Skip to content

Latest commit

 

History

History
397 lines (310 loc) · 38.4 KB

README.md

File metadata and controls

397 lines (310 loc) · 38.4 KB
Raw

Overview & Setup

git clone https://github.com/alexbabits/damn-vulnerable-defi-ctfs
  • Dependencies (OpenZeppelin, Solmate, Solady, safe-contracts (gnosis), OpenZeppelin upgradeable contracts):
forge install OpenZeppelin/openzeppelin-contracts
forge install transmissions11/solmate
forge install Vectorized/solady
forge install safe-global/safe-contracts
forge install OpenZeppelin/openzeppelin-contracts-upgradeable
  • Run the tests to complete challenges:
forge test --match-path test/CONTRACT_NAME_HERE.t.sol -vv
  • Challenges #5 and #6 not working because OpenZeppelin recently removed ERC20Snapshot.sol on Oct 5th, 2023. I may be able to fix it if I somehow get an earlier version of OZ.
  • Challenges #13, #14, #15 do not have a pre-made foundry test template, so if I solve these I will need to create and adapt the hardhat template from scatch.

#1 Unstoppable

  • Description/Goal: Make the vault stop offering flash loans by making flashLoan always revert.

  • Resources: https://twitter.com/bytes032/status/1631235276033990657 & https://gist.github.com/bytes032/68de03834881a41afa1d2d2f7b310d15

  • Topics: Flashloans (ERC-3156) & Vaults (ERC-4626)

  • Methodology: 0. Background: - ERC3156 flash loans invoke onFlashLoan callback after giving the receiver their token loan - supply = Number of virtual shares in ERC4626 vault. The virtual shares are an accounting representation of the deposited tokens. - assets = Actual tokens in ERC4626 vault.

    1. In our test, hack contract (us) first deposits 69 tokens to the vault: vault.deposit(2 ether, address(unstoppableHack));. Everything works properly for now.

      • totalAssets = 1_000_069
      • totalSupply = 1_000_069

    2. We then call attack passing in 1 token.

      • This calls flashLoan on the vault, referencing the hack contract for 1 token. All the checks pass because we haven't done anything naughty yet. Then this line: ERC20(_token).safeTransfer(address(receiver), amount); gives the hack contract 1 token as a loan. This immediately invokes the onFlashLoan callback function, directed at our hack contract.
      • Inside our onFlashLoan, which specifies what we are doing with our loan, we call the ERC4626.sol withdraw function on the vault for 1 token. The first thing withdraw does is call ERC4626.sol previewWithdraw, which is a confusing function.
        • previewWithdraw given an amount of assets, can calculate a number. It calculates that number based on supply, totalAssets, and assets. In the context of previewWithdraw being called within withdraw, it's calculating the number of virtual shares to be burned when withdrawing an amount of tokens. It calculates the number with this: assets.mulDivUp(supply, totalAssets());.
        • mulDivUp is an assembly math function in FixedPointMathLib.sol. It takes in x, y, and denominator and returns z.
        • In context of withdraw function: x = assets, y = supply, denominator = total assets, z = shares to be burned.
        • Breaking down the equation: z := add(gt(mod(mul(x, y), denominator), 0), div(mul(x, y), denominator)). We do x * y / denominator, and if remainder is greater than 0, that portion returns 1, so we add 1, otherwise add 0.
        • General idea: shares to be burned = (assets * total supply / total assets) + (0 if no remainder, 1 if remainder)
        • assets = x = 1 (we request 1 token for withdraw in the onFlashLoan function. Not to be confused with the 1 token received from the vault loan already).
        • supply = y = 1_000_069 (none of the virtual shares have been burned yet).
        • total assets = denominator = 1_000_068 (vault has given us one of our tokens, but withdraw function has NOT YET given us the 1 token we requested. The withdraw function calculates the shares BEFORE it does anything else, like burn the virtual shares or send the receiver their tokens. That's why previewWithdraw function is using the 1_000_068 value).
        • shares to be burned = 1 * 1_000_069 / 1_000_068 + (0 or 1) = 1.000001 + 1 = 2.000001. 2 shares will be burned when we withdraw 1 token.
        • The withdraw finishes and vault takes their loan back: ERC20(_token).safeTransferFrom(address(receiver), address(this), amount + fee);
      • State initial: totalAssets = 1_000_000, totalSupply = 1_000_000
      • State after deposit: totalAssets = 1_000_069, totalSupply = 1_000_069
      • State after vault sends us 1 token loan: totalAssets = 1_000_068, totalSupply = 1_000_069 (sends us 1 token loan).
      • State after withdraw finishes in onFlashLoan: totalAssets = 1_000_067, totalSupply = 1_000_067 (sends us our 1 requested token, and burns the 2 shares).
      • State after vault takes their loan back: totalAssets = 1_000_068, totalSupply = 1_000_067. Notice the mismatch now.
      • This causes flashLoan to always revert whenever a user wants to make a flash loan because we have messed up the accounting of the vault. This check in the vault's flashLoan now always reverts because the total supply is not equal to the balance before, which is the total assets: if (convertToShares(totalSupply) != balanceBefore) revert InvalidBalance();.

    3. Synopsis & Lessons:

      • Calling withdraw for a vault during the onFlashLoan fallback messes up the accounting in this scenario. The flashloan gives you the tokens, which decrements totalAssets. Then withdraw calculates the number of virtual shares to burn based on that decremented totalAssets. Because totalAsset doesn't equal totalSupply at this point in time, the mulDivUp calculates a remainder, which causes the formula to add extra 1 to the shares to burn. So it burns an extra share, gives you your withdraw token, and then the vault takes back it's loan. This is why in the final state we see there is 1 less virtual share in the accounting.

winner

#2 Naive Receiver

  • Description/Goal: Drain Naive Receiver's contract balance of 10 ETH in a single transaction. He has a contract setup that can call onFlashLoan for the pool. The pool has 1 ether fee per flash loan.
  • Resources: https://www.youtube.com/watch?v=2tFlcH5k-jk, https://github.com/zach030/damnvulnerabledefi-foundry
  • Topics: Flashloans (ERC-3156)
  • Methodology:
    • FlashLoanReceiver ignored the first parameter in the onFlashLoan which is the initiator, which is the msg.sender of flashLoan.
    • Because his onFlashLoan function is external and missing the variable associated with the initiator, we can create a contract that calls flashLoan on his behalf, loan 0 tokens, and just flashLoan 10 times, which will cost him 10 ETH and successfully drain his contract.
    • This can all happen in 1 transaction because upon deployment the Attacker contract must finish the loop and package that all into one transaction upon deployment.
    • It doesn't even matter what we do in onFlashLoan, all we wanted to do is drain his account from the absurd flash loan fees.
    • Anyone can initiate a flash loan on behalf of this guy (FlashLoanReceiver.sol).

winner

#3 Truster

  • Description/Goal: Take all the tokens out of the pool, in a single transaction if possible. A pool offering flash loans of DVT tokens for free. You have nothing.
  • Resources: https://github.com/zach030/damnvulnerabledefi-foundry
  • Topics: Flashloans (ERC-3156), tokens (ERC-20)
  • Methodology:
    • flashLoan has no safety checks on the data or target parameters, AND it makes a functionCall from OZ's Address.sol library, which is an external call where you can pass in a target address and data. So we can make a flash loan and pass in any arbitrary contract address and data to be executed for that contract address. (hint: You can pass in functions as data).
    • Attacker calls flashLoan requesting a loan of 0 tokens for ourself, passing in any malicious data we want. We decide to use abi.encodeWithSignature to package the erc20 approve function into parsable bytes data, with our address as the spender and the allowance amount being all the tokens in the pool.
    • The flashLoan in the pool then calls target.functionCall(data); where the target is the token address of the pool tokens and the function call data is the encoded approve function for ERC20 tokens. functionCall returns functionCallWithValue(target, data, 0);, and the functionCallWithValue is what actually makes the call with our data: (bool success, bytes memory returndata) = target.call{value: value}(data);. - The target is the token, and the function makes a low level call with our encoded approve function bytes data where we specified our address as the spender and the allowance as the entire pools balance.
    • The approve function thus gets executed, approving our hack contract as the spender to be able to spend all the pools tokens. We can then transferFrom all of the pools tokens from the pool to us.

winner

#4 Side Entrance

  • Description/Goal: Pool with 1000 ETH allows for deposits and withdraw of ETH with no fee flash loans. Starting with 1 ETH in balance, pass the challenge by taking all ETH from the pool.
  • Resources: https://github.com/zach030/damnvulnerabledefi-foundry
  • Topics: Flashloans (ERC-3156)
  • Methodology:
    • Attackers balance before is 0 ETH. Player's balance is 1 ETH. Player deploys the hack contract, attacker calls attack with 1000 ETH. This starts a flashLoan for 1000 ETH with no fees.
    • Then the pool does IFlashLoanEtherReceiver(msg.sender).execute{value: amount}(); which calls execute on the hackers contract.
    • Our execute function calls deposit on the pool which deposits the 1000 borrowed ETH back into the pool. This increases our tracked mapping balance as seen by the pool to now be 1000 ETH deposited, which can be withdrawn later.
    • Importantly, there is no pull action to get that 1000 ETH loan back at the end of the flash loan, all it does is check if (address(this).balance < balanceBefore) to complete the flash loan. This check passes because ( 1000 < 1000 ) is false. The pool's balance before was 1000 ETH, and after the flash loan it's real physical balance is 1000 ETH as well. This check is "supposed" to verify that the pool didn't lose money, which it technically did not at this state. However, the issue is that we manipulated the mapping of our virtual balance to be 1000 ETH during the flash loan by calling deposit and giving the pool our borrowed ETH.
    • Now the pool has it's 1000 ETH again and our hack contract has no ETH. But because we deposited that loaned ETH, the contract's balance mapping thinks we are eligible to claim that 1000 ETH.
    • We can call withdraw to withdraw 1000 ETH from the pool. Our receive function forwards that 1000 ETH straight to the owner (player).

winner

#5 The Rewarder

  • Preface: CURRENTLY NOT WORKING OpenZeppelin currently broke this challenge with the release of v5.0.0 because there is no longer ERC20snapshot.sol for the AccountingToken. I tried using v4.9.3 of OZ which has ERC20snapshot.sol, and also grabbing its imports, but it's all messed up. This writeup will just be explaining and understanding the hack without the solution contracts properly running.
  • Description/Goal: A pool is offering flash loans of DVT tokens. And there's another pool offering rewards in tokens every 5 days for people who deposit their DVT tokens into it. Alice, Bob, Charlie, and David already deposited some DVT tokens and have won their rewards. You have no DVT, but int he upcoming reward round, you must claim most of the rewards for yourself.
  • Resources: https://www.youtube.com/watch?v=zT5uNbGPaJ4, https://github.com/zach030/damnvulnerabledefi-foundry
  • Topics: Flashloans (ERC-3156)
  • Methodology:
    • Request a flashLoan for all the DVT tokens from flashLoanPool. This gives us msg.sender the tokens (us), and then calls msg.sender.functionCall(abi.encodeWithSignature("receiveFlashLoan(uint256)", amount));. So we are actually going to be calling receiveFlashLoan for the entire amount of DVT tokens that flashLoanPool has. We currently have all the DVT loaned to us.
    • Inside our attack we have the receiveFlashLoan function which is the callback during the flashloan. At this point we already have loaned and have all the DVT tokens. We first need to approve the RewarderPool to spend our tokens that we will deposit later. We then deposit the DVT tokens into RewarderPool, which mints us our virtual accounting tokens and then calls distributeRewards.
    • distributeRewards gives us our share of our rewards by minting and sending us some rewardTokens which are DVT, because we have in fact deposited (even though are deposit is the current borrowed DVT from the flash loans).
    • So now we've gotten some free DVT just from making a flashloan and depositing for the DVT rewards.
    • Then we immediately withdraw back our borrowed DVT and payback the flashloan. Then we transfer our free DVT we stole from our hack contract to our player personal address.
    • This is a stepwise exploit. You wait right before rewards are finished, then do a flashloan, deposit, get rewards, and then payback the loan. If stepwise calculations are done this is a big issue. It is better to calculate the rewards continuously rather than doing stepwise jumps.

#6 Selfie

  • Preface: CURRENTLY NOT WORKING OpenZeppelin currently broke this challenge with the release of v5.0.0 because there is no longer ERC20snapshot.sol.
  • Description/Goal: Pool offering flash loans of DVT tokens. It has a governance mechanism to control it. You start with no DVT tokens in balance. The pool has 1.5 million. Your goal is to take them all.
  • Resources: https://www.youtube.com/watch?v=_2RHyMMLR9A, https://github.com/zach030/damnvulnerabledefi-foundry
  • Topics: Flashloans (ERC-3156), DAO's
  • Methodology:
    • Anyone can take a snapshot. We flashLoan all 1.5M tokens from pool. This is so we can pass the _hasEnoughVotes check inside the queueAction function. We take a snapshot after we have all the DVT tokens during the flashloan process and successfully pass the check. This allows us to propose and queue an action.
    • We can now act as the governance contract, call queueAction, and pass in emergencyExit with us as the receiver, which will give us all the funds. The flash loan must finish first, so queueAction is called, then we approve the pool to transferFrom our borrowed DVT back.
    • The flash loan finishes. We wait 2 days and then call executeAction. This makes the governance contract call emergencyExit with us as the reciever. Because the governance contract is the one calling emergencyExit, this passes the onlyGovernance modifier.

#7 Compromised

  • Preface/Notes: _setupRole is depreicated in OZ, replace any instances with _grantRole.
  • Description/Goal: A related on-chain exchange is selling (absurdly overpriced) collectibles called “DVNFT”, now at 999 ETH each. This price is fetched from an on-chain oracle, based on 3 trusted reporters: 0xA732...A105, 0xe924...9D15 and 0x81A5...850c. Starting with 0.1 ETH, obtain all ETH available in the exchange.
HTTP/2 200 OK
content-type: text/html
content-language: en
vary: Accept-Encoding
server: cloudflare

4d 48 68 6a 4e 6a 63 34 5a 57 59 78 59 57 45 30 4e 54 5a 6b 59 54 59 31 59 7a 5a 6d 59 7a 55 34 4e 6a 46 6b 4e 44 51 34 4f 54 4a 6a 5a 47 5a 68 59 7a 42 6a 4e 6d 4d 34 59 7a 49 31 4e 6a 42 69 5a 6a 42 6a 4f 57 5a 69 59 32 52 68 5a 54 4a 6d 4e 44 63 7a 4e 57 45 35

4d 48 67 79 4d 44 67 79 4e 44 4a 6a 4e 44 42 68 59 32 52 6d 59 54 6c 6c 5a 44 67 34 4f 57 55 32 4f 44 56 6a 4d 6a 4d 31 4e 44 64 68 59 32 4a 6c 5a 44 6c 69 5a 57 5a 6a 4e 6a 41 7a 4e 7a 46 6c 4f 54 67 33 4e 57 5a 69 59 32 51 33 4d 7a 59 7a 4e 44 42 69 59 6a 51 34
MHhjNjc4ZWYxYWE0NTZkYTY1YzZmYzU4NjFkNDQ4OTJjZGZhYzBjNmM4YzI1NjBiZjBjOWZiY2RhZTJmNDczNWE5
0xc678ef1aa456da65c6fc5861d44892cdfac0c6c8c2560bf0c9fbcdae2f4735a9

MHgyMDgyNDJjNDBhY2RmYTllZDg4OWU2ODVjMjM1NDdhY2JlZDliZWZjNjAzNzFlOTg3NWZiY2Q3MzYzNDBiYjQ4
0x208242c40acdfa9ed889e685c23547acbed9befc60371e9875fbcd736340bb48
  • Now that we can control the oracles private keys, we can also just derive the public address with vm.addr(private key). We can now update the price with postPrice. We can act as an oracle and pass the TRUSTED_SOURCE_ROLE check in the postPrice function, so we set two of the three NFT's to a very low price of 0.0001 ETH. This will make the median price the very low price, which is what the oracle uses to determine the actual price of the symbol.
  • Then we act as the player, calling buyOne on the Exchange contract to buy an NFT for 0.0001 ETH.
  • Then we post our NFT for sale for the entire exchangeBalance (999+0.001), and sellOne sell it to the exchange. The player now has 999.1 ETH (started with 0.1), the exchange has 0 ETH.
  • Then as the last part of the challenge, we act as the oracle and set those two NFT prices back to 999 ETH.

winner

#8 Puppet 

Initial State:
Attacker: 1_000 DVT, 25 ETH
Lending Pool: 100_000 DVT
Exchange: 10 DVT, 10 ETH (1 ETH per DVT)
  • We ABI call tokenToEthSwapInput which sells 1_000 DVT for a minimum of 1 ETH. This puts 1_000 DVT into the uniswap exchange pool, and takes out ETH. But it doesn't just take out 1 ETH, we receive nearly all of the 10 ETH that was in the exchange because we put in so many DVT tokens compared to the initial state of the uniswap pool balances. Using the formula x * y = k we can calculate how much ETH we will receive, where x is DVT, y is ETH, and k is the constant.
  • k = 10 * 10 = 100. k must always remain 100. Therefore, after the swap the equation becomes (10 + 1000) * y = 100. So y = 0.099 ETH. This means the uniswap pools state after the swap will have 1010 DVT tokens and 0.099 ETH, giving us 10 - 0.099 = 9.901 ETH.
State after 1000 DVT swap to ETH:
Attacker: 0 DVT, 34.901 ETH
Lending Pool: 100_000 DVT
Exchange: 1_010 DVT, 0.099 ETH (10_202 DVT per ETH, 0.000098 ETH per DVT)
  • We have made the price of DVT extremely cheap in the eyes of the oracle which uses a very bad practice of calculating the price as return uniswapPair.balance * (10 ** 18) / token.balanceOf(uniswapPair);. Inside our hack contract to prepare for borrowing, we can calculate ethValue, the amount of ETH we need to deposit as collateral into the lending pool to get their 100_000 DVT tokens as a loan now with the adjusted cheap price of DVT. The equation boils down to DVT * (uni pool: ETH/DVT) * 2. In our case, that is 100_000 * (0.099 ETH / 1_010 DVT) * 2 = 19.6 ETH. So 19.6 ETH is needed as collateral to borrow 100_000 DVT. We call borrow, sending 19.6 ETH requesting to borrow the entire 100_000 DVT. We calculated ethValue beforehand so we know exactly how much to send to the pool for the 100_000 DVT tokens.
  • Note: There was no flashloan here, but instead a lending pool that let us borrow with ETH collateral. We are not forced or required to ever return the borrowed tokens. If we do not return the borrowed tokens, the lending pool simply keeps our 19.6 ETH collateral, which is fine for this challenge.
State after borrowing DVT from lending pool:
Attacker: 100_000 DVT, 15.301 ETH
Lending Pool: 0 DVT, 19.6 ETH
Exchange: 1_010 DVT, 0.099 ETH

winner

#9 Puppet V2 

State initially:
Attacker: 10_000 DVT, 20 ETH, 0 WETH
Lending Pool: 1_000_000 DVT, 0 ETH, 0 WETH
Exchange: 100 DVT, 10 ETH
  • The first thing we have to do is approve the router to spend our 10_000 DVT tokens, because we are looking to sell them all for the ETH in the exchange to manipulate the oracle price. Then we call swapExactTokensForETH on the router and request to swap all our 10_000 DVT tokens for a minimum of 1 ETH, from DVT to ETH, for our attacker. However, we know we'll get more than 1 ETH. Let's calculate.
  • x * y = k thus 100 DVT * 10 ETH = 1_000 = k. Recall k must remain constant during the swap.
  • (100 + 10_000) DVT * (10 - y) ETH = 1_000. Therefore y = 9.999 which is the amount of ETH we will receive. This isn't exact as it doesn't account for gas, so let y = 9.9.
  • The attacker then calls deposit function on weth that just exchanges all his 20+9.9 ETH for WETH.
State after attacker swaps all DVT for ETH, and then does deposit to swap ETH to WETH
Attacker: 10_000 DVT, 0 ETH, 29.9 WETH
Lending Pool: 1_000_000 DVT, 0 WETH
Exchange: 10_100 DVT, 0.1 ETH ( 101_000 DVT per ETH, 0.0000099 ETH per DVT)
  • Because DVT is so cheap now, the oracles quote price will reflect that. So we approve the lending pool for all our WETH, and then we want to calculate how much WETH collateral is required to take the 1_000_000 DVT from the lending pool with calculateDepositOfWETHRequired. This calculation goes through _getOracleQuote and quote functions, which in our case boils down to:
  • WETH collat req = ((lending pools DVT * uniswap reserve WETH) / uniswap reserve DVT) * 3. Therefore,
  • WETH collat req = ((1_000_000 * 0.1) / 10_100 ) * 3 = 29.7 WETH collateral required.
  • We then call borrow for the pool amount which passes successfully because we have enough collateral, yielding us 1_000_000 DVT for 29.7 WETH.
State after attacker successfully borrows DVT
Attacker: 1_010_000 DVT, 0 ETH, 0.2 WETH
Lending Pool: 0 DVT, 29.7 WETH

winner

#10 Free Rider 

State initially:
attacker: 0.5 ETH
attackerContract: 0
nft marketplace: 6 NFTs at 15 ETH each
freeRiderBuyer: 45 ETH
Uniswap pool: 9_000 WETH, 15_000 DVT
  • We call flashswap from our attack contract. What is a flash swap? "Uniswap flash swaps allow you to withdraw up to the full reserves of any ERC20 token on Uniswap and execute arbitrary logic at no upfront cost, provided that by the end of the transaction you either pay for the withdrawn ERC20 tokens with the corresponding pair tokens or return the withdrawn ERC20 tokens along with a small fee."
  • flashswap calls swap on UniswapV2Pair for DVT/WETH. We pass in amount0Out as 0 (DVT), amount1Out as 15 (WETH), our address, and then some data bytes("1337). This means we are requesting a flash swap for 15 WETH. We successfully get the 15 WETH.
  • Importantly, if data.length > 0 uniswap assumes the payment has NOT been received, so uniswap pair contract transfers us the 15 WETH and then calls our callback function uniswapV2Call in our contract. This is similiar to the onFlashLoan callback principle for flash loans. By the end of the uniswapV2Call callback, the WETH that was flash swapped for nothing must be repaid.
State after flash swapping 15 WETH and then swapping for 15 ETH:
attacker: 0.5 ETH
attackerContract: 15 ETH, 0 NFTs
nft marketplace: 6 NFTs, 0 ETH
freeRiderBuyer: 45 ETH
Uniswap pool: 8_985 WETH, 15_000 DVT
  • Next in our callback we prepare the token id array and then call buyMany on the freeRiderNFTMarketplace contract passing in 15 ether and all the token ids. This gives the marketplace our 15 ETH, and then calls _buyOne 6 times.
  • The _buyOne function has two vulnerabilities.
  • Firstly, it does not deduct priceToPay from msg.value! This means throughout the 6 calls to it, msg.value will stay at 15, meaning we only had to send the 15 ETH once in order to pass all the checks on this function.
  • Secondly, the safeTransferFrom and the sendValue function are in the wrong order! It should send the original owner (marketplace) the 15 ETH of value, and then give the attack contract the NFT. But because it first transfers the NFT to our attack contract, WE ARE NOW THE OWNER. And so when it sends the 15 ETH payment to the owner, instead of paying the actual owner, it pays us because we are the owner! This happens 6 times, giving us all 6 NFT's, and giving us 90 ETH in the process. While the marketplace only ends up with our 1 payment of 15 ETH.
State after buyOne fiasco:
attacker: 0.5 ETH
attackerContract: 90 ETH, 6 NFTs
nft marketplace: 0 NFTs, 15 ETH
freeRiderBuyer: 45 ETH
Uniswap pool: 8_985 WETH, 15_000 DVT
  • Now we can just send all 6 NFTs to freeRiderBuyer who will give attacker contract (NOT attackerContract) the 45 ETH bounty.
  • We can pay back the flash swap (exchange 15.1 ETH from attackerContract to 15.1 WETH and then payback flashloan to conclude the flashloan).
  • Optional: In the test file, you can transfer the NFTs from freeRiderBuyer to buyer if you want to complete the story.
State after giving 6 NFTs to freeRiderBuyer & paying back flash swap:
attacker: 45.5 ETH
attackerContract: 74.9 ETH, 0 NFTs
nft marketplace: 0 NFTs at 15 ETH each
freeRiderBuyer: 0 ETH, 6 NFTs
Uniswap pool: 9_000.1 WETH, 15_000 DVT

winner

#11 Backdoor

  • Preface: A mess to setup properly with imports and dependencies (safe-contracts repo changed their @!#&ing file names 3 weeks ago to remove the word GNOSIS, wtf?).

  • Description/Goal: There is a registry of gnosis safe wallets. If someone deploys and registers a wallet, they get 10 DVT from the registry. It also uses gnosis safe proxy factory. 4 people, Alice, Bob, Charlie, David are currently registered into the system as beneficiaries. Therefore, the registry has 40 DVT tokens in the balance ready to be distributed. We have to take the funds from the wallet registry in a single transaction.

  • Topics: Gnosis safe wallets

  • Resources: https://www.youtube.com/watch?v=j48sEXLzt0E, https://github.com/nicolasgarcia214/damn-vulnerable-defi-foundry

  • Methodology:

    1. Declaring our intentions

      • Our RegistryAttack contract calls createProxyWithCallback. This function makes two function calls, createProxyWithNonce and proxyCreated. The latter we do not need to worry about until after the proxy is created.
      • So firstly, createProxyWithNonce calls deployProxy, which makes a low level call, which begins the execution of our initializer data. Importantly, if initializer data is provided, which it is, it executes the initializer in the context of the newly created proxy. Also importantly, deployProxy is literally freshly printing a new proxy address with the create2 opcode, and then doing a low level call referencing the proxy and the initializer data, which makes the initializer data run in the context of the fresh off the press proxy.
      • So far, all we are doing is telling the deployProxy that we want to create a proxy with our initializer data. This is done with the gnosis Safe's setup function, which should be the first argument for our initializer.
      • Inside our initializer data, we specify Safe.setup.selector and then the arguments we want to pass into the Safe's setup function. Behind the scenes, setup calls setupOwners (which is irrelevant to the hack and just sets up the owner). But setup then also calls setupModules, which calls execute. And the execute function makes a delegatecall with our to address and the data argument inside our initializer data... Very interesting. There are no relevant checks anywhere for the to or data arguments for the safe's setup function. Therefore, we can pass in arbitrary to address and data.

    2. Specifying our malicious data

      • So inside our attack, we can pass in address(this) inside the initializer data as the to parameter for the setup function, which will reference the proxy because our initializer data is being run in the context of the proxy because of how deployProxy works.
      • For the data parameter for the setup function, we pass in delegateApprove function that we made in our RegistryAttack contract, which takes in address(this) which is the proxy, and then the 10 DVT token amount. This function approves the _spender which is the proxy to spend 10 DVT. This is important because after the proxy is created, the WalletRegistry will give the proxy 10 DVT.
      • Note1: The salt nonce and other peripheral things aren't core to the hack, but are just a part of the setup and creation process.
      • Note2: The setup function takes in an _owners array of addresses, so we had to make sure that argument was an array of length one which just contained whoever we were currently pwning (Alice, Bob, Charlie, David).

    3. Watching the magic happen

      • Finally, createProxyWithCallback finishes creating the proxy, and during this proxy creation, the DVT token contract has approved the proxy to be able to transfer 10 DVT tokens.
      • Now all that is left is the formalities and the things that happen after a proxy is created. Notice we pass in IProxyCreateFactory(registry) as the last argument which references the WalletRegistry contract, which does have the proper callback proxyCreated, so this is now what is executed because the proxy has finished being created.
      • Inside WalletRegistry.proxyCreated, it does a bunch of safety checks (one of the checks isn't robust enough and misses our malicious initializer data), then removes the expected wallet owner beneficiary (Alice, Bob, Charlie, David) from the beneficiaries mapping, and then registers their wallet associated with the proxy address (walletAddress), and finally then pays their proxy 10 DVT.
      • Now that the proxy has 10 DVT, (AND can send that 10 DVT!), we can call transferFrom from the proxy to msg.sender, which gives the attacker 10 DVT because msg.sender is the attacker because we start the prank as the attacker. Each time a wallet like Alice/Bob/Charlie/David creates a proxy for their associated wallet, we get sent 10 DVT. And importantly, we were able to just pass in Alice/Bob/Charlie/David as the wallet users in our attack function.
      • Note: During the whole attack function we iterate through each 4 users and do this entire process for each of them, so this all happens in a single transaction under the umbrella of the attack function

    4. Fixing the Vulnerability

      • The proxyCreated function inside WalletRegistry has a lackluster check for the initializer, and should be changed from if bytes4(initializer[:4]) != Safe.setup.selector) revert to something more explicit to make sure you can't pass malicious to and data parameters for the setup function during the proxy setup and creation. All their check is doing is making sure that our initializer is calling setup... But not checking the data inside setup!

winner

#12 Climber

  • Preface/Notes:

    • _setupRole is depreicated in OZ, replaced instances with _grantRole.
    • For __Ownable_init(admin); in ClimberVault contract, I had to pass in admin argument.

  • Description/Goal: A secure vault contract has 10 million DVT tokens. The vault is upgradeable, following the UUPS pattern. The owner of the vault, currently a timelock contract, can withdraw a very limited amount of tokens every 15 days. On the vault there’s an additional role with powers to sweep all tokens in case of an emergency. On the timelock, only an account with a “Proposer” role can schedule actions that can be executed 1 hour later. To pass this challenge, take all tokens from the vault.

  • Topics: UUPS, time locking

  • Resources: https://www.fatihdev.com/post/damn-vulnerable-defi-solutions-12-climber

  • Methodology:

    • The exploit function in ClimberAttack calls the timelock's execute function after carefully preparing malicious data for it. This triggers the execution of the calldata from the dataElements to their corresponding targets by calling functionCallWithValue which is ultimately making a low level call. Importantly, the targets, values, and dataElements array must all be the same length because each target address expects an associated value and data from the low level call.
    • Note: All of these packaged actions are using timelock as the msg.sender because we do timelock.execute so we have access to the ClimberTimelock contract functions.
    • We first set the delay of the timelock to 0 by ABI packaging timelock.updateDelay(0). This allows immediate execution of scheduled actions by passing the ReadyForExecution check in the excecute function after this piece has been iterated through and executed.
    • We then grant the PROPOSER role to our hack contract by ABI packaging timelock.grantRole(PROPOSER_ROLE, address(this)). This enables us to call schedule to schedule actions.
    • We then upgrade the vault implementation to our malicious ClimberVaultV2 with upgradeToAndCall function. Notice we also get to pass in data here after we upgrade a vault.
    • So then we call the new ClimberVaultV2 malicious sweep function to transfer all tokens to the attacker's address.
    • We then have to call scheduleOperation() which calls timelock.schedule for our data in our contract to actually schedule the above actions. This will schedule the actions and classify them as ReadyForExecution because of the 0 delay, therefore execute will successfully finish. We just need execute to not revert, because if it finishes then all of our malicious steps will happen in one atomic transaction successfully!
    • Lastly, we actually call execute with all this prepared data, which iterates through the targets array and makes the associated low level calls with the associated data we prepared. It successfully passes all the checks, executing our malicious data to give us all the vault funds.
    • Fixing the Vulnerability: Have an access control modifier for the execute function, and/or sanitize the incoming dataElements.

winner

#13 Wallet Mining

#14 Puppet V3

#15 ABI Smuggling

Resources I used to help me adapt all challenges to foundry