hololake-system-architecture/product-source/hololake-platform/guanghu-os/scripts/linux-subcontrol-docker-backend.sh

#!/usr/bin/env bash
set -Eeuo pipefail
umask 077

die() {
  echo "GUANGHU_LINUX_SUBCONTROL_FAIL_0: $*" >&2
  exit 1
}

test_mode=${GH_SUBCONTROL_TEST_MODE:-0}
if [[ "${test_mode}" == 1 ]]; then
  config_path=${GH_SUBCONTROL_CONFIG:?test config is required}
  docker_bin=${GH_SUBCONTROL_DOCKER:?test docker is required}
  dmi_path=${GH_SUBCONTROL_DMI_PATH:?test DMI path is required}
  grub_cfg=${GH_SUBCONTROL_GRUB_CFG:?test GRUB path is required}
  state_root=${GH_SUBCONTROL_STATE_ROOT:?test state root is required}
  receipt_root=${GH_SUBCONTROL_RECEIPT_ROOT:?test receipt root is required}
else
  [[ "${EUID}" == 0 ]] || die "root execution is required"
  for override in \
    GH_SUBCONTROL_CONFIG GH_SUBCONTROL_DOCKER GH_SUBCONTROL_DMI_PATH \
    GH_SUBCONTROL_GRUB_CFG GH_SUBCONTROL_STATE_ROOT GH_SUBCONTROL_RECEIPT_ROOT; do
    [[ -z "${!override:-}" ]] || die "production path override is forbidden: ${override}"
  done
  config_path=/etc/guanghu/linux-subcontrol.conf
  docker_bin=/usr/bin/docker
  dmi_path=/sys/class/dmi/id/product_uuid
  grub_cfg=/boot/grub/grub.cfg
  state_root=/run/guanghu/linux-subcontrol
  receipt_root=/guanghu/receipts/linux-subcontrol
  [[ -f "${config_path}" && ! -L "${config_path}" ]] || die "root-owned backend config is missing"
  [[ "$(stat -c '%u:%a' "${config_path}")" =~ ^0:(600|640|644)$ ]] ||
    die "backend config owner or mode is unsafe"
fi

schema=
node_id=
instance_id=
backend_id=
container_name=
image=
rescue_slot_id=
while IFS='=' read -r key value; do
  [[ -n "${key}" ]] || continue
  case "${key}" in
    schema|node_id|instance_id|backend_id|container_name|image|rescue_slot_id)
      printf -v "${key}" '%s' "${value}"
      ;;
    *) die "unknown config field: ${key}" ;;
  esac
done <"${config_path}"

[[ "${schema}" == "guanghu.linux-subcontrol-docker-backend/v1" ]] || die "config schema mismatch"
[[ "${node_id}" == "JD-FD-PRIMARY" ]] || die "target node mismatch"
[[ "${instance_id}" =~ ^[0-9a-f-]{36}$ ]] || die "instance id is invalid"
[[ "${backend_id}" =~ ^[a-z0-9-]+$ ]] || die "backend id is invalid"
[[ "${container_name}" =~ ^[a-z0-9-]+$ ]] || die "container name is invalid"
[[ "${image}" =~ ^ubuntu@sha256:[0-9a-f]{64}$ ]] || die "image must use an immutable Ubuntu digest"
[[ "${rescue_slot_id}" =~ ^[A-Za-z0-9._-]+$ ]] || die "rescue slot id is invalid"
[[ -x "${docker_bin}" ]] || die "docker executable is unavailable"

requested_backend=${2:-}
[[ -n "${requested_backend}" && "${requested_backend}" == "${backend_id}" ]] ||
  die "backend binding mismatch"

verify_machine() {
  local observed
  observed=$(tr '[:upper:]' '[:lower:]' <"${dmi_path}")
  [[ "${observed}" == "${instance_id}" ]] || die "machine identity mismatch"
}

verify_rescue() {
  grep -Fq -- "--id '${rescue_slot_id}'" "${grub_cfg}" ||
    die "independent Linux rescue slot is unavailable"
}

container_running() {
  [[ "$("${docker_bin}" inspect --format '{{.State.Running}}' "${container_name}" 2>/dev/null || true)" == true ]]
}

observe() {
  if container_running; then
    echo ready
  else
    echo dormant
  fi
}

action=${1:-}
case "${action}" in
  preflight)
    verify_machine
    verify_rescue
    "${docker_bin}" image inspect "${image}" >/dev/null
    echo "GUANGHU_LINUX_SUBCONTROL_PREFLIGHT_OK"
    ;;
  observe)
    observe
    ;;
  wake)
    verify_machine
    verify_rescue
    [[ "$(observe)" == dormant ]] || die "backend is not dormant before wake"
    "${docker_bin}" image inspect "${image}" >/dev/null
    mkdir -p "${state_root}" "${receipt_root}"
    "${docker_bin}" run \
      --detach \
      --pull never \
      --name "${container_name}" \
      --hostname jd-linux-subcontrol \
      --label guanghu.owner=GUANGHU_OS \
      --label "guanghu.node=${node_id}" \
      --label "guanghu.backend=${backend_id}" \
      --network none \
      --read-only \
      --tmpfs /run:rw,nosuid,nodev,noexec,size=16m \
      --tmpfs /tmp:rw,nosuid,nodev,noexec,size=16m \
      --cap-drop ALL \
      --security-opt no-new-privileges \
      --pids-limit 64 \
      --memory 256m \
      --cpus 0.50 \
      "${image}" \
      /bin/sh -ceu 'trap "exit 0" TERM INT; while :; do sleep 3600 & wait $!; done' \
      >/dev/null
    container_running || die "backend failed readiness readback"
    echo "GUANGHU_LINUX_SUBCONTROL_READY"
    ;;
  execute|verify)
    capability=${3:-}
    request_id=${4:-}
    [[ "${capability}" == runtime-identity-readback ]] || die "capability is not allowlisted"
    [[ "${request_id}" =~ ^[A-Za-z0-9._-]+$ ]] || die "request id is invalid"
    container_running || die "backend is not ready"
    mkdir -p "${state_root}" "${receipt_root}"
    target_file="${state_root}/${request_id}.target"
    if [[ "${action}" == execute ]]; then
      temporary="${target_file}.tmp"
      "${docker_bin}" exec "${container_name}" /bin/sh -ceu \
        '. /etc/os-release; printf "ID=%s\nVERSION_ID=%s\n" "$ID" "$VERSION_ID"' \
        >"${temporary}"
      chmod 600 "${temporary}"
      mv "${temporary}" "${target_file}"
      echo "GUANGHU_LINUX_SUBCONTROL_EXECUTED"
    else
      [[ -f "${target_file}" ]] || die "target readback is missing"
      expected=$'ID=ubuntu\nVERSION_ID=22.04'
      stored=$(cat "${target_file}")
      live=$("${docker_bin}" exec "${container_name}" /bin/sh -ceu \
        '. /etc/os-release; printf "ID=%s\nVERSION_ID=%s\n" "$ID" "$VERSION_ID"')
      [[ "${stored}" == "${expected}" && "${live}" == "${expected}" ]] ||
        die "target readback mismatch"
      echo "GUANGHU_LINUX_SUBCONTROL_TARGET_VERIFIED"
    fi
    ;;
  reclaim)
    if container_running; then
      "${docker_bin}" rm --force "${container_name}" >/dev/null
    fi
    [[ "$(observe)" == dormant ]] || die "backend reclaim failed"
    echo "GUANGHU_LINUX_SUBCONTROL_DORMANT"
    ;;
  *)
    die "usage: $0 <preflight|observe|wake|execute|verify|reclaim> <backend-id> [capability] [request-id]"
    ;;
esac