[BUG] Realm seal returns an internal error in five VM scenarios #162

Open
opened 2026-10-17 20:52:03 +02:00 by erikinkinen · 2 comments
Owner

Summary

Five realm VM scenarios fail while sealing the prepared realm. The shell receives
an internal error, so the scenarios never reach their client execution checks.

Reproduction

The complete catalog in CI 2822
at f06bf72280eff2f49248ce816c327f8a3760a4f4 reproduces this in:

  • appliance-disk-image-realm-clients-positive.toml
  • appliance-disk-image-realm-factories-positive.toml
  • appliance-disk-image-realm-threads-positive.toml
  • appliance-disk-image-realm-owner-loss-positive.toml
  • appliance-disk-image-realm-pipes-positive.toml

These files are under tests/scenarios. The workflow runs the list selected by
scripts/list-ci-scenarios.py through scripts/run-ci-vm-scenarios.sh.

Expected and actual behavior

Realm startup and sealing should complete before the scenario exercises its
clients. The original serial output instead reports:

ERIX_EXSH:REALM_ADMISSION:FAILED request=00000035 kind=0000000b first=000006da second=00000005

The owner-loss case reports the same error with request 00000018. Operation
0x6da is SEAL_REALM; result 5 is IPC_RESULT_INTERNAL. QEMU then reaches its
45-second serial-progress watchdog. The clients, factories, threads and pipes
cases had already emitted the admission, preparation and startup verified markers.

Environment and identity

This is the x86-64 QEMU VM job for the linked Integration revision and its exact
component catalog. The source of the internal error and the identity of the
component producing it have not been established. No last working revision is
known.

Evidence and prior checks

The original complete job log records 511 passing and 14 failing scenarios out
of 525. These five failures share the seal error above. The RTC staged-commit
failure in #122 and the
pipeline redirection failure in #91
are separate observations.

Tracking

Integration PR #12,
Launchd PR #2,
Procd PR #2,
phase completion.

  • Identify and repair the seal failure while preserving the existing authority and cleanup checks.
  • Add regression coverage at the responsible boundary.
  • Pass all five existing VM cases through seal and their intended assertions.
## Summary Five realm VM scenarios fail while sealing the prepared realm. The shell receives an internal error, so the scenarios never reach their client execution checks. ## Reproduction The complete catalog in [CI 2822](https://git.erikinkinen.fi/erix/integration/actions/runs/2822) at `f06bf72280eff2f49248ce816c327f8a3760a4f4` reproduces this in: - `appliance-disk-image-realm-clients-positive.toml` - `appliance-disk-image-realm-factories-positive.toml` - `appliance-disk-image-realm-threads-positive.toml` - `appliance-disk-image-realm-owner-loss-positive.toml` - `appliance-disk-image-realm-pipes-positive.toml` These files are under `tests/scenarios`. The workflow runs the list selected by `scripts/list-ci-scenarios.py` through `scripts/run-ci-vm-scenarios.sh`. ## Expected and actual behavior Realm startup and sealing should complete before the scenario exercises its clients. The original serial output instead reports: ```text ERIX_EXSH:REALM_ADMISSION:FAILED request=00000035 kind=0000000b first=000006da second=00000005 ``` The owner-loss case reports the same error with request `00000018`. Operation `0x6da` is `SEAL_REALM`; result 5 is `IPC_RESULT_INTERNAL`. QEMU then reaches its 45-second serial-progress watchdog. The clients, factories, threads and pipes cases had already emitted the admission, preparation and startup verified markers. ## Environment and identity This is the x86-64 QEMU VM job for the linked Integration revision and its exact component catalog. The source of the internal error and the identity of the component producing it have not been established. No last working revision is known. ## Evidence and prior checks The original complete job log records 511 passing and 14 failing scenarios out of 525. These five failures share the seal error above. The RTC staged-commit failure in [#122](https://git.erikinkinen.fi/erix/integration/issues/122) and the pipeline redirection failure in [#91](https://git.erikinkinen.fi/erix/integration/issues/91) are separate observations. ## Tracking [Integration PR #12](https://git.erikinkinen.fi/erix/integration/pulls/12), [Launchd PR #2](https://git.erikinkinen.fi/erix/launchd/pulls/2), [Procd PR #2](https://git.erikinkinen.fi/erix/procd/pulls/2), [phase completion](https://git.erikinkinen.fi/erix/integration/issues/65). - [ ] Identify and repair the seal failure while preserving the existing authority and cleanup checks. - [ ] Add regression coverage at the responsible boundary. - [ ] Pass all five existing VM cases through seal and their intended assertions.
Author
Owner

Launchd's seal check omitted supplementary-group records from its expected completed-record count. The canonical stream includes them, and these client fixtures supply three. A fully delivered valid plan therefore failed the local equality check with InvalidNativeEvidence, which becomes the reported internal error.

afbee6d1 restores the checked group count. The mock now derives its readiness gate from the delivered canonical header. Three new controls cover complete delivery, a delayed last group and premature Prepared; their execution is pending. Both seal stages still require complete delivery, original caller retirement and source disposal.

The five existing VM cases remain the guest acceptance gate. No guest success is claimed yet.

Launchd's seal check omitted supplementary-group records from its expected completed-record count. The canonical stream includes them, and these client fixtures supply three. A fully delivered valid plan therefore failed the local equality check with `InvalidNativeEvidence`, which becomes the reported internal error. [afbee6d1](https://git.erikinkinen.fi/erix/launchd/commit/afbee6d1b188ec8eb280fd074e3f91866864ef54) restores the checked group count. The mock now derives its readiness gate from the delivered canonical header. Three new controls cover complete delivery, a delayed last group and premature Prepared; their execution is pending. Both seal stages still require complete delivery, original caller retirement and source disposal. The five existing VM cases remain the guest acceptance gate. No guest success is claimed yet.
Author
Owner

The three-group fixtures were rejected during configuration because their group IDs were out of canonical order. The fixture correction supplies ascending IDs and includes the result code in the freeze assertion. The production record-count fix is unchanged. The new unit run and five realm VM cases remain pending.

The three-group fixtures were rejected during configuration because their group IDs were out of canonical order. The [fixture correction](https://git.erikinkinen.fi/erix/launchd/commit/406a1508354bc37a43e53f0ce91277faef0d6295) supplies ascending IDs and includes the result code in the freeze assertion. The production record-count fix is unchanged. The new unit run and five realm VM cases remain pending.
erikinkinen added this to the Alpha 1 milestone 2026-10-30 09:34:19 +01:00
Sign in to join this conversation.
No milestone
No project
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference
erix/integration#162
No description provided.