Repository navigation
Conversation
…routes Row HOSTED-RELAY-OWNER-HUMAN-BIND-CEREMONY (lead ruling B): on a served Hosted vault the owner enrols a device owner key over HTTP, which binds no one, so Memory::verify_owner answers OWNER_BINDING_REQUIRED; then the device prepares its binding (POST /v1/owner/owner-signing-keys/bind/prepare) and lands it with the key's signature (POST .../bind/land): another key's signature is refused, the key's own lands, and the owner verb passes. Red on the base: the routes answer 404.
…enrolled
Row HOSTED-RELAY-OWNER-HUMAN-BIND-CEREMONY (lead ruling B, 21 Nov 04:3x). On
a Hosted vault the enrolment's predecessor is the custodial host root, whose
human bind the fold refuses, and a Relay vault holds no root, so there an
enrolment binds no one and every owner verb that reads the owner binding
answers OWNER_BINDING_REQUIRED. A two-step door binds them, signed by the key
itself: Vault::prepare_owner_key_binding_as builds BindActor{K, owner, human,
next epoch} with K as its one signer at the heads and answers the unsigned
entry and its transcript; Vault::land_owner_key_binding_as takes K's strict
Ed25519 signature, rechecks the door's rules at its own heads and lands the
row only once the fold counts it and the owner's binding stands. K must be a
live Ed25519 owner device key a counted enrolment names the owner for, while
no live human binding names them; a passkey signs no binding and the
recovery root's authority key stays off this door. The routes POST
/v1/owner/owner-signing-keys/bind/prepare and /bind/land need no host root,
so they serve Relay too. The self-host enrolment binder is unchanged
(OWNER-ENROL-BIND-SIGNER = A). Engine tests in tests/owner_key_binding.rs:
Hosted first and added Owner, Relay added Owner, four refusals.
Engine ANSWER BIND-CEREMONY-CLIENT-SCOPE = A: the owner passkey module's header now says the vault bars every passkey from signing the owner's human binding, and names the owner device key routes that bind a Hosted or Relay owner.
|
Warning Review limit reached
This review includes 9 billable files and costs up to $2.25.
Reviews can continue after your included limit without a manual trigger. An admin must approve usage-based billing. Or wait 51 minutes for your next included review. View limit details
|
olety
marked this pull request as ready for review
November 20, 2026 20:04
|
Automatic reviews are disabled for this repository. Configure in Code Review Settings. |
…ute test Keeps the test function under the workspace's too_many_lines bar; the steps and assertions are unchanged.
The workspace denies redundant_closure_for_method_calls.
… into opus-ownerkey-p1/hosted-owner-bind-ceremony
A key's own RevokeActor of its binding at an epoch above it raises the fold's floor for the key's next binding. The door must prepare the next binding above that floor, or the fold refuses it and retrying cannot recover (review of stack #10009, finding 3). Red here: the door takes the epoch from the last binding only.
…e its revoke floor The binding door took any counted OwnerKeyEnroll naming the owner as the key's enrolment for them, a confirmation among them: an owner key that vouches again for a key only a passkey vouched for names its own owner, whom the key was not enrolled for, so that owner could bind the key while the removal door attributes it to the enrolling owner only (review of stack #10009, finding 2). The door now leaves out the rows the removal closure reads as confirmations, with the one ancestry reading, and keeps sibling enrolments on parallel branches. The next binding's epoch now lies above every counted RevokeActor of the key as well as its last binding, as the fold's floor does; an epoch at the last value refuses (finding 3).
… into opus-ownerkey-p1/hosted-owner-bind-ceremony
This branch has not been deployed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Row [HOSTED-RELAY-OWNER-HUMAN-BIND-CEREMONY] (P2; lead ruling B, Sat 21 Nov 04:3x: "build now"). Engine ANSWER BIND-CEREMONY-CLIENT-SCOPE = A (21 Nov 04:44). Stacks on the hosted RevokeActor owner-proof PR, which must land first so that the custodial root cannot end a binding this door makes.
Why
On a Hosted vault the enrolment's predecessor is the custodial host root, and the hosted fold counts no
humanBindActor that only a key in the host's custody signs. A Relay vault holds no root of its own. So there an owner key enrolment binds no one, for the first Owner and for every added one, and each owner verb that reads the binding (memory::verify_owner_actor_binding_in_txn,Memory::verify_owner) answers OWNER_BINDING_REQUIRED.What
A two-step door, in which the owner's device signs its own binding with the key it enrolled:
Vault::prepare_owner_key_binding_as(authority, public_key)buildsBindActor{K, owner, human, next epoch}with K as its one signer, at K's next sequence, at the main-log heads (a frontier wider than one entry names is joined first, carried by this install). It answers the unsigned entry (canonical MessagePack) and the authority transcript K signs.Vault::land_owner_key_binding_as(authority, entry, signature)checks the entry against what prepare builds (op, owner, class, vault, signer, no cosign, one encoding). It rechecks the door's rules at its own heads, verifies K's strict Ed25519 signature over the transcript, and lands the row only once the fold counts it and the owner's binding stands.humanbinding names the owner and none rests on K. A confirmation is no enrolment: the rows the removal closure reads as confirmations (owner_authority_closure.rsconfirmations, now shared) do not attribute K, so the door and the removal agree whose key K is. Sibling enrolments on parallel branches each stand. A passkey signs no binding (the fold bars every passkey from a BindActor). The recovery root's authority key stays off this door. A device key that only a passkey vouched for binds once confirmed (the hosted fold takes no owner proof from it before).RevokeActorof K, both of which the fold's floor for a new binding takes in (op_apply.rsapply_actor_binding); an epoch at the last value refuses. A floor that only a revoke the fold no longer counts holds (ONE-REVOKE-DURABILITY) is not read here: the fold then refuses the binding, and land answers that it did not count.POST /v1/owner/owner-signing-keys/bind/prepare(public_key) andPOST /v1/owner/owner-signing-keys/bind/land(entry,signature), hex like the other owner key routes, behindowner(). They need no root this host holds, so they serve Relay too. OpenAPI rows throughRestShape::json. Refusals use the enrol routes' mapping: integrity is 400, authority is 409refused.New verb: none. The routes are owner routes, and the act is the owner's own: their own key signs, and the fold judges it. No grant is involved.
Files
confirmationsandEnrolmentRowshared with the door (pub(super)).Stored change: additive (BindActor is an existing op; no row shape or fold rule changes here).
Tests
owner_key_binding_routes(one test). The red pin a1d8631 (branch opus-ownerkey-p1/bind-ceremony-red-r1) is the test commit on the base PR's head plus the test-helper commit, so its test file equals the head's. There the enrolment over HTTP binds no one and the owner verb answers OWNER_BINDING_REQUIRED (asserted, passes); then/bind/prepareanswers 404, and the test fails at that status. On the head: another key's signature is refused (400) and binds no one, the key's own lands, and the owner verb passes. Command:RUSTFLAGS="--cfg oneiron_test_support" cargo test --locked -p oneiron-server --all-features --features oneiron/test-support --test owner_key_binding_routes.owner_key_bindinga_rebinding_takes_an_epoch_above_the_keys_revocation: red pin efdf053 (branch opus-ownerkey-p1/bind-ceremony-red-r2, the test commit before the fix), where the key's ownRevokeActorat epoch 9 ends its first binding, the door prepares epoch 2, and land answers that the binding did not count. On the head the door prepares epoch 10, which counts, and the owner verb passes. Command as for the guards below.owner_key_binding: the Hosted first Owner, the Hosted added Owner, and the Relay added Owner (an owner-origin root, served as Relay) each bind and pass the owner verb. Refusals: another key's signature, a key enrolled for another owner or never enrolled, an entry landed for another owner, and a second key for a bound owner. Command:RUSTFLAGS="--cfg oneiron_test_support" cargo test --locked -p oneiron --all-features --test owner_key_binding.These run after the merge (READY-PROOF-VS-MERGE-BAR = A).
Own review (Sol xhigh, stack #10009)
Findings 2 and 3 (confirmation attribution; epoch below a revoke floor) are fixed in 22d6449; finding 1 is PR A's (#10006, 3a34c7c), merged up here.
Follow-ups
/bind/prepareand/bind/landright after/enroll. No shipping client drives/enrolltoday.humanbinding of theirs ends,owner_live_atreads them as removed. They bind again with another live device key of theirs through this door.