Linux containment qualification¶

Landlock support depends on kernel features, executable compatibility and the deployment's namespace and service policy. The published Linux x86-64 Symbiont 2.0.1 baseline has SHA-256 8c4c385d83965c3af965340ae50482400337ae596bbc232f4beb9026ec915ad7. Current-source checks also used commit 495d2b77a83ae6781d4e698ba476c82a5846cc96 with two release-profile builds: the Ubuntu candidate with SHA-256 feddcf2f752882f0c037d33a4097ded24a8937831dc6b664e0339fec6619299b and the Debian-compatible candidate with SHA-256 ffbf2e8b9246beeb7ad4d703131c09bf9d790f483c33f1f1e7d5c78cad4d4c52. Results qualify only the binary named in each row.

The Debian result exercises the advertised ABI floor. The Ubuntu comparison demonstrates why detecting a newer ABI is insufficient: symbi doctor must test the actual private workspace and transport. Services ran under a non-root user with a systemd user manager and cgroup v2 delegation. Cloud images needed a local test user and enabled user-manager lingering; that service setup is distinct from namespace security policy.

The first current-source binary was built on a host with glibc 2.43 and could not start on Debian 13 because its loader requires that newer ABI. Rebuilding the same frozen source in Debian Bookworm produced a binary whose highest referenced glibc version is 2.35, and that binary passed the Debian platform diagnostic. The Linux x86-64 release job is therefore pinned to Ubuntu 22.04 and rejects an artifact requiring a glibc version newer than 2.35. This makes distribution compatibility an enforced publication property instead of an assumption about the current hosted runner image.

Scope and retained evidence¶

The labs used disposable QEMU guests with CPU emulation, no shared host mounts and blocked external networking; only a localhost SSH forward was available for the controller. The workers themselves used native Landlock. The VM was an additional laboratory boundary, not a product requirement. Ubuntu's namespace adjustment was confined to that guest and was recorded separately from its default-policy result. No original host security policy was changed.

Official image downloads were checked against their publishers' HTTPS checksum files. Detached image-checksum signatures were not verified. Image identities, platform reports, evidence inventories and reproduction tools are retained in release assurance evidence.

These results cover x86-64 and the recorded configurations. They do not establish ARM64 execution, arbitrary desktop distribution compatibility, resistance to kernel exploits or containment after host compromise. A distribution policy refusal should be resolved through the operator's approved deployment policy; the test guest's broad namespace switch is not an installer action.

Reproduce the platform diagnostic¶

Run as the non-root service user under the systemd user manager, with Python 3.11+ and the verified executable. Supply the expected ABI from the selected test image. The diagnostic does not alter kernel or distribution policy:

python3 scripts/test-landlock-platform.py \
  --binary /absolute/path/to/verified/symbi \
  --expected-binary-sha256 "$BINARY_SHA256" \
  --expected-abi 6 --expect usable --report /absolute/new/platform.json

For an image expected to deny native namespaces, use --expect namespace-refused. Its report must say qualified: false, retain the specific failure and confirm cleanup. A pass in this mode proves the refusal behavior only. For execution evidence, follow the broker, lifecycle and external-effect procedures or run the combined release driver in a suitably prepared disposable lab.