#!/bin/bash
# This script checks for CageFS issues where sites return 503 until CageFS is remounted
# Refactored to check for mount errors in the system log and su errors on cPanel user accounts
# Refactored to add logging and cagefs restart

log_file="/var/log/check_503.log"

log_message() {
  local message="$1"
  echo "$(date '+%b %d %T') $message" >> "$log_file"
}

if which cagefsctl &> /dev/null; then
  msg_ok="0 check_503 - OK"
  msg_crit="2 check_503 - Possible 503 errors situation, resetting CageFS."
  domain_count=0
  error_count=0
  su_error_count=0

  # Function to check mounts error in last 15 minutes.
  check_mount_error() {
    last_mount_error=$(grep "reached the limit on mounts" /var/log/messages | tail -1)
    if [ -n "$last_mount_error" ]; then
      mount_error_timestamp=$(echo "$last_mount_error" | awk '{print $1, $2, $3}')
      epoch_timestamp=$(date -d "$mount_error_timestamp" +%s)
      current_epoch=$(date +%s)
      time_diff=$((current_epoch - epoch_timestamp))
      if [ "$time_diff" -le 900 ]; then
        log_message "Mount errors found in system log within last 15 minutes"
        return 0
      fi
    fi
    return 1
  }

  # Function to check su on each cPanel user account.
  check_su_error() {
    users_list=$(whmapi1 list_users --output=jsonpretty | jq -r '.data.users[]' | grep -v root)
    if [ $? -ne 0 ] || [ -z "$users_list" ]; then
      return 1
    fi
    total_num_of_users=$(echo "$users_list" | wc -l)
    for cpuser in $users_list; do
      if ! su - "$cpuser" -s /bin/sh -c "exit" &>/dev/null; then
        su_error_count=$((su_error_count + 1))
      fi
    done

    if [ "$su_error_count" -eq "$total_num_of_users" ]; then
      log_message "su attempt failed for $su_error_count/$total_num_of_users users."
      return 0
    else
      return 1
    fi
  }

  if check_mount_error || check_su_error; then

    # Loop through each domain and check if its giving 503 error
    domain=$(whmapi1 --output=jsonpretty get_domain_info | jq -r '.data.domains[].domain')
    for cur_domain in $domain; do
      domain_count=$((domain_count + 1))
      ip=$(whmapi1 --output=jsonpretty get_domain_info | jq -r --arg domain "$cur_domain" '.data.domains[] | select(.domain == $domain) | .ipv4')
      status_code=$(curl --connect-timeout 2 -sL -o /dev/null -w "%{http_code}" "$cur_domain" --connect-to ""$cur_domain":80:$ip")

      if [ "$status_code" -eq 301 ]; then
        status_code=$(curl --connect-timeout 2 -sL -o /dev/null -w "%{http_code}" "$cur_domain" --connect-to ""$cur_domain":443:$ip")
      fi

      if [ "$status_code" -eq 503 ]; then
        error_count=$((error_count + 1))
      fi
    done

    if [ "$error_count" -gt 0 ]; then
      log_message "$cur_domain showing 503 error."
      log_message "$msg_crit"
      echo "$msg_crit"
      exit 2
      systemctl restart cagefs
      /usr/sbin/cagefsctl --disable-all
      /usr/sbin/cagefsctl --enable-all
    else
      echo "$msg_ok"
      exit 0
    fi
  else
    echo "$msg_ok"
    exit 0
  fi
fi