Application note: Moving /usr under redhat linux

This application note details how to install/upgrade RedHat Linux with /usr as a symlink to a directory in another partition.

I have some otheroriginal linux pages, as well.

Typical Configuration

• Root partition small (50mb), has almost nothing which is normally written during system operation, this reduces the possibility of normal defragmentation corrupting essential system utilities, etc. which are useful for booting a minimal system for rescue operations.
• swap partition
• big partition /disk0 (additional partions /disk1, /disk2 on successive drives ).
• a small scratch partition which can be reconfigured as necessary to deal with the occasional need to put certain files (such as /tmp) on a filesystem with odd permissions).

   /usr         --> /disk0/usr
   /var         --> /disk0/var
   /tmp         --> /disk0/var/tmp
   /local       --> /disk0/local
   /usr/local   --> /disk0/local
   /home/user1  --> /disk0/home/user1
   /home/user2  --> /disk0/home/user2
   /home/user3  --> /disk0/home/user3
   etc.

Installation

Under Redhat 4.0, install normally but name your big disk partition /usr and then after the install is complete, boot single user and move /usr to /disk0/usr using the script movestuff which must be run in single user mode.

Note that moving /usr on a running system is somewhat dangerous since you are moving some of the tools which the system needs to operate and it needs to operate while moving those tools. I have used this script successfully on many systems. If you type the commands by hand, however, be warned that a single typo can lead to much chaos. The script expects you to confirm each line by pressing return.

Note that the movestuff script creates a couple symbolic links:

   /disk0/etc  --> /etc
   /disk0/etc  --> /bin

Package Upgrades

Major Upgrade

Sample upgrade

This was initially tested on a system with a JAZ drive and an Adaptec 2940au controller (this controller is incompatible with the normal 2940 which caused install problems on redhat 4.x systems). It had two partitions (not counting swap): a small root ("/") and large "/disk0" with virtually everything.

In real life, performing a major upgrade can be very dangerous. It might go smoothly but if anything goes wrong it can go very wrong. You should have multiple backups (preferrably using more than one backup program) before proceeding with an upgrade; you should expect that your primary copy (hard disk) may effectively be destroyed. Some of the disasters which may occur include symbolic link confusion, and disk full errors. Be sure to check the root partition and the partition which contains /usr to make sure they have enough space for an upgrade. /usr will probably grow by a couple hundred megabytes for a 4.0 to 5.0 upgrade. The root partition may be used for some temporary files, and may grow some due to rpmsaves, new kernels, new /etc files, new files in /bin, /lib, etc.etc. After the sample upgrade, I had 20573 kilobytes used and 27MB free (it may have required somewhat more space during the transition); I don't remember what it started at, so these numbers may not be terribly useful.

This upgrade was from RedHat 4.0 (full install) to RedHat 5.0 (full install except howto-translations). The system was a virgin install except that two scripts had been run( movestuff and services) to reorganize the filesystem as described in this document and disable any unwanted services. (Note that this is an old version of the services script. I have a newer one for RedHat 5.0 systems).

• Boot: expert
  Actually, the system did not go into expert mode as it should have, so the rest of the install was actually in the normal (fast track to hell) install mode. I tried several times to get it to use expert mode. Strange, because, I have used it successfully before. Expert mode is generally preferred because it makes it easier to backtrack if you entered a wrong value, if you get a keybounce on the return key, or are trying something sneaky.
• Color Monitor: yes
• Keyboard Type: US
• PCMCIA: no
• Insert CD ROM: ok
• "in second stage install" briefly appears in lower left hand corner of screen and then disappears. The second stage install will start a shell on console 2.
• Installation Path: Install/Upgrade
  Don't answer this question yet. Switch to a shell prompt by switching to console 2 by hitting Alt-F2. Now, type:

     ln -s /mnt/disk0/ /disk0
  

  Then switch back to console 1 using Alt-F1.
• Select "upgrade"
• SCSI Adapters: yes, 2940, autoprobe
• Upgrade Packages?
  Don't answer yet. Switch to a shell prompt by switching to console 2 by hitting Alt-F2. Now, type:

    cd /disk0
    rm etc; ln -s /sysroot/etc etc
    rm bin; ln -s /sysroot/bin bin
    ln -s /mnt /sysroot
    ln -s / /mnt/sysroot
    cd /       
  

  The last "cd /" is so the installer won't have problems later unmounting the filesystem because it is busy due to the current directory for your shell. A major component of this trick is the two /sysroot links which insure that a symbolic link which references /sysroot/some/where will point to /some/where on your target system regardless of whether it is being interpretted in the bogus install root filesystem or an environment chroot()ed to the root partition in your target environment
• Switch back to console 1 using Alt-F1. Now go through and select all the new packages you want installed (I selected everything besides howto-translations).
• Install will proceeed.
• If you get error messages that look like this:

  RPM install of xxxx failed: execution of script failed
  where xxx=automake, bc, bison, cvs, diffutils, ed, emacs, findutils, etc.
  (about 5% of the packages) then you probably did not do something 
  correctly regarding the symbolic links for /disk0/etc and /disk0/bin 
  via /sysroot.  These errors probably are the result of install scripts
  that recklessly use or rely on relative symbolic links over too large 
  a distance.
  You probably will get two error messages:
  Failed to install gettext.  
  Gettext0.10-5 (which is newer) is already installed.
  
  Failed to install rdate.  
  rdate-10.10-94.4 (which is newer) is already installed.
  
  These errors are likely the result of the version numbering system
  changing for a particular package and do not have anything to do
  with /usr being in a different place.  Note that these errors
  somehate evenly spaced out during the one hour upgrade operation
  that is normally unattended so you can't just go home and expect it
  to be finished the next day.
  reboot single
  Most of the stuff past this point is just cleaning up the
  normal mess which redhat makes and has little or nothing to do
  with the peculiarities related to a system where /usr has been moved.
  cp /tmp/upgrade.log /root/upgrade.log
  Turn off all the unwanted new services which were enabled and all
  the old services which were re-enabled by the upgrade process.
  (you can use my scripts, located nearby, to do this.
  mkdir /mnt/cdrom
  mount /dev/hdb /mnt/cdrom
  cd /mnt/cdrom/RedHat/RPMS
  rpm -U --force gettext-0.10-4.i386.rpm
  rpm -U --force rdate-0.960923-3.i386.rpm
  rpm -U --force libtermcap-2.0.0-6.i386.rpm
  
  there was an "execution of script failed" error for this one.
  Reinstalling in the normal environment seems to work.
  fgrep -i rpmsave /root/upgrade.log
  
  Find out all the configuration files which were butchered by rpm.  You
  will need to go fix those manually.
  reboot