Fatal flaws in a bot, a sort of on-chain virus, a trojan horse token, and arbitrage gone wrong
Join me in looking at the latest MEV bot exploit in this thread 👇🏻
The victim today tried to arbitrage the CHUM token (!) for ~0.01 ETH in profit, but in the process had 30 WETH transferred out of their wallet.
They only interacted with Uniswap v2 pools, how did this happen!?
Searchers optimize their contracts to the extreme to do very specific things very efficiently.
However, occasionally they have a need to do random things & add in functions that can be used for to execute basically any arbitrary transaction in a contract
Still, for almost any interaction on-chain it was fine.
To make an arbitrary call with this victim's contract you would need them to originate the transaction somehow and inject your desired transaction in their's, like a virus!
How might you bait them into doing so?
Well, the victim was an arbitrage bot! The answer was to bait them by deploying a new token and creating a fake arb to get them to interact with that token.
Thus the CHUM token and two Uniswap v2 pools were borne.
But how does CHUM work? Let's go through it end to end.
First, the victim tried to swap on the Uniswap v2 pool
Then, the Uniswap v2 pool transferred CHUM back to the victim, in doing so triggering some special logic
What happens now?
This was hard to piece together but I think I got it with 3 clues.
1st, you can see in CHUM decompiled that it checks for certain addresses as well as using block.coinbase and making calls.
This is CHUM targeting the victim & maybe avoiding simulation.
Second, looking @ CHUM's transactions we can see transactions with input data that contains the address for WETH, as well as the attacker's address (3a28...), and what looks to be numbers.
Looks to me like a custom payload that transfers WETH to the attacker, a sort of virus.
Lastly, we can also see in a trace of the execution that CHUM is calling another contract (!) "0x33..." that is hard to spot on Etherscan.
With these I think we can construct what happened.
The attacker first injects the on-chain virus designed to transfer WETH out of the victim into CHUM
Victim arbs CHUM and the Uniswap pool calls transfer on CHUM as part of the swap
CHUM calls then contract 0x33 with the WETH transfer virus
0x33 then calls the victim's contract with the WETH transfer virus, attempting to execute it
The victim's contract checks to see that the victim started the transaction and this condition passes.
The victim's contract executes the virus.
The attacker is sent 30 WETH.
The attacker then IMMEDIATELY updated their WETH transfer virus and repeated this twice more for 35 WETH in total.
They were watching closely.
This was a highly targeted attack pointed at a specific bot's contract. It would only work for bots with a specific function AND if the bot used tx.origin as a safety check.
I have to wonder how they even found that vulnerability amongst hundreds of other bots.
Still, it was relatively easy to defend against. First, you shouldn't use tx.origin as a safety check.
Second, you should check that your transaction has made a project before paying the miner!
That is it for today everyone.
Check out our Github to learn more and get involved if you're interested in mitigating MEV's negative externalities:
Writing this as quick as I can but I gotta figure this one out first. It's a hard one.
Looks like tomorrow or Monday I’ll put something out. Taking awhile to understand and express in a way that makes sense and I’ve got stuff to do today. Sorry folks.
🥪 frontruns a user, buying the asset they intend to and increasing the price. The user gets less tokens now.
The user's buy is then included, pushing the price up more
🥪 sells after the user's tx at the higher price, thus capturing profit
🥪 bots will watch the mempool for users trading with high slippage that they can frontrun. Until recently 🥪 bots would only do this with one trade, and with the Uniswap v2 or Sushiswap router.
The tricky thing for this bot is it needs to land right behind another user's tx to succeed
Since it can't express those ordering preferences and because the costs of a transaction are so low, it chooses to spam the network with transactions and hope one lands in the right place
It's a golden age for searchers. Late enough that Flashbots and DeFi exists, but not so late that massive institutions are competing for MEV.
Even still the barrier to entry for the top strategies is getting higher very quickly. It's probably a few months of work to catch up to the best backrunning arb bots now.
It is funny to think about the early days of Flashbots when there was literally 1 market maker, 1 ESD bot, and 2 arb bots sending bundles. Unbelievably good opportunity to make money back then.