MoShiro/packaging/update/moshiro-update.sh

#!/bin/sh
# MoShiro 容器节点（Agent）的宿主机更新器。用法：moshiro-update <check|apply> <组件>   组件 = agent
#
# 信任模型（防篡改、防中间人）：
#   - 只接受用发布私钥签名的清单（ECDSA P-256 / SHA-256，对清单文件的原始字节签名）；公钥放在本机，面板和 Registry 都改不了它。
#   - 清单里写的是镜像的 digest：拉取时用 repo@sha256:…，内容寻址，Registry 或链路上的人换不出另一个镜像。
#   - 镜像仓库、组件、服务名都来自本机的配置（/etc/moshiro-update/<组件>.conf），清单里写什么都不会改变拉哪个仓库。
#   - 清单的 seq 必须比已安装的大（防止旧清单被重放回滚）。
#   - 业务进程（Agent）只能往数据目录里写「清单」和「请求」文件，更新由这个宿主机脚本完成，业务进程不执行任何命令。
# Shell snippets below execute inside helper containers.
# shellcheck disable=SC2016
set -eu
umask 077

CMD="${1:-}"
COMPONENT="${2:-}"
case "$CMD" in check|apply) ;; *) echo "usage: moshiro-update <check|apply> <agent>" >&2; exit 2 ;; esac
case "$COMPONENT" in agent) ;; *) echo "component must be agent" >&2; exit 2 ;; esac

if [ "$(id -u)" -eq 0 ]; then
  DEFAULT_CONF_DIR=/etc/moshiro-update
else
  DEFAULT_CONF_DIR="${XDG_CONFIG_HOME:-$HOME/.config}/moshiro"
fi
CONF="${MOSHIRO_UPDATE_CONF_DIR:-$DEFAULT_CONF_DIR}/$COMPONENT.conf"
[ -r "$CONF" ] || { echo "missing $CONF" >&2; exit 2; }
# 配置：DATA_DIR（容器里 /etc/moshiro 对应的宿主机目录）、IMAGE_REPO、UNIT、PUBKEY、POLICY（auto|manual）、AUTHFILE（可选）、
#       STATE_DIR（已安装版本的记录）、ENV_FILE（容器单元读取的 IMAGE=… 文件）
# shellcheck disable=SC1090
. "$CONF"
: "${IMAGE_REPO:?}" "${UNIT:?}" "${PUBKEY:?}" "${ENV_FILE:?}"
POLICY="${POLICY:-manual}"
STATE_DIR="${STATE_DIR:-/var/lib/moshiro-update}"
ENGINE="${ENGINE:-podman}"
SERVICE_SCOPE="${SERVICE_SCOPE:-system}"
DATA_VOLUME="${DATA_VOLUME:-}"
STATUS_VOLUME="${STATUS_VOLUME:-moshiro-$COMPONENT-status}"
CHANNEL="${CHANNEL:-}"
HEALTH_URL="${HEALTH_URL:-}"
case "$ENGINE" in podman|docker) ;; *) echo "unsupported engine: $ENGINE" >&2; exit 2 ;; esac
case "$SERVICE_SCOPE" in system|user) ;; *) echo "unsupported service scope: $SERVICE_SCOPE" >&2; exit 2 ;; esac
service_ctl() {
  if [ "$SERVICE_SCOPE" = user ]; then systemctl --user "$@"; else systemctl "$@"; fi
}
pull_image() {
  if [ -n "${AUTHFILE:-}" ]; then
    if [ "$ENGINE" = podman ]; then
      "$ENGINE" pull --quiet --authfile "$AUTHFILE" "$1"
    else
      mkdir -m 700 "$WORK/docker-auth"
      cp "$AUTHFILE" "$WORK/docker-auth/config.json"
      DOCKER_CONFIG="$WORK/docker-auth" "$ENGINE" pull --quiet "$1"
    fi
  else
    "$ENGINE" pull --quiet "$1"
  fi
}
# 业务进程（容器里的 Agent）可写的目录只有 update/：它在里面放 offer.manifest、offer.sig、apply.request。
# 更新器在容器外运行：只读取普通文件（不跟符号链接）、读取之后先复制到自己的临时目录再验证和使用（防止验证之后被换掉），
# 状态目录在 rootless 容器里另行只读挂载；更新器配置和已安装版本状态不挂入容器。
if [ -n "$DATA_VOLUME" ]; then
  # Read/write volume contents via short-lived engine-managed helpers, without host mountpoint access.
  : "${HELPER_IMAGE:?}"
  "$ENGINE" volume inspect "$DATA_VOLUME" >/dev/null
  "$ENGINE" volume inspect "$STATUS_VOLUME" >/dev/null
else
  : "${DATA_DIR:?}"
  UPDATE_DIR="$DATA_DIR/update"
  STATUS_DIR="$DATA_DIR/update-status"
  STATUS="$STATUS_DIR/status.json"
  mkdir -p "$STATUS_DIR"
  chmod 755 "$STATUS_DIR"
fi
INSTALLED="$STATE_DIR/$COMPONENT.installed"     # 两行：seq、digest
ADOPTED="$STATE_DIR/$COMPONENT.key.pem"         # 发布密钥换新之后采用的公钥；有它就只认它，配置里的旧公钥不再用
mkdir -p "$STATE_DIR"
# Serialize check/apply path events for this component.
exec 9>"$STATE_DIR/$COMPONENT.lock"
flock -w 290 9 || { echo "another update is still running" >&2; exit 1; }
WORK="$(mktemp -d)"
trap 'rm -rf "$WORK"' EXIT
OFFER_MANIFEST="$WORK/offer.manifest"
OFFER_SIG="$WORK/offer.sig"

# 直接从镜像仓库取清单（不依赖面板，也不依赖 Release）：CI 把签名清单放在 <IMAGE_REPO>-manifest:<通道> 这个没有文件层的小镜像的标签里。
# 通道取配置里的 CHANNEL；没有的话按仓库路径推断（moshiro-dev → dev，moshiro → release）。MANIFEST_REF 可以直接指定；FEED=off 关掉。
# 取到的只是「候选」：之后照常用本机公钥验签，仓库被改了也只会验签失败。
feed_ref() {
  if [ -n "${MANIFEST_REF:-}" ]; then printf '%s' "$MANIFEST_REF"; return 0; fi
  channel="$CHANNEL"
  if [ -z "$channel" ]; then
    case "$IMAGE_REPO" in */moshiro-dev/*) channel=dev ;; */moshiro/*) channel=release ;; esac
  fi
  [ -n "$channel" ] || return 1
  printf '%s-manifest:%s' "$IMAGE_REPO" "$channel"
}

fetch_feed() {
  [ "${FEED:-auto}" != off ] || return 1
  ref="$(feed_ref)" || return 1
  pull_image "$ref" >/dev/null 2>&1 || return 1
  labels="$("$ENGINE" image inspect --format '{{index .Config.Labels "org.moshiro.manifest"}}|{{index .Config.Labels "org.moshiro.signature"}}' "$ref" 2>/dev/null)" || return 1
  manifest_b64="${labels%%|*}"; signature="${labels#*|}"
  [ -n "$manifest_b64" ] && [ -n "$signature" ] || return 1
  printf '%s' "$manifest_b64" | base64 -d > "$OFFER_MANIFEST" 2>/dev/null || return 1
  printf '%s' "$signature" > "$OFFER_SIG"
}

# 仓库里的清单优先（验签通过才用）；取不到或验不过就用面板放进信箱的那份
fetch_offer() {
  rm -f "$OFFER_MANIFEST" "$OFFER_SIG"
  feed_error=""
  if fetch_feed; then
    if verify_offer; then return 0; fi
    feed_error="$ERR"
    cp "$OFFER_MANIFEST" "$WORK/feed.manifest"; cp "$OFFER_SIG" "$WORK/feed.sig"
  fi
  fetch_mailbox
  # 信箱里没有东西，而仓库里的清单没通过（验证失败，或者就是已经装的这一份）：留下仓库的那份，让后面照常报出原因，而不是悄悄当作没有更新
  if [ -n "$feed_error" ] && [ ! -s "$OFFER_MANIFEST" ]; then
    cp "$WORK/feed.manifest" "$OFFER_MANIFEST"; cp "$WORK/feed.sig" "$OFFER_SIG"
  fi
}

# 把业务进程放的清单和签名复制到临时目录；不是普通文件（缺失、符号链接）就当作没有
fetch_mailbox() {
  rm -f "$OFFER_MANIFEST" "$OFFER_SIG"
  for name in offer.manifest offer.sig; do
    if [ -n "$DATA_VOLUME" ]; then
      "$ENGINE" run --pull=never --rm --network none --read-only --cap-drop ALL --security-opt no-new-privileges \
        -v "$DATA_VOLUME:/etc/moshiro:ro" --entrypoint /bin/sh "$HELPER_IMAGE" \
        -c 'f="/etc/moshiro/update/$1"; if [ -f "$f" ] && [ ! -L "$f" ]; then head -c 8192 "$f"; fi' sh "$name" > "$WORK/$name"
      continue
    fi
    f="$UPDATE_DIR/$name"
    if [ -f "$f" ] && [ ! -L "$f" ]; then head -c 8192 "$f" > "$WORK/$name"; fi
  done
}

now() { date +%s; }

# 状态文件：更新器写，Agent 从只读挂载中读取并随心跳上报。
write_status() { # state available_json error
  tmp="$WORK/status.json"
  printf '{"state":"%s","current":"%s","available":%s,"error":"%s","at":%s}\n' \
    "$1" "$(current_digest)" "${2:-null}" "$(printf '%s' "${3:-}" | tr -d '"\\\n' | cut -c1-200)" "$(now)" > "$tmp"
  if [ -n "$DATA_VOLUME" ]; then
    # 状态卷的属主可能是容器里的 app（1654）；root 丢掉全部能力后写不进去，所以只加回 DAC_OVERRIDE
    "$ENGINE" run --pull=never --rm -i --user 0 --network none --read-only --cap-drop ALL --cap-add DAC_OVERRIDE --security-opt no-new-privileges \
      -v "$STATUS_VOLUME:/status" --entrypoint /bin/sh "$HELPER_IMAGE" \
      -c 'tmp=$(mktemp /status/.status.XXXXXX); cat > "$tmp"; chmod 644 "$tmp"; mv -f "$tmp" /status/status.json' < "$tmp"
  else
    chmod 644 "$tmp"
    destination="$(mktemp "$STATUS_DIR/.status.XXXXXX")"
    cp "$tmp" "$destination"
    chmod 644 "$destination"
    mv -f "$destination" "$STATUS"
  fi
}

# 第一次部署（deploy-node.sh 已经验过签并按 digest 固定）时还没有安装记录：清单的 digest 就是正在跑的这个镜像，就记下来，不当作「有更新」
seed_installed() {
  [ ! -r "$INSTALLED" ] || return 0
  running="$(sed -n 's/^IMAGE=.*@\(sha256:[0-9a-f]*\)$/\1/p' "$ENV_FILE" 2>/dev/null | head -n 1)"
  if [ -n "$running" ] && [ "$running" = "$M_DIGEST" ]; then printf '%s\n%s\n' "$M_SEQ" "$M_DIGEST" > "$INSTALLED"; fi
}

current_digest() { [ -r "$INSTALLED" ] && sed -n 2p "$INSTALLED" || echo ""; }
installed_seq() { [ -r "$INSTALLED" ] && sed -n 1p "$INSTALLED" || echo 0; }

# 面板（管理员）请求更新：apply.request 在数据卷里放着就是请求了
fetch_request() {
  if [ -n "$DATA_VOLUME" ]; then
    "$ENGINE" run --pull=never --rm --network none --read-only --cap-drop ALL --security-opt no-new-privileges \
      -v "$DATA_VOLUME:/etc/moshiro:ro" --entrypoint /bin/sh "$HELPER_IMAGE" \
      -c 'f=/etc/moshiro/update/apply.request; if [ -f "$f" ] && [ ! -L "$f" ]; then printf yes; fi'
  elif [ -f "$UPDATE_DIR/apply.request" ] && [ ! -L "$UPDATE_DIR/apply.request" ]; then
    printf yes
  fi
}

# 从扁平的清单里取字段（清单由 CI 生成，只有字符串和整数，没有嵌套）
field() { sed -n 's/.*"'"$1"'":"\([^"]*\)".*/\1/p' "$OFFER_MANIFEST" | head -n 1; }
num_field() { sed -n 's/.*"'"$1"'":\([0-9][0-9]*\).*/\1/p' "$OFFER_MANIFEST" | head -n 1; }

# 验签 + 字段校验。成功时设置 M_DIGEST、M_TAG、M_VERSION、M_SEQ；失败返回非 0，原因在 ERR。
verify_offer() {
  ERR=""
  [ -s "$OFFER_MANIFEST" ] && [ -s "$OFFER_SIG" ] || { ERR="no offer"; return 1; }
  [ "$(wc -c < "$OFFER_MANIFEST")" -le 4096 ] || { ERR="manifest too large"; return 1; }
  sigbin="$WORK/sig.bin"
  base64 -d "$OFFER_SIG" > "$sigbin" 2>/dev/null || { ERR="bad signature encoding"; return 1; }
  key="$PUBKEY"
  [ -s "$ADOPTED" ] && key="$ADOPTED"
  openssl dgst -sha256 -verify "$key" -signature "$sigbin" "$OFFER_MANIFEST" >/dev/null 2>&1 || { ERR="signature check failed"; return 1; }
  # 密钥延续：这份清单由当前密钥签名，里面的 next_key（SPKI DER 的 base64）就是以后要认的密钥；写不成合法的公钥就整份拒绝
  M_NEXT=""
  next="$(field next_key)"
  if [ -n "$next" ]; then
    printf '%s' "$next" | base64 -d > "$WORK/next.der" 2>/dev/null \
      && openssl pkey -pubin -inform DER -in "$WORK/next.der" -outform PEM -out "$WORK/next.pem" >/dev/null 2>&1 \
      && [ -s "$WORK/next.pem" ] || { ERR="next_key is not a public key"; return 1; }
    M_NEXT="$WORK/next.pem"
  fi
  [ "$(field component)" = "$COMPONENT" ] || { ERR="manifest is for another component"; return 1; }
  [ "$(field image)" = "$IMAGE_REPO" ] || { ERR="manifest names another repository"; return 1; }
  if [ -n "$CHANNEL" ]; then
    [ "$(field channel)" = "$CHANNEL" ] || { ERR="manifest is for another channel"; return 1; }
  fi
  M_DIGEST="$(field digest)"
  case "$M_DIGEST" in sha256:*) ;; *) ERR="bad digest"; return 1 ;; esac
  [ "${#M_DIGEST}" -eq 71 ] && [ -z "$(printf '%s' "${M_DIGEST#sha256:}" | tr -d '0-9a-f')" ] || { ERR="bad digest"; return 1; }
  M_TAG="$(field tag)"; M_VERSION="$(field version)"; M_SEQ="$(num_field seq)"
  [ -n "$M_SEQ" ] || { ERR="no seq"; return 1; }
  [ "$M_SEQ" -gt "$(installed_seq)" ] || { ERR="stale manifest (seq $M_SEQ <= $(installed_seq))"; return 1; }
}

# 验过签的清单带了 next_key 就改认它（写临时文件再换名；已经是它就不动）
adopt_next_key() {
  [ -n "$M_NEXT" ] || return 0
  if [ -s "$ADOPTED" ] && cmp -s "$M_NEXT" "$ADOPTED"; then return 0; fi
  cp "$M_NEXT" "$ADOPTED.new" && mv -f "$ADOPTED.new" "$ADOPTED"
  echo "release key replaced by the one in the signed manifest"
}

available_json() { printf '{"version":"%s","tag":"%s","digest":"%s","seq":%s}' "$(printf '%s' "$M_VERSION" | tr -d '\042\134')" "$(printf '%s' "$M_TAG" | tr -d '\042\134')" "$M_DIGEST" "$M_SEQ"; }

do_check() {
  fetch_offer
  if ! verify_offer; then
    # 没有新清单，或者清单已经安装过：不算错误，只是没有更新
    case "$ERR" in
      "no offer") write_status up_to_date null "" ;;
      "stale manifest"*)
        write_status up_to_date null "" ;;
      *) write_status invalid null "$ERR"; echo "offer rejected: $ERR" >&2 ;;
    esac
    return 0
  fi
  adopt_next_key
  seed_installed
  if [ "$M_DIGEST" = "$(current_digest)" ]; then
    write_status up_to_date null ""
    return 0
  fi
  write_status available "$(available_json)" ""
  echo "update available: $M_TAG ($M_DIGEST)"
  if [ "$POLICY" = auto ]; then do_apply; fi
  return 0
}

do_apply() {
  if [ -n "$DATA_VOLUME" ]; then
    "$ENGINE" run --pull=never --rm --network none --read-only --cap-drop ALL --security-opt no-new-privileges \
      -v "$DATA_VOLUME:/etc/moshiro" --entrypoint /bin/sh "$HELPER_IMAGE" -c 'rm -f /etc/moshiro/update/apply.request'
  else
    rm -f "$UPDATE_DIR/apply.request"
  fi
  fetch_offer
  verify_offer || { write_status invalid null "$ERR"; echo "refusing to apply: $ERR" >&2; return 1; }
  adopt_next_key
  avail="$(available_json)"
  write_status applying "$avail" ""
  ref="$IMAGE_REPO@$M_DIGEST"
  if ! pull_image "$ref" >/dev/null 2>&1; then
    write_status failed "$avail" "pull failed"; echo "pull failed: $ref" >&2; return 1
  fi
  prev="$(cat "$ENV_FILE" 2>/dev/null || true)"
  printf 'IMAGE=%s\n' "$ref" > "$ENV_FILE.new" && mv -f "$ENV_FILE.new" "$ENV_FILE"
  service_ctl restart "$UNIT" || true
  # 健康确认：服务在跑，并且 20 秒内没有再被重启（崩溃循环算失败）
  sleep 5
  before="$(service_ctl show -p NRestarts --value "$UNIT" 2>/dev/null || echo 0)"
  sleep 15
  after="$(service_ctl show -p NRestarts --value "$UNIT" 2>/dev/null || echo 0)"
  healthy=true
  if [ -n "$HEALTH_URL" ]; then
    if ! curl --fail --silent --show-error --max-time 5 "$HEALTH_URL" | grep -q '"status":"ok"'; then healthy=false; fi
  fi
  if [ "$healthy" = true ] && service_ctl is-active --quiet "$UNIT" && [ "$before" = "$after" ]; then
    printf '%s\n%s\n' "$M_SEQ" "$M_DIGEST" > "$INSTALLED"
    write_status up_to_date null ""
    echo "updated to $M_TAG"
    return 0
  fi
  echo "new version did not become healthy, rolling back" >&2
  if [ -n "$prev" ]; then printf '%s\n' "$prev" > "$ENV_FILE"; fi
  service_ctl restart "$UNIT" || true
  write_status rolled_back "$avail" "new version did not become healthy"
  return 1
}

case "$CMD" in
  check)
    if [ -n "$DATA_VOLUME" ] && [ "$POLICY" != auto ]; then
      requested="$(fetch_request)"
      if [ -n "$requested" ]; then do_apply; else do_check; fi
    else
      do_check
    fi ;;
  apply) do_apply ;;
esac