Submitted by bart1e, also found by nobody2018 and immeas
https://github.com/code-423n4/2023-07-axelar/blob/2f9b234bb8222d5fbe934beafede56bfb4522641/contracts/its/token-manager/implementations/TokenManagerLockUnlock.sol#L60-L67
https://github.com/code-423n4/2023-07-axelar/blob/2f9b234bb8222d5fbe934beafede56bfb4522641/contracts/its/token-manager/implementations/TokenManagerLiquidityPool.sol#L94-L101
TokenManager implementations inherit from the FlowLimit contract that keeps track of flow in and flow out. If these two values are too far away from each other, it reverts:
Flow in and flow out are increased when some tokens are transferred from one blockchain to another.
There are 3 different kinds of TokenMaganer:
Lets see how Lock/Unlock and Liquidity Pool implementations handle cases when they have to transfer tokens to users:
As can be seen, they return a value equal to a balance difference before and after token transfer. This returned value is subsequently used by the giveToken function in order to call _addFlowIn:
The problem arises when token is ERC777 token. In that case, the token receiver can manipulate its balance in order to increase flow in more than it should be; see POC section for more details.
There is no mechanism that will disallow someone from creating TokenManager with ERC777 token as an underlying token, so its definitely a possible scenario and the protocol would malfunction if it happens.
Note that its not an issue like users may deploy TokenManager for their malicious tokens that could even lie about balanceOf. Users may simply want to deploy TokenManager for some ERC777 token and bridge their ERC777 tokens and there is nothing unusual about it.
It is possible to manipulate flow in when there is TokenManager of type Lock/Unlock or Liquidity Pool and the underlying token is ERC777 token. It could be used to create DoS attacks, as it wont be possible to transfer tokens from one blockchain to another when the flow limit is reached (it may be possible to send them from one blockchain, but it will be impossible to receive them on another one due to the revert in the _addFlow function).
In order to recover, flowLimit could be set to 0, but the feature was introduced in order to control flow in and flow out. Setting flowLimit to 0 means that the protocol wont be able to control it anymore.
Here, the availability of the protocol is impacted, but an extra requirement is that there has to be TokenManager of Lock/Unlock or Liquidity Pool kind with ERC777 underlying token, so Im submitting this issue as Medium.
Consider the following scenario:
First, this records current balance of T of MaliciousContract, which happens to be 0 and calls transfer, which finally calls MaliciousContract::tokensReceived hook.
Where <TOKEN_MANAGER_ADDRESS> is the address of the relevant TokenManager and maliciousContractHelper is an instance of MaliciousContractHelper. That exposes the sendMeT function, which will send all tokens that it has to the MaliciousContract instance that called it.
In short, Bob has increased the flow in amount for TokenManager by sending to their contract a lot of money in ERC777 tokensReceived hook from their other contract. It didnt cost them much since they sent only a tiny amount of T between blockchains. Hence they could use almost all of their T tokens for the attack.
Of course Bob could perform this attack without waiting for Alice to submit their transaction. The scenario presented above was just an example. In reality, Bob can do this at any moment.
It also seems possible to transfer tokens from the same blockchain to itself (by specifying wrong destinationChain in sendToken or just by specifying destinationChain = <CURRENT_CHAIN>), so Bob can have their tokens T only on one blockchain.
If gas price on that blockchain is low, Bob can perform that attack a lot of times. All they need to do is to send tokens back to MaliciousContractHelper after each attack (so that it can send it to MaliciousContract as described above). Finally, they will reach flowLimit for TokenManager and will cause denial of service.
VS Code
I would recommend doing one of the following:
deanamiel (Axelar) confirmed and commented:
