Applies to openSUSE Leap 15

9 Introduction to the Booting Process

Abstract

Booting a Linux system involves different components and tasks. After a firmware and hardware initialization process, which depends on the machine's architecture, the kernel is started by means of the boot loader GRUB 2. After this point, the boot process is completely controlled by the operating system and handled by systemd. systemd provides a set of targets that boot configurations for everyday usage, maintenance or emergencies.

9.1 Terminology

This chapter uses terms that can be interpreted ambiguously. To understand how they are used here, read the definitions below:

init

Two different processes are commonly named init:

  • The initramfs process mounting the root file system

  • The operating system process that starts all other processes that is executed from the real root file system

In both cases, the systemd program is taking care of this task. It is first executed from the initramfs to mount the root file system. Once that has succeeded, it is re-executed from the root file system as the initial process. To avoid confusing these two systemd processes, we refer to the first process as init on initramfs and to the second one as systemd.

initrd / initramfs

An initrd (initial RAM disk) is an image file containing a root file system image which is loaded by the kernel and mounted from /dev/ram as the temporary root file system. Mounting this file system requires a file system driver.

Beginning with kernel 2.6.13, the initrd has been replaced by the initramfs (initial RAM file system) which does not require a file system driver to be mounted. openSUSE Leap exclusively uses an initramfs. However, since the initramfs is stored as /boot/initrd, it is often called initrd. In this chapter we exclusively use the name initramfs.

9.2 The Linux Boot Process

The Linux boot process consists of several stages, each represented by a different component:

9.2.1 The Initialization and Boot Loader Phase

During the initialization phase the machine's hardware is set up and the devices are prepared. This process differs significantly between hardware architectures.

openSUSE Leap uses the boot loader GRUB 2 on all architectures. Depending on the architecture and firmware, starting the GRUB 2 boot loader can be a multi-step process. The purpose of the boot loader is to load the kernel and the initial, RAM-based file system (initramfs). For more information about GRUB 2, refer to Chapter 12, The Boot Loader GRUB 2.

9.2.1.1 Initialization and Boot Loader Phase on AArch64 and AMD64/Intel 64

After turning on the computer, the BIOS or the UEFI initializes the screen and keyboard, and tests the main memory. Up to this stage, the machine does not access any mass storage media. Subsequently, the information about the current date, time, and the most important peripherals are loaded from the CMOS values. When the boot media and its geometry are recognized, the system control passes from the BIOS/UEFI to the boot loader.

On a machine equipped with a traditional BIOS, only code from the first physical 512-byte data sector (the Master Boot Record, MBR) of the boot disk can be loaded. Only a minimal GRUB 2 fits into the MBR. Its sole purpose is to load a GRUB 2 core image containing file system drivers from the gap between the MBR and the first partition (MBR partition table) or from the BIOS boot partition (GPT partition table). This image contains file system drivers and therefore is able to access /boot located on the root file system. /boot contains additional modules for GRUB 2 core as well as the kernel and the initramfs image. Once it has access to this partition, GRUB 2 loads the kernel and the initramfs image into memory and hands control over to the kernel.

When booting a BIOS system from an encrypted file system that includes an encrypted /boot partition, you need to enter the password for decryption twice. It is first needed by GRUB 2 to decrypt /boot and then for systemd to mount the encrypted volumes.

On machines with UEFI the boot process is much simpler than on machines with a traditional BIOS. The firmware is capable to read from a FAT formatted system partition of disks with a GPT partition table. This EFI system-partition (in the running system mounted as /boot/efi) holds enough space to host a fully-fledged GRUB 2 which is directly loaded and executed by the firmware.

If the BIOS/UEFI supports network booting, it is also possible to configure a boot server that provides the boot loader. The system can then be booted via PXE. the BIOS/UEFI acts as the boot loader. It gets the boot image from the boot server and starts the system. This is completely independent of local hard disks.

9.2.2 The Kernel Phase

When the boot loader has passed on system control, the boot process is the same on all architectures. The boot loader loads both the kernel and an initial RAM-based file system (initramfs) into memory and the kernel takes over.

After the kernel has set up memory management and has detected the CPU type and its features, it initializes the hardware and mounts the temporary root file system from the memory that was loaded with the initramfs.

9.2.2.1 The initramfs file

initramfs (initial RAM file system) is a small cpio archive that the kernel can load into a RAM disk. It is located at /boot/initrd. It can be created with a tool called dracut—refer to man 8 dracut for details.

The initramfs provides a minimal Linux environment that enables the execution of programs before the actual root file system is mounted. This minimal Linux environment is loaded into memory by BIOS or UEFI routines and does not have specific hardware requirements other than sufficient memory. The initramfs archive must always provide an executable named init that executes the systemd daemon on the root file system for the boot process to proceed.

Before the root file system can be mounted and the operating system can be started, the kernel needs the corresponding drivers to access the device on which the root file system is located. These drivers may include special drivers for certain kinds of hard disks or even network drivers to access a network file system. The needed modules for the root file system are loaded by init on initramfs. After the modules are loaded, udev provides the initramfs with the needed devices. Later in the boot process, after changing the root file system, it is necessary to regenerate the devices. This is done by the systemd unit systemd-udev-trigger.service.

9.2.2.1.1 Regenerating the initramfs

Since the initramfs contains drivers, it needs to be updated whenever a new version of one of its drivers is available. This is done automatically when installing the package containing the driver update. YaST or zypper will inform you about this by showing the output of the command that generates the initramfs. However, there are some occasions on which you need to regenerate an initramfs manually:

Adding Drivers Because of Hardware Changes

If you need to change hardware (for example, hard disks), and this hardware requires different drivers to be in the kernel at boot time, you must update the initramfs file.

Edit /etc/dracut.conf.d/01-dist.conf (or create it, if it does not exist) and add the following line.

force_drivers+="DRIVER1"

Replace DRIVER1 with the module name of the driver. If you need to add more than one driver, list them space-separated:

force_drivers+="DRIVER1 DRIVER2"

Proceed with Procedure 9.1, “Generate an initramfs”.

Moving System Directories to a RAID or LVM

Whenever you move swap files, or system directories like /usr in a running system to a RAID or logical volume, you need to create an initramfs that contains support for software RAID or LVM drivers.

To do so, create the respective entries in /etc/fstab and mount the new entries (for example with mount -a and/or swapon -a).

Proceed with Procedure 9.1, “Generate an initramfs”.

Adding Disks to a LVM Group /Btrfs RAID Containing the Root File System

Whenever you add (or remove) a disk to a logical volume group or a Btrfs RAID containing the root file system, you need to create an initramfs that contains support for the enlarged volume. Follow the instructions at Procedure 9.1, “Generate an initramfs”.

Proceed with Procedure 9.1, “Generate an initramfs”.

Changing Kernel Variables

If you change the values of kernel variables via the sysctl interface by editing related files (/etc/sysctl.conf or /etc/sysctl.d/*.conf), the change will be lost on the next system reboot. Even if you load the values with sysctl --system at runtime, the changes are not saved into the initramfs file. You need to update it by proceeding as outlined in Procedure 9.1, “Generate an initramfs”.

Procedure 9.1: Generate an initramfs

Note that all commands in he following procedure need to be executed as user root.

  1. Generate a new initramfs file by running

    dracut MY_INITRAMFS

    Replace MY_INITRAMFS with a file name of your choice. The new initramfs will be created as /boot/MY_INITRAMFS.

    Alternatively, run dracut -f. This will overwrite the currently used, existing file.

  2. (Skip this step if you ran dracut -f in the previous step). Create a link to the initramfs file you created in the previous step:

    (cd /boot && ln -sf MY_INITRAMFS initrd)

9.2.3 The init on initramfs Phase

The temporary root file system mounted by the kernel from the initramfs contains the executable systemd (which is called init on initramfs in the following, also see Section 9.1, “Terminology”. This program performs all actions needed to mount the proper root file system. It provides kernel functionality for the needed file system and device drivers for mass storage controllers with udev.

The main purpose of init on initramfs is to prepare the mounting of and access to the real root file system. Depending on your system configuration, init on initramfs is responsible for the following tasks.

Loading Kernel Modules

Depending on your hardware configuration, special drivers may be needed to access the hardware components of your computer (the most important component being your hard disk). To access the final root file system, the kernel needs to load the proper file system drivers.

Providing Block Special Files

The kernel generates device events depending on loaded modules. udev handles these events and generates the required special block files on a RAM file system in /dev. Without those special files, the file system and other devices would not be accessible.

Managing RAID and LVM Setups

If you configured your system to hold the root file system under RAID or LVM, init on initramfs sets up LVM or RAID to enable access to the root file system later.

Managing the Network Configuration

If you configured your system to use a network-mounted root file system (mounted via NFS), init must make sure that the proper network drivers are loaded and that they are set up to allow access to the root file system.

If the file system resides on a network block device like iSCSI or SAN, the connection to the storage server is also set up by init on initramfs. openSUSE Leap supports booting from a secondary iSCSI target if the primary target is not available. .

Note
Note: Handling of Mount Failures

If the root file system fails to mount from within the boot environment, it must be checked and repaired before the boot can continue. The file system checker will be automatically started for Ext3 and Ext4 file systems. The repair process is not automated for XFS and Btrfs file systems, and the user is presented with information describing the options available to repair the file system. When the file system has been successfully repaired, exiting the boot environment will cause the system to retry mounting the root file system. If successful, the boot will continue normally.

9.2.3.1 The init on initramfs Phase in the Installation Process

When init on initramfs is called during the initial boot as part of the installation process, its tasks differ from those mentioned above. Note that the installation system also does not start systemd from initramfs—these tasks are performed by linuxrc.

Finding the Installation Medium

When starting the installation process, your machine loads an installation kernel and a special init containing the YaST installer. The YaST installer is running in a RAM file system and needs to have information about the location of the installation medium to access it for installing the operating system.

Initiating Hardware Recognition and Loading Appropriate Kernel Modules

As mentioned in Section 9.2.2.1, “The initramfs file”, the boot process starts with a minimum set of drivers that can be used with most hardware configurations. On AArch64, POWER, and AMD64/Intel 64 machines linuxrc starts an initial hardware scanning process that determines the set of drivers suitable for your hardware configuration. On IBM IBM Z, a list of drivers and their parameters needs to be provided for example via linuxrc or a parmfile.

These drivers are used to generate a custom initramfs that is needed to boot the system. If the modules are not needed for boot but for coldplug, the modules can be loaded with systemd; for more information, see Section 10.6.4, “Loading Kernel Modules”.

Loading the Installation System

When the hardware is properly recognized, the appropriate drivers are loaded. The udev program creates the special device files and linuxrc starts the installation system with the YaST installer.

Starting YaST

Finally, linuxrc starts YaST, which starts the package installation and the system configuration.

9.2.4 The systemd Phase

After the real root file system has been found, it is checked for errors and mounted. If this is successful, the initramfs is cleaned and the systemd daemon on the root file system is executed. systemd is Linux' system and service manager. It is the parent process that is started as PID 1 and acts as init system that brings up and maintains user space services. See Chapter 10, The systemd Daemon for details.