Synthetix SIP-37 Smart Contract Audit
iosiro was commissioned by Synthetix to conduct a smart contract audit on the implementation of SIP-37. The audit was performed between 10 February 2020 and 14 February 2020. A review of changes related to the audit was performed on 18 February 2020.
This report is organized into the following sections.
The information in this report should be used to understand the risk exposure of the smart contracts, and as a guide to improving the security posture of the smart contracts by remediating the issues that were identified. The results of this audit are only a reflection of the source code reviewed at the time of the audit and of the source code that was determined to be in-scope.
This report presents the findings of the audit performed by iosiro on the smart contract implementation of SIP-37.
The purpose of this audit was to achieve the following.
Assessing the market effect, economics, game theory, or underlying business model of the platform were strictly beyond the scope of this audit.
Due to the unregulated nature and ease of transfer of cryptocurrencies, operations that store or interact with these assets are considered very high risk with regards to cyber attacks. As such, the highest level of security should be observed when interacting with these assets. This requires a forward-thinking approach, which takes into account the new and experimental nature of blockchain technologies. There are a number of techniques that can help to achieve this, some of which are described below.
The purpose of SIP-37 is to prevent the front-running of oracles within the Synthetix system. This is achieved by ensuring that Synth exchanges take into account changes to market prices within a certain time window. Any profit made by the user within this time window can be reclaimed by the system. Conversely, if the user makes a loss on the exchange in the time window, the amount of the loss is rebated to the user.
SIP-12 introduced a maximum gas limit on certain operations within the system to prevent direct front-running of the system through submitting transactions with high gas prices. However, due to UX and technical concerns around the solution, an alternative approach was required. An area of particular concern was that as the system has started to move to decentralized price oracles on ChainLink, additional latency has been added to the system. While the Synthetix Oracle would update on-chain prices when a 0.3%-0.5% price deviation was detected, the ChainLink network operates closer to 1%. These slower updates due to reduced sensitivity would give front-runners more opportunities to exploit the system. As such, SIP-37 incorporates a waiting time between exchange operations, reducing the likelihood that a front-runner is able to take advantage of any latency in updating exchange rates.
The code was found to be of a very high standard, as it was modularised, well-documented, defensive, and generally adhered to best practices. The implementation was found to strictly implement the specification.
At the conclusion of the audit only informational issues were open. These findings included suggestions for ways to improve the performance and readability of the system.
The source code considered in-scope for the assessment is described below. Code from all other files is considered to be out-of-scope. Out-of-scope code that interacts with in-scope code is assumed to function as intended and introduce no functional or security vulnerabilities for the purposes of this audit.
Project Name: Synthetix
Commit: 6ac6f4b
Files: ExchangeRates.sol, ExchangeState.sol, Exchanger.sol, Issuer.sol, Synth.sol, Synthetix.sol
A variety of techniques were used in order to perform the audit. These techniques are briefly described below.
The source code was manually inspected to identify potential security flaws. Code review is a useful approach for detecting security flaws, discrepancies between the specification and implementation, design improvements, and high risk areas of the system.
The contracts were compiled, deployed, and tested in a Ganache test environment, both manually and through the Truffle test suite provided. Manual analysis was used to confirm that the code operated at a functional level, and to verify the exploitability of any potential security issues identified.
Tools were used to automatically detect the presence of several types of security vulnerabilities, including reentrancy, timestamp dependency bugs, and transaction-ordering dependency bugs. The static analysis results were manually analyzed to remove false-positive results. True positive results would be indicated in this report. Static analysis tools commonly used include Slither, Securify, and MythX. Tools such as the Remix IDE, compilation output, and linters are also used to identify potential issues.
Each issue identified during the audit has been assigned a risk rating. The rating is determined based on the criteria outlined below.
The following section outlines the intended functionality of the system at a high level.
The following section was taken from SIP-37 for the purposes of establishing a specification for the audit.
When a user exchanges src synth for dest synth, the waiting period of N minutes begins. Any transfer of dest synth will fail during this window, as will an exchange from dest synth to any other synth, or a burn of sUSD if it was the dest synth of any pending exchange. If another exchange into the same dest synth is performed before N minutes expires, the waiting period restarts with N minutes remaining.
Once N minutes has expired, the following exchange from the dest synth to any other, or a burn of sUSD, will invoke settle - calculating the difference between the exchanged prices and those at the end of the waiting perid. If the user made profit, it is taken to be front-run profit, and the profit is burned from the user's holding of the dest synth. If the user made a loss, this loss is issued to them from the dest synth. The exchange then continues as normal.
In the case of a user trying to transfer some amount of the dest synth after the waiting period has expired, this will always fail if the amount + totalOwing is more than the user's balanceOf. The user has to first invoke settle before a synth can be transferred. Otherwise, the transfer will continue as normal.
The calculation of owing or owed is as follows:
Amount * (1 - feeRate) * (srcRate/destRate - newSrcRate/newDestRate)
If the result is negative, the amount is owed as a rebate, otherwise its owing as a
Examples (with feeRate at 0.003):
100 sUSD into sETH at a ETHUSD rate of 100:1 (1/100) which raises to 105:1 (1/105), the owing would be: 100 * 0.997 * (1/100 - 1/105) = 0.04747619048 sETH.
100 sETH into sBTC at a BTCUSD rate of 10,000:1 and ETHUSD rate of 100:1 which raises to 105:1, the user would be rebated an owed amount of: 100 * 0.997 * (100/10000 - 105/10000) = 0.04985 sBTC
The implementation should be as follows.
Synthetix.exchange() invoked from synth src to dest by user for amount
Are we currently within a waiting period for any exchange into src?
settle(src)exchange as per usualdest synthSynthetix.settle(synth) invoked with synth by user
Are we currently within a waiting period for any exchange into synth?
owing and owed amounts on all unsettled synth exchanges as totalOwing and totalOwedtotalOwing of synth from the user by burning ittotalOwed of synth to the user by issuing itsynth exchanges for the userSynth.transfer() invoked from synth src by user for amount
Are we currently within a waiting period for any exchange into src?
owing amounts on all unsettled synth exchanges as totalOwingSynth.burnSynths() invoked by user for amount
Are we currently within a waiting period for any exchange into sUSD?
settle(src)burn as per usualThe following section includes in-depth descriptions of the findings of the audit.
No high risk issues were present at the conclusion of the audit.
No medium risk issues were present at the conclusion of the audit.
No low risk issues were present at the conclusion of the audit.
Actions to improve the functionality and readability of the codebase are outlined below.
The waiting period for all exchanges was set to a period of N minutes, with a default of 3 minutes. However, as the system represents assets with varying volatilities, it may be necessary to use different N values to better model the required waiting times of each asset. For example, with sETHsUSD being much more volatile than sXAUsUSD, N minutes might be too long for sETHsUSD and too short for sXAUsUSD.
Yes, this is something that we will be looking at in future deployments. We are considering looking at on-chain volatility to inform this.
The getMaxTimestamp(...) iterated through an array of a user's exchanges to find the exchange with the most recent timestamp. However, in the current implementation of the system, the only times when the array was modified was when a new entry was pushed into the array with a timestamp of now and when the array was emptied. As such, the last element inserted into the array would always have the most recent time, and could be referenced through userEntries[userEntries.length - 1].
It should be noted that if additional methods of modifying the array are anticipated (e.g. updating the timestamp of an exchange or inserting elements into the array without using push), then the more robust current approach should be used, as it would account for an unordered array.
True it could be improved given the current usage, however as ExchangeState is immutable, and we can always fetch the last entry via getEntryAt(getLengthOfEntries(…) - 1 externally if need be, so we’d prefer to keep the max function for potential future usage.
effectiveValueAtRound(...)The effectiveValueAtRound(...) function was used to determine the exchange rate of two currencies at a certain point in time. As different exchange rates had independent indices, it would be possible to call the function with two exchange rates which were set at vastly different times. Given the intended functionality of the function, it may be beneficial to include validation to ensure that the rounds are within a certain time period of each other.
The only usage of this function was in settlementOwing(...) which was not found to be exposed to this risk, as the round IDs were determined based on time values. Furthermore, given that there were two oracle systems and several assets, it would not be straightforward to determine an acceptable time window. As such, no action is necessarily required. However, it should be noted that third parties and future versions of the Synthetix system could be exposed to this risk.
This would be too onerous to manage, given that each oracle network under Chainlink will have different lengths of time that they will, at a minimum, guarantee updates.
A potential gas optimization for the resolver pattern was identified during the audit. The pattern at the time of the audit used two external calls to determine the address, e.g.
function exchanger() internal view returns (IExchanger) {
require(resolver.getAddress("Exchanger") != address(0), "Resolver is missing Exchanger address");
return IExchanger(resolver.getAddress("Exchanger"));
}
In testing a total of 29,159 gas was consumed.
However, it would be possible to store the address to reduce the function to one external call, e.g.
function exchanger() internal view returns (IExchanger) {
uint resolved = resolver.getAddress("Exchanger");
require(resolved != address(0), "Resolver is missing Exchanger address");
return IExchanger(resolved);
}
In testing a total of 25,646 was consumed during this function call, 3,513 gas less than the current implementation.
As this functionality was used extensively throughout the codebase, it may have a significant effect on the overall gas consumption.
Implemented in ff931cf.
Done, and improved slightly with a function requireAndGetAddress() that does the require as well, lowering the bytecode each of our contracts that use it require on deployment. (88f288b)
While Synthetix.sol#L246 specified that settle(...) returned a bool, the function returned (uint, uint). It is recommended that the comment be changed to correctly describe the code.
Removed in baa6d13.
It is recommended that certain portions of the code be refactored to improve readability.
roundsForRates should be currentRoundForRate for clarity.code should be replaced with currencyKey throughout the file for consistency.rates() should be rate() as it returns a single rate. Implemented in ca4923f.
Language mistakes were identified in the comments and revert messages in the codebase. Fixing these mistakes can help improve the end-user experience by providing clear information on errors encountered, and improve the maintainability and auditability of the codebase.
recevied should be received.keys should be key.configure.MUTITATIVE should be MUTATIVE..contract at the end of the sentence.THe should be The.Implemented in fa6b43d.
