From b6d210248deadfe1eb3c6a498051de1ef2d2f6bd Mon Sep 17 00:00:00 2001 From: mesutoezdil Date: Tue, 15 Sep 2026 19:20:26 +0200 Subject: [PATCH] fix(security-insights): make the insights file parseable and unset the image exec bit The OpenSSF Security Insights reader (ossf/si-tooling, used by LFX Insights through ossf/pvtr-github-repo-scanner) rejected the whole file, so every Baseline check that reads it fell back to a failure or a manual review, including OSPS-QA-04.01 "Insights does not contain a list of repositories" even though the repositories are listed. Two problems, both from targeting the unreleased 2.2.0 spec: in-scope and out-of-scope are declared as string lists in the spec but the generated Go type annotates them as a single URL, so a list fails to unmarshal and aborts the load. The scope text moves into the neighbouring comment, which carries the same information. policy only exists on the spec's unreleased main branch. 2.1.0 is the latest tag, so the file now declares 2.1.0 and uses security-policy, which the reader maps onto the same field. The file now passes both cue vet against the 2.1.0 schema and si.Load. docs/images/sample_nvidia-smi.png was committed as 100755. The scanner flags a binary blob carrying an execute bit as a generated executable artifact, which is why that PNG tripped OSPS-QA-05.01 and docs/images/hami-core-arch.png, the same format at 100644, did not. Content is unchanged. Signed-off-by: mesutoezdil --- docs/images/sample_nvidia-smi.png | Bin security-insights.yml | 21 +++++++++++---------- 2 files changed, 11 insertions(+), 10 deletions(-) mode change 100755 => 100644 docs/images/sample_nvidia-smi.png diff --git a/docs/images/sample_nvidia-smi.png b/docs/images/sample_nvidia-smi.png old mode 100755 new mode 100644 diff --git a/security-insights.yml b/security-insights.yml index 2a915903..22785e87 100644 --- a/security-insights.yml +++ b/security-insights.yml @@ -1,7 +1,7 @@ header: - schema-version: 2.2.0 - last-updated: '2026-08-26' - last-reviewed: '2026-08-26' + schema-version: 2.1.0 + last-updated: '2026-09-15' + last-reviewed: '2026-09-15' url: https://raw.githubusercontent.com/Project-HAMi/HAMi-core/main/security-insights.yml comment: | HAMi-core is one of the codebases that make up the HAMi project. It is @@ -34,17 +34,18 @@ project: vulnerability-reporting: reports-accepted: true bug-bounty-available: false - policy: https://github.com/Project-HAMi/HAMi/security/policy - in-scope: - - A workload reaching another tenant's GPU memory, device or namespace - out-of-scope: - - A workload exceeding its own GPU quota by unsetting LD_PRELOAD - - A workload exceeding its own GPU quota via a statically linked CUDA runtime - - A workload exceeding its own GPU quota via ptrace on its own process + security-policy: https://github.com/Project-HAMi/HAMi/security/policy comment: | Reports are accepted through the HAMi advisory form at https://github.com/Project-HAMi/HAMi/security/advisories/new. + In scope: a workload reaching another tenant's GPU memory, device or + namespace. + + Out of scope: a workload exceeding its own GPU quota by unsetting + LD_PRELOAD, by linking the CUDA runtime statically, or by calling ptrace + on its own process. + HAMi-core is a cooperative quota mechanism on a trusted cluster, not a hard isolation boundary. The out-of-scope cases are triaged as bugs.