fix: make Linux installation permissions deterministic

deploy/update.sh

@@ -18,6 +18,8 @@ RELEASE_API_URL=${UPDATE_RELEASE_API_URL:-}
 RELEASE_TOKEN=${UPDATE_RELEASE_TOKEN:-}
 CACHE_ROOT=${KAIDI_UPDATE_CACHE_ROOT:-$STATE_ROOT/cache}
 SERVICE_NAME=${KAIDI_SERVICE_NAME:-kaidi-finance.service}
+SERVICE_USER=${KAIDI_SERVICE_USER:-kaidi}
+SERVICE_GROUP=${KAIDI_SERVICE_GROUP:-kaidi}
 HEALTH_URL=${KAIDI_HEALTH_URL:-http://127.0.0.1:18080/actuator/health}
 APP_INDEX_URL=${KAIDI_APP_INDEX_URL:-http://127.0.0.1:18080/}
 LOCK_FILE=$STATE_ROOT/update.lock
@@ -38,9 +40,10 @@ case "$HEALTH_ATTEMPTS:$HEALTH_INTERVAL_SECONDS" in
 esac
 [ "$HEALTH_ATTEMPTS" -ge 1 ] || { printf '%s\n' "Update health-check attempts must be at least 1" >&2; exit 1; }
 
-mkdir -p "$STATE_ROOT/inbox" "$PROCESSING_DIR" "$FAILED_REQUEST_ROOT" "$TRANSACTION_ROOT" \
-  "$STATE_ROOT/work" "$APP_ROOT/releases" "$LOG_ROOT" "$BACKUP_ROOT" "$CACHE_ROOT"
-chmod 0700 "$PROCESSING_DIR" "$FAILED_REQUEST_ROOT" "$TRANSACTION_ROOT" "$BACKUP_ROOT" "$CACHE_ROOT"
+bootstrap_die() {
+  printf '%s\n' "$1" >&2
+  exit 1
+}
 
 status() {
   state=$1
@@ -77,6 +80,43 @@ fail() {
   exit 1
 }
 
+prepare_update_layout() {
+  id "$SERVICE_USER" >/dev/null 2>&1 || bootstrap_die "Service user $SERVICE_USER is missing"
+  id -nG "$SERVICE_USER" | tr ' ' '\n' | grep -Fxq "$SERVICE_GROUP" \
+    || bootstrap_die "Service user $SERVICE_USER is not a member of group $SERVICE_GROUP"
+  command -v runuser >/dev/null 2>&1 || bootstrap_die "runuser is required"
+  install -d -o root -g "$SERVICE_GROUP" -m 0750 \
+    "$APP_ROOT" "$APP_ROOT/releases" "$APP_ROOT/runtime" "$APP_ROOT/bin" "$STATE_ROOT" \
+    || bootstrap_die "Managed application or update directories could not be prepared"
+  install -d -o "$SERVICE_USER" -g "$SERVICE_GROUP" -m 0750 "$STATE_ROOT/inbox" "$LOG_ROOT" \
+    || bootstrap_die "Writable update directories could not be prepared"
+  install -d -o root -g root -m 0700 \
+    "$PROCESSING_DIR" "$FAILED_REQUEST_ROOT" "$TRANSACTION_ROOT" "$STATE_ROOT/work" \
+    "$BACKUP_ROOT" "$CACHE_ROOT" \
+    || bootstrap_die "Private update directories could not be prepared"
+}
+
+secure_release_tree() {
+  release_dir=$1
+  chown -R root:"$SERVICE_GROUP" "$release_dir" || return 1
+  find "$release_dir" -type d -exec chmod 0750 {} + || return 1
+  find "$release_dir" -type f -exec chmod 0640 {} + || return 1
+  chmod 0750 "$release_dir/ops/update.sh"
+}
+
+verify_release_access() {
+  release_dir=$1
+  # shellcheck disable=SC2016 # Positional parameters are expanded by the child shell.
+  runuser -u "$SERVICE_USER" -- sh -c \
+    'cd "$1" && test -r app.jar && test -r public/index.html' sh "$release_dir" \
+    || return 1
+  runuser -u "$SERVICE_USER" -- test -x "$APP_ROOT/runtime/java/bin/java" \
+    || return 1
+  runuser -u "$SERVICE_USER" -- "$APP_ROOT/runtime/java/bin/java" -version >/dev/null 2>&1
+}
+
+prepare_update_layout
+
 # shellcheck disable=SC2329 # Invoked by the EXIT trap below.
 cleanup() {
   rc=$?
@@ -329,6 +369,7 @@ rollback_active_transaction() {
     rollback_ok=false
   fi
   restore_operations || rollback_ok=false
+  systemctl reset-failed "$SERVICE_NAME" >/dev/null 2>&1 || true
   systemctl start "$SERVICE_NAME" || rollback_ok=false
   if [ "$rollback_ok" = true ] && verify_app_surface; then
     remove_failed_release "$failed_release"
@@ -520,8 +561,14 @@ if [ -e "$RELEASE_DIR" ]; then
   rm -rf "$RELEASE_DIR" || fail "Failed release staging directory could not be cleaned"
 fi
 mv "$WORK_DIR/extracted" "$RELEASE_DIR" || fail "Release directory could not be activated"
-chown -R root:root "$RELEASE_DIR" || fail "Release ownership could not be secured"
-chmod -R go-w "$RELEASE_DIR" || fail "Release permissions could not be secured"
+secure_release_tree "$RELEASE_DIR" || fail "Release ownership or permissions could not be secured"
+if command -v restorecon >/dev/null 2>&1; then
+  restorecon -RF "$APP_ROOT" >/dev/null 2>&1 || true
+fi
+if ! verify_release_access "$RELEASE_DIR"; then
+  rm -rf "$RELEASE_DIR"
+  fail "Service user cannot access the release or managed Java runtime"
+fi
 
 PREVIOUS_TARGET=$(readlink "$APP_ROOT/current" 2>/dev/null || true)
 mkdir "$ACTIVE_TRANSACTION"
@@ -552,6 +599,7 @@ write_transaction_value phase APP_SWITCHED
 
 status RUNNING "Starting and verifying release $TARGET_VERSION" "$TARGET_VERSION"
 write_transaction_value phase HEALTH_CHECKING
+systemctl reset-failed "$SERVICE_NAME" >/dev/null 2>&1 || true
 if systemctl start "$SERVICE_NAME" \
   && systemctl is-active --quiet kaidi-update.path \
   && verify_app_surface; then