Repository navigation
Description
Hey, second one from me today. I filed the stale pool denylist issue earlier as #513. This one is in the same function, different check.
Problem
check_for_denylisted_addresses (crates/execution-txpool/src/validator.rs:333-359) checks the sender, the to address, and the recovered authorities of the EIP-7702 authorization list, but never the authorization's address field, the delegate contract whose code is installed on the authority. A tx whose authorization delegates to a denylisted contract validates fine, so the denylisted contract's code executes in the authority's context whenever the authority is called.
The team clearly considers 7702 delegation in scope for the denylist. The authority check exists with dedicated tests (eip7702_denylisted_authority_rejected, eip7702_multiple_auths_second_denylisted_rejected). The delegate is the other half of the same authorization tuple and was missed.
What breaks
- A 7702 tx delegating to a denylisted address validates Valid.
- The control, denylisting the authority instead, is rejected. So the gap is specific to the delegate field.
Reproduction
Repro test crates/execution-txpool/tests/repro_round2.rs::repro_7702_denylisted_delegate_accepted (release/0.8):
- Denylist the delegate 0xdddddddddddddddddddddddddddddddddddddddd. The 7702 tx validates Valid.
- Same test with a blocklisted delegate also validates Valid.
- Control: denylisted authority is rejected.
Output:
7702 tx with DENYLISTED delegate outcome: Valid { ... authorization_list: [SignedAuthorization { inner: Authorization { chain_id: 1, address: 0xdddddddddddddddddddddddddddddddddddddddd, ... }], ... authorities: Some([0xa84585fb6728f413d4d89ec972c45e94686bf38e]) }
CONTROL OK: denylisted authority rejected, denylisted delegate accepted
test repro_7702_denylisted_delegate_accepted ... ok
7702 tx with BLOCKLISTED delegate outcome: Valid { ... }
test repro_7702_blocklisted_delegate_accepted ... ok
Exact location
crates/execution-txpool/src/validator.rs:340-348:
let auth_authorities: Vec<Address> = transaction
.authorization_list()
.into_iter()
.flatten()
.filter_map(|auth| auth.recover_authority().ok())
.collect();
// `auth.address` (the delegate) is never collected or checked.Fix
In check_for_denylisted_addresses, also collect auth.address for each authorization in the list and run it through is_denylisted, same treatment as authorities. One SLOAD per delegate, bounded by the existing MAX_AUTHORIZATIONS_PER_TX = 100 cap, which is enforced before the denylist check.
What this is not
This is not #498. That one is about which code runs when the delegate is a precompile address. This is about whether a denylisted delegate is admitted at all. Different question, different fix.
Activity
Jr-kenny commentedon Nov 13, 2026
can I take this one as well please ser?
osr21 commentedon Nov 13, 2026
I checked the release/0.8 validator: it checks the transaction sender, recipient, and recovered EIP-7702 authorities, but does not include each authorization tuple's address (the delegation target). The omission described here appears present.
One scope distinction may help before settling the fix: adding the target to txpool admission checks prevents new set-code transactions from selecting a denylisted delegate, but it does not stop execution through accounts that already hold a delegation to that target, or establish a rule for payloads that bypass txpool admission. If the policy is that denylisted delegate code must never execute, that guarantee needs an execution/block-validation check as well (the same admission-versus-execution distinction raised in #513). If the intended rule is only to reject new delegations, documenting that narrower scope would avoid overclaiming enforcement.
For tests, cover multiple authorization tuples and the EIP-7702 edge cases for invalid chain-id/nonce entries and zero-target clearing semantics, so the check is aligned with what the transaction actually applies. This is separate from #498, which concerns EIP-7702 code resolution for precompile targets.
References: validator on release/0.8 · EIP-7702 · #513 · #498.