Endaoment Audit

Introduction

The Endaoment team asked us to review and audit their smart contracts. We looked at the code and now publish our results.

We audited commit f60aa253d3d869ad6460877f23e6092acb313add of the endaoment/endaoment-contracts repository. The scope includes all the smart contracts in the repository, except the mocks used for testing.

About Endaoment

The Endaoment project is a Donor-Advised Funds platform, where everyone is free to transfer their ERC20 tokens to a Fund contract to fund approved organizations with the use of grants. More grants can be created for the same fund and when a specific grant is finalized an amount of donated tokens is transferred to the recipient organization. The funds received by an organization from the grant can be then cashed out.

Anyone can propose claims containing basic information about an organization but an administrative governance choses which claims to approve, which organizations to allow and which grants can be finalized. The entire system is governed by owners of the EndaomentAdmin contract which stores and provides specific role accesses to users interacting with the system. They also manage two factories to deploy Org contracts, corresponding to allowed organizations, and Fund contracts.

Governance model

The EndaomentAdmin contract is the one in charge of managing all restricted accesses to functions, included the ones to move tokens from Fund contracts to Org contracts and to cash them out.
The creator of this contract can transfer their ownership with a two-step transfer and is in charge of setting and revoking the following roles to the corresponding addresses:

  • PAUSER: to pause other roles and disable them temporary. Unpause them is an exclusive right of the contract owner.
  • ACCOUNTANT: to finalize grants, cash out organizations and to create Fund or Org contracts.
  • REVIEWER: to change the wallet associated to an organization and overwrite the fund’s manager address.
  • FUND_FACTORY and ORG_FACTORY: to deploy new Fund and Org contracts.
  • ADMIN: to have full control over the system and the exclusive role of deploying new factories.

The Fund contract has a specific fund manager that can create new grants.

A two-step transfer of ownership is, in general, a safe pattern to follow, but users must be aware of the fact that a transfer is not completed when initialized. Instead, this is completed when the new owner accepts the transfer. If an owner is trying to cancel the ownership transfer, it can be front-runned by the recipient owner trying to mine a transaction that accepts the transfer before the cancellation happens.

Summary

Overall the code is extensively commented and easy to understand with very well encapsulated contracts and functions. However, our main concern are the vulnerabilities we found in the roles implementation that can lead to loss of funds. The privileged roles built in the system are in general a safe way to protect access to funds management, but they should be finely implemented to avoid any possible attack from malicious actors. Also the lack of technical documentation makes it hard to understand the exact purpose and functionality of each one of the contracts, and thus makes them hard to audit.

Since fixing the issues we have found would require a lot of changes in the code, consider improving the tests with a particular focus on adversarial scenarios, and auditing again the code once updated.

Update: The Endaoment team did a great job reviewing the code, refactoring when necessary and fixing all the major issues. The fixes changed many lines of code, but the governance model has still the same overall architecture. We are happy to see that a very detailed and complete documentation has been created and made available for users and developers willing to learn more about the system. Even if all major issues have been fixed, the changes done produced a completely different code base which may benefit from another audit process, perhaps with a restricted scope made out of the contracts that had more changes. We recommend the Endaoment team to do extensive beta testing on their system before releasing it to production.

Critical severity

[C01] Governance restrictions can be bypassed

Update: Fixed in pull requests #76 and #77. The Endaoment team created a new EndaomentAdminStorage contract that keeps track of the current EndaomentAdmin implementation saved in the endaomentAdmin parameter. This contract extends from theAdministratable one and also provides a restricted method to eventually change the address of the implementation. The factories now extend from such contract and their addresses are stored in the Fund and Org contract whenever a new one is created. In this way the endaomentAdmin of both factories is passed as input parameter in the Administratable modifiers whenever a restricted function of the Fund or Org contract is called. It has to be noticed that, before interacting with the system, the correct roles must be set in the EndaomentAdmin contract configured in the EndaomentAdminStorage one. Even if this fixes the issue, an address is still passed as input parameter in the Administratable modifiers and a new contract has been generated, slightly increasing the gas cost.

The entire system relies on the EndaomentAdmin contract to restrict access to functions and to separate user roles.
The creator of the EndaomentAdmin contract is the one in charge of setting specific roles to addresses.

The Fund, FundFactory, Org, and OrgFactory contracts all extend from the Administratable contract.
The Administratable contract defines the onlyAdmin and onlyAdminOrRole modifiers, used to restrict function calls in the derived contracts.
These modifiers receive the adminContractAddress as input parameter and use it to instantiate the governance contract EndaomentAdmin.
They then ask to the EndaomentAdmin contract at the specified address if the msg.sender has the proper rights and it eventually reverts if it’s not the case.

The issue is that all the functions making use of these modifiers are public and receive an arbitrary adminContractAddress as input parameter.
Functions restricted by the onlyAdminOrRole modifier are:

While the onlyAdmin modifier is exclusively used to protect FundFactory and OrgFactory constructors.

This opens the doors for the following attack:

  • Alice deploys her own EndaomentAdmin contract and sets her address as having full access (i.e. by setting every role to her address).
  • She can now call any of the restricted functions listed above or deploy new factories, passing the address of her malicious EndaomentAdmin contract as adminContractAddress in the function call parameters.

There are several ways now Alice can benefit from this scenario by just using the malicious adminContractAddress:

  • Every Org contract instance, if it has been deployed by the OrgFactory, would result in an allowed organization. The createGrant function of the Fund contract checks allowed organizations calling the checkRecipient function. Alice can:
  • Create a malicious Org contract by calling the createOrg function of the OrgFactory contract making it an allowed organization.
  • Change the manager of a Fund contract setting her address instead, by calling the changeManager function .
  • Call the createGrant function which has the restricted modifier ensuring that the msg.sender == manager. She will bypass the require statement in line 101 and successfully create a grant to her malicious organization.

     

  • She can also finalize any grant moving out of the contract the funds needed to cover it by calling the finalizeGrant function. Together with the previous scenario, she can definitely steal all the funds from the contract.

  • Once every fund is transferred out to the malicious organization she can cash out everything by calling the cashOutOrg function of the
    Org factory.
  • She can finally change the orgWallet of any of the deployed Org contracts by calling the setOrgWallet function or approve any arbitrary malicious claim by calling the approveClaim function.

Moreover, all the constructors of the cited contracts are protected by both modifiers, where the adminContractAddress is passed as constructor parameter. Since these modifiers can be skipped, the constructors of the factories and of Fund and Org contracts can be easily called.

To solve the issue consider doing the following:

  • Set the current implementation of the EndaomentAdmin contract in the constructor of the Administratable contract whenever it is called by the derived contract constructors. Save it as state variable.
  • Remove the adminContractAddress input parameter from any function making use of it and from the modifier definitions.
  • Change the modifier implementations to ask directly to the previously set EndaomentAdmin contract for msg.sender roles.
  • Implement an auxiliary function, with restricted access, to change, eventually, the address of the correct EndaomentAdmin implementation.

[C02] Fees are wrongly calculated

Update: Partially fixed in pull request #64. Fees can be still evaluated to 0 but the Endaoment team acknowledged this, given the fact that it’s an uncommon outcome and that grant values are bounded by the web application.

In the finalizeGrant function of the Fund contract, the fee parameter is calculated as grant.value/100. The entire system assumes that fees are paid to the admin of the fund whenever a grant is finalized.

Due to the Solidity truncating rule for division, if grant.value < 100, then fee will be zero and nothing will be paid to the admin.

The rounding error introduced when dividing will always round down the fees paid to the admin.

Consider handling the rounding errors using a library for decimals or introducing specific logic to handle them. Moreover, consider calculating the finalGrant variable as the grant.value - fee. This would make consistent the fact that both, if summed up, equal the grant.value.

High severity

[H01] Unsupported ERC20 tokens can be stuck in the contract

Update: Fixed in pull request #68.

Contract interfaces are contracts that declare function signatures without implementing them. They are used as an specification that decouples the signature of the functions from their implementation, allowing to interoperate with different implementations as long as they adhere to the specification.
In the codebase, the ERC20 token implementation by OpenZeppelin is used in the Fund and Org contracts, instead of using the interface.
The address of the ERC20 token to be used is passed as input parameter.

Given the fact that one can pass arbitrary addresses whether or not they represent valid ERC20 tokens and that not every ERC20 contract is implemented exactly as the OpenZeppelin one, it can happen that transactions calling those functions fail due to a mismatch between the ERC20 OpenZeppelin contract and the target implementation represented by the tokenAddress.

Another consequence is that, since anyone can transfer arbitrary ERC20 tokens to the Fund or Org contract, if someone accidentally deposits an ERC20 token which has a different implementation from the OpenZeppelin one, those funds are stuck forever in the contract and can’t be transferred out.

As suggested in [N03], consider using the IERC20 interface importing it from the openzeppelin-contracts package to allow interoperation with any possible implementation of the ERC20 token standard and avoid the loss of the funds deposited in the contracts.

[H02] Not following the Checks-Effects-Interactions pattern

Update: Fixed in pull request #63.

The finalizeGrant function of the Fund contract is setting the grant.complete storage variable after a token transfer.

Solidity recommends the usage of the Check-Effects-Interaction Pattern to avoid potential security issues, such as reentrancy.

The finalizeGrant function can be used to conduct a reentrancy attack, where the token transfer in line 129 can call back again the same function, sending to the admin multiple times an amount of fee, before setting the grant as completed.

In this way the grant.recipient can receive less than expected and the contract funds can be drained unexpectedly leading to an unwanted loss of funds.

Consider always following the “Check-Effects-Interactions” pattern, thus modifying the contract’s state before making any external call to other contracts.

[H03] Token transfers can silently fail

Update: Fixed in pull request #69.

The finalizeGrant function of the Fund contract and the
cashOutOrg function of the Org contract are both doing token transfers from the contract to some recipients.

In both cases, the balanceOf(sender) is not checked to be greater than 0 or greater or equal than the amount that is going to be transferred to the recipient.
Moreover, the returned values of the transfer methods, if present, are not checked.

If the transfer fails without reverting the transaction, an event is emitted or a grant is erroneously marked as completed.

To avoid finalizing grants without actually transferring tokens to the recipient organization, consider always checking the result of a token transfer and to validate first the amount to be transferred. This could also save some gas by not emitting unnecessary events.

Medium severity

[M01] Lack of input validation

Update: Fixed in pull request #74.

There is a general lack of input validation in the entire code base. Some examples are:

  • The onlyAdmin and onlyAdminOrRole modifiers, the changeManager, finalizeGrant, checkRecipient and getSummary of the Fund contract and the cashOutOrg and setOrgWallet of the Org contract, are not checking whether the addresses passed as input parameter are the zero address.
  • The createGrant function of the Fund contract doesn’t check any of the grant input parameters before creating a grant and push it into the grants array. Only the recipient is checked to be an allowed organization.
  • The constructor of the Org contract is not checking if the passed ein value is zero.
  • The claimRequest function of the Org contract is not validating if the fName, lName and eMail parameters are empty.
  • The approveClaim function of the Org contract and the finalizeGrant of the Fund contract are not checking if the index parameter is accessing to a non existing element in the corresponding array.

The lack of validation on user-controlled parameters may result in erroneous or failing transactions. Note also that some user interfaces may default to sending null parameters if none are specified.

Consider reviewing the entire code base and add input validations, especially when an user-controlled input parameter is used to store a value in the contracts.

[M02] Non flexible data objects

Update: Fixed in pull request #75.

Some data objects used in the code like the grants[] of the Fund contract, the allowedOrgs of the OrgFactory contract, or the claims[] of the Org contract are all lacking of auxiliary functions to modify their elements after creation.

In particular, allowed organizations cant be disallowed, grants can’t be modified nor deleted after being finalized, and organization claims can’t be modified.

Moreover, claims[] and grants[] have no protection from duplicates when new elements are added.

In order to handle unexpected errors and increase the flexibility of the system, consider adding auxiliary functions, with the proper access restriction, to modify and delete, when necessary, such objects from the arrays or mappings.
Also consider adding a duplicate protection to the arrays using a key-value mapping to rapidly check if an element has been already added or not.

Low severity

[L01] Missing error messages

Update: Fixed in pull request #53.

There are several require statements in the Fund and EndaomentAdmin contracts without error messages. In particular:

Consider including specific and informative error messages in all require statements.

[L02] README file is missing important information

Update: Fixed in pull request #60.

The README.md of the Endaoment project has little information about what is the purpose of the project nor how to use it. README files on the root of git repositories are the first documents that most developers often read, so they should be complete, clear, concise, and accurate.

Consider following Standard Readme to define the structure and contents for the README file.
Also, consider including an explanation of the core concepts of the repository, the usage workflows, the public APIs and how the code relates to other key components of the project.

Furthermore, it is highly advisable to include instructions for the responsible disclosure of any security vulnerabilities found in the project.

[L03] Missing docstrings

Update: Fixed in pull request #47. Moreover some of the mentioned functions are not existing anymore or they have been refactored.

The setOrgWallet, getTokenBalance and the getClaimsCount functions of the Org contract are lacking any sort of docstrings.

This hinders understanding of the code’s intention, which is fundamental to correctly assess not only security, but also correctness. Additionally, docstrings improve readability and ease maintenance.

Consider thoroughly documenting the missing functions (and their parameters) that are part of the contracts’ public API.

[L04] SafeMath library is not used

Update: Fixed in pull request #54.

In the code base, the SafeMath contract library is declared to be used against uint256 variables in the Fund contract, but then it’s not used in lines 127-128.

To avoid any possible underflow/overflow or division by zero, consider always relying on the proven safety of the SafeMath library by OpenZeppelin and using it when performing mathematical operations. Also consider removing line 19 of the Org contract if the contract doesn’t need to perform calculations.

[L05] Misleading docstrings

Update: Partially fixed in pull request #78. Most of the docstrings have been fixed, some of them have been added or refactored.

Several docstrings throughout the code base were found to contain misleading information and should be fixed. In particular:

  • Line 16 of the Fund contract is incorrectly assuming that the token transfer is done using the SafeMath library, but since the token address used to do the transfer is arbitrary, then the transfer implementation can be different and then the assumption doesn’t hold.
  • Lines 14,15 and 16 of the FundFactory contract state that it provides a way for fetching individual Org contract addresses and a list of allowedOrgs but there is no way of retrieving such information from the contract.
  • Lines 231 and 232 of the EndaomentAdmin contract are stating that the permitted roles to pass the modifier’s restrictions have to be a “bot commander (0)” or a “pauser (1)”. In reality, the modifier can be called with any of the six roles permitted and it only checks if that role exists or if it’s paused. Moreover it’s not clear what “bot commander” means.
  • Line 182, 172 and 158 of the EndaomentAdmin contract say “External” when the function visibility is public.
  • Lines 27 an 28 of the Administratable contract, the comment says that the only admitted roles are ADMIN, ACCOUNTANT and REVIEWER while the modifier actually checks also for the FUND_FACTORY and ORG_FACTORY roles.
  • Line 184 of the EndaomentAdmin contract says admin where it should be the holder or roleAddress as the returned named parameter. The same happens in the IEndaomentAdmin.getRoleAddress function definition.
  • Line 79 and 116 of the Fund contract and line 88 of the Org contract are mentioning stablecoins. There is actually no restriction on the type of ERC20 tokens that can be passed as input parameter in the functions making use of them and any mention should therefore be removed.

[L06] Multiple getters for the same state variable

Update: Fixed in pull request #73.

Some contracts in the code base contain multiple public getter functions for the same state variable.

For example:

To avoid duplication, ensure that there is at most one publicly exposed getter for each contract state variable. Consider making all the state variables private, following the style of the openzeppelin-contracts package.

[L07] Useless input parameters

Update: Fixed in pull request #66.

In the claimRequest function of the Org contract, the orgAdminAddress is passed as input parameter. In line 60 then, the msg.sender is required to be the same address as the orgAdminAddress.
The same happens with the fSub parameter which is required, in line 59 to be true.

The function is public and anyone can call it. The msg.sender can freely choose their address as orgAdminAddress and fSub as true.

Given the fact that the orgAdminAddress and fSub are not used for any other purpose we can conclude that the require statements in line 59 and 60 are useless and so the orgAdminAddress and fSub arguments. Consider removing them.

[L08] Lack of event emission

Update: Fixed in pull request #52.

The changeManager, createGrant, finalizeGrant functions of the Fund contract, the claimRequest and the setOrgWallet functions of the Org contract are not emitting any event after changing the value of state variables.

To easily inform clients about state changes in the contract and to let the applications subscribe to such changes, consider always emitting an event when changing important parameters of each contract.

[L09] Missing getter

Update: Fixed in pull request #55.

The _newPotentialOwner private state variable of the TwoStepsOwnable contract is missing a getter function to read its value.

To have access to this state variable without parsing the contract storage in the blockchain, consider adding a agetter function to retrieve its value.

[L10] Incorrect require statement

Update: Fixed in pull request #56.

In the Administratable contract the onlyAdminOrRole modifier in line 48 is requiring the msg.sender to be the ADMIN address of the EndaomentAdmin contract, but this statement is inside an conditional structure with the condition that the msg.sender is not the ADMIN.
So the require statement will be never satisfied and will always revert if evaluated.

This require statement should reflect that the caller is not the ADMIN or that the role requested is paused and should therefore be a revert with an explicative error message if the else block is evaluated.

Consider refactoring the conditional statements by providing a cleaner structure, and reverting when the caller is accessing to a paused role and it’s not the ADMIN.

Notes & Additional Information

[N01] uint as uint256

Update: Fixed in pull request #49.

There are several cases in the code base where the uint keyword is used implicitly assuming the uint256 type.
Some examples can be line 81 of the Fund contract, and line 44 of the Org contract.

To favor explicitness, all instances of uint should be declared as uint256.

[N02] Unused state variables

Update: Fixed in pull request #46.

In the Fund contract, the state variables contributors and totalContributors are never used and should therefore be removed.

 

[N03] Unused contracts

Update: Fixed in pull request #50.

There are files which are not used at all in the code base:

To improve readability, security and clarity of the code, consider replacing the ERC20 import with the IERC20 interface and removing the SafeMath file, importing it, where needed, directly from the OpenZeppelin contracts.

[N04] Style recommendations not applied

Update: Fixed in pull request #51. Solhint is now used as linter tool.

The style used in the entire code base is following the Solidity style recommendations, however there are still some parts in the code that have some deviations. In particular:

Taking into consideration how much value a consistent coding style adds to the project’s readability, enforcing a standard coding style with help of linter tools such as Solhint is recommended.

[N05] Incorrect functions visibility

Update: Fixed in pull request #62.

Whenever a function is not being called internally in the code, it can be easily declared as external, saving also gas. While the entire code base have explicit visibilities for every function, some of them can be changed to be external.
Examples are the getters in the OrgFactory contract or each one of the public functions of the EndaomentAdmin contract.

To improve clarity following Solidity recommendations and to better reflect the scope of each function, consider reviewing the visibilities applied.

[N06] Unnecessary import

Update: Fixed in pull request #45.

In the FundFactory contract, consider removing the import statement for OrgFactory, as this contract is never used in the contract’s code.

[N07] Fund contract can’t receive ethers

Update: Fixed in pull request #61.

The Fund contract is in charge of recollecting the funds to later support organizations with the recollected tokens. The getSummary function is returning the ETH balance of the contract but the contract has no payable functions to receive them.

Consider whether removing any reference to ETH if the contract is not intended to use it as a valid currency, or implementing the missing functionality to correctly receive and move them out from the contract.

[N08] Typos in the code

Update: Fixed in pull request #42.

There are some typos to fix in the code base. In particular:

  • Line 76 of the Org contract is repeated.
  • In line 59 of the OrgFactory contract, “is provided” should be “if provided”.
  • In line 16 of the Org contract, “direct received” should be “directly receive”.
  • In the docstring and in the input parameter of the cashOutOrg function of the Org contract the desiredWithdrawlAddress and withdrawl words should be changed to desiredWithdrawalAddress and withdrawal.
  • In the Fund contract the word reccomendation should be changed for recommendation in several lines.
  • In the Administratable contract in line 8 Provides a of modifiers should be Provides two modifiers.

[N09] Unclear function purpose

Update: Fixed in pull request #48.

The approveClaim function of the Org contract is having the only purpose of calling the setOrgWallet function with the claim.desiredWallet of the claim chosen by the index input parameter.
On the other side, the setOrgWallet function is public but called internally only by the approveClaim function and it is a one line function setting the orgWallet state variable to the providedWallet input parameter.
Moreover the orgWallet variable is not used anywhere in the code base and it’s not clear what’s its purpose.

If is an explicit design decision, consider commenting on the docstrings why approveClaim is approving a claim or why setOrgWallet sets that variable and what’s its purpose.

[N10] Default value as sensible information

Update: Fixed in pull request #44.

In the list of all possible user’s roles in the IEndaomentAdmin interface, the ADMIN role is the 0 value of the enum structure which is the default value of the integer type. Relying the ADMIN role on such value can be risky and should be avoided since many clients can default to that value if none is provided.

Consider moving the ADMIN role to the end of the list and to assign to the 0 value an EMPTY role.

Conclusions

2 critical and 3 high severity issues were found. Some changes were proposed to follow best practices and reduce potential attack surface.

Update: All the reported issues have been addressed and fixed by the Endaoment team.