ci: pin base images by digest resolved directly from Docker Hub

With the pkg-cache BuildKit mirror on, `FROM almalinux/10-base` was resolved
by tag through the cache, so a compromised cache could substitute the image.

Each job using the mirror now looks up the base tag's digest DIRECTLY at
registry-1.docker.io (.gitea/scripts/resolve-base-image.sh, HTTPS, HEAD on
the manifest, regex-validated) and builds FROM image@sha256:... via the
BASE_IMAGE build-arg, so the mirror can only serve content-addressed blobs.
If the lookup fails, the mirror is NOT used for that job and the tag is pulled
straight from Docker Hub. The digest is resolved every run (no manual bumps);
local builds default to the tag.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

.gitea/scripts/resolve-base-image.sh

@@ -0,0 +1,35 @@
+#!/bin/bash
+## resolve-base-image.sh IMAGE:TAG — CI only (not copied into images).
+## Prints IMAGE@sha256:<digest> for a Docker Hub tag, looked up DIRECTLY at
+## registry-1.docker.io over HTTPS (never via the pkg-cache mirror), so the
+## build can say `FROM IMAGE@sha256:...` and the mirror can only serve
+## content-addressed blobs that BuildKit verifies. Exits non-zero (prints
+## nothing on stdout) if the lookup fails; the workflow then skips the mirror
+## and pulls the tag directly from Docker Hub. HEAD requests do not count
+## against Docker Hub pull rate limits.
+## Canonical copy: cloud-apache-container/.gitea/scripts/resolve-base-image.sh
+## (docs/pkg-cache.md there); other repos carry identical copies.
+set -euo pipefail
+ref="${1:?usage: resolve-base-image.sh IMAGE:TAG}"
+name_re='^[a-z0-9]+([._-][a-z0-9]+)*(/[a-z0-9]+([._-][a-z0-9]+)*)?$'
+tag_re='^[A-Za-z0-9_][A-Za-z0-9_.-]{0,127}$'
+name="${ref%:*}"; tag="${ref##*:}"
+[ "$name" != "$ref" ] || { echo "resolve-base-image: '$ref' has no :tag" >&2; exit 2; }
+[[ "$name" =~ $name_re ]] || { echo "resolve-base-image: bad image name '$name'" >&2; exit 2; }
+[[ "$tag" =~ $tag_re ]] || { echo "resolve-base-image: bad tag '$tag'" >&2; exit 2; }
+repo="$name"; [[ "$repo" == */* ]] || repo="library/$repo"   # official images
+
+curl_opts=(-fsS --proto '=https' --retry 3 --retry-all-errors --max-time 20)
+token=$(curl "${curl_opts[@]}" "https://auth.docker.io/token?service=registry.docker.io&scope=repository:${repo}:pull" |
+        python3 -c 'import json,sys; print(json.load(sys.stdin)["token"])')
+[ -n "$token" ] || { echo "resolve-base-image: no Docker Hub token for $repo" >&2; exit 1; }
+hdrs=$(curl "${curl_opts[@]}" -I \
+    -H "Authorization: Bearer ${token}" \
+    -H 'Accept: application/vnd.oci.image.index.v1+json' \
+    -H 'Accept: application/vnd.docker.distribution.manifest.list.v2+json' \
+    -H 'Accept: application/vnd.oci.image.manifest.v1+json' \
+    -H 'Accept: application/vnd.docker.distribution.manifest.v2+json' \
+    "https://registry-1.docker.io/v2/${repo}/manifests/${tag}")
+digest=$(printf '%s\n' "$hdrs" | tr -d '\r' | awk 'tolower($1)=="docker-content-digest:" {print $2}' | tail -n1)
+[[ "$digest" =~ ^sha256:[0-9a-f]{64}$ ]] || { echo "resolve-base-image: no valid digest for $ref (got '${digest}')" >&2; exit 1; }
+printf '%s@%s\n' "$name" "$digest"

.gitea/workflows/build-push.yaml

@@ -16,17 +16,40 @@ jobs:
       - name: Checkout
         uses: actions/checkout@v4
 
+      # Base image pinned by DIGEST, looked up directly at Docker Hub (never via
+      # the pkg-cache mirror), so the mirror can only serve content-addressed
+      # blobs that BuildKit verifies. If the lookup fails this job does NOT use
+      # the mirror and pulls the tag straight from Docker Hub.
+      - name: Resolve base image digest (Docker Hub, direct)
+        id: base
+        env:
+          BASE_TAG: almalinux/10-base:latest
+        run: |
+          set -uo pipefail
+          if ref=$(.gitea/scripts/resolve-base-image.sh "$BASE_TAG") && [[ "$ref" =~ ^[a-z0-9][a-z0-9._/-]*@sha256:[0-9a-f]{64}$ ]]; then
+            echo "base-image: ${BASE_TAG} -> ${ref} (digest from registry-1.docker.io)"
+            echo "image=${ref}" >> "$GITHUB_OUTPUT"
+            echo "pinned=true" >> "$GITHUB_OUTPUT"
+          else
+            echo "::warning::base image ${BASE_TAG}: Docker Hub digest lookup failed - pulling the tag directly from Docker Hub, pkg-cache registry mirror NOT used"
+            echo "image=${BASE_TAG}" >> "$GITHUB_OUTPUT"
+            echo "pinned=false" >> "$GITHUB_OUTPUT"
+          fi
+
       # Internal package cache (CT130 pkg-cache). Used ONLY if it answers over
       # TLS with our pinned CA; otherwise this job builds from the upstream repos
       # and Docker Hub exactly as before. Repo cache and registry mirror are
       # probed separately so either can fail back on its own.
       - name: Probe package cache
         id: pkgcache
+        # The registry mirror is only used for a digest-pinned base image.
+        env:
+          BASE_PINNED: ${{ steps.base.outputs.pinned }}
         run: |
           url=""; mirror=""
           if curl -fsS --max-time 5 --cacert keys/pkg-cache/ca.crt -o /dev/null https://192.168.1.40/; then url="https://192.168.1.40"; fi
-          if curl -fsS --max-time 5 --cacert keys/pkg-cache/ca.crt -o /dev/null https://192.168.1.40:5000/v2/; then mirror="192.168.1.40:5000"; fi
-          echo "pkg-cache: repos=${url:-<unreachable: upstream>} registry-mirror=${mirror:-<unreachable: Docker Hub>}"
+          if [ "$BASE_PINNED" = true ] && curl -fsS --max-time 5 --cacert keys/pkg-cache/ca.crt -o /dev/null https://192.168.1.40:5000/v2/; then mirror="192.168.1.40:5000"; fi
+          echo "pkg-cache: repos=${url:-<unreachable: upstream>} registry-mirror=${mirror:-<not used: Docker Hub direct>}"
           [ -n "$url" ] || echo "::warning::pkg-cache unreachable - building from upstream repositories"
           echo "url=${url}" >> "$GITHUB_OUTPUT"
           echo "mirror=${mirror}" >> "$GITHUB_OUTPUT"
@@ -64,6 +87,7 @@ jobs:
           push: true
           build-args: |
             NODEVER=${{ matrix.nodever }}
+            BASE_IMAGE=${{ steps.base.outputs.image }}
             PKG_CACHE=${{ steps.pkgcache.outputs.url }}
           tags: |
             repo.anhonesthost.net/cloud-hosting-platform/cnoc:node${{ matrix.nodever }}

Dockerfile

@@ -1,4 +1,9 @@
-FROM almalinux/10-base
+# Base image. CI passes BASE_IMAGE=almalinux/10-base@sha256:<digest>, resolved
+# DIRECTLY from Docker Hub in the same run (.gitea/scripts/resolve-base-image.sh),
+# so the pkg-cache registry mirror can only serve content-addressed blobs.
+# Local builds use the tag. cloud-apache-container docs/pkg-cache.md.
+ARG BASE_IMAGE=almalinux/10-base
+FROM ${BASE_IMAGE}
 ARG NODEVER=20
 
 # OPTIONAL internal package cache, e.g. --build-arg PKG_CACHE=https://192.168.1.40