Repository navigation
Description
Hey, fifth one from me today, same denylist area as my earlier filings. This one is about CREATE transactions.
Problem
The denylist path checks the sender, the to address, and the authorities. For a CREATE tx, to() is None, so the address the tx deploys to is never checked. The blocklist path handles exactly this case, in the same function family (validator.rs, check_for_blocklisted_addresses: TxKind::Create => transaction.sender().create(transaction.nonce()), with four dedicated tests including a queued nonce case). The denylist path does not mirror it.
If Circle pre-denylists a future deployment address, derivable from a known deployer and nonce, a value-bearing CREATE funding that address is admitted to the pool, and because execution never checks the denylist, the value lands.
What breaks
- A value-bearing CREATE to a pre-denylisted derived deployment address is admitted (Valid).
- The same address reached by a plain CALL is rejected with DenylistedAddressError.
- A CREATE to a clean derived address is admitted, as it should be. So the gap is specific to the denylist path never deriving the CREATE address.
I am stating the honest weakness plainly: this needs Circle to have pre-denylisted a not yet existing address. The reason to file it is the inconsistency with the blocklist path, same file, same function family, explicit tests, plus the fact that execution never checks the denylist, so if it happens, the value lands.
Reproduction
Repro tests (release/0.8):
- crates/execution-txpool/tests/repro_denylist_r2.rs::repro_create_to_denylisted_derived_address_accepted (asserts acceptance; controls: denylisted sender CREATE rejected, blocklisted derived address CREATE rejected).
- crates/execution-txpool/tests/repro_round2_create_denylist.rs::probe_create_to_denylisted_derived_address (probe style).
Output (cargo test -p arc-execution-txpool --test repro_denylist_r2 -- --nocapture, 3 passed, 0 failed):
T3 CREATE to pre-denylisted derived address outcome: Valid { ... transaction: Valid(Legacy { ..., sender: 0x6ac7ea33f8831ea9dcc53393aaa88b25a785dbf0, nonce: 0, ..., to: Create, value: 1000, ... }), propagate: true, authorities: None }
test repro_create_to_denylisted_derived_address_accepted ... ok
CONTROL call to denylisted address outcome: Invalid(..., Other(DenylistedAddressError(0xcd234a471b72ba2f1ccf0a70fcaba648a5eecd8d)))
test control_call_to_denylisted_address_rejected ... ok
CONTROL CREATE to clean derived 0xcd234A471b72ba2F1Ccf0A70FCABA648a5eeCD8d outcome: Valid { ... }
test control_create_to_clean_derived_address_accepted ... ok
Probe file, 1 passed:
CREATE to pre-denylisted derived address outcome: Valid { ... to: Create, value: 1000, ... }
PROBE RESULT: value-bearing CREATE to pre-denylisted derived address was ACCEPTED (denylist gap, cf. blocklist path which rejects this)
test probe_create_to_denylisted_derived_address ... ok
Fix
Mirror the blocklist path in check_for_denylisted_addresses: for TxKind::Create, also check transaction.sender().create(transaction.nonce()). CREATE2 deployment addresses are not derivable from the tx alone without parsing init code; the blocklist path has the same limitation at mempool and relies on execution, out of scope here since execution has no denylist check at all, which is the #513 / #495 area.
What this is not
This is not #496. That one is about genesis builders accepting proxy addresses for the Denylist contract, a different area entirely. This is a distinct admission gap, CREATE address derivation, not the stale pool issue in #513 and not the 7702 delegate issue.
Activity