Skip to content

fix(authority): a hosted RevokeActor that touches a person needs an owner key's proof - #10006

Open
olety wants to merge 4 commits into
mainfrom
opus-ownerkey-p1/hosted-revoke-actor-needs-owner-proof
Open

olety wants to merge 4 commits into
mainfrom
opus-ownerkey-p1/hosted-revoke-actor-needs-owner-proof

Conversation

@olety

@olety olety commented Nov 20, 2026 •

Copy link
Copy Markdown
Member

Row [HOSTED-OWNER-BIND-ROOT-REVOKE] (P3, filed 16 Nov), made live by [HOSTED-RELAY-OWNER-HUMAN-BIND-CEREMONY] (lead ruling B, 21 Nov 04:3x), which adds the hosted human-bind writer this row waited for. Engine ANSWER BIND-CEREMONY-REVOKE-ACTOR-PROOF = A (21 Nov 04:54), following the lead's 15 Nov 20:41 rule: "a revoke of an owner key needs an owner key's signature, never only the custodial root". Base of a two-PR stack; the ceremony stacks on it.

What

On a hosted fold, owner_proof::human_bind_proved now judges a RevokeActor as it judges a human bind. A revoke counts only when one of its participants proves for the owner (an owner key outside the host's custody that no passkey limit holds) if the revoke touches a person, meaning any of:

  • the key's binding, live or ended, is human;
  • the key is a live roster key that the consent arm takes (an owner or admin key);
  • the revoke's parents do not hold the key: the floor of a verified revoke lands on every branch the merge joins, including a sibling that enrols and binds the key.

Without this, once an owner key has bound its owner (the ceremony), the custodial root alone could do three things:

  1. end that binding, so the owner's owner verbs answer OWNER_BINDING_REQUIRED again;
  2. make owner_live_at read the owner as removed, so every later carried act of theirs, including a new key's enrolment, is refused on every replica;
  3. raise the key's epoch floor, so the key never binds again.

A machine's or an agent's binding still ends on the root's word. A self-host root folded as hosted still proves, because the gate reads the cloud stamp. ONE-REVOKE-DURABILITY keeps a verified revoke's floor when its ancestry turns invalid. It names no signer, and a revoke whose own participants do not consent never verifies. The restriction-fact path proves each revoke on its own branch with the same hosted context (revoke_proof.rs verify_signed_branch, ..context), so a refused revoke lands no floor there either. The refusal reuses ActorBindingRejection::OwnerProofRequired.

Files

  • crates/oneiron/src/authority/owner_proof.rs: human_bind_proved and its doc.
  • crates/oneiron-records/src/authority/checkpoint_image/mod.rs: AUTHORITY_CHECKPOINT_FOLD_SCHEMA 55 -> 56, with its doc line.
  • crates/oneiron/src/authority/owner_key_enrollment_door.rs: the membership move's RevokeActor rows carry the owner proof (below).
  • crates/oneiron/tests/hosted_root_human_bind.rs: 3 red tests and 4 guards. One existing test, a_custodial_root_alone_does_not_bind_over_the_owners_revoked_binding, now has the owner key revoke its own binding before the root's bind over it.

Stored change: keep (fold schema 55 -> 56; no row shape changes). Before this stack no writer made an owner-proved human binding on a hosted vault. The one stored shape that folds differently is a custodial-root RevokeActor of a binding a self-host root made, on a vault moved onto a managed host.

Tests

Red pin 78749f3 (the tests commit on main c8a36a0). By reading, the three new tests fail there at their !valid(revoke) assertion:

  • a_custodial_root_alone_does_not_revoke_an_owner_keys_binding
  • a_custodial_root_revoke_on_old_parents_leaves_an_owner_keys_binding_active
  • a_custodial_root_alone_does_not_floor_an_owner_keys_epoch

These pass on main too:

  • an_owner_key_revokes_its_own_binding_on_a_hosted_vault
  • an_owner_key_cosign_lets_the_custodial_root_revoke_the_owners_binding
  • a_custodial_root_alone_revokes_an_agent_keys_binding_on_a_hosted_vault
  • a_self_host_root_revokes_the_owners_binding_alone

The rewritten existing test also passes on main.

Command: RUSTFLAGS="--cfg oneiron_test_support" cargo test --locked -p oneiron --all-features --test hosted_root_human_bind. It runs after the merge (READY-PROOF-VS-MERGE-BAR = A).

Membership move writer (own review of stack #10009, finding 1)

revoke_owner_authority_in_txn (crates/oneiron/src/authority/owner_key_enrollment_door.rs) ends each member key with a RevokeDevice, which ends that key's binding, and then writes a RevokeActor for each other live human binding that names the member or that a key of theirs signed. On a hosted fold that RevokeActor now needs an owner key's proof too, so (commit 3a34c7c):

  • where the root does not prove for the owner, each RevokeActor carries the cosign of the owner key this install holds (Vault::owner_proof_signer), as each RevokeDevice already did;
  • the proof key's own RevokeDevice now comes after every binding's revoke, so it is still live when it proves them;
  • with no proof key held, the root's revoke still counts where the fold takes the root as the owner's (a self-host root folded as hosted, unstamped); the refusal with no held key and member keys left is unchanged (OwnerKeyMoveRefused, step Unavailable).

Known limit

The device-confirmed move (#9093) signs only RevokeDevice rows, so its RevokeActor rows stay host-signed. A residual human binding on a key the move does not revoke can therefore be ended there only where the root proves for the owner. Today no shipped writer makes such a binding on a hosted fold: it needs a vault moved from self-host whose root then moved to a stamped root, or rows a client signed by hand. Such a move refuses with the uncounted entry and writes nothing. Extending the plan would change the OwnerKeyMovePlan verb wire, so it stays row [HOSTED-MOVE-REVOKE-ACTOR-MIGRATED-BINDING] (the commit message calls it HOSTED-MOVE-RESIDUAL-ACTOR-REVOKE).

…g on a hosted fold

Row HOSTED-OWNER-BIND-ROOT-REVOKE: on a hosted fold the custodial root alone
revokes the human binding an owner key made, on old parents too, and floors
an unbound owner key's epoch, so owner_live_at reads the owner as removed on
every replica. Three red tests and four guards in hosted_root_human_bind.rs;
the bind-over-a-revoked-binding test now has the owner key revoke its own
binding.
…wner key's proof

Row HOSTED-OWNER-BIND-ROOT-REVOKE (lead ruling HOSTED-RELAY-OWNER-HUMAN-BIND-CEREMONY
= B adds the hosted human-bind writer the row waited for). human_bind_proved
now judges a RevokeActor as it judges a human bind: on a hosted fold, a revoke
of a key whose binding is human, of a live owner or admin key, or of a key its
parents do not hold counts only when a participant proves for the owner. A
machine's or an agent's binding still ends on the root's word; a self-host
root folded as hosted still proves. The restriction-fact path proves each
revoke on its own branch with the same hosted context, so a refused revoke
lands no floor. Fold schema 55 -> 56.
@coderabbitai

coderabbitai Bot commented Nov 20, 2026 •

Copy link
Copy Markdown

Warning

Review limit reached

  • Run on-demand review

This review includes 4 billable files and costs up to $1.00.

  • Ask an admin to make reviews automatic

Open in CodeRabbit

Reviews can continue after your included limit without a manual trigger. An admin must approve usage-based billing.

Or wait 52 minutes for your next included review.

Check out review usage here.

View limit details

Limit details: You’ve used the included review currently available. Your 151 included PR review attempts over the past 7 days set your current allowance at 1 review per hour.

Learn how review limits work.

Review configuration:

Run configuration

Commits

Reviewing files that changed from the base of the PR and between 4317485 and 33054a8.

Files selected for processing: 4

  • crates/oneiron-records/src/authority/checkpoint_image/mod.rs
  • crates/oneiron/src/authority/owner_key_enrollment_door.rs
  • crates/oneiron/src/authority/owner_proof.rs
  • crates/oneiron/tests/hosted_root_human_bind.rs
  • Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

@olety
olety marked this pull request as ready for review November 20, 2026 20:04
@greptile-apps

greptile-apps Bot commented Nov 20, 2026

Copy link
Copy Markdown

Automatic reviews are disabled for this repository. Configure in Code Review Settings.

@olety
olety added this pull request to stack #10009 November 20, 2026 20:04
… owner key

A hosted fold counts a revoke of a person's binding only with an owner
key's proof (HOSTED-OWNER-BIND-ROOT-REVOKE), but the membership move still
signed each remaining human binding's RevokeActor with the host root
alone, so on a hosted vault a removal that leaves such a binding refused
where main ran it (review of stack #10009, finding 1).

Each RevokeActor now carries the owner key's cosign this install holds
where the root does not prove, and the proof key's own RevokeDevice comes
after every binding's revoke. With no proof key held, the root's revoke
still counts where the fold takes the root as the owner's (a self-host
root folded as hosted). The device-confirmed move signs no RevokeActor:
row HOSTED-MOVE-RESIDUAL-ACTOR-REVOKE.
…by its FIX-LIST id

The device-confirmed move's residual RevokeActor case is row HOSTED-MOVE-REVOKE-ACTOR-MIGRATED-BINDING, filed with this PR; the previous commit's message called it HOSTED-MOVE-RESIDUAL-ACTOR-REVOKE.

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant