In this article I will explain, step by step, how to install a NetBSD server with a root LFS partition.
Since LFS is considered experimental, it is highly advised to test this setup on a testbed / Virtual Machine. Use at your own risk.
In this setup, the server will solely run under LFS without any FFS partitions.
There are a lot of ways to accomplish this task. This is how I do it.
LFS is an implementation of a log-structured file system.
For example Sun's ZFS is a log-structured file system.
LFS can recover faster from a crash, because it does not need to fsck the whole disk. It is faster than FFS.
It has never worked reliably. It is limited to 2 Terabytes. It does not perform very well at constant high disk activity like ftp uploads/downloads. It can't handle situations where the disc is almost full, i.e. it usually crashes requiring a newfs, though most data can be recovered manually.
We need to install a NetBSD system from scratch without sysinst, since sysinst is lacking LFS support at the moment. This may change in the future. Requirements
Physical access to the server.
We will also need the NetBSD sets (base.tgz, comp.tgz, etc.tgz, man.tgz, misc.tgz, text.tgz ...). It is recommended to download all sets.
You can either download the latest sets from the NetBSD autobuild cluster (ftp://ftp.netbsd.org/pub/NetBSD-daily/HEAD/) or you can build your own release and use your own sets. I recommend to use the latest NetBSD sources.
A tutorial on how to build current can be found here: How to build NetBSD-current.
The sets have to be accessible from the liveCD in some way. For example via http, ftp or scp.
Please boot into the liveCD on the server you want to install.
Gain root privileges (su -).
Use fdisk to create an active NetBSD (ID 169) partition.
# fdisk -iu wd0
Use disklabel to prepare your disk. This part of the tutorial is with purpose not very detailed. You should get comfortable with disklabel beforehand.
# disklabel -i -I wd0
on the command line to enter the interactive disklabel menu. I am assuming you are using wd0. Otherwise substitute with your drive (sd0, ld0...)
We will create one big "a" partition in this example. Feel free to try another setup in your second try.
In disklabel create one big partition "a" spanning the whole disk starting from sector 63 (63s) until the end minus the space you want to give to the swap partition.
Use 4.4LFS as your file system.
Partition b is used as swap. Start from the end of partition a until the end ($).
Partition c and d are the disks itself and should be of type unused starting from 0 to the end.
Remove all other partitions (e-p).
When you are finished your label should look like this:
# size offset fstype [fsize bsize cpg/sgs] a: 73400320 63 4.4LFS 0 0 0 # (Cyl. 0*- 72817*) b: 2097152 73400383 swap # (Cyl. 72817*- 77504*) c: 78124937 63 unused 0 0 # (Cyl. 0*- 77504*) d: 78125000 0 unused 0 0 # (Cyl. 0 - 77504*)
Label the disk (N), Write changes to disk (W), and quit (Q).
You can now create the LFS filesystem on the disk you just labeled.
# newfs_lfs wd0a
There are more options like -A and different segment and frag sizes. But we will stick to the default 1M segment size, since other values may get LFS unstable.
The rest is trivial. We mount the filesystem and extract our sets.
# mkdir /tmp/targetroot # mount /dev/wd0a /tmp/targetroot
Create another directory to store the sets in.
# mkdir /tmp/sets
Change in that directory
# cd /tmp/sets
And download your sets, for example via ftp. This are the sets you have prepared upfront by either compiling a release or downloading them from the autobuild cluster.
# ftp 192.168.0.200 ...
extract your sets using option -p (important).
# cd /tmp/sets # tar xvzpf base.tgz -C /tmp/targetroot
repeat with all your sets, but extract only one GENERIC kernel named kern-GENERIC.tgz
change into /tmp/targetroot and do a base configuration. Edit etc/fstab
/dev/wd0a / lfs rw 1 1 /dev/wd0b none swap sw 0 0 ptyfs /dev/pts ptyfs rw 0 0 tmpfs /tmp tmpfs rw
ptyfs and tmpfs are optional, but recommended.
Copy boot to the targetroot.
# cp /tmp/targetroot/usr/mdec/boot /tmp/targetroot
# /usr/sbin/installboot -v -m i386 -o timeout=5,console=pc /dev/rwd0a /tmp/targetroot/usr/mdec/bootxx_lfsv2
Don't forget to create all devices.
# cd /tmp/targetroot/dev # ./MAKEDEV all
this may take a while.
That's it. Sync and reboot.
# sync # sync # sync # reboot
If everything went well, your system should boot. Once you have logged in, you can configure your system and do all the fine tuning.
You should not fill up your LFS partition over 75%. This could damage the file system (at the moment).
If you want to install an LFS root file system on your server in your data center remotely, console access is beneficial, but not necessary. The minimum requirement is a rescue console. This is mostly a linux ramdisk. One way is to build a custom boot floppy including LFS and newfs_lfs. Because newfs_lfs does not fit on the disk, you have to exclude unnecessary tools. Then write a small shell script that is executed when you boot the floppy, summing up all steps in the tutorial including adding a user account and setting up ifconfig, resolv.conf, default gateway, to be able to log in afterwards. Make a backup of the first 5 MB you are going to overwrite with dd. Now just dd the floppy image to server harddisk and reboot. Good luck.