#!/usr/bin/env bash
# DashCaddy Host-Side Updater
# Triggered by systemd path unit when the container writes trigger.json.
# Reads the trigger, backs up current API + data/, copies new files, rebuilds container.
# Writes result.json so the new container knows the outcome.
#
# This runs on the HOST, outside the container.
#
# Channel selection: by default only "stable" releases are applied. Set
# ALLOW_PRERELEASE=true in /opt/dashcaddy/updates/channel.conf to opt in to
# prerelease/beta/rc channels. Useful for staging hosts, not production.

set -euo pipefail

readonly UPDATES_DIR="/opt/dashcaddy/updates"
readonly TRIGGER_FILE="${UPDATES_DIR}/trigger.json"
readonly RESULT_FILE="${UPDATES_DIR}/result.json"
readonly BACKUPS_DIR="${UPDATES_DIR}/backups"
readonly CONTAINER_NAME="dashcaddy-api"
readonly IMAGE_TAG="dashcaddy-dashcaddy-api:latest"
readonly MAX_BACKUPS=3
readonly HEALTH_TIMEOUT=60
readonly CHANNEL_CONF="${UPDATES_DIR}/channel.conf"

# Data directory backup — stored alongside code backups so everything rolls back together
readonly DATA_SOURCE_DIR="/opt/dashcaddy/dashcaddy-api/data"
readonly DATA_BACKUP_PREFIX="data-backup"

# Updater state (trigger.json / result.json) backup — keeps the audit trail
# (what version we were attempting, what the previous update's outcome was) tied
# to the same versioned backup directory as code + data. After a failed update,
# operators can inspect what was attempted without correlating timestamps, and
# rollback tooling can reconstruct a "what just happened" view of the update
# state machine. NOTE: we do NOT auto-restore trigger.json on rollback — the
# rollback handler reads a fresh trigger.json written by the operator/container;
# restoring the previous attempt's trigger would clobber the active rollback
# request. Backups here are read-only forensic evidence.
readonly UPDATE_STATE_BACKUP_PREFIX="update-state"
readonly TRIGGER_PROCESSING="${TRIGGER_FILE}.processing"

log() { echo "[dashcaddy-update] $(date '+%Y-%m-%d %H:%M:%S') $*"; }

# Decide if a given release channel is acceptable on this host.
# Returns 0 (accept) or 1 (reject) and logs the reason.
channel_allowed() {
  local channel="$1"
  local allow_prerelease="false"

  if [[ -f "$CHANNEL_CONF" ]]; then
    # shellcheck disable=SC1090
    source "$CHANNEL_CONF"
    allow_prerelease="${ALLOW_PRERELEASE:-false}"
  fi

  case "${channel,,}" in
    stable|"")
      return 0
      ;;
    prerelease|beta|rc|alpha)
      if [[ "${allow_prerelease,,}" == "true" ]]; then
        log "Channel '${channel}' accepted (ALLOW_PRERELEASE=true in ${CHANNEL_CONF})"
        return 0
      else
        log "Channel '${channel}' rejected — set ALLOW_PRERELEASE=true in ${CHANNEL_CONF} to accept"
        return 1
      fi
      ;;
    *)
      log "Channel '${channel}' rejected — unknown channel"
      return 1
      ;;
  esac
}

# ── JSON string escaping (DC-122) ─────────────────────────────────────────────
# Complete JSON string encoder for the rare no-python fallback path: mandatory
# escapes (quote, backslash) plus ALL control bytes U+0000–U+001F as \uXXXX or
# their short forms.
json_escape() {
  local s="$1" out="" ch i hex
  for (( i=0; i<${#s}; i++ )); do
    ch="${s:i:1}"
    case "$ch" in
      \\)   out+='\\' ;;
      \")   out+='\"' ;;
      $'\b') out+='\b' ;;
      $'\f') out+='\f' ;;
      $'\n') out+='\n' ;;
      $'\r') out+='\r' ;;
      $'\t') out+='\t' ;;
      *)
        if [[ "$ch" < ' ' || "$ch" == $'\x7f' ]]; then
          printf -v hex '%02x' "'$ch"
          out+="\u00${hex}"
        else
          out+="$ch"
        fi
        ;;
    esac
  done
  printf '%s' "$out"
}

write_result() {
  local success="$1" version="$2" duration="$3"
  shift 3
  local error="${1:-}"

  if command -v python3 >/dev/null 2>&1; then
    python3 - "$success" "$version" "$duration" "$error" > "$RESULT_FILE" <<'PY'
import json, sys, datetime
success, version, duration, error = sys.argv[1], sys.argv[2], sys.argv[3], sys.argv[4]
obj = {
    "success": success == "true",
    "version": version,
    "duration": int(duration) if duration.isdigit() else 0,
    "timestamp": datetime.datetime.now(datetime.timezone.utc).strftime("%Y-%m-%dT%H:%M:%SZ"),
}
if error:
    obj["error"] = error
print(json.dumps(obj, indent=2))
PY
  else
    # Safe fallback: escaped interpolation (no raw quote/backslash leakage).
    local esc_version esc_error
    esc_version=$(json_escape "$version")
    esc_error=$(json_escape "$error")
    if [[ "$success" == "true" ]]; then
      cat > "$RESULT_FILE" <<EOF
{
  "success": true,
  "version": "${esc_version}",
  "duration": ${duration},
  "timestamp": "$(date -u +%Y-%m-%dT%H:%M:%SZ)"
}
EOF
    else
      cat > "$RESULT_FILE" <<EOF
{
  "success": false,
  "version": "${esc_version}",
  "duration": ${duration},
  "error": "${esc_error}",
  "timestamp": "$(date -u +%Y-%m-%dT%H:%M:%SZ)"
}
EOF
    fi
  fi
}

cleanup_old_backups() {
  local count
  count=$(find "$BACKUPS_DIR" -maxdepth 1 -mindepth 1 -type d 2>/dev/null | wc -l)
  if (( count > MAX_BACKUPS )); then
    log "Cleaning old backups (${count} > ${MAX_BACKUPS})"
    find "$BACKUPS_DIR" -maxdepth 1 -mindepth 1 -type d -printf '%T+ %p\n' \
      | sort | head -n $(( count - MAX_BACKUPS )) | cut -d' ' -f2- \
      | xargs rm -rf
  fi
}

# ── Data backup (rsync for efficiency + permissions) ──────────────────────────
backup_data_dir() {
  local backup_dir="$1"
  if [[ -d "$DATA_SOURCE_DIR" ]]; then
    log "Backing up data/ to ${backup_dir}/${DATA_BACKUP_PREFIX}/"
    mkdir -p "${backup_dir}/${DATA_BACKUP_PREFIX}"
    # DC-122: fallback must copy CONTENTS ("dir/.") — a bare "cp -a dir dest"
    # nests a data/ level inside the existing destination dir, which then made
    # restore_data_dir restore data/data/... (rollback restoring nothing).
    rsync -a --delete "$DATA_SOURCE_DIR/" "${backup_dir}/${DATA_BACKUP_PREFIX}/" 2>/dev/null \
      || { log "rsync unavailable — cp fallback (contents copy)"; cp -a "$DATA_SOURCE_DIR/." "${backup_dir}/${DATA_BACKUP_PREFIX}/"; }
    # Manifest of backed-up files — lets the no-rsync restore path mirror
    # rsync --delete semantics (remove live files that the backup lacks).
    ( cd "${backup_dir}/${DATA_BACKUP_PREFIX}" && find . -type f -printf '%P\n' | sort ) \
      > "${backup_dir}/data.manifest" 2>/dev/null || true
    log "Data backup complete ($(du -sh "${backup_dir}/${DATA_BACKUP_PREFIX}" 2>/dev/null | cut -f1))"
  else
    log "WARNING: Data source dir $DATA_SOURCE_DIR not found — skipping data backup"
  fi
}

# ── Updater state backup (trigger.json.processing + result.json) ─────────────
# Captures what was being attempted + the last result so post-mortem can answer
# "why did this fail" without joining timestamps across files. Tolerates absent
# files (first-ever run) and locked files (chattr +i). Idempotent — re-running
# overwrites the previous backup.
backup_update_state() {
  local backup_dir="$1"
  local state_dir="${backup_dir}/${UPDATE_STATE_BACKUP_PREFIX}"
  mkdir -p "$state_dir"

  local copied=0
  for src in "$TRIGGER_PROCESSING" "$RESULT_FILE"; do
    if [[ -f "$src" ]]; then
      # Unlock temporarily if immutable, copy, re-lock.
      local was_locked=false
      if lsattr -d "$src" 2>/dev/null | awk '{exit !($1 ~ /i/)}'; then
        was_locked=true
        chattr -i "$src" 2>/dev/null || true
      fi
      cp -f "$src" "${state_dir}/$(basename "$src")" 2>/dev/null && copied=$(( copied + 1 ))
      if [[ "$was_locked" == "true" ]]; then
        chattr +i "$src" 2>/dev/null || true
      fi
    fi
  done

  if (( copied > 0 )); then
    log "Update-state backup: ${copied} file(s) -> ${state_dir}"
  else
    log "Update-state backup: nothing to back up (no trigger/result files)"
    rmdir "$state_dir" 2>/dev/null || true
  fi
}

# ── Data restore ──────────────────────────────────────────────────────────────
restore_data_dir() {
  local backup_dir="$1"
  local data_backup="${backup_dir}/${DATA_BACKUP_PREFIX}"
  if [[ -d "$data_backup" ]]; then
    # DC-122: contents copy on the cp fallback — see backup_data_dir note.
    if rsync -a --delete "$data_backup/" "$DATA_SOURCE_DIR/" 2>/dev/null; then
      log "Data restored successfully (rsync --delete)"
    else
      log "rsync unavailable — cp fallback + manifest reconciliation"
      cp -a "$data_backup/." "$DATA_SOURCE_DIR/"
      # Mirror deletion semantics without rsync: remove live files that the
      # backup manifest says did not exist at backup time.
      local manifest="${backup_dir}/data.manifest"
      if [[ -f "$manifest" ]]; then
        local live_list deleted=0
        live_list="$( cd "$DATA_SOURCE_DIR" && find . -type f -printf '%P\n' | sort )"
        while IFS= read -r rel; do
          [[ -z "$rel" ]] && continue
          # defense against path traversal in a corrupted manifest
          [[ "$rel" == ..* || "$rel" == */..* || "$rel" == *../* ]] && continue
          rm -f "$DATA_SOURCE_DIR/$rel" && deleted=$(( deleted + 1 ))
        done < <(comm -13 "$manifest" <(printf '%s\n' "$live_list"))
        # prune directories that became empty
        find "$DATA_SOURCE_DIR" -mindepth 1 -type d -empty -delete 2>/dev/null || true
        log "Manifest reconciliation: removed ${deleted} post-backup file(s)"
      fi
      log "Data restored successfully (cp fallback)"
    fi
  else
    log "WARNING: No data backup found at ${data_backup} — data/ not restored"
  fi
}

# ── Frontend backup/restore (DC-122) ──────────────────────────────────────────
# The frontend is synced to the live web root BEFORE the API rebuild + health
# check. If the update then fails, the new frontend would pair with the rolled
# back API. Snapshot exactly what we touch so rollback can restore it.
FRONTEND_SUBDIRS="dist css vendor js assets"

# Snapshot = files + a manifest + metadata, so restore is EXACT:
#   frontend.meta : JSON with target dir + whether index.html/sw.js existed
#   frontend.manifest : every file that existed at snapshot time (relative)
# Restore deletes live files that the snapshot manifest does not know about
# (an update-introduced asset dies with the update) and recreates absent
# snapshot files.
backup_frontend_dir() {
  local backup_dir="$1" target="$2"
  # DC-122: reset any stale snapshot from a previous update with the same
  # version key before writing this one.
  rm -rf "${backup_dir}/frontend" "${backup_dir}/frontend.manifest" "${backup_dir}/frontend.meta"
  mkdir -p "${backup_dir}/frontend"
  [[ -f "$target/index.html" ]] && cp -f "$target/index.html" "${backup_dir}/frontend/index.html"
  [[ -f "$target/sw.js" ]] && cp -f "$target/sw.js" "${backup_dir}/frontend/sw.js"
  for sub in $FRONTEND_SUBDIRS; do
    [[ -d "$target/$sub" ]] && cp -rf "$target/$sub" "${backup_dir}/frontend/$sub"
  done
  ( cd "${backup_dir}/frontend" && find . -type f -printf '%P\n' | sort ) \
    > "${backup_dir}/frontend.manifest" 2>/dev/null || true
  # DC-122: use a validated target path for frontend.meta — json-escaped.
  local esc_target
  esc_target=$(json_escape "$target")
  printf '{"target":"%s","indexExisted":%s,"swExisted":%s}\n' \
    "$esc_target" \
    "$( [[ -f "$target/index.html" ]] && echo true || echo false )" \
    "$( [[ -f "$target/sw.js" ]] && echo true || echo false )" \
    > "${backup_dir}/frontend.meta"
  log "Frontend snapshot saved to ${backup_dir}/frontend ($(wc -l < "${backup_dir}/frontend.manifest" 2>/dev/null || echo 0) files)"
}

# ── Frontend target validation (DC-122) ───────────────────────────────────────
# A custom DASHCADDY_HOST_FRONTEND_DIR is supported, but ONLY if it validates:
# absolute, exists, is a directory, no '..' components, not the filesystem root.
# The SAME validator gates deploy and restore, so anything we deploy to is
# something we can roll back, and nothing else is ever touched.
validate_frontend_target() {
  local t="$1"
  [[ -n "$t" ]] || return 1
  [[ "$t" = /* ]] || return 1
  [[ "$t" != "/" ]] || return 1
  [[ "$t" != *..* ]] || return 1
  [[ -d "$t" ]] || return 1
  return 0
}

restore_frontend_dir() {
  local backup_dir="$1"
  local snap="${backup_dir}/frontend"
  [[ -d "$snap" ]] || { log "No frontend snapshot in this backup — skipping frontend restore"; return 0; }
  # DC-122: resolve the target FIRST, validate BEFORE any delete/copy this
  # function performs. A recorded custom target is honored — it was validated
  # by this same function's rules before deployment, and recorded in meta.
  local target=""
  if [[ -f "${backup_dir}/frontend.meta" ]]; then
    target=$(python3 -c "import json;print(json.load(open('${backup_dir}/frontend.meta')).get('target',''))" 2>/dev/null)
    # DC-122 (hardened after live-fire test caught it): if the recorded meta
    # target is present but INVALID, refuse outright. Never fall through to
    # discovery with an untrusted/invalid target — that could point the
    # restore at a directory the snapshot was never taken from.
    if [[ -n "$target" ]] && ! validate_frontend_target "$target"; then
      log "REFUSING frontend restore: recorded meta target is invalid: '${target}'"
      return 1
    fi
    validate_frontend_target "$target" || target=""
  fi
  if [[ -z "$target" ]]; then
    for candidate in /var/www/dashcaddy-status /etc/dashcaddy/sites/status; do
      [[ -d "$candidate" ]] && target="$candidate" && break
    done
  fi
  validate_frontend_target "$target" || { log "Refusing frontend restore on unexpected/unknown target: '${target:-}'"; return 1; }
  log "Restoring frontend to ${target} (exact state, updater-owned scope)..."
  # DC-122: structural exact restore of the owned subtrees:
  #   - snapshot has the subtree  → rm -rf live + copy snapshot (exact)
  #   - snapshot lacks the subtree but live has it → the update introduced it
  #     ⇒ remove it. (Fixes rollback leaving behind e.g. a css/ dir the update
  #     created when none existed before.)
  # index.html / sw.js are governed by frontend.meta existence flags.
  # Files outside FRONTEND_SUBDIRS are NEVER touched.
  for sub in $FRONTEND_SUBDIRS; do
    if [[ -d "$snap/$sub" ]]; then
      rm -rf "${target:?}/${sub:?}"
      cp -rf "$snap/$sub" "${target:?}/$sub"
    elif [[ -d "${target:?}/${sub}" ]]; then
      rm -rf "${target:?}/${sub:?}"
      log "Removed update-introduced subtree: ${sub}/"
    fi
  done
  # index.html / sw.js per frontend.meta existence flags (single pass).
  if grep -q '"indexExisted":true' "${backup_dir}/frontend.meta" 2>/dev/null && [[ -f "$snap/index.html" ]]; then
    cp -f "$snap/index.html" "${target:?}/index.html"
  elif grep -q '"indexExisted":false' "${backup_dir}/frontend.meta" 2>/dev/null; then
    rm -f "${target:?}/index.html"
  fi
  if grep -q '"swExisted":true' "${backup_dir}/frontend.meta" 2>/dev/null && [[ -f "$snap/sw.js" ]]; then
    cp -f "$snap/sw.js" "${target:?}/sw.js"
  elif grep -q '"swExisted":false' "${backup_dir}/frontend.meta" 2>/dev/null; then
    rm -f "${target:?}/sw.js"
  fi
  # Rollback removes the deployment stamp: the restored frontend is NOT a
  # self-updater deployment, so start.sh's source sync must resume authority.
  rm -f "${target:?}/update-stamp.json"
  # Empty dirs left behind by the removal pass.
  find "${target:?}/dist" "${target:?}/css" "${target:?}/js" "${target:?}/vendor" "${target:?}/assets" \
    -mindepth 1 -type d -empty -delete 2>/dev/null || true
  log "Frontend restored"
}

# ── Docker space reclaim (DC-122) ─────────────────────────────────────────────
# Every rebuild leaves the previous image dangling (~250MB); prune it so a
# churn of auto-updates doesn't seize disk. Safe to call at any exit point.
prune_docker() {
  docker image prune -f --filter "dangling=true" >/dev/null 2>&1 || true
  docker builder prune -f --keep-storage 500m >/dev/null 2>&1 || true
}

wait_for_health() {
  local port="${1:-3001}"
  local timeout="$HEALTH_TIMEOUT"
  local elapsed=0

  log "Waiting for health check (timeout: ${timeout}s)..."
  while (( elapsed < timeout )); do
    if curl -fsSL --max-time 3 "http://localhost:${port}/health" &>/dev/null; then
      log "Health check passed after ${elapsed}s"
      return 0
    fi
    sleep 2
    elapsed=$(( elapsed + 2 ))
  done

  log "Health check FAILED after ${timeout}s"
  return 1
}

# ── Shared rollback: restore code + data ────────────────────────────────────
rollback_restore() {
  local backup_dir="$1"
  log "Rolling back: restoring code files..."
  for item in "$backup_dir"/*.js "$backup_dir"/package.json "$backup_dir"/package-lock.json "$backup_dir"/Dockerfile "$backup_dir"/openapi.yaml "$backup_dir"/VERSION; do
    [[ -f "$item" ]] && cp -f "$item" "$api_source_dir/" 2>/dev/null || true
  done
  if [[ -d "$backup_dir/routes" ]]; then
    rm -rf "$api_source_dir/routes"
    cp -rf "$backup_dir/routes" "$api_source_dir/routes"
  fi
  if [[ -d "$backup_dir/src" ]]; then
    rm -rf "$api_source_dir/src"
    cp -rf "$backup_dir/src" "$api_source_dir/src"
  fi
  if [[ -d "$backup_dir/dns-providers" ]]; then
    rm -rf "$api_source_dir/dns-providers"
    cp -rf "$backup_dir/dns-providers" "$api_source_dir/dns-providers"
  fi
  restore_data_dir "$backup_dir"
  restore_frontend_dir "$backup_dir"
}

# ── Deployment mode ───────────────────────────────────────────────────────────
# Reproduce the SAME container the install created so an auto-update keeps every
# volume + env var (docker socket, Caddyfile, config/credentials, updates mount),
# not a minimal subset. Standard installs use docker-compose (compose file in the
# api source dir); the publish/dev host uses /opt/dashcaddy/start.sh; otherwise a
# bare docker run is the last resort. build_image() and restart_container() both
# honor the detected mode so build and run stay consistent.
deploy_mode() {
  if [[ -f "$api_source_dir/docker-compose.yml" || -f "$api_source_dir/compose.yml" || -f "$api_source_dir/compose.yaml" ]]; then
    echo compose
  elif [[ -x /opt/dashcaddy/start.sh ]]; then
    echo startsh
  else
    echo run
  fi
}

# Build the API image using whatever the install is wired for. Returns the build
# command's exit status so callers can detect failure.
build_image() {
  cd "$api_source_dir" || return 1
  case "$(deploy_mode)" in
    compose) docker compose build 2>&1 || docker-compose build 2>&1 ;;
    *)       docker build -t "$IMAGE_TAG" . 2>&1 ;;
  esac
}

# ── Shared container restart — recreate with the full, install-defined spec ───
# Recreates (rm + run / compose up) so new code AND new env vars take effect.
restart_container() {
  cd "$api_source_dir" 2>/dev/null || true
  case "$(deploy_mode)" in
    compose)
      log "Recreating container via docker compose (full compose spec)..."
      docker compose up -d 2>&1 || docker-compose up -d 2>&1
      ;;
    startsh)
      log "Recreating container via /opt/dashcaddy/start.sh (full container spec)..."
      bash /opt/dashcaddy/start.sh
      ;;
    *)
      log "Recreating container via minimal docker run (fallback)..."
      docker rm -f "$CONTAINER_NAME" 2>/dev/null || true
      docker run -d --restart unless-stopped --name "$CONTAINER_NAME" \
        --log-driver json-file --log-opt max-size=10m --log-opt max-file=3 \
        -p 127.0.0.1:3001:3001 \
        -v /opt/dashcaddy/dashcaddy-api/data:/app/data \
        -e SERVICES_FILE=/app/data/services.json \
        "$IMAGE_TAG"
      ;;
  esac
  log "Container recreated"
}

# ── Code-only restore (used after failed build when data hasn't changed yet) ──
code_restore() {
  local backup_dir="$1"
  log "Restoring code files..."
  for item in "$backup_dir"/*.js "$backup_dir"/package.json "$backup_dir"/package-lock.json "$backup_dir"/Dockerfile "$backup_dir"/openapi.yaml "$backup_dir"/VERSION; do
    [[ -f "$item" ]] && cp -f "$item" "$api_source_dir/" 2>/dev/null || true
  done
  if [[ -d "$backup_dir/routes" ]]; then
    rm -rf "$api_source_dir/routes"
    cp -rf "$backup_dir/routes" "$api_source_dir/routes"
  fi
  if [[ -d "$backup_dir/src" ]]; then
    rm -rf "$api_source_dir/src"
    cp -rf "$backup_dir/src" "$api_source_dir/src"
  fi
  if [[ -d "$backup_dir/dns-providers" ]]; then
    rm -rf "$api_source_dir/dns-providers"
    cp -rf "$backup_dir/dns-providers" "$api_source_dir/dns-providers"
  fi
}

main() {
  local start_time
  start_time=$(date +%s)

  # 1. Read trigger
  if [[ ! -f "$TRIGGER_FILE" ]]; then
    log "No trigger file found — nothing to do"
    exit 0
  fi

  # Parse trigger.json (uses python3 which is available on all supported distros)
  local action version from_version staging_dir api_source_dir commit channel
  local frontend_staging_dir frontend_target_dir
  action=$(python3 -c "import json; print(json.load(open('${TRIGGER_FILE}'))['action'])")
  version=$(python3 -c "import json; print(json.load(open('${TRIGGER_FILE}'))['version'])")
  from_version=$(python3 -c "import json; print(json.load(open('${TRIGGER_FILE}'))['fromVersion'])")
  staging_dir=$(python3 -c "import json; print(json.load(open('${TRIGGER_FILE}'))['stagingDir'])")
  api_source_dir=$(python3 -c "import json; print(json.load(open('${TRIGGER_FILE}'))['apiSourceDir'])")
  commit=$(python3 -c "import json; print(json.load(open('${TRIGGER_FILE}')).get('commit') or '')")
  frontend_staging_dir=$(python3 -c "import json; print(json.load(open('${TRIGGER_FILE}')).get('frontendStagingDir') or '')")
  frontend_target_dir=$(python3 -c "import json; print(json.load(open('${TRIGGER_FILE}')).get('frontendTargetDir') or '')")
  channel=$(python3 -c "import json; print(json.load(open('${TRIGGER_FILE}')).get('channel') or 'stable')")
  # Handle action=rollback (no new version to deploy)
  local to_version="${version}"

  log "=== ${action^^}: v${from_version} -> v${to_version} (channel: ${channel}) ==="
  log "Staging: ${staging_dir}"
  log "API source: ${api_source_dir}"

  # Consume the trigger immediately so we don't re-process on failure
  mv "$TRIGGER_FILE" "${TRIGGER_FILE}.processing"

  # Channel gate: refuse to apply prereleases unless explicitly opted-in.
  # Rollbacks always allowed (no new release channel involved).
  if [[ "${action}" != "rollback" ]] && ! channel_allowed "${channel}"; then
    write_result "false" "$to_version" "0" "Channel '${channel}' not allowed on this host"
    rm -f "${TRIGGER_FILE}.processing"
    exit 1
  fi

  # DC-122: validate the frontend target BEFORE any backup/deploy mutation.
  # An invalid trigger-provided target is a hard failure of the whole update
  # (the release cannot be applied faithfully), not a silent frontend skip.
  if [[ "$action" != "rollback" && -n "$frontend_target_dir" ]] && ! validate_frontend_target "$frontend_target_dir"; then
    log "ERROR: frontend target '${frontend_target_dir}' failed validation — refusing update"
    write_result "false" "$to_version" "$(( $(date +%s) - start_time ))" "Invalid frontendTargetDir in trigger: ${frontend_target_dir}"
    rm -f "${TRIGGER_FILE}.processing"
    exit 1
  fi

  # ── Handle rollback ────────────────────────────────────────────────────────
  if [[ "$action" == "rollback" ]]; then
    local backup_dir="${BACKUPS_DIR}/${version}"
    if [[ ! -d "$backup_dir" ]]; then
      log "ERROR: No backup found for version ${version}"
      write_result "false" "$version" "$(( $(date +%s) - start_time ))" "No backup found for version ${version}"
      rm -f "${TRIGGER_FILE}.processing"
      exit 1
    fi

    log "Performing rollback to v${version}..."
    rollback_restore "$backup_dir"

    # Rebuild old code — DC-122: a rollback that cannot rebuild or cannot
    # recover health is a FAILED rollback and must be reported as such.
    log "Rebuilding container..."
    local rebuild_ok=false health_ok=false
    if build_image; then rebuild_ok=true; fi

    restart_container
    if wait_for_health; then health_ok=true; fi

    if [[ "$rebuild_ok" == "true" && "$health_ok" == "true" ]]; then
      write_result "true" "$version" "$(( $(date +%s) - start_time ))"
      log "=== Rollback complete ==="
      prune_docker
      rm -f "${TRIGGER_FILE}.processing"
      exit 0
    else
      write_result "false" "$version" "$(( $(date +%s) - start_time ))" \
        "Rollback incomplete (rebuild_ok=$rebuild_ok health_ok=$health_ok) — INTERVENTION REQUIRED"
      log "=== Rollback FAILED (rebuild_ok=$rebuild_ok health_ok=$health_ok) — INTERVENTION REQUIRED ==="
      prune_docker
      rm -f "${TRIGGER_FILE}.processing"
      exit 1
    fi
  fi

  # ── Handle update ───────────────────────────────────────────────────────────
  if [[ ! -d "$staging_dir" ]]; then
    log "ERROR: Staging directory not found: ${staging_dir}"
    write_result "false" "$to_version" "$(( $(date +%s) - start_time ))" "Staging directory not found"
    rm -f "${TRIGGER_FILE}.processing"
    exit 1
  fi

  # 2. Backup current API code + data/
  local backup_dir="${BACKUPS_DIR}/${from_version}"
  mkdir -p "$backup_dir"
  log "Backing up current API files to ${backup_dir}"
  for item in "$api_source_dir"/*.js "$api_source_dir"/package.json "$api_source_dir"/package-lock.json "$api_source_dir"/Dockerfile "$api_source_dir"/openapi.yaml "$api_source_dir"/VERSION; do
    [[ -f "$item" ]] && cp -f "$item" "$backup_dir/" 2>/dev/null || true
  done
  [[ -d "$api_source_dir/routes" ]] && cp -rf "$api_source_dir/routes" "$backup_dir/"
  [[ -d "$api_source_dir/src" ]] && cp -rf "$api_source_dir/src" "$backup_dir/"
  [[ -d "$api_source_dir/dns-providers" ]] && cp -rf "$api_source_dir/dns-providers" "$backup_dir/"

  # Backup data/ directory (services.json, config.json, credentials, etc.)
  backup_data_dir "$backup_dir"

  # DC-122: snapshot the live frontend so a failed update can roll it back
  # (frontend is synced to the web root before build+health check).
  # Use the TRIGGER-provided target when set (custom installs) — the snapshot
  # must cover exactly the dir the deploy will touch — else discover.
  local fe_target="${frontend_target_dir}"
  if [[ -z "$fe_target" ]]; then
    for candidate in /var/www/dashcaddy-status /etc/dashcaddy/sites/status; do
      [[ -d "$candidate" ]] && fe_target="$candidate" && break
    done
  fi
  [[ -n "$fe_target" && -d "$fe_target" ]] && backup_frontend_dir "$backup_dir" "$fe_target"

  # Backup updater state (trigger.json.processing + result.json) so post-mortem
  # has a forensic trail tied to this exact version's backup.
  backup_update_state "$backup_dir"

  cleanup_old_backups

  # 3. Copy new files from staging to API source
  log "Deploying new API files..."
  for item in "$staging_dir"/*.js "$staging_dir"/package.json "$staging_dir"/package-lock.json "$staging_dir"/Dockerfile "$staging_dir"/openapi.yaml "$staging_dir"/VERSION; do
    [[ -f "$item" ]] && cp -f "$item" "$api_source_dir/" 2>/dev/null || true
  done
  # Safety: only replace routes/src if staging has the dir AND it's non-empty.
  # An empty or partial staging dir used to cause live routes/src to be wiped
  # when a prior update cycle was interrupted. We also handle locked files
  # (chattr +i) by temporarily unlocking before replace and re-locking after.
  deploy_tree() {
    local rel="$1"  # e.g. "routes"
    local src="${staging_dir}/${rel}"
    local dst="${api_source_dir}/${rel}"

    if [[ ! -d "$src" ]] || [[ -z "$(ls -A "$src" 2>/dev/null)" ]]; then
      [[ -d "$src" ]] && log "WARNING: staging ${rel}/ exists but is empty — leaving live ${rel}/ untouched"
      return 0
    fi

    # Collect any locked files (chattr +i) in the destination. lsattr's
    # first field is the attribute flags ("i" at position 5 = immutable);
    # the second field is the filename. We unlock before rm -rf and re-lock
    # after so the locked state survives the update.
    local locked_files=()
    if [[ -d "$dst" ]]; then
      while IFS= read -r lf; do
        [[ -n "$lf" ]] && locked_files+=("$lf")
      done < <(find "$dst" -type f \( -name "*.js" -o -name "*.json" -o -name "*.sh" \) -print0 2>/dev/null \
                | xargs -0 lsattr -a 2>/dev/null \
                | awk '$1 ~ /i/ { print $2 }')
    fi

    for lf in "${locked_files[@]:-}"; do
      [[ -n "$lf" ]] && chattr -i "$lf" 2>/dev/null || true
    done

    rm -rf "$dst"
    cp -rf "$src" "$dst"
    local file_count
    file_count=$(find "$dst" -type f 2>/dev/null | wc -l)
    log "${rel}/ deployed (${file_count} files)"

    for lf in "${locked_files[@]:-}"; do
      [[ -n "$lf" ]] && [[ -f "$lf" ]] && chattr +i "$lf" 2>/dev/null || true
    done
  }
  deploy_tree "routes"
  deploy_tree "src"
  deploy_tree "dns-providers"
  if [[ -n "$commit" ]]; then
    echo "$commit" > "$api_source_dir/VERSION"
  fi

  # 3a. Apply post-deploy patches — fix upstream bugs in released tarballs
  # (e.g. v1.14.4 has broken require paths and missing license-keygen module).
  # Runs AFTER staging copy, BEFORE docker build. Idempotent.
  local patch_script="/opt/dashcaddy/scripts/dashcaddy-post-deploy-patches.sh"
  if [[ -x "$patch_script" ]]; then
    log "Applying post-deploy patches..."
    if "$patch_script" "$api_source_dir"; then
      log "Post-deploy patches applied successfully"
    else
      log "WARNING: Post-deploy patches exited non-zero — continuing build anyway"
    fi
  else
    log "NOTE: $patch_script not found or not executable — skipping post-deploy patches"
  fi

  # 3b. Sync frontend
  if [[ -z "$frontend_staging_dir" ]]; then
    parent_staging=$(dirname "$staging_dir")
    [[ -d "$parent_staging/status" ]] && frontend_staging_dir="$parent_staging/status"
  fi
  if [[ -z "$frontend_target_dir" ]]; then
    for candidate in /var/www/dashcaddy-status /etc/dashcaddy/sites/status; do
      [[ -d "$candidate" ]] && frontend_target_dir="$candidate" && break
    done
  fi
  if [[ -n "$frontend_staging_dir" && -n "$frontend_target_dir" && -d "$frontend_staging_dir" ]]; then
    # DC-122: same validator gates deployment — if the trigger-provided custom
    # target doesn't validate, refuse before mutating anything.
    if ! validate_frontend_target "$frontend_target_dir"; then
      log "ERROR: frontend target '${frontend_target_dir}' failed validation — skipping frontend sync (update continues for API only)"
    else
    log "Syncing frontend: $frontend_staging_dir -> $frontend_target_dir"
    mkdir -p "$frontend_target_dir"
    [[ -f "$frontend_staging_dir/index.html" ]] && cp -f "$frontend_staging_dir/index.html" "$frontend_target_dir/index.html"
    [[ -f "$frontend_staging_dir/sw.js" ]] && cp -f "$frontend_staging_dir/sw.js" "$frontend_target_dir/sw.js"
    for sub in dist css vendor js; do
      if [[ -d "$frontend_staging_dir/$sub" ]]; then
        mkdir -p "$frontend_target_dir/$sub"
        cp -rf "$frontend_staging_dir/$sub/"* "$frontend_target_dir/$sub/" 2>/dev/null || true
      fi
    done
    if [[ -d "$frontend_staging_dir/assets" ]]; then
      mkdir -p "$frontend_target_dir/assets"
      cp -rf "$frontend_staging_dir/assets/"* "$frontend_target_dir/assets/" 2>/dev/null || true
    fi
    # DC-122: host-side deployment stamp — start.sh treats a stamped, newer
    # deployment as authoritative and skips its source-bundle sync (this is
    # the only writer that can reach the web root with real host paths).
    local esc_ver
    esc_ver=$(json_escape "$to_version")
    printf '{"version":"%s","at":"%s"}\n' "$esc_ver" "$(date -u +%Y-%m-%dT%H:%M:%SZ)" \
      > "$frontend_target_dir/update-stamp.json" 2>/dev/null || true
    fi # DC-122 close: validated frontend-target branch
  fi

  # 4. Rebuild container
  log "Rebuilding container..."
  local build_ok=false
  if build_image; then
    build_ok=true
  fi

  if [[ "$build_ok" != "true" ]]; then
    log "ERROR: Docker build failed — rolling back code + data"
    code_restore "$backup_dir"
    restore_frontend_dir "$backup_dir"
    local rb_rebuild_ok=false rb_health_ok=false
    if build_image; then rb_rebuild_ok=true; fi
    restart_container
    if wait_for_health; then rb_health_ok=true; fi
    prune_docker
    if [[ "$rb_rebuild_ok" == "true" && "$rb_health_ok" == "true" ]]; then
      write_result "false" "$to_version" "$(( $(date +%s) - start_time ))" "Docker build failed — rolled back cleanly"
    else
      write_result "false" "$to_version" "$(( $(date +%s) - start_time ))" \
        "Docker build failed AND rollback incomplete (rebuild_ok=$rb_rebuild_ok health_ok=$rb_health_ok) — INTERVENTION REQUIRED"
    fi
    rm -f "${TRIGGER_FILE}.processing"
    exit 1
  fi

  # 5. Restart container (recreate so new code + env vars take effect)
  restart_container

  # 6. Health check
  if wait_for_health; then
    local duration=$(( $(date +%s) - start_time ))
    log "=== Update successful: v${to_version} in ${duration}s ==="
    write_result "true" "$to_version" "$duration"
  else
    local duration=$(( $(date +%s) - start_time ))
    log "ERROR: Health check failed after update — rolling back code + data"
    rollback_restore "$backup_dir"
    local rb_rebuild_ok=false rb_health_ok=false
    if build_image; then rb_rebuild_ok=true; fi
    restart_container
    if wait_for_health; then rb_health_ok=true; fi
    prune_docker
    if [[ "$rb_rebuild_ok" == "true" && "$rb_health_ok" == "true" ]]; then
      write_result "false" "$to_version" "$duration" "Health check failed after update — rolled back cleanly"
    else
      write_result "false" "$to_version" "$duration" \
        "Health check failed after update AND rollback incomplete (rebuild_ok=$rb_rebuild_ok health_ok=$rb_health_ok) — INTERVENTION REQUIRED"
    fi
  fi

  # 7. Cleanup
  rm -f "${TRIGGER_FILE}.processing"
  rm -rf "${UPDATES_DIR}/staging" 2>/dev/null || true

  # 8. DC-122: reclaim docker space after every self-update rebuild (old
  # dangling image layers otherwise accumulate ~250MB per apply).
  prune_docker

  log "=== Update process complete ==="
}

main "$@"