Keep netboot entries in efiboot

when defining Zero-OS as first boot, we can not get the node back into rescue mode as that is the way Hetzner functions: tne netboot get an ipxe that doe a local reload or, in case of rescue starts a full netboot
vor 1 Woche · d0fd7fcc7b
--- a/zosinstalluefipxeondisk.sh
+++ b/zosinstalluefipxeondisk.sh
@@ -5,264 +5,302 @@ set -e

 # Function to log messages with timestamps
 log() {
    echo "$(date '+%F %T') - $@"
  echo "$(date '+%F %T') - $@"
 }

 # Function to display usage information
 usage() {
    echo "Usage: $0 [-d DEVICE] [-f FARM] [-e ENVIRONMENT]"
    echo "  -d, --device DEVICE      Specify the block device to use (e.g., sda, nvme0n1)."
    echo "  -f, --farm FARM          Specify the FARM ID."
    echo "  -e, --environment ENV    Specify the environment (dev, qa, test, prod)."
    echo "  -h, --help               Display this help message."
    exit 1
  echo "Usage: $0 [-d DEVICE] [-f FARM] [-e ENVIRONMENT]"
  echo "  -d, --device DEVICE      Specify the block device to use (e.g., sda, nvme0n1)."
  echo "  -f, --farm FARM          Specify the FARM ID."
  echo "  -e, --environment ENV    Specify the environment (dev, qa, test, prod)."
  echo "  -h, --help               Display this help message."
  exit 1
 }

 # Function to find a suitable block device
 find_blockdev() {
    # Minimum size in bytes, e.g., 10 GiB
    MIN_SIZE=$((10 * 1024 * 1024 * 1024))

    # Search for NVMe devices first
    log "Searching for NVMe devices..."
    for DEV in /dev/nvme*n1; do
        if [ -b "$DEV" ]; then
            SIZE_BYTES=$(blockdev --getsize64 "$DEV")
            if [ "$SIZE_BYTES" -ge "$MIN_SIZE" ]; then
                log "Found NVMe device: $DEV ($(numfmt --to=iec $SIZE_BYTES))"
                BLOCKDEV="${DEV##*/}"
                return 0
            else
                log "$DEV is smaller than $(numfmt --to=iec $MIN_SIZE). Skipping."
            fi
        fi
    done

    # If no NVMe device found, search for SSD devices
    log "No suitable NVMe device found. Searching for SSD devices..."
    for DEV_PATH in /sys/block/*; do
        DEV_NAME=$(basename "$DEV_PATH")
        # Exclude unwanted devices
        case "$DEV_NAME" in
            loop*|ram*|sr*|fd*|md*|dm-*)
                continue
                ;;
        esac
        DEVICE="/dev/$DEV_NAME"
        if [ -b "$DEVICE" ] && [ -e "/sys/block/$DEV_NAME/queue/rotational" ] && [ "$(cat /sys/block/$DEV_NAME/queue/rotational)" == "0" ]; then
            SIZE_BYTES=$(blockdev --getsize64 "$DEVICE")
            if [ "$SIZE_BYTES" -ge "$MIN_SIZE" ]; then
                log "Found SSD device: $DEVICE ($(numfmt --to=iec $SIZE_BYTES))"
                BLOCKDEV="$DEV_NAME"
                return 0
            else
                log "$DEVICE is smaller than $(numfmt --to=iec $MIN_SIZE). Skipping."
            fi
        fi
    done
  # Minimum size in bytes, e.g., 10 GiB
  MIN_SIZE=$((10 * 1024 * 1024 * 1024))

  # Search for NVMe devices first
  log "Searching for NVMe devices..."
  for DEV in /dev/nvme*n1; do
    if [ -b "$DEV" ]; then
      SIZE_BYTES=$(blockdev --getsize64 "$DEV")
      if [ "$SIZE_BYTES" -ge "$MIN_SIZE" ]; then
        log "Found NVMe device: $DEV ($(numfmt --to=iec $SIZE_BYTES))"
        BLOCKDEV="${DEV##*/}"
        return 0
      else
        log "$DEV is smaller than $(numfmt --to=iec $MIN_SIZE). Skipping."
      fi
    fi
  done

  # If no NVMe device found, search for SSD devices
  log "No suitable NVMe device found. Searching for SSD devices..."
  for DEV_PATH in /sys/block/*; do
    DEV_NAME=$(basename "$DEV_PATH")
    # Exclude unwanted devices
    case "$DEV_NAME" in
    loop* | ram* | sr* | fd* | md* | dm-*)
      continue
      ;;
    esac
    DEVICE="/dev/$DEV_NAME"
    if [ -b "$DEVICE" ] && [ -e "/sys/block/$DEV_NAME/queue/rotational" ] && [ "$(cat /sys/block/$DEV_NAME/queue/rotational)" == "0" ]; then
      SIZE_BYTES=$(blockdev --getsize64 "$DEVICE")
      if [ "$SIZE_BYTES" -ge "$MIN_SIZE" ]; then
        log "Found SSD device: $DEVICE ($(numfmt --to=iec $SIZE_BYTES))"
        BLOCKDEV="$DEV_NAME"
        return 0
      else
        log "$DEVICE is smaller than $(numfmt --to=iec $MIN_SIZE). Skipping."
      fi
    fi
  done

    log "No suitable block device found."
    return 1
  log "No suitable block device found."
  return 1
 }

 # Function to manage UEFI boot entries
 manage_boot_entries() {
    log "Managing UEFI boot entries..."
  log "Managing UEFI boot entries..."

    # Check if efibootmgr is available
    if ! command -v efibootmgr &> /dev/null; then
        log "Error: 'efibootmgr' is not installed. Please install it and rerun the script."
        exit 1
  # Check if efibootmgr is available
  if ! command -v efibootmgr &>/dev/null; then
    log "Error: 'efibootmgr' is not installed. Please install it and rerun the script."
    exit 1
  fi

  # Get the list of current boot entries
  log "Retrieving current boot entries..."
  BOOT_ENTRIES_INFO=$(efibootmgr -v | grep -E "^Boot[0-9A-F]{4}")

  # Arrays to hold boot entry numbers
  NETBOOT_ENTRIES=()
  ENTRIES_TO_DELETE=()

  # Identify netboot entries and entries to delete
  while read -r LINE; do
    ENTRY_NUM=$(echo "$LINE" | awk '{print $1}' | sed 's/Boot//;s/\*//')
    ENTRY_DESC=$(echo "$LINE" | cut -d' ' -f2-)
    ENTRY_PATH=$(echo "$LINE" | awk -F'File' '{print $2}')

    # Check if the entry is a netboot entry (e.g., contains "PXE", "Network", or "IPv4"/"IPv6")
    if echo "$ENTRY_DESC" | grep -qiE "(PXE|Network|IPV4|IPV6)"; then
      NETBOOT_ENTRIES+=("$ENTRY_NUM")
      log "Identified netboot entry: Boot$ENTRY_NUM - $ENTRY_DESC"
    else
      ENTRIES_TO_DELETE+=("$ENTRY_NUM")
      log "Marking entry for deletion: Boot$ENTRY_NUM - $ENTRY_DESC"
    fi

    # Get the list of current boot entries
    CURRENT_BOOT_ENTRIES=$(efibootmgr | grep "BootOrder")
    log "Current BootOrder: $CURRENT_BOOT_ENTRIES"

    # Delete all existing boot entries except the EFI Shell (if present)
    log "Deleting existing boot entries..."
    BOOT_ENTRIES=$(efibootmgr | grep -E "^Boot[0-9A-F]{4}" | awk '{print $1}' | sed 's/\*//')
    for ENTRY in $BOOT_ENTRIES; do
        efibootmgr -b ${ENTRY#Boot} -B
        log "Deleted boot entry $ENTRY"
  done <<<"$BOOT_ENTRIES_INFO"

  # Delete non-netboot entries
  if [ ${#ENTRIES_TO_DELETE[@]} -gt 0 ]; then
    log "Deleting non-netboot boot entries..."
    for ENTRY_NUM in "${ENTRIES_TO_DELETE[@]}"; do
      efibootmgr -b "$ENTRY_NUM" -B
      log "Deleted boot entry Boot$ENTRY_NUM"
    done
  else
    log "No non-netboot entries to delete."
  fi

  # Create a new boot entry pointing to BOOTX64.EFI on the first partition
  log "Creating new boot entry for /dev/${PARTPREFIX}1..."
  # Adjust the partition number if necessary
  ESP_PART_NUM=1
  efibootmgr -c -d /dev/${BLOCKDEV} -p $ESP_PART_NUM -L "Zero-OS" -l "\\EFI\\BOOT\\BOOTX64.EFI" || {
    log "Error: Failed to create new boot entry."
    exit 1
  }

    # Create a new boot entry pointing to BOOTX64.EFI on the first partition
    log "Creating new boot entry for /dev/${PARTPREFIX}1..."
    # Adjust the partition number if necessary
    ESP_PART_NUM=1
    efibootmgr -c -d /dev/${BLOCKDEV} -p $ESP_PART_NUM -L "Zero-OS" -l "\\EFI\\BOOT\\BOOTX64.EFI" || {
        log "Error: Failed to create new boot entry."
        exit 1
    }

    # Set the boot order to include only the new boot entry
    NEW_BOOT_NUM=$(efibootmgr | grep "Zero-OS" | awk '{print $1}' | sed 's/\*//;s/Boot//')
    if [ -z "$NEW_BOOT_NUM" ]; then
        log "Error: Failed to retrieve new boot entry number."
        exit 1
    fi
    log "Setting BootOrder to only include Boot$NEW_BOOT_NUM..."
    efibootmgr -o $NEW_BOOT_NUM || {
        log "Error: Failed to set new BootOrder."
        exit 1
    }
  # Retrieve the new boot entry number
  NEW_BOOT_NUM=$(efibootmgr | grep "Zero-OS" | awk '{print $1}' | sed 's/\*//;s/Boot//')
  if [ -z "$NEW_BOOT_NUM" ]; then
    log "Error: Failed to retrieve new boot entry number."
    exit 1
  fi

  # Set the new boot order: netboot entries first, followed by the new boot entry
  log "Setting new BootOrder with netboot entries first..."
  BOOT_ORDER=""
  for ENTRY_NUM in "${NETBOOT_ENTRIES[@]}"; do
    BOOT_ORDER+="$ENTRY_NUM,"
  done
  BOOT_ORDER+="$NEW_BOOT_NUM"

  efibootmgr -o "$BOOT_ORDER" || {
    log "Error: Failed to set new BootOrder."
    exit 1
  }

    log "New BootOrder set successfully."
  log "New BootOrder set successfully: $BOOT_ORDER"
 }

 # Main script execution
 main() {
    # Default values
    BLOCKDEV_SPECIFIED=""
    FARM="1234"
    ENVIRONMENT="prod"

    # Parse command-line arguments
    while [[ $# -gt 0 ]]; do
        case "$1" in
            -d|--device)
                shift
                if [ -z "$1" ]; then
                    log "Error: '--device' requires a non-empty option argument."
                    usage
                fi
                BLOCKDEV_SPECIFIED="$1"
                shift
                ;;
            -f|--farm)
                shift
                if [ -z "$1" ]; then
                    log "Error: '--farm' requires a non-empty option argument."
                    usage
                fi
                FARM="$1"
                shift
                ;;
            -e|--environment)
                shift
                if [ -z "$1" ]; then
                    log "Error: '--environment' requires a non-empty option argument."
                    usage
                fi
                ENVIRONMENT="$1"
                shift
                ;;
            -h|--help)
                usage
                ;;
            *)
                log "Error: Unknown option: $1"
                usage
                ;;
        esac
    done

    # Validate environment
    case "$ENVIRONMENT" in
        dev|qa|test|prod)
            log "Using environment: $ENVIRONMENT"
            ;;
        *)
            log "Error: Invalid environment '$ENVIRONMENT'. Valid options are 'dev', 'qa', 'test', 'prod'."
            exit 1
            ;;
  # Default values
  BLOCKDEV_SPECIFIED=""
  FARM="1234"
  ENVIRONMENT="prod"

  # Parse command-line arguments
  while [[ $# -gt 0 ]]; do
    case "$1" in
    -d | --device)
      shift
      if [ -z "$1" ]; then
        log "Error: '--device' requires a non-empty option argument."
        usage
      fi
      BLOCKDEV_SPECIFIED="$1"
      shift
      ;;
    -f | --farm)
      shift
      if [ -z "$1" ]; then
        log "Error: '--farm' requires a non-empty option argument."
        usage
      fi
      FARM="$1"
      shift
      ;;
    -e | --environment)
      shift
      if [ -z "$1" ]; then
        log "Error: '--environment' requires a non-empty option argument."
        usage
      fi
      ENVIRONMENT="$1"
      shift
      ;;
    -h | --help)
      usage
      ;;
    *)
      log "Error: Unknown option: $1"
      usage
      ;;
    esac
  done

  # Validate environment
  case "$ENVIRONMENT" in
  dev | qa | test | prod)
    log "Using environment: $ENVIRONMENT"
    ;;
  *)
    log "Error: Invalid environment '$ENVIRONMENT'. Valid options are 'dev', 'qa', 'test', 'prod'."
    exit 1
    ;;
  esac

    if [ -n "$BLOCKDEV_SPECIFIED" ]; then
        # Use the specified block device
        if [[ "$BLOCKDEV_SPECIFIED" != /dev/* ]]; then
            BLOCKDEV_SPECIFIED="/dev/$BLOCKDEV_SPECIFIED"
        fi
        if [ ! -b "$BLOCKDEV_SPECIFIED" ]; then
            log "Error: Specified device $BLOCKDEV_SPECIFIED does not exist or is not a block device."
            exit 1
        fi
        BLOCKDEV="${BLOCKDEV_SPECIFIED##*/}"
        log "Using specified block device: $BLOCKDEV_SPECIFIED"
    else
        # Find a suitable block device
        if ! find_blockdev; then
            log "Error: No suitable block device found."
            exit 1
        fi
        log "Selected block device: /dev/${BLOCKDEV}"
  if [ -n "$BLOCKDEV_SPECIFIED" ]; then
    # Use the specified block device
    if [[ "$BLOCKDEV_SPECIFIED" != /dev/* ]]; then
      BLOCKDEV_SPECIFIED="/dev/$BLOCKDEV_SPECIFIED"
    fi

    # Confirm with the user before proceeding
    read -p "This will erase all data on /dev/${BLOCKDEV}. Are you sure you want to proceed? (yes/[no]): " CONFIRM
    if [ "$CONFIRM" != "yes" ]; then
        log "Operation cancelled by user."
        exit 0
    if [ ! -b "$BLOCKDEV_SPECIFIED" ]; then
      log "Error: Specified device $BLOCKDEV_SPECIFIED does not exist or is not a block device."
      exit 1
    fi

    # Determine partition naming convention
    if [[ $BLOCKDEV =~ [0-9]$ ]]; then
        PARTPREFIX="${BLOCKDEV}p"
    else
        PARTPREFIX="${BLOCKDEV}"
    BLOCKDEV="${BLOCKDEV_SPECIFIED##*/}"
    log "Using specified block device: $BLOCKDEV_SPECIFIED"
  else
    # Find a suitable block device
    if ! find_blockdev; then
      log "Error: No suitable block device found."
      exit 1
    fi
    log "Selected block device: /dev/${BLOCKDEV}"
  fi

  # Confirm with the user before proceeding
  read -p "This will erase all data on /dev/${BLOCKDEV}. Are you sure you want to proceed? (yes/[no]): " CONFIRM
  if [ "$CONFIRM" != "yes" ]; then
    log "Operation cancelled by user."
    exit 0
  fi

  # Determine partition naming convention
  if [[ $BLOCKDEV =~ [0-9]$ ]]; then
    PARTPREFIX="${BLOCKDEV}p"
  else
    PARTPREFIX="${BLOCKDEV}"
  fi

  # Wipe filesystem signatures
  log "Wiping filesystem signatures on /dev/${BLOCKDEV}..."
  wipefs -a -f /dev/${BLOCKDEV} || {
    log "Failed to wipe filesystem signatures."
    exit 1
  }

    # Wipe filesystem signatures
    log "Wiping filesystem signatures on /dev/${BLOCKDEV}..."
    wipefs -a -f /dev/${BLOCKDEV} || { log "Failed to wipe filesystem signatures."; exit 1; }

    # Create GPT partition table
    log "Creating GPT partition table on /dev/${BLOCKDEV}..."
    parted -s /dev/${BLOCKDEV} mklabel gpt || { log "Failed to create GPT partition table."; exit 1; }
  # Create GPT partition table
  log "Creating GPT partition table on /dev/${BLOCKDEV}..."
  parted -s /dev/${BLOCKDEV} mklabel gpt || {
    log "Failed to create GPT partition table."
    exit 1
  }

    # Create partitions
    log "Creating partitions on /dev/${BLOCKDEV}..."
    if ! parted -s /dev/${BLOCKDEV} "mkpart zosboot fat32 1MiB 300MiB set 1 esp on mkpart zoscache btrfs 300MiB 100%"; then
        log "Failed to create partitions."
        exit 1
    fi
  # Create partitions
  log "Creating partitions on /dev/${BLOCKDEV}..."
  if ! parted -s /dev/${BLOCKDEV} "mkpart zosboot fat16 1MiB 100MiB set 1 esp on mkpart zoscache btrfs 100MiB 100%"; then
    log "Failed to create partitions."
    exit 1
  fi

    # Format partitions
    log "Formatting /dev/${PARTPREFIX}1 as FAT32..."
    if ! mkfs.vfat -F32 /dev/${PARTPREFIX}1; then
        log "Failed to format /dev/${PARTPREFIX}1 as FAT32."
        exit 1
    fi
  # Format partitions
  log "Formatting /dev/${PARTPREFIX}1 as FAT32..."
  if ! mkfs.vfat -F32 -nZOSPXE /dev/${PARTPREFIX}1; then
    log "Failed to format /dev/${PARTPREFIX}1 as FAT32."
    exit 1
  fi

    log "Formatting /dev/${PARTPREFIX}2 as BTRFS..."
    if ! mkfs.btrfs -f /dev/${PARTPREFIX}2; then
        log "Failed to format /dev/${PARTPREFIX}2 as BTRFS."
        exit 1
    fi
  log "Formatting /dev/${PARTPREFIX}2 as BTRFS..."
  if ! mkfs.btrfs -f /dev/${PARTPREFIX}2 -LZOSCACHE; then
    log "Failed to format /dev/${PARTPREFIX}2 as BTRFS."
    exit 1
  fi

  # Mount and prepare directories
  MOUNT_POINT="zospxe"
  log "Creating and mounting directory ./${MOUNT_POINT}..."
  mkdir -p "${MOUNT_POINT}"
  if ! mount /dev/${PARTPREFIX}1 -t vfat "${MOUNT_POINT}"; then
    log "Failed to mount /dev/${PARTPREFIX}1 to ${MOUNT_POINT}."
    exit 1
  fi

    # Mount and prepare directories
    MOUNT_POINT="zospxe"
    log "Creating and mounting directory ./${MOUNT_POINT}..."
    mkdir -p "${MOUNT_POINT}"
    if ! mount /dev/${PARTPREFIX}1 -t vfat "${MOUNT_POINT}"; then
        log "Failed to mount /dev/${PARTPREFIX}1 to ${MOUNT_POINT}."
        exit 1
    fi
  log "Creating directory structure in ${MOUNT_POINT}..."
  mkdir -p "${MOUNT_POINT}/efi/boot"

    log "Creating directory structure in ${MOUNT_POINT}..."
    mkdir -p "${MOUNT_POINT}/efi/boot"

    # Download the required BOOTX64.EFI file
    # NOTE:: EDIT THIS WHEN APPROPRIATE AND IN PROD
    EFI_FILE_URL="https://bootstrap.grid.tf/uefi/${ENVIRONMENT}/${FARM}/debug/zero-os-development-zos-v4-debug-7d2de62033.efi?version=v4"
    # EFI_FILE_URL="https://bootstrap.grid.tf/uefi/${ENVIRONMENT}/${FARM}"
    log "Downloading BOOTX64.EFI from ${EFI_FILE_URL}..."
    if ! wget -q "${EFI_FILE_URL}" -O "${MOUNT_POINT}/efi/boot/BOOTX64.EFI"; then
        log "Failed to download BOOTX64.EFI."
        exit 1
    fi
  # Download the required BOOTX64.EFI file
  EFI_FILE_URL="https://bootstrap.grid.tf/uefi/${ENVIRONMENT}/${FARM}"
  log "Downloading BOOTX64.EFI from ${EFI_FILE_URL}..."
  if ! wget -q "${EFI_FILE_URL}" -O "${MOUNT_POINT}/efi/boot/BOOTX64.EFI"; then
    log "Failed to download BOOTX64.EFI."
    exit 1
  fi

    log "Download successful."
  log "Download successful."

    # Unmount the partition
    log "Unmounting ${MOUNT_POINT}..."
    umount -r "${MOUNT_POINT}" || { log "Failed to unmount ${MOUNT_POINT}."; exit 1; }
  # Unmount the partition
  log "Unmounting ${MOUNT_POINT}..."
  umount -r "${MOUNT_POINT}" || {
    log "Failed to unmount ${MOUNT_POINT}."
    exit 1
  }

    # Manage UEFI boot entries
    manage_boot_entries
  # Manage UEFI boot entries
  manage_boot_entries

    log "Operation completed successfully."
  log "Operation completed successfully."
 }

 # Execute the main function