How Did Lendf.Me Lose $25 Million to A Reentrancy Attack? [An Analysis]

DeFi or decentralized finance is a growing sector in the blockchain and cryptocurrency space that defines an ecosystem of decentralized applications providing financial services with no governing authority.

Lendf.me is a DeFi app utilizing smart contracts in order to provide instant, decentralized lending. The platform suffered an attack causing the loss of $25m in cryptocurrency on the day of April 19, 2020.

Thus, joining the list of other DeFi protocols exploited recently:

Synthetix hack — 37M sETH stolen

bZx hack — $900k stolen

Lendf.me’s current vulnerability is a unique instance of the reentrancy bug. Reentrancy is a well-known issue in the field of computing, referring to the ability of a subroutine to be interrupted in the middle of its execution and (then safely) be called again.

Other reentrancy bugs have been exploited in the past, causing massive damage, including:

The DAO hack — $150m stolen

Spank Chain hack — $30k stolen

The Lendf.me attack took place on the Ethereum main-net smart contract named MoneyMarket, which implements the core logic of the Lendf.me app.

Understanding the Lendf.me vulnerability

In order to best understand the underlying cause of the vulnerability, we should consider the contents of the various functions in the MoneyMarket contract.

First, we will consider the

MoneyMarket.supply()

function (line 1508):

MoneyMarket.supply()

part 1

MoneyMarket.supply()

part 2

MoneyMarket.doTransferIn()

The main purpose of the

MoneyMarket.supply()

function is to handle token deposits. The function takes two arguments, the asset (the asset that the user wishes to deposit), and the amount (the number of tokens he wishes to deposit).

The main logic flow of the

MoneyMarket.supply()

function is as follows:

First, we read the balance variable that represents the user’s deposited asset balance in

MoneyMarket

storage (line 1514), then,

MoneyMarket.checkTransferIn()

function is invoked (line 1526). This function (externally) calls the asset contract in order to figure if the user has the number of tokens he wishes to deposit and that he approved the

MoneyMarket

contract to withdraw this amount on his behalf.

Later

MoneyMarket.doTransferIn()

function is invoked (line 1583) which (externally) calls the asset contract’s

transferFrom()

function (line 405) that in turn transfers the amount from the user to the

MoneyMarket

contract. After the return from the external call the

MoneyMarket.supply()

function is updating the user’s deposited balance (lines 1599–1600).

Let’s go over the

MoneyMarket.withdraw()

function’s logic briefly. In a simplified manner, this function gets the requested amount of tokens to withdraw, checks that the user holds at least this amount of tokens then transfers these tokens to the user by (externally) calling the token contract

transfer()

function.

Can you spot the vulnerability by now?

The issue here is that

MoneyMarket.supply()

function is actually updating the user’s asset balance after the external call to

asset.transferFrom()

(lines 1599–1600), but based on a value that was read before the external call (line 1514), which means that the update potentially ignores any updates that were made within the external call. In many terms, we can consider this anomaly to be a “Lost Update”.

But why is Lendf.me’s vulnerability exploitable?

In order to understand this, we will have a look at the imBTC contract (or any other ERC-777 compliant contract)

imBTC._transferFrom()

imBTC._callTokensToSend()

The attacker took advantage of the fact that some of the assets implement ERC-777 standard, which means that the

imBTC._callTokensToSend()

function and thus,

attackerContract.tokensToSend()

function are invoked (lines 866, 1056 respectively) before the actual transfer of value between the two parties.

This way, the attacker’s contract gets a chance to call

MoneyMarket.withdraw()

function before the invocation of

MoneyMarket.supply()

is finished!

Attack Strategy

The only prerequisite for attempting the exploit is for an attacker to deploy an attacker contract that holds some amount of any asset that is ERC-777 compliant, let’s assume for example that the attacker holds 10 tokens of imBTC.

Now,

The attacker would place the first transaction that invokes

MoneyMarket.supply

( asset =

imBTCAddress

, amount = 9). At this point, the attacker holds a supply of 9 imBTC in the

MoneyMarket

contract, and a balance of 1 imBTC in the imBTC token contract.The attacker would place the second transaction that invokes

MoneyMarket.supply

(asset = imBTCAddress, amount = 1) , but now with an external call to

MoneyMarket.withdraw

(asset =

imBTCAddress

requestedAmount

= 9) inside the

attackerContract.tokensToSend()

callback.

By the end of this transaction, the attacker’s imBTC balance in the imBTC token contract is 9, but the imBTC supply in the

MoneyMarket

contract is 10! This unwanted state occurred as the

MoneyMarket.supply()

function increases the supply for the attacker (lines 1599–1600) it uses stale data.

Therefore, the function doesn’t “know” at this point, that the attacker has already withdrawn some of his supply.Now, the attacker holds a deposit on the

MoneyMarket

contract, backed by nothing.

The attacker can use this to (falsely) borrow or withdraw assets deposited by other users. Furthermore, these two steps could be potentially performed, again and again, thus draining

MoneyMarket

’s liquidity.

Mitigation

When writing the smart contract code, try not to update any storage variables after an external call.If not possible, deploy some locking mechanism, like the commonly known

ReentrancyGuard

instead. Make sure that any pair of code paths that have a possible read/write conflict for a variable will be “reentrancy guarded”. For example, in this case, deploying a reentrancy guard only for the

MoneyMarket.supply()

function would not solve the problem, it should be deployed for the

MoneyMarket.withdraw()

function as well. Valid network’s automated tools can help identify locations where these guards are missing, or incorrectly implemented.

(Written by David Oz Kashi on Behalf of Valid Network)