#!/usr/bin/env bash
set -o errexit
set -o nounset
set -o pipefail

since_value="1 hour ago"
warning_count=20
critical_count=50
top_count=10

usage() {
  cat <<'USAGE'
Usage: check_failed_ssh_logins.sh [--since TEXT] [--warning COUNT] [--critical COUNT] [--top N] [--help]

Detect failed SSH login bursts from journal or readable authentication logs.
USAGE
}

is_number() {
  [[ "$1" =~ ^[0-9]+$ ]]
}

while (($# > 0)); do
  case "$1" in
    --since) [[ $# -ge 2 ]] || { printf 'CRITICAL: --since requires a value\n'; exit 2; }; since_value="$2"; shift 2 ;;
    --warning) [[ $# -ge 2 ]] || { printf 'CRITICAL: --warning requires a value\n'; exit 2; }; warning_count="$2"; shift 2 ;;
    --critical) [[ $# -ge 2 ]] || { printf 'CRITICAL: --critical requires a value\n'; exit 2; }; critical_count="$2"; shift 2 ;;
    --top) [[ $# -ge 2 ]] || { printf 'CRITICAL: --top requires a value\n'; exit 2; }; top_count="$2"; shift 2 ;;
    --help|-h) usage; exit 0 ;;
    *) printf 'CRITICAL: unknown option: %s\n' "$1"; usage; exit 2 ;;
  esac
done

for value in "$warning_count" "$critical_count" "$top_count"; do
  if ! is_number "$value"; then
    printf 'CRITICAL: numeric option expected, got: %s\n' "$value"
    exit 2
  fi
done
if ((warning_count >= critical_count)); then
  printf 'CRITICAL: --warning must be lower than --critical\n'
  exit 2
fi

tmp_log="$(mktemp)"
trap 'rm -f "$tmp_log"' EXIT
log_source="journalctl"

if command -v journalctl >/dev/null 2>&1; then
  journalctl --since "$since_value" --no-pager 2>/dev/null \
    | grep -Ei 'sshd.*(Failed password|Invalid user|authentication failure)|authentication failure.*sshd' > "$tmp_log" || true
else
  log_source="log file fallback"
fi

if [[ ! -s "$tmp_log" ]]; then
  for log_file in /var/log/auth.log /var/log/secure /var/log/messages; do
    if [[ -r "$log_file" ]]; then
      grep -Ei 'sshd.*(Failed password|Invalid user|authentication failure)|authentication failure.*sshd' "$log_file" >> "$tmp_log" || true
      log_source="$log_file"
    fi
  done
fi

attempts="$(wc -l < "$tmp_log" | awk '{print $1}')"

status="OK"
exit_code=0
if ((attempts >= critical_count)); then
  status="CRITICAL"
  exit_code=3
elif ((attempts >= warning_count)); then
  status="WARNING"
  exit_code=1
fi

printf '%s: Found %s failed SSH login attempt(s) for requested window\n\n' "$status" "$attempts"

printf 'Top source IPs:\n'
if [[ -s "$tmp_log" ]]; then
  grep -Eo 'from ([0-9]{1,3}\.){3}[0-9]{1,3}|rhost=([0-9]{1,3}\.){3}[0-9]{1,3}' "$tmp_log" \
    | sed -E 's/^(from|rhost=) //' \
    | sort | uniq -c | sort -rn | head -n "$top_count" || true
else
  printf 'OK: no failed SSH attempts found in available sources\n'
fi
printf '\n'

printf 'Top attempted users:\n'
if [[ -s "$tmp_log" ]]; then
  sed -nE 's/.*Invalid user ([^ ]+).*/\1/p; s/.*Failed password for invalid user ([^ ]+).*/\1/p; s/.*Failed password for ([^ ]+).*/\1/p; s/.*user=([^ ]+).*/\1/p' "$tmp_log" \
    | sort | uniq -c | sort -rn | head -n "$top_count" || true
else
  printf 'OK: no attempted users extracted\n'
fi
printf '\n'

printf 'Sample recent lines:\n'
if [[ -s "$tmp_log" ]]; then
  tail -n "$top_count" "$tmp_log"
else
  printf 'OK: no sample lines available\n'
fi
printf '\n\n'

printf 'Evidence:\n'
printf 'Thresholds: warning=%s critical=%s since="%s"\n' "$warning_count" "$critical_count" "$since_value"
printf 'Log source: %s\n' "$log_source"
if [[ "$log_source" != "journalctl" ]]; then
  printf 'WARNING: log file fallback may include entries outside the requested --since window\n'
fi
if [[ "${EUID:-$(id -u 2>/dev/null || printf '1')}" != "0" ]]; then
  printf 'WARNING: running without root; authentication log visibility may be limited\n'
fi
printf '\n'

printf 'Recommended next steps:\n'
printf -- '- Verify source IPs against expected scanners, admins, or automation\n'
printf -- '- Check firewall, fail2ban, or security tooling state\n'
printf -- '- Confirm whether the attempts are expected for this host\n'
printf -- '- Review successful logins too, not only failures\n'
printf -- '- Attach this output to incident ticket\n'

exit "$exit_code"