This is an opinionated list of issues that can be improved if they are found in Cadence code intended for production.
Security and Robustness
Avoid using AuthAccount as a function parameter
Problem
Some may choose to authenticate or perform operations for their users by using the users' account addresses.
In order to do this, a commonly seen case would be to pass the user's AuthAccount object
as a parameter to a contract function to use for querying the account or storing objects directly.
This is problematic, as the AuthAccount object allows access to ALL private areas of the account,
for example, all of the user's storage, authorized keys, etc.,
which provides the opportunity for bad actors to take advantage of.
Example:
_37..._37// BAD CODE_37// DO NOT COPY_37_37// Imagine this code is in a contract that uses AuthAccount to authenticate users_37// To transfer NFTs_37_37// They could deploy the contract with an Ethereum-style access control list functionality_37_37pub fun transferNFT(id: UInt64, owner: AuthAccount) {_37 assert(owner(id) == owner.address)_37_37 transfer(id)_37}_37_37// But they could upgrade the function to have the same signature_37// so it looks like it is doing the same thing, but they could also drain a little bit_37// of FLOW from the user's vault, a totally separate piece of the account that_37// should not be accessible in this function_37// BAD_37_37pub fun transferNFT(id: UInt64, owner: AuthAccount) {_37 assert(owner(id) == owner.address)_37_37 transfer(id)_37_37 // Sneakily borrow a reference to the user's Flow Token Vault_37 // and withdraw a bit of FLOW_37 // BAD_37 let vaultRef = owner.borrow<&FlowToken.Vault>(/storage/flowTokenVault)!_37 let stolenTokens <- vaultRef.withdraw(amount: 0.1)_37_37 // deposit the stolen funds in the contract owners vault_37 // BAD_37 contractVault.deposit(from: <-stolenTokens)_37}_37...
Solution
Projects should find other ways to authenticate users, such as using resources and capabilities as authentication objects. They should also expect to perform most storage and linking operations within transaction bodies rather than inside contract utility functions.
There are some scenarios where using an AuthAccount object is necessary, such as a cold storage multi-sig,
but those cases are extremely rare and AuthAccount usage should still be avoided unless absolutely necessary.
Auth references and capabilities should be avoided
Problem
Authorized references allow downcasting restricted
types to their unrestricted type and should be avoided unless necessary.
The type that is being restricted could expose functionality that was not intended to be exposed.
If the auth keyword is used on local variables they will be references.
References are ephemeral and cannot be stored.
This prevents any reference casting to be stored under account storage.
Additionally, if the auth keyword is used to store a public capability, serious harm
could happen since the value could be downcasted to a type
that has functionality and values altered.
Example
A commonly seen pattern in NFT smart contracts is including a public borrow function that borrows an auth reference to an NFT (eg. NBA Top Shot). This allows anyone to access the stored metadata or extra fields that weren't part of the NFT standard. While generally safe in most scenarios, not all NFTs are built the same. Some NFTs may have privileged functions that shouldn't be exposed by this method, so please be cautious and mindful when imitating NFT projects that use this pattern.
Another Example
When we create a public capability for our FungibleToken.Vault we do not use an auth capability:
_10// GOOD: Create a public capability to the Vault that only exposes_10// the balance field through the Balance interface_10signer.link<&FlowToken.Vault{FungibleToken.Balance}>(_10 /public/flowTokenBalance,_10 target: /storage/flowTokenVault_10)
If we were to use an authorized type for the capability, like so:
_10// BAD: Create an Authorized public capability to the Vault that only exposes_10// the balance field through the Balance interface_10// Authorized referenced can be downcasted to their unrestricted types, which is dangerous_10signer.link<auth &FlowToken.Vault{FungibleToken.Balance}>(_10 /public/flowTokenBalance,_10 target: /storage/flowTokenVault_10)
Then anyone in the network could take that restricted reference that is only supposed to expose the balance field and downcast it to expose the withdraw field and steal all our money!
_10// Exploit of the auth capability to expose withdraw_10let balanceRef = getAccount(account)_10 .getCapability<auth &FlowToken.Vault{FungibleToken.Balance}>(/public/flowTokenBalance)_10 .borrow()!_10_10let fullVaultRef = balanceRef as! &FlowToken.Vault_10_10// Withdraw the newly exposed funds_10let stolenFunds <- fullVaultRef.withdraw(amount: 1000000)
Events from resources may not be unique
Problem
Public functions in a contract can be called by anyone, e.g. any other contract or any transaction. If that function creates a resource, and that resource has functions that emit events, that means any account can create an instance of that resource and emit those events. If those events are meant to indicate actions taken using a single instance of that resource (eg. admin object, registry), or instances created through a particular workflow, it's possible that events from other instances may be mixed in with the ones you're querying for - making the event log search and management more cumbersome.
Solution
To fix this, if there should be only a single instance of the resource,
it should be created and link()ed to a public path in an admin account's storage
during the contracts's initializer.
Access Control
Public functions and fields should be avoided
Problem
Be sure to keep track of access modifiers when structuring your code, and make public only what should be public. Accidentally exposed fields can be a security hole.
Solution
When writing your smart contract, look at every field and function and make sure
that they are all access(self), access(contract), or access(account), unless otherwise needed.
Capability-Typed public fields are a security hole
This is a specific case of "Public Functions And Fields Should Be Avoided", above.
Problem
The values of public fields can be copied. Capabilities are value types, so if they are used as a public field, anyone can copy it from the field and call the functions that it exposes. This almost certainly is not what you want if a capability has been stored as a field on a contract or resource in this way.
Solution
For public access to a capability, place it in an accounts public area so this expectation is explicit.
Array or dictionary fields should be private
This anti-pattern has been addressed with FLIP #703
Problem
This is a specific case of "Public Functions And Fields Should Be Avoided", above. Public array or dictionary fields are not directly over-writable, but their members can be accessed and overwritten if the field is public. This could potentially result in security vulnerabilities for the contract if these fields are mistakenly made public.
Ex:
_10pub contract Array {_10 // array is intended to be initialized to something constant_10 pub let shouldBeConstantArray: [Int]_10}
Anyone could use a transaction like this to modify it:
_10import Array from 0x01_10_10transaction {_10 execute {_10 Array.shouldbeConstantArray[0] = 1000_10 }_10}
Solution
Make sure that any array or dictionary fields in contracts, structs, or resources
are access(contract) or access(self) unless they need to be intentionally made public.
_10pub contract Array {_10 // array is inteded to be initialized to something constant_10 access(self) let shouldBeConstantArray: [Int]_10}
Public admin resource creation functions are unsafe
This is a specific case of "Public Functions And Fields Should Be Avoided", above.
Problem
A public function on a contract that creates a resource can be called by any account. If that resource provides access to admin functions then the creation function should not be public.
Solution
To fix this, a single instance of that resource should be created in the contract's init() method,
and then a new creation function can be potentially included within the admin resource, if necessary.
The admin resource can then be link()ed to a private path in an admin's account storage during the contract's initializer.
Example
_36// Pseudo-code_36_36// BAD_36pub contract Currency {_36 pub resource Admin {_36 pub fun mintTokens()_36 }_36_36 // Anyone in the network can call this function_36 // And use the Admin resource to mint tokens_36 pub fun createAdmin(): @Admin {_36 return <-create Admin()_36 }_36}_36_36// This contract makes the admin creation private and in the initializer_36// so that only the one who controls the account can mint tokens_36// GOOD_36pub contract Currency {_36 pub resource Admin {_36 pub fun mintTokens()_36_36 // Only an admin can create new Admins_36 pub fun createAdmin(): @Admin {_36 return <-create Admin()_36 }_36 }_36_36 init() {_36 // Create a single admin resource_36 let firstAdmin <- create Admin()_36_36 // Store it in private account storage in `init` so only the admin can use it_36 self.account.save(<-firstAdmin, to: /storage/currencyAdmin)_36 }_36}
Do not modify smart contract state or emit events in public struct initializers
This is another example of the risks of having publicly accessible parts to your smart contract.
Problem
Data structure definitions in Cadence currently must be declared as public so that they can be used by anyone. Structs do not have the same restrictions that resources have on them, which means that anyone can create a new instance of a struct without going through any authorization.
Solution
Any contract state-modifying operations related to the creation of structs should be contained in restricted resources instead of the initializers of structs.
Example
This used to be a bug in the NBA Top Shot smart contract, so we'll use that as an example. Before, when it created a new play, it would initialize the play record with a struct, which increments the number that tracks the play IDs and emits an event:
_23// Simplified Code_23// BAD_23//_23pub contract TopShot {_23_23 // The Record that is used to track every unique play ID_23 pub var nextPlayID: UInt32_23_23 pub struct Play {_23_23 pub let playID: UInt32_23_23 init() {_23_23 self.playID = TopShot.nextPlayID_23_23 // Increment the ID so that it isn't used again_23 TopShot.nextPlayID = TopShot.nextPlayID + 1_23_23 emit PlayCreated(id: self.playID, metadata: metadata)_23 }_23 }_23}
This is a risk because anyone can create the Play struct as many times as they want,
which could increment the nextPlayID field to the max UInt32 value,
effectively preventing new plays from being created. It also would emit bogus events.
This bug was fixed by instead updating the contract state in the admin function that creates the plays.
_34// Update contract state in admin resource functions_34// GOOD_34//_34pub contract TopShot {_34_34 // The Record that is used to track every unique play ID_34 pub var nextPlayID: UInt32_34_34 pub struct Play {_34_34 pub let playID: UInt32_34_34 init() {_34 self.playID = TopShot.nextPlayID_34 }_34 }_34_34 pub resource Admin {_34_34 // Protected within the private admin resource_34 pub fun createPlay() {_34 // Create the new Play_34 var newPlay = Play()_34_34 // Increment the ID so that it isn't used again_34 TopShot.nextPlayID = TopShot.nextPlayID + UInt32(1)_34_34 emit PlayCreated(id: newPlay.playID, metadata: metadata)_34_34 // Store it in the contract storage_34 TopShot.playDatas[newPlay.playID] = newPlay_34 }_34 }_34}