banger/scripts/verify.sh

#!/usr/bin/env bash
set -euo pipefail

log() {
  printf '[verify] %s\n' "$*"
}

SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
REPO_ROOT="$(cd "$SCRIPT_DIR/.." && pwd)"
DEFAULT_RUNTIME_DIR="$REPO_ROOT/build/runtime"
if [[ ! -d "$DEFAULT_RUNTIME_DIR" && -d "$REPO_ROOT/runtime" ]]; then
  DEFAULT_RUNTIME_DIR="$REPO_ROOT/runtime"
fi
RUNTIME_DIR="${BANGER_RUNTIME_DIR:-$DEFAULT_RUNTIME_DIR}"
SSH_KEY="$RUNTIME_DIR/id_ed25519"
if [[ ! -d "$RUNTIME_DIR" ]]; then
  log "runtime bundle not found: $RUNTIME_DIR"
  log "run 'make runtime-bundle' or set BANGER_RUNTIME_DIR"
  exit 1
fi
if [[ ! -f "$SSH_KEY" ]]; then
  log "ssh key not found: $SSH_KEY"
  exit 1
fi
DAEMON_LOG="${XDG_STATE_HOME:-$HOME/.local/state}/banger/bangerd.log"
SSH_COMMON_ARGS=(
  -F /dev/null
  -i "$SSH_KEY"
  -o IdentitiesOnly=yes
  -o BatchMode=yes
  -o PreferredAuthentications=publickey
  -o PasswordAuthentication=no
  -o KbdInteractiveAuthentication=no
  -o StrictHostKeyChecking=no
  -o UserKnownHostsFile=/dev/null
)
OPENCODE_PORT=4096

resolve_banger_bin() {
  if [[ -n "${BANGER_BIN:-}" ]]; then
    printf '%s\n' "$BANGER_BIN"
    return
  fi
  if [[ -x "$REPO_ROOT/build/bin/banger" ]]; then
    printf '%s\n' "$REPO_ROOT/build/bin/banger"
    return
  fi
  if [[ -x "$REPO_ROOT/banger" ]]; then
    printf '%s\n' "$REPO_ROOT/banger"
    return
  fi
  if command -v banger >/dev/null 2>&1; then
    command -v banger
    return
  fi
  log "banger binary not found; run 'make build' or set BANGER_BIN"
  exit 1
}

BANGER_BIN="$(resolve_banger_bin)"

firecracker_running() {
  local pid="$1"
  local api_sock="$2"
  local cmdline=""

  if [[ -z "$pid" || "$pid" -le 0 || -z "$api_sock" ]]; then
    return 1
  fi
  if [[ ! -r "/proc/$pid/cmdline" ]]; then
    return 1
  fi
  cmdline="$(cat "/proc/$pid/cmdline" 2>/dev/null | tr '\0' ' ' || true)"
  [[ "$cmdline" == *firecracker* && "$cmdline" == *"$api_sock"* ]]
}

pooled_tap() {
  local tap="$1"
  [[ "$tap" == tap-pool-* ]]
}

wait_for_ssh() {
  local guest_ip="$1"
  local deadline="$2"

  while ((SECONDS < deadline)); do
    if ssh "${SSH_COMMON_ARGS[@]}" -o ConnectTimeout=2 "root@${guest_ip}" "true" >/dev/null 2>&1; then
      return 0
    fi
    sleep 1
  done

  return 1
}

wait_for_tcp() {
  local host="$1"
  local port="$2"
  local deadline="$3"

  while ((SECONDS < deadline)); do
    if (exec 3<>/dev/tcp/"$host"/"$port") >/dev/null 2>&1; then
      return 0
    fi
    sleep 1
  done

  return 1
}

refresh_vm_metadata() {
  if ! VM_JSON="$("$BANGER_BIN" vm show "$VM_NAME" 2>/dev/null)"; then
    return 1
  fi
  TAP="$(printf '%s\n' "$VM_JSON" | jq -r '.runtime.tap_device // empty')"
  VM_DIR="$(printf '%s\n' "$VM_JSON" | jq -r '.runtime.vm_dir // empty')"
  GUEST_IP="$(printf '%s\n' "$VM_JSON" | jq -r '.runtime.guest_ip // empty')"
  API_SOCK="$(printf '%s\n' "$VM_JSON" | jq -r '.runtime.api_sock_path // empty')"
  PID="$(printf '%s\n' "$VM_JSON" | jq -r '.runtime.pid // 0')"
  VM_STATE="$(printf '%s\n' "$VM_JSON" | jq -r '.state // empty')"
  LAST_ERROR="$(printf '%s\n' "$VM_JSON" | jq -r '.runtime.last_error // empty')"
  return 0
}

wait_for_vm_ready() {
  local deadline="$1"

  while ((SECONDS < deadline)); do
    if ! refresh_vm_metadata; then
      sleep 1
      continue
    fi
    if [[ "$VM_STATE" == "error" || -n "$LAST_ERROR" ]]; then
      return 2
    fi
    if [[ -n "$API_SOCK" && "${PID:-0}" -gt 0 ]] && ! firecracker_running "$PID" "$API_SOCK"; then
      return 3
    fi
    if [[ "$VM_STATE" == "running" && -n "$GUEST_IP" && -n "$TAP" && -n "$VM_DIR" && -n "$API_SOCK" && "${PID:-0}" -gt 0 ]]; then
      if [[ -S "$API_SOCK" ]] && ip link show "$TAP" >/dev/null 2>&1; then
        return 0
      fi
    fi
    sleep 1
  done

  return 1
}

dump_diagnostics() {
  log "diagnostics for $VM_NAME"
  "$BANGER_BIN" vm show "$VM_NAME" || true
  if [[ "${PID:-0}" -gt 0 ]]; then
    log "process state for pid $PID"
    ps -fp "$PID" || true
  fi
  log "recent firecracker log"
  "$BANGER_BIN" vm logs "$VM_NAME" 2>/dev/null | tail -n 200 || true
  if [[ -f "$DAEMON_LOG" ]]; then
    log "recent daemon log"
    tail -n 200 "$DAEMON_LOG" || true
  fi
  if [[ -n "${TAP:-}" ]]; then
    log "tap state for $TAP"
    ip link show "$TAP" || true
  fi
  if [[ -n "${API_SOCK:-}" ]]; then
    log "api socket $API_SOCK"
    ls -l "$API_SOCK" 2>/dev/null || true
  fi
  if (( NAT_ENABLED )) && [[ -n "${UPLINK:-}" && -n "${GUEST_IP:-}" && -n "${TAP:-}" ]]; then
    log "nat rules for ${GUEST_IP} via ${UPLINK}"
    sudo iptables -t nat -S POSTROUTING | grep "${GUEST_IP}/32" || true
    sudo iptables -S FORWARD | grep "$TAP" || true
  fi
}

usage() {
  cat <<'EOF'
Usage: ./scripts/verify.sh [--nat] [--image <name>]

Run a basic smoke test for the Go VM workflow.
Use --nat to additionally verify outbound NAT and host rule cleanup.
Use --image to verify a non-default image such as void-exp.
EOF
}

NAT_ENABLED=0
IMAGE_NAME=""
BOOT_TIMEOUT_SECS="${VERIFY_BOOT_TIMEOUT_SECS:-90}"
while [[ $# -gt 0 ]]; do
  case "$1" in
    --nat)
      NAT_ENABLED=1
      shift
      ;;
    --image)
      IMAGE_NAME="${2:-}"
      if [[ -z "$IMAGE_NAME" ]]; then
        usage
        exit 1
      fi
      shift 2
      ;;
    *)
      usage
      exit 1
      ;;
  esac
done

VM_NAME="verify-$(date +%s)"
VM_JSON=""
TAP=""
VM_DIR=""
GUEST_IP=""
UPLINK=""
API_SOCK=""
PID="0"
VM_STATE=""
LAST_ERROR=""

delete_vm() {
  if [[ -n "${VM_NAME:-}" ]]; then
    "$BANGER_BIN" vm delete "$VM_NAME"
  fi
}

cleanup() {
  if [[ -n "${VM_NAME:-}" ]]; then
    "$BANGER_BIN" vm delete "$VM_NAME" >/dev/null 2>&1 || true
  fi
}

trap cleanup EXIT

log "starting VM"
CREATE_ARGS=("$BANGER_BIN" vm create --name "$VM_NAME")
if [[ -n "$IMAGE_NAME" ]]; then
  CREATE_ARGS+=(--image "$IMAGE_NAME")
fi
if (( NAT_ENABLED )); then
  CREATE_ARGS+=(--nat)
fi
"${CREATE_ARGS[@]}" >/dev/null

BOOT_DEADLINE=$((SECONDS + BOOT_TIMEOUT_SECS))

log "waiting for VM runtime readiness"
if wait_for_vm_ready "$BOOT_DEADLINE"; then
  :
else
  status=$?
  case "$status" in
    2) log "vm entered an error state before becoming ready" ;;
    3) log "firecracker exited before the guest became ready" ;;
    *) log "vm did not become ready before timeout" ;;
  esac
  dump_diagnostics
  exit 1
fi

if (( NAT_ENABLED )); then
  UPLINK="$(ip route show default 2>/dev/null | awk '/default/ {print $5; exit}')"
  if [[ -z "$UPLINK" ]]; then
    log "failed to detect uplink interface"
    exit 1
  fi
  log "asserting NAT rules are installed"
  sudo iptables -t nat -C POSTROUTING -s "${GUEST_IP}/32" -o "$UPLINK" -j MASQUERADE
  sudo iptables -C FORWARD -i "$TAP" -o "$UPLINK" -j ACCEPT
  sudo iptables -C FORWARD -i "$UPLINK" -o "$TAP" -m state --state RELATED,ESTABLISHED -j ACCEPT
fi

log "asserting VM is reachable via SSH"
if ! wait_for_ssh "$GUEST_IP" "$BOOT_DEADLINE"; then
  log "ssh did not become ready for ${GUEST_IP}"
  dump_diagnostics
  exit 1
fi
ssh "${SSH_COMMON_ARGS[@]}" "root@${GUEST_IP}" "uname -a" >/dev/null

log "asserting opencode is available and listening in the guest"
ssh "${SSH_COMMON_ARGS[@]}" "root@${GUEST_IP}" "command -v opencode >/dev/null 2>&1 && ss -H -lntp | awk '\$4 ~ /:${OPENCODE_PORT}\$/ { found = 1 } END { exit found ? 0 : 1 }'" >/dev/null

log "asserting opencode server is reachable from the host"
if ! wait_for_tcp "$GUEST_IP" "$OPENCODE_PORT" "$BOOT_DEADLINE"; then
  log "opencode server did not become reachable at ${GUEST_IP}:${OPENCODE_PORT}"
  dump_diagnostics
  exit 1
fi

log "asserting opencode port is reported by banger vm ports"
if ! "$BANGER_BIN" vm ports "$VM_NAME" | grep -F ":${OPENCODE_PORT}" >/dev/null 2>&1; then
  log "banger vm ports did not report ${OPENCODE_PORT}"
  dump_diagnostics
  exit 1
fi

if (( NAT_ENABLED )); then
  log "asserting VM has outbound network access"
  ssh "${SSH_COMMON_ARGS[@]}" "root@${GUEST_IP}" "curl -fsS https://example.com >/dev/null" >/dev/null
fi

log "cleaning up VM"
if ! delete_vm; then
  log "vm delete failed for $VM_NAME"
  dump_diagnostics
  exit 1
fi

log "asserting cleanup success"
if "$BANGER_BIN" vm show "$VM_NAME" >/dev/null 2>&1; then
  log "vm still exists after delete: $VM_NAME"
  exit 1
fi
if ip link show "$TAP" >/dev/null 2>&1; then
  if pooled_tap "$TAP"; then
    log "tap returned to idle pool: $TAP"
  else
    log "tap still exists: $TAP"
    exit 1
  fi
fi
if [[ -d "$VM_DIR" ]]; then
  log "vm dir still exists: $VM_DIR"
  exit 1
fi
if (( NAT_ENABLED )); then
  if sudo iptables -t nat -C POSTROUTING -s "${GUEST_IP}/32" -o "$UPLINK" -j MASQUERADE 2>/dev/null; then
    log "nat rule still exists for ${GUEST_IP}"
    exit 1
  fi
  if sudo iptables -C FORWARD -i "$TAP" -o "$UPLINK" -j ACCEPT 2>/dev/null; then
    log "forward-out rule still exists for ${TAP}"
    exit 1
  fi
  if sudo iptables -C FORWARD -i "$UPLINK" -o "$TAP" -m state --state RELATED,ESTABLISHED -j ACCEPT 2>/dev/null; then
    log "forward-in rule still exists for ${TAP}"
    exit 1
  fi
fi

log "ok"