[BUG] Nested-script exit-status guest cannot launch its command #135

Open
opened 2026-10-16 05:27:12 +02:00 by erikinkinen · 6 comments
Owner

Summary

The nested-script exit-status scenario builds and boots, but Exsh reports
command not executable and never publishes its required completion marker.
The cause of the launch failure is still unknown.

Reproduction

From Integration, run:

bash scripts/run-scenario.sh tests/scenarios/appliance-disk-image-initial-exsh-exit-status-script-positive.toml components.toml

The latest local run uses the selected development image and original limits:
120 seconds for QEMU, 45 seconds for serial silence, one CPU and 1 GiB RAM.
The same symptom occurred before the Launchd storage-rejection fix and persists
in a fresh image containing it. No last known working revision is established.

Expected and actual behavior

The nested here-document script should return seven, propagate that status
through the outer POSIX exit, and emit
ERIX_ROOTD:INITIAL_EXSH:EXITED_EXIT_STATUS_SCRIPT_OK.

Instead, the scenario exits 1 after exsh: command not executable; the completion
marker is absent. Rootd records its first call as endpoint 166/op 1678 and its
last as endpoint 2636/op 1668. Those context records do not identify the failing
resolver branch or prove that either call failed.

Environment and identity

  • Integration: ede995b73ee90d67d48a0382e5c04ad9805998da.
  • Launchd: 5d8cddb07e57843f52914dc2edc1d9b7ad201abe.
  • Selected Exsh: 8540bcf44fbfd24a51d0e399077f25e4b6651988.
  • Kernel: 68a65a866a78bc966696939d74592540896dda15.
  • Rootd: ff0bd1ef6a6aad09e3f0bf39c7443ece4d1847dc.
  • Local x86_64 QEMU/OVMF; Rust 1.97.1; scenario-defined initial-shell input and
    service grants, with no added diagnostic authority.

Evidence and prior checks

Build, signing and packaging complete. The original producer exits 1; both
capture stages exit 0. All host owners confirm cleanup. The complete serial
log, failed image and selected build inputs are retained. The latest producer
log contains no warning diagnostics.

CI 2711
also contains native launch/status failures. Similar output does not establish
a shared cause across scenarios.

Tracking

Related: Launchd resolver-slot repair,
Launchd storage-rejection fix
and Phase 6 tracking.

  • Identify and fix the actual launch failure without changing the script,
    authority policy, expected status or time limits.
  • Pass the original exit-status scenario with its complete marker sequence
    and confirmed cleanup.
  • Check related native launch/status failures against the corrected source.
## Summary The nested-script exit-status scenario builds and boots, but Exsh reports `command not executable` and never publishes its required completion marker. The cause of the launch failure is still unknown. ## Reproduction From Integration, run: ```sh bash scripts/run-scenario.sh tests/scenarios/appliance-disk-image-initial-exsh-exit-status-script-positive.toml components.toml ``` The latest local run uses the selected development image and original limits: 120 seconds for QEMU, 45 seconds for serial silence, one CPU and 1 GiB RAM. The same symptom occurred before the Launchd storage-rejection fix and persists in a fresh image containing it. No last known working revision is established. ## Expected and actual behavior The nested here-document script should return seven, propagate that status through the outer POSIX `exit`, and emit `ERIX_ROOTD:INITIAL_EXSH:EXITED_EXIT_STATUS_SCRIPT_OK`. Instead, the scenario exits 1 after `exsh: command not executable`; the completion marker is absent. Rootd records its first call as endpoint 166/op 1678 and its last as endpoint 2636/op 1668. Those context records do not identify the failing resolver branch or prove that either call failed. ## Environment and identity - Integration: `ede995b73ee90d67d48a0382e5c04ad9805998da`. - Launchd: `5d8cddb07e57843f52914dc2edc1d9b7ad201abe`. - Selected Exsh: `8540bcf44fbfd24a51d0e399077f25e4b6651988`. - Kernel: `68a65a866a78bc966696939d74592540896dda15`. - Rootd: `ff0bd1ef6a6aad09e3f0bf39c7443ece4d1847dc`. - Local x86_64 QEMU/OVMF; Rust 1.97.1; scenario-defined initial-shell input and service grants, with no added diagnostic authority. ## Evidence and prior checks Build, signing and packaging complete. The original producer exits 1; both capture stages exit 0. All host owners confirm cleanup. The complete serial log, failed image and selected build inputs are retained. The latest producer log contains no warning diagnostics. [CI 2711](https://git.erikinkinen.fi/erix/integration/actions/runs/2711) also contains native launch/status failures. Similar output does not establish a shared cause across scenarios. ## Tracking Related: [Launchd resolver-slot repair](https://git.erikinkinen.fi/erix/launchd/issues/21), [Launchd storage-rejection fix](https://git.erikinkinen.fi/erix/launchd/issues/20) and [Phase 6 tracking](https://git.erikinkinen.fi/erix/integration/issues/65). - [x] Identify and fix the actual launch failure without changing the script, authority policy, expected status or time limits. - [x] Pass the original exit-status scenario with its complete marker sequence and confirmed cleanup. - [ ] Check related native launch/status failures against the corrected source.
Author
Owner

Launchd dcbb1cc8 adds seven fixed resolver tags to the existing optional process-exit trace. The component architecture documents the mapping.

The fresh guest failure remains unresolved. The next diagnostic will explicitly enable the existing trace and retain the original timeout and exit-status checks; no traced guest result is available yet.

Launchd [dcbb1cc8](https://git.erikinkinen.fi/erix/launchd/commit/dcbb1cc8b8be9540867cf1dab6fc1e1b70388fca) adds seven fixed resolver tags to the existing optional process-exit trace. The component architecture documents the mapping. The fresh guest failure remains unresolved. The next diagnostic will explicitly enable the existing trace and retain the original timeout and exit-status checks; no traced guest result is available yet.
Author
Owner

7bdb0ca2 selects Launchd dcbb1cc8 in both component catalogs. That revision adds fixed resolver tags to the existing optional exit trace.

The original exit-status failure remains open. A fresh run with the existing trace enabled is being prepared to locate it; no failing branch has been identified yet.

`7bdb0ca2` selects Launchd `dcbb1cc8` in both component catalogs. That revision adds fixed resolver tags to the existing optional exit trace. The original exit-status failure remains open. A fresh run with the existing trace enabled is being prepared to locate it; no failing branch has been identified yet.
Author
Owner

The traced exit-status guest still reports “command not executable” and misses the expected marker. Build and packaging complete; all three owners finish cleanup under the original limits.

The available transport samples do not identify the ordinary launch rejection. Exsh currently discards its typed search error before reporting status 126. I’m preserving that detail through the existing stderr path so the next diagnostic can identify the failed operation without adding authority.

The traced exit-status guest still reports “command not executable” and misses the expected marker. Build and packaging complete; all three owners finish cleanup under the original limits. The available transport samples do not identify the ordinary launch rejection. Exsh currently discards its typed search error before reporting status 126. I’m preserving that detail through the existing stderr path so the next diagnostic can identify the failed operation without adding authority.
Author
Owner

Exsh 835dcea2 preserves ordinary command rejection causes through completion. Application rejections retain their original operation/result/class tuple; native transport status and committed-job denial stay distinct. One bounded record goes through the command's current stderr without exposing paths, arguments or capability IDs.

The earlier guest failure remains unresolved. The extended physical/COM1 probe now checks a directory rejection for one typed cause and status 126; its host and socket controls pass, but a genuine VM run is still pending. Integration's component pins are being updated to the published runtime and manual revisions.

Exsh [835dcea2](https://git.erikinkinen.fi/erix/exsh/commit/835dcea2d6b55ce75599f1947c07378a2da9487c) preserves ordinary command rejection causes through completion. Application rejections retain their original operation/result/class tuple; native transport status and committed-job denial stay distinct. One bounded record goes through the command's current stderr without exposing paths, arguments or capability IDs. The earlier guest failure remains unresolved. The extended physical/COM1 probe now checks a directory rejection for one typed cause and status 126; its host and socket controls pass, but a genuine VM run is still pending. Integration's component pins are being updated to the published runtime and [manual](https://git.erikinkinen.fi/erix/docs/commit/2cce0e0a1e4093690843454a2f5914cdb91cc2f8) revisions.
Author
Owner

A fresh build at Integration 9e9d708f, with Exsh 835dcea2 and Launchd dcbb1cc8, completes native compilation, packaging and disk creation but still reaches the original 45-second serial watchdog.

The new shell diagnostic identifies a rejected Launchd COMMIT (0x692) with storage failure class 7. Kernel serial records split the result field; the surrounding fragments suggest BUSY (4), so that value remains qualified. The resolver search has a concrete source defect tracked in Launchd #21. Its fix is being reviewed; no corrected-source guest success is claimed.

A fresh build at Integration `9e9d708f`, with Exsh `835dcea2` and Launchd `dcbb1cc8`, completes native compilation, packaging and disk creation but still reaches the original 45-second serial watchdog. The new shell diagnostic identifies a rejected Launchd COMMIT (`0x692`) with storage failure class `7`. Kernel serial records split the result field; the surrounding fragments suggest BUSY (`4`), so that value remains qualified. The resolver search has a concrete source defect tracked in [Launchd #21](https://git.erikinkinen.fi/erix/launchd/issues/21). Its fix is being reviewed; no corrected-source guest success is claimed.
Author
Owner

The original exit-status scenario now passes with Launchd 25efde7a and Integration b950621d, including the nested here-document, propagated exit status and complete ordered marker sequence. The script, grants, expected result and original VM/silence limits are unchanged. Build and capture finish without warnings and cleanup completes.

The resolver-slot repair is published and its selected binary is confirmed in the image. The first two checklist items are complete. Related launch/status failures remain under review, so this issue stays open. Incorrect packaged revision metadata is tracked separately.

The original exit-status scenario now passes with Launchd `25efde7a` and Integration `b950621d`, including the nested here-document, propagated exit status and complete ordered marker sequence. The script, grants, expected result and original VM/silence limits are unchanged. Build and capture finish without warnings and cleanup completes. The [resolver-slot repair](https://git.erikinkinen.fi/erix/launchd/issues/21) is published and its selected binary is confirmed in the image. The first two checklist items are complete. Related launch/status failures remain under review, so this issue stays open. [Incorrect packaged revision metadata](https://git.erikinkinen.fi/erix/integration/issues/144) is tracked separately.
erikinkinen added this to the Alpha 1 milestone 2026-10-30 09:34:17 +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#135
No description provided.