever collateral they wish
function _deleteLienPosition(uint256 collateralId, uint256 position) public {
uint256[] storage stack = liens[collateralId];
require(position < stack.length, "index out of bounds");
emit RemoveLien(
stack[position],
lienData[stack[position]].collateralId,
lienData[stack[position]].position
);
for (uint256 i = position; i < stack.length - 1; i++) {
stack[i] = stack[i + 1];
}
stack.pop();
}_deleteLienPosition is a public function and doesn't validate that it's being called by any permissioned account. The result is that anyone can call it to delete any lien that they want. It wouldn't remove the lien data but it would remove it from the array associated with collateralId, which would allow it to pass the CollateralToken.sol#releaseCheck and the underlying to be withdrawn by the user.
All liens can be deleted completely rugging lenders
https://github.com/sherlock-audit/2022-10-astaria/blob/main/src/LienToken.sol#L651-L664
Manual Review
Change _deleteLienPosition to internal rather than public.
IAmTurnipBoy
Escalate for 1 USDC
Deleting leans allows borrower to steal funds because they never have to repay the funds the borrowed. Should be high risk
sherlock-admin
Escalate for 1 USDC
Deleting leans allows borrower to steal funds because they never have to repay the funds the borrowed. Should be high risk
You've created a valid escalation for 1 USDC!
To remove the escalation from consideration: Delete your comment. To change the amount you've staked on this escalation: Edit your comment (do not create a new comment).
You may delete or edit your escalation comment anytime before the 48-hour escalation window closes. After that, the escalation becomes final.
Evert0x
Escalation accepted
sherlock-admin
Escalation accepted
This issue's escalations have been accepted!
Contestants' payouts and scores will be updated according to the changes made on this issue.
commitToLiens always revertsSource: https://github.com/sherlock-audit/2022-10-astaria-judging/issues/204
bin2chen, HonorLt, rvierdiiev, Jeiwan, 0xRajeev
The function commitToLiens() always reverts at the call to _returnCollateral() which prevents borrowers from depositing collateral and requesting loans in the protocol.
The collateral token with collateralId is already minted directly to the caller (i.e. borrower) in commitToLiens() at the call to _transferAndDepositAsset() function. That's because while executing _transferAndDepositAsset the NFT is transferred to COLLATERAL_TOKEN whose onERC721Received mints the token with collateralId to borrower (from address) and not the operator_ (i.e. AstariaRouter) because operator_ != from_.
However, the call to _returnCollateral() in commitToLiens() incorrectly assumes that this has been minted to the operator and attempts to transfer it to the borrower which will revert because the collateralId is not owned by AstariaRouter as it has already been transferred/minted to the borrower.
The function commitToLiens() always reverts, preventing borrowers from depositing collateral and requesting loans in the protocol, thereby failing to bootstrap its core NFT lending functionality.
Manual Review
Remove the call to _returnCollateral() in commitToLiens().
secureum
Escalate for 2 USDC.
Given the impact of failing to bootstrap core protocol functionality as described above, we still think this is of high-severity (not Medium as judged) unlike a DoS that affects only a minority of protocol flows. Also, this is not a dup of #195.
cc @berndartmueller @lucyoa
sherlock-admin
Escalate for 2 USDC.
Given the impact of failing to bootstrap core protocol functionality as described above, we still think this is of high-severity (not Medium as judged) unlike a DoS that affects only a minority of protocol flows. Also, this is not a dup of #195.
cc @berndartmueller @lucyoa
You've created a valid escalation for 2 USDC!
To remove the escalation from consideration: Delete your comment. To change the amount you've staked on this escalation: Edit your comment (do not create a new comment).
You may delete or edit your escalation comment anytime before the 48-hour escalation window closes. After that, the escalation becomes final.
Evert0x
Escalation accepted
sherlock-admin
Escalation accepted
This issue's escalations have been accepted!
Contestants' payouts and scores will be updated according to the changes made on this issue.
Source: https://github.com/sherlock-audit/2022-10-astaria-judging/issues/202
bin2chen, chainNue, csanuragjain, TurnipBoy, peanuts, neila, Prefix, Jeiwan, hansfriese, 0xRajeev, minhquanym
If the collateral token owner cancels the active auction and repays outstanding debt, the current highest bidder will not be refunded and loses their funds.
The AuctionHouse.createBid() function refunds the previous bidder if there is one. The same logic would also be necessary in the AuctionHouse.cancelAuction() function but is missing.
If the collateral token owner cancels the active auction and repays outstanding debt (reservePrice), the current highest bidder will not be refunded and will therefore lose their funds which can also be exploited by a malicious borrower.
Potential exploit scenario: A malicious borrower can let the loan expire without repayment, trigger an auction, let bids below reserve price, and (hope to) front-run any bid >= reserve price to cancel the auction which effectively lets the highest bidder pay out (most of) the liens instead of the borrower.
Manual Review
Add the refund logic (via _handleOutGoingPayment() to the current bidder) in the cancel auction flow similar to the create bid auction flow.
IAmTurnipBoy
Escalate for 1 USDC
Huge loss of funds for bidder. High risk
sherlock-admin
Escalate for 1 USDC
Huge loss of funds for bidder. High risk
You've created a valid escalation for 1 USDC!
To remove the escalation from consideration: Delete your comment. To change the amount you've staked on this escalation: Edit your comment (do not create a new comment).
You may delete or edit your escalation comment anytime before the 48-hour escalation window closes. After that, the escalation becomes final.
Evert0x
Escalation accepted.
sherlock-admin
Escalation accepted.
This issue's escalations have been accepted!
Contestants' payouts and scores will be updated according to the changes made on this issue.
Source: https://github.com/sherlock-audit/2022-10-astaria-judging/issues/199
bin2chen, TurnipBoy, Jeiwan, 0xRajeev, 0x4141
Canceling an auction with 0 bids only partially pays back the outstanding debt without removing outstanding liens resulting in the collateralized NFT being locked forever.
Given this scenario of an auction with 0 bids and, for example, a reservePrice of 10 ETH and initiatorFee set to 10 (= 10%), with the following sequence executed:
AuctionHouse.cancelAuction(), _handleIncomingPayment() accepts incoming transfer of 10 ETH reserve price_handleIncomingPayment(), the initiatorFee of 10% is calculated, deducted from the 10 ETH to transfer 1 ETH to the initiatorHowever, the outstanding debt was initially 10 ETH (= reservePrice). After deducting the initiator fee, only 9 ETH remains. This means that the debt is not fully paid back. But the auction will successfully be cancelled but with outstanding debt remaining. There is also no explicit removal of liens (as there is in endAuction).
A borrower canceling an auction with the expected reserve price will leave behind existing unpaid liens that will make the releaseCheck modifier applied to releaseToAddress() prevent the borrower from releasing their NFT collateral successfully even after paying the reserve price during auction cancellation. The borrower's collateralized NFT is locked forever despite paying the outstanding lien amount during the auction cancellation.
Manual Review
The auction cancellation amount required should be reserve price + liquidation fee. On payment, remaining liens should be removed.
Source: https://github.com/sherlock-audit/2022-10-astaria-judging/issues/198
0xRajeev
The initiator fee on the current bid has already been accounted during that bid but is expected to be paid again on cancellation by the borrower.
Consider an active auction with a reserve price of 10 ETH and a current bid of 9 ETH. If the auction gets canceled, the transferAmount will be 10 ETH. Once the cancellation logic adds repayment to current bidder (see other finding reported on this issue), the initiator fees for cancellation should only be calculated on the delta payment of 1 ETH because the fees for the earlier 9 ETH bid has already been paid to the initiator. However, this is not accounted and requires the borrower to pay the initiator fees on the entire reserve price leading to overpayment towards the initiator and loss of funds for the borrower.
Canceling an auction will require paying (a lot) more initiator fees than needed resulting in a loss of funds for the borrower.
Manual Review
The cancellation amount required should be the reserve price + liquidation fee, where the fee is calculated on (reserve price - current bid) and not the reserve price.
yIntercept updateSource: https://github.com/sherlock-audit/2022-10-astaria-judging/issues/197
0xRajeev, zzykxx
The deduction of yIntercept during payments is missing in beforePayment() which can lead to vault insolvency.
yIntercept is declared as "sum of all LienToken amounts" and documented elsewhere as "yIntercept (virtual assets) of a PublicVault". It is used to calculate the total assets of a public vault as: slope.mulDivDown(delta_t, 1) + yIntercept.
It is expected to be updated on deposits, payments, withdrawals, liquidations. However, the deduction of yIntercept during payments is missing in beforePayment(). As noted in the function's Natspec:
/**
* @notice Hook to update the slope and yIntercept of the PublicVault on payment.
* The rate for the LienToken is subtracted from the total slope of the PublicVault, and recalculated in afterPayment().
* @param lienId The ID of the lien.
* @param amount The amount paid off to deduct from the yIntercept of the PublicVault.
*/the amount of payment should be deducted from yIntercept but is missing.
PoC: https://gist.github.com/berndartmueller/477cc1026d3fe3e226795a34bb8a903a
This missing update will inflate the inferred value of the public vault corresponding to its actual value leading to eventual insolvency because of resulting protocol miscalculations.
Manual Review
Update yIntercept in beforePayment() by the amount value.
androolloyd
tagging @SantiagoGregory but i believe this is a documentation error, will address
secureum
Escalate for 2 USDC.
Given the vault insolvency impact as described and demonstrated by the PoC, we still think this is a high-severity impact (not Medium as judged). The other dup #92 also reported this as a High.
cc @berndartmueller @lucyoa
sherlock-admin
Escalate for 2 USDC.
Given the vault insolvency impact as described and demonstrated by the PoC, we still think this is a high-severity impact (not Medium as judged). The other dup #92 also reported this as a High.
cc @berndartmueller @lucyoa
You've created a valid escalation for 2 USDC!
To remove the escalation from consideration: Delete your comment. To change the amount you've staked on this escalation: Edit your comment (do not create a new comment).
You may delete or edit your escalation comment anytime before the 48-hour escalation window closes. After that, the escalation becomes final.
Evert0x
Escalation accepted
sherlock-admin
Escalation accepted
This issue's escalations have been accepted!
Contestants' payouts and scores will be updated according to the changes made on this issue.
LienToken.buyoutLien will always revertSource: https://github.com/sherlock-audit/2022-10-astaria-judging/issues/196
8olidity, rvierdiiev, supernova, neila, yixxas, ctf_sec, 0xRajeev, cccz, zzykxx
buyoutLien() will always revert, preventing the borrower from refinancing.
buyoutFeeDenominator is 0 without a setter which will cause getBuyoutFee() to revert in the buyoutLien() flow.
Refinancing is a crucial feature of the protocol to allow a borrower to refinance their loan if a certain minimum improvement of interest rate or duration is offered. The reverting buyoutLien() flow will prevent the borrower from refinancing and effectively lead to loss of their funds due to lock-in into currently held loans when better terms are available.
Manual Review
Initialize the buyout fee numerator and denominator in AstariaRouter and add their setters to file().
secureum
Escalate for 2 USDC.
Given the potential impact to different flows/contexts, we still think this is a high-severity impact (not Medium as judged). A majority of the dups (while some are dups of a different but related issue) also reported this as a High.
cc @berndartmueller @lucyoa
sherlock-admin
Escalate for 2 USDC.
Given the potential impact to different flows/contexts, we still think this is a high-severity impact (not Medium as judged). A majority of the dups (while some are dups of a different but related issue) also reported this as a High.
cc @berndartmueller @lucyoa
You've created a valid escalation for 2 USDC!
To remove the escalation from consideration: Delete your comment. To change the amount you've staked on this escalation: Edit your comment (do not create a new comment).
You may delete or edit your escalation comment anytime before the 48-hour escalation window closes. After that, the escalation becomes final.
Evert0x
Escalation accepted
sherlock-admin
Escalation accepted
This issue's escalations have been accepted!
Contestants' payouts and scores will be updated according to the changes made on this issue.
Source: https://github.com/sherlock-audit/2022-10-astaria-judging/issues/194
0xRajeev
The lien count per epoch is not updated, causing payments to the lien and liquidation of the lien to revert.
The PublicVault contract keeps track of open liens per epoch to prevent starting a new epoch with open liens. The lien count for a given epoch is decreased whenever a lien is fully paid back (either through a regular payment or a liquidation payment). However, if a lien is bought out, the lien start will be set to the current block.timestamp and the duration to the newly provided duration.
If the borrower now wants to make a payment to this lien, the LienToken._payment function will evaluate the lien's current epoch and will use a different epoch as when the lien was initially created. The attempt to then call IPublicVault(lienOwner).decreaseEpochLienCount will fail, as the lien count for the new epoch has not been increased yet. The same will happen for liquidations.
After a lien buyout, payments to the lien and liquidating the lien will revert, which will ultimately lock the collateralized NFT.
This will certainly prevent the borrower from making payments towards that future-epoch lien in the current epoch because decreaseEpochLienCount() will revert. However, even after the epoch progresses to the next one via processEpoch(), the liensOpenForEpoch for the new epoch still does not account for the previously bought out lien aligned to this new epoch because liensOpenForEpoch is only updated in two places:
_increaseOpenLiens() which is not called by anyone _afterCommitToLien() <- commitToLien() <-- <-- commitToLiens() which happens only for new lien commitmentsThis will prevent the borrower from making payments towards the previously bought lien that aligns to the current epoch, which will force a liquidation on exceeding lien duration. Depending on when the liquidation can be triggered, if this condition is satisfied PublicVault(owner).timeToEpochEnd() <= COLLATERAL_TOKEN.auctionWindow() then decreaseEpochLienCount() will revert to prevent auctioning and lock the borrower's collateral in the protocol.
Manual Review
liensOpenForEpoch should be incremented when a lien is bought with a duration spilling into an epoch higher than the current one.
Source: https://github.com/sherlock-audit/2022-10-astaria-judging/issues/193
0xRajeev, rvierdiiev, obront
A purchaser who buys out an existing lien via buyoutLien() will not receive future payments made to that lien holder if the seller had changed the lien payee via setPayee() and if they do not change it themselves after buying.
buyoutLien() does not reset lienData[lienId].payee to either 0 or to the new owner. While the ownership is transferred, the payments made in _payment() get their payee via getPayee() which returns the owner only if the lienData[lienId].payee is set to the zero address but returns lienData[lienId].payee otherwise. This will still have the value set by the previous owner who will continue to receive the payments.
The purchaser who buys out an existing lien via buyoutLien() will not receive future payments if the previous owner had set the payee but they do not change it via setPayee() themselves after buying. Given that setPayee() is considered optional (nothing specified in spec/docs/comments) and this reset is not part of the default flow, this will lead to a loss of purchaser's anticipated payments until they realize and reset the lien payee.
Exploit Scenario: A malicious lien seller can use setPayee() to set the payee to their address of choice and continue to receive payments, even after selling, if the buyer does not realize that they have to change the payee address to themselves after buying.
Manual Review
buyoutLien() should reset lienData[lienId].payee to either the zero address or to the new owner.
Source: https://github.com/sherlock-audit/2022-10-astaria-judging/issues/190
0xRajeev, obront, zzykxx, Jeiwan
A payment made towards multiple liens is entirely consumed for the first one causing the borrower to lose funds to the payee.
A borrower can make a bulk payment against multiple liens for a collateral hoping to pay more than one at a time using makePayment (uint256 collateralId, uint256 paymentAmount) where the underlying _makePayment() loops over the open liens attempting to pay off more than one depending on the totalCapitalAvailable provided.
However, the entire totalCapitalAvailable is provided via paymentAmount in the call to _payment() in the first iteration which transfers that completely to the payee in its logic even if it exceeds that lien.amount. That total amount is returned as capitalSpent which makes the paymentAmount for next iteration equal to 0.
Only the first lien is paid off and the entire payment is sent to its payee. The remaining liens remain unpaid. The payment maker (i.e. borrower ) loses funds to the payee.
Manual Review
Add paymentAmount -= lien.amount in the else block of _payment().
IAmTurnipBoy
Escalate for 1 USDC
Clear loss of funds. Should be high
sherlock-admin
Escalate for 1 USDC
Clear loss of funds. Should be high
You've created a valid escalation for 1 USDC!
To remove the escalation from consideration: Delete your comment. To change the amount you've staked on this escalation: Edit your comment (do not create a new comment).
You may delete or edit your escalation comment anytime before the 48-hour escalation window closes. After that, the escalation becomes final.
Evert0x
Escalation accepted
sherlock-admin
Escalation accepted
This issue's escalations have been accepted!
Contestants' payouts and scores will be updated according to the changes made on this issue.
LiquidationAccountant.claim() can be called by anyone causing vault insolvencySource: https://github.com/sherlock-audit/2022-10-astaria-judging/issues/188
0xRajeev, TurnipBoy, obront, rvierdiiev
LiquidationAccountant.claim() can be called by anyone to reduce the implied value of a public vault.
LiquidationAccountant.claim() is called by the PublicVault as part of the processEpoch() flow. But it has no access control and can be called by anyone and any number of times. If called after finalAuctionEnd, one will be able to trigger decreaseYIntercept() on the vault even if they cannot affect fund transfer to withdrawing liquidity providers and the PublicVault.
This allows anyone to manipulate the yIntercept of a public vault by triggering the claim() flow after liquidations resulting in vault insolvency.
Manual Review
Allow only vault to call claim() by requiring authorizations.
SantiagoGregory
Our updated LiquidationAccountant implementation (now moved to WithdrawProxy) tracks a hasClaimed bool to make sure claim() is only called once (we also now block claim() from being called until after finalAuctionEnd).
IAmTurnipBoy
Escalate for 1 USDC
Abuse can cause vault to implode and cause loss of funds to all depositors. Should be high
sherlock-admin
Escalate for 1 USDC
Abuse can cause vault to implode and cause loss of funds to all depositors. Should be high
You've created a valid escalation for 1 USDC!
To remove the escalation from consideration: Delete your comment. To change the amount you've staked on this escalation: Edit your comment (do not create a new comment).
You may delete or edit your escalation comment anytime before the 48-hour escalation window closes. After that, the escalation becomes final.
Evert0x
Escalation accepted.
sherlock-admin
Escalation accepted.
This issue's escalations have been accepted!
Contestants' payouts and scores will be updated according to the changes made on this issue.
Source: https://github.com/sherlock-audit/2022-10-astaria-judging/issues/187
0xRajeev
A malicious lien owner can exploit a reentrancy in auctions to have their liens removed without making their payments and thus stealing LP funds.
A malicious lien owner can exploit a reentrancy from the callback to decreaseYIntercept() in endAuction() to call cancelAuction() and then take a new loan whose lien token is removed when control returns back to endAuction().
PoC: https://gist.github.com/lucyoa/901a7713fded73293b5e4f9452344c5a
Exploit scenario sequence:
_buyoutLien by paying 10ETH + feeendAuction()decreaseYIntercept():
reservePrice+initiatorFee (this would also go to borrower if setPayee is set before the auction as the owner of that Lien token)AuctionHouse.endAuction, borrower's new loan's Lien token is deletedCollateralToken.endAuction releases borrower's NFT to borrowerManual Review
VaultImplementation._validateCommitment may prevent liens that satisfy their terms of maxPotentialDebtSource: https://github.com/sherlock-audit/2022-10-astaria-judging/issues/182
obront, tives, rvierdiiev, Jeiwan, hansfriese, 0xRajeev, zzykxx
The calculation of potentialDebt in VaultImplementation._validateCommitment() is incorrect and will cause a DoS to legitimate borrowers.
The calculation of potentialDebt in VaultImplementation._validateCommitment() is incorrect because it computes uint256 potentialDebt = seniorDebt * (ld.rate + 1) * ld.duration; which incorrectly adds a factor of ld.duration to seniorDebt thus making the potential debt much higher by that factor than it will be. The use of INTEREST_DENOMINATOR and implied lien rate is also missing here.
Liens that would have otherwise satisfied the constraint of potentialDebt <= ld.maxPotentialDebt will fail because of this miscalculation and will cause a DoS to legitimate borrowers and likely all of them.
Manual Review
Change the calculation to uint256 potentialDebt = seniorDebt * (ld.rate * ld.duration + 1).mulDivDown(1, INTEREST_DENOMINATOR);. This should also consider the implied rate of all the liens against the collateral instead of only this lien.
secureum
Escalate for 2 USDC.
Given the potential impact to different flows/contexts, we still think this is a high-severity impact (not Medium as judged). A majority of the dups (4 of 6) also reported this as a High.
cc @berndartmueller @lucyoa
sherlock-admin
Escalate for 2 USDC.
Given the potential impact to different flows/contexts, we still think this is a high-severity impact (not Medium as judged). A majority of the dups (4 of 6) also reported this as a High.
cc @berndartmueller @lucyoa
You've created a valid escalation for 2 USDC!
To remove the escalation from consideration: Delete your comment. To change the amount you've staked on this escalation: Edit your comment (do not create a new comment).
You may delete or edit your escalation comment anytime before the 48-hour escalation window closes. After that, the escalation becomes final.
Evert0x
Escalation accepted.
sherlock-admin
Escalation accepted.
This issue's escalations have been accepted!
Contestants' payouts and scores will be updated according to the changes made on this issue.
Source: https://github.com/sherlock-audit/2022-10-astaria-judging/issues/181
0xRajeev
A malicious lien buyer can DoS to cause fund loss/lock of other lien holders & borrower collateral.
Anyone can call buyoutLien to purchase a lien token delivered to an arbitrary receiver. A malicious entity (even the borrower themselves) can purchase a small lien amount with its token delivered to their contract, which can implement supportsInterface() with arbitrary code, e.g. revert, that gets executed by the protocol in multiple places giving the attacker control at those points.
endAuction() to prevent the release of collateral to the winner.liquidate() to prevent the liquidation from even starting._payment() to prevent any lien payments from succeeding in calls to makePayment(uint256 collateralId, uint256 paymentAmount)Thus, a malicious lien buyer can DoS to cause fund loss/lock of other lien holders & loss/lock of borrower collateral.
Manual Review
Source: https://github.com/sherlock-audit/2022-10-astaria-judging/issues/180
0xRajeev
Lien buyout with new terms does not update the slope of public vaults leading to reverts and potential insolvency of vaults.
In the VaultImplementation.buyoutLien function, a lien is bought out with new terms. Terms of a lien (last, start, rate, duration) will have changed after a buyout. This changes the slope calculation, however, after the buyout, the slope for the public vault is not updated.
There are multiple parts of the code affected by this.
1) When liquidating, the vault slope is adjusted by the liens slope (see https://github.com/sherlock-audit/2022-10-astaria/blob/main/src/AstariaRouter.sol#L385-L387). In case of a single lien, the PublicVault.updateVaultAfterLiquidation function can revert if the lien terms have changed previously due to a buyout (https://github.com/sherlock-audit/2022-10-astaria/blob/main/src/PublicVault.sol#L532). Hence, liquidations with a public vault involved, will revert.
2) The PublicVault contract calculates the implied value of a vault (ie. totalAssets) with the use of the slope value (see https://github.com/sherlock-audit/2022-10-astaria/blob/main/src/PublicVault.sol#L412). As the slope value can be outdated, this leads to undervalue or overvalue of a public vault and thus vault share calculations will be incorrect.
Manual Review
Calculate the slope of the lien before the buyout in the VautImplementation.buyoutLien function, subtract the calculated slope value from PublicVault.slope, update lien terms, recalculate the slope and add the new updated slope value to PublicVault.slope.
Source: https://github.com/sherlock-audit/2022-10-astaria-judging/issues/179
0xRajeev, Prefix, sorrynotsorry
Auction bids will revert for a bid amount that does not fully pay back an expired lien.
The value of lien.last is updated to block.timestamp in several parts of the code, even for expired liens. Payments made as part of liquidations will set lien.last to block.timestamp > lien.start + lien.duration causing a revert in the flow shown below at step 3 when it is subtracted from end:
IPublicVault(lienOwner).afterPayment(lienId);slope += LIEN_TOKEN().calculateSlope(lienId);return (owedAtEnd - lien.amount).mulDivDown(1, end - lien.last);Auction bids will revert for a bid amount that does not fully pay back an expired lien. If a lien is only partially paid back, it will reach the if branch in LienToken._payment, which sets lien.last to the current timestamp (which itself is > lien.start + lien.duration, hence the liquidation) and then revert.
This affects the economic efficiency of auctions because it DoS's partial bids and therefore results in loss of funds to LPs.
Manual Review
Revisit the logic that updates lien.last in the protocol to ensure no reverts in expected flows.
Source: https://github.com/sherlock-audit/2022-10-astaria-judging/issues/176
0xRajeev, bin2chen, TurnipBoy, 0x4141
Anyone can deposit and mint Withdrawal proxy shares by directly interacting with the base ERC4626Cloned contract's functions, allowing them to capture distributed yield from borrower interests.
The WithdrawProxy contract extends the ERC4626Cloned vault contract implementation. The ERC4626Cloned contract has the functionality to deposit and mint vault shares. Usually, withdrawal proxy shares are only distributed via the WithdrawProxy.mint function, which is only called by the PublicVault.redeemFutureEpochfunction. Anyone can deposit WETH into a deployed Withdraw proxy to receive shares, wait until assets (WETH) are deposited via the PublicVault.transferWithdrawReserve or LiquidationAccountant.claim function and then redeem their shares for WETH assets.
By depositing/minting directly to the Withdraw proxy, one can get interest yield on-demand without being an LP and having capital locked for epoch(s). This may potentially be timed in a way to deposit/mint only when we know that interest yields are being paid by a borrower who is not defaulting on their loan. The returns are diluted for the LPs at the expense of someone who directly interacts with the underlying proxy.
Manual Review
Overwrite the ERC4626Cloned.afterDeposit function and revert to prevent public deposits and mints.
IAmTurnipBoy
Escalate for 1 USDC
This leads to material loss of funds. Definitely high risk
sherlock-admin
Escalate for 1 USDC
This leads to material loss of funds. Definitely high risk
You've created a valid escalation for 1 USDC!
To remove the escalation from consideration: Delete your comment. To change the amount you've staked on this escalation: Edit your comment (do not create a new comment).
You may delete or edit your escalation comment anytime before the 48-hour escalation window closes. After that, the escalation becomes final.
Evert0x
Escalation accepted
sherlock-admin
Escalation accepted
This issue's escalations have been accepted!
Contestants' payouts and scores will be updated according to the changes made on this issue.
Source: https://github.com/sherlock-audit/2022-10-astaria-judging/issues/175
0xRajeev
The currentEpoch can be progressed while having an ongoing auction which will completely mess up the liquidation logic to potentially cause LP fund loss and collateral lockup.
The LiquidationAccountant.claim() function is only callable if finalAuctionEnd is set to 0 or the block.timestamp is greater than finalAuctionEnd (i.e. auction has ended). Furthermore, PublicVault.processEpoch should only be callable if there is no ongoing auction if a liquidation accountant is deployed in the current epoch.
finalAuctionEnd is set within the LiquidationAccountant.handleNewLiquidation function, which is called from the AstariaRouter.liquidate function. However, instead of providing a timestamp of the auction end, a value of 2 days + 1 days is given because COLLATERAL_TOKEN.auctionWindow() returns 2 days.
This does not achieve the intended constraint on checking for the end of the auction because it uses a fixed duration instead of a timestamp.
Epochs can be progressed during ongoing auctions to completely mess up the liquidation logic to potentially cause LP fund loss and collateral lockup.
Manual Review
Revisit the logic to use an appropriate timestamp instead of a fixed duration.
Source: https://github.com/sherlock-audit/2022-10-astaria-judging/issues/173
0xRajeev
Public vault total asset calculation is incorrect until the first payment, leading to depositors receiving fewer vault shares than expected.
As long as PublicVault.last is set to 0, the PublicVault.totalAssets function returns the actual ERC-20 token balance (WETH) of the public vault. Due to borrowing, this balance is reduced by the borrowed amount. Therefore, as there is no payment, this leads to depositors receiving fewer vault shares than expected.
PoC: https://gist.github.com/berndartmueller/8a71ff76c7eb8207e1f01a154a873b2c
Public vault depositors will receive fewer vault shares until the first payment.
Manual Review
Revisit the logic behind updating last when yIntercept and/or slope are updated.
Source: https://github.com/sherlock-audit/2022-10-astaria-judging/issues/172
0xRajeev
The liens array and its indexes are changed whenever a lien is fully paid back. Therefore, referencing liens by their position in the array is broken because indices change. Payments and liquidations of multiple liens within a loop will revert and can be exploited by sandwiching liens to force a payment to the attacker's lien.
Whenever a lien position is fully paid back, the lien is deleted from the liens array. This means that the array is shortened and the indices of the remaining liens are changed. This is problematic when paying back liens in a loop because the indices of the remaining liens are changed which will cause revert with an index out-of-bounds error.
One can be exploited to pay for a different (attacker's) lien position because the lien array gets shortened when another payment (by the attacker or someone else) beforehand causes a deleted lien position.
Liquidations of multiple liens and payments against a collateral token and all its liens will revert. This will result in the collateral NFT being locked forever.
Exploit scenario: A malicious junior debt holder can buy out a senior debt (sandwich the original borrower effectively) to trigger this sequence of actions. For e.g., if the liens array is [1, 5, 3] where the malicious junior lien holder of 3 also buys out 1 and front-runs the borrower's intended payment for 5 by repaying 1 and forcing the array to shorten to [5, 3] will make the borrower pay off their lien of 3 instead of borrower's lien of 5. Effectively, a lien holder can "force" a payment to their own lien instead of another one.
Manual Review
Revisit and fix the entire logic that manages positions with the liens array addition, traversal and deletion.
Source: https://github.com/sherlock-audit/2022-10-astaria-judging/issues/169
0xRajeev
Missing to account for the new lien can allow loans on a collateral to exceed maximum potential debt leading to vault insolvency and potential loss of LP funds.
The LienToken.createLien function tries to prevent loans and the total potential debt from surpassing the defined params.terms.maxPotentialDebt limit. When the getTotalDebtForCollateralToken function is called, the new lien (which will be created within this transaction) is not yet added to the liens[collateralId] array. However, the getTotalDebtForCollateralToken function iterates over this very same array and will return the total debt without considering the new lien being added and for which this check is being performed.
The strategist's defined max potential debt limit can be exceeded, which changes/increases the risk for LPs, as it imposes a higher debt to the public vault. This could lead to vault insolvency and loss of LP funds.
PoC: https://gist.github.com/berndartmueller/8b0f870962acc4c999822d742e89151b
Example exploit: Given a public vault and a lien with a max potential debt amount of 50 ETH (that's the default standardLien in TestHelpers.t.sol)
1. 100 ETH have been deposited in the public vault by LPs
2. Bob borrows 50 ETH with his 1st NFT -> success
3. Bob borrows another 50 ETH with his 2nd NFT -> successful as well even though theirs a limit of 50 ETH
4. Bob now has 100 ETH -> The max potential debt limit is exceeded by 50 ETHManual Review
Check the total debt limit after adding the new lien to the liens[collateralId] array.
androolloyd
this is working as intended, the value is intended to represent the max potential debt the collateral can potentnially be under at the moment the new terms are originated.
secureum
Escalate for 2 USDC.
We still think this is a valid issue with a high-severity impact as described above and demonstrated in the PoC.
cc @berndartmueller @lucyoa
sherlock-admin
Escalate for 2 USDC.
We still think this is a valid issue with a high-severity impact as described above and demonstrated in the PoC.
cc @berndartmueller @lucyoa
You've created a valid escalation for 2 USDC!
To remove the escalation from consideration: Delete your comment. To change the amount you've staked on this escalation: Edit your comment (do not create a new comment).
You may delete or edit your escalation comment anytime before the 48-hour escalation window closes. After that, the escalation becomes final.
Evert0x
Escalation accepted
Having the possibility of exceeding the maximum potential debt should be mitigated
sherlock-admin
Escalation accepted
Having the possibility of exceeding the maximum potential debt should be mitigated
This issue's escalations have been accepted!
Contestants' payouts and scores will be updated according to the changes made on this issue.
Source: https://github.com/sherlock-audit/2022-10-astaria-judging/issues/167
0xRajeev
Triggering liquidations for a collateral after one of its liens has expired but before the auction window (default 2 days) at epoch end leads to loss and lock of funds.
Liquidations are allowed to be triggered for a collateral if any of its liens have exceeded their loan duration with outstanding payments. However, the liquidation logic does not execute the decrease of lien count and setting up of liquidation accountant if the time to epoch end is greater than the auction window (default 2 days). The auction proceeds nevertheless.
If liquidations are triggered before the auction window at epoch end, they will proceed to auctions without decreasing epoch lien count, without setting up the liquidation accountant for the lien and other related logic. This will, at a minimum, lead to auction proceeds going to lien owners directly instead of via the liquidation accountants (loss of funds) and the epoch unable to proceed to the next on (lock of funds and protocol halt).
1.https://github.com/sherlock-audit/2022-10-astaria/blob/main/src/AstariaRouter.sol#L388-L415
Manual Review
Revisit the liquidation logic and its triggering related to the auction window and epoch end.
Source: https://github.com/sherlock-audit/2022-10-astaria-judging/issues/163
Jeiwan, ctf_sec
Lack of access control in PublicVault.sol#transferWithdrawReserve let user call transferWithdrawReserve() multiple times to modify withdrawReserve
The function PublicVault.sol#transferWithdrawReserve() is meants to transfers funds from the PublicVault to the WithdrawProxy.
https://github.com/sherlock-audit/2022-10-astaria/blob/main/src/PublicVault.sol#L341-L363
However, this function has no access control, anyone can call it multiple times to modify the withdrawReserve value
A malicious actor can keep calling the function transferWthdrawReserve() before the withdrawal proxy is created.
If the underlying proxy has address(0), the transfer is not performed, but the state withdrawReserves is decremented.
Then user can invoke this function to always decrement the withdrawReserve to 0
// prevent transfer of more assets then are available
if (withdrawReserve <= withdraw) {
withdraw = withdrawReserve;
withdrawReserve = 0;
} else {
withdrawReserve -= withdraw;
}https://github.com/sherlock-audit/2022-10-astaria/blob/main/src/PublicVault.sol#L341-L363
Manual Review
We recommend the project add requiestAuth modifier to the function
function transferWithdrawReserve() public {We can also change the implementation by implmenting: if the underlying withdrawProxy is address(0), revert transfer.
IAmTurnipBoy
Escalate for 1 USDC
Invalid. Impossible for withdrawReserve != 0 and there isn't a withdraw proxy, because of the following logic. Anytime there is a liquidation a liquidation accountant is deployed and that ALWAYS deploys a withdraw proxy. If there's no liquidations then there's no withdrawReserves.
sherlock-admin
Escalate for 1 USDC
Invalid. Impossible for withdrawReserve != 0 and there isn't a withdraw proxy, because of the following logic. Anytime there is a liquidation a liquidation accountant is deployed and that ALWAYS deploys a withdraw proxy. If there's no liquidations then there's no withdrawReserves.
You've created a valid escalation for 1 USDC!
To remove the escalation from consideration: Delete your comment. To change the amount you've staked on this escalation: Edit your comment (do not create a new comment).
You may delete or edit your escalation comment anytime before the 48-hour escalation window closes. After that, the escalation becomes final.
Evert0x
Escalation rejected. There's no hard requirement to always have a liquidation accountant..
sherlock-admin
Escalation rejected. There's no hard requirement to always have a liquidation accountant..
This issue's escalations have been rejected!
Watsons who escalated this issue will have their escalation amount deducted from their next payout.
Source: https://github.com/sherlock-audit/2022-10-astaria-judging/issues/107
TurnipBoy
When a token still has open liens against it only the value of the liens will be paid by the bidder but their current bid will be set to the full value of the bid. This can be abused in one of two ways. The bidder could place a massive bid like 500 ETH that will never be outbid or they could place a bid they know will outbid and profit the difference when they're sent a refund.
uint256[] memory liens = LIEN_TOKEN.getLiens(tokenId);
uint256 totalLienAmount = 0;
if (liens.length > 0) {
for (uint256 i = 0; i < liens.length; ++i) {
uint256 payment;
uint256 lienId = liens[i];
ILienToken.Lien memory lien = LIEN_TOKEN.getLien(lienId);
if (transferAmount >= lien.amount) {
payment = lien.amount;
transferAmount -= payment;
} else {
payment = transferAmount;
transferAmount = 0;
}
if (payment > 0) {
LIEN_TOKEN.makePayment(tokenId, payment, lien.position, payer);
}
}
} else {
//@audit-issue logic skipped if liens.length > 0
TRANSFER_PROXY.tokenTransferFrom(
weth,
payer,
COLLATERAL_TOKEN.ownerOf(tokenId),
transferAmount
);
}We can examine the payment logic inside _handleIncomingPayment and see that if there are still open liens against then only the amount of WETH to pay back the liens will be taken from the payer, since the else portion of the logic will be skipped.
uint256 vaultPayment = (amount - currentBid);
if (firstBidTime == 0) {
auctions[tokenId].firstBidTime = block.timestamp.safeCastTo64();
} else if (lastBidder != address(0)) {
uint256 lastBidderRefund = amount - vaultPayment;
_handleOutGoingPayment(lastBidder, lastBidderRefund);
}
_handleIncomingPayment(tokenId, vaultPayment, address(msg.sender));
auctions[tokenId].currentBid = amount;
auctions[tokenId].bidder = address(msg.sender);In createBid, auctions[tokenId].currentBid is set to amount after the last bidder is refunded and the excess is paid against liens. We can walk through an example to illustrate this:
Assume a token with a single lien of amount 10 WETH and an auction is opened for that token. Now a user places a bid for 20 WETH. They are the first bidder so lastBidder = address(0) and currentBid = 0. _handleIncomingPayment will be called with a value of 20 WETH since there is no lastBidder to refund. Inside _handleIncomingPayment the lien information is read showing 1 lien against the token. Since transferAmount >= lien.amount, payment = lien.amount. A payment will be made by the bidder against the lien for 10 WETH. After the payment _handleIncomingPayment will return only having taken 10 WETH from the bidder. In the next line currentBid is set to 20 WETH but the bidder has only paid 10 WETH. Now if they are outbid, the new bidder will have to refund then 20 WETH even though they initially only paid 10 WETH.
Bidder can steal funds due to _handleIncomingPayment not taking enough WETH
Manual Review
In _handleIncomingPayment, all residual transfer amount should be sent to COLLATERAL_TOKEN.ownerOf(tokenId).
Source: https://github.com/sherlock-audit/2022-10-astaria-judging/issues/88
TurnipBoy, rvierdiiev
Possible to fully block PublicVault.processEpoch function. No one will be able to receive their funds
When liquidity providers want to redeem their share from PublicVault they call redeemFutureEpoch function which will create new WithdrawProxy for the epoch(if not created already) and then mint shares for redeemer in WithdrawProxy. PublicVault transfer user's shares to himself.
https://github.com/sherlock-audit/2022-10-astaria/blob/main/src/PublicVault.sol#L178-L212
function redeemFutureEpoch(
uint256 shares,
address receiver,
address owner,
uint64 epoch
) public virtual returns (uint256 assets) {
// check to ensure that the requested epoch is not the current epoch or in the past
require(epoch >= currentEpoch, "Exit epoch too low");
require(msg.sender == owner, "Only the owner can redeem");
// check for rounding error since we round down in previewRedeem.
ERC20(address(this)).safeTransferFrom(owner, address(this), shares);
// Deploy WithdrawProxy if no WithdrawProxy exists for the specified epoch
_deployWithdrawProxyIfNotDeployed(epoch);
emit Withdraw(msg.sender, receiver, owner, assets, shares);
// WithdrawProxy shares are minted 1:1 with PublicVault shares
WithdrawProxy(withdrawProxies[epoch]).mint(receiver, shares); // was withdrawProxies[withdrawEpoch]
}This function mints WithdrawProxy shares 1:1 to redeemed PublicVault shares.
Then later after call of processEpoch and transferWithdrawReserve the funds will be sent to the WithdrawProxy and users can now redeem their shares from it.
Function processEpoch decides how many funds should be sent to the WithdrawProxy.
https://github.com/sherlock-audit/2022-10-astaria/blob/main/src/PublicVault.sol#L268-L289
if (withdrawProxies[currentEpoch] != address(0)) {
uint256 proxySupply = WithdrawProxy(withdrawProxies[currentEpoch])
.totalSupply();
liquidationWithdrawRatio = proxySupply.mulDivDown(1e18, totalSupply());
if (liquidationAccountants[currentEpoch] != address(0)) {
LiquidationAccountant(liquidationAccountants[currentEpoch])
.setWithdrawRatio(liquidationWithdrawRatio);
}
uint256 withdrawAssets = convertToAssets(proxySupply);
// compute the withdrawReserve
uint256 withdrawLiquidations = liquidationsExpectedAtBoundary[
currentEpoch
].mulDivDown(liquidationWithdrawRatio, 1e18);
withdrawReserve = withdrawAssets - withdrawLiquidations;
// burn the tokens of the LPs withdrawing
_burn(address(this), proxySupply);
_decreaseYIntercept(withdrawAssets);
}This is how it is decided how much money should be sent to WithdrawProxy.
Firstly, we look at totalSupply of WithdrawProxy.
uint256 proxySupply = WithdrawProxy(withdrawProxies[currentEpoch]).totalSupply();.
And then we convert them to assets amount.
uint256 withdrawAssets = convertToAssets(proxySupply);
In the end function burns proxySupply amount of shares controlled by PublicVault.
_burn(address(this), proxySupply);
Then this amount is allowed to be sent(if no auctions currently, but this is not important right now).
This all allows to attacker to make WithdrawProxy.deposit to mint new shares for him and increase totalSupply of WithdrawProxy, so proxySupply becomes more then was sent to PublicVault.
This is attack scenario.
1.PublicVault is created and funded with 50 ethers.
2.Someone calls redeemFutureEpoch function to create new WithdrawProxy for next epoch.
3.Attacker sends 1 wei to WithdrawProxy to make totalAssets be > 0. Attacker deposit to WithdrawProxy 1 wei. Now WithdrawProxy.totalSupply > PublicVault.balanceOf(PublicVault).
4.Someone call processEpoch and it reverts on burning.
As result, nothing will be send to WithdrawProxy where shares were minted for users. The just lost money.
Also this attack can be improved to drain users funds to attacker.
Attacker should be liquidity provider. And he can initiate next redeem for next epoch, then deposit to new WithdrawProxy enough amount to get new shares. And call processEpoch which will send to the vault amount, that was not sent to previous attacked WithdrawProxy, as well. So attacker will take those funds.
Funds of PublicVault depositors are stolen.
This is simple test that shows how external actor can corrupt WithdrawProxy.
function testWithdrawProxyDdos() public {
Dummy721 nft = new Dummy721();
address tokenContract = address(nft);
uint256 tokenId = uint256(1);
address publicVault = _createPublicVault({
strategist: strategistOne,
delegate: strategistTwo,
epochLength: 14 days
});
_lendToVault(
Lender({addr: address(1), amountToLend: 50 ether}),
publicVault
);
uint256 collateralId = tokenContract.computeId(tokenId);
uint256 vaultTokenBalance = IERC20(publicVault).balanceOf(address(1));
console.log("balance: ", vaultTokenBalance);
// _signalWithdrawAtFutureEpoch(address(1), publicVault, uint64(1));
_signalWithdraw(address(1), publicVault);
address withdrawProxy = PublicVault(publicVault).withdrawProxies(
PublicVault(publicVault).getCurrentEpoch()
);
vm.deal(address(2), 2);
vm.startPrank(address(2));
WETH9.deposit{value: 2}();
//this we need to make share calculation not fail with divide by 0
WETH9.transferFrom(address(2), withdrawProxy, 1);
WETH9.approve(withdrawProxy, 1);
//deposit 1 wei to make WithdrawProxy.totalSupply > PublicVault.balanceOf(address(PublicVault))
//so processEpoch can't burn more shares
WithdrawProxy(withdrawProxy).deposit(1, address(2));
vm.stopPrank();
assertEq(vaultTokenBalance, IERC20(withdrawProxy).balanceOf(address(1)));
vm.warp(block.timestamp + 15 days);
//processEpoch fails, because it can't burn more amount of shares, that was sent to PublicVault
vm.expectRevert();
PublicVault(publicVault).processEpoch();
}Manual Review
Make function WithdrawProxy.deposit not callable.
IAmTurnipBoy
Escalate for 1 USDC
Causes total loss of funds for LPs when this happens. High risk
sherlock-admin
Escalate for 1 USDC
Causes total loss of funds for LPs when this happens. High risk
You've created a valid escalation for 1 USDC!
To remove the escalation from consideration: Delete your comment. To change the amount you've staked on this escalation: Edit your comment (do not create a new comment).
You may delete or edit your escalation comment anytime before the 48-hour escalation window closes. After that, the escalation becomes final.
Evert0x
Escalation accepted
sherlock-admin
Escalation accepted
This issue's escalations have been accepted!
Contestants' payouts and scores will be updated according to the changes made on this issue.
Source: https://github.com/sherlock-audit/2022-10-astaria-judging/issues/72
obront
The strategist signs the merkle root, their nonce, and the deadline of all strategies to ensure that new borrowers meet their criteria. However, the lien type (nlrType) is not signed. Currently, the structs for the different types are unique, so there is no ability to borrow one type as another, but if struct schemas of different types overlap in the future, this will open the door for exploits.
When a new lien is requested, the borrower submits a Lien Request, which is filled with the parameters at which they would like to borrow. This is kept honest and aligned with the lenders intent because the merkle root, strategist nonce, and deadline are all signed by the strategist.
Because the merkle root is signed, the borrower must submit lien parameters (nlrDetails) that align with one of the strategies that the strategist has chosen to allow (represented as leaves in the merkle tree). The schemas of these signed structs differ depending on the validator being used, which is defined in the nlrType parameter.
Currently, each of the validators has a unique schema for their struct. However, if there is an overlap in the schema of the Details struct of multiple validators, where the different parameters represent different values, it opens the door to having a fraudulent lien accepted.
Here's an example of how this might work:
lienDetails with nlrType = Type A and send the validation to the wrong strategy validator, thus bypassing the expected checksAs more strategy types are added, conflicts in struct schemas could open the door to fraudulent behavior.
https://github.com/sherlock-audit/2022-10-astaria/blob/main/src/AstariaRouter.sol#L222-L232
Current Details struct schemas:
Manual Review
Include the nlrType in the data signed by the strategist. The easiest way to do this would be to pack it in with each leaf when assembling the leaves that will create the merkle tree:
function assembleLeaf(ICollectionValidator.Details memory details, address nlrType)
public
pure
returns (bytes memory)
{
return abi.encode(details, nlrType);
}
function validateAndParse... {
...
leaf = keccak256(assembleLeaf(cd, params.nlrType));
}Source: https://github.com/sherlock-audit/2022-10-astaria-judging/issues/69
obront, HonorLt, rvierdiiev, cryptphi, yixxas, zzykxx
If a public vault is created without a delegate, delegate will have the value of address(0). This is also the value returned by ecrecover for invalid signatures (for example, if v is set to a position number that is not 27 or 28), which allows a malicious actor to cause the signature validation to pass for arbitrary parameters, allowing them to drain a vault using a worthless NFT as collateral.
When a new Public Vault is created, the Router calls the init() function on the vault as follows:
VaultImplementation(vaultAddr).init(
VaultImplementation.InitParams(delegate)
);If a delegate wasn't set, this will pass address(0) to the vault. If this value is passed, the vault simply skips the assignment, keeping the delegate variable set to the default 0 value:
if (params.delegate != address(0)) {
delegate = params.delegate;
}Once the delegate is set to the zero address, any commitment can be validated, even if the signature is incorrect. This is because of a quirk in ecrecover which returns address(0) for invalid signatures. A signature can be made invalid by providing a positive integer that is not 27 or 28 as the v value. The result is that the following function call assigns recovered = address(0):
address recovered = ecrecover(
keccak256(
encodeStrategyData(
params.lienRequest.strategy,
params.lienRequest.merkle.root
)
),
params.lienRequest.v,
params.lienRequest.r,
params.lienRequest.s
);To confirm the validity of the signature, the function performs two checks:
require(
recovered == params.lienRequest.strategy.strategist,
"strategist must match signature"
);
require(
recovered == owner() || recovered == delegate,
"invalid strategist"
);These can be easily passed by setting the strategist in the params to address(0). At this point, all checks will pass and the parameters will be accepted as approved by the vault.
With this power, a borrower can create params that allow them to borrow the vault's full funds in exchange for a worthless NFT, allowing them to drain the vault and steal all the user's funds.
All user's funds held in a vault with no delegate set can be stolen.
https://github.com/sherlock-audit/2022-10-astaria/blob/main/src/AstariaRouter.sol#L537-L539
https://github.com/sherlock-audit/2022-10-astaria/blob/main/src/VaultImplementation.sol#L118-L124
https://github.com/sherlock-audit/2022-10-astaria/blob/main/src/VaultImplementation.sol#L167-L185
Manual Review, Foundry
Add a require statement that the recovered address cannot be the zero address:
require(recovered != address(0));sherlock-admin
Escalate for 1 USDC
Disagree with high. Requires a external factors and a bit of phishing to make this work. User would have to voluntarily make a lien position that has strategist == address(0) which should raise red flags. Medium seems more fitting.
Relevant lines:
You've deleted an escalation for this issue.
zobront
Escalate for 5 USDC
This may be the most severe flaw in the whole protocol. Definitely deserves to be a high.
I’m not sure why it was downgraded, but the deleted escalation seems to think that the strategist needs to be address(0). This isn’t the case. We just need to set the strategist to address(0) in our fake data to make the signature pass.
The delegate is a separate role that can be given access, and any user that doesn’t set a delegate will default to delegate being set to address(0).
Any of these vaults can be drained for every single dollar in the vault due to this flaw.
As you can see from the dups, every single one that realized this exploit marked it as a high. The only mediums are the ones that caught the zero address gap but didn’t understand how it was exploitable.
sherlock-admin
Escalate for 5 USDC
This may be the most severe flaw in the whole protocol. Definitely deserves to be a high.
I’m not sure why it was downgraded, but the deleted escalation seems to think that the strategist needs to be address(0). This isn’t the case. We just need to set the strategist to address(0) in our fake data to make the signature pass.
The delegate is a separate role that can be given access, and any user that doesn’t set a delegate will default to delegate being set to address(0).
Any of these vaults can be drained for every single dollar in the vault due to this flaw.
As you can see from the dups, every single one that realized this exploit marked it as a high. The only mediums are the ones that caught the zero address gap but didn’t understand how it was exploitable.
You've created a valid escalation for 5 USDC!
To remove the escalation from consideration: Delete your comment. To change the amount you've staked on this escalation: Edit your comment (do not create a new comment).
You may delete or edit your escalation comment anytime before the 48-hour escalation window closes. After that, the escalation becomes final.
Evert0x
Escalation accepted
sherlock-admin
Escalation accepted
This issue's escalations have been accepted!
Contestants' payouts and scores will be updated according to the changes made on this issue.
Source: https://github.com/sherlock-audit/2022-10-astaria-judging/issues/51
obront
When liens are liquidated, the router checks if the auction will complete in a future epoch and, if it does, sets up a liquidation accountant and other logistics to account for it. However, the check for auction completion does not take into account extended auctions, which can therefore end in an unexpected epoch and cause accounting issues, losing user funds.
The liquidate() function performs the following check to determine if it should set up the liquidation to be paid out in a future epoch:
if (PublicVault(owner).timeToEpochEnd() <= COLLATERAL_TOKEN.auctionWindow())This function assumes that the auction will only end in a future epoch if the auctionWindow (typically set to 2 days) pushes us into the next epoch.
However, auctions can last up to an additional 1 day if bids are made within the final 15 minutes. In these cases, auctions are extended repeatedly, up to a maximum of 1 day.
if (firstBidTime + duration - block.timestamp < timeBuffer) {
uint64 newDuration = uint256(
duration + (block.timestamp + timeBuffer - firstBidTime)
).safeCastTo64();
if (newDuration <= auctions[tokenId].maxDuration) {
auctions[tokenId].duration = newDuration;
} else {
auctions[tokenId].duration =
auctions[tokenId].maxDuration -
firstBidTime;
}
extended = true;
}The result is that there are auctions for which accounting is set up for them to end in the current epoch, but will actual end in the next epoch.
Users who withdrew their funds in the current epoch, who are entitled to a share of the auction's proceeds, will not be paid out fairly.
https://github.com/sherlock-audit/2022-10-astaria/blob/main/src/AstariaRouter.sol#L388-L415
Manual Review
Change the check to take the possibility of extension into account:
if (PublicVault(owner).timeToEpochEnd() <= COLLATERAL_TOKEN.auctionWindow() + 1 days)IAmTurnipBoy
Escalate for 1 USDC
The specific issue address in this submission is incorrect. It passes COLLATERAL_TOKEN.auctionWindow() + 1 days (max possible duration of an auction) into handleNewLiquidation which makes this a non issue.
Relevant Lines: https://github.com/sherlock-audit/2022-10-astaria/blob/7d12a5516b7c74099e1ce6fb4ec87c102aec2786/src/AstariaRouter.sol#L407-L410
sherlock-admin
Escalate for 1 USDC
The specific issue address in this submission is incorrect. It passes COLLATERAL_TOKEN.auctionWindow() + 1 days (max possible duration of an auction) into handleNewLiquidation which makes this a non issue.
Relevant Lines: https://github.com/sherlock-audit/2022-10-astaria/blob/7d12a5516b7c74099e1ce6fb4ec87c102aec2786/src/AstariaRouter.sol#L407-L410
You've created a valid escalation for 1 USDC!
To remove the escalation from consideration: Delete your comment. To change the amount you've staked on this escalation: Edit your comment (do not create a new comment).
You may delete or edit your escalation comment anytime before the 48-hour escalation window closes. After that, the escalation becomes final.
Evert0x
Escalation rejected.
The if in https://github.com/sherlock-audit/2022-10-astaria/blob/main/src/AstariaRouter.sol#L388-L391 is missing the auction window extension of 1 day. This leads to auctions with extended durations overlapping the current epoch and not having liquidation accountants in place
sherlock-admin
Escalation rejected.
The
ifin https://github.com/sherlock-audit/2022-10-astaria/blob/main/src/AstariaRouter.sol#L388-L391 is missing the auction window extension of 1 day. This leads to auctions with extended durations overlapping the current epoch and not having liquidation accountants in place
This issue's escalations have been rejected!
Watsons who escalated this issue will have their escalation amount deducted from their next payout.
Source: https://github.com/sherlock-audit/2022-10-astaria-judging/issues/49
obront
Strategists set their vault fee in BPS (x / 10,000), but are paid out as x / 1,000. The result is that strategists will always earn 10x whatever vault fee they set.
Whenever any payment is made towards a public vault, beforePayment() is called, which calls _handleStrategistInterestReward().
The function is intended to take the amount being paid, adjust by the vault fee to get the fee amount, and convert that amount of value into shares, which are added to strategistUnclaimedShares.
function _handleStrategistInterestReward(uint256 lienId, uint256 amount)
internal
virtual
override
{
if (VAULT_FEE() != uint256(0)) {
uint256 interestOwing = LIEN_TOKEN().getInterest(lienId);
uint256 x = (amount > interestOwing) ? interestOwing : amount;
uint256 fee = x.mulDivDown(VAULT_FEE(), 1000);
strategistUnclaimedShares += convertToShares(fee);
}
}Since the vault fee is stored in basis points, to get the vault fee, we should take the amount, multiply it by VAULT_FEE() and divide by 10,000. However, we accidentally divide by 1,000, which results in a 10x larger reward for the strategist than intended.
As an example, if the vault fee is intended to be 10%, we would set VAULT_FEE = 1000. In that case, for any amount paid off, we would calculate fee = amount * 1000 / 1000 and the full amount would be considered a fee for the strategist.
Strategists will be paid 10x the agreed upon rate for their role, with the cost being borne by users.
https://github.com/sherlock-audit/2022-10-astaria/blob/main/src/PublicVault.sol#L513-L524
Manual Review
Change the 1000 in the _handleStrategistInterestReward() function to 10_000.
IAmTurnipBoy
Escalate for 1 USDC
Don't agree with high severity. Fee can easily be changed by protocol after the fact to fix this, which is why medium makes more sense. Simple to fix as a parameter change.
sherlock-admin
Escalate for 1 USDC
Don't agree with high severity. Fee can easily be changed by protocol after the fact to fix this, which is why medium makes more sense. Simple to fix as a parameter change.
You've created a valid escalation for 1 USDC!
To remove the escalation from consideration: Delete your comment. To change the amount you've staked on this escalation: Edit your comment (do not create a new comment).
You may delete or edit your escalation comment anytime before the 48-hour escalation window closes. After that, the escalation becomes final.
Evert0x
Escalation accepted.
sherlock-admin
Escalation accepted.
This issue's escalations have been accepted!
Contestants' payouts and scores will be updated according to the changes made on this issue.
zobront
Escalate for 5 USDC
VAULT_FEE() is set as an immutable arg when vault is deployed using ClonesWithImmutableArgs. There is no ability to change it later as the escalation claims. This value would be hard coded in all vaults created.
The result is that vault strategists would be given all of the funds of users who deposit in their vault, irretrievably stealing funds from users of the protocol. This is definitely a high.
sherlock-admin
Escalate for 5 USDC
VAULT_FEE() is set as an immutable arg when vault is deployed using ClonesWithImmutableArgs. There is no ability to change it later as the escalation claims. This value would be hard coded in all vaults created.
The result is that vault strategists would be given all of the funds of users who deposit in their vault, irretrievably stealing funds from users of the protocol. This is definitely a high.
You've created a valid escalation for 5 USDC!
To remove the escalation from consideration: Delete your comment. To change the amount you've staked on this escalation: Edit your comment (do not create a new comment).
You may delete or edit your escalation comment anytime before the 48-hour escalation window closes. After that, the escalation becomes final.
Evert0x
Escalation accepted
sherlock-admin
Escalation accepted
This issue's escalations have been accepted!
Contestants' payouts and scores will be updated according to the changes made on this issue.
Source: https://github.com/sherlock-audit/2022-10-astaria-judging/issues/48
obront
When claim() is called on the Liquidation Accountant, it decreases the y-intercept based on the balance of the contract after funds have been distributed, rather than before. The result is that the y-intercept will be decreased more than it should be, siphoning funds from all users.
When LiquidationAccountant.sol:claim() is called, it uses its withdrawRatio to send some portion of its earnings to the WITHDRAW_PROXY and the rest to the vault.
After performing these transfers, it updates the vault's y-intercept, decreasing it by the gap between the expected return from the auction, and the reality of how much was sent back to the vault:
PublicVault(VAULT()).decreaseYIntercept(
(expected - ERC20(underlying()).balanceOf(address(this))).mulDivDown(
1e18 - withdrawRatio,
1e18
)
);This rebalancing uses the balance of the liquidationAccountant to perform its calculation, but it is done after the balance has already been distributed, so it will always be 0.
Looking at an example:
expected = 1 ether (meaning the y-intercept is currently based on this value)withdrawRatio = 0 (meaning all funds will go back to the vault)(expected - 0) * 1e18 / 1e18, which equals expectedThat decrease should not happen, and causing problems for the protocol's accounting. For example, when withdraw() is called, it uses the y-intercept in its calculation of the totalAssets() held by the vault, creating artificially low asset values for a given number of shares.
Every time the liquidation accountant is used, the vault's math will be thrown off and user shares will be falsely diluted.
https://github.com/sherlock-audit/2022-10-astaria/blob/main/src/LiquidationAccountant.sol#L62-L97
Manual Review
The amount of assets sent to the vault has already been calculated, as we've already sent it. Therefore, rather than the full existing formula, we can simply call:
PublicVault(VAULT()).decreaseYIntercept(expected - balance)Alternatively, we can move the current code above the block of code that transfers funds out (L73).
Source: https://github.com/sherlock-audit/2022-10-astaria-judging/issues/46
obront
New liquidations are sent to the liquidationAccountant with a finalAuctionTimestamp value, but the actual value that is passed in is simply the duration of an auction. The claim() function uses this value in a require check, so this error will allow it to be called before the auction is complete.
When a lien is liquidated, AstariaRouter.sol:liquidate() is called. If the lien is set to end in a future epoch, we call handleNewLiquidation() on the liquidationAccountant.
One of the values passed in this call is the finalAuctionTimestamp, which updates the finalAuctionEnd variable in the liquidationAccountant. This value is then used to protect the claim() function from being called too early.
However, when the router calls handleLiquidationAccountant(), it passes the duration of an auction rather than the final timestamp:
LiquidationAccountant(accountant).handleNewLiquidation(
lien.amount,
COLLATERAL_TOKEN.auctionWindow() + 1 days
);As a result, finalAuctionEnd will be set to 259200 (3 days).
When claim() is called, it requires the final auction to have ended for the function to be called:
require(
block.timestamp > finalAuctionEnd || finalAuctionEnd == uint256(0),
"final auction has not ended"
);Because of the error above, block.timestamp will always be greater than finalAuctionEnd, so this will always be permitted.
Anyone can call claim() before an auction has ended. This can cause many problems, but the clearest is that it can ruin the protocol's accounting by decreasing the Y intercept of the vault.
For example, if claim() is called before the auction, the returned value will be 0, so the Y intercept will be decreased as if there was an auction that returned no funds.
https://github.com/sherlock-audit/2022-10-astaria/blob/main/src/AstariaRouter.sol#L407-L410
https://github.com/sherlock-audit/2022-10-astaria/blob/main/src/LiquidationAccountant.sol#L113-L120
https://github.com/sherlock-audit/2022-10-astaria/blob/main/src/LiquidationAccountant.sol#L65-L69
Manual Review
Adjust the call from the router to use the ending timestamp as the argument, rather than the duration:
LiquidationAccountant(accountant).handleNewLiquidation(
lien.amount,
block.timestamp + COLLATERAL_TOKEN.auctionWindow() + 1 days
);IAmTurnipBoy
Escalate for 1 USDC
Touched on the same idea as #135. Tough call on duplication. This issue + #47 combine represent the same vulnerability fixed in both #135 and #188. In this issue it addresses being called too early and in #47 it addresses being called multiple times. The fix proposed in #135 and #188 address both issues by permissioning the function. IMHO this and 47 should be duped with #135 and #188, but up to judges.
ALSO little bit of a conflict of interest that @zobront validated his own issue here.
sherlock-admin
Escalate for 1 USDC
Touched on the same idea as #135. Tough call on duplication. This issue + #47 combine represent the same vulnerability fixed in both #135 and #188. In this issue it addresses being called too early and in #47 it addresses being called multiple times. The fix proposed in #135 and #188 address both issues by permissioning the function. IMHO this and 47 should be duped with #135 and #188, but up to judges.
ALSO little bit of a conflict of interest that @zobront validated his own issue here.
You've created a valid escalation for 1 USDC!
To remove the escalation from consideration: Delete your comment. To change the amount you've staked on this escalation: Edit your comment (do not create a new comment).
You may delete or edit your escalation comment anytime before the 48-hour escalation window closes. After that, the escalation becomes final.
Evert0x
Escalation accepted.
claim() is not safe to be publicly callable.sherlock-admin
Escalation accepted.
135 and #188 are already duplicates. This issue and #47 touch on the same core issue that
claim()is not safe to be publicly callable.
This issue's escalations have been accepted!
Contestants' payouts and scores will be updated according to the changes made on this issue.
zobront
Escalate for 5 USDC
This isn’t a dup of #135 or #188. Those two issues are talking about access control for the function. This calls out that there is a specific time enforcement mechanism (finalAuctionTimestamp) that is calculated incorrectly, allowing it to be called early.
Yes, I agree that adding access controls (their solution) would reduce the harm from this issue, but they are unrelated issues that just happen to have overlapping solutions. The real solution to this one is to do the calculation properly.
NOTE: The other issue of mine that was dup’d with these is #47, which focuses on that there aren’t restrictions to stop claim() being called multiple times. I probably wouldn’t consider that a dup of theirs either, but it’s on the fence so I’m ok with it — but this one is clearly different.
sherlock-admin
Escalate for 5 USDC
This isn’t a dup of #135 or #188. Those two issues are talking about access control for the function. This calls out that there is a specific time enforcement mechanism (
finalAuctionTimestamp) that is calculated incorrectly, allowing it to be called early.Yes, I agree that adding access controls (their solution) would reduce the harm from this issue, but they are unrelated issues that just happen to have overlapping solutions. The real solution to this one is to do the calculation properly.
NOTE: The other issue of mine that was dup’d with these is #47, which focuses on that there aren’t restrictions to stop
claim()being called multiple times. I probably wouldn’t consider that a dup of theirs either, but it’s on the fence so I’m ok with it — but this one is clearly different.
You've created a valid escalation for 5 USDC!
To remove the escalation from consideration: Delete your comment. To change the amount you've staked on this escalation: Edit your comment (do not create a new comment).
You may delete or edit your escalation comment anytime before the 48-hour escalation window closes. After that, the escalation becomes final.
Evert0x
Escalation accepted. at the initial escalation we misinterpreted the https://github.com/sherlock-audit/2022-10-astaria-judging/issues/135 / https://github.com/sherlock-audit/2022-10-astaria-judging/issues/188 as it was thought they described both issues.
sherlock-admin
Escalation accepted. at the initial escalation we misinterpreted the https://github.com/sherlock-audit/2022-10-astaria-judging/issues/135 / https://github.com/sherlock-audit/2022-10-astaria-judging/issues/188 as it was thought they described both issues.
This issue's escalations have been accepted!
Contestants' payouts and scores will be updated according to the changes made on this issue.
Source: https://github.com/sherlock-audit/2022-10-astaria-judging/issues/36
csanuragjain
If user has provided transferAmount which is greater than all lien.amount combined then initiatorPayment will be incorrect since it is charged on full amount when only partial was used as shown in poc
uint256 initiatorPayment = transferAmount.mulDivDown(
auction.initiatorFee,
100
); if (transferAmount >= lien.amount) {
payment = lien.amount;
transferAmount -= payment;
} else {
payment = transferAmount;
transferAmount = 0;
}
if (payment > 0) {
LIEN_TOKEN.makePayment(tokenId, payment, lien.position, payer);
}
}Excess initiator fees will be deducted which was not required
Manual Review
Calculate the exact amount of transfer amount required for the transaction and calculate the initiator fee based on this amount
IAmTurnipBoy
Escalate for 1 USDC
The fee is not the problem, so this report is invalid. The issue is that the payment isn't working as intended and that sometimes it doesn't take as much as it should. See #107.
sherlock-admin
Escalate for 1 USDC
The fee is not the problem, so this report is invalid. The issue is that the payment isn't working as intended and that sometimes it doesn't take as much as it should. See #107.
You've created a valid escalation for 1 USDC!
To remove the escalation from consideration: Delete your comment. To change the amount you've staked on this escalation: Edit your comment (do not create a new comment).
You may delete or edit your escalation comment anytime before the 48-hour escalation window closes. After that, the escalation becomes final.
Evert0x
Escalation rejected.
This is a valid finding as fees are too high - Fees should be calculated based on the actual amount used as the repayment
sherlock-admin
Escalation rejected.
This is a valid finding as fees are too high - Fees should be calculated based on the actual amount used as the repayment
This issue's escalations have been rejected!
Watsons who escalated this issue will have their escalation amount deducted from their next payout.
Source: https://github.com/sherlock-audit/2022-10-astaria-judging/issues/22
obront
isValidRefinance() is intended to check whether either (a) the loan interest rate decreased sufficiently or (b) the loan duration increased sufficiently. Instead, it requires both of these to be true, leading to the rejection of valid refinances.
When trying to buy out a lien from LienToken.sol:buyoutLien(), the function calls AstariaRouter.sol:isValidRefinance() to check whether the refi terms are valid.
if (!ASTARIA_ROUTER.isValidRefinance(lienData[lienId], ld)) {
revert InvalidRefinance();
}One of the roles of this function is to check whether the rate decreased by more than 0.5%. From the docs:
An improvement in terms is considered if either of these conditions is met:
- The loan interest rate decrease by more than 0.5%.
- The loan duration increases by more than 14 days.
The currently implementation of the code requires both of these conditions to be met:
return (
newLien.rate >= minNewRate &&
((block.timestamp + newLien.duration - lien.start - lien.duration) >= minDurationIncrease)
);Valid refinances that meet one of the two criteria will be rejected.
https://github.com/sherlock-audit/2022-10-astaria/blob/main/src/AstariaRouter.sol#L488-L490
Manual Review
Change the AND in the return statement to an OR:
return (
newLien.rate >= minNewRate ||
((block.timestamp + newLien.duration - lien.start - lien.duration) >= minDurationIncrease)
);SantiagoGregory
Independently fixed during our own review so there's no PR specifically for this, but this is now updated to an or.
IAmTurnipBoy
Escalate for 1 USDC
Should be medium because there are no funds at risk
sherlock-admin
Escalate for 1 USDC
Should be medium because there are no funds at risk
You've created a valid escalation for 1 USDC!
To remove the escalation from consideration: Delete your comment. To change the amount you've staked on this escalation: Edit your comment (do not create a new comment).
You may delete or edit your escalation comment anytime before the 48-hour escalation window closes. After that, the escalation becomes final.
Evert0x
Escalation rejected.
Not a major loss of funds but definitely a severe flaw that will hurt the protocol.
sherlock-admin
Escalation rejected.
Not a major loss of funds but definitely a severe flaw that will hurt the protocol.
This issue's escalations have been rejected!
Watsons who escalated this issue will have their escalation amount deducted from their next payout.
Source: https://github.com/sherlock-audit/2022-10-astaria-judging/issues/21
hansfriese, obront, 0xRajeev
The math in isValidRefinance() checks whether the rate increased rather than decreased, resulting in invalid refinances being approved and valid refinances being rejected.
When trying to buy out a lien from LienToken.sol:buyoutLien(), the function calls AstariaRouter.sol:isValidRefinance() to check whether the refi terms are valid.
if (!ASTARIA_ROUTER.isValidRefinance(lienData[lienId], ld)) {
revert InvalidRefinance();
}One of the roles of this function is to check whether the rate decreased by more than 0.5%. From the docs:
An improvement in terms is considered if either of these conditions is met:
- The loan interest rate decrease by more than 0.5%.
- The loan duration increases by more than 14 days.
The current implementation of the function does the opposite. It calculates a minNewRate (which should be maxNewRate) and then checks whether the new rate is greater than that value.
uint256 minNewRate = uint256(lien.rate) - minInterestBPS;
return (newLien.rate >= minNewRate ...The result is that if the new rate has increased (or decreased by less than 0.5%), it will be considered valid, but if it has decreased by more than 0.5% (the ideal behavior) it will be rejected as invalid.
https://github.com/sherlock-audit/2022-10-astaria/blob/main/src/AstariaRouter.sol#L482-L491
Manual Review
Flip the logic used to check the rate to the following:
uint256 maxNewRate = uint256(lien.rate) - minInterestBPS;
return (newLien.rate <= maxNewRate...IAmTurnipBoy
Escalate for 1 USDC
Should be medium because no funds at risk
sherlock-admin
Escalate for 1 USDC
Should be medium because no funds at risk
You've created a valid escalation for 1 USDC!
To remove the escalation from consideration: Delete your comment. To change the amount you've staked on this escalation: Edit your comment (do not create a new comment).
You may delete or edit your escalation comment anytime before the 48-hour escalation window closes. After that, the escalation becomes final.
Evert0x
Escalation rejected.
Not a major loss of funds but definitely a severe flaw that will hurt the protocol.
sherlock-admin
Escalation rejected.
Not a major loss of funds but definitely a severe flaw that will hurt the protocol.
This issue's escalations have been rejected!
Watsons who escalated this issue will have their escalation amount deducted from their next payout.
Source: https://github.com/sherlock-audit/2022-10-astaria-judging/issues/264
bin2chen
document : " Epochs PublicVaults operate around a time-based epoch system. An epoch length is defined by the strategist that deploys the PublicVault. The duration of new loans is restricted to not exceed the end of the next epoch. For example, if a PublicVault is 15 days into a 30-day epoch, new loans must not be longer than 45 days. " but more than 2 epoch's duration can be added
the max duration is not detected. add success when > next epoch
function testBasicPublicVaultLoan() public {
IAstariaRouter.LienDetails memory standardLien2 =
IAstariaRouter.LienDetails({
maxAmount: 50 ether,
rate: (uint256(1e16) * 150) / (365 days),
duration: 50 days, /****** more then 14 * 2 *******/
maxPotentialDebt: 50 ether
});
_commitToLien({
vault: publicVault,
strategist: strategistOne,
strategistPK: strategistOnePK,
tokenContract: tokenContract,
tokenId: tokenId,
lienDetails: standardLien2, /**** use standardLien2 ****/
amount: 10 ether,
isFirstLien: true
});
}Too long duration
https://github.com/sherlock-audit/2022-10-astaria/blob/main/src/VaultImplementation.sol#L209
Manual Review
PublicVault#_afterCommitToLien
function _afterCommitToLien(uint256 lienId, uint256 amount)
internal
virtual
override
{
// increment slope for the new lien
unchecked {
slope += LIEN_TOKEN().calculateSlope(lienId);
}
ILienToken.Lien memory lien = LIEN_TOKEN().getLien(lienId);
uint256 epoch = Math.ceilDiv(
lien.start + lien.duration - START(),
EPOCH_LENGTH()
) - 1;
+ require(epoch <= currentEpoch + 1,"epoch max <= currentEpoch + 1");
liensOpenForEpoch[epoch]++;
emit LienOpen(lienId, epoch);
}
Source: https://github.com/sherlock-audit/2022-10-astaria-judging/issues/258
__141345__
The lenders in principal have the claim for the loan collateral, but current rule will let the liquidation caller get the collateral for free. Effectively take advantage from the vault LP, which is not fair.
After the endAuction(), the collateral will be released to the initiator. Essentially, the initiator gets the NFT for free. But the lenders of the loan take the loss.
However, the lenders should have the claim to the collateral, since originally the funds are provided by the lenders. If the collateral at the end is owned by whoever calls the liquidation function, it is not fair for the lenders. And will discourage future users to use the protocol.
If there is no bidder, the winner will be assigned to the auction initiator. And the debts will all be wrote off. https://github.com/sherlock-audit/2022-10-astaria/blob/main/lib/astaria-gpl/src/AuctionHouse.sol#L178-L204
After the endAuction(), the collateral will be released to the initiator.
https://github.com/sherlock-audit/2022-10-astaria/blob/main/src/CollateralToken.sol#L341-L346
Manual Review
If there is no bidder for the auction, allow the NFT to get auctioned for another chance.
androolloyd
working as intended
141345
Escalate for 3 USDC
A borrower uses the NFT as collateral, the lender will get the collateral if the borrower defaults, that's how lending works normally. However, according to the current rule, anyone starts the liquidation process could potentially get the collateral, if no bidder bid on the auction. And the liquidator initiator already gets compensated by the initiator fee.
Current rule allows for a situation that 3rd user could gain the ownership of the NFT by calling liquidate(). But in common practice, it is the lender should claim the ownership of the collateral.
One step further, if by any chance, the initiator could start some DoS attack and make the protocol inoperable, this rule may become part of the attack, to get the collateral for free.
Although it is a corner case, I believe this is a business logic issue.
sherlock-admin
Escalate for 3 USDC
A borrower uses the NFT as collateral, the lender will get the collateral if the borrower defaults, that's how lending works normally. However, according to the current rule, anyone starts the liquidation process could potentially get the collateral, if no bidder bid on the auction. And the liquidator initiator already gets compensated by the initiator fee.
Current rule allows for a situation that 3rd user could gain the ownership of the NFT by calling
liquidate(). But in common practice, it is the lender should claim the ownership of the collateral.One step further, if by any chance, the initiator could start some DoS attack and make the protocol inoperable, this rule may become part of the attack, to get the collateral for free.
Although it is a corner case, I believe this is a business logic issue.
You've created a valid escalation for 3 USDC!
To remove the escalation from consideration: Delete your comment. To change the amount you've staked on this escalation: Edit your comment (do not create a new comment).
You may delete or edit your escalation comment anytime before the 48-hour escalation window closes. After that, the escalation becomes final.
Evert0x
Escalation accepted.
Will be rewarded a medium as it requires the auction to end with 0 bids
sherlock-admin
Escalation accepted.
Will be rewarded a medium as it requires the auction to end with 0 bids
This issue's escalations have been accepted!
Contestants' payouts and scores will be updated according to the changes made on this issue.
Source: https://github.com/sherlock-audit/2022-10-astaria-judging/issues/232
TurnipBoy
_makePayment(uint256, uint256) looping logic is inconsistent with how _deleteLienPosition manages the lien stack. _makePayment loops from 0 to openLiens.length but _deleteLienPosition (called when a lien is fully paid off) actively compresses the lien stack. When a payment pays off multiple liens the compressing effect causes an array OOB error towards the end of the loop.
function _makePayment(uint256 collateralId, uint256 totalCapitalAvailable)
internal
{
uint256[] memory openLiens = liens[collateralId];
uint256 paymentAmount = totalCapitalAvailable;
for (uint256 i = 0; i < openLiens.length; ++i) {
uint256 capitalSpent = _payment(
collateralId,
uint8(i),
paymentAmount,
address(msg.sender)
);
paymentAmount -= capitalSpent;
}
}LienToken.sol#_makePayment(uint256, uint256) loops from 0 to openLiens.Length. This loop attempts to make a payment to each lien calling _payment with the current index of the loop.
function _deleteLienPosition(uint256 collateralId, uint256 position) public {
uint256[] storage stack = liens[collateralId];
require(position < stack.length, "index out of bounds");
emit RemoveLien(
stack[position],
lienData[stack[position]].collateralId,
lienData[stack[position]].position
);
for (uint256 i = position; i < stack.length - 1; i++) {
stack[i] = stack[i + 1];
}
stack.pop();
}LienToken.sol#_deleteLienPosition is called on liens when they are fully paid off. The most interesting portion of the function is how the lien is removed from the stack. We can see that all liens above the lien in question are slid down the stack and the top is popped. This has the effect of reducing the total length of the array. This is where the logical inconsistency is. If the first lien is paid off, it will be removed and the formerly second lien will now occupy it's index. So then when _payment is called in the next loop with the next index it won't reference the second lien since the second lien is now in the first lien index.
Assuming there are 2 liens on some collateral. liens[0].amount = 100 and liens[1].amount = 50. A user wants to pay off their entire lien balance so they call _makePayment(uint256, uint256) with an amount of 150. On the first loop it calls _payment with an index of 0. This pays off liens[0]. _deleteLienPosition is called with index of 0 removing liens[0]. Because of the sliding logic in _deleteLienPosition lien[1] has now slid into the lien[0] position. On the second loop it calls _payment with an index of 1. When it tries to grab the data for the lien at that index it will revert due to OOB error because the array no long contains an index of 1.
Large payment are impossible and user must manually pay off each liens separately
https://github.com/sherlock-audit/2022-10-astaria/blob/main/src/LienToken.sol#L410-L424
Manual Review
Payment logic inside of AuctionHouse.sol works. _makePayment should be changed to mimic that logic.
IAmTurnipBoy
Escalate for 1 USDC
Not a dupe of #190, issue is with the lien array is managed as the payments are made. Fixing #190 wouldn't fix this.
sherlock-admin
Escalate for 1 USDC
Not a dupe of #190, issue is with the lien array is managed as the payments are made. Fixing #190 wouldn't fix this.
You've created a valid escalation for 1 USDC!
To remove the escalation from consideration: Delete your comment. To change the amount you've staked on this escalation: Edit your comment (do not create a new comment).
You may delete or edit your escalation comment anytime before the 48-hour escalation window closes. After that, the escalation becomes final.
Evert0x
Escalation accepeted
Source: https://github.com/sherlock-audit/2022-10-astaria-judging/issues/203
0xRajeev
The best-effort one-time English auction for borrower collateral is not economically efficient to drive auction bids towards reaching the total outstanding debt, which leads to loss of LP funds.
When any lien against a borrower collateral is not paid within the lien duration, the underlying collateral is put up for auction where bids can come in at any price. The borrower is allowed to cancel the auction if the current bid is lower than the reserve price which is set to the total outstanding debt. The reserve price is not enforced anywhere else. If there are no bids, the liquidator will receive the collateral.
This auction design of a best-effort one-time English auction is not economically efficient to drive auction bids towards reaching the total outstanding debt which effectively leads to loss of LP funds on unpaid liens.
Manual Review
Consider alternative auction design mechanisms e.g. a Dutch auction where the auction starts at the reserve price to provide a higher payment possibility to the LPs.
SantiagoGregory
We're switching to a Dutch auction through Seaport.
Evert0x
Downgrading to info as it's a protocol design choice.
secureum
Escalate for 2 USDC.
This finding is based on current protocol design and implementation (i.e. there was no documentation suggesting their future switch to Dutch auction). Based on the protocol team's response above, they effectively confirm the current design choice (and implementation) to be a serious enough issue that they are changing the protocol design to what is recommended by this finding. Just because it is a design issue does not deem this to be downgraded to informational — design drives implementation and is harder to change. Moving to a Dutch auction, as recommended, will affect significant parts of protocol implementation.
Therefore, we still think this is of Medium severity impact, if not higher.
cc @berndartmueller @lucyoa
sherlock-admin
Escalate for 2 USDC.
This finding is based on current protocol design and implementation (i.e. there was no documentation suggesting their future switch to Dutch auction). Based on the protocol team's response above, they effectively confirm the current design choice (and implementation) to be a serious enough issue that they are changing the protocol design to what is recommended by this finding. Just because it is a design issue does not deem this to be downgraded to informational — design drives implementation and is harder to change. Moving to a Dutch auction, as recommended, will affect significant parts of protocol implementation.
Therefore, we still think this is of Medium severity impact, if not higher.
cc @berndartmueller @lucyoa
You've created a valid escalation for 2 USDC!
To remove the escalation from consideration: Delete your comment. To change the amount you've staked on this escalation: Edit your comment (do not create a new comment).
You may delete or edit your escalation comment anytime before the 48-hour escalation window closes. After that, the escalation becomes final.
Evert0x
Escalation accepted based on comment from Watson
sherlock-admin
Escalation accepted based on comment from Watson
This issue's escalations have been accepted!
Contestants' payouts and scores will be updated according to the changes made on this issue.
Source: https://github.com/sherlock-audit/2022-10-astaria-judging/issues/201
bin2chen, Prefix, yixxas, 0xRajeev, minhquanym
Incorrect auction extension logic extends the auction by an additional amount of the previous duration instead of extending it by 15 minutes.
The calculation of newDuration incorrectly adds duration in the auction extension logic. This causes the new duration to be extended by an additional amount of the existing duration, instead of an additional 15 minutes (timeBuffer), when a bid is created in the last 15 mins of the existing auction duration.
Delayed payout of funds/collateral upon auction completion only after the newly extended duration.
Manual Review
Change the calculation to uint64 newDuration = uint256(block.timestamp + timeBuffer - firstBidTime).safeCastTo64();
AstariaRouter.commitToLiens will revert if the protocol fee is enabledSource: https://github.com/sherlock-audit/2022-10-astaria-judging/issues/195
0xRajeev
The function commitToLiens() will revert in getProtocolFee(), which prevents borrowers from depositing collateral and requesting loans in the protocol.
If the protocol fee is enabled by setting feeTo to a non-zero address, then getProtocolFee() will revert because of division-by-zero given that protocolFeeDenominator is 0 without any initialization and no setter (in file()) for setting it.
The function commitToLiens() will revert if the protocol fee is enabled thus preventing borrowers from depositing collateral and requesting loans in the protocol thereby failing to bootstrap its core NFT lending functionality.
Manual Review
Initialize protocol fee numerator and denominator in AstariaRouter and add their setters to file().
secureum
Escalate for 2 USDC.
We do not think this is a duplicate issue of #204. While both are about commitToLiens() reverting, the triggering locations, conditions and therefore the recommendations are entirely different. This issue is specific to non-initialization of protocol fee variables as described above.
cc @berndartmueller @lucyoa
sherlock-admin
Escalate for 2 USDC.
We do not think this is a duplicate issue of #204. While both are about
commitToLiens()reverting, the triggering locations, conditions and therefore the recommendations are entirely different. This issue is specific to non-initialization of protocol fee variables as described above.cc @berndartmueller @lucyoa
You've created a valid escalation for 2 USDC!
To remove the escalation from consideration: Delete your comment. To change the amount you've staked on this escalation: Edit your comment (do not create a new comment).
You may delete or edit your escalation comment anytime before the 48-hour escalation window closes. After that, the escalation becomes final.
Evert0x
Escalation accepted
sherlock-admin
Escalation accepted
This issue's escalations have been accepted!
Contestants' payouts and scores will be updated according to the changes made on this issue.
LienToken.createLien may prevent liens that satisfy their terms of maxPotentialDebtSource: https://github.com/sherlock-audit/2022-10-astaria-judging/issues/192
0xRajeev, hansfriese
The potentialDebt calculation in createLien is incorrect.
The calculated potentialDebt is effectively (impliedRate + totalDebt) * params.terms.duration because getImpliedRate() returns impliedRate = impliedRate.mulDivDown(1, totalDebt);. The calculated potentialDebt because of multiplying totalDebt by duration is significantly higher than it actually is and so will fail the params.terms.maxPotentialDebt check and revert.
This will cause DoS on valid lien creation.
Manual Review
Have getImpliedRate() not do impliedRate = impliedRate.mulDivDown(1, totalDebt); and calculate potentialDebt as totalDebt * (1 + impliedRate * params.terms.duration * mulDivDown(1, INTEREST_DENOMINATOR).
LienToken.changeInSlope calculation can lead to vault insolvencySource: https://github.com/sherlock-audit/2022-10-astaria-judging/issues/189
0xRajeev, hansfriese
The calculation of newSlope in changeInSlope() is incorrect which can lead to vault insolvency.
Contrary to the changeInSlope function, the calculateSlope function uses the _getOwed function to calculate the owed amount (incl. the interest). The interest is calculated with the _getInterest function. This function takes care of the lien.last and also if a lien is expired already. This logic is completely missing in the changeInSlope for the newSlope calculation which makes it incorrect. Also, very importantly, in the changeInSlope function, the INTEREST_DENOMINATOR is missing which makes the value inflated causing an underflow error in the last line of changeInSlope function: slope = oldSlope - newSlope;. oldslope, which accounts for the INTEREST_DENOMINATOR, is less than newSlope.
The incorrect changeInSlope() calculation can lead to reverts and vault insolvency because users cannot determine the implicit value of vaults while interacting with it as borrowers or lenders.
Manual Review
Make changeInSlope() consistent with calculateSlope() by implementing a separate helper function to calculate the interest accounting for all the parameters and reusing it in both places.
Source: https://github.com/sherlock-audit/2022-10-astaria-judging/issues/186
csanuragjain, chainNue, obront, peanuts, hansfriese, 0xRajeev
Auctions run for less time than intended causing them to be not economically efficient for the lenders, thus causing a suboptimal credit of liquidation funds to them.
From the documentation/comments, we can infer that firstBidTime is supposed to be // The time of the first bid and duration is supposed to be // The length of time to run the auction for, after the first bid was made.
However, when an auction is created in createAuction(), the auction's firstBidTime is incorrectly initialized as block.timestamp.safeCastTo64() instead of 0. This is premature initialization because the auction was only created here and no bid has been made yet via createBid(). The code in createBid() which check for firstBidTime == 0is more evidence that the initialization in createAuction() is incorrect.
This causes auctions to run for less time than intended if the first bid comes at a much later time after the auction was created. A shorter auction time could potentially allow fewer bids and cause it to be not economically efficient for the lenders, thus causing a suboptimal credit of liquidation funds to them.
Manual Review
createAuction() should initialize firstBidTime = 0.
androolloyd
will fix but its a convention issue we want auctions to start immediately not on first bid
Source: https://github.com/sherlock-audit/2022-10-astaria-judging/issues/174
0xRajeev, TurnipBoy
Assets can be deposited into public vaults by LPs with PublicVault.mint function to bypass a possible paused protocol.
The PublicVault contract prevents calls to the PublicVault.deposit function while the protocol is paused by using the whenNotPaused modifier.
The PublicVault contract extends the ERC4626Cloned contract, which has two functions to deposit assets into the vault: the deposit function and the mint function. The latter function, however, is not overwritten in the PublicVault contract and therefore lacks the appropriate whenNotPaused modifier.
LPs can deposit assets into public vaults with the PublicVault.mint function to bypass a possible paused protocol. This can lead to LP fund loss depending on the reason for the protocol pause and the incident response.
Manual Review
Override the mint function and add the whenNotPaused modifier
IAmTurnipBoy
Escalate for 1 USDC
Causes material loss of funds. Should be high risk
sherlock-admin
Escalate for 1 USDC
Causes material loss of funds. Should be high risk
You've created a valid escalation for 1 USDC!
To remove the escalation from consideration: Delete your comment. To change the amount you've staked on this escalation: Edit your comment (do not create a new comment).
You may delete or edit your escalation comment anytime before the 48-hour escalation window closes. After that, the escalation becomes final.
Evert0x
Escalation rejected.
This issue depends on the pausable state being active, also, it has the potential to lead to a loss of funds, it's not a guarantee.
sherlock-admin
Escalation rejected.
This issue depends on the pausable state being active, also, it has the potential to lead to a loss of funds, it's not a guarantee.
This issue's escalations have been rejected!
Contestants' payouts and scores will not be updated.
Auditors who escalated this issue will have their escalation amount deducted from future payouts.
Source: https://github.com/sherlock-audit/2022-10-astaria-judging/issues/171
0xRajeev
Liens whose duration is equal to (or maybe less than) minDurationIncrease cannot be bought out to be replaced by newer liens with lower interest rates but the same duration. This locks the borrower out of better-termed liens, effectively resulting in the loss of their funds
Liens whose duration is equal to (or maybe less than) minDurationIncrease cannot be bought out to be replaced by newer liens with lower interest rates but the exact duration because it results in an underflow in _getRemainingInterest().
Example scenario: if the strategyliendetails.duration is <= 14 days, then it's impossible to do a buyout of a new lien because the implemented check requires to wait minDurationIncrease, which is set to 14 days. However, if the buyer waits 14 days, the lien is expired, which triggers the earlier mentioned underflow.
The borrower gets locked out of better-termed liens, effectively resulting in the loss of their funds because of extra interest paid on older liens.
Manual Review
Revisit the checking logic and minimum duration as it applies to shorter-duration loans.
SantiagoGregory
We updated buyoutLien() to check for a lower interest rate or higher duration.
Source: https://github.com/sherlock-audit/2022-10-astaria-judging/issues/170
0xRajeev
Loan duration can exceed the end of the next epoch, which deviates from the protocol specification.
From the specs: "The duration of new loans is restricted to not exceed the end of the next epoch. For example, if a PublicVault is 15 days into a 30-day epoch, new loans must not be longer than 45 days."
However, there's no enforcement of this requirement.
The implementation does not adhere to the spec: Loan duration can exceed the end of the next epoch, which breaks protocol specification and therefore lead to miscalculations and potential fund loss.
Manual Review
Implement as per specification or revisit the specification.
Source: https://github.com/sherlock-audit/2022-10-astaria-judging/issues/143
ak1, 0xNazgul, joestakey, rvierdiiev, neila, pashov, Jeiwan, ctf_sec, __141345__
The first depositor of an ERC4626 vault can maliciously manipulate the share price by depositing the lowest possible amount (1 wei) of liquidity and then artificially inflating ERC4626.totalAssets.
This can inflate the base share price as high as 1:1e18 early on, which force all subsequence deposit to use this share price as a base and worst case, due to rounding down, if this malicious initial deposit front-run someone else depositing, this depositor will receive 0 shares and lost his deposited assets.
Given a vault with DAI as the underlying asset:
Alice (attacker) deposits initial liquidity of 1 wei DAI via deposit()
Alice receives 1e18 (1 wei) vault shares
Alice transfers 1 ether of DAI via transfer() to the vault to artificially inflate the asset balance without minting new shares. The asset balance is now 1 ether + 1 wei DAI -> vault share price is now very high (= 1000000000000000000001 wei ~ 1000 * 1e18)
Bob (victim) deposits 100 ether DAI
Bob receives 0 shares
Bob receives 0 shares due to a precision issue. His deposited funds are lost.
The shares are calculated as following
return supply == 0 ? assets : assets.mulDivDown(supply, totalAssets());
In case of a very high share price, due to totalAssets() > assets * supply, shares will be 0.
ERC4626 vault share price can be maliciously inflated on the initial deposit, leading to the next depositor losing assets due to precision issues.
Manual Review
This is a well-known issue, Uniswap and other protocols had similar issues when supply == 0.
For the first deposit, mint a fixed amount of shares, e.g. 10**decimals()
if (supply == 0) {
return 10**decimals;
} else {
return assets.mulDivDown(supply, totalAssets());
}Source: https://github.com/sherlock-audit/2022-10-astaria-judging/issues/141
neila
Insufficient access control for lending to PublicVault found by yawn-c111
Docs says as follows https://docs.astaria.xyz/docs/intro
Any strategists may provide their own capital to fund these loans through their own
PrivateVaults, and whitelisted strategists can deployPublicVaultsthat accept funds from other liquidity providers.
However, Liquidity Providers can also lend to PrivateVault.
This is because lendToVault function is controlled by mapping(address => address) public vaults, which are managed by _newVault function and include PrivateVaults
This leads to unexpected atttack.
Unexpected liquidity providers can lend to private vaults
https://github.com/unchain-dev/2022-10-astaria-UNCHAIN/blob/main/src/AstariaRouter.sol#L324
function lendToVault(IVault vault, uint256 amount) external whenNotPaused {
TRANSFER_PROXY.tokenTransferFrom(
address(WETH),
address(msg.sender),
address(this),
amount
);
require(
vaults[address(vault)] != address(0),
"lendToVault: vault doesn't exist"
);
WETH.safeApprove(address(vault), amount);
vault.deposit(amount, address(msg.sender));
}https://github.com/unchain-dev/2022-10-astaria-UNCHAIN/blob/main/src/AstariaRouter.sol#L500
function _newVault(
uint256 epochLength,
address delegate,
uint256 vaultFee
) internal returns (address) {
uint8 vaultType;
address implementation;
if (epochLength > uint256(0)) {
require(
epochLength >= minEpochLength && epochLength <= maxEpochLength,
"epochLength must be greater than or equal to MIN_EPOCH_LENGTH and less than MAX_EPOCH_LENGTH"
);
implementation = VAULT_IMPLEMENTATION;
vaultType = uint8(VaultType.PUBLIC);
} else {
implementation = SOLO_IMPLEMENTATION;
vaultType = uint8(VaultType.SOLO);
}
//immutable data
address vaultAddr = ClonesWithImmutableArgs.clone(
implementation,
abi.encodePacked(
address(msg.sender),
address(WETH),
address(COLLATERAL_TOKEN),
address(this),
address(COLLATERAL_TOKEN.AUCTION_HOUSE()),
block.timestamp,
epochLength,
vaultType,
vaultFee
)
);
//mutable data
VaultImplementation(vaultAddr).init(
VaultImplementation.InitParams(delegate)
);
vaults[vaultAddr] = msg.sender;
emit NewVault(msg.sender, vaultAddr);
return vaultAddr;
}Manual Review
create requirement to lend to only PublicVaults.
Source: https://github.com/sherlock-audit/2022-10-astaria-judging/issues/83
0x4141
LiquidationAccountant.claim may initiate a transfer with the amount 0, which reverts for some tokens.
Some tokens (e.g., LEND -> see https://github.com/d-xo/weird-erc20#revert-on-zero-value-transfers) revert when a transfer with amount 0 is initiated. This can happen within claim when the withdrawRatio is 100%.
In such a scenario, the funds are not claimable, leading to a loss of funds.
https://github.com/sherlock-audit/2022-10-astaria/blob/main/src/LiquidationAccountant.sol#L88
Manual Review
Do not initiate a transfer when the amount is zero.
Source: https://github.com/sherlock-audit/2022-10-astaria-judging/issues/73
rvierdiiev
LienToken._payment function increases users debt by setting lien.amount = _getOwed(lien)
LienToken._payment is used by LienToken.makePayment function that allows borrower to repay part or all his debt.
Also this function can be called by AuctionHouse when the lien is liquidated.
https://github.com/sherlock-audit/2022-10-astaria/blob/main/src/LienToken.sol#L594-L649
function _payment(
uint256 collateralId,
uint8 position,
uint256 paymentAmount,
address payer
) internal returns (uint256) {
if (paymentAmount == uint256(0)) {
return uint256(0);
}
uint256 lienId = liens[collateralId][position];
Lien storage lien = lienData[lienId];
uint256 end = (lien.start + lien.duration);
require(
block.timestamp < end || address(msg.sender) == address(AUCTION_HOUSE),
"cannot pay off an expired lien"
);
address lienOwner = ownerOf(lienId);
bool isPublicVault = IPublicVault(lienOwner).supportsInterface(
type(IPublicVault).interfaceId
);
lien.amount = _getOwed(lien);
address payee = getPayee(lienId);
if (isPublicVault) {
IPublicVault(lienOwner).beforePayment(lienId, paymentAmount);
}
if (lien.amount > paymentAmount) {
lien.amount -= paymentAmount;
lien.last = block.timestamp.safeCastTo32();
// slope does not need to be updated if paying off the rest, since we neutralize slope in beforePayment()
if (isPublicVault) {
IPublicVault(lienOwner).afterPayment(lienId);
}
} else {
if (isPublicVault && !AUCTION_HOUSE.auctionExists(collateralId)) {
// since the openLiens count is only positive when there are liens that haven't been paid off
// that should be liquidated, this lien should not be counted anymore
IPublicVault(lienOwner).decreaseEpochLienCount(
IPublicVault(lienOwner).getLienEpoch(end)
);
}
//delete liens
_deleteLienPosition(collateralId, position);
delete lienData[lienId]; //full delete
_burn(lienId);
}
TRANSFER_PROXY.tokenTransferFrom(WETH, payer, payee, paymentAmount);
emit Payment(lienId, paymentAmount);
return paymentAmount;
}The main problem is in line 617. lien.amount = _getOwed(lien);
https://github.com/sherlock-audit/2022-10-astaria/blob/main/src/LienToken.sol#L617
Here lien.amount becomes lien.amount + accrued interests, because _getOwed do that calculation.
lien.amount is the amount that user borrowed. So actually that line has just increased user's debt. And in case if he didn't pay all amount of lien, then next time he will pay more interests.
Example.
User borrows 1 eth. His lien.amount is 1eth.
Then he wants to repay some part(let's say 0.5 eth). Now his lien.amount becomes lien.amount + interests.
When he pays next time, he pays (lien.amount + interests) + new interests. So interests are acummulated on previous interests.
User borrowed amount increases and leads to lose of funds.
Provided above
Manual Review
Do not update lien.amount to _getOwed(lien).
Evert0x
Downgrading to medium
IAmTurnipBoy
Escalate for 1 USDC
Invalid. All I see here is compounding interest. Nothing incorrect here, just a design choice. Plenty of loans use compounding interest.
sherlock-admin
Escalate for 1 USDC
Invalid. All I see here is compounding interest. Nothing incorrect here, just a design choice. Plenty of loans use compounding interest.
You've created a valid escalation for 1 USDC!
To remove the escalation from consideration: Delete your comment. To change the amount you've staked on this escalation: Edit your comment (do not create a new comment).
You may delete or edit your escalation comment anytime before the 48-hour escalation window closes. After that, the escalation becomes final.
Evert0x
Escalation rejected
The impact of this is it'll cause a user in this situation to pay compound interest. Since loan is only 2 weeks, that will be very small impact. But the more often you make a payment the more interest you pay, a valid medium severity finding.
sherlock-admin
Escalation rejected
The impact of this is it'll cause a user in this situation to pay compound interest. Since loan is only 2 weeks, that will be very small impact. But the more often you make a payment the more interest you pay, a valid medium severity finding.
This issue's escalations have been rejected!
Watsons who escalated this issue will have their escalation amount deducted from their next payout.
Source: https://github.com/sherlock-audit/2022-10-astaria-judging/issues/66
rvierdiiev
LienToken.createLien doesn't check if user should be liquidated and provides new loan if auction do not exist for collateral.
LienToken.createLien relies on AuctionHouse to check if new loan can be added to borrower. It assumes that if auction doesn't exist then user is safe to take new loan.
The problem is that to start auction with token that didn't pay the debt someone should call AstariaRouter.liquidate function. If no one did it then auction for the NFT will not exists, and LienToken.createLien will create new Lien to user, while he already didn't pay debt and should be liquidated.
New loan will be paid to user that didn't repay previous lien.
Provided above
Manual Review
Check if user can be liquidated through all of his liens positions. If not then only proceed with new loan.
IAmTurnipBoy
Escalate for 1 USDC
Disagree with high severity. User would have to voluntarily sign a lien for another user who is already late on their payments (should raise some red flags), should only be medium or low given external circumstances
sherlock-admin
Escalate for 1 USDC
Disagree with high severity. User would have to voluntarily sign a lien for another user who is already late on their payments (should raise some red flags), should only be medium or low given external circumstances
You've created a valid escalation for 1 USDC!
To remove the escalation from consideration: Delete your comment. To change the amount you've staked on this escalation: Edit your comment (do not create a new comment).
You may delete or edit your escalation comment anytime before the 48-hour escalation window closes. After that, the escalation becomes final.
Evert0x
Escalation accepted.
sherlock-admin
Escalation accepted.
This issue's escalations have been accepted!
Contestants' payouts and scores will be updated according to the changes made on this issue.
Source: https://github.com/sherlock-audit/2022-10-astaria-judging/issues/59
cryptphi, rvierdiiev, HonorLt, ctf_sec
Strategist nonce is not checked while checking commitment. This makes impossible for strategist to cancel signed commitment.
VaultImplementation.commitToLien is created to give the ability to borrow from the vault. The conditions of loan are discussed off chain and owner or delegate of the vault then creates and signes deal details. Later borrower can provide it as IAstariaRouter.Commitment calldata params param to VaultImplementation.commitToLien.
After the checking of signer of commitment VaultImplementation._validateCommitment function calls AstariaRouter.validateCommitment.
function validateCommitment(IAstariaRouter.Commitment calldata commitment)
public
returns (bool valid, IAstariaRouter.LienDetails memory ld)
{
require(
commitment.lienRequest.strategy.deadline >= block.timestamp,
"deadline passed"
);
require(
strategyValidators[commitment.lienRequest.nlrType] != address(0),
"invalid strategy type"
);
bytes32 leaf;
(leaf, ld) = IStrategyValidator(
strategyValidators[commitment.lienRequest.nlrType]
).validateAndParse(
commitment.lienRequest,
COLLATERAL_TOKEN.ownerOf(
commitment.tokenContract.computeId(commitment.tokenId)
),
commitment.tokenContract,
commitment.tokenId
);
return (
MerkleProof.verifyCalldata(
commitment.lienRequest.merkle.proof,
commitment.lienRequest.merkle.root,
leaf
),
ld
);
}This function check additional params, one of which is commitment.lienRequest.strategy.deadline. But it doesn't check for the nonce of strategist here. But this nonce is used while signing.
Also AstariaRouter gives ability to increment nonce for strategist, but it is never called. That means that currently strategist use always same nonce and can't cancel his commitment.
Strategist can't cancel his commitment. User can use this commitment to borrow up to 5 times.
Provided above
Manual Review
Give ability to strategist to call increaseNonce function.
Source: https://github.com/sherlock-audit/2022-10-astaria-judging/issues/52
rvierdiiev, obront
If a collateral token owner approves another user as an operator for all their tokens (rather than just for a given token), the validation check in _validateCommitment() will fail.
The collateral token is implemented as an ERC721, which has two ways to approve another user:
approve())setApprovalForAll())However, when the _validateCommitment() function checks that the token is owned or approved by msg.sender, it does not accept those who are set as operators.
if (msg.sender != holder) {
require(msg.sender == operator, "invalid request");
}Approved operators of collateral tokens will be rejected from taking actions with those tokens.
https://github.com/sherlock-audit/2022-10-astaria/blob/main/src/VaultImplementation.sol#L152-L158
Manual Review
Include an additional check to confirm whether the msg.sender is approved as an operator on the token:
address holder = ERC721(COLLATERAL_TOKEN()).ownerOf(collateralId);
address approved = ERC721(COLLATERAL_TOKEN()).getApproved(collateralId);
address operator = ERC721(COLLATERAL_TOKEN()).isApprovedForAll(holder);
if (msg.sender != holder) {
require(msg.sender == operator || msg.sender == approved, "invalid request");
}Source: https://github.com/sherlock-audit/2022-10-astaria-judging/issues/50
hansfriese, 0xNazgul, obront, 0xRajeev
When a lien is liquidated, it calls timeToEpochEnd() to determine if a liquidation accountant should be deployed and we should adjust the protocol math to expect payment in a future epoch. Because of an error in the implementation, all liquidations that will pay out in the current epoch are set up as future epoch liquidations.
The liquidate() function performs the following check to determine if it should set up the liquidation to be paid out in a future epoch:
if (PublicVault(owner).timeToEpochEnd() <= COLLATERAL_TOKEN.auctionWindow())This check expects that timeToEpochEnd() will return the time until the epoch is over. However, the implementation gets this backwards:
function timeToEpochEnd() public view returns (uint256) {
uint256 epochEnd = START() + ((currentEpoch + 1) * EPOCH_LENGTH());
if (epochEnd >= block.timestamp) {
return uint256(0);
}
return block.timestamp - epochEnd;
}If epochEnd >= block.timestamp, that means that there IS remaining time in the epoch, and it should perform the calculation to return epochEnd - block.timestamp. In the opposite case, where epochEnd <= block.timestamp, it should return zero.
The result is that the function returns 0 for any epoch that isn't over. Since 0 < COLLATERAL_TOKEN.auctionWindow()), all liquidated liens will trigger a liquidation accountant and the rest of the accounting for future epoch withdrawals.
Accounting for a future epoch withdrawal causes a number of inconsistencies in the protocol's math, the impact of which vary depending on the situation. As a few examples:
decreaseEpochLienCount(). This has the effect of artificially lowering the number of liens in the epoch, which will cause the final liens paid off in the epoch to revert (and will let us process the epoch earlier than intended).increaseLiquidationsExpectedAtBoundary(), which can throw off the math when processing the epoch.https://github.com/sherlock-audit/2022-10-astaria/blob/main/src/PublicVault.sol#L562-L570
https://github.com/sherlock-audit/2022-10-astaria/blob/main/src/AstariaRouter.sol#L388-L415
Manual Review
Fix the timeToEpochEnd() function so it calculates the remaining time properly:
function timeToEpochEnd() public view returns (uint256) {
uint256 epochEnd = START() + ((currentEpoch + 1) * EPOCH_LENGTH());
if (epochEnd <= block.timestamp) {
return uint256(0);
}
return epochEnd - block.timestamp; //
}Source: https://github.com/sherlock-audit/2022-10-astaria-judging/issues/33
0x0
Arithmetic operations are performed with the assumption that the token always has 18 decimals.
Arithmetic operations assume the token has 18 decimals. Not all tokens use 18 decimals, such as Tether.
uint256 transferAmount = withdrawRatio.mulDivDown(balance, 1e18);Manual Review
Source: https://github.com/sherlock-audit/2022-10-astaria-judging/issues/28
bin2chen, ak1, obront, tives, rvierdiiev, 0xRajeev, __141345__, hansfriese
The _payment() function sends the full paymentAmount argument to the lien owner, which both (a) overpays lien owners if borrowers accidentally overpay and (b) sends the first lien owner all the funds for the entire loop of a borrower is intending to pay back multiple loans.
There are two makePayment() functions in LienToken.sol. One that allows the user to specific a position (which specific lien they want to pay back, and another that iterates through their liens, paying each back.
In both cases, the functions call out to _payment() with a paymentAmount, which is sent (in full) to the lien owner.
TRANSFER_PROXY.tokenTransferFrom(WETH, payer, payee, paymentAmount);This behavior can cause problems in both cases.
The first case is less severe: If the user is intending to pay off one lien, and they enter a paymentAmount greater than the amount owed, the function will send the full paymentAmount to the lien owner, rather than just sending the amount owed.
The second case is much more severe: If the user is intending to pay towards all their loans, the _makePayment() function loops through open liens and performs the following:
uint256 paymentAmount = totalCapitalAvailable;
for (uint256 i = 0; i < openLiens.length; ++i) {
uint256 capitalSpent = _payment(
collateralId,
uint8(i),
paymentAmount,
address(msg.sender)
);
paymentAmount -= capitalSpent;
}The _payment() function is called with the first lien with paymentAmount set to the full amount sent to the function. The result is that this full amount is sent to the first lien holder, which could greatly exceed the amount they are owed.
A user who is intending to pay off all their loans will end up paying all the funds they offered, but only paying off their first lien, potentially losing a large amount of funds.
The _payment() function with the core error:
https://github.com/sherlock-audit/2022-10-astaria/blob/main/src/LienToken.sol#L594-L649
The _makePayment() function that uses _payment() and would cause the most severe harm:
https://github.com/sherlock-audit/2022-10-astaria/blob/main/src/LienToken.sol#L410-L424
Manual Review
In _payment(), if lien.amount < paymentAmount, set paymentAmount = lien.amount.
The result will be that, in this case, only lien.amount is transferred to the lien owner, and this value is also returned from the function to accurately represent the amount that was paid.
Source: https://github.com/sherlock-audit/2022-10-astaria-judging/issues/25
obront, sorrynotsorry
The _getInterest() function takes a timestamp as input. However, in a crucial check in the function, it uses block.timestamp instead. The result is that other functions expecting accurate interest amounts will receive incorrect values.
The _getInterest() function takes a lien and a timestamp as input. The intention is for it to calculate the amount of time that has passed in the lien (delta_t) and multiply this value by the rate and the amount to get the interest generated by this timestamp.
However, the function uses the following check regarding the timestamp:
if (block.timestamp >= lien.start + lien.duration) {
delta_t = uint256(lien.start + lien.duration - lien.last);
} Because this check uses block.timestamp before returning the maximum interest payment, the function will incorrectly determine which path to take, and return an incorrect interest value.
There are two negative consequences that can come from this miscalculation:
block.timestamp >= lien.start + lien.duration) to check an interest amount from a timestamp during the lien, it will incorrectly return the maximum interest valuedelta_t = uint256(timestamp.safeCastTo32() - lien.last);)This _getInterest() function is used in many crucial protocol functions (_getOwed(), calculateSlope(), changeInSlope(), getTotalDebtForCollateralToken()), so these incorrect values can have surprising and unexpected negative impacts on the protocol.
https://github.com/sherlock-audit/2022-10-astaria/blob/main/src/LienToken.sol#L177-L196
Manual Review
Change block.timestamp to timestamp so that the if statement checks correctly.
Source: https://github.com/sherlock-audit/2022-10-astaria-judging/issues/19
zzykxx, obront
VAULT_FEE() uses an incorrect offset, returning a number ~1e16X greater than intended, providing strategists with unlimited access to drain all vault funds.
When using ClonesWithImmutableArgs, offset values are set so that functions representing variables can retrieve the correct values from storage.
In the ERC4626-Cloned.sol implementation, VAULT_TYPE() is given an offset of 172. However, the value before it is a uint8 at the offset 164. Since a uint8 takes only 1 byte of space, VAULT_TYPE() should have an offset of 165.
I put together a POC to grab the value of VAULT_FEE() in the test setup:
function testVaultFeeIncorrectlySet() public {
Dummy721 nft = new Dummy721();
address tokenContract = address(nft);
uint256 tokenId = uint256(1);
address publicVault = _createPublicVault({
strategist: strategistOne,
delegate: strategistTwo,
epochLength: 14 days
});
uint fee = PublicVault(publicVault).VAULT_FEE();
console.log(fee)
assert(fee == 5000); // 5000 is the value that was meant to be set
}In this case, the value returned is > 3e20.
This is a highly critical bug. VAULT_FEE() is used in _handleStrategistInterestReward() to determine the amount of tokens that should be allocated to strategistUnclaimedShares.
if (VAULT_FEE() != uint256(0)) {
uint256 interestOwing = LIEN_TOKEN().getInterest(lienId);
uint256 x = (amount > interestOwing) ? interestOwing : amount;
uint256 fee = x.mulDivDown(VAULT_FEE(), 1000); //VAULT_FEE is a basis point
strategistUnclaimedShares += convertToShares(fee);
}The result is that strategistUnclaimedShares will be billions of times higher than the total interest generated, essentially giving strategist access to withdraw all funds from their vaults at any time.
https://github.com/sherlock-audit/2022-10-astaria/blob/main/src/PublicVault.sol#L518-L523
Manual Review
Set the offset for VAULT_FEE() to 165. I tested this value in the POC I created and it correctly returned the value of 5000.
Source: https://github.com/sherlock-audit/2022-10-astaria-judging/issues/18
csanuragjain, TurnipBoy, obront, rvierdiiev, peanuts, neila, 0xRajeev, Prefix, Jeiwan, yixxas, hansfriese, minhquanym, 0x4141
The auction mechanism is intended to watch for bids within timeBuffer of the end of the auction, and automatically increase the remaining duration to timeBuffer if such a bid comes in.
There is an error in the implementation that causes all bids within timeBuffer of the end of a max duration auction to revert, effectively ending the auction early and cutting off bidders who intended to wait until the end.
In the createBid() function in AuctionHouse.sol, the function checks if a bid is within the final timeBuffer of the auction:
if (firstBidTime + duration - block.timestamp < timeBuffer)If so, it sets newDuration to equal the amount that will extend the auction to timeBuffer from now:
uint64 newDuration = uint256( duration + (block.timestamp + timeBuffer - firstBidTime) ).safeCastTo64();If this newDuration doesn't extend beyond the maxDuration, this works great. However, if it does extend beyond maxDuration, the following code is used to update duration:
auctions[tokenId].duration = auctions[tokenId].maxDuration - firstBidTime;This code is incorrect. maxDuration will be a duration for the contest (currently set to 3 days), whereas firstTimeBid is a timestamp for the start of the auction (current timestamps are > 1 billion).
Subtracting firstTimeBid from maxDuration will underflow, which will revert the function.
Manual Review
Change this assignment to simply assign duration to maxDuration, as follows:
auctions[tokenId].duration = auctions[tokenId].maxDurationSource: https://github.com/sherlock-audit/2022-10-astaria-judging/issues/3
Bnke0x0, w42d3n, pashov
The ERC4626-Cloned.deposit/mint functions do not work well with fee-on-transfer tokens as the assets variable is the pre-fee amount, including the fee, whereas the totalAssets do not include the fee anymore.
This can be abused to mint more shares than desired.
' function deposit(uint256 assets, address receiver)
public
virtual
override(IVault)
returns (uint256 shares)
{
// Check for rounding error since we round down in previewDeposit.
require((shares = previewDeposit(assets)) != 0, "ZERO_SHARES");
// Need to transfer before minting or ERC777s could reenter.
ERC20(underlying()).safeTransferFrom(msg.sender, address(this), assets);
_mint(receiver, shares);
emit Deposit(msg.sender, receiver, assets, shares);
afterDeposit(assets, shares);
}' `ERC20(underlying()).safeTransferFrom(msg.sender, address(this), assets);`A deposit(1000) should result in the same shares as two deposits of deposit(500) but it does not because assets is the pre-fee amount.
Assume a fee-on-transfer of 20%. Assume current totalAmount = 1000, totalShares = 1000 for simplicity.
deposit(1000) = 1000 / totalAmount * totalShares = 1000 shares.
deposit(500) = 500 / totalAmount * totalShares = 500 shares. Now the totalShares increased by 500 but the totalAssets only increased by (100% - 20%) * 500 = 400. Therefore, the second deposit(500) = 500 / (totalAmount + 400) * (newTotalShares) = 500 / (1400) * 1500 = 535.714285714 shares.
In total, the two deposits lead to 35 more shares than a single deposit of the sum of the deposits.
Manual Review
assets should be the amount excluding the fee, i.e., the amount the contract actually received.
This can be done by subtracting the pre-contract balance from the post-contract balance.
However, this would create another issue with ERC777 tokens.
Maybe previewDeposit should be overwritten by vaults supporting fee-on-transfer tokens to predict the post-fee amount. And do the shares computation on that, but then the afterDeposit is still called with the original assetsand implementers need to be aware of this.
androolloyd
we will not be supporting fee on transfer tokens at the time