CVE-2023-31146: OOB DynArray access when array is on both LHS and RHS of an assignment

Impact

during codegen, the length word of a dynarray is written before the data, which can result in OOB array access in the case where the dynarray is on both the lhs and rhs of an assignment. here is a minimal example producing the issue:

a:DynArray[uint256,3] @external def test() -> DynArray[uint256,3]: self.a = [1,2,3] self.a = empty(DynArray[uint256,3]) self.a = [self.a[0],self.a[1],self.a[2]] return self.a # return [1,2,3]

and here is an example demonstrating the issue can cause data corruption across call frames:

@external def test() -> DynArray[uint256,3]: self.a() return self.b() # return [1,2,3]

@internal def a(): a: uint256 = 0
b: uint256 = 1
c: uint256 = 2
d: uint256 = 3

@internal def b() -> DynArray[uint256,3]: a: DynArray[uint256,3] = empty(DynArray[uint256,3]) a = [a[0],a[1],a[2]] return a

examples involving append and pop:

@internal def foo(): c: DynArray[uint256, 1] = [] c.append(c[0])

@internal def foo(): c: DynArray[uint256, 1] = [1] c[0] = c.pop()

the expected behavior in all of the above cases is to revert due to oob array access.

Patches

Has the problem been patched? What versions should users upgrade to?

Workarounds

Is there a way for users to fix or remediate the vulnerability without upgrading?

References

Are there any links users can visit to find out more?