#!/usr/bin/env bash
set -u  # no unset vars

########################
# Configuration
########################

SECONDS_WAIT=60
DESTINATION_DIR="/mnt/unprotected/tmp/VM/backup/corellia"
VMHOME="/vm/ssd0/"   # base dir for VM data; adapt as needed
DAYS_TO_KEEP_VM_BACKUP=60
MAIL_TO="vplesnila@gmail.com"
MAIL_FROM="plesnila.valeriu@orange.fr"
MAIL_SUBJECT_OK="VM backup success"
MAIL_SUBJECT_FAIL="VM backup failure"
LOG_DIR="/root/tmp"
LOGFILE="${LOG_DIR}/vm_backup_$(date +%F_%H%M%S).log"

########################
# Logging setup
########################

mkdir -p "$LOG_DIR" "$DESTINATION_DIR"

# Log everything to file and stdout
exec > >(tee -a "$LOGFILE") 2>&1

echo "===== VM backup started at $(date) ====="

FAILED=0

fail() {
  echo "ERROR: $*" >&2
  FAILED=1
}

send_mail() {
  local subject="$1"
  {
    echo "From: ${MAIL_FROM}"
    echo "To: ${MAIL_TO}"
    echo "Subject: ${subject}"
    echo
    cat "$LOGFILE"
  } | msmtp "$MAIL_TO"
}

########################
# Helper functions
########################

list_all_vms() {
  virsh list --all --name | sed '/^$/d'
}

list_running_vms() {
  virsh list --name | sed '/^$/d'
}

wait_and_check_offline() {
  local wait_seconds="$1"
  echo "Waiting ${wait_seconds}s..."
  sleep "$wait_seconds"

  local still_running
  still_running=$(list_running_vms || true)

  if [[ -n "$still_running" ]]; then
    echo "Still running VMs:"
    echo "$still_running"
    return 1
  fi

  echo "All VMs are offline."
  return 0
}

backup_vm() {
  local vm="$1"
  local date_str
  date_str="$(date +%F)"
  local backup_file="${DESTINATION_DIR}/${vm}_${date_str}.tar.gz"
  local xml_file="${DESTINATION_DIR}/${vm}_${date_str}.xml"

  echo "----- Backing up VM: ${vm} -----"

  # 1. Export XML definition
  echo "Exporting XML definition to ${xml_file}"
  if ! virsh dumpxml "$vm" > "$xml_file"; then
    fail "Failed to dump XML for VM '$vm'"
    return 1
  fi

  # 2. Backup VM storage
  if [[ ! -d "$VMHOME" ]]; then
    fail "VMHOME directory '$VMHOME' does not exist for VM '$vm'"
    return 1
  fi

  cd "$VMHOME" || { fail "Cannot cd to $VMHOME for VM '$vm'"; return 1; }

  echo "Creating compressed archive: $backup_file"
  if ! tar -cvf - ${vm} | pigz > "$backup_file"; then
    fail "Backup failed for VM '$vm'"
    return 1
  fi

  # 3. Cleanup old backups
  echo "Cleaning old backups for VM '$vm' older than ${DAYS_TO_KEEP_VM_BACKUP} days"
  if ! find "$DESTINATION_DIR" -maxdepth 1 -type f \
        \( -name "${vm}_*.tar.gz" -o -name "${vm}_*.xml" \) \
        -mtime +"$DAYS_TO_KEEP_VM_BACKUP" -print -delete; then
    fail "Failed to clean old backups for VM '$vm'"
    return 1
  fi

  echo "Backup completed for VM '$vm'"
  return 0
}

start_all_vms() {
  echo "===== Starting ALL defined VMs ====="

  local all_vms
  all_vms=$(virsh list --all --name | sed '/^$/d')

  if [[ -z "$all_vms" ]]; then
    echo "No VMs defined on this host; nothing to start."
    return 0
  fi

  while read -r vm; do
    [[ -z "$vm" ]] && continue
    echo "Starting VM: $vm"
    if ! virsh start "$vm"; then
      fail "Failed to start VM '$vm'"
    fi
  done <<< "$all_vms"

  echo "===== VM startup phase completed ====="
}

########################
# Main logic
########################

ALL_VMS=$(list_all_vms)
if [[ -z "$ALL_VMS" ]]; then
  echo "No VMs defined on this host."
fi

echo "All VMs:"
echo "$ALL_VMS"

RUNNING_VMS=$(list_running_vms)
echo "Running VMs:"
echo "${RUNNING_VMS:-<none>}"

# Graceful shutdown
if [[ -n "$RUNNING_VMS" ]]; then
  echo "Shutting down VMs gracefully..."
  while read -r vm; do
    [[ -z "$vm" ]] && continue
    echo "virsh shutdown $vm"
    if ! virsh shutdown "$vm"; then
      fail "Graceful shutdown failed for VM '$vm'"
    fi
  done <<< "$RUNNING_VMS"
else
  echo "No running VMs to shut down gracefully."
fi

# Wait and check
if ! wait_and_check_offline "$SECONDS_WAIT"; then
  echo "Some VMs still running after graceful shutdown; forcing shutdown..."
  STILL_RUNNING=$(list_running_vms)
  while read -r vm; do
    [[ -z "$vm" ]] && continue
    echo "virsh destroy $vm"
    if ! virsh destroy "$vm"; then
      fail "Force shutdown failed for VM '$vm'"
    fi
  done <<< "$STILL_RUNNING"
fi

# If any failure so far, send failure mail and exit
if [[ "$FAILED" -ne 0 ]]; then
  echo "Failures detected before backup phase."
  send_mail "$MAIL_SUBJECT_FAIL"
  echo "===== VM backup finished with FAILURE at $(date) ====="
  exit 1
fi

# Backup each VM
for vm in $ALL_VMS; do
  [[ -z "$vm" ]] && continue
  backup_vm "$vm" || FAILED=1
done

echo "Starting all VMs..."
start_all_vms

# Final status and mail
if [[ "$FAILED" -ne 0 ]]; then
  echo "One or more VM backups failed."
  send_mail "$MAIL_SUBJECT_FAIL"
  echo "===== VM backup finished with FAILURE at $(date) ====="
  exit 1
else
  echo "All VM backups completed successfully."
  send_mail "$MAIL_SUBJECT_OK"
  echo "===== VM backup finished with SUCCESS at $(date) ====="
  exit 0
fi