[BUG] First control challenge can fail before the started mediator registers #8
Labels
No labels
alpha-1
bug
ci
docs
duplicate
enhancement
help wanted
invalid
performance
phase-6
question
refactor
security
wontfix
No project
No assignees
1 participant
Notifications
Due date
No due date set.
Dependencies
No dependencies set
Reference
erix/procd#8
Loading…
Reference in a new issue
No description provided.
Delete branch "%!s()"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
Summary
Actual changed-image startup and running retirement remain pending; this report stays open.
Procd can submit its first control Challenge before the newly started Posixd has registered its receiver. The Kernel rejects this unaccepted submission with
NOT_FOUNDand owner zero. Procd treats that result as a permanent bootstrap failure, even though its exact child has just become runnable and can register on a later scheduling turn.This is a source-confirmed startup ordering defect discovered while reviewing the next native path. It is separate from the empty CLAIM defects and is not assigned as the cause of a previously recorded guest failure.
Reproduction
At Procd
d04311abff67e7355bd92bb2e48fb6161e0ec674, start an otherwise fully provisioned, exact RealmMediator generation. Let the original START_PROCESS call succeed but keep the child unscheduled through the immediately following Challenge submission. Return the Kernel's actual absent-registration result(IPC_RESULT_NOT_FOUND, 0), then permit the child to register its control receiver on a later turn.This ordering exists in the native source: Procd's START control call uses timeout zero; it immediately advances the newly created Challenge caller in that same producer admission before its outer service-loop yield. A deterministic host control should model delayed registration and then successful submission without repeating native START.
Expected and actual behavior
Only the original Start/Challenge caller, before any accepted invocation owner, may wait for the exact newly started child's receiver. Zero-owner
NOT_FOUNDmust be accompanied by a successful fresh query of the retained SEND-only control route. A missing route, failed query, surplus rights, other error, or nonzero error handle must preserve failure/ownership semantics. Existing exact-child terminal polling and bounded service-loop yielding must continue while registration is pending.Currently the caller permits only zero-owner
RETRYas pending. The first zero-ownerNOT_FOUNDsets a sticky failure; a later scheduler yield or receiver registration cannot recover the bootstrap. START must never be replayed, and an accepted SUBMIT must never be repeated.BUSYis not an absent-registration result and must not be generalized into readiness. ContextProbe's existing stricter contract stays intact because Launchd registers that receiver before Provision.Environment and identity
d04311abff67e7355bd92bb2e48fb6161e0ec674; Kernel5e285aa91a1423523b2a094d4ce0449ada39b16c; Posixd809647d60544c4c6121868c99133f30d054965c9. Reviewed selected Integration:fa857c0964774def6777142afeb7a3c16440abee.Evidence and prior checks
The new
producer_start_waits_for_control_registration_without_replaying_startregression fails against the original producer with a retainedSome(NOT_FOUND)failure after the first unaccepted Challenge. The corrected dedicated Start/Challenge caller waits only before any accepted owner and only after a fresh exact SEND-only endpoint query. Six focused controls pass: delayed registration then one accepted submission, route disappearance/query failure/surplus rights, unchanged BUSY/DENIED errors, retained nonzero failure owners, exact child loss and unchanged strict ContextProbe behavior. The existing original-child reservation, terminal polling and scheduler yielding remain in force. Native START and accepted SUBMIT are never replayed.The combined frozen source passes 335 default or 341 all-feature main tests plus five existing probe tests in each development/release configuration. Formatting, all thirteen strict Clippy selections, all nine native build selections, warning-denied host/native private rustdoc, original dependency checks and canonical Markdown checks pass with zero warnings. New original CI runs are being observed separately; no CI success or native acceptance is inferred from local checks.
NOT_FOUNDbefore allocating invocation storage, transferring bindings or publishing an owner; wire rejection returns zero auxiliary ownership.RETRY/zero. Procd's later outer-loop yield occurs after the first refusal has already been retained.An intervening timer or scheduling turn can mask the race; it is not a protocol guarantee of receiver registration. The original source review ran no compiler or VM and changed no runtime input. Baseline regression, strict validation and corrected actual native acceptance remain required.
Tracking
Procd lifecycle audit, Procd PR 2, Posixd realm lifecycle, Launchd empty CLAIM, Posixd empty CLAIM, Procd retirement event ownership, Integration validation, and Phase 6 completion.
erikinkinen referenced this issue from erix/integration2026-09-24 20:11:14 +02:00
erikinkinen referenced this issue2026-09-24 20:29:37 +02:00
erikinkinen referenced this issue from erix/docs2026-09-24 20:38:25 +02:00