sherlock-admin2 commented 7 months ago

hyh

high

Rewards can be stolen from other proposals and votes by extending auction revenue period with the help of bogus proposals

Summary

Attacker controlling more than proposalThresholdBPS share of votes can systematically create auxiliary proposals to capture a bigger period of auction revenue as rewards for ones they are affiliated with.

It makes sense for a beneficiary of the currently ended proposal 0 to front-run any next valid proposal creation with a dummy proposal, so it will be an anchor last proposal for updateRewardsForProposalWritingAndVoting(anchorProposalId, ...) call so all the auction revenue before the next valid proposal creation moment will go to proposal 0. This will systematically increase their rewards at the expense of other client id owners, i.e. stealing from them.

Vulnerability Detail

The auction reward period can be artificially extended at the expense of other eligible proposals' rewards with the help of automated creation of bogus proposals. That's possible as the auction revenue period is derived from the user supplied last proposal id before proposal eligibility control. I.e. an attacker can front run next valid proposal creation with the creation of the fake proposal in order to maximize the auction revenue period for the proposal they have interest in (i.e. have any affiliation with client id owner of proposal or its votes).

Fake proposal doesn't have to be eligible: it can be heavily voted against, it can be vetoed. The only requirement is that it should have ended (require(block.number > endBlock, ...) as of the time of the reward allocation call. Note that objectionPeriodEndBlock is activated only for flipped proposals, which is highly unlikely for bogus ones, so endBlock = max(proposals[i].endBlock, proposals[i].objectionPeriodEndBlock) = proposals[i].endBlock, which is known as of time of proposal creation. In general it should be as late as possible, for example front running the next rival reward allocation proposal.

Schematic POC: 1) suppose a proposal 1 that will drive much attention and votes is known to be published 2) attacker is affiliated to a beneficiary of a clientId used in creation of some already passed or almost passed not so popular proposal 0 3) attacker front runs proposal 1 creation with creation of its own proposal 2. Both have the same creation time 4) attacker runs updateRewardsForProposalWritingAndVoting(proposal_2_id, ids) with ids covering all the activity of the yet uncovered period 5) Their clientId is getting all the auction revenue prior to creation_time undiluted by votes of proposal 1 (which can be expected to be larger than ones of proposal 0)

Attacker has effectively stolen rewards from client id owner of proposal 1 proposal and votes.

Impact

The attack can be carried deterministically, there are no direct prerequisites, while the total cost is additional proposal creation gas only. The probability of the mentioned setup occurring so the attack will make sense can be estimated as medium as the situation is pretty typical. Auction period manipulation has substantial impact on rewards calculation, so the material impact on the other client id owners' reward revenue is high.

Likelihood: Medium + Impact: High = Severity: High.

Code Snippet

The revenue allocated for reward distribution is based on the creation time of the t.lastProposal:

https://github.com/sherlock-audit/2024-03-nouns-dao-2/blob/main/nouns-monorepo/packages/nouns-contracts/contracts/client-incentives/Rewards.sol#L332-L342

        t.firstAuctionIdForRevenue = $.nextProposalRewardFirstAuctionId;
        (uint256 auctionRevenue, uint256 lastAuctionIdForRevenue) = getAuctionRevenue({
            firstNounId: t.firstAuctionIdForRevenue,
>>          endTimestamp: t.lastProposal.creationTimestamp
        });
        $.nextProposalRewardFirstAuctionId = uint32(lastAuctionIdForRevenue) + 1;

        require(auctionRevenue > 0, 'auctionRevenue must be > 0');

        t.proposalRewardForPeriod = (auctionRevenue * $.params.proposalRewardBps) / 10_000;
        t.votingRewardForPeriod = (auctionRevenue * $.params.votingRewardBps) / 10_000;

Which is user supplied lastProposalId as proposalDataForRewards doesn't filter the proposals, returning all the results sequentially:

https://github.com/sherlock-audit/2024-03-nouns-dao-2/blob/main/nouns-monorepo/packages/nouns-contracts/contracts/client-incentives/Rewards.sol#L319-L330

        require(lastProposalId <= nounsDAO.proposalCount(), 'bad lastProposalId');
        require(lastProposalId >= t.nextProposalIdToReward, 'bad lastProposalId');
        require(isSortedAndNoDuplicates(votingClientIds), 'must be sorted & unique');

>>      NounsDAOTypes.ProposalForRewards[] memory proposals = nounsDAO.proposalDataForRewards(
            t.nextProposalIdToReward,
>>          lastProposalId,
            votingClientIds
        );
        $.nextProposalIdToReward = lastProposalId + 1;

>>      t.lastProposal = proposals[proposals.length - 1];

I.e. here t.lastProposal doesn't have to be eligible, it can be any fake proposal which will be omitted later on in the logic. But since t.lastProposal.creationTimestamp is used for getAuctionRevenue() it will have an impact of capturing a bigger share of auction profits:

https://github.com/sherlock-audit/2024-03-nouns-dao-2/blob/main/nouns-monorepo/packages/nouns-contracts/contracts/client-incentives/Rewards.sol#L531-L542

    function getAuctionRevenue(
        uint256 firstNounId,
        uint256 endTimestamp
    ) public view returns (uint256 sumRevenue, uint256 lastAuctionId) {
        INounsAuctionHouseV2.Settlement[] memory s = auctionHouse.getSettlementsFromIdtoTimestamp(
            firstNounId,
>>          endTimestamp,
            true
        );
        sumRevenue = sumAuctions(s);
        lastAuctionId = s[s.length - 1].nounId;
    }

Allowing the attacker to maximize the auction revenue period for rewards calculation as the whole period from startId to maxId will be used, while endTimestamp was manipulated to be close to the current moment:

https://github.com/sherlock-audit/2024-03-nouns-dao-2/blob/main/nouns-monorepo/packages/nouns-contracts/contracts/NounsAuctionHouseV2.sol#L452-L480

    function getSettlementsFromIdtoTimestamp(
        ...
    ) public view returns (Settlement[] memory settlements) {
        uint256 maxId = auctionStorage.nounId;
        require(startId <= maxId, 'startId too large');
        settlements = new Settlement[](maxId - startId + 1);
        uint256 actualCount = 0;
        SettlementState memory settlementState;
>>      for (uint256 id = startId; id <= maxId; ++id) {
            settlementState = settlementHistory[id];

            if (skipEmptyValues && settlementState.blockTimestamp <= 1) continue;

            // don't include the currently auctioned noun if it hasn't settled
            if ((id == maxId) && (settlementState.blockTimestamp <= 1)) continue;

>>          if (settlementState.blockTimestamp > endTimestamp) break;

            settlements[actualCount] = Settlement({
                blockTimestamp: settlementState.blockTimestamp,
                amount: uint64PriceToUint256(settlementState.amount),
                winner: settlementState.winner,
                nounId: id,
                clientId: settlementState.clientId
            });
            ++actualCount;
        }

Tool used

Manual Review

Recommendation

Consider basing the check on the last eligible proposal by filtering them out in proposalDataForRewards(), which isn't used outside reward logic:

https://github.com/sherlock-audit/2024-03-nouns-dao-2/blob/main/nouns-monorepo/packages/nouns-contracts/contracts/governance/NounsDAOProposals.sol#L719-L737

    function proposalDataForRewards(
        NounsDAOTypes.Storage storage ds,
        uint256 firstProposalId,
        uint256 lastProposalId,
+       uint16 proposalEligibilityQuorumBps,
        uint32[] calldata votingClientIds
    ) internal view returns (NounsDAOTypes.ProposalForRewards[] memory) {
        require(lastProposalId >= firstProposalId, 'lastProposalId >= firstProposalId');
        uint256 numProposals = lastProposalId - firstProposalId + 1;
        NounsDAOTypes.ProposalForRewards[] memory data = new NounsDAOTypes.ProposalForRewards[](numProposals);

        NounsDAOTypes.Proposal storage proposal;
        uint256 i;
        for (uint256 pid = firstProposalId; pid <= lastProposalId; ++pid) {
            proposal = ds._proposals[pid];
+           if (proposal.canceled || proposals.forVotes < (proposals.totalSupply * proposalEligibilityQuorumBps) / 10_000) continue;

            NounsDAOTypes.ClientVoteData[] memory c = new NounsDAOTypes.ClientVoteData[](votingClientIds.length);
            for (uint256 j; j < votingClientIds.length; ++j) {
                c[j] = proposal.voteClients[votingClientIds[j]];
            }

In order to minimize the changes this suggestion is shared with another issues.

This way getAuctionRevenue() will be called with the endTimestamp based on the last proposal after filtration.

eladmallel commented 7 months ago

We are working on a fix that will better filter out proposals.

We think severity should be lowered to Medium; still thinking the term "loss of funds" is not quite appropriate here, but this criterion for medium issues from Sherlock docs seems relevant:

Causes a loss of funds but requires certain external conditions or specific states, or a loss is highly constrained. The losses must exceed small, finite amount of funds, and any amount relevant based on the precision or significance of the loss.

While the definitions for high seem less fitting to this issue.

WangSecurity commented 7 months ago

I believe here the High severity is appropriate since the prerequisites and conditions are not external and can be created by the attacker.

WangSecurity commented 7 months ago

Comment from LSW:

"the attacker only need to be able to surpass proposalThresholdBPS, which is and will be kept low enough to facilitate governance participation. There are no other ifs here, one can manipulate the auction span straightforwardly."

Under these reasons the report remains High seveirty.

sherlock-admin4 commented 7 months ago

The protocol team fixed this issue in the following PRs/commits: https://github.com/nounsDAO/nouns-monorepo/pull/839

dmitriia commented 7 months ago

Fix looks ok: bogus proposals are now filtered out in proposalDataForRewards() as eligibility criteria was moved there.