#!/bin/bash
cvmfs_test_name="Watchdog started by a hotpatched client under a pre-2.14 loader"

# After a hotpatch from a pre-2.14 loader the new FUSE module spawns its own
# watchdog.  The supervisee keeps real uid 0 there, so a watchdog that drops
# its capabilities can neither ptrace nor kill nor unmount it on a crash.

cleanup() {
  echo "running cleanup()"
  sudo umount -l /cvmfs/$MIGTEST_003_REPO > /dev/null 2>&1 || true
}

MIGTEST_003_REPO=lhcb.cern.ch

cvmfs_run_test() {
  local logfile=$1
  local script_location=$2

  . ${script_location}/../../common/migration_tests/common.sh

  local repo=$MIGTEST_003_REPO
  trap cleanup EXIT HUP INT TERM || return $?

  current_version=$(package_version $(find_new_package cvmfs))
  echo "Version of testee:        $current_version"
  # Must ship a loader older than version 6
  previous_version="2.13.3"
  echo "Version to be downloaded: $previous_version"
  upstream_packages=""
  for name in cvmfs cvmfs-libs cvmfs-fuse3; do
    upstream_package_url=$(guess_package_url $name ${previous_version})
    echo "Download URL:             $upstream_package_url"
    wget --no-check-certificate --quiet $upstream_package_url || return 2
    upstream_packages="$upstream_packages $(basename $upstream_package_url)"
  done

  for name in cvmfs cvmfs-libs cvmfs-fuse3; do
    if is_installed $name; then
      echo "uninstalling $name $(installed_package_version $name)"
      uninstall_package $name || return 3
    fi
  done

  echo "installing CernVM-FS $previous_version"
  install_packages $upstream_packages || return 4
  sudo cvmfs_config setup    || return 5
  sudo cvmfs_config chksetup || return 5

  cvmfs_mount $repo || return 6
  ls /cvmfs/$repo || return 6
  local old_pid=$(sudo cvmfs_talk -i $repo pid)

  echo "updating CernVM-FS package to version $current_version (hotpatch)"
  install_new_packages || return 7
  # cvmfs-reload.service is Type=simple, so the package install returns early
  if [ -d /run/systemd/system ]; then
    local i
    for i in $(seq 120); do
      systemctl is-active -q cvmfs-reload.service || break
      sleep 1
    done
    systemctl is-active -q cvmfs-reload.service && return 7
  fi
  ls /cvmfs/$repo || return 8

  local pid=$(sudo cvmfs_talk -i $repo pid)
  [ "$pid" = "$old_pid" ] || { echo "client was remounted, not hotpatched"; return 9; }
  sudo cvmfs_talk -i $repo version
  sudo cvmfs_talk -i $repo version | grep -q "$current_version" || return 10

  local watchdog_pid=$(sudo cvmfs_talk -i $repo pid watchdog)
  echo "watchdog PID: $watchdog_pid"
  [ -n "$watchdog_pid" ] && [ "$watchdog_pid" != "0" ] || return 10

  local stacktrace_path="$(get_cvmfs_cachedir $repo)/stacktrace.${repo}"
  sudo rm -f $stacktrace_path

  # Avoid the main thread: the old module frees its sigaltstack on reload
  # without disabling it, so the crash handler cannot run there
  local tid=$(sudo ls /proc/$pid/task | grep -vx $pid | sort -n | tail -n 1)
  echo "crashing the hotpatched client (PID $pid, TID $tid)"
  [ -n "$tid" ] || return 11
  sudo python3 -c \
    'import ctypes, sys; sys.exit(ctypes.CDLL(None, use_errno=True).tgkill(int(sys.argv[1]), int(sys.argv[2]), 11))' \
    $pid $tid || return 11
  # The watchdog writes the trace after killing the client, then unmounts
  for i in $(seq 60); do
    sudo kill -0 $watchdog_pid 2>/dev/null || break
    sleep 1
  done

  sudo [ -f $stacktrace_path ] || return 12
  sudo cat $stacktrace_path
  if sudo grep -q -e 'failed to gain ptrace capability' \
                  -e 'Failed to kill cvmfs client' $stacktrace_path; then
    return 13
  fi
  sudo grep -q '^#[0-9]' $stacktrace_path || return 14
  sudo kill -0 $pid 2>/dev/null && return 15

  # The watchdog unmounts the crashed mount point on its own
  if mount | grep -q " /cvmfs/$repo "; then
    return 16
  fi

  return 0
}
