5. Installing: kernel, RAM disk and packages

5.2.1. Boot process overview

To boot a system with root-on-LVM-on-RAID (or just with root-on-LVM) a RAM disk is needed: this is necessary because root file system resides in a volume group, so to be able to access root file system LVM must first be started. Commands (and maybe modules) to accomplish this task are stored into a RAM disk, which is automatically loaded and used by the kernel.

When booting a system with root-on-LVM-on-RAID the following steps are performed:

• LILO (or another boot loader with RAID support) is loaded.

• Kernel image is loaded.

• If RAID support is built in the kernel RAID arrays are auto-detected and started.

• RAM disk image with LVM support is loaded and uncompressed.

• RAM disk is mounted as root.

• If RAID and LVM support is not built in the kernel these modules must be included in the RAM disk and loaded.

• If RAID support has been loaded as a module RAID arrays must be started manually (RAM disk must be modified).

• LVM is started.

• Volume groups are detected and activated.

• "Real" root file system over LVM is mounted read-only as new root file system.

• RAM disk is unmounted.

• LVM is started again (root has changed).

• Root file system is checked for errors.

• If no errors are found root file system is remounted read-write.

• From this point on boot process continue as normal (modules are loaded, remaining file systems are mounted, …).

5.2.2. RAID and LVM support loaded as a module vs statically linked

By choosing the first solution, modular support, there is one great advantage: the stock bf2.4 kernel shipped with Debian "Woody" 3.0 CD has RAID and LVM compiled as a module, so there is no need to compile a custom kernel. Moreover the kernel can be installed from main menu.

The main disadvantage of this solution is that with RAID support loaded as a module, RAID auto-detect does not work. To workaround this problem the RAM disk must be modified to make it start RAID arrays. The real problem is that array configuration has to be included into the RAM disk, and thus the RAM disk must be modified for each single system that is to be installed. This is not very good if several different systems are to be installed. Even worst in some cases if array configuration is changed, for example an array is added, the RAM disk must be updated.

The second solution, a custom kernel with built-in RAID and LVM support, has working RAID arrays auto-detect, and does not require manual RAM disk modifications. The drawback is that kernel compilation is required and that the kernel has to be installed manually.

5.2.3. Install stock bf2.4 Debian kernel with modular RAID and LVM support

To install the stock bf2.4 kernel choose "Install Kernel and Driver Modules" from the main install menu and then "Configure Device Driver Modules" to configure any required module.

5.2.4. Install custom kernel with statically compiled RAID and LVM

It is not possible to install a custom kernel from the install menu: it's necessary to perform this step from command line. Probably to transfer the kernel image to the system, network access will be necessary along with some tools like ftp or wget.

For all these reasons installation of a custom kernel has to be performed later: necessary steps are reported in section Install custom kernel with statically compiled RAID and LVM.

5.7.1. Custom kernel compilation

To compile a custom kernel it is necessary to download the source, unpack it, configure it or copy a configuration file and then run the compile. While it is possible to compile the kernel "on-the-fly" during installation, this requires installing some additional packages, at least on Debian. It is better to compile the kernel on an already installed system, then pack the binaries and transfer them to the system that is being installed.

In order to have a kernel able to boot a RAID or LVM system there are a few options that must be set, these are:

• Multiple device support (CONFIG_MD=y)

• RAID device support (CONFIG_BLK_DEV_MD=y)

• RAID 0 support, if using RAID 0 arrays (CONFIG_MD_RAID0=y)

• RAID 1 support, if using RAID 1 arrays (CONFIG_MD_RAID1=y)

• RAID 5 support, if using RAID 5 arrays (CONFIG_MD_RAID5=y)

• LVM support, if using LVM (CONFIG_BLK_DEV_LVM=y)

Moreover, since a RAM disk has to be loaded, following options must be set:

• Loopback device support (CONFIG_BLK_DEV_LOOP=y)

• RAM disk support (CONFIG_BLK_DEV_RAM=y)

• Maximum RAM disk size, default is 4096 (4 MB), do not change this (CONFIG_BLK_DEV_RAM_SIZE=4096)

• Initial RAM disk support (CONFIG_BLK_DEV_INITRD=y)

Optionally devfs support can be compiled into the kernel. devfs is, in kernel 2.4.18, an experimental feature: it is a virtual /dev file system (like /proc) that replaces classical /dev file system. Main advantages of devfs are that devices are dynamically registered as needed, and this happens just for devices that are effectively used by the system. For more information look at devfs documentation in the kernel source.

• devfs support (CONFIG_DEVFS_FS=y)

• Automatically mount devfs at boot (CONFIG_DEVFS_MOUNT=y)

If devfs support is enabled, it is recommended that devfsd is installed, look at section Optional: install devfsd

Kernel Sources can be downloaded from The Linux Kernel Archives http://www.kernel.org or installed with distribution specific means (deb packages for Debian, rpm packages for Red Hat).

For more information on kernel compilation look at Appendix D: kernel compilation tips.

5.11.1. Install RAM disk for kernel bf2.4

Download this RAM disk or create one as explained in section RAM disk for bf2.4 kernel: requirements and creation, then copy it to /boot on the system that's being installed and create a symbolic link /boot/initrd to it.

ln -s /boot/initrd.gz /boot/initrd

5.11.2. Install RAM disk for custom kernel

Download this RAM disk or create one as explained in section RAM disk for custom kernel: requirements and creation, then copy it to /boot on the system that's being installed and create a symbolic link /boot/initrd to it.

ln -s /boot/initrd.gz /boot/initrd

5.11.3. RAM disk for custom kernel: requirements and creation

The requirements for a RAM disk to be used with a kernel that has RAID and LVM support statically compiled will be presented first, since they are simpler than the ones for kernel with modular support.

With this kind of kernel RAID array autodetect and autostart at boot works. Moreover the RAM disk must not contain modules for RAID and LVM support (they already are in the kernel!).

The role of the RAM disk in this case is limited to start LVM subsystem (vgscan) and activate volume group(s) (vgchange).

Requirements are as follows:

• RAM disk must contain commands vgscan and vgchange.

• RAM disk must contain commands mount and umount.

• Any library needed by previous commands, particularly liblvm1.0.so.1

• A linuxrc file that issues commands necessary to activate LVM and volume group(s).

The linuxrc file must be like the one shown below, the same file can be downloaded here:

#!/bin/sh
/bin/mount /proc
/sbin/vgscan
/sbin/vgchange -a y
/bin/umount /proc

The RAM disk can be created at hand (modify an existing one or create it with mkinitrd and then modify it) or it can be created automatically using lvmcreate_initrd: this command creates a RAM disk that mets above requirements and puts it into /boot.

It is suggested that RAM disk creation is performed prior to installing the wanna be RAID system (on another already installed system) or that an available RAM disk is used.

The RAM disk can be created during install process, but this can be a little tricky: lvmcreate_initrd tryies to put into the RAM disk modules from the directory /lib/modules/kernel_version, where kernel_version is the kernel used for boot: if the command is used during the installation, the same kernel used for boot (ex. bf2.4) must already have been installed on the system, even if later the system will be booted with another kernel (a custom one).

Ex. Debian install is booted using the 2.4.18-bf.24 kernel, if during install lvmcreate_initrd is called depmod will state that it cannot find modules.dep. To be able to create the RAM disk with this command the kernel bf2.4 must first be installed.

To modify an existing RAM disk or to examine an existing one:

• If the RAM disk is compressed, uncompress it with:

  gzip -cd /boot/initrd.gz > /boot/initrd_unc

  This uncompress the RAM disk in a copy: using gzip -d the compressed file will be removed. It will be necessary to generate it again with a gzip command. If these steps are performed on an already installed system remember to run lilo after.

• Mount the RAM disk with:

  mount -o loop /boot/initrd_unc /initrd

• Examine or modify the RAM disk in /initrd

• Unmount the RAM disk with:

  umount /initrd

• If necessary compress again the RAM disk:

  gzip /boot/initrd_unc

5.11.4. RAM disk for bf2.4 kernel: requirements and creation

Read previous section for RAM disk requirements for static RAID support: in this section are illustrated only differences with that case. The same for commands necessary to RAM disk creation.

Since the kernel has no RAID and LVM support the modules for this must be included in the RAM disk:

• md.o

• raid1.o

• lvm-mod.o

With RAID and LVM support compiled as a module, RAID autodetect will not work at boot. This requires that the RAM disk contains not only commands for starting LVM but also for starting RAID arrays (mdadm or raidstart).

Finally linuxrc file must include commands to load the modules and for starting RAID arrays. Below is shown a modified script, the same file can be downloaded here:

#!/bin/sh
/sbin/modprobe md
/sbin/modprobe raid1
/sbin/modprobe lvm-mod
/sbin/mdadm -A -R /dev/md1 /dev/hda1 /dev/hdb1
/sbin/mdadm -A -R /dev/md2 /dev/hda2 /dev/hdb2
/bin/mount /proc
/sbin/vgscan
/sbin/vgchange -a y
/bin/umount /proc

As it can be easily noticed the script is dependent from disks and arrays configuration, so it must be modified to be adapted for each system: a mdadm command must be added for each RAID array.

To modify the RAM disk:

• If the RAM disk is compressed, uncompress it.

• Mount the RAM disk over /initrd

• Add required modules to the RAM disk:

  cp /lib/modules/2.4.18-bf2.4/kernel/drivers/md/md.o /initrd/lib/modules/2.4.18-bf2.4/kernel/drivers/md/ cp /lib/modules/2.4.18-bf2.4/kernel/drivers/md/raid1.o /initrd/lib/modules/2.4.18-bf2.4/kernel/drivers/md/ cp /lib/modules/2.4.18-bf2.4/kernel/drivers/md/lvm-mod.o /initrd/lib/modules/2.4.18-bf2.4/kernel/drivers/md/

• Modify or copy linuxrc as shown above to include calls to mdadm.

• Add mdadm to the RAM disk:

  cp /sbin/mdadm /initrd/sbin/mdadm

• Unmount the RAM disk with.

• If necessary compress again the RAM disk.

5.12.1. /etc/raidtab

The file /etc/raidtab is needed for raid arrays management, create or copy it, look at section Create RAID arrays.

	Note
	In the chrooted shell the RAM disk is not accessible. To copy the file raidtab from the RAM disk use a non-chrooted shell: cp /etc/raidtab /target/etc/ To do this it is possible to exit from the chrooted shell and the reopen it. Otherwise remount the floppy with all config files in the chrooted shell and copy raidtab from floppy.

5.12.2. /etc/fstab

File Systems Table: debian install does not detect file systems over lvm, so /etc/fstab created by install process will be empty; the file must be created by hand or copied.

Here can be downloaded an example /etc/fstab file, the same file is shown below. Modify it to reflect current configuration: boot raid device, logical volumes name or number.

# /etc/fstab: static file system information.
#
# <file system>  <mount point>  <type>   <options>          <dump>  <pass>
/dev/vg00/root   /              ext3     errors=remount-ro  0       1
/dev/vg00/swap   none           swap     sw                 0       0
proc             /proc          proc     defaults           0       0
/dev/fd0         /floppy        auto     user,noauto        0       0
/dev/cdrom       /cdrom         iso9660  ro,user,noauto     0       0
/dev/md1         /boot          ext2     defaults           0       2
/dev/vg00/home   /home          ext3     defaults           0       2
/dev/vg00/opt    /opt           ext3     defaults           0       2
/dev/vg00/tmp    /tmp           ext3     defaults           0       2
/dev/vg00/usr    /usr           ext3     defaults           0       2
/dev/vg00/var    /var           ext3     defaults           0       2

5.12.3. /etc/lilo.conf

To make the system bootable lilo has to be configured.

Since version 22.0 lilo raid support has been made more powerful.

Taken from lilo changelog:

Changes from version 21.7.5 to 22.0 (29-Aug-2001) John Coffman [released 9/27]
Boot Installer
--------------
- RAID installations now create a single map file, install the
boot record on the RAID partition, install auxiliary boot
records only on MBRs if needed, except BIOS device 0x80. Back-
ward compatibility is possible with new config-file and command
line options (raid-extra-boot= or '-x' switch). Even with
stored boot command lines ('-R', lock, fallback), RAID set
coherency can be maintained.

To have lilo boot from a RAID system use:

boot=/dev/md1

Change /dev/md1 if the boot device is different.

To use a root-on-LVM file system include the line:

root=/dev/vg00/root

Change /dev/vg00/root if the volume group name or logical volume name are different.

Now disks whose MBR is to be written by lilo have to be indicated with option raid-extra-boot; from lilo.conf man page:

raid-extra-boot=<option>
This option only has meaning for RAID1 installations. The <option> may be
specified as none, auto, mbr-only, or a comma-separated list of devices; e.g.,
"/dev/hda,/dev/hdc6". Starting with LILO version 22.0, the boot record is
normally written to the first sector of the RAID1 device. Use of an explicit
list of devices, forces writing of auxiliary boot records only on those devices
enumerated, in addition to the boot record on the RAID1 device. Since the
version 22 RAID1 codes will never automatically write a boot record on the MBR
of device 0x80, if such a boot record is desired, this is the way to have it
written.

So add this line to lilo.conf:

raid-extra-boot="/dev/hda, /dev/hdb"

Change the disks to reflect system configuration.

Have the kernel mount root read-only by adding this option:

read-only

Finally specify an image to be loaded at boot with related RAM disk.

image=/boot/vmlinuz
    label=Linux
    initrd=/boot/initrd

With non standard disks setup options disk and bios it could be necessary to specify correspondence between disks and BIOS numbers. Look at lilo.conf man page for more details.

Below is shown an example /etc/lilo.conf, the same file can be downloaded here.

boot=/dev/md1
root=/dev/vg00/root
raid-extra-boot="/dev/hda, /dev/hdb"
read-only
image=/boot/vmlinuz
    label=Linux
    initrd=/boot/initrd

Change the devices, kernel image name and RAM disk name as required or create symbolic links /boot/vmlinuz and /boot/initrd to regular files.

5.12.4. /etc/modules

File /etc/modules contains modules to be loaded at startup. If the stock kernel is being used and modules configuration has been performed from the main menu (look at section Install stock bf2.4 Debian kernel with modular RAID and LVM support) this file should be already ok. Just check it with:

cat /etc/modules

If a custom kernel has been installed manual module configuration will be necessary.

If any required module or parameter is missing add it. An example of /etc/modules is shown below:

# /etc/modules: kernel modules to load at boot time.
#
# This file should contain the names of kernel modules that are
# to be loaded at boot time, one per line.  Comments begin with
# a "#", and everything on the line after them are ignored.

usb-uhci
input
usbkbd
keybdev

This file can be downloaded here.