han-app/VM1_app/codebase/backend/deployment/preflight.sh

#!/bin/sh
set -eu

ROOT=${1:-/opt/han-chat/current/backend}
ENV_FILE=${2:-/etc/han/vm1.env}
MANIFEST=${3:-/run/han-chat/secrets/manifest}
failures=0

fail() {
  echo "FAIL: $*" >&2
  failures=$((failures + 1))
}

value() {
  awk -F= -v key="$1" '$1 == key {print substr($0, index($0, "=") + 1)}' "$ENV_FILE"
}

[ "$(id -u)" -eq 0 ] || fail "preflight must run as root"
[ -f "$ROOT/docker-compose.yml" ] || fail "root docker-compose.yml is missing"
[ -f "$ENV_FILE" ] || fail ".env is missing"
[ -f "$MANIFEST" ] || fail "runtime secret manifest is missing"
[ -d /var/log/journal ] || fail "persistent journald directory is missing"
[ -L /opt/han-chat/current ] || fail "/opt/han-chat/current must be a root-controlled release link"
[ "$(/usr/bin/stat -c '%U:%G' /opt/han-chat/current)" = root:root ] ||
  fail "active release link must be root:root"
nginx_compose="$ROOT/nginx/docker-compose.yml"
[ -f "$nginx_compose" ] || fail "nginx compose fragment is missing"
if [ -f "$nginx_compose" ]; then
  grep -Fq '/var/lib/han-chat/public-tls' "$nginx_compose" &&
    grep -Fq '/run/tls' "$nginx_compose" ||
    fail "nginx must mount host public-tls staging at /run/tls"
  grep -Fq '/var/lib/han-chat/acme' "$nginx_compose" ||
    fail "nginx must mount host ACME webroot"
  ! grep -Fq '/etc/letsencrypt' "$nginx_compose" ||
    fail "nginx/certbot compose must not mount the root Certbot tree"
  ! grep -Eq '^[[:space:]]+certbot:' "$nginx_compose" ||
    fail "certbot must not be a Compose service"
  ! grep -Eq 'nginx-(certs|acme)' "$nginx_compose" ||
    fail "legacy nginx TLS named volumes are forbidden"
fi
! grep -Eq '^[[:space:]]+nginx-(certs|acme)' "$ROOT/docker-compose.yml" ||
  fail "legacy nginx TLS named volumes are forbidden"

if [ -d "$ROOT" ]; then
  if find "$ROOT" -type l -print -quit | grep -q .; then
    fail "active release contains a symlink"
  fi
  while IFS= read -r protected; do
    [ -f "$protected" ] || {
      fail "protected deployment file is missing: $protected"
      continue
    }
    [ "$(/usr/bin/stat -c '%U:%G' "$protected")" = root:root ] ||
      fail "$protected must be root:root"
    mode=$(/usr/bin/stat -c '%A' "$protected")
    case "$mode" in
      ??????w???|????????w?) fail "$protected is writable by group/other" ;;
    esac
  done <<EOF
$ROOT/docker-compose.yml
$ROOT/deployment/preflight.sh
$ROOT/deployment/han-stack@.service
$ROOT/deployment/han-host-otel-collector@.service
$ROOT/deployment/observability/otel-host-collector.yaml
$ROOT/deployment/scripts/tls-deploy-hook.sh
$ROOT/deployment/secrets/han-compose
$ROOT/deployment/secrets/han-secrets
EOF
fi

for executable in \
  "$ROOT/deployment/preflight.sh" \
  "$ROOT/deployment/scripts/tls-deploy-hook.sh" \
  "$ROOT/deployment/secrets/han-compose" \
  "$ROOT/deployment/secrets/han-secrets"
do
  [ -x "$executable" ] || fail "required executable is not executable: $executable"
  if [ -f "$executable" ] && LC_ALL=C grep -q "$(printf '\r')" "$executable"; then
    fail "CRLF is forbidden in executable: $executable"
  fi
done

if [ -f "$ENV_FILE" ]; then
  if grep -Eq '(^|_)(PASSWORD|SECRET|TOKEN|DATABASE_URL|REDIS_URL|PRIVATE_KEY|ACCESS_KEY)=' "$ENV_FILE"; then
    fail ".env contains a secret-shaped assignment"
  fi
  if grep -Eq '=<[^>]+>|change-me|example\.(com|org|net)|\.invalid([:/]|$)' "$ENV_FILE"; then
    fail ".env contains placeholders"
  fi
  [ "$(value APP_ENV)" = production ] || fail "APP_ENV must be production"
  [ "$(value SECRETS_SOURCE)" = selectel ] || fail "production SECRETS_SOURCE must be selectel"
  [ "$(value FRONTEND_DEV_PROXY_ENABLED)" = false ] ||
    fail "FRONTEND_DEV_PROXY_ENABLED must be false"
  [ "$(value NGINX_TLS_ENABLED)" = true ] || fail "NGINX_TLS_ENABLED must be true"
  [ "$(value NGINX_HTTP_PORT)" = 80 ] || fail "nginx must publish host port 80"
  [ "$(value NGINX_HTTPS_PORT)" = 443 ] || fail "nginx must publish host port 443"
  [ -n "$(value HOST_NAME)" ] || fail "HOST_NAME is required for telemetry correlation"
  echo "$(value TELEMETRYGEN_IMAGE)" |
    grep -Eq '^ghcr\.io/.+@sha256:[a-f0-9]{64}$' ||
    fail "TELEMETRYGEN_IMAGE must be pinned by digest"
  [ "$(value NGINX_TLS_CERTIFICATE)" = /run/tls/fullchain.pem ] ||
    fail "nginx certificate must use staged /run/tls/fullchain.pem"
  [ "$(value NGINX_TLS_CERTIFICATE_KEY)" = /run/tls/privkey.pem ] ||
    fail "nginx key must use staged /run/tls/privkey.pem"
  safety_url=$(value MESSAGE_SAFETY_URL)
  echo "$safety_url" | grep -Eq '^https://[A-Za-z0-9.-]+:8443$' ||
    fail "MESSAGE_SAFETY_URL must be private HTTPS VM2 :8443"
  echo "$safety_url" | grep -Eq '(message-safety|localhost|127\.0\.0\.1)' &&
    fail "MESSAGE_SAFETY_URL must not point to a local/stub service"
  safety_extra_host=$(value MESSAGE_SAFETY_EXTRA_HOST)
  if ! python3 - "$safety_url" "$safety_extra_host" <<'PY'
import ipaddress
import socket
import sys
from urllib.parse import urlparse

host = urlparse(sys.argv[1]).hostname
try:
    mapped_host, mapped_ip = sys.argv[2].rsplit("=", 1)
    mapped_address = ipaddress.ip_address(mapped_ip)
except (IndexError, ValueError):
    raise SystemExit(1)
allowed = (
    ipaddress.ip_network("10.0.0.0/8"),
    ipaddress.ip_network("172.16.0.0/12"),
    ipaddress.ip_network("192.168.0.0/16"),
    ipaddress.ip_network("fc00::/7"),
)
if (
    not host
    or mapped_host != host
    or not any(mapped_address in network for network in allowed)
):
    raise SystemExit(1)
try:
    addresses = {
        ipaddress.ip_address(item[4][0])
        for item in socket.getaddrinfo(host, 8443, type=socket.SOCK_STREAM)
    }
except OSError:
    raise SystemExit(1)
if (
    not addresses
    or mapped_address not in addresses
    or any(not any(address in network for network in allowed) for address in addresses)
):
    raise SystemExit(1)
PY
  then
    fail "MESSAGE_SAFETY_URL hostname must resolve only to private VPC addresses and MESSAGE_SAFETY_EXTRA_HOST must map the same address"
  fi
  safety_ca=$(value MESSAGE_SAFETY_CA_HOST_PATH)
  [ -n "$safety_ca" ] || fail "MESSAGE_SAFETY_CA_HOST_PATH is required"
  [ -f "$safety_ca" ] || fail "VM2 internal CA file is missing"
  pg_ca=$(value PG_CA_HOST_PATH)
  [ -f "$pg_ca" ] || fail "managed PostgreSQL CA file is missing"
  for image_key in \
    API_BACKEND_IMAGE \
    BITRIX_LOCAL_APP_IMAGE \
    FRONTEND_STATIC_IMAGE \
    KEYCLOAK_IMAGE \
    NGINX_IMAGE \
    OTEL_COLLECTOR_IMAGE \
    OTEL_QUEUE_INIT_IMAGE \
    REDIS_IMAGE \
    SMS_SERVICE_IMAGE \
    TOOLBOX_IMAGE
  do
    image=$(value "$image_key")
    echo "$image" | grep -Eq '@sha256:[0-9a-f]{64}$' ||
      fail "$image_key must be pinned by sha256 digest"
    case "$image" in
      registry.example.ru/*|*@sha256:0000000000000000000000000000000000000000000000000000000000000000)
        fail "$image_key still contains the example image/digest"
        ;;
    esac
  done
fi

resolved_config=$(mktemp)
resolved_services=$(mktemp)
resolved_images=$(mktemp)
trap 'rm -f "$resolved_config" "$resolved_services" "$resolved_images"' EXIT HUP INT TERM
if ! command -v docker >/dev/null 2>&1; then
  fail "docker is required to resolve production Compose"
elif [ -f "$ENV_FILE" ]; then
  if ! docker compose --env-file "$ENV_FILE" -f "$ROOT/docker-compose.yml" \
    config >"$resolved_config"; then
    fail "production Compose does not resolve"
  else
    docker compose --env-file "$ENV_FILE" -f "$ROOT/docker-compose.yml" \
      config --services >"$resolved_services" ||
      fail "cannot enumerate resolved production services"
    docker compose --env-file "$ENV_FILE" -f "$ROOT/docker-compose.yml" \
      config --images >"$resolved_images" ||
      fail "cannot enumerate resolved production images"

    while IFS= read -r image; do
      [ -n "$image" ] || continue
      echo "$image" | grep -Eq '@sha256:[0-9a-f]{64}$' ||
        fail "resolved production image is not digest-pinned: $image"
      case "$image" in
        registry.example.ru/*|*@sha256:0000000000000000000000000000000000000000000000000000000000000000)
          fail "resolved production image still contains an example value: $image"
          ;;
      esac
    done <"$resolved_images"
    [ -s "$resolved_images" ] || fail "resolved production image set is empty"

    for forbidden_service in certbot message-safety bitrix-sync otel-collector-local; do
      ! grep -Fxq "$forbidden_service" "$resolved_services" ||
        fail "local/legacy service resolved in production: $forbidden_service"
    done

    published_services=$(
      awk '
        /^services:$/ { in_services=1; next }
        in_services && /^[^ ]/ { in_services=0 }
        in_services && /^  [A-Za-z0-9_.-]+:$/ {
          service=$1
          sub(/:$/, "", service)
          next
        }
        in_services && /^    ports:$/ { print service }
      ' "$resolved_config"
    )
    [ "$published_services" = nginx ] ||
      fail "only nginx may publish production host ports (found: ${published_services:-none})"
  fi
fi

tls_dir=/var/lib/han-chat/public-tls
getent group han-nginx-tls | awk -F: '$3 == 11001 {found=1} END {exit !found}' ||
  fail "han-nginx-tls group with GID 11001 is missing"
[ "$(/usr/bin/stat -c '%U:%G:%a' "$tls_dir" 2>/dev/null || true)" = root:han-nginx-tls:750 ] ||
  fail "public TLS directory must be root:han-nginx-tls 0750"
for tls_file in fullchain.pem privkey.pem; do
  path="$tls_dir/$tls_file"
  [ -s "$path" ] || {
    fail "public TLS file is missing: $path"
    continue
  }
  [ "$(/usr/bin/stat -c '%U:%G:%a' "$path")" = root:han-nginx-tls:640 ] ||
    fail "$path must be root:han-nginx-tls 0640"
done
if [ -s "$tls_dir/fullchain.pem" ] && [ -s "$tls_dir/privkey.pem" ]; then
  cert_public=$(
    openssl x509 -in "$tls_dir/fullchain.pem" -pubkey -noout 2>/dev/null |
      openssl pkey -pubin -outform DER 2>/dev/null |
      sha256sum | awk '{print $1}'
  ) || cert_public=
  key_public=$(
    openssl pkey -in "$tls_dir/privkey.pem" -passin pass: -pubout -outform DER 2>/dev/null |
      sha256sum | awk '{print $1}'
  ) || key_public=
  [ -n "$cert_public" ] && [ "$cert_public" = "$key_public" ] ||
    fail "public TLS certificate/private key are invalid or do not match"
fi

required_secrets='
DATABASE_URL
REDIS_URL
REDIS_REALTIME_URL
MESSAGE_SAFETY_SERVICE_TOKEN
BITRIX_DATABASE_URL
SMS_DATABASE_URL
KEYCLOAK_DB_PASSWORD
KEYCLOAK_ADMIN_PASSWORD
SELECTEL_S3_ACCESS_KEY
SELECTEL_S3_SECRET_KEY
OTEL_REMOTE_AUTH_HEADER'
if [ -f "$MANIFEST" ]; then
  old_ifs=$IFS
  IFS='
'
  for name in $required_secrets; do
    [ -n "$name" ] || continue
    path=$(awk -F= -v key="$name" '$1 == key {print substr($0, index($0, "=") + 1)}' "$MANIFEST")
    [ -n "$path" ] || {
      fail "manifest is missing $name"
      continue
    }
    [ -f "$path" ] || fail "secret file is missing for $name"
  done
  IFS=$old_ifs
fi

host_otel_config="$ROOT/deployment/observability/otel-host-collector.yaml"
if [ -f "$host_otel_config" ]; then
  grep -Fq 'filelog/docker' "$host_otel_config" ||
    fail "host collector must include Docker filelog"
  grep -Fq 'journald/host' "$host_otel_config" ||
    fail "host collector must include journald allow-list"
  grep -Fq 'filter/allowlist' "$host_otel_config" ||
    fail "host collector must drop non-allowlisted containers"
  grep -Fq 'file_storage' "$host_otel_config" ||
    fail "host collector must persist offsets and exporter queue"
  ! grep -Fq '/var/run/docker.sock' "$host_otel_config" ||
    fail "Docker socket access is forbidden"
fi
[ -x /usr/local/bin/otelcol-contrib ] ||
  fail "pinned host otelcol-contrib is not installed"
otel_checksum=/usr/local/share/han-otel/otelcol-contrib.sha256
otel_version=/usr/local/share/han-otel/otelcol-contrib.version
[ -f "$otel_checksum" ] || fail "host Collector checksum record is missing"
[ -f "$otel_version" ] || fail "host Collector version record is missing"
if [ -f "$otel_checksum" ]; then
  (cd / && sha256sum -c "$otel_checksum") ||
    fail "host Collector binary checksum mismatch"
fi
if [ -x /usr/local/bin/otelcol-contrib ] && [ -f "$otel_version" ]; then
  /usr/local/bin/otelcol-contrib --version 2>&1 |
    grep -Fq "$(cat "$otel_version")" ||
    fail "host Collector binary version mismatch"
fi
if [ -x /usr/local/bin/otelcol-contrib ] && [ -f "$host_otel_config" ]; then
  /usr/local/bin/otelcol-contrib validate --config="$host_otel_config" ||
    fail "host collector config validation failed"
fi

if [ -x "$ROOT/scripts/validate-env" ] && [ -f "$ENV_FILE" ] && [ -f "$MANIFEST" ]; then
  "$ROOT/scripts/validate-env" "$ENV_FILE" --runtime-manifest "$MANIFEST" ||
    fail "config/runtime validator rejected the production inputs"
else
  fail "config/runtime validator or its inputs are unavailable"
fi

if [ "$failures" -ne 0 ]; then
  echo "preflight: $failures failure(s); VM1 deployment remains closed" >&2
  exit 1
fi
echo "preflight: static VM1 production gates passed"