feat: rely call detector [PoC]

[wei3erHase]

Description

The following PoC shows a way of detecting specific types of calls, focusing on probably the most critical: rely(address). Considering that if a wrong address is given throughout the execution of the spell, the whole protocol could be at risk (same as if an authorization is denied). This PoC aims to provide the process some tools to review that the spell is acting accordingly.

Considering only rely(address), there are 2 possible scenarios: either the spell needs to make one of these calls (and give authorization to an address), or the spell does not. The testing suite then incorporates 2 tests, one for each of the possible scenarios:

• testFailAuthCall: will ✅ only if NO rely(address) calls were made in spell execution
• testAuthCall: requires the crafter to declare all expected rely(address) calls (will ✅ if ALL of these calls are made)

Taking into account that an alien auth call could be made in the second scenario (when we expect to see a rely event), the reviewer checklist should add something like:

- [ ] check `[PASS] testFailAuthCall`
- [ ] If `[SKIP] testFailAuthCall`: check the logs bloom and verify there's not other rely event than in `testAuthCall`

In this way, adding an automated way of testing that there are no new authorizations, and a manual way of checking them when there are some, we can entrust that no new unwanted authorization is made.

Considerations:

• rely(address) comes from the legacy Note library, the test should be duplicated with new Rely(address) event
• the test should also be duplicated with deny(address) and Deny(address)
• adds manual labour to reviewer, tho only in special ocations (when there are new auths)

[amusingaxl]

Shouldn't we validate at least topic 1 here?

Suggested change

	vm.expectEmit(false, false, false, false);
	vm.expectEmit(true, false, false, false);

This way, it would only fail if rely is called somewhere, otherwise, I guess any LogNote would trigger a failure.

[amusingaxl]

Also we do have some contracts that use proper Rely(address) events instead of LibNote, so this test would have to cover that as well.

[wei3erHase]

another test needs to cover it, same structure, only expecting emit a Rely(address) (we cannot put both in the same test), as the expectation would only fail if BOTH the expectations are fulfilled

[wei3erHase]

the 1st topic is ALWAYS validated, the 2nd topic (which corresponds to the 1st boolean) would be:

• for Note: address indexed usr
• for Rely(address): address

[wei3erHase]

otherwise, I guess any LogNote would trigger a failure

beware that LogNote is an anonymous event, meaning, it does NOT have a topic 0 == keccak(LogNote), the topic0 IS the bytes4 indexed sig