Automated installs (very long)

[Louis Mandelstam]

On Tue, 7 Oct 1997, Jason Gerrity wrote:

> We use the DOS msclient boot disk to boot the pc and the installer has to
> log in to connect to the network share where the image is stored (Louis,
> you mentioned that you use the linux dos boot, I would be VERY interested
> to see how you did this).

I've commented about it in some posts of mine to samba-list, but let me
summarize, so you can ask more specific questions about the various steps
if needed.  Note I do NOT need to cater for workstations having different
kinds of network cards, they all have 3C905b-TPOs in them.  I'm only
including the really necessary steps.

1. Installer sets BIOS to boot A:,C: (we usually keep it C:,A: - with the
   BIOS setup protected from casual meddling by a password)

2. Installer inserts "Magic workstation install disk" (which is Linux
   based), and boots the computer.

3. Linux boots, gets IP settings using bootp, mounts a shared directory
   from the server via NFS, jumps to a script on the share, with all the
   necessary binaries on the server as well.  Only things such as bootpc
   and ifconfig are on the inird on the Linux boot disk.

4. Install script prepares the local hard disk by writing the partition
   table the way we want it, writing a suitable Master Boot Record, and
   creating a bootable DOS 7 filesystem as C: (a previous post detailed
   how this is done)

5. Install script mounts the newly created C: into its filespace as a vfat
   fs.

6. Install script extracts a previously creating .tgz file from the server
   onto the local hard drive - this is the directory tree and files we
   want on the local hard drive.

7. Install script umounts the hard drive, sets the CMOS to boot C:,A:
   again, and reboots.


The installation as it ends up on the workstation has an entry in
HKLM\SOFTWARE\Microsoft\Windows\CurrentVersion\RunServices which tries to
run a .PIF file on \\server\netlogon which is named to match the
"distribution" version on that workstation.  (My netlogon is mode 755 and
public = yes)

If that PIF file is not found, nothing happens and the user can log in.

If ever I want all workstations run version xx of the distribution to come
and get the latest setup from the server next time they are booted, I
softlink xx.pif to a .pif file in the netlogon share which starts up a
.bat file which installs loadlin.exe and the same kernel image and initrd
as used by the boot disk onto C: modifies autoexec.bat to run loadlin and
thus boot the install system, then forces a reboot (I'm not too sure about
the reliability of my reboot method (running from RunServices, ie Windows
is already running - and at the login dialog box) but it seems to work so
far.

This saves me from even the much reduced hassle of taking my magic disk
for a tour of all the workstations if ever I need to change something so
drastic on all of them that it makes sense to re-install them.  Hence one
of the needs for keeping user files off C: since then I need to worry
about loosing them during a re-install.

Regards

---------------------------------------------------------------|-----|--
Louis Mandelstam              Tel +27 83 227-0712   Symphony  /|\   /|\
Linux systems integration     http://sr.co.za       Research {   } {   }
Johannesburg, South Africa    mailto:louis at sr.co.za (Pty)Ltd {___} {___}