Text 6223, 75 rader
Skriven 2006-06-05 19:17:04 av Maurice Kinal (1:140/13.1)
   Kommentar till text 6222 av Paul Rogers (1:105/360.0)
Ärende: You gotta smile
=======================
Hey Paul!

Jun 05 11:06 06, Paul Rogers wrote to Maurice Kinal:

 PR> Soooooo, if I understand this right...
 PR> 0) running under a different version of the kernel than the
 PR>    target

Correct.  One possible glitch is chrooting to a 2.6 enviroment from a 2.4
enviroment.  I think the other way around will work but won't if the host
system is 2.4 based and the target is 2.6 based.  However the special lilo
configuration will still work.

 PR> 1) the target is mounted

Yes.

 PR> 2) virtually all the files restored, /[,s]bin, /lib, et al.,

The rootfs has to be in place on the targetted drive.  You only need to make
sure that the basics are configured though, such as fstab, inittab, and the
rc's wrt what you will require of the target once the drive is physically wired
into the target machine.  If no monitor and keyboard is on the target machine
then you'll more then likely want to be able to login to it later via a
networking interface to be able to alter anything.

 PR> 3) chroot to the target (chroot /mnt/hda6 'bash lfs-60-fstab')
 PR>         Ummm, where does my PATH point to now?
 PR>         Same named directories in the target?
 PR>         What about environment?
 PR>         chroot wants a command, does that work?

Once you are chrooted into it from the host machine it doesn't matter what the
/etc/fstab on the target claims unless you mount /proc while in chroot seeing
the target will more then likely be different.  That is why I suggested running
lilo from the host and use the '-C' switch with the extra-special-lilo.conf
file although that will have to be editted to reflect both the host's and
target's system as they are unlikely to be the same.  They could be but it is
not a given.  Hence the 'disk=/dev/hdx bios=0x80' configuration line.

 PR> 4) everything in the target (bash, programs called by the
 PR>    script, shared libraries, /etc/*) comes from the target

Okay.  That is what should happen and the method should work as long as the
/etc/fstab has the target's information correct.  Also a 'rdev
/mnt/boot/vmlinux /dev/hdxx' from the host to reflect the partition the target
is going to be using to mount the rootfs is a good idea.  For instance for a
ramdisk this command would become 'rdev /mnt/boot/vmlinuz-whatever /dev/ram0'.

 PR> 5) if the kernel & target are reasonably close the ABI for the
 PR>    libraries should work

Yes.  However for my latest 64-bit system built from a 32-bit system that isn't
the case other then the versions.  I cannot chroot into it so I need to make
sure everything is preconfigured for the target in order to kickstart it on the
target.  For the sata drive the /mnt/etc/fstab needs to reflect sdx's rather
then hdx's and 'rdev /mnt/boot/vmlinuz-whatever /dev/sda1' since I want it to
boot and mount the rootfs to that particular partition.

 PR> 6) kernel should have the modules it needs already loaded?
 PR>         grub/lilo writing the MBR is OK?

The modules should be in /mnt/lib/modules which will magically become
/lib/modules when you actually boot it up.  As for grub I never use it for this
particular purpose but I know for a fact lilo works with this scheme onto
hotswapped drives on the host machine no matter what type (compactflash, usb
drives, regular ide and scsi drives, ide flash drives, ????).  It won't work
for phoney floppy boots though and I've never tried it with CD/DVD's seeing
they suck and I don't like them.

Life is good,
Maurice

--- Msged/LNX 6.2.0
 * Origin: The Pointy Stick Society XVIII - Don't know what I want (1:140/13.1)